Stability and change in the interpretation of facial emotions in fetal alcohol spectrum disorders from childhood to adolescence

Abstract

Background

The ability to identify and interpret facial emotions plays a critical role in effective social functioning, which may be impaired in individuals with fetal alcohol spectrum disorders (FASD). We previously reported deficits in children with fetal alcohol syndrome (FAS) and partial FAS (PFAS) on the “Reading the Mind in the Eyes” (RME) test, which assesses the interpretation of facial emotion. This follow‐up study in adolescents was designed to determine whether this impairment persists or represents a developmental delay; to classify the RME stimuli by valence (positive, negative, or neutral) and determine whether RME deficits differ by affective valence; and to explore how components of executive function mediate these associations.

Methods

The RME stimuli were rated and grouped according to valence. Sixty‐two participants who had been administered the RME in late childhood (mean ± SD = 11.0 ± 0.4 years) were re‐administered this test during adolescence (17.2 ± 0.6 years). Overall and valence‐specific RME accuracy was examined in relation to prenatal alcohol exposure (PAE) and FASD diagnosis.

Results

Children with FAS (n = 8) and PFAS (n = 15) performed more poorly on the RME than non‐syndromal heavily exposed (HE; n = 19) and control individuals (n = 20). By adolescence, the PFAS group performed similarly to HE and controls, whereas the FAS group continued to perform more poorly. No deficits were seen for positively valenced items in any of the groups. For negative and neutral items, in late childhood individuals with FAS and PFAS performed more poorly than HE and controls, but by adolescence only the FAS group continued to perform more poorly. Test–retest reliability was moderate across the two ages. At both timepoints, the effects in the FAS group were partially mediated by Verbal Fluency but not by other aspects of executive function.

Conclusions

Individuals with full FAS have greater difficulty interpreting facial emotions than those with non‐syndromal HE and healthy controls in both childhood and adolescence. By contrast, RME deficits in individuals with PFAS in childhood represent developmental delay.

We previously reported poorer performance in children with fetal alcohol syndrome (FAS) and partial FAS (PFAS) on the “Reading the Mind in the Eyes” assessment of the ability to interpret facial emotion. In this follow‐up of 62 adolescents, those with PFAS performed similarly to controls, but those with FAS continued to perform more poorly. The groups also differed depending on affective valence of the facial expressions. Performance by the FAS group was partially mediated by verbal fluency but not by other aspects of executive function.

INTRODUCTION

Heavy prenatal alcohol exposure (PAE) can result in craniofacial anomalies, growth restriction, and a range of cognitive and behavioral impairments, collectively referred to as fetal alcohol spectrum disorders (FASD). Although social and emotional deficits have been identified in classification systems of FASD (e.g., Astley & Clarren, ²⁰⁰⁰; Hoyme et al., ^{2005, 2016}), most of the early research on neurodevelopment focused on impairment in cognitive function. In 2013, the American Psychiatric Association (APA) Diagnostic and Statistical Manual of Mental Disorders—5th edition (DSM‐5) recognized FASD for the first time as a “condition in need of further study” (APA, ²⁰¹³). The DSM‐5 proposed a new diagnosis, Neurodevelopmental Disorder Associated with PAE (ND‐PAE), which requires impairment in three domains: cognition, behavior (including self‐regulation), and adaptive function, particularly social communication and interaction. Social skills and communication are also included in the current Canadian guidelines (Cook et al., 2016).

By contrast to findings on cognitive impairment in FASD, there is limited empirical evidence linking PAE to deficits in social communication and interaction. Some individuals with fetal alcohol syndrome (FAS) have been described as having difficulty understanding the consequences of their behavior and being socially withdrawn and indifferent to subtle social cues (Streissguth et al., 1991). These observations have led to a growing interest in empirical studies focusing on the teratogenic effects of PAE on social cognition, which involves interpreting, encoding, and applying information about other people and social situations (e.g., Carmichael Olson et al., ¹⁹⁹⁸; Kully‐Martens et al., ²⁰¹²; McGee et al., ^{2008, 2009}; Rasmussen et al., ²⁰¹³).

