Ex-Gaussian, Frequency and Reward Analyses Reveal Specificity of Reaction Time Fluctuations to ADHD and Not Autism Traits

Attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorders (ASD) show high rates of co-morbidity and significant overlap in genetic influences (Rommelse et al. 2010) and in several neurocognitive impairments, including those in executive functioning, sustained attention, and response to rewards (Rommelse et al. 2011). Of individual measures that show an association with both ADHD and ASD, reaction time variability (RTV) – linked to the neural mechanisms underlying attention allocation (Cheung et al. 2017) and arousal regulation (James et al. 2016), − is a particularly promising candidate for investigation that might ultimately inform the neurobiology of the two disorders and their overlap. Accordingly, here we test whether RTV measures may help find common and unique impairments in ADHD and ASD under varying task conditions.

In ADHD research, high RTV has emerged as one of the neurocognitive impairments showing the strongest phenotypic and genetic association with the diagnosis and with continuous ADHD symptom scores (Kofler et al. 2013; Kuntsi et al. 2010; Crosbie et al. 2013). Evidence is now also accumulating of a phenotypic association of high RTV with the ASD diagnosis and traits (Karalunas et al. 2014; Pinto et al. 2016), which is partly explained by shared genetic influences (Pinto et al. 2016). Given the high co-occurrence of ADHD and ASD, a further question is whether the association of high RTV with ASD may be explained by co-occurring ADHD symptoms. By pooling the results of 17 studies, Karalunas et al. (2014) obtained tentative evidence that, although high RTV is indeed associated with ASD, this is only in the presence of comorbid ADHD. This is further consistent with data from a general population study suggesting that RTV may predict ADHD traits beyond ASD traits (Truedsson et al. 2015).

Commonly, investigators have measured RTV with the standard deviation of RTs (SDRT). However, SDRT represents an overall phenomenon; to better understand the association of ADHD and ASD with RTV, it may be informative to decompose SDRT into its components that capture the extremely slow RTs within the individual’s performance (Leth-Steensen et al. 2000) or the periodic dynamics of their RTs (Castellanos et al. 2005). One such approach is the ex-Gaussian analysis, which separates RT distributions into their normal (Gaussian) and exponential (ex-Gaussian) parts. Analyses on RT data from participants with ADHD indicate that a set of infrequent, ultra-long RTs (the ex-Gaussian Tau) specifically contribute to increased RTV (Hervey et al. 2006). So far, two studies have directly compared children with ADHD and those with ASD in relation to ex-Gaussian parameters: in one study elevated Tau characterised ASD regardless of the co-occurrence of ADHD (Geurts et al. 2008), while the other study found elevated Tau in children with ADHD only and those with comorbid ADHD and ASD (Tye et al. 2016). The shorter task duration (3-min) in the first study (Geurts et al. 2008) might not have captured slower patterns of responses typically observed in ADHD in longer tasks such as that in the second study (~9-min) (Tye et al. 2016). Thus, experimental conditions might explain inconsistencies between studies.

A second approach uses frequency decompositions of RT data to identify the periodic patterns of RTV (e.g., cycles of ≥5 s) (Castellanos et al. 2005; Feige et al. 2013; Johnson et al. 2007). An early small study that examined slow RT fluctuations (15–20-s cycles) reported that increased RTV in these slow frequencies occurs in ASD groups and differentiates children with comorbid ASD-ADHD from those with ADHD only (Geurts et al. 2008). More recent data suggest that elevated RT fluctuations in a wide range of frequencies may be specifically associated with ADHD and not with ASD, and that RT fluctuations occurring in relatively rapid cycles (2–5-s) are elevated in children with ASD who show high ADHD symptoms (Adamo et al. 2013). Altogether, findings from ex-Gaussian and RT-fluctuation analyses suggest that RTV subcomponents may help find common and unique RTV profiles in ADHD and ASD, and ultimately guide the understanding of their underlying mechanisms.

Beyond the study of the RTV subcomponents, another approach that may inform on the underlying mechanisms of both disorders and their overlap is the investigation of the malleability of RTV impairments. Studies on population (Kuntsi et al. 2009, 2005) and clinical (Andreou et al. 2007; Epstein et al. 2011; Hervey et al. 2006) samples converge in indicating that task manipulations with rewards and faster stimulus presentation rates, alone or in combination, elicit greater RTV improvements in children with ADHD than in control children, and in relation to continuously measured ADHD symptoms. By directly comparing children with ADHD, children with ASD and those with co-morbid ADHD-ASD to control children, Tye et al. (Tye et al. 2016) found that greater improvements of overall and ex-Gaussian RTV measures in a fast-incentive condition were observed in the ADHD-only and co-morbid ADHD-ASD groups, and not in the ASD-only or control groups, suggesting that the RTV malleability could be specific to ADHD.

