Types of studies

Only randomized controlled trials (RCTs) with at least five participants in each study arm were included. In contrast to our original review in which we included quasi‐randomized trials (for example, alternative assignment) in a sensitivity analysis, for this updated review we included only true RCT designs and excluded all quasi‐randomized trials. A study was deemed to be a true RCT if the authors explicitly stated that participants were randomly assigned to groups (for example, assignment was determined using a table of random numbers) and did not indicate using quasi‐randomization methods (for example, alternating assignment) at any point in the paper. For example, some authors may describe the study as an RCT in the title or abstract but later report quasi‐randomization techniques in the methods sections. A true RCT is one that is described as such throughout the entire paper. While we were more lenient in the original review and included a few studies reporting quasi‐randomization procedures, these studies were excluded from the present review in order to be more consistent and stringent in our approach. In addition, unlike in our original review, we included only published trials and therefore excluded any non‐published trials (for example, dissertations). No language restrictions were used during the search and translations were obtained when necessary.

Types of participants

Studies involving children and adolescents aged two to 19 years undergoing needle‐related medical procedures were included. For the purposes of this review, a needle‐related medical procedure was defined as any procedure performed as part of a medical diagnosis, prevention, or treatment. This included dental procedures (excluding dental surgery) but did not include procedures such as body piercings or tattoos, which involve needles but are not performed for medical purposes. The search was limited to needle‐related pain, which is among the most commonly occurring and feared procedures for both healthy and chronically‐ill children (Broome 1990). Our justification for not including children less than two years of age is that the majority of psychological interventions examined in this review are either not appropriate for use with infants or are qualitatively different when applied to infants. The efficacy of psychological interventions for procedural pain and distress in infants has been addressed in another review (Pillai‐Riddell 2011a; Pillai‐Riddell 2011b). For both the original review and this review update, we included studies with participants ranging from two to 19 years of age. A maximum age of 19 years was chosen to ensure that our search was limited to children and adolescents only, and is also consistent with the World Health Organization (WHO) definition of adolescence, which defines it as ranging from 10 to 19 years (http://www.searo.who.int/en/Section13/Section1245_4980.htm). The age range was kept broad so as to not exclude any relevant studies; however, studies that included any participants falling outside of this age range were excluded unless authors were able to provide data for only the age range specified for this review while also continuing to meet the minimum criterion of five participants per group.

We included studies examining medical procedures from the same list of procedures identified in our prior review (please see Table 1 for the list of medical procedures and their definitions). Definitions were derived from online medical dictionaries (for example, MedlinePlus Medical Encyclopedia, MedLine 2004; On‐Line Medical Dictionary, OLMD 2004) and by consulting with medical professionals in the area of pediatric pain. Participants included healthy children and children with chronic or transitory illnesses from both inpatient and outpatient settings. Studies including patients with known needle phobias (that is, diagnosed by a qualified professional such as a psychologist) were excluded. While we acknowledge that needle fears are quite prevalent and some youth included in the reviewed studies may have had undiagnosed phobias, this criterion was established in an attempt to distinguish between typical needle fears or ones that are extreme enough to have warranted assessment and diagnosis prior to the study. Furthermore, children undergoing surgery were excluded because numerous factors specific to surgery or being in intensive care units can complicate and interfere with the accuracy of self‐reported accounts of pain and distress. These factors may include sedation, inability to verbalize because of intubation, more intensive pharmacological interventions, long‐term hospital stays, inability or difficulty attributing pain or distress to one specific medical procedure, and difficulty distinguishing between the pain and distress caused by the procedure versus the medical condition requiring the surgery (Puntillo 2004). The exception to this was for studies that assessed the efficacy of a psychological intervention for a pre‐surgical needle procedure such as an intravenous (IV) insertion. The outcome measures of interest had to be completed prior to surgery or sedation in order for the study to be included in this review.

Types of interventions

Studies were included if at least one trial arm consisted of a psychological intervention, and there was a comparator control arm, other active treatment, treatment as usual, or waiting list control. We excluded studies in which the psychological intervention(s) was combined with a non‐psychological intervention (for example, pharmacological intervention) when the unique effects of the psychological intervention could not be evaluated.