Theory of mind, the capacity to understand the mental states of others (i.e., their desires, beliefs, emotions, and intentions; Baron‐Cohen et al., ¹⁹⁹⁴), plays a critical role in effective social functioning and aids in the development of interpersonal relationships. Inability to decode the emotions of others accurately can lead to inappropriate behavior and problems in social relations (Ciarrochi et al., 2000; Marsh et al., ²⁰⁰⁷), while the ability to identify and interpret facial emotions accurately can limit social conflict (Fischer & Manstead, 2008). Young children diagnosed with FASD perform more poorly on first‐order false belief theory of mind tasks, which are designed to assess the ability to understand that another person may not possess knowledge that they possess (Greenbaum et al., 2009; Rasmussen et al., ²⁰⁰⁹).

Reading another's mental state, also known as mentalizing, is a complex two‐stage process that requires the attribution of a mental state, followed by inferring its content. We have previously reported that 9‐ to 11‐year‐old children with FAS and partial FAS (PFAS) performed more poorly than typically developing controls on the “Reading the Mind in the Eyes” (RME) test (Lindinger et al., 2016), a higher‐order theory of mind task. The RME assesses the attribution part of mentalizing by measuring the ability to identify basic facial emotions and more complex mental states when viewing expressions solely in the eye region (Baron‐Cohen et al., 2001). The RME provides a relatively direct measure of mental state attribution because it does not rely on naming ability or free recall (mental state words are provided) and demands on working memory are minimized (stimuli remain in the participant's view). The RME has been used in more than 250 studies (Vellante et al., 2013) and has been shown to be sensitive to deficits in social cognition in children with autism (Baron‐Cohen et al., 2001) and in individuals with bipolar disorder (Petersen et al., 2016), attention deficit hyperactivity disorder (Mary et al., 2016), and epilepsy (Lew et al., 2015).

Evaluation and interpretation of another person's mental states and intentions requires the ability to keep one's own and another's perspective in mind, while switching between and contrasting the two viewpoints. Thus, there appears to be considerable overlap between the cognitive demands involved in social cognition and several aspects of executive function. The term executive function refers to a set of inter‐related higher‐order cognitive processes, including cognitive control, planning, and response inhibition, that are involved in selecting and successfully monitoring behaviors to facilitate the attainment of chosen goals (Anderson, 2002). Just as executive function continues to develop during adolescence, mental perspective taking also improves with age, resulting in the ability to engage in increasingly complex, higher‐order social cognitive processing. In typically developing children and adults, specific aspects of executive function and theory of mind have been shown to be intricately linked (Bock et al., 2015; Carlson et al., ²⁰⁰⁴; Hughes & Ensor, ²⁰⁰⁷). For example, the executive function component referred to as cognitive flexibility has been shown to be related to social understanding and to be particularly important in higher‐order theory of mind (Ahmed & Miller, 2011; Apperly et al., ²⁰⁰⁹; Bock et al., ²⁰¹⁵). Similarly, set shifting plays an important role in higher order social cognition by facilitating the optimization of perspective taking when predicting or interpreting behavior (Im‐Bolter et al., 2016). In children and adolescents with autism spectrum disorder, performance on theory of mind tasks (Joseph & Tager‐Flusberg, 2004) and specifically on the RME (Kouklari et al., 2018) have been positively associated with working memory, response inhibition and cognitive flexibility. In addition to executive function, semantic language ability (i.e., understanding the meaning of words and phrases) appears especially important for perspective taking in early childhood (Astington & Jenkins, 1999; Milligan et al., ²⁰⁰⁷) and adolescence (Im‐Bolter et al., 2016).