Overall, the available results on detailed RTV measures and their sensitivity to task manipulations have emerged from relatively small samples and mostly from clinical populations; no study to date has examined these detailed phenotypes in relation to both ADHD and ASD traits in non-clinical samples. Applying these analyses in an unselected general population sample avoids possible selection biases associated with clinic-referred or selected community samples, and provides a tool to capture detailed RTV impairments in relation to the full spectrum of ADHD and ASD symptoms. Further, examining ADHD and ASD traits is in line with recent proposals for transdiagnostic, neurobiologically grounded features that underlie the aetiology of psychopathology (Morris and Cuthbert 2012). Identifying the links of such features beyond the diagnoses of ADHD or ASD may therefore help further understand the underlying mechanisms of the observed clinical overlap between the disorders.

Here, we aim to extend initial reports of common and disorder-specific refined RTV components in ADHD and ASD in a large-scale study on a population sample of children. We perform frequency and ex-Gaussian decompositions of RT data, which previously indicated positive associations of overall RTV with the inattention and hyperactivity-impulsivity subdomains of ADHD, as well as with ASD social-communication difficulties (Kuntsi et al. 2009; Pinto et al. 2016). We first aim to investigate which frequency and ex-Gaussian RTV subcomponents are associated with ADHD symptoms (inattention and hyperactivity-impulsivity) and which with ASD symptoms (social-communication difficulties and repetitive-restricted behaviours and interests), using a ‘baseline’ slow task condition and a faster condition that allows measuring RTV indices in a range of slow and fast patterns. We then test whether the association of one trait with each RTV measure remains when controlling for both subdomains of the other trait, and whether ASD and ADHD symptoms have additive effects on RTV increases. Finally, we aim to investigate whether the RTV subcomponents’ malleability (improvement with faster event rate or incentives) differentiates between ADHD and ASD traits.

We show that, beyond what first appears a shared neurocognitive impairment of increased RTV between ADHD and ASD traits, specificity of this feature to ADHD traits emerges under closer inspection. Investigating frequency and ex-Gaussian RTV subcomponents in a large population sample of children, we found, first, that the refined RTV components were linked to ADHD traits and not to ASD traits. Second, although both ADHD and ASD social-communication traits were associated with the overall measure of SDRT, the association with ASD social-communications trait disappeared when controlling for ADHD traits, while association with ADHD traits remained when controlling for ASD traits. Third, a reward-induced improvement in RTV measures, indicating malleability, was only observed in relation to ADHD traits.

We found that the ex-Gaussian Tau was uniquely related to inattention and that the periodic RT fluctuations in slow and fast cycles showed specificity to inattention and hyperactivity-impulsivity. These findings are in line with the majority of prior studies on clinically diagnosed samples reporting increased amplitudes in these measures in participants with ADHD, but not those with ASD (Adamo et al. 2013; Johnson et al. 2007; Tye et al. 2016), although with modest effect sizes, as is common in general population samples where the use of the full range of scores detects modest effects. One previous study reported, however, elevated ex-Gaussian and frequency parameters in children diagnosed with ASD and those with comorbid ASD-ADHD but not in those with ADHD only (Geurts et al. 2008). As compared to the shorter duration (3-min) of the task used in Geurts et al. (2008), our tasks of ~8-min and ~9-min might have captured slower patterns of responses typically observed in relation to ADHD in longer tasks (Adamo et al. 2013; Johnson et al. 2007; Tye et al. 2016). The persistent association of Tau with inattention and the lack of an association of Tau with hyperactivity-impulsivity when controlling for ASD traits further suggest a closer association of this ex-Gaussian measure with inattention, and support previous hypotheses that the ultra-long, rare RTs captured by Tau may reflect lapses of attention (Leth-Steensen et al. 2000; West et al. 2002). Future work is warranted to further our understanding of how ASD traits affect the relationship of RTV measures with hyperactivity-impulsivity, and the pathophysiology underlying the detailed RTV measures. So far, only preliminary evidence has emerged for a direct association between the refined RTV measures and neural impairments, showing that relatively fast cycles in RTV parallel fluctuations in neural markers of attention allocation in healthy individuals (Adamo et al. 2015).

The lack of a significant association of Sigma with the behavioural traits in our study contrasts with three previous reports of elevated Sigma in ADHD and co-morbid ADHD-ASD in clinical samples (Geurts et al. 2008; Hervey et al. 2006; Tye et al. 2016), but parallels the results of another study that could not differentiate children with ADHD and control children using this index (Leth-Steensen et al. 2000). Such inconsistency across studies can be viewed in light of the results of a large meta-analysis of case-control studies on RTV in ADHD, which found that Sigma discriminates between ADHD and control groups with smaller effect sizes than Tau (Kofler et al. 2013). Increased RTV in individuals with ADHD is therefore more likely to reflect variability of the rare, abnormally long RTs, captured in Tau, rather than fluctuations in the normally distributed RTs, captured in Sigma (Kofler et al. 2013).

The only shared impairment between the ADHD and ASD traits, observed in the overall RTV measure of SDRT, was no longer associated with ASD social-communication difficulties when controlling for either ADHD trait, but remained significantly associated with both ADHD subdomains when controlling for ASD traits. The association of high SDRT in relation to ASD traits may therefore be explained by co-occurring ADHD symptoms, supporting previous evidence from clinical samples (Karalunas et al. 2014).

