Video feedback for parental sensitivity and attachment security in children under five years

Summary of findings for the main comparison. Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment

Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment
Patient or population: parent‐child dyads (including foster or adoptive carers), where the child was aged between birth and four years 11 months (inclusive), and where problems had been identified that were impacting or might impact on the parent's sensitivity Setting: community, hospital outpatient and hospital inpatient Intervention: video feedback Comparison: no intervention or inactive alternative intervention
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Outcomes	Risk with no intervention or inactive alternative intervention	Risk with video feedback	Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Parental sensitivity Follow‐up: postintervention or short‐term follow‐up	‐	The mean parental sensitivity score in the intervention group was 0.34 standard deviations higher (0.20 higher to 0.49 higher)	‐	1757 dyads (20 RCTs)	⊕⊕⊕⊝ Moderate^a	Higher scores indicate a better outcome. Effect size of 0.33 standard deviations compares favourably to other similar interventions.
Parental reflective functioning	‐	‐	‐	‐	‐	No study reported this outcome.
Attachment security Measured by: Strange Situation Procedure (odds of being securely attached) Follow‐up: postintervention	Study population		OR 3.04 (1.39 to 6.67)	166 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a better outcome.
	341 per 1000	612 per 1000 (419 to 776)	OR 3.04 (1.39 to 6.67)	166 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a better outcome.
Attachment security Measured by: Attachment Q‐sort Follow‐up: postintervention	The mean attachment security score across control groups ranged from 0.33 to 0.37 (scores can range from + 1.00 to −1.00)	The mean attachment security score in the intervention group was0.02 standard deviations higher (0.33 lower to 0.38 higher)	‐	131 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Effect size of 0.02 standard deviations indicates no evidence of effectiveness.
Adverse events: parental stress Follow‐up: postintervention or short term	‐	The mean parental stress score in the intervention group was 0.09 standard deviations lower (0.26 lower to 0.09 higher)	‐	537 dyads (8 RCTs)	⊕⊕⊝⊝ Low^b,c	Higher scores indicate a worse outcome. Effect size of 0.09 standard deviations indicates no evidence of effectiveness.
Adverse events: parental anxiety Follow‐up: short term	‐	The mean parental anxiety score in the intervention group was0.28 standard deviations lower (0.87 lower to 0.31 higher)	‐	311 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^c,d,e	Higher scores indicate a worse outcome. Effect size of 0.28 compares favourably to other similar interventions.
Child behaviour Follow‐up: long term	‐	The mean child behaviour score in the intervention group was 0.04 standard deviations higher (0.33 lower to 0.42 higher)	‐	119 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a worse outcome. Effect size of 0.04 standard deviations indicates no evidence of effectiveness.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aDowngraded one level due to inconsistency: moderate heterogeneity, which was not explained by our subgroup analysis. ^bDowngraded one level for risk of bias: we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias. ^cDowngraded one level due to imprecision: low number of participants, leading to wide confidence interval. ^dDowngraded one level due to publication bias: few studies in this review reported this outcome. ^eDowngraded one level due to inconsistency: high heterogeneity.

Background

available in

Description of the condition

Attachment

A child’s relationship with his or her primary carer is the first, and arguably the most important, relationship formed following birth. The primary carer is normally, but not always, the child's birth mother or father. This emotional bond between a child and their primary caregiver is known as a 'selective attachment' relationship. Attachment is a biobehavioural system that has evolved over time. It is intended to bring about protection in the face of both perceived danger as well as the fear that comes with it (Bowlby 1969). When a child is distressed, he or she is programmed to seek and secure proximity to, and contact with, the primary caregiver (Bowlby 1969). Attachment behaviour may be activated by circumstances internal to the child such as illness, hunger or pain; by separation from the primary caregiver such as when a mother leaves the room or discourages proximity; or by external events that cause distress such as frightening events or rejection by others (Bowlby 1969). Depending on the intensity of the threat, the attachment behaviour may be terminated by the appearance of the caregiver or physical contact with them. The younger the child, or the more serious the threat, the more likely that only physical reassurance and containment will provide comfort. The attachment relationship, therefore, is a dynamic one, in which the child plays an active part (see Shin 2008). It has been described by Zeanah and colleagues as a reciprocal relationship of seeker (child) and provider (parent), the purpose of which is to comfort children when they are upset, support the development of emotional regulation, and offer security (Zeanah 1993).

Children whose caregivers provide sensitive and responsive care develop secure attachments to those carers. Children who experience insensitive, unpredictable or intrusive parenting develop attachments that are insecure, putting them at risk of adverse consequences for a range of aspects of their psychosocial development, including being more reliant on teachers; showing less positive affective expression and impaired social problem‐solving skills; showing more frustration and less persistence; more negative responses to others and less overall social competence (Sroufe 2005). In addition to children being classified as either secure or insecure, they may also be classified as disorganised, when there is evidence of a conflict between wanting to approach and wanting to avoid the caregiver when the attachment system is activated (Main 1990a). Disorganised attachment occurs when children are frightened of the caregiver and have been exposed to a range of anomalous, atypical parent‐child interactions (Madigan 2006). Disorganised attachment is associated with predictors of later psychopathology, including externalising behaviours (Fearon 2010), and personality disorders (Steele 2010). Many studies consider only secure and insecure attachment patterns when classifying children, as these were the attachment patterns that were first described (Weinfield 2004). These studies (carried out in the general population) typically find that approximately 60% of children are securely attached, and the remainder (40%) are insecurely attached (Moullin 2014). For insecurely attached children, 25% learn to avoid their parent when they are distressed (avoidant attachment) and 15% learn to resist the parent, often because the parent responds unpredictably or amplifies their distress (resistant attachment; Moullin 2014). In studies where disorganised attachment has been included, around 40% of disadvantaged children are classified as disorganised (Weinfield 2004), and as many as 80% of abused children receive this classification (Cyr 2010).

Although children usually have a particularly strong bond with one primary caregiver, most have more than one attachment relationship, often with fathers, siblings and grandparents as well as with mothers (see, for example, Hallers‐Haalboom 2014). As such, children can be securely attached to one parent but insecurely attached to another. The role of early relationship experiences and the development of child self‐regulatory skills have been linked to the child’s ability to control behavioural and physiological responses such as anger (Gilliom 2002), aggression (Alink 2009), and anxiety (Hannesdottir 2007).

Caregiver sensitivity

One key predictor of child attachment status is the parent's attachment status (Van Ijzendoorn 1995). The impact of the parent's attachment status on the child's attachment appears to be mediated by parental sensitivity to child cues.

Ainsworth and colleagues defined sensitivity as a mother’s ability to attend and respond to her child in ways that accurately match her child's needs (Ainsworth 1978). Sensitive and responsive parents do the following: notice a child’s signals; interpret these signals correctly; and respond to signals in a timely and appropriate manner (Ainsworth 1974). The concept of sensitivity, therefore, refers not to a specific set of maternal behaviours but to something much more dynamic and relational.

Parental sensitivity can be compromised by a variety of factors. These include social influences such as social isolation (Belsky 2002; Kivijärvi 2004), or domestic violence (Levendosky 2006); psychological factors such as maternal depression (NICHD Early Child Care Research Network 1999; Karl 1995; Murray 1997), or personality disorder (Laulik 2013); maternal history of maltreatment (Pereira 2012), substance dependence (Eiden 2014), low self‐esteem (Leerkes 2002; Shin 2008); or cognitive factors such as maternal preconceptions about parenting (Kiang 2004; Leerkes 2010). Child characteristics can also impact negatively on parental sensitivity, including child prematurity (Singer 1999); the presence of excessive negative child behaviour, such as general distress (Leerkes 2002); and the child's proneness to anger (Ciciolla 2013), and irritability (Van den Boom 1991). Some studies have examined father involvement as a mediator of maternal sensitivity (see, for example, Stolk 2008), whilst others have examined the role of the father as caregiver (see, for example, Pelchat 2003). Comparative research on the relative sensitivity of mothers and fathers is scarce and therefore the findings are somewhat inconclusive; some studies report fathers as less sensitive than mothers (see Hallers‐Haalboom 2014; Heerman 1994; Lovas 2005), while others have found no difference (Pelchat 2003).

Although parental sensitivity has been found to be an important predictor of child attachment security, a systematic review of the antecedents to attachment security suggests that it explains around only one third of the variance (De Wolff 1997). Research has also highlighted the importance of mid‐range contingency (Beebe 2010), and maternal reflective functioning (Slade 2005), or mind‐mindedness (Meins 2001). Mid‐range contingency refers to the ability of the parent to regulate flexibly both their own internal emotional states and their interaction with the baby, and is characterised by moments of synchrony or attunement, followed by rupture and then repair. A study by Beebe 2010 found that interaction that occurred outside this mid‐range, resulting from the parent’s preoccupation either with self‐regulation (e.g. depressed parents) or interactive regulation (e.g. anxious parents), was associated with an insecure or disorganised attachment).

Reflective functioning is a term that describes a parent's ability to comprehend their child's behaviour with regards to their internal mental states (Slade 2005). High reflective functioning correlates with positive maternal parenting traits, such as flexibility and responsiveness. Low reflective functioning can be seen in tandem with negative maternal behaviours, such as withdrawal and intrusiveness (Kelly 2005; Slade 2005). Similarly, mind‐mindedness, which refers to the parent's ability both to understand a young child's state of mind and to respond appropriately, has been associated with behavioural sensitivity and interactive synchrony (Meins 2001), and to better predict attachment security of the child at one year of age than maternal sensitivity (Lundy 2003; Meins 2001; Meins 2012).

Other studies have identified the importance of a range of atypical or anomalous parent‐child interactions characterised as ‘Fr‐behaviours’, which are the behaviours of parents who are either frightened or frightening, or both (Jacobvitz 1997; Main 1990b), or who are hostile and helpless (Lyons‐Ruth 2005). Fr‐behaviours have been described as subtle (e.g. periods of being dazed and unresponsive) or more overt (deliberately frightening children; Lyons‐Ruth 2005), and are strongly associated with disorganised attachment (Madigan 2006).

Description of the intervention

Video feedback is a generic term that refers to the use of videotaped interactions of the parent and child to promote parental sensitivity; it has other names, including Video Interaction Guidance (VIG), Interaction Guidance (IG), Video Home Training (VHT) and Video Feedback Intervention to Promote Positive Parenting (VIPP). Developed by Harrie Biemans and colleagues in the 1980s, video feedback is a relationship‐based parenting intervention that aims to enhance maternal sensitivity at the behavioural level (Kennedy 2010). The core aspects of interventions based on video feedback are as follows.

Video‐recording the parent‐child interaction during play or aspects of daily caregiving.
Editing the recording to select micro‐moments of interaction that demonstrate the child's contact initiatives and examples of the parent's attuned response to these signals.
Parent and 'guider' (the person responsible for the therapy) jointly reviewing the recordings, with the guider providing praise to the parent, not for the attunement per se but for engaging in the evaluation of the interactions being viewed.

The intervention model is underpinned by two core concepts: intersubjectivity and mediated learning. Intersubjectivity, or 'shared moments of attunement', is modelled by the therapist (or 'guider') in their relationship and interactions with the parent, as well as being identified in the video recordings of the parent‐child interaction. Mediated learning, or 'scaffolding', refers to the role adults play in helping children learn how to do things that they might not otherwise manage alone. Mediated learning is also modelled by the guider in his or her relationships with the parent, as the guider helps the parent to describe what is happening in the clips being viewed, and what the parent and child in the video might be thinking or feeling, and to identify the consequences for the parent and the child (Kennedy 2011).

Video feedback may be delivered on a one‐to‐one (e.g. VIPP, VIG) or group (e.g. Circle of Security (CS)) basis, and has been used with first‐time mothers (Klein Velderman 2006); hard‐to‐reach families (Kennedy 2010); parents of premature infants (Hoffenkamp 2015); parents with mental health problems, including postpartum depression (Vik 2006); parents of autistic children (Poslawsky 2015), maltreated children (Moss 2011), and adopted children (Juffer 1997); parents of children with atopic dermatitis (Cassibba 2015); ethnic‐minority parents (Yagmur 2014); and parents with an eating disorder (Stein 2006). Although video feedback is usually delivered in the home environment, it has also been used in clinical settings, such as hospital environments with mothers of preterm babies (Hoffenkamp 2015), and residential treatment centres (Kennedy 2010). It is now used in over 15 countries by practitioners who work in a range of helping professions (e.g. social work, education and health; Kennedy 2010).

How the intervention might work

In terms of the underpinning theoretical model, most forms of video feedback are attachment‐based in the sense that they aim to enhance maternal sensitivity, and promote optimal child social and emotional development (Klein Velderman 2006a), with the longer‐term goal of promoting improved child attachment (Juffer 2008). However, the presumed mechanisms by which this is achieved vary across the different models of video feedback. All video‐feedback interventions primarily target the behavioural level using video‐recorded episodes of the parent‐child interaction. The guider provides an opportunity for the caregiver to experience attuned interactions with a sensitively attuned adult (Kennedy 2011), and also to view themselves in interaction with their child and observe positive responses from them. Together, these are hypothesised to bring about a range of meta‐cognitive changes that result from the discrepancy between their own beliefs about their ability to parent and what they can see on the video, in addition to an increase in feelings of empowerment and self‐efficacy, and their ability for self‐reflection (Kennedy 2011).

Some models of video‐feedback intervention include additional components that may provide a more explicit focus on representational issues. For example, Video Feedback Intervention to Promote Positive Parenting with Discussions on the Representational Level (VIPP‐R; Juffer 2008), involves the therapist addressing the mother's representations and attachment using discussions that may, for example, focus explicitly on the mother's own experiences of separation in early childhood and those experienced with her own child (Klein Velderman 2006a).

Other models involve the inclusion of teaching about sensitive discipline techniques, such as Video Feedback Intervention to Promote Positive Parenting ‐ Sensitive Discipline (VIPP‐SD). There is evidence that suggests the effectiveness of video feedback may vary with factors at both the level of the parent and of the child. For example, Klein Velderman 2006 reported that amongst mothers with insecure attachments, those classed as 'insecure dismissing' (who idealise their own parents or minimise the importance of attachment relationships in their own lives) benefited most from video feedback, whilst those classed as 'insecure preoccupied' benefited most when they participated in video feedback together with further discussions about their individual experiences of attachment in childhood.

Why it is important to do this review

Improvement of the health and well‐being of children is part of a global agenda. While the basic needs of children (e.g. food, sanitation, health care) are paramount to survival and development, living with an adult who is responsive to their needs is also important (Jones 2003). UNICEF 2008 highlights that a loving, stable and stimulating relationship with caregivers in the earliest months and years of life are critical for every aspect of a child’s development. Specifically, the empirical literature shows that maternal sensitivity is a key predictor of child attachment security (De Wolff 1997), and that a secure attachment promotes more optimal childhood development (Sroufe 2005), while an insecure or disorganised attachment predicts later behaviour problems (Fearon 2010), and psychopathology (Steele 2010). Research suggests that early, targeted interventions are potentially an effective means of increasing parental sensitivity (Bakermans‐Kranenburg 2003), and although there have been a number of reviews of the impact of video feedback on a range of outcomes including maternal sensitivity (Balldin 2018; Fukkink 2008; Juffer 2018; NICE 2016; Van den Broek 2017), only two conducted meta‐analyses. Fukkink 2008 concluded that video feedback was an effective means of improving parenting behaviour and attitudes, and child development. However, the report did not provide the search dates for the review, which was submitted in June 2008; did not search a wide range of databases; and was very broad in its scope, including all uses of video feedback with no age limits on the children (who ranged in age from birth to seven years, with an average age of 2.4 years (standard deviation (SD) 2.7 years). More importantly, the study authors paid little attention to the quality of included studies (that is, there were no 'Risk of bias' assessments) and included studies without random assignment. Juffer 2018 looked only at studies using one type of video feedback known as 'Video Interaction to promote Positive Parenting' (VIPP) and found that VIPP was effective in improving parental sensitivity.

This systematic review of current best evidence addresses the methodological weakness of Fukkink 2008 and has a broader scope than Juffer 2018. It will be of interest to policy makers and practitioners internationally.

Objectives

available in

To assess the effects of video feedback on parental sensitivity and attachment security in children aged under five years who are at risk for poor attachment outcomes.

Methods

available in

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) and quasi‐RCTs (in which the allocation to study arms is not truly random; for example, allocation is done through a form of alternation such as days of the week or by date of birth). We included cluster‐ and cross‐over RCTs.

We excluded studies that had an alternative treatment but no control group. Alternative treatment controls are not appropriate when seeking to investigate the effectiveness of an intervention, which was the aim of this review (this is in accordance with advice from Cochrane Developmental, Pyschoscial and Learning Problems).

Types of participants

We included parent‐child dyads (including foster or adoptive carers) or family units, where the child was aged between birth and four years 11 months (inclusive), and where problems had been identified that might impact or were impacting the parent's sensitivity (e.g. poor bonding, depression, eating disorders, maltreatment) or child attachment (e.g. behaviour problems, challenging temperament, preterm birth). The majority of studies looked at parent‐child dyads.

