Emotional decision-making in autism spectrum disorder: the roles of interoception and alexithymia

We initially established that, in total, 0.31 and 0.80 % of trials were excluded for control and autistic participants, respectively. There was no significant group difference in the number of excluded trials, t(38) = 1.43, p = 0.17, d = 0.45, 95 % CI for d[−0.17, 1.08].² There was also no significant group difference between the ASD and control groups in IA (control: M = 70.99, SD = 20.40, ASD: M = 65.54, SD = 25.85, t(38) = 0.74, p = 0.46, d = 0.23, 95 % CI for d[−0.39, 0.85]). However, autistic participants significantly outperformed the control group on the TE procedure (control: M = 72.40, SD = 19.96, ASD: M = 85.24, SD = 11.55, t(38) = 2.49, p = 0.017, d = 0.79, 95 % CI for d[−0.14, 1.43]).

FEs were evident in both individuals with and without ASD, as demonstrated by the greater percentage of trials in which participants chose the gamble option in the loss than gain frame (Fig. 2a). However, the size of the FE (i.e., the difference between gambling in the Loss compared with Gain Frame) was substantially smaller and less variable in the ASD compared with the control group (Fig. 2b).

These data were analyzed using mixed model ANOVA with frame (gain, loss) as the within-subjects factor and group (ASD, control) as the between-subjects factor. There was no main effect of group, F(1, 38) = 0.83, p = 0.37, η_p ² = 0.02 [90 % CI = 0, 0.05], indicating there was no overall group difference in the number of trials on which they chose to gamble. There was a large and significant effect of frame, F(1, 38) = 29.59, p < 0.001, η_p ² = 0.44 [90 % CI = 0.23, 0.57], which confirmed the effectiveness of the framing manipulation. Crucially, the group × frame interaction reached significance, F(1, 38) = 6.35, p = 0.016, η_p ² = 0.14 [90 % CI = 0.02, 0.31], which indicated that susceptibility to the framing manipulation was different between those with and without ASD. Planned comparisons confirmed both control t(19) = 4.42, p < 0.001, d = 1.43, 95 % CI for d[0.71, 2.14], and autistic t(19) = 3.34, p = 0.003, d = 1.08, 95 % CI for d[0.39, 1.76], participants gambled significantly more in the loss compared with the gain frame. Importantly, however, the size of the FE was significantly larger in the control group than in the ASD group, t(38) = 2.52 p = 0.016, d = 0.80, 95 % CI for d[0.15, 1.44]. This result supported our first hypothesis.

Data from catch trials were inspected to examine whether the participants were engaged in the task and to assess if the ASD participants were responding to monetary incentives in a manner that was comparable to the control group. The analysis of catch trials also determined whether overall risk tendency was matched across groups. Only one participant in each group showed an atypical pattern of responding, indicative of extreme aversion to risk (i.e., choosing not to gamble on trials when there was a 95 % probability of keeping the initial amount compared to 50 % probability of a win if they chose the “sure” choice). However, the pattern of significance detailed above remained when both individuals were excluded from analysis (e.g., group × frame interaction; F(1, 36) = 6.38, p = 0.016, η_p ² = 0.15 [90 % CI = 0.02, 0.32]). Furthermore, re-including all participants, and in line with previous data [7], the number of catch trials on which participants gambled was not different between groups, t(38) = 0.94, p = 0.35, d = 0.30, 95 % CI for d[−0.33,0.92]. The critical group × frame interaction also remained significant after including individual performance on catch trial data as a covariate in the analysis, F(1, 37) = 5.40, p = 0.026, η_p ² = 0.13 [90 % CI = 0.01, 0.29].

In line with the existing literature (e.g., [16]), there was an association between IA and the TAS-20, across groups (r = −0.58, p < 0.001), in the control group (r = −0.60, p = 0.005), and ASD group (r = −0.61, p = 0.004). Supporting our second hypothesis, after collapsing across groups, the size of the FE was negatively correlated with both autistic traits (r = −0.43, p = 0.005) and alexithymia scores (r = −0.36, p = 0.023). Conversely, there was a positive correlation between IA and the FEs (r = 0.37, p = 0.019) but no such relationship with TE ability (r = 0.02, p = 0.92).

We conducted moderation analyses to investigate whether the relationship between interoception and the FE and the association between alexithymia and the FE (i) significantly varied as a function of group and (ii) held after controlling for potentially confounding variables. Participant age, IQ, state, and trait anxiety scores (see Table 1), TE ability, group (ASD, control), and alexithymia (TAS-20 scores) were entered as predictor variables into step 1 of a hierarchical regression. Group was the only significant predictor of the FE (β = 0.43, t = 2.43, p = 0.021), whereas alexithymia (β = −0.27, t = −1.68, p = 0.10) and the other variables failed to reach significance (other ps > 0.16). Together, step 1 explained 32.90 % of variance in the FE. When the alexithymia × group term was added to the model in step 2, the results differed such that in addition to group (β = 0.41, t = 2.49, p = 0.019), both state (β = −0.52, t = −2.29, p = 0.029) and trait (β = 0.48, t = 2.05, p = 0.049) anxiety scores were significant predictors of a larger FE. Again, alexithymia and other confounding variables remained non-significant (ps > 0.34). Most importantly, however, the predictor coding the interaction between alexithymia and group was a significant predictor of the FE (β = −0.40, t = −2.27), significantly increasing the variance accounted for by 9.80 %, F(1, 30) = 5.14, p = 0.031. This analysis confirmed that the relationship between alexithymia and the FE was moderated by group.

