In total, 29 autistic adults and 28 neurotypical individuals (NT) were recruited for this study. Autistic participants were selected from the adult outpatient clinic of the Psychiatric Unit at the University Hospital of Strasbourg, France. To be eligible for participation, autistic individuals had to meet the specific inclusion criteria: (i) Experiencing suicidal thoughts and/or engaging in impulsive behaviours (such as alcohol, drug, or pharmaceutical abuse), non-suicidal self-injury, or have had a documented history of suicidal ideation and/or attempts within the 12 months leading up to the recruitment; (ii) Having difficulties in regulating emotions, as assessed using the Difficulties in Emotion Regulation Scale – DERS (6), with a cut-off score exceeding 95; (iii) Having an intellectual quotient above the cut-off score of 70, as evaluated by the Wechsler Adult Intelligence Scale-IV - WAIS-IV (105); (iv) Presenting no co-occurring psychotic disorders. The control group was recruited by responding to an announcement posted by the University Hospital of Strasbourg. To be eligible for the study, NT individuals had to meet the following conditions (i) No reported difficulties with regulating their emotions, as assessed by the DERS (87); (ii) No history of psychiatric or neurological disorders; and (iii) Not currently using any psychiatric medications.
The participants’ ages ranged from 18 to 67 years, and the two groups were similar in terms of sex, age, and educational background. Table 1 provides a detailed overview of the participants’ demographics. As shown in Table 2, 17% of the ASC group did not present any co-occurring Axis 1 disorder. Among participants with one or more co-occurring psychiatric or neurodevelopmental diagnosis, anxiety disorders were reported in 46% of cases, affective disorders (i.e., major depression or bipolar disorder) in 58%, eating disorders in 8%, and attention-deficit hyperactivity disorder in 33%. Psychotropic treatments were prescribed in 86% of autistic participants, including anxiolytics (32%), antidepressants (72%), neuroleptics (20%), psychostimulants (24%) and antipsychotics (36%). As part of the exclusion criteria described previously, NT group did not have any psychiatric diagnoses and were not taking any medication.
Procedure
The present research was conducted as part of a larger study that included additional tasks, which are not detailed here. All participants were invited to the laboratory, where they received detailed information about the study’s objectives and provided written informed consent. Following this, the first author conducted an in-depth demonstration of the EMA protocol, where participants promptly responded to EMA questions under the guidance of the experimenter. After participants had a clear understanding of the EMA procedure, they completed a comprehensive battery of self-reported assessments. These assessments covered domains such as autism severity, emotion dysregulation, alexithymia, depression, anxiety, camouflaging, borderline symptoms, and sensory sensitivity (for details, please see ‘Materials’). To enhance the study’s precision and account for possible IQ discrepancies between the two participant groups, we administered the Matrix Reasoning and Information subtests of the Wechsler Intelligence Scale - WAIS-IV (105). Finally, a 15-minute post-experimental paradigm was proposed to all participants to gain a deeper knowledge of the user’s experience and to verify possible stress-related effects.
Data for this study was collected between May 2022 to July 2023. All participants were compensated with €200 for their participation. The regional ethics committee of the East of France approved this study as a preliminary step for a randomized control trial on dialectical behaviour therapy (DBT) and ED (Reference: SI 21.01.21.41923).
Materials
Diagnostic assessments
Autistic participants were diagnosed using the Autism Diagnostic Interview-Revised - ADI-R (106) and the Autism Diagnostic Observation Schedule-2 – ADOS-2 (107, 108). Their mental health status was evaluated by licenced clinical psychologists or psychiatrists, adhering to the criteria outlined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition - DSM-5 (20). Regarding NT participants, their psychological well-being and emotion regulation strategies were evaluated to ensure the absence of mental health issues, through a 20-minutes semi-structured phone interview, and the completion of the Difficulties in Emotion Regulation Scale (DERS) (6).