If studies included a proportion of participants aged above four years 11 months, we endeavoured to obtain data on the sample aged up to four years 11 months; where this was not possible, we used outcome data that included children outside our target age group in the meta‐analysis (e.g. Moss 2011; Poslawsky 2015).

We excluded studies in which the intervention was used with a population group where neither parents nor children had any risk factors for attachment problems.

Types of interventions

We included video‐feedback interventions delivered in any setting, in which the parent and child were filmed and then feedback was provided to the parent either on a one‐to‐one basis or in groups, with the aim of improving the sensitivity of their interactions with the child, child attachment, or the reflective functioning of the parent.

We included interventions that, in addition to video feedback, also provided a small number of additional sessions related to the primary aim of the intervention; for example, VIPP‐R or VIPP‐SD.

We included studies in which the intervention was compared with no treatment, an inactive alternative intervention or treatment as usual. Examples of control treatment included a sequence of telephone calls with a parent (Barone 2019; Kalinauskiene 2009; Negrão 2014; Van Zeijl 2006; Yagmur 2014); a limited number of home visits with (1) video recordings between parent and child with no feedback (Benzies 2013; Koniak‐Griffin 1992; Moran 2005); (2) by a play and development service (Green 2010); (3) discussions about parenting (Poslawsky 2015); standard hospital care (Hoffenkamp 2015) or routine care at well baby units (Høivik 2015).

We excluded studies comparing video feedback with other interventions, as well as:

interventions in which video feedback was used as part of a wider set of methods of working with the family and in which we could not differentiate the effect of video feedback, and
programmes that used videotape modelling or videotaped vignettes of parents and their children (e.g. Webster‐Stratton 2015).

Types of outcome measures

We excluded studies that did not measure parental sensitivity or did not do so in an objective way; for example, studies relying on self‐report measures, such as the Parent‐Child Dysfunctional Interaction (PCDI) subscale of the Parenting Stress Index (Abidin 1995).

Primary outcomes

Parental sensitivity, measured by, for example, the Ainsworth Sensitivity Scale (ASS; Ainsworth 1974), the Child‐Adult Relationship Experimental Index (CARE‐Index; Crittenden 2001), the Parental Sensitivity Assessment Scale (PSAS; Hoff 2004), Coding Interactive Behaviour (CIB; Feldman 1998), the Emotional Availability (EA) scales (Biringen 2000a), Global Ratings Scales (GRS) of mother‐infant interaction (Murray 1996), Maternal Behaviour Q‐sort (MBQS; Pederson 1999), or Nursing Child Assessment Teaching Scale (NCATS; Sumner 1994).
Parental reflective functioning, measured by, for example, the Parent Development Interview (PDI; unpublished manuscript by Aber 1985), or the PDI‐Revised (PDI‐R; unpublished manuscript by Slade 2004).
Attachment security, measured by, for example, the Attachment Q‐sort (AQS; Waters 1985; Waters 1987), or the Strange Situation Procedure (SSP; Ainsworth 1978).
Adverse effects. We acknowledge that a worsening of any of our primary outcomes listed above would be considered an adverse effect. However, we also considered the effects of the intervention on parental anxiety and stress, measured by, for example, the Parenting Stress Index (PSI; Abidin 1995), or the Parenting Stress Scale (PSS; Berry 1995).

Secondary outcomes

Child mental health, measured by behavioural assessments of emotional disorders, hyperactivity and conduct disorders.
Child physical and socioemotional development, measured through, for example, the Bayley Scales of Infant and Toddler Development, Third Edition (Bayley‐III; Bayley 2006), or the Strengths and Difficulties Questionnaire (SDQ; Goodman 1997).
Child behaviour, measured by, for example, the Child Behaviour Assessment Instrument (CBAI; Samarakkody 2010).
Costs, measured by direct costs stated by studies.

Timing of outcome assessment

We collected outcome data at time points provided within the included studies and grouped these as postintervention (immediately upon completion of the intervention), short term (up to six months), medium term (up to one year) and long term (over one year).

Search methods for identification of studies

We ran the first database searches in August and September 2016 (Electronic searches) followed by searches of other resources in July 2017 (Searching other resources). In November 2018 we updated the searches, including bibliography screening, and ran further searches of other resources in July 2019. We did not apply any date or language restrictions to the electronic searches and had two papers translated into English (Bovenschen 2012; Kalinauskiene 2009).

Electronic searches

We searched the electronic databases and trials registers listed below.

Cochrane Central Register of Controlled Trials (CENTRAL; 2018, Issue 11), in the Cochrane Library and which includes the Cochrane Developmental, Psychosocial and Learning Problems' Specialised Register (searched 10 November 2018)
MEDLINE Ovid (1946 to November Week 1 2018)
Embase Ovid (1974 to 2018 Week 44)
CINAHL Plus EBSCOhos (Cumulative Index to Nursing and Allied Health Literature; 1937 to 10 November 2018)
PsycINFO Ovid (1806 to 2018 Week 44)
Sociological Abstracts ProQuest (1952 to 10 November 2018)
Social Sciences Citation Index Web of Science (SSCI; 1970 to 10 November 2018)
Social Services Abstracts ProQuest (1979 to 10 November 2018)
Conference Proceedings Citation Index ‐ Social Science & Humanities Web of Science (CPCI‐SS&H; 1990 to 10 November 2018)
LILACS (Latin American and Caribbean Health Science Information database; 1985 to current; lilacs.bvsalud.org/en; searched 10 November 2018).
Cochrane Database of Systematic Reviews (CDSR; 2018; Issue 11), part of the Cochrane Library (searched 10 November 2018)
Database of Abstracts of Reviews of Effects (DARE; 2015; Issue 2. Final issue), part of the Cochrane Library (searched 10 November 2018)
Networked Digital Library of Theses and Dissertations (NDLTD; www.ndltd.org; searched 10 November 2018)
WorldCat (limited to dissertations and theses; www.worldcat.org; searched 10 November 2018)
Clinicaltrials.gov (Clinicaltrials.gov; searched 10 November 2018)
World Health Organization International Clinical Trials Registry Platform (WHO ICTRP; www.who.int/ictrp/en; searched 10 November 2018)

Searching other resources

Two review authors (ES and LOH) screened the bibliographies of included studies (Barlow 2016; Barone 2019; Benzies 2013; Bovenschen 2012; Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Lam‐Cassettari 2015; Moran 2005; Moss 2011; Negrão 2014; Platje 2018; Poslawsky 2015; Seifer 1991; Stein 2006; Van Zeijl 2006; Yagmur 2014) and relevant reviews (Balldin 2018; Fukkink 2008; Juffer 2018; NICE 2015; Van den Broek 2017), to identify any additional relevant publications. They also searched the websites of the following relevant organisations and government departments: United Nations International Children's Emergency Fund (UNICEF) Global Evaluation Database (unicef.org/evaldatabase); National Society for the Prevention of Cruelty to Children (NSPCC) Impact and Evidence Hub (nspcc.org.uk/services‐and‐resources/impact‐evidence‐evaluation‐child‐protection); and the Association for Video Interaction Guidance UK (AVigUK; videointeractionguidance.net) (see Appendix 1). One review author (ES) visited the websites of research groups we knew to be conducting work in this area to screen their listed publications (see Appendix 1). Another review author (LOH) also used Google Scholar to search the internet for unpublished work (see Appendix 1).

Although originally planned (O'Hara 2016), we did not contact experts to enquire about other published work or unpublished work (Table 1).

Table 1. Methods for use in future updates of this review

Issue	Method
Searching other resources	We will draft a list of included studies to send to experts in the field and ask them to forward to us any published, unpublished or ongoing work that we may have missed.
Measures of treatment effect	Continuous outcome data If necessary, we will compute effect estimates from P values, T statistics, analysis of variance (ANOVA) tables or other statistics, as appropriate.
Measures of treatment effect	Multiple outcomes When a study provides multiple, interchangeable measures of the same construct at the same point in time (e.g. multiple measures of maternal sensitivity), we will calculate the average SMD across these outcomes and the average of their estimated variances. This strategy aims to avoid the need to select a single measure and to avoid inflated precision in meta‐analyses (i.e. preventing studies that report on more outcome measures receiving more weight in the analysis than comparable studies that report on a single outcome measure).
Unit of analysis issue	Cluster‐RCTs In the event that we identify relevant cluster‐RCTs that meet the inclusion criteria of the review, we will deploy appropriate statistical methods based on the guidance provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Where study authors have dealt appropriately with the clustered design in their analyses, we will try to obtain direct estimates of the effect (e.g. an OR with its CI). Where study authors have not dealt appropriately with the cluster design in their analyses, we will extract or calculate effect estimates and their SEs as for a parallel‐group trial, and adjust the SEs to account for the clustering (Donner 1980). To do this, we will need to identify an appropriate ICC, which describes the relative variability in outcome within and between clusters (Donner 1980). Where available, we will look for this information in the reports of relevant trials. If this is unavailable, we will try to obtain the information from the study authors. If this proves unsuccessful, we will use external estimates obtained from similar studies. We will find closest‐matching scenarios (regarding both outcome measures and types of clusters) from existing databases of ICCs. If we are unable to identify any matches, we will perform sensitivity analyses using a high ICC of 0.1, a moderate ICC of 0.01 and a small ICC or 0.001, to cover a broader range of plausible values while still allowing for strong design effects for smaller studies (see Sensitivity analysis). Furthermore, we will combine these estimates and their corrected SEs from the cluster‐RCTs with those from parallel designs using the generic inverse variance method in Review Manager 5 (Review Manager 2014).
Dealing with missing data	Data imputation Where it has not been possible to obtain any unreported data from authors of included studies, and there is reason to believe that it is not missing at random, we will follow the recommendations in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011, Section16.1), and we will do the following: Where appropriate, develop a strategy for data imputation (if we assume the data to be not missing at random). In the case of data imputation, we will specify the methods used in the 'Characteristics of included studies’ tables. We will describe other missing data and dropouts/attrition for each included study in the ‘Risk of bias’ tables, and we will discuss the extent to which these missing data could alter the results or conclusions of the review.
Dealing with missing data	Meta‐regression We will assess the sensitivity of any primary meta‐analyses to missing data using meta‐regression to test for any effect of missingness on the summary estimates (Higgins 2011, Section 16.1.2).
Data synthesis	In the occurrence of severe funnel plot asymmetry, we will present both fixed‐effect and random‐effects analyses under the assumption that asymmetry suggests that neither model is appropriate. If both indicate a presence (or absence) of effect we will be reassured; if they do not agree we will report this.
Subgroup analyses	We will investigate heterogeneity using subgroup analyses or meta‐regression, if appropriate. We will group the included studies and analyse them according to the intervention approach, including the following. Delivery method (i.e. group‐based versus individual delivery) Participating child (e.g. pre‐birth or highly temperamental babies)
Sensitivity analysis	We will assess the robustness of findings to decisions made in obtaining them by conducting sensitivity analyses. We will perform sensitivity analyses by conducting the following reanalysis. Reanalysis excluding studies with imputed data
CI: confidence interval; ICC: intra‐class correlation coefficient; OR: odds ratio; RCT: randomised controlled trial; SD: standard deviation; SMD: standardised mean difference; VIG: Video Interaction Guidance; VIPP‐R: Video‐feedback to promote Positive Parenting ‐ Representational level; VIPP‐SD: Video‐feedback to promote Positive Parenting ‐ Sensitive Discipline

Data collection and analysis

For this section, we only report those methods used in this review. Other methods that were not relevant to the available data, or that we could not use for other reasons, are summarised in Table 1. One of the review authors (JB) is an author of an included study (Barlow 2016). JB was not involved in data extraction or assessment of risk of bias; the review authors involved in this did not need to seek any further advice on either of these areas with regards to this particular study.

Selection of studies

At least two review authors (from ES, LOH and NH) independently screened titles and abstracts yielded by the searches against the inclusion criteria for the review (Criteria for considering studies for this review). The review authors retrieved the full‐text reports of all studies selected for potential inclusion, or those where there was some uncertainty, and assessed the reports for eligibility. Where review authors could not agree, they further discussed the papers with JB or GM. In one case, Mendelsohn 2005, we wrote to the study authors for the purposes of clarifying whether or not the study met our inclusion criteria (Table 2). We list excluded studies in the Characteristics of excluded studies tables, together with the reason for their exclusion. We report the flow of studies in a PRISMA diagram (Liberati 2009).

Table 2. Summary of contact with study authors

Study	Date contact initiated	Reason for contacting study authors	Response received
Barone 2019	9 July 2019 (Smith 2019a [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore. We also requested study data for overall child behaviour, as the published report only contained externalising behaviour.	The study author provided the maternal sensitivity and child behaviour data for inclusion in the meta‐analysis (Barone 2019 [pers comm]). They excluded children aged 5 years and over from the data sent over, as the original study did include these children.
Bovenschen 2012	12 January 2018 (Smith 2018b [pers comm])	The reported data were not labelled sufficiently clearly in the published paper to be used in the meta‐analysis. We requested clarification from the study authors.	The study author provided the necessary, additional study data, so they could be included in meta‐analysis (Bovenschen 2019 [pers comm]).
Hodes 2017	2 June 2017 (Smith 2017a [pers comm])	We requested study data relating to parental stress outcomes, reanalysed for children within included age range.	We received no response. As a result, we did not subsequently request the study data on Harmonious Parent‐Child Interaction to be analysed for children within included age range.
Hoffenkamp 2015	16 January 2017 (O'Hara 2017a [pers comm])	We requested missing study data relating to parental sensitivity outcomes.	The study author provided the missing data so they could be included in the meta‐analysis (Van Bakel 2017 [pers comm]).
Høivik 2015	8 February 2018 (Smith 2018c [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore.	The study author provided maternal sensitivity data for inclusion in the meta‐analysis (Hoivik 2018 [pers comm]).
Klein Velderman 2006	21 February 2018 (Smith 2018g [pers comm])	The outcomes data for maternal stress were not included in published studies.	The corresponding author provided us with the missing data for the purposes of meta‐analysis (Bakermans‐Kranenburg 2018 [pers comm]).
Koniak‐Griffin 1992	8 February 2018 (Smith 2018d [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore.	We received no response.
Lam‐Cassettari 2015	2 June 2017 (Smith 2017b [pers comm])	The published study data for maternal sensitivity outcomes included children aged 5 years and over. We requested outcomes data with those children excluded.	The study author provided the data with those children aged 5 years and over excluded (Lam‐Cassettari 2018 [pers comm]).
Mendelsohn 2008	1 February 2018 (Smith 2018i [pers comm]	We requested a copy of the conference abstract.	We received no response.
Moran 2005	3 June 2017 (O'Hara 2017b [pers comm])	The maternal sensitivity outcomes were reported as means without standard deviations or standard errors. We requested these data so they could be used in the meta‐analysis.	The corresponding author no longer had access to the data due to retirement, so could not provide the information (Moran 2017 [pers comm]).
Moss 2011	12 January 2018 Smith 2018h [pers comm]	The published study data for maternal sensitivity outcomes included children aged 5 years and over, We requested outcomes data with those children excluded.	We received an initial response from the study authors but they did not subsequently provide the data (Dubois‐Comtois 2018 [pers comm]).
Negrão 2014	12 January 2018 (Smith 2018e [pers comm])	The maternal stress outcomes data were not reported in the published paper, so we requested this information for the purposes of the meta‐analysis.	The study author provided these data for the purposes of meta‐analysis (Pereira 2018 [pers comm]).
Poslawsky 2015	12 January 2018 (Smith 2018f [pers comm])	The reported outcomes included children aged 5 years or older. We requested outcomes data with those children excluded. We also requested means and standard deviations for the relevant 3‐month follow‐up outcome (daily hassles).	The corresponding author was unable to provide the requested data (Poswlawsky 2018 [pers comm]).
Seifer 1991	22 July 2019 (Smith 2019b [pers comm])	We requested outcomes data for mental and psychomotor development	We received no response.
Stein 2006	22 May 2018 (Barlow 2018 [pers comm])	The outcomes data for 'Verbal responses to infant cues' were reported as medians, so we requested the means and standard deviations.	The study authors provided us with these data for the purposes of meta‐analysis (Stein 2018 [pers comm]).

Data extraction and management

Two review authors (from NH, LOH, ES) independently extracted data from each included study and recorded the following information on a pre‐piloted data extraction form.

Participant characteristics (age, gender, ethnicity, location)
Intervention characteristics (including delivery, duration, outcomes and measures, and within‐intervention variability)
Comparison characteristics (including whether the study used an active or inactive comparison)
Study characteristics (study design, sample size, length of follow‐up, attrition or dropout, handling of missing data, methods of analysis, dates of study, funding sources, conflicts of interest)
Outcome data (relevant details on all primary and secondary outcome measures used, and summary data, including means, standard deviations (SDs), confidence intervals (CIs) and significance levels for continuous data and proportions for dichotomous data)

Review authors resolved disagreements through discussion. Where clarity was needed over whether an outcome in a study was relevant, the reviewer authors sought advice from JB.

