Brief Report: Specificity of Interpersonal Synchrony Deficits to Autism Spectrum Disorder and Its Potential for Digitally Assisted Diagnostics

Participants were drawn from a representative clinical referral population of adults referred to autism diagnostics to the specialised autism outpatient clinics of the University Hospitals of Munich and Cologne. All participants were assessed while undergoing real-life diagnostic procedures and had been referred to the specialist clinics by medical consultants (psychiatry and neurology) on the basis of suspicion of a possible ASD diagnosis due to reported social-emotional symptoms. Diagnostic procedures have been conducted in accordance to German guidelines for diagnosis of ASD in adulthood (German S3-guidelines; AWMF, 2016) comprising neuropsychological and clinical assessment of at least two independent trained clinicians. The assessed population was retrospectively split into ASD + cases (n = 16) consisting of individuals who received a diagnosis of F84.5 (n = 10; Asperger Syndrome), F84.1 (n = 3; Atypical Autism), or F84.0 (n = 3; Childhood Autism) according to ICD-10 (World Health Organization, 2016) and ASD- cases (n = 23) consisting of individuals for whom any F84 diagnosis (including F84.9) was ruled out.

The groups were frequency-matched with respect to IQ (t(37) =  − 0.437, p = 0.664), motor difficulties (U = 239.5, p = 0.116) and age (U = 134, p = 0.157). Descriptive statistics can be found in Table 1. All participants gave written informed consent before study participation. The study was approved by the ethics committee of the medical faculties of the LMU Munich and the University of Cologne, in agreement with the Declaration of Helsinki.

Data was collected in an observational study design accompanying standard diagnostic procedures. The first diagnostic interview after an initial admission appointment of every participant was videotaped on a HD camera (Sony HDR-CX625) with a frame rate of 25 fps. The interviews were conducted by eight different diagnosticians (including JK). Subsequently, patients underwent several hours of analysis of medical history, former parental/caretaker interviews and neuropsychological assessments (according to German S3-guidelines; AWMF, 2016). Finally, a diagnostic decision was eventually made by one of three specialized clinical experts (including KV and CFW) who did not take part in the data collection. Participants were post-hoc assigned to either the ASD + or ASD- group according to diagnostic decision.

The video setup, lighting and seating arrangement closely resembled the setup of our previous study on reduced IPS in autism (Georgescu et al., 2020). To ensure comparable filming conditions, stable ambient light was kept and the camera position remained static throughout the interviews. Chairs were positioned in fixed spots at 45° angle from the camera. Seating positions of participant and diagnostician (left, right) were allocated in a counterbalanced order. A video vignette of the first 14 min of each diagnostic session was extracted, to allow for unobstructed analysis with both participants seated and without any external disturbances.

Participants filled out the AQ (Baron-Cohen et al., 2001) to measure autistic trait severity, the EQ (Baron-Cohen & Wheelwright, 2004), an index of global empathy, and the degree of self-rated motor difficulties, as indicated by a translated German version of the Adult Developmental Co-ordination Disorders/Dyspraxia Checklist (ADC; Kirby et al., 2010), and completed a post-test questionnaire to rate the quality of the interaction and the extent by which they felt influenced by the recording camera. As part of the standard clinical assessment, participants additionally completed the German version of the Beck’s Depression Inventory (BDI; Hautzinger et al., 1994), the Systemizing Quotient (SQ; Frith et al., 2003), and the 20-Item-Toronto Alexithymia Scale (TAS20; Bagby et al., 1994).

Videos were analyzed post-hoc with Motion Energy Analysis Version 3.10 (MEA; Ramseyer & Tschacher, 2011), an observer-independent, objective tool to quantify movement. MEA is a frame-differencing method used to quantify the amount of movement present in a video. The software extracts changes in grayscale pixel values frame by frame, so-called motion energy, of videos in pre-defined regions of interest (ROI; Ramseyer & Tschacher, 2014). With stable background and lighting, each change in pixels/motion energy indicates body motion of the participants. In line with previous studies using MEA, two ROIs were chosen manually for each participant, namely head and upper body. MEA then yields movement time series of gray-scale pixel change of every frame for each ROI above a manually-set threshold within the program that allows for the filtering of signal fluctuations as opposed to real movement (for an in-depth description of MEA see Ramseyer et al., 2019). After careful inspection of our data, we chose a value of 15, thereby lowering the default (25) to adjust for our lighting specifics. To compute interpersonal synchrony (IPS), time series were cross-correlated in moving windows of 60 s for all time lags between − 5 and + 5 s, using R custom software (rMEA; Kleinbub & Ramseyer, 2020). Cross-correlations were Fisher’s Z-transformed to allow aggregation and their means aggregated in absolute values over the 14-min interval, yielding two interpersonal synchrony values per dyad, for head and body movement respectively. One vignette had to be split in half due to interactants exchanging documents half way through the conversation. The resulting IPS values were averaged across the two vignettes. We have also added the values of the head and body ROI (as they are mutually exclusive) to compute a total ROI per person.