Ecological Momentary Assessment protocol
The EMA protocol required participants to respond to 12 semi-randomized smartphone text messages per day for a continuous 7-day period. Each message contained a prompt asking the participants to click on an attached link to participate in a survey. The survey assessed their current emotional state from a predefined list of 12 options: 'joy’, ‘anxiety’, ‘anger’, ‘interest’, ‘shame’, ‘disgust’, ‘calm’, ‘sadness’, ‘surprise’, ‘guilt’, ‘I have an emotion that I cannot name’, and ‘I have no emotion’. To prevent participants from selecting ‘I have an emotion that I cannot name’, or ‘I have no emotion’ simply because they could not find their specific emotion on the list, we included a wide spectrum of emotions. This list was compiled based on previous studies that used EMA to gain a deeper understanding of the emotional dynamics in clinical groups with ED and difficulties in labelling emotions, such as BPD (e.g., (109)), as well as based on studies on basic emotions (110). Participants were instructed to report all emotions they were experiencing at the time. For each survey, participants rated the intensity (i.e., the dimension of ‘Arousal’) of their selected emotional state. Additionally, they were asked about their emotion regulation needs, specifying whether they intended to decrease, maintain, or increase their emotional state (referred to as ‘Need for ER’). The main objective of this study was to improve understanding of the factors contributing to ED. To accomplish this goal, we used an EMA query to evaluate participants’ perceived self-efficacy in ‘Emotion Control’. Finally, to reduce the impact of potential confounding variables such as fatigue and stress derived from social interactions, we included the questions outlined in Table 3 in the survey.
Text message scheduling was customized to each participant’s circadian rhythm and occurred within specified timeframes (8 am − 8 pm, 9 am − 9 pm, 10 am − 10 pm, or 11 am − 11 pm). Participants received a total of 84 text messages over 7 days, with intervals ranging from 43 to 84 minutes between messages (mean = 65 minutes). They were instructed to respond at their earliest convenience by clicking on the link generated using Qualtrics software (Qualtrics, Provo, UT). Text messages were scheduled using the Textra app on a Samsung Galaxy XCover 5 devices. In addition to the scheduled text messages, participants were also given the option to report their emotional state if they were experiencing strong emotions. This spontaneous self-assessment survey included the same set of questions as the initial survey and was accessed via a separate Qualtrics survey.
Additional self-report measures
To control for clinical variables related to autism severity, ED, alexithymia, levels of depression, anxiety, borderline traits, or autism core traits, such as camouflaging and sensory sensitivity, participants completed a list of self-reported measures at baseline, as detailed below.
The Autism-Spectrum Quotient - AQ (111) is a widely used self-administered questionnaire designed to evaluate autistic traits. Comprising 50 items, the questionnaire assesses five different areas: social skills, attention switching, attention to detail, communication, and imagination. Each item response is categorized dichotomously (i.e., definitely, or slightly agreeing; definitely, or slightly disagreeing), with values of 1 denoting the presence of mild or strong autistic traits, while 0 signifies their absence. Aggregate scores range from 0 to 50, with higher values indicating greater levels of autistic traits. In the current study, respondents completed the French-Canadian version – AQ-F (112). Internal consistency, as assessed by Cronbach’s alpha, yielded good values for both autistic (α = 0.84) and NT (α = 0.87) cohorts.
The Difficulties in Emotion Regulation Scale - DERS (3) is a popular self-assessment tool, consisting of 36 items designed to measure emotion dysregulation. DERS evaluates challenges within six dimensions: (a) Nonacceptance (i.e., Non-Acceptance of Emotional Responses), (b) Goals (i.e., Difficulties Engaging in Goal-Directed Behaviour), (c) Impulse (Impulsive Control Difficulties), (d) Awareness (i.e., Lack of Emotional Awareness); (e) Strategies (i.e., Limited Access to Emotion Regulation Strategies); and (f) Clarity (i.e., Lack of Emotional Clarity). Participants rate the content of the items on a Likert-scale ranging from 1 (almost never) to 5 (almost always), and higher scores denote heightened emotion dysregulation. Notably, DERS has been validated for use in both autistic adolescents and adults (87). In this study, we used the French version (113). Internal consistency analyses produced good results for the ASC group (α = 0.89) and excellent values for neurotypical participants (α = 0.98).