In our recent functional magnetic resonance imaging study on affective appraisal (Lindinger et al., 2021), we found marked PAE‐related effects on the neural processing of affective face stimuli that varied by emotional valence. Children with FAS and PFAS exhibited greater blood oxygenation level dependent (BOLD) signal changes when processing neutral faces compared to nonface, pixelated (control) images but a smaller signal when appraising angry compared to happy faces. These findings suggest that effects of PAE on mentalizing may vary depending on the emotional valence of stimuli.

The RME test includes problems relating to the mentalizing of positive, negative, and neutral basic emotions, personality attributes, and mental states. Few previous studies have examined the degree to which performance on this task may vary depending on the affective valence of the target stimuli. The first study to classify the stimuli of the adult version of the RME according to valence (Harkness et al., 2005) examined mental state decoding accuracy and valence bias in dysphoric college students. They found that, compared to nondysphoric students, the dysphoric group was more accurate on the RME overall and that depressed individuals were not more likely to select negative items in their mental state coding. Another study (Rueda et al., 2015) used the same valence classification system when examining aspects of empathy and emotion recognition in youth (aged 9 to 17 years) with Asperger's syndrome. The Asperger's group performed more poorly compared to an age‐, sex‐ and IQ‐matched control group on the RME overall and on positive items but not on negative or neutral items. Similarly, Baribeau et al. (2015) found that children with autism spectrum disorder or attention‐deficit/hyperactivity disorder performed more poorly than controls on the overall RME and on the positive items.

Stevens et al. (2017) administered the children's version of the RME to 21 typically developing 8‐ to 12‐year old‐children and 35 who met criteria for FASD based on either the Canadian Guidelines (Chudley et al., 2005) or the Astley and Clarren (2000) 4‐digit code. The FASD group performed more poorly on total RME and on the positive, negative and neutral items, compared to the controls. One limitation of the Stevens et al. (2017), Rueda et al. (2015), and Baribeau et al. (2015) studies was that, because no valence classification for the children's version of the RME had been developed, these studies used the Harkness et al. (2005) classification system, which was derived from the adult version of the RME.

In typically developing children, theory of mind and executive function emerge during early childhood and continue to develop throughout adolescence. Yet, to our knowledge, no previous studies have investigated the stability of RME performance across this developmental period. In this longitudinal follow‐up study, we examine performance on the RME administered to 62 children at school age and adolescence. Our primary goal is to determine the degree to which the PAE‐related deficit in the attribution of mental states seen during childhood represents a developmental delay (i.e., they catch up later) or persists across development. The aims of the study are (1) to develop a classification system for the items from the child version of the RME based on their emotional valence (positive, negative or neutral); (2) to determine the degree to which fetal alcohol‐related deficits in mentalizing differ depending on the valence of the emotion being evaluated; (3) to examine stability and change between school age and adolescence in fetal alcohol‐related impairment in attributing mental states; and (4) to examine the degree to which executive function mediates the effect of FASD on RME performance.

METHODS

Participants

This study was conducted in Cape Town, South Africa, where the prevalence of heavy alcohol consumption during pregnancy (particularly, weekend binge drinking) and incidence of FAS are among the highest in the world (May et al., 2013, 2018). Participants were 42 prenatally heavily alcohol exposed adolescents (FAS, PFAS and nonsyndromal heavily exposed (HE)) and 20 controls from the same community (mean age = 17.2 year, SD ± 0.4), who are participating in our Cape Town Longitudinal Cohort study (Jacobson et al., 2008) and had been administered the RME during late childhood (mean = 11.0 year, SD ± 0.4).¹