Assessment of risk of bias in included studies

Two review authors (ES for all studies, with either LOH or NH) independently assessed the risk of bias of each included study using the Cochrane 'Risk of bias' tool (Higgins 2017). They assigned judgements of low, high or unclear risk of bias for each of the following domains, using the criteria set out in Appendix 2: sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting and other bias. Where review authors did not agree after discussion, they discussed further with another author (JB or NL). We recorded the judgements in 'Risk of bias' tables.

Measures of treatment effect

We calculated unadjusted treatment effects using Review Manager 5 (RevMan 5) (Review Manager 2014).

Dichotomous outcome data

We calculated the odds ratio (OR) with 95% CI for dichotomous outcomes. For dichotomous outcomes that we included in the 'Summary of findings' tables, we expressed the results as absolute risks and used high and low observed risks among the control groups as reference points.

Continuous outcome data

For continuous outcomes, we calculated the mean difference (MD) if all included studies used the same measurement scale, or the standardised mean difference (SMD) if studies used different measurement scales, and 95% CIs. We calculated SMD using Hedge's g. In one instance, we converted an SMD from Cohen's d to Hedge's g.

Economics issues

We reviewed studies for data on the costs of programmes within the included studies.

Unit of analysis issues

Studies with multiple treatment groups

In the primary analysis, we combined results across all eligible intervention groups and compared them with the combined results across all eligible control groups, and made single pair‐wise comparisons. Where studies compared more than one form of video interaction with a control group or groups, such that combining them prevented investigation of potential sources of heterogeneity, we analysed each video interaction group separately (against a common control group) but divided the sample size for common comparator groups proportionately across each comparison (Higgins 2011; Section 16.5.5). This simple approach allows the use of standard software and prevents inappropriate double counting of individuals. We applied this latter approach to three studies (Benzies 2013; Hoffenkamp 2015; Klein Velderman 2006).

Dealing with missing data

Where necessary, one review author (LOH or ES) contacted the authors of included studies requesting them to supply any unreported data such as missing outcome data (e.g. group means and SDs and details of number of dropouts). Details of which study authors we contacted and why are in the Characteristics of included studies tables and Table 2.

If we were not able to obtain unreported outcome data, we followed the recommendations in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011, Section 16.1) and did the following:

Analysed the data available, as we assumed the data were missing at random.

Two studies had unreported outcome data on parental sensitivity that the study authors were unable to provide: Koniak‐Griffin 1992 reported a result for a scale with parental sensitivity as a subdomain, but did not report the subdomains; and Moran 2005 reported means but not SDs or standard errors (SE). We did not impute this unreported data, as we assumed the data were missing at random.

Assessment of heterogeneity

We assessed clinical heterogeneity across included studies by examining the distribution of important participant factors (e.g. age) and intervention characteristics (e.g. style, setting, personnel, context of delivery) among studies. The details of this information are included in the Characteristics of included studies tables, and discussed in the Results section.

We assessed methodological heterogeneity across included studies by comparing the distribution of study factors (e.g. allocation concealment, blinding of outcome assessment, losses to follow‐up, treatment type, cointerventions). This information is contained in the Characteristics of included studies tables and 'Risk of bias' tables, and considered in the Discussion.

We described statistical heterogeneity by computing the I² statistic (Higgins 2002), which describes approximately the proportion of variation in point estimates that is due to heterogeneity rather than sampling error. In addition, we used the Chi² test (P < 0.10) of homogeneity to detect the strength of evidence that heterogeneity is genuine (Deeks 2017).

Assessment of reporting biases

We drew a funnel plot (estimating differences in treatment effects against their standard error (SE)) when we identified 10 or more studies that provided data on an outcome; in this case, parental sensitivity. We assessed the funnel plot by visual inspection and also by Egger's regression test (Egger 1997). We redrew the funnel plot without an outlying study (Green 2010), to better assess the asymmetry.

We considered the reasons for any asymmetry. Asymmetry might be due to publication bias, but might also reflect a relationship between study size and effect size, such as when larger studies have lower compliance, and compliance is positively related to effect size. It may also be due to clinical variation between the studies (Sterne 2017, Section 10.4), for example the study population, reflecting true heterogeneity.

As a direct test for publication bias, we compared results extracted from published journal reports with results obtained from other sources for the two outcomes for which this was possible, parental sensitivity and parental stress. In these cases we obtained some outcome data directly from study authors that were not reported in the published papers (see Table 2).

Data synthesis

Where interventions were similar in terms of (1) the age of the child(ren), (2) parent gender and (3) intensity, frequency and duration of video feedback, we synthesised results in a meta‐analysis.

We used both fixed‐effect and random‐effects models and compared the results to assess the impact of statistical heterogeneity. Except where the model was contraindicated (e.g. if there was funnel plot asymmetry), we present the results from the random‐effects model. When we report the results of the random‐effects model, we include an estimate of the between‐study variance (Tau²).

We calculated all overall effects using inverse variance methods.

Where some primary studies reported an outcome as a dichotomous measure and others used a continuous measure of the same construct (as in the case of attachment security), we performed two separate analyses rather than converting the OR to a SMD. This was because we could not assume that the underlying measure had a normal or logistic distribution, as the nature of the populations in the two relevant studies means that the distribution of attachment patterns is likely to be skewed (teenage mothers in Moran 2005 and families where children had been subjected to maltreatment in Moss 2011).

Where a trial reported two outcomes within a time period covered by the same meta‐analysis, we combined the data from the time point nearest the end of the intervention. Where possible, we tried to combine outcomes measured at similar time points in the follow‐up period.

'Summary of findings' table

We created a 'Summary of findings' table for the following comparison: video feedback versus no intervention or inactive alternative intervention.

We followed the guidelines in the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2017), and included the following six elements in these tables.

A list of all outcomes
A measure of the typical burden of these outcomes
Absolute and relative magnitude of effect
Numbers of participants and studies that address these outcomes
A rating of the overall certainty of evidence for each outcome
Additional comments

Two review authors (LOH, ES) independently assessed the certainty of the evidence, using the following five GRADE considerations (Schünemann 2017).

Limitations in study design and implementation: for RCTs, for example, these included lack of allocation concealment, lack of blinding and large loss to follow‐up.
Indirectness of evidence: for example, if findings were restricted to indirect comparisons between two interventions. RCTs that met the eligibility criteria but that addressed a restricted version of the main review questions in terms of population, intervention, comparator or outcomes are another example of this and would also have been downgraded.
Unexplained heterogeneity or inconsistency of results: we looked for robust explanations for heterogeneity in studies that yielded widely differing estimates of effect.
Imprecision of results: we downgraded the certainty of evidence for those studies that included few participants and few events and thus had wide CIs.
Publication bias: we downgraded the certainty of evidence level if investigators failed to report studies or outcomes on the basis of results.

We downgraded the ratings (from high to very low), depending on the presence of the five factors.

We used GRADEpro GDT to prepare the 'Summary of findings' table, and specifically, to enable us to produce relative effects and absolute risks associated with the interventions (GRADEpro GDT). We used all primary outcomes and one secondary outcome of interest to populate the ‘Summary of findings’ table (primary outcomes: parental sensitivity at postintervention to six months; reflective functioning; attachment security measured by Strange Situation Procedure at postintervention; attachment security measured by Attachment Q‐sort at postintervention; parental stress measured at postintervention and short‐term follow‐up; and parental anxiety at short‐term follow‐up; secondary outcome: child behaviour measured at long‐term follow‐up). We also used Ryan 2016 to guide our judgements.

Subgroup analysis and investigation of heterogeneity

We investigated heterogeneity by conducting moderator analyses for the outcome of 'parental sensitivity'. To perform this analysis, we used a random‐effects meta‐analysis with a Sidik‐Jonkman estimator, which is robust for small numbers of studies and provides improved CI (Veroniki 2019). We considered the following factors, some of which we decided post hoc.

Prespecified factors

Intervention dose: defined by number of video feedback sessions (zero to five versus six to 10 versus more than 10; grouping this factor in this way was a post hoc decision).
Participating carer: all mothers versus all fathers versus mix of mothers and fathers (we made a post hoc decision to include studies with a mix of parental genders along side those who were all fathers or all mothers).
Type of video feedback (VIPP versus non‐VIPP; grouping types of video feedback in this way was a post hoc decision).

Factors specified post hoc

Age of child (children under one year old versus children aged one year or older; using age of child as a factor was a post hoc decision).
Disability status of children (disability versus no disability; using disability status of the child was a post hoc decision).

In the first step, we assessed the moderators individually and reported their overall contribution to the reduction of heterogeneity (Q‐between). To assess whether moderation effects for study characteristics existed, we conducted a moderator analysis in which we included the three prespecified moderators (type of video feedback, duration of video feedback and participating carer) simultaneously, this accounts for potential correlations between moderators. Given the small number of studies, this analysis should be treated with caution, due to the relatively low power. Predicted values are reported alongside regression results. We did not impute missing data in line with the main analyses.

We conducted the moderator analyses in R version 3.6.1 (R 2018), using the metafor‐package 2.1.0 (Viechtbauer 2010); analysis syntax and data are available from the review authors on request.

Sensitivity analysis

We assessed the robustness of findings to decisions made in obtaining them by conducting the following sensitivity analyses (Deeks 2017).

Reanalysis excluding studies at high or unclear risk of bias
Reanalysis using different statistical approaches (comparing the use of a random‐effects model with a fixed‐effect model).

Results

Description of studies

Results of the search

Our initial electronic searches (August to September 2016 and July 2017) identified 6191 records (see Figure 1). We identified an additional 381 records from other sources. After the removal of duplicates, we screened the titles and abstracts of 4368 records. We obtained and scrutinised 81 full‐text reports for eligibility, 47 of which (37 studies) did not meet the inclusion criteria and were excluded from the review with reasons (see Characteristics of excluded studies), and 34 (19 studies) that did and were included in the review.

Figure 1

95 Study flow diagram

Our updated electronic searches (November 2018) identified 2887 records. We identified an additional 211 records through other sources in November 2018, and an additional two records in July 2019. After the removal of duplicates, we screened the titles and abstracts of 1662 records. During the title and abstract screening, we identified six ongoing studies, one of which we excluded, leaving five ongoing studies (Euser 2016; Firk 2015; ISRCTN92360616; NCT03052374; Schoemaker 2018), and one study awaiting classification (Mendelsohn 2008). We reviewed seven full‐text reports and added six reports pertaining to three new studies (Barone 2019; Platje 2018; Seifer 1991) and one report of a study identified during our initial search (Hodes 2017), to the review (see Figure 1).

Included studies

This review includes 22 studies (see Characteristics of included studies tables and Table 3), comprising a total of 41 reports and 1889 randomised parent‐child dyads or family units.

See: Summary of findings for the main comparison Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment

Table 3. Type of video‐feedback intervention

Study	Aim	Content/delivery
Video‐feedback Intervention to promote Positive Parenting (VIPP; Juffer 2008)
Green 2015	To test the effect of a parent‐mediated intervention for children at high risk of autism spectrum disorder	Video Interaction for promoting Positive Parenting (iBASIS‐VIPP), a modification for the autism prodome of the VIPP infancy programme. The intervention consisted of 12 sessions (an additional 6 booster sessions compared with VIPP).The intervention uses video feedback "to help parents understand and adapt to their infants' individual communication style to promote optimal social and communicative development" (quote). The study authors describe that "The therapist uses excerpts of parent‐child interactions in a series of developmentally sequenced home‐sessions focusing on interpreting the infant's behaviour and recognising their intentions; enhancing sensitive responding; emotional attunement and patterns of verbal and non‐verbal interaction." (quote)
Hodes 2017	To test if a video‐feedback intervention to promote positive parenting and sensitive discipline reduces child‐related parental stress in parents with mild learning disabilities in comparison with care as usual	A Video‐feedback Intervention for Positive Parenting and Learning Difficulties (VIPP‐LD) where the original protocol of VIPP‐SD (Juffer 2008) was adapted for mild intellectual disabilities. For VIPP‐LD, in each session, the parent is videoed interacting with their child. The coach and parent review the footage together, drawing attention to instances of sensitive responsiveness and sensitive discipline, and the coach helps the parent look at the child from the child's perspective. The adaptation included shortening of each session, shorter video recordings and more real‐life practice. The study authors describe how "Parents also received a personal scrapbook with skills taken from video recordings and quotes from the parents representing the theme of the session." (quote)
Kalinauskiene 2009	To evaluate the effectiveness of a short‐term, interaction‐focused video‐feedback intervention implemented in families with mothers rated low in maternal responsiveness	A Video‐feedback Intervention to promote Positive Parenting (VIPP). The intervention was applied as per protocol with the main goal "to reinforce mothers' sensitive responsiveness to their infants' signals focusing on different aspects of mother‐infant interactions" (quote). Mothers were also "provided with information on attachment‐related issues by giving them brochures about sensitive parenting." (quote)
Klein Velderman 2006	To explore if a combination of attention to parental sensitivity and parental attachment representations might lead to firmer and more enduring changes in both parenting behaviour and children's attachment security	A Video‐feedback Intervention to promote Positive Parenting (VIPP). VIPP programs consisted of four home visits lasting 1.5 hours each, with 3‐4 weeks in between. Each session was focused around a specific theme. VIPP‐R included additional discussions on parental representations.
Negrão 2014	To test the effectiveness of a video‐feedback intervention to promote positive parenting and sensitive discipline in a sample of poor Portuguese mothers and their 1‐4‐year old children	A sensitive discipline video‐feedback intervention to promote positive parenting (VIPP‐SD). The study authors state that "VIPP‐SD is a short term intervention programme that relies on video‐feedback technique to enhance parental sensitivity and positive discipline strategies. The intervention was applied through standardised protocols of six home visits...The VIPP‐SD working method is divided into three steps: (1) Sessions 1 and 2 main goals are building a relationship with the mother, focusing on child behaviour and emphasizing positive interactions in the video feedback; (2) Sessions 3 and 4 actively work on improving parenting behaviours by showing the mother when her parenting strategies work and to what other situations she could apply these strategies; and (3) Sessions 5 and 6 (booster) aim to review feedback and information from the previous sessions in order to strengthen intervention effectiveness." (quote)
Platje 2018	To evaluate a video‐feedback intervention aimed at improving parent‐child interaction for parents of children with a visual or visual and intellectual disability	A Video‐feedback Intervention to promote Positive Parenting adapted to parents of children with a visual or visual and intellectual disability (VIPP‐V). The study authors state that the intervention was based on VIPP, but "this new intervention [is] applicable for use in families with a young child with a visual or visual‐and‐intellectual disability. Particular attention was devoted to increasing (safe) exploration, joint attention, and parent’s abilities to recognize and understand the signals and emotions of their child" (quote). The intervention consists of 7 home visits (5 primary visits plus 2 booster sessions).
Poslawsky 2015	To evaluate the early intervention programme, video‐feedback intervention to promote positive parenting adapted to autism, with primary caregivers and their child with autism spectrum disorder	VIPP adapted to autism (VIPP‐AUTI). The intervention comprised 5 home visits lasting 60‐90 minutes every 2 weeks. Sessions included: (1) "Attachment and Exploration" (quote); (2) "Speaking for the Child" (quote); (3) "Sensitivity Chain" (quote); (4) "Sharing Emotions" (quote); (5) "Booster session" (quote).
Van Zeijl 2006	To test the video‐feedback intervention to promote positive parenting and sensitive discipline in "a large sample of families screened for their children's relatively high scores on externalizing behaviour." (quote)	The study applied VIPP‐SD, aimed at parental sensitivity and sensitive parental discipline. The first four intervention sessions each had their own themes, (1) "exploration versus attachment" (quote); (2) "centered around speaking for the child" (quote); (3) "the intervener stressed the importance of adequate and prompt responses to the child’s signals" (quote); (4) "the importance of sharing—both positive and negative—emotions (sensitivity) and promoting empathy for the child" (quote); (5 & 6) "aimed at consolidating intervention effects by integrating—in video feed‐back and discussion—all tips and feedback given in the previous sessions" (quote).
Yagmur 2014	"To test the effectiveness of the video feedback intervention to promote positive parenting and sensitive discipline adapted to the specific child‐rearing context of Turkish families (VIPP‐TM) in the Netherlands" (quote), including second‐generation Turkish immigrant families with toddlers at risk for the development of externalising problems	"The VIPP‐TM program is a culturally sensitive adaptation of the VIPP‐SD program for Turkish minority families in the Netherlands, but follows the general procedures of the original program...The VIPP‐SD program is described in a detailed protocol and consists of six home visits. The first four visits each have their own themes regarding sensitivity and discipline, and the last two sessions are booster sessions in which the themes from previous sessions are reviewed once more." (quote)
Video Interaction Guidance (VIG)
Barlow 2016	"To assess the potential of video interaction guidance to increase sensitivity in parents of preterm infants." (quote)	The study authors report that "VIG is a strengths‐based form of video feedback in which parents are invited to jointly observe and reflect on their own successful interactions with their baby...The core aspects of the model involve three home visits comprising (a) video recording the parent‐infant interaction during play or other aspects of care giving, (b) editing of the recording to select micro‐moments of interaction that demonstrate the infant's contact initiatives and the parents attuned response to these signals and (c) joint reviewing of the recordings with the parent." (quote)
Hoffenkamp 2015	To evaluate the effectiveness of hospital‐based video interaction guidance in parents with moderately and very preterm babies	"Video recordings of parent‐infant interactions and the feedback from a VIG professional provide an opportunity for parents to observe, analyse and discuss the infant's behaviour and contact initiatives" (quote). In this study "VIG consisted of three sessions during the first week after birth" (quote), and included "(1) video‐recording parent‐infant interaction; (2) editing the video recordings; (3) reviewing the edited recordings with parents." (quote)
Lam‐Cassettari 2015	To examine "the effect of a family‐focused psychosocial video intervention program on parent‐child communication in the context of childhood hearing loss" (quote)	Parents completed three sessions: "(a) a goal setting session; (b) three filming sessions of parent–child interaction in the family home, and (c) three shared review sessions in which three short video clips (demonstrating attuned responses linked to the family’s goal) were played so families could microanalyze and discuss." (quote)
Video feedback of Infant‐Parent Interaction (VIPI)
Høivik 2015	To investigate "in a heterogenic community sample of families with interactional problems, whether VIPI would be more effective than standard care (TAU) received in the community" (quote)	VIPI involves at least 6 consultation sessions over a maximum period of 3 months focusing on (1) "Initiative of the infants to contact caregivers and initiate pauses in the dyadic exchange" (quote); (2) "Responses of caregivers" (quote); (3) "Following the child" (quote); (4) "Naming" (quote); (5) "Step‐by‐step guidance" (quote); (6) "Directing attention towards social interaction and exploration" (quote). In this study, "families in the VIPI group received eight video feedback sessions, with the last two sessions tailored to meet the individual family needs regarding any of the six topics in the manual" (quote).
Video self‐modelling with feedback
Benzies 2013	To explore if fathers of late, preterm children who received video self‐modelling with feedback intervention would have better father‐child interaction skills when the child was 8 months old than fathers who received information only	Self‐modelling "involves the father's active participation that increases his cognitive awareness of specific behaviours such as infant cues and how to stimulate development" (quote). The intervention involved video recording a father‐infant play interaction and providing positive feedback and suggestions to enhance the interaction and language development.
Video feedback (non‐specified or other)
Bovenschen 2012	To assess "the effectiveness of an attachment‐based short term intervention using video‐feedback" (quote)	Up to 10 sessions of home‐based video feedback
Green 2010	To test a parent‐child communication‐focused intervention in children with core autism	A parent‐mediated communication‐focused intervention: "The intervention consisted of one‐to‐one clinic sessions between therapist and parent with the child present. The aim of the intervention was first to increase parental sensitivity and responsiveness to child communication and reduce mistimed parental responses by working with the parent and using video‐feedback methods to address parent‐child interaction... incremental development of the child's communication was helped by the promotion of a range of strategies such as action routines, familiar repetitive language and pauses...After an initial orientation meeting, families attended biweekly 2 hour clinic sessions for 6 months followed by booster sessions for 6 months (total 18). Between sessions families were also asked to do 30 mins of daily home practice." (quote)
Koniak‐Griffin 1992	To evaluate "the effects of video tape instruction and feedback (video‐therapy) on mothering behaviours" (quote)	The intervention group received two home visits. Participants were "video taped during structured mother‐infant teaching episodes in their homes at 1 and 2 months postpartum" (quote). Participants "reviewed the video tapes with feedback from a professionally trained nurse who emphasised positive aspects of maternal behaviour" (quote)
Moran 2005	To evaluate "the effectiveness of a brief intervention program designed to support adolescent mothers' sensitivity to their infants attachment signals" (quote)	A brief intervention programme (eight home visits) designed to support the mother's sensitivity to her child. The home visits (lasting approximately one hour) were "designed to provide mutually beneficial play interactions and the mother's enjoyment of her infant" (quote). The four goals of the programme included "to affirm parenting strengths already present in the mother...increase the mother's awareness of how her behaviour influenced her child's behaviour...look for ways to augment the mother's awareness of her infant's signals and for ways to establish positive experiences for both the mother and infant" (quote).
Moss 2011	To evaluate the "efficacy of a short‐term attachment‐based intervention for changing risk outcomes for children of maltreating families" (quote)	The intervention consisted of "8 weekly home visits directed at the caregiver–child dyad and focused on improving caregiver sensitivity" (quote). The study authors describe that "All intervention sessions were primarily focused on reinforcing parental sensitive behavior by means of personalized parent–child interaction, video feedback, and discussion of attachment/emotion regulation‐related themes" (quote).
Seifer 1991	To examine the effects of easy‐to‐use interaction coaching techniques on interaction style and developmental status of a population of mothers and their young children with developmental disabilities	"Interaction coaching" (quote; 10‐month programme). "Sessions lasted six minutes and parents were asked to play with their children as they would during a short break at home...After the initial taping session the video record was viewed by the mother and an interaction coach. Suggestions were then provided by the therapist for the mother to employ during interaction with her child...Another 6 minute interaction was then recorded that was reviewed by the intervener and could be used during the next week’s session. The procedure was repeated for a maximum of 6 sessions." (quote)
Stein 2006	To test "whether video‐feedback treatment especially targeting mother‐child interaction would be superior to counselling in improving mother‐child interaction, especially mealtime conflict and infant weight and autonomy" (quote)	"Thirteen 1‐hour treatment sessions were offered in the mothers’ homes beginning when the infants were between 4 and 6 months old and completed by the time the infants were 12 months old. The intervention group received video‐feedback interactional treatment that was a modification of that developed by [Juffer et al]" (quote). Treatment consisted of three stages: "The first concentrated on the infant’s perspective, focusing on his or her signals...The second stage included the mother’s perspective...Third, as treatment progressed, the videotapes were used to help the mother identify and address potential triggers of mealtime conflict" (quote).

IBASIS‐VIPP: Intervention within the British Autsim Study of Infant Siblings ‐ Video‐feedback Interaction to promote Positive Parenting; Mins: Minutes; TAU: Treatment as usual; VIG: Video Interaction Guidance; VIPI: Video‐feedback of Infant‐Parent Interaction; VIPP: Video‐feedback Interaction to promote Positive Parenting; VIPP‐AUTI: Video‐feedback Interaction to promote Positive Parenting ‐ Autism;VIPP‐LD: Video‐feedback Interaction to promote Positive Parenting ‐ Learning Difficulties; VIPP‐R: Video‐feedback Interaction to promote Positive Parenting ‐ Representational level; VIPP‐SD: Video‐feedback Interaction to promote Positive Parenting ‐ Sensitive Discipline; VIPP‐TM: Video‐feedback Interaction to promote Positive Parenting ‐ Turkish Minorities; VIPP‐V: Video‐feedback Interaction to promote Positive Parenting ‐ Visual or visual and intellectual disability.

Location

Seven studies were conducted in the Netherlands (Hodes 2017; Hoffenkamp 2015; Klein Velderman 2006; Platje 2018; Poslawsky 2015; Van Zeijl 2006; Yagmur 2014), five in the UK (Barlow 2016; Green 2010; Green 2015; Lam‐Cassettari 2015; Stein 2006), three in Canada (Benzies 2013; Moran 2005; Moss 2011), and two in the USA (Koniak‐Griffin 1992; Seifer 1991). One study apiece was conducted in Germany (Bovenschen 2012), Italy (Barone 2019), Lithuania (Kalinauskiene 2009), Norway (Høivik 2015), and Portugal (Negrão 2014).

Design

All but two studies were RCTs (Bovenschen 2012; Seifer 1991). Bovenschen 2012 was originally designed as an RCT but some mothers only agreed to take part if they could participate in the intervention, thereby undermining the randomisation. Seifer 1991 was a quasi‐RCT, with participants allocated based on the day of the week they attended a linked treatment programme.

Two of the 22 included RCTs employed a three‐arm design (Benzies 2013; Klein Velderman 2006). Klein Velderman 2006 allocated parents to either a video‐feedback group, a video‐feedback and discussion group, or a control group. In Benzies 2013, one group was allocated two visits with a video‐feedback intervention, the second group was allocated four visits with a video‐feedback intervention and the final group was allocated to a control condition.

The other studies employed a two‐arm design with parents allocated either to a video‐feedback intervention or control group (see Table 3).

Sample size

The number of dyads randomised in each trial ranged from 14 (Lam‐Cassettari 2015), to 237 (Van Zeijl 2006).

Recruitment

Two studies recruited participants from an inpatient hospital setting (Barlow 2016; Hoffenkamp 2015). The other 20 studies recruited participants from a community setting, including primary care, and hospital outpatient clinics.

Participants

The majority of studies (n = 14) randomised only mother‐child dyads. Seven studies randomised male as well as female caregiver and child dyads (Barlow 2016; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Moss 2011; Platje 2018; Poslawsky 2015). Only one study randomised father‐child dyads (Benzies 2013).

The average age of carers, when reported, ranged from 17.16 years (Koniak‐Griffin 1992), to 42.6 years (Barone 2019).

Twelve studies had a mean age of participant children that was under one year at baseline (Barlow 2016; Benzies 2013; Bovenschen 2012; Green 2015; Hoffenkamp 2015; Høivik 2015; Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Moran 2005; Seifer 1991; Stein 2006); the remaining 10 studies had a mean age of participant children that was over one year at baseline (Barone 2019; Green 2010; Hodes 2017; Lam‐Cassettari 2015; Moss 2011; Negrão 2014; Platje 2018; Poslawsky 2015; Van Zeijl 2006; Yagmur 2014). Many studies specified in their inclusion criteria that they were either recruiting babies (children aged under one year) or children (children aged one year or over); however, in some studies, their inclusion criteria included children both under and over one year of age.

Participants were recruited for a range of reasons including: child behaviour problems; parental diagnosis of an eating disorder; adverse family circumstances; parental depression; sensitivity problems; insecure attachment; parental intellectual disability; teenage or single parenthood (or both); migration status; preterm baby; adopted child; deaf children; parents who were being monitored by social services for child maltreatment; parents of children with a visual or visual and intellectual disability; and autistic children or children considered at risk of autism.

It is difficult to summarise the ethnicities of participants as different studies categorised this variable in different ways. Six studies did not report ethnicity (Bovenschen 2012; Hodes 2017; Hoffenkamp 2015; Klein Velderman 2006; Moss 2011; Poslawsky 2015). In 12 studies the majority of participants appeared to be from white European backgrounds (see Characteristics of included studies tables). One study recruited internationally adopted children (Barone 2019), and in one study the majority of participants were from African American or Hispanic backgrounds (Koniak‐Griffin 1992). Two studies used ethnicity as part of their inclusion or exclusion criteria (Negrão 2014; Yagmur 2014).

Type of video‐feedback intervention

Table 3 provides an overview of type of video‐feedback interventions that the included studies evaluated, organised by type. In summary:

10 studies implemented Video‐feedback Intervention to promote Positive Parenting (VIPP) or a variation of VIPP (VIPP with a representational component (VIPP‐R), VIPP and sensitive discipline (VIPP‐SD), VIPP adapted to autism (VIPP‐AUTI), VIPP adapted for Turkish mothers (VIPP‐TM), VIPP‐visual (VIPP‐V), VIPP adapted for fostered or adopted children (VIPP‐FC/A)): (Barone 2019; Green 2015; Hodes 2017; Kalinauskiene 2009; Klein Velderman 2006; Negrão 2014; Platje 2018; Poslawsky 2015; Van Zeijl 2006; Yagmur 2014);
Three studies included Video Interaction Guidance (VIG; Barlow 2016; Hoffenkamp 2015; Lam‐Cassettari 2015);
One study implemented Video‐feedback of Infant‐Parent Interaction (VIPI; Høivik 2015);
One study implemented video self‐modelling with feedback (Benzies 2013); and
Six studies implemented a non‐specified type of video feedback or another type not named above (Bovenschen 2012; Green 2010; Koniak‐Griffin 1992; Moran 2005; Moss 2011; Seifer 1991; Stein 2006).

Treatment intensity

Six studies had between one and five sessions of video feedback (Barlow 2016; Benzies 2013; Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Poslawsky 2015); 12 studies had six to 10 sessions of video feedback (Barone 2019; Bovenschen 2012; Hoffenkamp 2015; Høivik 2015; Lam‐Cassettari 2015; Moran 2005; Moss 2011; Negrão 2014; Platje 2018; Seifer 1991; Van Zeijl 2006; Yagmur 2014); and four studies offered more than 10 sessions of video feedback (Green 2010; Green 2015; Hodes 2017; Stein 2006).

Monitoring of treatment fidelity

Two studies reported a quantitative measure of treatment fidelity (Green 2010; Green 2015). Ten studies reported having a process in place to monitor treatment fidelity, although they did not report a quantitative measure of treatment fidelity (Hodes 2017; Hoffenkamp 2015; Høivik 2015; Moran 2005; Moss 2011; Platje 2018; Poslawsky 2015; Stein 2006; Van Zeijl 2006; Yagmur 2014). The 10 remaining studies did not report any monitoring of treatment fidelity.

Comparisons

Eleven studies used either usual care (such as routine visits from community health staff or play support programmes) or no additional intervention for their control group (Barlow 2016; Bovenschen 2012; Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Klein Velderman 2006; Lam‐Cassettari 2015; Moss 2011; Platje 2018). The other 11 studies used some sort of inactive alternative treatment, such as telephone calls or videoing the parent‐child dyads, without providing any feedback.

Outcomes and outcome measures

Parental sensitivity

All studies measured parental sensitivity or used a measure that could act as a proxy.

Seven studies (Barone 2019; Høivik 2015; Klein Velderman 2006; Lam‐Cassettari 2015; Negrão 2014; Poslawsky 2015; Yagmur 2014), used the Emotional Availability Scale (Biringen 2000b; Biringen 2008).
Three studies (Bovenschen 2012, Kalinauskiene 2009; Klein Velderman 2006), used the Ainsworth Rating Scale (Ainsworth 1974; Ainsworth 1978), and one (Stein 2006), used an adapted version of this scale.
Two studies (Moran 2005; Moss 2011), used the Maternal Behaviour Q‐sort (Pederson 1999; Pederson 1995).
Six other studies used the following scales:
1. Barlow 2016 used the CARE‐Index (Crittenden 2001);
2. Benzies 2013 used the Parent Child Interaction Teach Scale (Sumner 1994);
3. Green 2015 used the Manchester Assessment of Caregiver‐Infant Interaction (Wan 2017);
4. Hoffenkamp 2015 used an adapted measure based on a coding scale from NICHD Early Child Care Research Network 2005;
5. Koniak‐Griffin 1992 used the Nursing Child Assessment Teaching Scale (Barnard 1978);
6. Platje 2018 used an adapted version of the National Institute of Child Health and Human Development Scales (Egeland and Heister 1993); and
7. Van Zeijl 2006 used measures for parental sensitivity that were taken from Egeland 1990.

Three studies relied on proxy measures (Green 2010; Hodes 2017; Seifer 1991).

Green 2010 used the proportion of parental communications with the child that were synchronous based on observation.
Hodes 2017 assessed harmonious parent‐child interaction (measured using the three‐bag procedure; NIHCD Early Child Care Research Network 2003).
Seifer 1991 used an observer to measure maternal responsive behaviour, but the report does not specify the type of scale used to code observations.

Fourteen studies measured parental sensitivity immediately postintervention (Barone 2019; Bovenschen 2012; Green 2015; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Kalinauskiene 2009; Koniak‐Griffin 1992; Lam‐Cassettari 2015; Moran 2005; Moss 2011 Platje 2018; Poslawsky 2015; Seifer 1991), and 17 measured it in the short term (Barlow 2016; Barone 2019; Benzies 2013; Bovenschen 2012; Green 2010; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Klein Velderman 2006; Koniak‐Griffin 1992; Lam‐Cassettari 2015; Negrão 2014; Platje 2018; Poslawsky 2015; Stein 2006; Van Zeijl 2006; Yagmur 2014). None of the studies measured it in the medium term, and just three studies measured it in the long term (Kalinauskiene 2009; Klein Velderman 2006; Moss 2011).

Parental reflective functioning

No study measured this outcome.

Attachment security

Four studies measured child attachment security (Kalinauskiene 2009; Klein Velderman 2006; Moran 2005; Moss 2011), and all but one study, Klein Velderman 2006, measured this outcome postintervention. Klein Velderman 2006 assessed attachment security in both the short and long term.