The General Alexithymia Factor Score - GAFS-8 (88) constitutes a specific collection of eight items extracted from the Toronto Alexithymia Scale - TAS-20 (89). This abbreviated version is considered a more adequate model fit to measure challenges in recognizing and labelling emotions (known as alexithymia) in individuals situated within the autism spectrum. Respondents rate each item on a five-point Likert-type scale, where ‘1’ corresponds to ‘strongly disagree’ and ‘5’ corresponds to ‘strongly agree’. In this study, the GAFS-8’s cumulative score was computed by summing up the responses to the specific items: 1, 2, 6, 9, 11, 12, 13, and 14, as designed within the French version of the TAS-20 (114). Adequate internal consistency was affirmed for both autistic (α = 0.80) and NT (α = 0.83) groups.
The Beck Depression Inventory, 2nd edition - BDI-II (115) is the most widely used self-report questionnaire for assessing depression severity. The tool includes 21 items with statements assessing for symptoms experienced over the past week. Each item is rated on a Likert scale with response options spanning from 0 (not at all) to 3 (severely). The French adaptation of the BDI has been found to demonstrate good validity and reliability and was used for this study (116). Internal consistency assessment yielded acceptable results for ASC (α = 0.78) and favourable values for NT (α = 0.86).
The Beck Anxiety Inventory – BAI (117) is a self-assessment tool comprising 21 items designed to capture anxiety symptoms. Respondents evaluate each item on a 4-point Likert Scale, ranging from 0 (indicating no presence of the symptom) to 3 (representing severe symptom severity). Total scores are calculated by summing item scores, resulting in a possible range from 0 to 63. For this study, the validated French version of the BAI (118) was employed. Notably, internal consistency was observed to be high for both the ASC group (α = 0.91), and the NT group (α = 0.87).
The Short Version of the Borderline Symptom List - BSL-23 (119) is a self-administered 23-item questionnaire that measures features associated with Borderline Personality Disorder (BPD). Items range from 0 (not at all) to 4 (very strongly), and the score is computed by summing the valid items, dividing by 23. For the present study, we used the French validated version of the BSL-23 (120). Notably, the internal consistency of this instrument proved to be excellent for both the ASC group (α = 0.94) and the NT group (α = 0.95).
The Camouflaging Autistic Trait Questionnaire - CAT-Q (121) is a self-report measure designed to assess social camouflaging behaviours, across both autistic and non-autistic populations. The questionnaire consists of 25 statements, organized into a three-factor structure: (a) Compensation (i.e., compensating behaviours in social situations), (b) Masking (i.e., concealing autistic-like traits), and (c) Assimilation (i.e., attempting to integrate with others). Items are scored on a 7-point Likert-type scale, spanning from 1 (strongly disagree) to 7 (strongly agree). A total score can be calculated by summing all items, and can range from 25 to 175, with higher values indicating greater levels of camouflaging. The French validated version of the CAT-Q was used in the current study (122). Internal consistency results varied from good within the ASC group (α = 0.85) to excellent in the NT group (α = 0.90).
The Glasgow Sensory Questionnaire - GSQ (123) is a 42 items self-report instrument, measuring hyper – and hyposensitivity in the general population. Respondents rate frequency of sensory events experienced through seven modalities: vision, audition, olfaction, taste, touch, proprioception, and vestibular sensations. The 5-point Likert scale ranges from 0 (never) to 4 (always), with total scores spanning from 0 to 168. In our study, participants completed the French version of the GSQ validated for use in adults with a high Autism-Spectrum Quotient. (124). For our screening, internal consistency was excellent for autistic participants (α = 0.91) and good for the NT group (α = 0.84).