Types of outcome measures

The two measured outcomes of interest were pain and distress, assessed using scales or measures with established reliability and validity (that is, as evidenced in at least one prior study published in a peer‐reviewed journal). For this review, pain was operationalized using the definition proposed by the International Association for the Study of Pain (IASP), describing pain as: “an unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage” (IASP 2004). In addition, distress was broadly defined as any type of negative affect associated with the procedure (for example, anxiety, stress, fear). Because the main focus of this review was the data pooling component (that is, the meta‐analysis), as with the original review, RCTs were excluded if the data necessary for data pooling (for example, means, SDs, cell sizes) were not available in the published study, could not be identified through contact with the study authors, or could not be calculated based on other data provided (for example, confidence intervals).

Electronic searches

Through consultation with a reference librarian and with assistance from the Cochrane Pain, Palliative and Supportive Care (PaPaS) Group, detailed search strategies were developed to search the following six electronic databases for relevant trials:

• Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2013, Issue 2);

• MEDLINE (1966 to March 2013) ;

• EMBASE (1989 to March 2013);

• PsycINFO (March 2013);

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) (March 2013);

• Web of Science (IBI Web of Knowledge) (March 2013).

For this updated review, the same database search terms were used as in the original review; however, we extended the search date to cover the period from when the last update ended (2005) to March 2013. Thus, altogether, the database searches covered the period from their inception to March 2013. Because we opted to include only published studies for this current review update, we excluded one additional database searched in the original review (that is, Dissertation Abstracts International). A similar search strategy was adapted for each of the databases but additional related keywords and MeSH terms were included as appropriate for each specific database. Please see the appendices for the search strategies and terms used for each of the databases: MEDLINE (Appendix 1), PsycINFO (Appendix 2), CENTRAL (Appendix 3), EMBASE (Appendix 4), IBI Web of Knowledge (Appendix 5), and CINAHL (Appendix 6).

Searching other resources

In addition to the electronic search strategy, we also put out a call for relevant studies to four e‐mail list serves including: 1) Pain in Child Health (PICH), 2) Pediatric Pain, 3) the American Psychological Association’s Society of Pediatric Psychology Division 54, and 4) the American Psychological Association’s Health Psychology Division 38. Requests to these list serves were sent out for both the original review and, again, for the present review update. Any other relevant studies identified and included in the original review (for example, from reference or citation lists) were included in this update.

1. Selection of studies

Two review authors (LU and CC) independently screened titles and abstracts of trials from literature searches for inclusion in the original review, and two additional review authors (MN and KB) independently screened titles and abstracts for the review update. For all abstracts that were relevant, potentially relevant, or where relevance to the current review was unclear, the full articles were obtained and read by one of the study authors. Using the full articles, two review authors (LU and CC for the original review, and MN and KB for the updated review) decided which studies did and did not meet the inclusion criteria. Review authors were not blind to the authors, institutions, journals, or results. A third review author (PM for the original review, and LU for the updated review) was brought in to help resolve any issues or selection discrepancies that arose.

As noted above, to increase the methodological quality of the studies included in this updated review, we imposed more stringent inclusion criteria for this review update as compared to the original review. Specifically, for this review update we omitted any non‐published studies (for example, unpublished dissertations) and quasi‐randomized trials (that is, anything other than reporting true random assignment, as described above). In our original review, contact with some study authors clarified that the assignment they used was indeed random assignment although it was not described as such in the paper (for example, described in the abstract as an RCT but described in the methods as using alternating assignment). Thus, to be consistent across all studies, those reporting quasi‐randomized procedures in the published studies were excluded from this update even if they were included in the original review.

A total of 28 RCTs were retrieved and included in the original review, although for the current update we excluded seven of these studies because they were unpublished dissertations or reported quasi‐randomized methods (that is, alternating assignment). In March 2012, our updated search strategy identified 18 additional RCTs meeting the inclusion criteria and having the necessary data to be included in this update. An additional search in March 2013 did not identify any further studies. Thus, a total of 39 RCTs (retrieved from the original and updated searches) were included in this updated review; all of these included RCTs were coded in full. The references for these studies are all provided below in the 'Description of studies' section.