Alexithymia and IA were highly correlated. In order to avoid problems with multicollinearity, a second moderation analysis was conducted with IA replacing alexithymia in steps 1 and 2, and accordingly, an IA × group interaction term in step 2. Group was a significant predictor of the FE (β = 0.41, t = 2.43, p = 0.021), as was IA (β = 0.33, t = 2.13, p = 0.041). The other variables in the first step failed to reach significance (other ps > 0.23) and altogether, step 1 explained 36.1 % of variance in the FE. In step 2, both group (β = 0.41, t = 2.57, p = 0.015) and IA (β = 0.40, t = 2.68, p = 0.012) remained as significant predictors. More importantly, the interaction between these variables was also a predictor of the FE (β = 0.32, t = 2.27), significantly increasing the variance accounted for by 9.40 %, F(1, 31) = 5.17, p = 0.030, and thus confirmed that the relationship between interoception and the FE was different between groups. These patterns of results supported our third hypothesis, which also held when both moderation analyses were repeated without including age, IQ, TE ability, and anxiety scores (this analysis which just includes the variables of key interest may be more appropriate given the sample size).

Correlational analyses showed that both the relationship between alexithymia and the FE and the association between interoception and the FE were only observed in, and therefore being driven by, the control group (Fig. 3). Greater IA (r = 0.55, p = 0.012) and lower alexithymia scores (r = −0.49, p = 0.029) were associated with a larger FE. The FE was not associated with any of the other demographic or potentially confounding variables (all ps > 0.11). In contrast, there was no relationship between alexithymia and the FE (r = 0.03, p = 0.89), or interoception and the FE (r = 0.14, p = 0.57) in the ASD group; however, there was a negative association between age and the size of the FE (r = −0.69, p = 0.001).

The present study investigated whether individuals with ASD exhibit enhanced logical consistency, i.e., smaller FEs, after accounting for alexithymia and interoception. More generally, we sought to investigate the relationship between interoception and decision-making which has seldom been investigated in typical individuals and never before in a clinical group. To these ends, a well-established measure of the FE was used, together with a widely used measure of interoception. Replicating a previous study [7], FEs were significantly smaller in autistic individuals compared to the non-autistic control group. Importantly, these results were found even when the control group was matched for alexithymia, suggesting that the enhanced logical consistency seen in the ASD group was due to ASD itself, rather than co-occurring alexithymia.

Moderation and within-group analyses demonstrated a striking difference between the relative impact of alexithymia and interoceptive ability on the ASD and control groups. For individuals without ASD, IA and alexithymia both significantly predicted the size of the FE, such that better interoception (and lower trait alexithymia) was associated with a larger FE. This result is in accordance with the model forwarded by De Martino and colleagues [2] in which emotional information influences decision-making in neurotypical individuals, and the extent to which one can accurately perceive emotional information/arousal determines the influence it has on decision-making (see also [13]). In contrast, alexithymia and interoception did not predict susceptibility to the FE in those with ASD. This differential pattern of results in people with and without ASD indicates the use of different strategies when making judgments: one strategy involving gist-based, fuzzy, intuitive processes used by non-autistic individuals, and another strategy which recruits rule/verbatim-based analysis which is recruited by individuals with ASD (see [59] for a neurodevelopmental perspective). These results are consistent with the idea that, instead of using interoceptive or emotional information, and regardless of whether they have access to these signals, individuals with ASD use a rule-based strategy which results in a smaller FE at the group level (see [60]). These results, in line with our predictions, are consistent with recent research showing divergent roles of autism and alexithymia across high-level judgment and decision-making tasks that invoke emotional processing [37‐39].