Three studies, Klein Velderman 2006, Moran 2005 and Moss 2011, used the Strange Situation Procedure (Ainsworth 1978) to measure attachment security, and two studies, Kalinauskiene 2009 and Klein Velderman 2006, used the Attachment Q‐sort (Waters 1985; Waters 1987).

Adverse effects

Parental stress

Eight studies measured parental stress (Barlow 2016; Benzies 2013; Hodes 2017; Kalinauskiene 2009; Klein Velderman 2006; Negrão 2014; Platje 2018; Poslawsky 2015). Four studies measured the outcome postintervention (Hodes 2017; Kalinauskiene 2009; Platje 2018; Poslawsky 2015). Six studies measured this outcome in the short term (Barlow 2016; Benzies 2013; Klein Velderman 2006; Negrão 2014; Platje 2018; Poslawsky 2015). One study measured this outcome at an unspecified follow‐up time point (Hodes 2017).

Two studies, Barlow 2016 and Benzies 2013, used a version of the Parenting Stress Index (full or short form; Abidin 1995; Terry 1991). Three studies, Kalinauskiene 2009, Negrão 2014 and Poslawsky 2015, used the Parenting Daily Hassles or Daily Hassles Questionnaire (Crnic and Greenberg 1990; Kanner 1981). One study, Klein Velderman 2006, used the Support and Stress Questionnaire (Van den Boom 1988). Two studies, Platje 2018 and Hodes 2017, used the Nijmeegse Ouderlijke Stress Index ‐ Dutch version of the Parenting Stress Index (Abidin 1983; De Brock 1992).

Parental anxiety

Only two studies measured parental anxiety. Barlow 2016 measured parental anxiety in the short term using the Hospital Anxiety and Depresssion Scale (Zigmond 1983); and Hoffenkamp 2015 measured parental anxiety in the short and medium term using the State‐Trait Anxiety Inventory (Spielberger 1983).

Child mental health

One study, Green 2010, measured child mental health in the long term, using the Development and Well‐being Assessment (Goodman 2011).

Child physical and socioemotional development

Seifer 1991 measured child psychomotor development at postintervention using the Uzgiris and Hunt Ordinal Scales of Development (Uzgiris 1975); all seven subscales were measured individually.

Five studies measured aspects of children's socioemotional development (Green 2010; Green 2015; Høivik 2015; Poslawsky 2015; Seifer 1991). Høivik 2015 measured this immediately postintervention and in the medium term using the Ages and Stages Questionnaire (Squires 2002). Poslawsky 2015 measured it at both postintervention and in the short term using the Early Social Communication Scales (Mundy 2003). Green 2010 assessed this in the short term using the Vineland Adaptive Behaviour Scales (VABS; Sparrow 2005), and in the long term using the Strengths and Difficulties Questionnaire (Goodman 1997). Green 2015 used the VABS to measure this outcome in the short and long term (Sparrow 2005). Seifer 1991 measured child mental development using the Bayley Scales of Infant Development at postintervention (Bayley 1969).

Child behaviour

Five studies measured child behaviour at different time points, using various versions of the Child Behaviour Checklist (CBCL; Achenbach 1992; Achenbach 2000). Moss 2011 and Barone 2019 measured children's behaviour postintervention; and Van Zeijl 2006 and Barone 2019 did so in the short term. Two studies measured it in the long term (Kalinauskiene 2009; Klein Velderman 2006).

Three studies reported externalising behaviour, which is a domain of child behaviour (Barone 2019; Moss 2011; Van Zeijl 2006), using the CBCL (Achenbach 2000). Two studies reported it at postintervention (Barone 2019; Moss 2011), and two reported it at short‐term follow‐up (Barone 2019; Van Zeijl 2006). No study measured it at long‐term follow‐up.

Costs

None of the studies reported data on costs.

Funding sources

These are listed in the Characteristics of included studies tables. Almost all studies reported some sort of external funding, from a charitable organisation (n = 7) and/or public body (n = 18). No studies reported commercial funding.

Excluded studies

We formally excluded 37 completed studies, consisting of 47 reports, and one ongoing study, details of which can be found in the Characteristics of excluded studies tables.

We excluded completed studies for the following reasons: intervention had no video‐feedback component (4 studies); intervention contained multiple sessions of non‐video‐feedback intervention activities (14 studies); used video feedback as part of a multicomponent intervention (6 studies); study did not measure parental sensitivity, child attachment or reflective functioning outcomes (4 studies); study was not an RCT or quasi RCT (4 studies); caregivers did not match this review's inclusion criteria (3 studies); study was a comparison between two active interventions rather than an intervention and inactive alternative intervention (2 studies).

We excluded one ongoing study because it does not measure parental sensitivity, child attachment or reflective functioning outcomes.

Ongoing studies

We identified five ongoing studies, described in further detail here: Characteristics of ongoing studies. All five studies are RCTs. Three of the studies include parent‐child dyads (Firk 2015; ISRCTN92360616; NCT03052374), one study includes parents and twins (Euser 2016), and one includes foster parents and foster children (Schoemaker 2018). Two studies are being conducted in the Netherlands (Euser 2016; Schoemaker 2018), one study in Germany (Firk 2015), one in Ireland (ISRCTN92360616), and one in Canada (NCT03397719).

Three of the studies include a component of video feedback (Firk 2015; ISRCTN92360616; NCT03052374), and two include an adaptation of VIPP (Positve Parenting and Sensitive Discipline in twin families (VIPP‐twin; Euser 2016); Positive Parenting for Foster Care (VIPP‐FC; Schoemaker 2018)). Three studies used standard care as their control intervention (Firk 2015; ISRCTN92360616; NCT03052374), and two studies used phone calls as their control intervention (Euser 2016; Schoemaker 2018).

Of these studies, two were funded through public sector funding sources (Euser 2016; Firk 2015); two were funded through charitable sources (ISRCTN92360616; Schoemaker 2018), and one did not declare a source of funding (NCT03052374).

Studies awaiting classification

There was one report that we could not obtain in full (Mendelsohn 2008), despite a request to the first author (Smith 2018i [pers comm]). This is listed under Characteristics of studies awaiting classification.

Risk of bias in included studies

We present the 'Risk of bias' tables for each included study beneath the Characteristics of included studies tables. Figure 2 summarises the 'Risk of bias' assessments across all outcomes, and Figure 3 summarises these assessments across all included studies.

Figure 2

'Risk of bias' graph: review authors' judgements about each 'Risk of bias' item presented as percentages across all included studies

Figure 3

'Risk of bias' summary: review authors' judgements about each 'Risk of bias' item for each included study

Allocation

Random sequence generation

We assessed the risk of selection bias from randomisation to be low in 16 studies (Barlow 2016; Barone 2019; Benzies 2013; Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Lam‐Cassettari 2015; Moss 2011; Negrão 2014; Platje 2018; Poslawsky 2015; Stein 2006; Van Zeijl 2006; Yagmur 2014). We assessed two studies as having a high risk of selection bias from poor randomisation (Bovenschen 2012; Seifer 1991); for instance due to randomisation based on the day of the week. Four studies stated that participants were randomised, but did not state how this was carried out (Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Moran 2005), so we assessed them as at unclear risk of bias.

Allocation concealment

We assessed the risk of selection bias to be low in six included studies across allocation concealment (Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Poslawsky 2015; Stein 2006). We assessed 13 studies as having an unclear risk of allocation bias, as they did not give sufficient information on how allocation took place (Barlow 2016; Barone 2019; Benzies 2013; Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Lam‐Cassettari 2015; Moran 2005; Moss 2011 Negrão 2014; Platje 2018; Van Zeijl 2006; Yagmur 2014). We rated three studies at high risk of selection bias: Bovenschen 2012, as the study authors stated that some participants only agreed to take part if they were allocated to a specific arm of the study; Høivik 2015 due to the use of consecutive randomisation; and Seifer 1991 as allocation was based on presentation on a certain day of the week.

Blinding

Given the nature of the intervention, we judged it impossible to truly blind any of the participants, so we rated all studies at high risk of performance bias. Two studies did attempt to address this issue by giving participants limited information about the purpose of the study (Benzies 2013; Moran 2005); however, in our judgement, this was not sufficient to blind participants.

We judged detection bias relating to outcomes that were relevant to this review only. We rated each outcome separately. The results are presented below.

Primary outcomes

Parental sensitivity

All 22 included studies measured this outcome or proxy domain. All except two studies, Moran 2005 and Van Zeijl 2006, were able to adequately blind the outcome assessor(s), so we rated them at low risk of detection bias. As Moran 2005 and Van Zeijl 2006 did not describe the blinding in sufficient detail, we rated them as having an unclear risk of detection bias.

Parental reflective functioning

None of the included studies measured this outcome.

Attachment security

Four studies measured this outcome (Kalinauskiene 2009; Klein Velderman 2006; Moran 2005; Moss 2011). We rated one study, Moran 2005 at unclear risk of detection bias, and the other three studies at low risk of detection bias, as they blinded assessors.

Adverse effects

We examined parental stress and parental anxiety. Eight studies reported parental stress (Barlow 2016; Benzies 2013; Hodes 2017; Kalinauskiene 2009; Klein Velderman 2006; Negrão 2014; Platje 2018; Poslawsky 2015). All of these studies used self‐report scales, so we rated all at high risk of detection bias.

Two studies reported parental anxiety (Barlow 2016; Hoffenkamp 2015). Again, both of these studies used self‐report scales, so we rated them at high risk of detection bias.

Secondary outcomes

Child mental health

A single study measured child mental health (Green 2010). They used the Development and Well‐Being Assessment (DAWBA), which uses a parental assessment of their children. Consequently, we rated it at high risk of detection bias.

Child physical and socioemotional development

Four studies measured children's socioemotional development (Green 2010; Høivik 2015; Poslawsky 2015; Seifer 1991). Green 2010 and Høivik 2015 used scales based on parental assessments of their children, so we rated these studies at high risk of detection bias. Poslawsky 2015 and Seifer 1991 used blinded raters so we rated these studies at low risk of detection bias.

Child behaviour

Five studies measured child behaviour (Barone 2019; Kalinauskiene 2009; Klein Velderman 2006; Moss 2011; Van Zeijl 2006). All of these studies used scales based on parental ratings of their child's behaviour, so we rated all as being at high risk of detection bias.

Incomplete outcome data

Of the 22 included studies, we assessed 14 as being at low risk of attrition bias, due to either appropriate methods of imputation by the study authors, very low or no attrition, or attrition that was balanced across all arms of the study (Barlow 2016; Barone 2019; Benzies 2013; Green 2010; Green 2015; Hoffenkamp 2015; Kalinauskiene 2009; Klein Velderman 2006; Lam‐Cassettari 2015; Moran 2005; Negrão 2014; Poslawsky 2015; Stein 2006; Van Zeijl 2006); four as unclear risk of attrition bias because there was unclear reporting of number of, or reason for, dropouts (Bovenschen 2012; Hodes 2017; Koniak‐Griffin 1992; Seifer 1991); and four as high risk of attrition bias due to unequal attrition across arms that could have been for reasons related to the intervention (Høivik 2015; Moss 2011; Platje 2018; Yagmur 2014).

Selective reporting

The majority of studies did not have protocols, making it difficult to judge whether there was reporting bias. We judged 16 studies as having an unclear risk of reporting bias because they appeared to report all outcomes in their Methods section but did not have a protocol available (Barlow 2016; Barone 2019; Benzies 2013; Bovenschen 2012; Høivik 2015; Kalinauskiene 2009; Klein Velderman 2006; Koniak‐Griffin 1992; Lam‐Cassettari 2015; Moran 2005; Moss 2011; Negrão 2014; Poslawsky 2015; Seifer 1991; Van Zeijl 2006; Yagmur 2014). We rated six studies at low risk of reporting bias because they reported all prespecified outcomes from their published protocols (Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Platje 2018; Stein 2006).

For parental sensitivity, we compared published results with unpublished results, to test for publication bias in the three studies where this was possible (Barone 2019; Hoffenkamp 2015; Klein Velderman 2006). There was no evidence of a difference between the two groups of studies.

For parental stress, we compared published results with unpublished results, to test for publication bias in two studies (Klein Velderman 2006; Negrão 2014). We found no difference; neither group showed evidence of a difference between intervention and control groups.

Other potential sources of bias

We rated Moran 2005 at high risk of other bias as they did not report the maternal sensitivity outcome data completely (data were missing SDs or SEs).

Effects of interventions

We summarise the results of our meta‐analyses below. We also report the results from single studies that we did not combine in a meta‐analysis because: not enough studies reported that outcome; data were missing and we were unable to obtain them from the study authors; the outcome was measured at a different time point to other studies reporting that outcome, or the study measured the outcome at multiple similar time points, meaning that we selected a single time point for the meta‐analysis. In addition, we present the results of a moderator analysis for the outcome parental sensitivity; we decided post hoc to undertake this analysis (see Differences between protocol and review).

We have organised results for the main comparison under headings corresponding to the primary and secondary outcomes outlined in the Types of outcome measures section.

Numbers given are the total number of participants randomised. Where it has been possible to calculate an effect size, we have reported this with 95% CI. Where the calculated effect size had a P value less than 0.05, we have stated whether or not the result favours the intervention group.

summary of findings Table for the main comparison summarises the main results of our meta‐analyses.

Video feedback versus no intervention or inactive alternative intervention

Primary outcomes

Parental sensitivity

Using a random‐effects model, we conducted a meta‐analysis of data from 20 studies (1757 parent‐child dyads) that measured the effects of video‐feedback on parental sensitivity from postintervention to six months' follow‐up (Barlow 2016; Barone 2019; Benzies 2013; Bovenschen 2012; Green 2010; Green 2015; Hodes 2017; Hoffenkamp 2015; Høivik 2015; Kalinauskiene 2009; Klein Velderman 2006; Lam‐Cassettari 2015; Moss 2011; Negrão 2014; Platje 2018; Poslawsky 2015; Seifer 1991; Stein 2006; Van Zeijl 2006; Yagmur 2014). We have presented data for mothers and fathers from Hoffenkamp 2015 separately as this is how the data were provided to us by the study authors. We have presented data for Benzies 2013 and Klein Velderman 2006 with two treatment groups, as these were three‐armed studies where both treatment groups in the study met our inclusion criteria. Details of how we managed unit of analysis issues are described in Unit of analysis issues.

The results suggest evidence favouring video feedback compared with the control group (SMD 0.34, 95% CI 0.20 to 0.49, Analysis 1.1). There was evidence of moderate heterogeneity, meaning that the observed variation is likely to be due to statistical heterogeneity (Tau² = 0.07; Chi² = 49.21, df = 22 (P = 0.0008); I² = 55%). The GRADE certainty rating for this meta‐analysis was moderate; we downgraded due to inconsistency (moderate heterogeneity that was not explained by the subgroup analyses).

We drew a funnel plot (estimating differences in treatment effects against their SE) for the outcome 'parental sensitivity' as this was the only outcome with 10 or more studies that provided data. Figure 4 shows no major asymmetry for this comparison when all studies were included. We ran Egger's regression test for assessing funnel plot asymmetry; there was no evidence for funnel plot asymmetry (P = 0.281). However, when we removed Green 2010, Egger's regression test provided evidence for funnel plot asymmetry (P value = 0.022). The appearance of the funnel plot suggests that the asymmetry might be due to small study effects.

Figure 4

Funnel plot of comparison: 1. Primary outcomes, outcome: 1.1 parental sensitivity (postintervention ‐ 6 months)

Single study results (follow‐up only)

It should be noted that we included postintervention data from all of the following studies in the meta‐analysis for parental sensitivity, except Koniak‐Griffin 1992 and Moran 2005. The following data represent results for later follow‐up time points from these studies, and the single study results are all for later time periods.