2. Data extraction and management

Two review authors (LU and CC) extracted data using a data extraction form designed for the original review, and two additional review authors (MN and KB) performed data extraction for additional studies identified in the review update. In addition, another researcher who was fluent in Farsi confirmed inclusion eligibility and conducted data extraction for one non‐English study identified from our updated search. Data were extracted on various details relating to study design, participant demographics, diagnosis (when applicable), needle procedure, type of intervention and control conditions, outcomes, as well as other related variables. A third review author (PM for the original review, and LU for the updated review) was available to help resolve any coding discrepancies. Data from the studies were extracted using paper data extraction sheets developed for this review and were also entered into electronic spreadsheets to differentiate between included versus excluded studies. Attempts were made to obtain missing data (for example, means, SDs, cell sizes) from the authors, whenever feasible, to include in the meta‐analysis. All data for the included studies were first recorded on paper data extraction forms by review authors (LU for the original review, and MN and KB for the updated review) and reviewed for errors by a trained research assistant. All data suitable for pooling were analysed using RevMan 5.1 software (RevMan 2011).

3. Assessment of risk of bias in included studies

In the original review, each included study was scored independently for quality by two review authors using the Oxford Quality Scale created by Jadad 1996. We also completed a more extensive quality analysis of the included studies (Uman 2010) given that the Oxford Quality Rating Scale is more appropriate for double‐blind studies (for example, pharmacological interventions); and psychological interventions, unlike placebo studies, typically cannot be truly double‐blind. For the purpose of this updated review, we used the Cochrane 'Risk of bias' tool as a measure of trial quality. The Cochrane Handbook for Systematic Reviews of Interventions (www.cochrane‐handbook.org/) distinguishes between methodological quality assessment and risk of bias, and recommends a focus on the latter (please refer to the Cochrane Handbook for a detailed account of reasons why). The Risk of bias tool is a two‐part tool assessing seven areas: adequate sequence generation, allocation concealment, blinding of participants and study personnel, blinding of outcome assessment, incomplete outcome data addressed, free of selective reporting, and other biases. All included studies from the original review and those identified in this update were independently coded by two review authors (MN and KB) who then compared responses to reach consensus. This procedure was verified with, and endorsed by, the Cochrane PaPaS Review Group.

4. Measures of treatment effect

Given the nature of the outcome measures in this review, all of the outcome data for the included studies were continuous (for example, rating scales). Consistent with the analytical approach used in the original review, and as is recommended by The Cochrane Collaboration, we computed standardized mean differences (SMD) with 95% confidence intervals (CI), which allowed us to combine the results from different scales measuring the same construct (for example, pain). We proposed that when sufficient data were available from various studies using the same measurement instruments, a weighted mean difference (WMD) with 95% CI would also be conducted. However, given the wide range of different assessment measures used, this was not feasible. Thus, all mean differences presented in the tables and plots of the results section represent SMDs.

Each psychological intervention was assessed separately in a meta‐analysis because the interventions were considered too different to combine. However, within each intervention we analysed each outcome measure separately. Thus, for each intervention the following seven outcomes were separately assessed.

a) Pain: self‐reports.

b) Pain: observer global reports.

c) Pain: behavioural measures.

d) Distress: self‐reports.

e) Distress: observer global reports.

f) Distress: behavioural measures.

g) Physiological measures: each physiological outcome (e.g. heart rate, blood pressure) was assessed separately.

5. Unit of analysis issues

In this context, unit of analysis issues refer to the level at which randomization occurs. For example, in most situations the number of observations analysed should match the number of 'units' that were randomized. In a simple, parallel two‐group RCT (intervention versus control), participants would be randomized to one of the two groups and a single measurement for each outcome would be collected and analysed. For the current review, we included and analysed parallel two‐group RCTs as well as cluster‐randomized trials (that is, groups of individuals randomized together to the same intervention). In addition, we also included cross‐over trials (that is, participants undergo more than one intervention); however, we only included these studies if the data were available separately for each group following the first treatment arm. For example, in a treatment (n = 50) versus control (n = 50) two‐group design we required the data for the first treatment arm (treatment group versus control) before the groups crossed over and the conditions were reversed rather than the overall combined values for all treatment values compared to all control values. The rationale for this decision was based on the belief that once a psychological intervention (for example, distraction, hypnosis) has been introduced and taught to participants, it is difficult to prevent participants from using these strategies themselves (for example, cognitive distraction techniques) even if they are then switched to a control condition. Lastly, we also included studies using repeated measures designs (for example, pain and distress outcomes assessed at various time points); however, in these situations we only included outcomes occurring during the needle procedure (pre‐needle outcomes were not included). If outcomes during the needle procedure were not evaluated, we then selected the next proximal time point occurring closest to the completion of the procedure. If outcomes that assessed both during and following the needle‐related procedure were reported, outcomes during the needle procedure were selected for inclusion.