De Martino et al. [7] proposed that “although this impairment in processing contextual emotional information protects ASD subjects from the framing bias, leading to more consistent behavior in situations of risk, it may come at a cost of the social, emotional, and behavioral deficits that characterize the condition” (p. 10,749). The present study is consistent with this proposal, insofar as it was clearly shown that autistic individuals with social atypicalities are indeed less susceptible to the FE. However, we found that IA and alexithymia were not correlated with the size of FE in ASD. This indicates that use of rule-based strategies, leading to a smaller FE, is less related to social-emotional atypicalities in ASD and may be more closely related to the rigid and repetitive behavioral difficulties in this condition. The (over) reliance on rule-based strategies might lead to difficulties in day-to-day functioning (e.g., insistence on sameness, potentially leading to difficulties gaining or maintaining employment in certain job roles). Equally, however, the reduced use of interoceptive signals—and thereby enhanced logical consistency in ASD—may confer an advantage in many situations where emotional information would otherwise interfere with the optimal outcome (e.g., gambling, financial investments, and risk assessments). More generally, neurocognitive strengths shown by individuals with ASD are both under researched, and/or commonly interpreted as impairments, yet the findings from the present study highlight that certain cognitive processes in ASD may, depending on the situation, benefit as well as impair day-to-day functioning (see also [61‐63]). On a related note, it is to be emphasized that, while the current study replicates and extends De Martino and colleagues’ results [7], both studies contained high functioning adult ASD samples which limits the generalizability of these findings. Future research on emotional decision-making, alexithymia, and interoception in lower functioning adults and children with ASD are required, and such research will more comprehensively elucidate the causes and consequences of atypical decision-making on individuals across the autism spectrum.

It was not possible to perform neuroimaging in the present study; however, the current results also accord with existing neuroscience data on the FE. The insula has been shown to be involved in emotional decision-making, interoception, and alexithymia [32, 34‐36, 64, 65], which together, is likely to underpin the pattern of results that we observed in the control group. Similarly, the strength of prefrontal-amygdala connectivity, known to be atypical in ASD [66‐68], is thought to be positively correlated with the size of the FE. This supports the idea that emotional information has a smaller influence on decision-making in autistic individuals and may explain the smaller FE observed in the ASD group.

When considered as a whole, extant research suggests that both ASD itself and co-occurring alexithymia can contribute to atypical performance in ASD in a task-dependent manner. When interoceptive or emotion processing is necessary and sufficient for accurate performance (e.g., tasks assessing interoception, emotion perception, or empathy), then any “ASD impairment” is likely due to alexithymia, such that individuals with ASD without co-occurring alexithymia are unimpaired [29‐33]. However, for socio-cognitive tasks that typically involve, but are not necessarily reliant upon, emotion processing (moral judgments, decision-making, Theory of Mind), then individuals with ASD may perform the tasks atypically (demonstrating either impairments or superior performance) because they use a non-affective strategy to complete the task. Performance on these latter tasks would therefore be determined by the presence of an ASD diagnosis, rather than co-occurring alexithymia.

In addition to the primary aim of the study, these data provide evidence for intact IA in ASD. This accords with a recent study in which ASD and control groups were matched for alexithymia: In line with Shah and colleagues’ data [16], the present study indicated that alexithymia, not autism, is associated with atypical interoception.³ Previous studies assessing interoceptive ability in ASD which do not control for alexithymia have produced inconsistent results [8‐12], but a clearer picture is obtained when this important source of variance is accounted for [16, 17, 19, 69] A higher incidence of alexithymia in many psychiatric and neurological disorders supports the inclusion of alexithymia questionnaires when studying interoception in any clinical group, in order to decompose any reduced IA into that accounted for by the condition itself and that accounted for by co-occurring alexithymia [16, 17]. More generally, interoception research is typically conducted in high functioning adults; therefore, there is a need for interoception research including a wider range of individuals to firmly establish the existence of interoceptive impairments (if any) in ASD and other clinical groups.

An enhanced TE ability in ASD relative to the control group was also observed in these data. Time perception was not the focus of the current study, and the TE task was designed as a control rather than a sensitive measure of time perception ability. These factors mean that we do not wish to draw strong conclusions from this result, but the finding of superior time perception in autistic adults accords with reports of enhanced time perception in children with ASD [70]. Similarly, the present study was designed to investigate the relative contributions of autism, alexithymia, and interoception to the FE, while controlling for anxiety. Nevertheless, in line with existing research, increased anxiety was associated with larger FEs [6, 55] after accounting for autism and alexithymia. This warrants further investigation in future research with larger samples comprising sufficient variance to fully address the interrelationship between ASD, anxiety, alexithymia, interoception, and emotional decision-making. Finally, although the negative relationship between age and the size of the FE in ASD was unexpected, it seems likely that increased time living with the cognitive features of ASD—such as a bias towards an analytical rule-based strategy when making judgments—may increase the impact of ASD on reduced susceptibility to FEs. Although not observed in this study, there are reports of increased FEs with age in typical populations, suggesting a developmental approach to FEs in ASD, and other clinical groups may prove fruitful [59].

In summary, the present study replicates De Martino et al.’s finding [7] of enhanced logical consistency in ASD. It was shown that this is unlikely to be due to co-occurring alexithymia—instead evidence suggested that interoceptive ability and alexithymia were unrelated to the FE in individuals with ASD. In contrast, both alexithymia and interoception were associated with the size of FEs in neurotypical individuals, in line with contemporary suggestions that interoceptive and emotional signals guide decision-making under uncertainty [35, 36]. Together, these results reinforce the idea that atypical decision-making is a robust feature of ASD, while highlighting the importance of alexithymia when studying interoception and emotional decision-making in clinical and non-clinical samples.