The mothers of the study participants were recruited between 1999 and 2002 at their first antenatal care visit to a neighborhood maternity obstetrical unit serving an economically disadvantaged population (Jacobson et al., 2008). At recruitment, each mother was interviewed orally by our research nurse using a timeline follow‐back interview regarding her drinking on a day‐by‐day basis during the preceding 2 weeks and during a typical week around time of conception. Recall was facilitated by asking the mother to remember where she was and with whom she was drinking during each drinking episode. The timeline follow‐back interview was repeated at mid‐pregnancy and again at 1‐month postpartum (to reflect the third trimester of pregnancy). Volume of each alcoholic beverage (beer, wine, liquor, cider) consumed daily during pregnancy was averaged and converted to oz absolute alcohol (AA; 0.5 oz AA ≈ 1 standard drink) and provided three summary measures of PAE: oz AA/day, oz AA/occasion, and frequency of drinking days across pregnancy.² Given that the mothers concentrated their drinking on the weekends, those reporting drinking ≥14 drinks/week (≥1.0 oz AA/day) on average or engaging in binge drinking (≥4 drinks/occasion) were considered heavy drinkers and recruited into the study. Women from the same community who abstained or were light drinkers (≤0.2 oz AA/day and did not binge drink) were recruited as controls. Data on smoking (cigarettes/day) and other illicit drug use (days/month), including marijuana (“dagga”), cocaine, and methaqualone (“mandrax”), during pregnancy were also collected at each visit.

In September 2005, we organized a diagnostic clinic during which each child was independently examined for growth (weight, height, and head circumference) and 18 fetal‐alcohol‐related dysmorphic features (see Jacobson et al., ²⁰²¹, Supplementary Table) by two expert FASD dysmorphologists (H.E. Hoyme (HEH), MD, and L.K. Robinson (LKR), MD), who were blind regarding PAE history. FAS was diagnosed if the participant exhibited a characteristic pattern of craniofacial dysmorphology (short palpebral fissures, thin upper lip (vermilion), flat philtrum), pre‐ and/or postnatal growth restriction, and small head circumference (Hoyme et al., 2005). PFAS was diagnosed if the participant exhibited a similar pattern of craniofacial dysmorphic features and either growth restriction or small head circumference. Case conferences including the dysmorphologists, SWJ, JLJ, and CDM were held to reach consensus regarding diagnosis of FAS and PFAS. The diagnoses were subsequently confirmed by examinations of the children in 2009 by HEH and LKR and by a team of expert dysmorphologists led by HEH in follow‐up clinics in 2013 and 2016. In 2017, HEH, SWJ, JLJ, and R.C. Carter, M.D., M.M. Sc., reviewed all of the anthropometric and dysmorphology data from the four clinics for each child including the maternal alcohol history and arrived at a final FASD diagnosis (see Jacobson et al., ²⁰⁰⁸).

Procedure and measures

Mothers and children were transported to our University of Cape Town (UCT) Child Development Research Laboratory in our research van. The mothers provided information regarding sociodemographic background including their educational attainment, marital status, and socioeconomic status (SES; Hollingshead, ²⁰¹¹).

Reading the mind in the eyes

The children's version of the RME (Baron‐Cohen et al., 2001) was administered orally with stimuli presented visually at both ages. The participant was presented with a series of 28 photos showing only the eye region of males and females, expressing simple and basic emotional expressions (e.g., sad), personality attributes (e.g., kind), and complex mental states (e.g., coercion; Figure 1). The participant was asked to choose which of four words (one target and three foil words) best depicts what the person was feeling or thinking. The examiner read each of the four words to the participant, who was encouraged to ask the meaning of any unfamiliar words. The stimuli remained in front of the participant during the decision process to reduce demand on working memory; no time limit was enforced.

FIGURE 1.

Examples of the stimuli presented in the child version of the “Reading the Mind in the Eyes” task (from Baron‐Cohen et al., ²⁰⁰¹). The word choices include basic and more complex emotions and mental state words. Correct answers for trials shown are “not pleased” and “happy”

Using the classification method applied to group the adult RME stimuli (Harkness et al., 2005), we conducted a valence classification study of the stimuli used in the children's version. We asked 24 postgraduate students to rank the valence of the each of the stimuli on a 7‐point Likert scale (1 = very negative; 4 = neutral; 7 = very positive). Stimuli with mean ratings significantly below four were classified as negative (e.g., upset, worried), those with mean ratings above four as positive (e.g., kind, friendly), and those that did not differ from the value of four as neutral (e.g., remembering, serious). Each of the 28 stimuli was classified according to valence based on the mean value of the Likert‐scale ratings: 11 items were rated as negative, 10 as neutral, and seven as positive (see Table 1). Based on these ratings, we created three new outcome variables, measuring accuracy for each valence category, at each age: overall accuracy (i.e., total number correct) and accuracy on the three emotional valence categories described above.