In order to evaluate the significance of synchrony values, we need to validate the procedure against coincidental synchrony (i.e. synchrony that might have occurred by chance). To this end n = 1,000 surrogate synchronies were computed (out of a possible of N = 3,120) by aligning time series of participants in random order who never actually interacted with one another (Ramseyer, 2019). The resulting surrogate datasets were analyzed in the same manner, again yielding two global values per dyad (head and body) for pseudosynchrony.

The relative amount of movement quantity of every participant was calculated as the percentage of frames with above-threshold movement (number of frames with frame differences > 0 divided by total number of frames) within every ROI (Ramseyer & Tschacher, 2014). The values of the respective interactants were averaged within every dyad to obtain an indicator of mean movement quantity per dyad. A table of the descriptive statistics can be found in the supplementary results.

The context and nature of these interactions required that the clinician interacts with objects (a clipboard and pen). This makes the synchrony in the body ROI more difficult to interpret. For reasons of completion we have reported these, along with the total ROI in the supplementary materials. We therefore report the head ROI and have good reason to do so: In previous research on IPS, head synchrony has been associated with the overall outcome of psychotherapy (Ramseyer & Tschacher, 2014) and has been found to be indicative of the symptom profile of schizophrenia (Kupper et al., 2016), a psychiatric disorder that shares several common features with ASD, though the exact dynamic patterns remain to be investigated. Further, head movement dynamics (pitch, yaw and roll) have been found to differ between children with and without ASD in a social context (Martin et al., 2018), deeming them appropriate for further investigation in a synchrony context.

To ascertain that the method yielded valid IPS measurements, we compared mean IPS with surrogate data across participants using the Mann–Whitney test, as a nonparametric alternative for independent samples t-test (Ramseyer, 2020), because a Shapiro–Wilk test showed a significant departure from normality for the pseudo-IPS values (W = 0.991, p < 0.001). In line with previous studies applying MEA, IPS was significantly higher than pseudosynchrony in the head ROI (U = 23,354, p = 0.018, one-tailed, d_Cohen = 0.198).

To rule out the overall amount of movement as a confounding factor for potential differences in IPS, we examined differences in average movement quantity per dyad across groups. We performed an independent samples t-test. No significant differences were found between groups (t(37) = − 0.326, p = 0.746, d_Cohen =  − 0.106), suggesting similar amounts of mean overall head movement in both groups.

To investigate group differences in interpersonal synchrony, we performed an independent samples t-test comparing ASD+ and ASD− . We found a significant difference of IPS head synchrony between the ASD + and the ASD- group (M = 0.061 and M = 0.070 respectively, t(37) = − 2.068, p = 0.023, one-tailed, d = − 0.673). IPS was significantly lower in interviews with ASD+ patients than with ASD− patients not fulfilling diagnostic criteria.

We conducted an independent samples t-test for the included standard screening instruments for autism traits (AQ) and empathy (EQ) and subsequently correlated them with the IPS values within groups. We found no significant group differences for both (AQ: U = 192, p = 0.830, d = 0.043; EQ: t(37) = − 1.065, p = 0.294, d = − 0.347; see Table 1 and Fig. 1). For the dyspraxia questionnaire (ADC), scores of 4 participants (3 ASD− and 1 ASD+) were missing and imputed with the respective group average. We found no significant group differences (U = 239.5, p = 0.116, d = 0.302; see Table 1 and Fig. 1) and no significant association with IPS. A correlation table can be found in the supplementary materials.

Important limitations of this study must be considered when interpreting the results. We recognize that our findings will need to be replicated in a larger sample and with specific homogeneous clinical control samples complementing our results of specificity of reduced IPS for adults with a confirmed ASD diagnosis within a referral population of individuals with a suspicion of ASD. Furthermore, MEA, though frequently used to quantify IPS and being substantially more time-efficient and objective than manual coding, has certain methodological constraints (as discussed before, Georgescu et al., 2020). In particular, while we chose MEA due to our focus on timing of social interaction and easy translation into clinical practice, other methods might be used in the future to complement our findings with abnormalities in spatial movement patterns, such as 3D tracking methods or qualitative variables underlying our findings of reduced head IPS in ASD. It is important to note that sophisticated motion capture technology, while certainly being excellent tools for fine-grained motion tracking, pose specific constraints to a clinical population, are high-cost and may currently not readily be translatable to clinical practice. Lastly, due to the dyadic nature of the output variable IPS, we cannot entirely rule out the possibility that the perceived IPS by the clinician conducting the interview might have influenced the impression formation of the conversational partner. Indeed, previous research has shown that perceived synchrony in a motor-tapping task increases empathy towards a partner (Koehne et al., 2015). Thus, while the diagnostic decision was made by specialized clinical experts not taking part in the data collection and considering multiple anamnestic sources adhering to German diagnostic S3-guidelines for ASD (AWMF, 2016), perceived IPS might have influenced the attitude and degree of understanding of the interviewer towards the patient. This becomes especially crucial within the “double-empathy problem” framework in autism (Milton, 2012), explaining the disconnection between two interaction partners by a lack of understanding for the other. By investigating a measure of interpersonal coordination like IPS, we are essentially moving away from individual deficits alone to include dyadic measures. This is particularly important in autism, as a condition with social difficulties but also in the broader context of “2nd person psychiatry” (Schilbach, 2016).