The investigation of the malleability of the high RTV provides another angle on potential specificity. Most previous research on the malleability of RT fluctuations has focused on the overall measure of SDRT in relation to ADHD (Cheung et al. 2017; Hervey et al. 2006; James et al. 2016). Conversely, Tye et al. (2016) extended this approach to ex-Gaussian RTV measures in an ADHD-ASD comparison, finding that the greater improvement of both overall and ex-Gaussian RTV measures under fast-incentive conditions were specific to ADHD, as such improvements were not observed in children with ASD only. Our findings confirm this observation in a population-based sample and further extend the findings to the frequency measures: when rewards were given, SDRT, Tau and the Slow-4 RT fluctuations decreased (i.e., improved) in relation to inattention, and Slow-5 RT fluctuations improved in relation to both ADHD subdomains, while no such effects emerged for the ASD traits. Together with prior evidence that cognitive performance is optimised in individuals with ADHD with the introduction of incentives (Andreou et al. 2007; Kofler et al. 2013; Kuntsi et al. 2009; Tye et al. 2016; Uebel et al. 2010), these findings support theories of atypical reward processes in ADHD, which in turn might help understand the neurobiology underlying the disorder (Luman et al. 2010). With our current and previous (Kuntsi et al. 2013b) results pointing to stronger associations of the RTV improvements with inattentive than hyperactive-impulsive traits, our findings further motivate future exploration of the neural correlates of RTV fluctuations in both ADHD subdomains. In our analyses, effects of faster event rate did not reach significance. We have previously reported how, in relation to ADHD diagnosis and trait, rewards tend to lead to a slightly greater SDRT improvement than fast event rates (Kuntsi et al. 2009; Uebel et al. 2010). Yet we have shown using quantitative genetic model-fitting analyses that both manipulations measure, to a large extent, the same underlying process (Kuntsi et al. 2013a). Further, jittered stimulus presentation has also previously shown to improve RTV measured as the ex-Gaussian Tau in children with ADHD (Lee et al. 2015), suggesting that response preparation may not be optimized with fast yet consistent event rates.

Our findings on the lack of positive effects from reward in relation to ASD traits are in agreement with reports of children with ASD benefiting less (Delmonte et al. 2012) or not at all (Scott-Van Zeeland et al. 2010) from the introduction of monetary rewards in other cognitive impairments compared to controls. For the clinician, these findings emphasise how, in the design of treatment protocols for attention impairments in children with neurodevelopmental disorders, different approaches may work with children with ASD than in children with ADHD, as the use of monetary or token incentives likely not have the expected reinforcing effects or might even worsen cognitive performance in these children. While we observed no association between ASD traits and improvement in the RTV measures following either rewards or a faster event rate, ASD restricted-repetitive behaviours and interests were significantly associated with a worsening in Tau following rewards. In reviewing this result, we consider emerging findings of aberrant temporal processing in children with ASD, who might integrate different stimuli into one single event over a longer window than normal controls (Foss-Feig et al. 2010). Accordingly, we tentatively speculate that an impaired integration of stimuli (i.e., incentives and target stimuli) might interfere with the processing of target stimuli, generating more variable responding. This motivates further investigations into ASD traits that use, for example, the high temporal resolution of electroencephalography to disentangle the neural basis of the effects of rewards on Tau.

The current study has some limitations. Behavioural ratings on ASD traits were collected approximately one year later than ADHD ratings and cognitive data, therefore potentially reducing the magnitude of observed associations of RTV measures with ASD traits. However, while in children with diagnosed ASD significant age-related increases in repetitive behaviours have been reported (Richler et al. 2010), autism traits have been reported to be stable over time in the general population (Gotham et al. 2012), limiting the potential effect of a lag between parent report and cognitive testing. The CAST scale provides a symptom count of the social-communication difficulties and the restricted-repetitive behaviours and interests, which may be a suboptimal measure of the full range of ASD traits in a population sample. Future research could benefit from examining the relationship of RTV indices with measures that capture the severity, rather than the presence or absence, of the ASD symptoms. An additional limitation is that the current study only focused on a population-sample of twins aged 7–10 years. Twins may not be representative of the general population in terms of mental health problems, as it has been suggested that twins might have an increased risk for ADHD compared to singletons (Levy et al. 1996). However, other studies have found little or no evidence for such differences (Gjone and Novik 1995; Robbers et al. 2010; Simonoff et al. 1997), thus this unlikely affects the interpretation of our results. Nevertheless, future research will need to establish the generalizability of our findings across a wider age range.

In sum, in an investigation of ADHD and ASD traits in a large population sample of children, detailed ex-Gaussian and frequency RTV indices, as well as reward-induced improvements in the RTV measures, show specificity to ADHD traits. As reports of increased RTV are not limited to ADHD and ASD (Kofler et al. 2013), our findings support the application of the ex-Gaussian and frequency approaches, and of reward manipulations, to further cross-disorder investigations. For the clinician planning effective behavioural interventions, our findings indicate that attentional fluctuation in children with high ASD traits may be due to co-occurring ADHD traits and emphasise how the effectiveness of rewards does not generalise from ADHD to ASD traits.