Data Tranformation
To simplify the subsequent analysis of the data for the EMA protocol, the 12 emotional responses were grouped based on their valence and labelling. Therefore, the initial 10 emotion options were categorized into a variable called ‘Emotional Valence’ : ‘Negative Emotions’ (i.e., unpleasant emotions), including ‘anxiety’, ‘anger’, ‘shame’, ‘disgust’, ‘sadness’, and ‘guilt’, were coded as ‘-1’; ‘Positive Emotions’ (i.e., pleasant emotions) such as ‘joy’, ‘interest’, and ‘calm’, were coded as ‘1’; participants were given the opportunity to rate all the emotions they experienced. Therefore, responses that included two opposing emotions (e.g., ‘anxiety’ and ‘joy’) were classified into a third sub-category named ‘Conflicting emotions’ and labelled with the value ‘2’. The categorization of the neutral emotion, ‘surprise’, was dependent on other responses. If other choices were negative, ‘surprise’ was classified as ‘-1’, and if other choices were positive, it was classified as ‘1’. In cases where ‘surprise’ was the only choice, a value of ‘-1’ was assigned if participants wished to decrease the ‘Arousal’ of their emotion, and ‘1’ was assigned if they opted to maintain or increase the emotion (refer to Question 4, Table 3). Under the categorical variable ‘Emotion Labelling’, the sub-category ‘Absence of Emotion Labelling’ includes the codes’0’ for ‘I have no emotion’ and ‘2’ for ‘I have an emotion that I cannot name’. All other emotions were assigned the value ‘1’, indicating the sub-category ‘Emotion Labelling’.
As detailed in Table 3, Question 4, ‘Need for ER’ categorical variable designated three options: need to maintain, increase, or decrease the current emotional state. Finally, ‘Social Interaction’ query represented another categorical variable, where the six options presented in Table 3, Question 6 were combined into two sub-categories: ‘0’ indicating the absence of any direct social interaction, and ‘1’ signifying the occurrence of at least one direct social interaction within the past 15 minutes. All remaining continuous variables were assessed using a Likert scale ranging from 0 to 10, including the intensity of the experienced emotion (‘Arousal’), ‘Emotion Control’, and ‘Fatigue’ (refer to Table 3).
Table 3
Questions and Response Options in the Ecological Momentary Assessment Protocol
|
Questions
|
Response Options
|
1.
|
Please select one or more options from the list below to characterize your current emotions.
|
Multiple Choice a
1. Joy / Enthusiasm / Happiness
2. Anxiety / Fear / Worry
3. Anger / Irritation / Annoyance
4. Interest / Curiosity / Attraction
5. Shame / Embarrassment / Burden
6. Disgust / Aversion
7. Serenity / Calm / Relaxation
8. Sadness / Grief
9. Surprise / Astonishment
10. Guilt / Remorse / Regret
11. I have an emotion I cannot name
12. I have no emotion
|
2.
|
On a scale ranging of 0 to 10, please rate the intensity of the selected emotional state.
|
Likert scale (0–10)
0 = My emotion is not intense at all.
10 = My emotion is very intense.
|
3.
|
On a scale ranging of 0 to 10, please rate your perceived level of control over the selected emotional state.
|
Likert scale (0–10)
0 = I have no control over my emotions.
10 = I am completely in control of my emotions.
|
4.
|
Do you prefer to regulate the selected emotional state?
|
Multiple Choice
1. I would prefer to maintain my current emotional state.
2. I would prefer to increase my current emotional state.
3. I would prefer to decrease my current emotional state.
|
5.
|
On a scale ranging of 0 to 10, please rate the extent of your fatigue.
|
Likert scale (0–10)
0 = I am not tired at all.
10 = I feel very tired.
|
6.
|
Have you had any social interactions in the last 15 minutes?
|
Multiple Choice
1. I had no interactions; I was alone and in a closed environment (e.g., in a room).
2. I had no interactions; I was alone and in an open environment (e.g., on the street).
3. I had no interactions; Other people were present in the same environment.
4. I had a direct social interaction (e.g., a conversation).
5. I had a virtual social interaction (e.g., an online meeting).
6. I had an indirect social interaction (e.g., text messages).
|
Note. a Reflects that participants had the flexibility to choose more than one answer option. |
----------------------------insert Table 3----------------------------------------
Data Analysis and Statistical Procedure
Jamovi (125) and R (126) were used for all statistical analyses. First, descriptive statistics examined frequencies of survey responses as well as differences between the ASC and NT groups in terms of demographic, clinical, and EMA variables such as ‘Need for ER’, and ‘Social Interactions’. Independent t-tests, and Mann Whitney U tests were calculated for continuous variables, and chi-square (χ2) was used for nominal variables. To test our first hypothesis, which suggests that autistic adults would experience higher frequencies of negative emotions and lower frequencies of positive emotions (H1a), and that they would have a higher absence of emotion labelling (H1b), we calculated chi-square tests (χ2).