Barone 2019 measured maternal sensitivity at six months' follow‐up. The six‐month results were not reported in the study, but the study authors provided us with unpublished means, number of participants (n) and SDs (intervention group: mean = 25.88, SD = 2.8, n = 42; control group: mean = 22.13, SD = 4.13, n = 37). We used these to calculate an SMD of 1.07 (95% CI 0.59 to 1.54), suggesting evidence of a difference between groups.
Bovenschen 2012 reported no evidence of an effect of the intervention at three months' follow‐up. The study authors provided the following unpublished data to us: intervention group: mean = 2.91, SD = 1.89, n = 17; control group: mean = 2.36, SD = 1.36, n = 19). A P value is not reported for this comparison; we used the data reported to calculate an SMD of 0.33 (95% CI −0.33 to 0.99), suggesting no evidence of a difference between groups.
Green 2010 reported the impact on maternal sensitivity 5.75 years after the end of the trial. At 5.75 years there was not strong evidence of a difference between the groups (video‐feedback group: mean = 44.4%, SD = 16.1%, n = 59; comparator group: mean = 43.1%, SD = 15.7%, n = 62; log OR of parent synchrony in video‐feedback group versus comparator group: 0.02 (bootstrap 95% CI −0.30 to 0.36)).
Hodes 2017 reported the impact on maternal sensitivity at three months' follow‐up. At three months' follow‐up there was not strong evidence of a difference between the groups (video‐feedback group: mean = 4.80, SD = 0.63, n = 43; comparator group: mean = 4.84, SD = 0.71, n = 42; repeated measures analysis of variance (ANOVA): F (2, 166) = 0.49, P = 0.61).
Hoffenkamp 2015 reported medium‐term (six months) outcomes for fathers and mothers. At six months' follow‐up there was no evidence of a difference between the two groups (mothers: MD = 0.29, SD = 0.22, P = 0.19; fathers: MD = 0.12, SD = 0.23, P = 0.60). We did not include six‐month follow‐up outcomes in the meta‐analysis as this would have meant two time points from a single trial in the same meta‐analysis.
Koniak‐Griffin 1992 measured maternal sensitivity as part of the NCATS (Nursing Child Assessment Teaching Scale) assessment. We were unable to obtain a breakdown of the data from the study authors, and so could not include them in the meta‐analysis.
Lam‐Cassettari 2015 reported results at three months' follow‐up, but by this time participants in the waiting‐list control group had received the intervention, so the comparison does not fit with the review question.
Moran 2005 measured maternal sensitivity at postintervention and at 12 months' follow‐up and reported that, "none of the t‐tests comparing the Intervention and Comparison group means at each age were significant". The data were not reported and the study author was unable to provide them when requested.
Platje 2018 reported parental sensitivity at three months' follow‐up. The study authors found no strong evidence of a difference between groups with regards to parental sensitivity at three months postintervention (video‐feedback group: mean = 16.39, SD = 1.96, n = 37; comparator group: mean = 16.42, SD = 2.33, n = 40). Repeated measures ANOVA showed no evidence of an interaction between time and condition (Time × Condition interaction F (1, 75) = 0.13, P = 0.715).

Parental reflective functioning

None of our included studies measured or reported data on this outcome.

Attachment security

As described in Measures of treatment effect, we combined data from studies that measured OR separately to those that measured means.

Two studies (Moran 2005; Moss 2011; 166 parent‐child dyads) measured this outcome using the Strange Situation Procedure (Ainsworth 1974), at postintervention. The pooled analysis of these studies using OR under a random‐effects assumption resulted in evidence favouring the intervention (OR 3.04, 95% CI 1.39 to 6.67, Analysis 1.2). We did not assess heterogeneity due to the small number of studies included in this meta‐analysis. We rated the certainty of this evidence as very low using GRADE, due to risk of bias (we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias), imprecision (low number of participants, leading to wide CI) and publication bias (few studies in this review reported this outcome).

Two other studies (Kalinauskiene 2009; Klein Velderman 2006; 131 parent‐child dyads) measured this outcome using the Attachment Q‐sort, with scores ranging from +1.00 for the perfectly secure child to −1.00 for the most insecure child (Waters 1985; Waters 1987). The time points of the data combined were quite different (one study reported data at postintervention and the other study at 12 months postintervention). The pooled analysis of these studies using SMD under a random‐effects assumption found no evidence of a difference between intervention and control groups (SMD 0.02, 95% CI −0.33 to 0.38, Analysis 1.3). We did not assess heterogeneity due to the small number of studies included in this meta‐analysis. We rated the certainty of this evidence as very low using GRADE: we downgraded one level for risk of bias (we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias); one level due to imprecision (low number of participants, leading to wide CI); and one level due to publication bias (few studies in this review report this outcome).

Single study results

One study, Klein Velderman 2006, did not report data in a way that we could use in the meta‐analysis. The study found that there was no evidence that attachment security was different between the intervention and control group in the short term (VIPP group compared to control group: d = 0.33, P = 0.11 (one‐tailed), n = 55; VIPP‐R group compared to control group: d = 0.12, P = 0.33 (one‐tailed), n = 53).

Adverse effects

Parental stress

We pooled data from eight studies (537 parent‐child dyads) reporting data at postintervention or short‐term follow‐up (Barlow 2016; Benzies 2013; Hodes 2017; Kalinauskiene 2009; Klein Velderman 2006; Negrão 2014; Platje 2018; Poslawsky 2015). A random‐effects meta‐analysis did not show any strong evidence of a difference between intervention and control groups (SMD −0.09, 95% CI −0.26 to 0.09, Analysis 1.4). Heterogeneity was low (Tau² = 0.00; Chi² = 6.36, df = 8 (P = 0.61); I² = 0%). We rated the certainty of this evidence as low using GRADE. We downgraded one level for risk of bias (we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias) and one level due to imprecision (low number of participants, leading to wide CI).

Single study results: three studies reported the impact of video feedback on parental stress in ways that we could not include in the meta‐analysis (Hodes 2017; Platje 2018; Poslawsky 2015).

Hodes 2017 measured the impact of video‐feedback at an unspecified follow‐up time point, reporting no evidence of a difference between intervention and control groups on a repeated measures multivariate ANOVA (MANOVA) (video‐feedback group: mean = 70.4, SD = 24.87, n= 43; comparator group: mean = 72.14, SD = 24.75, n = 42; F (1.57, 130, 6) = 4.39, P = 0.02). We did not include these data in the meta‐analysis as we used data provided by the same study at postintervention.
Platje 2018 reported the impact on parents' stress levels at postintervention and six months' follow‐up. The postintervention data are included in the meta‐analysis. The study found no evidence of a difference between intervention and control groups at six months' follow‐up using a repeated measures ANOVA (video‐feedback group: mean = 2.36, SD = 0.94, n = 37; control group: mean = 2.58, SD = 0.90, n = 40; F (1, 75) = 3.52, P = 0.07, η2 = 0.05).
Poslawsky 2015 measured parental stress at three months, but did not report it and were not able to provide the data when requested.

Parental anxiety

We combined data from two studies (311 parent‐child dyads), measured at short‐term follow‐up, using a random‐effects model (Barlow 2016; Hoffenkamp 2015). Data for mothers and fathers in Hoffenkamp 2015 are included separately, as this is how the data were provided to us. The meta‐analysis found no strong evidence of a difference between the intervention group and the control group (SMD −0.28, 95% −0.87 to 0.31, Analysis 1.5). We did not assess heterogeneity due to the small number of studies included in this meta‐analysis. We rated the evidence as very low certainty using GRADE. We downgraded one level due to imprecision (low number of participants, leading to wide CI), one level due to publication bias (few studies in this review reported this outcome) and one level due to inconsistency (high heterogeneity).

Secondary outcomes

Child mental health

Only one study, Green 2010, measured the rate of child mental illness at long‐term follow‐up, and found no strong evidence of a difference between intervention and control group (log odds of depression in video‐feedback group (n = 50) versus comparator group (n = 44): 0.07 (bootstrap CI −0.85 to 1.03); log odds of conduct/oppositional disorder in video‐feedback group (n = 50) versus comparator group (n = 44): −0.13 (bootstrap CI −1.08 to 0.72); log odds of hyperkinesis in video‐feedback group (n = 50) versus comparator group (n = 44): 0.11 (bootstrap CI −0.70 to 0.93); log odds of anxiety/OCD in video‐feedback group (n = 50) versus comparator group (n = 46): 0.51 (bootstrap CI −0.33 to 1.51).

Child physical and socioemotional development

Four studies measured elements of child socioemotional development in ways that were too clinically different for meta‐analysis to be appropriate (Green 2010; Høivik 2015; Poslawsky 2015; Seifer 1991).

Green 2010 found no strong evidence of a difference between intervention and control groups with regards to prosocial behaviour (log odds of prosocial behavior in video‐feedback group (n = 59) versus control group (n = 62): 0.73 (bootstrap 95% CI −0.08 to 1.64) or peer problems (log odds of peer problems in video‐feedback group (n = 59) versus control group (n = 61): 0.64 (bootstrap 95% CI −0.21 to 1.62)).
Høivik 2015 reported no strong evidence of a difference postintervention between intervention and control groups with regards to their score on the socioemotional element of the Ages and Stages Questionnaire (video‐feedback group: mean = 26.21, SD = 19.61, n = 37; comparator group: mean = 25.74, SD = 17.02, n = 27; P = 0.17). At six months postintervention there was evidence that the intervention group had fewer concerns regarding their child's socioemotional development than the control group (video‐feedback group: mean = 20.44, SD = 13.45, n = 22; comparator group: mean = 25, SD = 16.53, n = 27; P = 0.02).
Poslawsky 2015, after controlling for school attendance, reported evidence that the video‐feedback group (n = 38) had better scores than the comparator group (n = 34) on measures of initiating joint attention (f = 2.35, df = 8, P = 0.03, ƞ² = i) but not reciprocating joint attention.
Seifer 1991 measured child mental development (measured by Bayley Scales of Infant Development) and child psychomotor development (measured by the Uzgiris and Hunt Ordinal Scales of Development ‐ the study authors report measuring all seven subscales individually) at postintervention. However, they did not report these data did not respond to our request for the data (Smith 2019b [pers comm]).

Child behaviour

We pooled data from two studies (119 parent‐child dyads) looking at child behaviour in the long term (Kalinauskiene 2009; Klein Velderman 2006). A random‐effects meta‐analysis found no strong evidence of a difference between intervention and control groups (SMD 0.04, 95% CI −0.33 to 0.42, Analysis 1.6). We did not assess heterogeneity due to the small number of studies included in this meta‐analysis. The GRADE certainty rating was very low: we downgraded one level for risk of bias (we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias); one level due to imprecision (low number of participants, leading to wide CI) and one level due to publication bias (few studies in this review reported this outcome).

Single study results

Three studies measured aspects of child behaviour that we could not include in a meta‐analysis: one study reported the data at a different time point to other studies reporting this outcome (Barone 2019); and the other studies reported elements of child behaviour at two very different time points (Moss 2011; Van Zeijl 2006).

Barone 2019 measured child behaviour at postintervention and six months' follow‐up. They reported results for externalising behaviour in their published report. The study authors reported that there was no evidence of an effect of the intervention at either time point for any of the outcomes measured, although they did not report a P value. The data reported at postintervention were as follows: video‐feedback group: mean = 16.6, SD = 9.5, n = 44; control group: mean = 14.2, SD = 10.4, n = 39. We used these data to calculate an SMD of 0.24 (95% CI −0.19 to 0.67). The data they reported at six months' follow‐up were as follows: video‐feedback group: mean = 16.1, SD = 10.9, n = 44; control group: mean = 12.7, SD = 10.6, n = 39. We used these data to calculate an SMD of 0.31 (95% CI −0.12 to 0.75). There is no evidence of an effect at either time point.
Moss 2011 reported no strong evidence of a difference between intervention and control groups at postintervention for internalising behaviour (video‐feedback group: mean = 54.43, SD = 7.44, n = 35; comparator group: mean = 55.56, SD =11.45, n = 32) or externalising behaviour (video‐feedback group: mean = 57.85, SD = 9.84, n = 35; comparator group: mean = 57.54, SD = 12.61, n = 32). P values were not reported by the review authors. We used the data reported to calculate an SMD for internalising behaviour of −0.12 (95% CI −0.60 to 0.36), showing no evidence of a difference between groups. For externalising behaviour, we calculated an SMD of 0.03 (95% CI −0.45 to 0.51), again showing no evidence of a difference between groups.
Van Zeijl 2006 reported no strong evidence of a difference between intervention and control groups for externalising behaviour at long‐term follow‐up (video‐feedback group: mean = 21.55, SD = 9.08, n = 83; comparator group: mean = 21.36, SD 8.62, n = 74). P values were not reported. We used the data reported to calculate an SMD of 0.02 (95% CI −0.29 to 0.33), demonstrating no evidence of a difference between groups.

Moderator analysis for parental sensitivity

Moderator analysis

Appendix 3 reports the overall effects by individual moderator. Three studies contained in this moderator analysis have two separate intervention groups (Benzies 2013; Hoffenkamp 2015; Klein Velderman 2006), meaning that for this part of the analysis, k (number of studies) = 23 studies, rather than 20 studies. Adding these as individual studies potentially biases the test statistic, as these intervention groups are not statistically independent. The usual solution to this problem is to conduct a multilevel meta‐analysis; however, the small number of related studies makes this unviable.

Notably, all subgroups except 'more than 10 sessions of video feedback' and studies with ‘only fathers’ and 'both parents' show evidence of an overall treatment effect, measured as SMDs (d). Substantively, the disability subgroup had the largest effect size (d = 0.49**, 95% CI 0.16 to 0.82). This suggests that some moderator effects may exist for some study characteristics when considered individually. For all moderators, heterogeneity is I² greater than 50%, with evidence of residual heterogeneity (Q_E).

Figure 5 reports the results from the meta‐regression with all three prespecified moderators for k = 23 studies. There is no evidence that jointly the type of intervention, intervention duration, or gender of the participating carer reduce heterogeneity (Q_between (F(df1_moderators = 7, df2_studies = 17) = 1.008, P = 0.4429; R²< 0.01%)), and substantial between‐study heterogeneity still exists (Q_E (df = 17) = 39.77, P = 0.014; though I² = 55.6%, see Appendix 4). In addition, none of the three moderators in the meta‐regression are statistically significant (at α (alpha) = 0.05). Parent gender (both parents versus only mothers or only fathers) potentially has a statistically significant negative moderation effect, though only at α = 0.1. This suggests that, when we consider the three prespecified moderators simultaneously rather than as individual subgroups, no moderation effect exists for any specific study characteristics. In other words, we are not able to say that any particular characteristic accounts for the between‐study heterogeneity when controlling for other characteristics. Figure 5 reports the predicted study effect sizes controlling for the moderator variables. As the grey‐shaded polygons indicate, no consistent and strong moderation can be observed.

Figure 5

Observed versus predicted intervention effects following moderator analysis

Sensitivity analyses

Reanalysis excluding studies at high or unclear risk of bias

Only two meta‐analyses included data from more than two studies (Analysis 2.1; Analysis 1.4). For these analyses, we explored the effects of excluding studies at a high risk of bias.

Parental sensitivity

For Analysis 2.1, we first considered the effect of excluding four studies classed at high risk of attrition bias (Høivik 2015; Moss 2011; Platje 2018; Yagmur 2014). This had no effect on the results, which continued to show evidence of a difference between groups (16 studies, 1414 dyads; SMD 0.35, 95% CI 0.17 to 0.53). When we removed the two studies at high risk of selection bias (Bovenschen 2012; Seifer 1991), the analysis continued to show evidence of a difference between groups (14 studies, 1338 dyads; SMD 0.32, 95% CI 0.13 to 0.51). When we further removed the two remaining studies at unclear risk of selection bias (Kalinauskiene 2009; Klein Velderman 2006), the analysis still also showed evidence of a difference between groups (12 studies, 1203 dyads; SMD 0.27, 95% CI 0.06 to 0.48).

Adverse effects: parental stress

For Analysis 1.4, we considered the effect of excluding the two studies at high or unclear risk of attrition bias (Hodes 2017; Platje 2018). This had no effect on the analysis, which continued to show no difference between groups (6 studies, 375 dyads; SMD −0.07, 95% CI −0.28 to 0.14). This remained the case when we additionally excluded the two studies (Kalinauskiene 2009; Klein Velderman 2006) at unclear risk of selection bias (4 studies, 240 dyads; SMD −0.09, 95% −0.35 to 0.17).

Reanalysis using different statistical approaches

In the preceding sections, we have presented the results from meta‐analyses conducted using a random‐effects model. We repeated all analyses using a fixed‐effect model. There was no difference in overall outcomes for any of the meta‐analyses other than Analysis 1.5. Under a random‐effects assumption, there was no strong evidence of a difference in parental anxiety between intervention and control groups (311 dyads; SMD −0.28, 95% CI −0.87 to 0.31, Analysis 1.5). Under a fixed‐effect assumption, there was evidence of a reduction in parental anxiety in the short term in the intervention group compared to the comparison group (311 dyads; SMD −0.25, 95% −0.47 to −0.02). The two studies pooled in this meta‐analysis, Barlow 2016 and Hoffenkamp 2015, are clinically very similar: both studies are with parents of preterm children, using three sessions of VIG. This might explain why the fixed‐effect model produces some evidence of an effect; however, given the very high heterogeneity (82%), we chose to present the results of the random‐effects model.

Discussion

available in

Summary of main results

Twenty‐two studies, enrolling 1889 parent‐child dyads or family units, met the inclusion criteria for this review. Parents who participated in the studies were experiencing a variety of problems that might impede their ability to respond sensitively to the cues and needs of their children, and that might therefore also undermine their children's ability to form secure attachments. The evidence suggests that video feedback may help to promote parents' sensitivity (moderate‐certainty evidence). An effect size of 0.34 means that if 10,000 parents were to receive a video‐feedback intervention, around 1100 of them would benefit (Magnusson 2014). Furthermore, although the standard system of rating such effect sizes suggests that this finding is small (Higgins 2017), it is highly favourable when compared with that for other parenting interventions such as home visiting programmes, which show evidence of much smaller overall effect sizes (see for example, Michalopoulos 2019).