TABLE 1.

Reading the Mind in the Eyes stimuli valence classification

Item	Item number	Mean (SD)	Significantly different from neutral
			t	p
Positive valence
Kind	1	5.3 (1.3)	4.86	0.000
Friendly	3	5.2 (1.7)	3.44	0.002
Interested	7	5.0 (1.0)	4.51	0.000
Hoping	11	5.0 (1.0)	5.11	0.000
Thinking about something	13	4.5 (0.9)	2.41	0.024
Interested	19	4.5 (1.2)	2.18	0.039
Happy	28	6.0 (1.0)	10.02	0.000
Negative valence
Sad	2	2.2 (0.6)	−15.91	0.000
Upset	4	1.7 (0.8)	−13.39	0.000
Making somebody do something	5	2.5 (1.1)	−6.48	0.000
Worried	6	2.6 (0.9)	−7.70	0.000
Not believing	15	3.4 (1.2)	−2.43	0.023
A bit worried	17	3.4 (1.0)	−3.16	0.004
Thinking about something sad	18	2.8 (0.9)	−6.58	0.000
Not pleased	20	2.5 (1.5)	−4.89	0.000
Serious	24	3.2 (1.1)	−3.40	0.002
Worried	25	3.0 (0.9)	−5.62	0.000
Nervous	26	3.4 (1.1)	−2.90	0.008
Not believing	27	3.5 (1.3)	−1.96	0.062
Neutral valence
Remembering	8	4.3 (0.9)	1.66	0.110
Thinking about something	9	4.5 (0.9)	2.41	0.024
Not believing	10	3.7 (1.1)	−1.50	0.148
Serious	12	3.8 (1.2)	−1.03	0.314
Thinking about something	14	4.4 (1.1)	1.74	0.095
Made up her mind	16	3.9 (1.3)	−0.46	0.649
Interested	21	4.1 (1.3)	0.47	0.641
Thinking about something	22	4.3 (0.9)	1.88	0.073
Sure about something	23	4.2 (1.1)	0.72	0.477

Item means significantly lower than a value of 4 (neutral) classified as negative; significantly higher, as positive; not significantly different, as neutral.

Neuropsychological assessment

Participants were administered a battery of cognitive tests in Afrikaans or English, depending on the primary language used in their school. IQ assessed at 11 years on the Wechsler Intelligence Scale for Children‐Fourth Edition (WISC‐IV; Wechsler, ²⁰⁰³) and at 17 years on the Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, ¹⁹⁹⁹), which were strongly correlated (r = 0.79, p < 0.001). We administered the South African version of the WASI (Ferrett et al., 2014), in which a few words used in South African English were substituted for the American terms (e.g., “torch,” rather than “flashlight,” “holiday,” rather than “vacation”); similarly, animals that were more familiar (e.g., “lion,” rather than “bear”; “alligator” rather than “crocodile”) were used. Test instruments were translated into Afrikaans by a clinical psychologist whose first language is Afrikaans. Examiners were blind regarding PAE history and FASD diagnosis. Informed consent procedures were approved by the Wayne State University and the UCT ethics committees. Consent was obtained from mothers at the recruitment and child/adolescent assessment visits; assent, from the children and adolescents.