6. Dealing with missing data

All included RCTs were reviewed to determine if the authors reported any missing data. We had originally proposed that data would not be presented for studies if more than 20% of the originally randomized participants withdrew; however this was not applicable to any of the retrieved studies.

In all situations where data necessary for data pooling were omitted from the published RCT (for example, means, standard deviations (SDs), groups sizes, CIs), attempts were made to contact the study authors to obtain these data. If that was not possible, statistical formulae identified in the Cochrane Handbook for Systematic reviews of Interventions for calculating missing data using other reported measures of variation were used (for example, obtaining standard deviations from standard errors, confidence intervals, t values, and P values). In situations in which the authors could not be contacted, did not respond to contact attempts, did not have these data available, or there were insufficient data available to compute the necessary calculations, these studies or the outcomes with the missing data were excluded from this review.

To be consistent with authors’ reporting, we included the number of participants per group identified in the study results sections. When not otherwise specified by the authors, we assumed that there were no study dropouts and used the reported group sizes in the meta‐analyses. Initially we intended to conduct all analyses using intention‐to‐treat (ITT) analyses. ITT analyses include all participants enrolled in the study including those who were lost to follow‐up or dropped out. Given that the majority of RCTs did not specifically identify whether ITT analyses were used and did not consistently report whether there were any participant dropouts, ITT analyses could not be performed.

7. Assessment of heterogeneity

For each outcome combining the effects of two or more studies, we calculated heterogeneity using both the Chi² test and the I² statistic. Given that Chi² tests often have low statistical power, a type 1 error level of 0.10 was employed for rejecting the null hypothesis of homogeneity as opposed to the more traditional type 1 error level of 0.05. While Chi² tests are useful for identifying whether heterogeneity is present, it has been argued that there will always be some level of heterogeneity in meta‐analyses given the clinical and methodological diversity (Higgins 2011). The I² statistic shifts the focus away from whether heterogeneity is present and provides a measure of inconsistency across studies to assess the impact of heterogeneity on the meta‐analysis (Higgins 2011). I² is expressed as a percentage from 0 to 100, and the Cochrane Handbook for Systematic Reviews of Interventions suggests the following rough interpretation guide: 0 to 40%, might not be important; 30% to 60%, may represent moderate heterogeneity; 50% to 90%, may represent substantial heterogeneity; and 75% to 100% represents considerable heterogeneity. The reason for some of the overlap in ranges is because the importance of I² depends on several other factors such as the magnitude and direction of effects, as well as the strength of evidence for the heterogeneity (for example, the P value for the Chi² test or the confidence interval for the I² statistic). In cases where statistically significant heterogeneity was detected, the data were still pooled; however, these results should be interpreted with caution. Given that there was significant heterogeneity for several of the analyses, results were analysed using a random‐effects model. When possible, attempts were made to explore the reasons for the heterogeneity. We chose only to report I² values in the text of the results section but both the I² statistics and Chi² tests are depicted in the forest plot figures.

8. Assessment of reporting biases

In order to help overcome publication bias in both the original and updated review, we: (1) imposed no language barriers in our search, (2) contacted several list serves and researchers in the field of pediatric health and pain to request any published, unpublished, and in‐progress studies, and (3) contacted the authors of all the studies with missing means, standard deviations, and cell sizes in attempts to retrieve these data. Of note, many studies included several outcome measures; however, in some situations authors only reported means and standard deviations when the group differences with respect to the intervention were significant, but not when group differences were not significant. We included any studies in the meta‐analysis that provided completed results (that is, means, SDs, and cell sizes for both treatment and control groups) for at least one outcome measure. Information related to reporting biases is also captured in the Risk of bias tool used to code all of the included studies reported in this update (studies included from the original review were also coded for this update using this tool). Nine of the studies included in this review had authors who responded to our requests and provided clarification or missing data (Balan 2009; Bisignano 2006; Caprilli 2007; Cavender 2004; Gupta 2006; Kleiber 2001; Liossi 1999; McCarthy 2010; Sinha 2006; Wint 2002).