There is currently only little, very low‐certainty evidence regarding the impact of video feedback on attachment security, compared with control: results differed based on the type of measure used, and follow‐up was limited in duration. There is no evidence of adverse impacts on parental stress (low‐certainty evidence) or anxiety (very low‐certainty evidence). No study measured parental reflective functioning.

There was also no evidence of a moderator effect for the three prespecified variables (intervention type, number of feedback sessions and participating carer) when jointly tested, although parent gender (both parents versus only mothers or only fathers) potentially has a statistically significant, negative moderation effect.

Overall completeness and applicability of evidence

We set out to evaluate the effectiveness of video feedback for improving parental sensitivity and promoting attachment security in children aged under five years old. In terms of completeness, although some RCTs did not provide data in a form that we could incorporate in a meta‐analysis (e.g. Koniak‐Griffin 1992; Moran 2005), and our attempts to obtain such data from the study authors were not always successful, we are confident that we have identified all of the available published evidence.

In terms of applicability, we identified studies targeting parents and children experiencing a range of difficulties or problems that put them at risk of poor parenting (e.g. parental depression; sensitivity problems; intellectual disability; insecure attachment; first‐time teen or immigrant parent; preterm babies; children with autism). The studies were conducted across a range of countries and with widely diverging ethnic groups, in one of three settings: the home; the community, such as a family centre; and inpatient settings, such as hospitals. We also included a comprehensive range of video‐feedback interventions. Most of the studies included mother‐child dyads or primary caregiver‐child dyads where the primary caregiver was the mother, while few studies had more than 10% of participants who were fathers. There were some meta‐analyses (e.g. for child behaviour or parental anxiety) that contain studies primarily with children under 1 year of age, potentially limiting their generalisability.

However, while the results of these studies should not, as such, be extended to fathers, for whom further research is needed, the findings of the current review regarding parental sensitivity appear to be widely applicable and the findings will be useful to both policymakers and health professionals across a range of contexts.

Quality of the evidence

Using the GRADE approach, we rated the overall certainty of the body of evidence between moderate and very low (see summary of findings Table for the main comparison). We did not downgrade any outcome more than once for any of the five domains.

Limitations in study design and implementation

We downgraded two outcomes for limitations in study design and implementation (attachment security at postintervention and parental stress at postintervention or short‐term follow‐up), and one outcome (child behaviour) for risk of bias.

Indirectness of evidence

We did not downgrade any outcomes for indirectness of evidence.

Unexplained heterogeneity or inconsistency of results

We downgraded two outcomes because of heterogeneity: parental sensitivity at postintervention or short‐term follow‐up due to unexplained moderate heterogeneity and parental anxiety at short‐term follow‐up due to high heterogeneity.

Imprecision of results

We downgraded all outcomes except parental sensitivity at postintervention or short‐term follow‐up for imprecision (wide confidence intervals).

Publication bias

We downgraded all outcomes except parental sensitivity at postintervention or short‐term follow‐up and parental stress at postintervention or short‐term follow‐up for publication bias (most studies in this review did not report these outcomes).

Potential biases in the review process

Our literature searches and screening process conformed strictly to Cochrane criteria, as defined by our Methods. We conducted systematic searches across a large number of highly relevant databases, including trials registers, to identify both completed and ongoing trials. Two review authors independently screened potentially eligible studies for inclusion, extracted data, assessed risk of bias in included studies, and rated the certainty of the evidence. Therefore, any reviewer bias was very limited.

For a small number of studies (Hodes 2017; Moss 2011; Poslawsky 2015), we used outcomes data that included children aged five years and over, as the study authors either did not respond to our request for outcomes data with those children who were outside of our included age range excluded, or were not able to provide it. We judged that the number of children aged five years and over from those studies were likely to be very small, and the benefits of including the outcomes data outweighed any negatives.

In terms of conflict of interest, it should be noted that one of the review authors (JB) was the lead author on one of the included studies (Barlow 2016). However, JB was not involved in study selection, data extraction, assessment of risk of bias or GRADE ratings for this study.

We did not prespecify which time point we would use if there were two time points that could be combined in the same meta‐analysis. We have chosen the time point closest to the end of the intervention for consistency; however, in a number of these studies, there was a diminution of effect over time, so we may have found a different result had we chosen later time points.

We assumed that missing data were missing at random, but this may have been an incorrect assumption. Unexplained attrition was quite high in some studies and this also may have impacted the validity of some results.

Agreements and disagreements with other studies or reviews

There have been a number of reviews of different types of video feedback. At the time the protocol for this review was published (O'Hara 2016), only one quantitative review of the effectiveness of video feedback had been conducted: Fukkink 2008. That review concluded that video feedback was effective in improving a range of outcomes when used with parents of children up to seven years of age. However, it also had a number of limitations, perhaps the most important of all being that it included uncontrolled studies and did not rate the quality of the included studies.

Since the publication of our protocol (O'Hara 2016), we have identified other systematic reviews on this topic. Balldin 2018 undertook a systematic review of RCTs and quasi‐RCTs of video feedback; that review focused on describing the components of video‐feedback programmes and the outcomes from individual studies. However, the authors did not undertake a meta‐analysis of results, and their methods state that they searched only a small number of databases together with Google Scholar. Their list of included studies differs from ours, largely because they included studies that used video feedback alongside other interventions, which we excluded. They concluded that video feedback seems effective in improving parental sensitivity, parent behaviour and child behaviour, which is a broader conclusion than we have reached in this review.

Van den Broek 2017 conducted a systematic review of studies examining issues that should be considered when delivering video feedback for children with visual impairments. They included a wider range of study designs than we have here, and they did not identify any RCTs or quasi‐RCTs. They also searched a smaller number of databases than this review. They concluded by highlighting the themes and issues that are important to consider when adapting video feedback for children with visual impairments.

NICE 2016 recommended that video feedback is one of a small number of interventions for which there is low‐quality evidence of effectiveness in improving maternal attachment for children on the edge of care. This was based on evidence gathered through their own systematic search process for the corresponding NICE guideline (NICE 2015). Our findings were more mixed, although our population of interest was broader.

Finally, the authors of VIPP recently published a book chapter summarising the results of a review of 12 RCTs of VIPP, reporting an effect size of 0.47 (95% CI 0.34 to 0.60) for sensitivity (Juffer 2018). They included two studies that we excluded from this review because they did not have an appropriate control group or were not published, and they reported the data differently. For example, in our review, we report that Poslawsky 2015 shows no evidence of impact on parents at risk for autism, which is consistent with the findings reported in the original paper; Juffer 2018 presents the non‐intrusiveness subdomain of the Emotional Avaliability Scale, which shows evidence of effectiveness, while we use the sensitivity subdomain, which shows no evidence of effectiveness. Qualitatively, however, our findings of the impact of a range of types of video feedback on sensitivity are similar.

Overall, therefore, the results of the current review are consistent with those of other reviews.

Figure 1

95 Study flow diagram

Figure 2

'Risk of bias' graph: review authors' judgements about each 'Risk of bias' item presented as percentages across all included studies

Figure 3

'Risk of bias' summary: review authors' judgements about each 'Risk of bias' item for each included study

Figure 4

Funnel plot of comparison: 1. Primary outcomes, outcome: 1.1 parental sensitivity (postintervention ‐ 6 months)

Figure 5

Observed versus predicted intervention effects following moderator analysis

Analysis 1.1

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 1 Parental sensitivity (postintervention ‐ 6 months).

Analysis 1.2

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 2 Attachment security, measured by Strange Situation Procedure (odds of being securely attached) (postintervention).

Analysis 1.3

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 3 Attachment security, measured by Attachment Q‐sort (any duration of follow‐up).

Analysis 1.4

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 4 Adverse events: parental stress (postintervention or short‐term follow‐up).

Analysis 1.5

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 5 Adverse events: parental anxiety (short‐term follow‐up).

Analysis 1.6

Comparison 1 Video feedback versus no intervention or inactive comparator, Outcome 6 Child behaviour (long‐term follow‐up).

Summary of findings for the main comparison. Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment

Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment
Patient or population: parent‐child dyads (including foster or adoptive carers), where the child was aged between birth and four years 11 months (inclusive), and where problems had been identified that were impacting or might impact on the parent's sensitivity Setting: community, hospital outpatient and hospital inpatient Intervention: video feedback Comparison: no intervention or inactive alternative intervention
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Outcomes	Risk with no intervention or inactive alternative intervention	Risk with video feedback	Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Parental sensitivity Follow‐up: postintervention or short‐term follow‐up	‐	The mean parental sensitivity score in the intervention group was 0.34 standard deviations higher (0.20 higher to 0.49 higher)	‐	1757 dyads (20 RCTs)	⊕⊕⊕⊝ Moderate^a	Higher scores indicate a better outcome. Effect size of 0.33 standard deviations compares favourably to other similar interventions.
Parental reflective functioning	‐	‐	‐	‐	‐	No study reported this outcome.
Attachment security Measured by: Strange Situation Procedure (odds of being securely attached) Follow‐up: postintervention	Study population		OR 3.04 (1.39 to 6.67)	166 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a better outcome.
	341 per 1000	612 per 1000 (419 to 776)	OR 3.04 (1.39 to 6.67)	166 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a better outcome.
Attachment security Measured by: Attachment Q‐sort Follow‐up: postintervention	The mean attachment security score across control groups ranged from 0.33 to 0.37 (scores can range from + 1.00 to −1.00)	The mean attachment security score in the intervention group was0.02 standard deviations higher (0.33 lower to 0.38 higher)	‐	131 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Effect size of 0.02 standard deviations indicates no evidence of effectiveness.
Adverse events: parental stress Follow‐up: postintervention or short term	‐	The mean parental stress score in the intervention group was 0.09 standard deviations lower (0.26 lower to 0.09 higher)	‐	537 dyads (8 RCTs)	⊕⊕⊝⊝ Low^b,c	Higher scores indicate a worse outcome. Effect size of 0.09 standard deviations indicates no evidence of effectiveness.
Adverse events: parental anxiety Follow‐up: short term	‐	The mean parental anxiety score in the intervention group was0.28 standard deviations lower (0.87 lower to 0.31 higher)	‐	311 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^c,d,e	Higher scores indicate a worse outcome. Effect size of 0.28 compares favourably to other similar interventions.
Child behaviour Follow‐up: long term	‐	The mean child behaviour score in the intervention group was 0.04 standard deviations higher (0.33 lower to 0.42 higher)	‐	119 dyads (2 RCTs)	⊕⊝⊝⊝ Very low^b,c,d	Higher scores indicate a worse outcome. Effect size of 0.04 standard deviations indicates no evidence of effectiveness.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; OR: odds ratio; RCT: randomised controlled trial
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aDowngraded one level due to inconsistency: moderate heterogeneity, which was not explained by our subgroup analysis. ^bDowngraded one level for risk of bias: we rated most domains in the 'Risk of bias' assessment at high or uncertain risk of bias. ^cDowngraded one level due to imprecision: low number of participants, leading to wide confidence interval. ^dDowngraded one level due to publication bias: few studies in this review reported this outcome. ^eDowngraded one level due to inconsistency: high heterogeneity.

Summary of findings for the main comparison. Video feedback versus no intervention or inactive alternative intervention for parental sensitivity and attachment

Table 1. Methods for use in future updates of this review

Issue	Method
Searching other resources	We will draft a list of included studies to send to experts in the field and ask them to forward to us any published, unpublished or ongoing work that we may have missed.
Measures of treatment effect	Continuous outcome data If necessary, we will compute effect estimates from P values, T statistics, analysis of variance (ANOVA) tables or other statistics, as appropriate.
Measures of treatment effect	Multiple outcomes When a study provides multiple, interchangeable measures of the same construct at the same point in time (e.g. multiple measures of maternal sensitivity), we will calculate the average SMD across these outcomes and the average of their estimated variances. This strategy aims to avoid the need to select a single measure and to avoid inflated precision in meta‐analyses (i.e. preventing studies that report on more outcome measures receiving more weight in the analysis than comparable studies that report on a single outcome measure).
Unit of analysis issue	Cluster‐RCTs In the event that we identify relevant cluster‐RCTs that meet the inclusion criteria of the review, we will deploy appropriate statistical methods based on the guidance provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Where study authors have dealt appropriately with the clustered design in their analyses, we will try to obtain direct estimates of the effect (e.g. an OR with its CI). Where study authors have not dealt appropriately with the cluster design in their analyses, we will extract or calculate effect estimates and their SEs as for a parallel‐group trial, and adjust the SEs to account for the clustering (Donner 1980). To do this, we will need to identify an appropriate ICC, which describes the relative variability in outcome within and between clusters (Donner 1980). Where available, we will look for this information in the reports of relevant trials. If this is unavailable, we will try to obtain the information from the study authors. If this proves unsuccessful, we will use external estimates obtained from similar studies. We will find closest‐matching scenarios (regarding both outcome measures and types of clusters) from existing databases of ICCs. If we are unable to identify any matches, we will perform sensitivity analyses using a high ICC of 0.1, a moderate ICC of 0.01 and a small ICC or 0.001, to cover a broader range of plausible values while still allowing for strong design effects for smaller studies (see Sensitivity analysis). Furthermore, we will combine these estimates and their corrected SEs from the cluster‐RCTs with those from parallel designs using the generic inverse variance method in Review Manager 5 (Review Manager 2014).
Dealing with missing data	Data imputation Where it has not been possible to obtain any unreported data from authors of included studies, and there is reason to believe that it is not missing at random, we will follow the recommendations in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011, Section16.1), and we will do the following: Where appropriate, develop a strategy for data imputation (if we assume the data to be not missing at random). In the case of data imputation, we will specify the methods used in the 'Characteristics of included studies’ tables. We will describe other missing data and dropouts/attrition for each included study in the ‘Risk of bias’ tables, and we will discuss the extent to which these missing data could alter the results or conclusions of the review.
Dealing with missing data	Meta‐regression We will assess the sensitivity of any primary meta‐analyses to missing data using meta‐regression to test for any effect of missingness on the summary estimates (Higgins 2011, Section 16.1.2).
Data synthesis	In the occurrence of severe funnel plot asymmetry, we will present both fixed‐effect and random‐effects analyses under the assumption that asymmetry suggests that neither model is appropriate. If both indicate a presence (or absence) of effect we will be reassured; if they do not agree we will report this.
Subgroup analyses	We will investigate heterogeneity using subgroup analyses or meta‐regression, if appropriate. We will group the included studies and analyse them according to the intervention approach, including the following. Delivery method (i.e. group‐based versus individual delivery) Participating child (e.g. pre‐birth or highly temperamental babies)
Sensitivity analysis	We will assess the robustness of findings to decisions made in obtaining them by conducting sensitivity analyses. We will perform sensitivity analyses by conducting the following reanalysis. Reanalysis excluding studies with imputed data
CI: confidence interval; ICC: intra‐class correlation coefficient; OR: odds ratio; RCT: randomised controlled trial; SD: standard deviation; SMD: standardised mean difference; VIG: Video Interaction Guidance; VIPP‐R: Video‐feedback to promote Positive Parenting ‐ Representational level; VIPP‐SD: Video‐feedback to promote Positive Parenting ‐ Sensitive Discipline

Table 1. Methods for use in future updates of this review

Table 2. Summary of contact with study authors

Study	Date contact initiated	Reason for contacting study authors	Response received
Barone 2019	9 July 2019 (Smith 2019a [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore. We also requested study data for overall child behaviour, as the published report only contained externalising behaviour.	The study author provided the maternal sensitivity and child behaviour data for inclusion in the meta‐analysis (Barone 2019 [pers comm]). They excluded children aged 5 years and over from the data sent over, as the original study did include these children.
Bovenschen 2012	12 January 2018 (Smith 2018b [pers comm])	The reported data were not labelled sufficiently clearly in the published paper to be used in the meta‐analysis. We requested clarification from the study authors.	The study author provided the necessary, additional study data, so they could be included in meta‐analysis (Bovenschen 2019 [pers comm]).
Hodes 2017	2 June 2017 (Smith 2017a [pers comm])	We requested study data relating to parental stress outcomes, reanalysed for children within included age range.	We received no response. As a result, we did not subsequently request the study data on Harmonious Parent‐Child Interaction to be analysed for children within included age range.
Hoffenkamp 2015	16 January 2017 (O'Hara 2017a [pers comm])	We requested missing study data relating to parental sensitivity outcomes.	The study author provided the missing data so they could be included in the meta‐analysis (Van Bakel 2017 [pers comm]).
Høivik 2015	8 February 2018 (Smith 2018c [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore.	The study author provided maternal sensitivity data for inclusion in the meta‐analysis (Hoivik 2018 [pers comm]).
Klein Velderman 2006	21 February 2018 (Smith 2018g [pers comm])	The outcomes data for maternal stress were not included in published studies.	The corresponding author provided us with the missing data for the purposes of meta‐analysis (Bakermans‐Kranenburg 2018 [pers comm]).
Koniak‐Griffin 1992	8 February 2018 (Smith 2018d [pers comm])	The study data for maternal sensitivity were reported in the published paper as part of a composite measure. We requested maternal sensitivity subscore.	We received no response.
Lam‐Cassettari 2015	2 June 2017 (Smith 2017b [pers comm])	The published study data for maternal sensitivity outcomes included children aged 5 years and over. We requested outcomes data with those children excluded.	The study author provided the data with those children aged 5 years and over excluded (Lam‐Cassettari 2018 [pers comm]).
Mendelsohn 2008	1 February 2018 (Smith 2018i [pers comm]	We requested a copy of the conference abstract.	We received no response.
Moran 2005	3 June 2017 (O'Hara 2017b [pers comm])	The maternal sensitivity outcomes were reported as means without standard deviations or standard errors. We requested these data so they could be used in the meta‐analysis.	The corresponding author no longer had access to the data due to retirement, so could not provide the information (Moran 2017 [pers comm]).
Moss 2011	12 January 2018 Smith 2018h [pers comm]	The published study data for maternal sensitivity outcomes included children aged 5 years and over, We requested outcomes data with those children excluded.	We received an initial response from the study authors but they did not subsequently provide the data (Dubois‐Comtois 2018 [pers comm]).
Negrão 2014	12 January 2018 (Smith 2018e [pers comm])	The maternal stress outcomes data were not reported in the published paper, so we requested this information for the purposes of the meta‐analysis.	The study author provided these data for the purposes of meta‐analysis (Pereira 2018 [pers comm]).
Poslawsky 2015	12 January 2018 (Smith 2018f [pers comm])	The reported outcomes included children aged 5 years or older. We requested outcomes data with those children excluded. We also requested means and standard deviations for the relevant 3‐month follow‐up outcome (daily hassles).	The corresponding author was unable to provide the requested data (Poswlawsky 2018 [pers comm]).
Seifer 1991	22 July 2019 (Smith 2019b [pers comm])	We requested outcomes data for mental and psychomotor development	We received no response.
Stein 2006	22 May 2018 (Barlow 2018 [pers comm])	The outcomes data for 'Verbal responses to infant cues' were reported as medians, so we requested the means and standard deviations.	The study authors provided us with these data for the purposes of meta‐analysis (Stein 2018 [pers comm]).