Three executive function domains were assessed at both ages: Working Memory, Cognitive Control (also referred to as cognitive flexibility or set shifting in the literature), and Verbal Fluency. Except for Working Memory, which was examined using a single test (Digit Span Backward from the WISC‐IV), tests measuring the same domain were converted into to z‐scores and averaged to provide composite measures for each of these domains: at 11 years, the Cognitive Control composite measure consisted of the mean scores on the Inhibition and Inhibition/Switching subtests of the Colour‐Word Interference Test of the Delis‐Kaplan Executive Function System (D‐KEFS, Delis et al., ²⁰⁰¹) and the Children's Colour Trails Task (CCTT; Llorente et al., ²⁰⁰³). Similarly, at 17 years, the Cognitive Control composite consisted of the mean scores on the D‐KEFS Inhibition, D‐KEFS Inhibition/Switching and the D‐KEFS Trails test. The Verbal Fluency composites at each age consisted of scores on the three D‐KEFS Verbal Fluency subtests (Letters, Categories and Switching). The letters F, A, S, which are used as the initial word letters in the English version of the Verbal Fluency test, were replaced in the Afrikaans version by the letters B, S, L, because their prevalence as initial word letters was found to be similar in a validation study conducted by Ferrett et al. (2014).

Statistical analyses

Demographic background variables were examined by FASD group using analysis of variance (ANOVA) or chi‐square tests. AA/day was transformed (logX + 1) because it was positively skewed. Test–retest reliability between 11 and 17 years was assessed for each outcome variable—overall RME, positive, negative, and neutral RME accuracy—using Pearson r. Each of the RME measures was examined in a mixed model analysis of variance with age as a within‐subjects measure and FASD diagnosis as a between‐group measure. Matched pair t‐tests were used to compare change across age within each FASD group. Pearson r was used to examine the relation of AA/day to each of the RME measures. ANOVA was used to compare the performance of the four diagnostic groups on each of the RME measures; least significant difference tests were used for post hoc comparisons.

Five control variables were examined as potential confounders of the effects of AA/day and FASD diagnosis on each of the RME outcomes: SES, maternal smoking during pregnancy, and participant's sex, age at testing and language of assessment. Each control variable related even weakly (at p < 0.10) to a given outcome variable was included in multiple regression analyses examining the effects of AA/day and the potential confounders on each of the RME outcomes and in ANCOVAs to adjust for potential confounders in the analyses of diagnostic group differences. Within‐group sex differences were examined using independent sample t‐tests. ANCOVAs were run to determine whether the effects of FASD on RME performance remained significant after controlling for IQ.

The composite executive function scores were examined in relation to AA/day using Pearson correlation and in relation to FASD diagnosis using one‐way ANOVA. To determine whether executive function mediated the association of FASD diagnosis with RME performance, estimates of the indirect effect were determined using the MEDIATE macro from SPSS (Hayes & Preacher, 2014), a multivariate extension of the Sobel (1982) method. This macro allows the use of multi‐categorical independent variables (IV) and multiple mediators. The FAS, PFAS, and HE diagnostic categories were indicator coded using the control group as the reference group. Mediation was determined based on the size of the indirect effects, which are the product of the regression coefficients for the direct effects of the IV on the mediating variable (MV) and the MV on the dependent variable. Statistical significance of the indirect effects was determined based on 95% bootstrap percentile confidence intervals. Indirect effects are deemed statistically significant when the confidence intervals do not overlap with 0.

RESULTS

Participant characteristics

Maternal and participant demographic and background characteristics are summarized in Table 2. Mothers of children with FAS and PFAS were less educated compared to control mothers, and the mothers of children with PFAS were also less educated than mothers of children in the HE group. Mean SES for the FAS and PFAS groups was within the lowest social stratum on the Hollingshead Scale (Hollingshead, 2011; Level V, unskilled labor), while the mean for the HE and control groups fell in Level IV (semi‐skilled workers).

TABLE 2.