9. Data synthesis

SMDs using a random‐effects model for each of the above subcategories were calculated when the necessary data were available. Interventions were considered efficacious when the SMD and both anchors of the CI fell in the negative range. In the original review, we limited the main analyses to only RCTs but included quasi‐randomized trials in sensitivity analyses. Due to the limitations of quasi‐randomized trials, we limited the analyses to only true RCTs for this review update and did not include the additional sensitivity analyses for quasi‐randomized trials.

In situations where studies included more than one type of control condition, we selected the condition that was most similar to the intervention condition, with the only difference being non‐administration of the specific intervention. Thus, if a study had three conditions comparing: (1) music with headphones, (2) headphones without music, and (3) no headphones or music, we would select the headphones without music condition as the appropriate control condition because it was most similar to the intervention but just did not include the active ingredient (that is, music). In addition, it was important to note that some control or standard care groups included some cognitive or behavioural techniques, or both. However, we decided a priori to continue to classify these as control conditions since they were determined to be part of standard care and were conceptualized as control conditions by the study authors. In addition, and as noted above, if the control condition included a pharmacological component (for example, lidocaine and prilocaine eutectic mixture (EMLA), a cream used for local anaesthesia), this was still conceptualized as a control condition provided that the intervention group also received the same pharmacological component plus a psychological intervention. The implications of these types of control conditions are further explored in the discussion section of this review.

Consistent with the original review, when a single study provided more than one observer rating of the same construct (for example, both parent and nurse VAS ratings of child pain) or more than one behavioural measure for the same construct (for example, CAMPIS and OSBD measures to assess distress), these measures were pooled using statistical formulae recommended by The Cochrane Collaboration for combining means and SDs. This was done in order to be able to summarize the large amount of data reported in these studies. The formulae we used to pool means and SDs were the following: pooled mean = [(mean1 x N1) + (mean2 x N2) / (N1 + N2)] and pooled SD = square root of [SD1² (N1 ‐ 1) + SD2² (N2 ‐ 1)] / N1 + N2 ‐2.

As previously indicated, for studies that included outcomes for numerous time points we restricted our analyses to the measurement occurring during the procedure or, if that was not provided, we used the first post‐procedure measurement. For example, if a study included procedural measures (taken during the needle) and post‐procedural measures (taken after the needle), we included only the former in our analyses. For studies that included only post‐needle measures but provided them at several time points (for example, immediately after needle, one minute later, five minutes later), we included the first measure or the outcome assessed as soon as possible following the needle. Pre‐procedural measures of pain or distress were not analysed as the focus of this review was on pain and distress reduction during needle procedures.

10. Subgroup analysis and investigation of heterogeneity

As with the original review, subgroup analyses were limited to analysing each category of psychological intervention separately (for example, distraction RCTs, hypnosis RCTs, etc) as the interventions were considered too qualitatively distinct to combine, and overall analyses including all RCTs would not be as meaningful. In addition, as mentioned above, within each intervention category analyses were broken down by outcome category (pain versus distress outcomes) and within each of those categories they were further broken down by outcome measure (that is, self‐report, observer‐report, behavioural rating scales, and physiological interventions). Physiological outcomes were further analysed separately as they represented distinct functions or processes (for example, heart rate versus blood pressure). For all outcomes, the Chi² test and I² statistic of heterogeneity were carried out as previously described.

11. Sensitivity analysis

In our original review, we were unable to conduct all of the sensitivity analyses that we proposed due to insufficient data reported within and across studies, as well as the small number of studies within each intervention category. The main sensitivity analyses we conducted in the original review involved comparing the study results when quasi‐randomized trials were added to the analyses. However, in order to strengthen the methodological quality of the findings for this updated review, we limited the included trials to only true RCTs that explicitly stated that true random assignment was conducted. As a result, the four studies with alternating assignment included in the sensitivity analyses of the original review were omitted from this updated review (Christiano 1996; MacLaren 2005; Manne 1990; Sparks 2001).