Table 2. Summary of contact with study authors

Table 3. Type of video‐feedback intervention

Study	Aim	Content/delivery
Video‐feedback Intervention to promote Positive Parenting (VIPP; Juffer 2008)
Green 2015	To test the effect of a parent‐mediated intervention for children at high risk of autism spectrum disorder	Video Interaction for promoting Positive Parenting (iBASIS‐VIPP), a modification for the autism prodome of the VIPP infancy programme. The intervention consisted of 12 sessions (an additional 6 booster sessions compared with VIPP).The intervention uses video feedback "to help parents understand and adapt to their infants' individual communication style to promote optimal social and communicative development" (quote). The study authors describe that "The therapist uses excerpts of parent‐child interactions in a series of developmentally sequenced home‐sessions focusing on interpreting the infant's behaviour and recognising their intentions; enhancing sensitive responding; emotional attunement and patterns of verbal and non‐verbal interaction." (quote)
Hodes 2017	To test if a video‐feedback intervention to promote positive parenting and sensitive discipline reduces child‐related parental stress in parents with mild learning disabilities in comparison with care as usual	A Video‐feedback Intervention for Positive Parenting and Learning Difficulties (VIPP‐LD) where the original protocol of VIPP‐SD (Juffer 2008) was adapted for mild intellectual disabilities. For VIPP‐LD, in each session, the parent is videoed interacting with their child. The coach and parent review the footage together, drawing attention to instances of sensitive responsiveness and sensitive discipline, and the coach helps the parent look at the child from the child's perspective. The adaptation included shortening of each session, shorter video recordings and more real‐life practice. The study authors describe how "Parents also received a personal scrapbook with skills taken from video recordings and quotes from the parents representing the theme of the session." (quote)
Kalinauskiene 2009	To evaluate the effectiveness of a short‐term, interaction‐focused video‐feedback intervention implemented in families with mothers rated low in maternal responsiveness	A Video‐feedback Intervention to promote Positive Parenting (VIPP). The intervention was applied as per protocol with the main goal "to reinforce mothers' sensitive responsiveness to their infants' signals focusing on different aspects of mother‐infant interactions" (quote). Mothers were also "provided with information on attachment‐related issues by giving them brochures about sensitive parenting." (quote)
Klein Velderman 2006	To explore if a combination of attention to parental sensitivity and parental attachment representations might lead to firmer and more enduring changes in both parenting behaviour and children's attachment security	A Video‐feedback Intervention to promote Positive Parenting (VIPP). VIPP programs consisted of four home visits lasting 1.5 hours each, with 3‐4 weeks in between. Each session was focused around a specific theme. VIPP‐R included additional discussions on parental representations.
Negrão 2014	To test the effectiveness of a video‐feedback intervention to promote positive parenting and sensitive discipline in a sample of poor Portuguese mothers and their 1‐4‐year old children	A sensitive discipline video‐feedback intervention to promote positive parenting (VIPP‐SD). The study authors state that "VIPP‐SD is a short term intervention programme that relies on video‐feedback technique to enhance parental sensitivity and positive discipline strategies. The intervention was applied through standardised protocols of six home visits...The VIPP‐SD working method is divided into three steps: (1) Sessions 1 and 2 main goals are building a relationship with the mother, focusing on child behaviour and emphasizing positive interactions in the video feedback; (2) Sessions 3 and 4 actively work on improving parenting behaviours by showing the mother when her parenting strategies work and to what other situations she could apply these strategies; and (3) Sessions 5 and 6 (booster) aim to review feedback and information from the previous sessions in order to strengthen intervention effectiveness." (quote)
Platje 2018	To evaluate a video‐feedback intervention aimed at improving parent‐child interaction for parents of children with a visual or visual and intellectual disability	A Video‐feedback Intervention to promote Positive Parenting adapted to parents of children with a visual or visual and intellectual disability (VIPP‐V). The study authors state that the intervention was based on VIPP, but "this new intervention [is] applicable for use in families with a young child with a visual or visual‐and‐intellectual disability. Particular attention was devoted to increasing (safe) exploration, joint attention, and parent’s abilities to recognize and understand the signals and emotions of their child" (quote). The intervention consists of 7 home visits (5 primary visits plus 2 booster sessions).
Poslawsky 2015	To evaluate the early intervention programme, video‐feedback intervention to promote positive parenting adapted to autism, with primary caregivers and their child with autism spectrum disorder	VIPP adapted to autism (VIPP‐AUTI). The intervention comprised 5 home visits lasting 60‐90 minutes every 2 weeks. Sessions included: (1) "Attachment and Exploration" (quote); (2) "Speaking for the Child" (quote); (3) "Sensitivity Chain" (quote); (4) "Sharing Emotions" (quote); (5) "Booster session" (quote).
Van Zeijl 2006	To test the video‐feedback intervention to promote positive parenting and sensitive discipline in "a large sample of families screened for their children's relatively high scores on externalizing behaviour." (quote)	The study applied VIPP‐SD, aimed at parental sensitivity and sensitive parental discipline. The first four intervention sessions each had their own themes, (1) "exploration versus attachment" (quote); (2) "centered around speaking for the child" (quote); (3) "the intervener stressed the importance of adequate and prompt responses to the child’s signals" (quote); (4) "the importance of sharing—both positive and negative—emotions (sensitivity) and promoting empathy for the child" (quote); (5 & 6) "aimed at consolidating intervention effects by integrating—in video feed‐back and discussion—all tips and feedback given in the previous sessions" (quote).
Yagmur 2014	"To test the effectiveness of the video feedback intervention to promote positive parenting and sensitive discipline adapted to the specific child‐rearing context of Turkish families (VIPP‐TM) in the Netherlands" (quote), including second‐generation Turkish immigrant families with toddlers at risk for the development of externalising problems	"The VIPP‐TM program is a culturally sensitive adaptation of the VIPP‐SD program for Turkish minority families in the Netherlands, but follows the general procedures of the original program...The VIPP‐SD program is described in a detailed protocol and consists of six home visits. The first four visits each have their own themes regarding sensitivity and discipline, and the last two sessions are booster sessions in which the themes from previous sessions are reviewed once more." (quote)
Video Interaction Guidance (VIG)
Barlow 2016	"To assess the potential of video interaction guidance to increase sensitivity in parents of preterm infants." (quote)	The study authors report that "VIG is a strengths‐based form of video feedback in which parents are invited to jointly observe and reflect on their own successful interactions with their baby...The core aspects of the model involve three home visits comprising (a) video recording the parent‐infant interaction during play or other aspects of care giving, (b) editing of the recording to select micro‐moments of interaction that demonstrate the infant's contact initiatives and the parents attuned response to these signals and (c) joint reviewing of the recordings with the parent." (quote)
Hoffenkamp 2015	To evaluate the effectiveness of hospital‐based video interaction guidance in parents with moderately and very preterm babies	"Video recordings of parent‐infant interactions and the feedback from a VIG professional provide an opportunity for parents to observe, analyse and discuss the infant's behaviour and contact initiatives" (quote). In this study "VIG consisted of three sessions during the first week after birth" (quote), and included "(1) video‐recording parent‐infant interaction; (2) editing the video recordings; (3) reviewing the edited recordings with parents." (quote)
Lam‐Cassettari 2015	To examine "the effect of a family‐focused psychosocial video intervention program on parent‐child communication in the context of childhood hearing loss" (quote)	Parents completed three sessions: "(a) a goal setting session; (b) three filming sessions of parent–child interaction in the family home, and (c) three shared review sessions in which three short video clips (demonstrating attuned responses linked to the family’s goal) were played so families could microanalyze and discuss." (quote)
Video feedback of Infant‐Parent Interaction (VIPI)
Høivik 2015	To investigate "in a heterogenic community sample of families with interactional problems, whether VIPI would be more effective than standard care (TAU) received in the community" (quote)	VIPI involves at least 6 consultation sessions over a maximum period of 3 months focusing on (1) "Initiative of the infants to contact caregivers and initiate pauses in the dyadic exchange" (quote); (2) "Responses of caregivers" (quote); (3) "Following the child" (quote); (4) "Naming" (quote); (5) "Step‐by‐step guidance" (quote); (6) "Directing attention towards social interaction and exploration" (quote). In this study, "families in the VIPI group received eight video feedback sessions, with the last two sessions tailored to meet the individual family needs regarding any of the six topics in the manual" (quote).
Video self‐modelling with feedback
Benzies 2013	To explore if fathers of late, preterm children who received video self‐modelling with feedback intervention would have better father‐child interaction skills when the child was 8 months old than fathers who received information only	Self‐modelling "involves the father's active participation that increases his cognitive awareness of specific behaviours such as infant cues and how to stimulate development" (quote). The intervention involved video recording a father‐infant play interaction and providing positive feedback and suggestions to enhance the interaction and language development.
Video feedback (non‐specified or other)
Bovenschen 2012	To assess "the effectiveness of an attachment‐based short term intervention using video‐feedback" (quote)	Up to 10 sessions of home‐based video feedback
Green 2010	To test a parent‐child communication‐focused intervention in children with core autism	A parent‐mediated communication‐focused intervention: "The intervention consisted of one‐to‐one clinic sessions between therapist and parent with the child present. The aim of the intervention was first to increase parental sensitivity and responsiveness to child communication and reduce mistimed parental responses by working with the parent and using video‐feedback methods to address parent‐child interaction... incremental development of the child's communication was helped by the promotion of a range of strategies such as action routines, familiar repetitive language and pauses...After an initial orientation meeting, families attended biweekly 2 hour clinic sessions for 6 months followed by booster sessions for 6 months (total 18). Between sessions families were also asked to do 30 mins of daily home practice." (quote)
Koniak‐Griffin 1992	To evaluate "the effects of video tape instruction and feedback (video‐therapy) on mothering behaviours" (quote)	The intervention group received two home visits. Participants were "video taped during structured mother‐infant teaching episodes in their homes at 1 and 2 months postpartum" (quote). Participants "reviewed the video tapes with feedback from a professionally trained nurse who emphasised positive aspects of maternal behaviour" (quote)
Moran 2005	To evaluate "the effectiveness of a brief intervention program designed to support adolescent mothers' sensitivity to their infants attachment signals" (quote)	A brief intervention programme (eight home visits) designed to support the mother's sensitivity to her child. The home visits (lasting approximately one hour) were "designed to provide mutually beneficial play interactions and the mother's enjoyment of her infant" (quote). The four goals of the programme included "to affirm parenting strengths already present in the mother...increase the mother's awareness of how her behaviour influenced her child's behaviour...look for ways to augment the mother's awareness of her infant's signals and for ways to establish positive experiences for both the mother and infant" (quote).
Moss 2011	To evaluate the "efficacy of a short‐term attachment‐based intervention for changing risk outcomes for children of maltreating families" (quote)	The intervention consisted of "8 weekly home visits directed at the caregiver–child dyad and focused on improving caregiver sensitivity" (quote). The study authors describe that "All intervention sessions were primarily focused on reinforcing parental sensitive behavior by means of personalized parent–child interaction, video feedback, and discussion of attachment/emotion regulation‐related themes" (quote).
Seifer 1991	To examine the effects of easy‐to‐use interaction coaching techniques on interaction style and developmental status of a population of mothers and their young children with developmental disabilities	"Interaction coaching" (quote; 10‐month programme). "Sessions lasted six minutes and parents were asked to play with their children as they would during a short break at home...After the initial taping session the video record was viewed by the mother and an interaction coach. Suggestions were then provided by the therapist for the mother to employ during interaction with her child...Another 6 minute interaction was then recorded that was reviewed by the intervener and could be used during the next week’s session. The procedure was repeated for a maximum of 6 sessions." (quote)
Stein 2006	To test "whether video‐feedback treatment especially targeting mother‐child interaction would be superior to counselling in improving mother‐child interaction, especially mealtime conflict and infant weight and autonomy" (quote)	"Thirteen 1‐hour treatment sessions were offered in the mothers’ homes beginning when the infants were between 4 and 6 months old and completed by the time the infants were 12 months old. The intervention group received video‐feedback interactional treatment that was a modification of that developed by [Juffer et al]" (quote). Treatment consisted of three stages: "The first concentrated on the infant’s perspective, focusing on his or her signals...The second stage included the mother’s perspective...Third, as treatment progressed, the videotapes were used to help the mother identify and address potential triggers of mealtime conflict" (quote).
IBASIS‐VIPP: Intervention within the British Autsim Study of Infant Siblings ‐ Video‐feedback Interaction to promote Positive Parenting; Mins: Minutes; TAU: Treatment as usual; VIG: Video Interaction Guidance; VIPI: Video‐feedback of Infant‐Parent Interaction; VIPP: Video‐feedback Interaction to promote Positive Parenting; VIPP‐AUTI: Video‐feedback Interaction to promote Positive Parenting ‐ Autism;VIPP‐LD: Video‐feedback Interaction to promote Positive Parenting ‐ Learning Difficulties; VIPP‐R: Video‐feedback Interaction to promote Positive Parenting ‐ Representational level; VIPP‐SD: Video‐feedback Interaction to promote Positive Parenting ‐ Sensitive Discipline; VIPP‐TM: Video‐feedback Interaction to promote Positive Parenting ‐ Turkish Minorities; VIPP‐V: Video‐feedback Interaction to promote Positive Parenting ‐ Visual or visual and intellectual disability.

Table 3. Type of video‐feedback intervention

Comparison 1. Video feedback versus no intervention or inactive comparator

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Parental sensitivity (postintervention ‐ 6 months) Show forest plot	20	1757	Std. Mean Difference (Random, 95% CI)	0.34 [0.20, 0.49]

1.1 VIPP	10	861	Std. Mean Difference (Random, 95% CI)	0.24 [0.05, 0.42]
1.2 Other types of video feedback	10	896	Std. Mean Difference (Random, 95% CI)	0.44 [0.23, 0.66]
2 Attachment security, measured by Strange Situation Procedure (odds of being securely attached) (postintervention) Show forest plot	2	166	Odds Ratio (M‐H, Random, 95% CI)	3.04 [1.39, 6.67]

3 Attachment security, measured by Attachment Q‐sort (any duration of follow‐up) Show forest plot	2	131	Std. Mean Difference (IV, Random, 95% CI)	0.02 [‐0.33, 0.38]

4 Adverse events: parental stress (postintervention or short‐term follow‐up) Show forest plot	8	537	Std. Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.26, 0.09]

5 Adverse events: parental anxiety (short‐term follow‐up) Show forest plot	2	311	Std. Mean Difference (IV, Random, 95% CI)	‐0.28 [‐0.87, 0.31]

6 Child behaviour (long‐term follow‐up) Show forest plot	2	119	Std. Mean Difference (IV, Random, 95% CI)	0.04 [‐0.33, 0.42]

Comparison 1. Video feedback versus no intervention or inactive comparator