Participant characteristics (n = 62)

	FAS (n = 8)	PFAS (n = 15)	HE (n = 19)	CTL (n = 20)	F or χ 2	p
Maternal characteristics
Education (years completed) a	8.4 (2.0)	6.9 (2.5)	9.2 (2.5)	10.3 (1.5)	7.32	<0.001
Socioeconomic status b , c	16.4 (8.4)	15.6 (6.4)	20.7 (7.9)	24.9 (7.3)	5.33	0.003
Alcohol use during pregnancy d
N (%) who drank alcohol	8 (100.0)	15 (100.0)	19 (100.0)	1 (5.0)
AA/day (oz) (drinkers only) e	1.8 (2.4)	1.1 (0.7)	0.6 (0.4)	N/A	2.55	0.070
AA/occasion (oz) (drinkers only)	4.4 (2.0)	4.5 (2.8)	3.5 (2.6)	N/A	0.93	0.437
Frequency (days/week) (drinkers only)	2.2 (1.9)	1.9 (1.0)	1.3 (0.8)	N/A	2.12	0.114
Smoking during pregnancy
N (%) who smoked f	8 (100.0)	14 (93.3)	14 (73.7)	6 (30.0)	21.65	<0.001
Cigarettes/day (smokers only)	8.8 (5.3)	7.8 (5.1)	8.6 (5.6)	6.5 (6.7)	0.25	0.861
Participant characteristics
Sex (male:female)	3:5	8:7	12:7	11:9	1.52	0.679
Age at school‐age assessment (years) g	10.7 (0.4)	11.0 (0.4)	11.4 (0.3)	11.0 (0.3)	7.34	<0.001
Age at adolescent assessment (years) h	16.9 (0.6)	17.0 (0.6)	17.6 (0.7)	17.0 (0.5)	5.39	0.002
Language (English:Afrikaans) i	3:5	4:11	12:7	17:3	13.62	0.003

Values are Mean (SD). AA, absolute alcohol; FAS, fetal alcohol syndrome; PFAS, partial fetal alcohol syndrome; HE, heavily exposed nonsyndromal; CTL, controls; N/A, not applicable.

^a FAS < CTL, p = 0.042; PFAS < CTL, p = 0.003; PFAS < HE, p < 0.001.

^b Based on Hollingshead Four Factor Index of Social Status (Hollingshead, 2011).

^c FAS < CTL, p = 0.008; PFAS < CTL, p = 0.001; HE < CTL, p = 0.088; PFAS < HE, p = 0.048.

^d F‐tests for alcohol use compare the three groups of heavy drinkers.

^e FAS > HE, p = 0.002.

^f FAS > CTL, p < 0.001; PFAS > CTL, p < 0.001; HE > CTL, p = 0.006.

^g FAS < PFAS, p = 0.088; FAS < HE, p < 0.001; PFAS < HE, p = 0.006; CTL < HE, p = 0.001.

^h FAS < HE, p = 0.004; PFAS < HE, p = 0.005; CTL < HE, p = 0.001.

ⁱ FAS < CTL, p = 0.012; PFAS < HE, p = 0.034; PFAS < CTL, p < 0.001.

All three alcohol‐exposed groups were lower in SES than the controls. Mothers of participants with FAS drank on average 8 to 9 standard drinks on 1 to 2 days/week. All but one control mother abstained from drinking alcohol during pregnancy; that mother reported drinking 2 to 3 drinks 3 to 4 times/month. Illicit drug use during pregnancy was rare and infrequent. Two mothers of children with FAS reported using marijuana less than 1 day/month during pregnancy. One mother of a child with PFAS reported smoking marijuana 3 days/month and one used cocaine 2 to 3 days/month. Two mothers in the HE group reported smoking marijuana during pregnancy, one on 3 occasions; the other, on 2. One control mother reported smoking marijuana 2 to 3 days/month. The groups did not differ by sex, but the HE group was slightly older than the other children at both assessments, a difference probably too small to be clinically relevant. The FAS and PFAS groups had proportionately fewer English speakers than the HE and control groups.