To assess further hypotheses regarding the factors contributing to reduced ‘Emotion Control’, such as demographics, co-occurring disorders, ASC-related traits, negative emotions, and alexithymia, and to account for the interdependence of data at the individual level, we applied multilevel modelling (MLM). We followed recommended practices for EMA and employed a restricted maximum likelihood (REML) estimation method (127). This hierarchical structure allowed us to explore variations in ‘Emotion Control’ across different levels, including occasions within a day, days within a person, and among different individuals. An additive hierarchical multiple regression framework was used to create models. Subsequently, nested models with additional parameters were compared to the previous model using Akaike Information Criterion (AIC) and log likelihood ratio tests.
Before examining hypotheses H2) and H3), empty means models with random intercepts were used to partition the variability in ‘Emotion Control’ and to determine the correct number of levels for the MLM. Consistent with prior work (128), we used Intraclass Correlation Coefficient (ICC) and log likelihood ratio tests to compare the non-nested and two-level models. Our data had a hierarchical structure, with days (level 1) nested within individuals (level 2).
To control for the influence of demographic, group status (i.e., ASC vs NT), clinical variables related to common co-occurring disorders, ASC trait characteristics, fatigue, and social interactions, these variables were added to the model as between-person fixed effects. To test hypothesis H2), that higher levels of intense negative emotions would be related to reduced emotion control, especially within the ASC group, Arousal, ‘Emotional Valence’ and their interaction with the group status (ASC vs NT) were added to the model. Finally, to assess H3) that increased difficulties in emotion labelling (‘I have no emotion’ and ‘I have an emotion that I cannot name’) would be negatively associated with reduced ‘Emotion Control’, particularly in autistic individuals, this categorical variable and its interaction with the group was added to the model.
To improve interpretability and reduce multicollinearity, particularly when dealing with interaction effects, continuous between-person predictors and outcome variable were centred, while categorical variables were not centred (129). The final multi-level regression model aims to predict ‘Emotion Control’ score EmCont, ranging from 0 to 10, based on a combination of predictor variables and their interactions as in Eq. 1:
Here, the terms (1 | D) and (1 | ID) denote random intercepts for the groups D and ID, respectively. Indeed, the model allows for variability in the intercepts across different levels and subjects. Variable D, ranging from 1 to 7, indicates the day of the experiment; the ID is the identification code of the subject. The terms A: G, V: G, L: G represent the interactions between variable G and the predictors A, V and L respectively. The predictors are described in Table 4.
Table 4
Description of predictors in Eq. 1.
Predictors
|
Description
|
Levels
|
Age
|
Age
|
-
|
Sex
|
Sex – F/M
|
Female = 3; Male = 1.
|
G
|
Group – ASC/NT
|
ASC = B; NT = A.
|
AQ
|
Autism-Spectrum Quotient
|
-
|
GAFS
|
The General Alexithymia Factor Scale
|
-
|
BDI
|
Beck Depression Inventory
|
-
|
BAI
|
Beck Anxiety Inventory
|
-
|
CATQ
|
The Camouflaging Autistic Trait Questionnaire
|
-
|
BSL23
|
Borderline Symptom List, short form
|
-
|
GSQ
|
The Glasgow Sensory Questionnaire
|
-
|
SocInt
|
Social Interactions
|
Presence = 1; Absence = 0.
|
F
|
Fatigue
|
-
|
W
|
Weekday / Weekend
|
Weekday = 0; Weekend = 1.
|
A
|
Arousal
|
-
|
V
|
Valence of Emotions:
− Negative
− Conflicting
− Positive
|
− Negative Emotions = -1.
− Positive Emotions = 1.
− Conflicting Emotions = 2.
|
L
|
Emotion Labelling:
− ‘I have no emotion’
− ‘I have an emotion that I cannot name’
− Emotion Labelling
|
− ‘I have no emotion’ = 0;
− ‘I have an emotion that I cannot name’ = 2;
− Emotion Labelling = 1.
|
Note. The symbol ‘-’ denotes that the variable is continuous. |