Effects of PAE and FASD diagnostic group on Reading the Mind in the Eyes

Higher levels of PAE, measured by maternal alcohol consumption during pregnancy, were associated with poorer overall RME performance at both ages (Table 3). Higher PAE was associated with less accurate interpretation of the negative and neutral stimuli but did not affect the ability to interpret the positive stimuli at either age. The diagnostic groups differed significantly on overall RME and the negative and neutral valenced stimuli at both ages. Table 4 shows which of the control variables were considered as potential confounders of effects on each of the outcome measures. All of the significant effects reported in Table 3 continued to be significant after control for potential confounders, except for the effect of diagnostic group on neutral RME at 17 years, which fell short of statistical significance. The effect of diagnostic group on overall RME performance remained significant after adjustment for IQ, F(3, 57) = 5.65, p = 0.002, and F(3, 57) = 3.82, p = 0.015, at 11 and 17 years, respectively.

TABLE 3.

Overall and valence accuracy on the RME by AA/day and FASD diagnostic group at the 11‐ and 17‐year follow‐up assessments

Outcome variables	AA/day		FAS (n = 8)	PFAS (n = 15)	HE (n = 19)	CTL (n = 20)	F a	F b
	r	β
Overall accuracy
11 years c	−0.34**	−0.26*	10.0 (2.4)	12.9 (2.5)	16.2 (3.1)	16.3 (3.4)	11.81***	8.29***
17 years d	−0.38**	−0.27*	14.8 (3.5)	18.2 (4.1)	20.4 (3.3)	20.3 (2.8)	7.70***	4.18*
Valence accuracy
11 years: Positive	−0.08	−0.10	1.8 (1.4)	2.8 (1.3)	3.3 (1.8)	3.0 (1.2)	2.35	1.41
17 years: Positive	−0.14	−0.12	3.5 (1.7)	3.6 (1.4)	3.8 (1.5)	3.9 (1.1)	0.26	0.25
11 years: Negative e	−0.38**	−0.30*	4.5 (14.9)	5.4 (1.5)	7.0 (1.3)	6.7 (2.1)	9.34***	5.25**
17 years: Negative f	−0.29*	−0.29*	4.9 (1.5)	7.3 (1.6)	8.4 (1.5)	8.1 (1.4)	11.63***	8.17***
11 years: Neutral g	−0.27*	−0.12	4.0 (0.9)	5.1 (1.5)	6.4 (1.8)	6.5 (1.8)	6.23**	3.79*
17 years: Neutral h	−0.41**	−0.32*	6.0 (1.3)	7.3 (2.1)	8.1 (1.7)	8.3 (1.8)	3.82*	2.35 †

Values are means (SD) for the total mean raw scores, except for AA/day where values are Pearson r or standardized regression coefficients (β) after adjustment for confounders. AA, absolute alcohol; FAS, fetal alcohol syndrome; PFAS, partial fetal alcohol syndrome; HE, heavily exposed nonsyndromal; CTL, controls; RME, Reading the Mind in the Eyes.

^a Before controlling for confounders.

^b After controlling for confounders.

^c FAS < PFAS, p = 0.033; FAS < HE, p < 0.001; FAS < CTL, p < 0.001; PFAS < HE, p = 0.002; PFAS < CTL, p = 0.002.

^d FAS < PFAS, p = 0.013; FAS < HE, p < 0.001; FAS < CTL, p < 0.001; PFAS < HE, p = 0.071; PFAS < CTL, p = 0.084.

^e FAS < HE, p < 0.001; FAS < CTL, p < 0.001; PFAS < HE, p = 0.002; PFAS < CTL, p = 0.001.

^f FAS < PFAS, p = 0.001; FAS < HE, p < 0.001; FAS<CTL, p < 0.001; PFAS < HE, p = 0.029.

^g FAS < HE, p = 0.001; FAS < CTL, p < 0.001; PFAS < HE, p = 0.033; PFAS < CTL, p = 0.014.

^h FAS < PFAS, p = 0.089; FAS < HE, p = 0.006; FAS < CTL, p = 0.003.

^† p = 0.082; *p < 0.05; **p < 0.005; ***p < 0.001.

TABLE 4.

Relation of control variables to RME performance

Outcome variables	Maternal		Participant
	SES	Smoking	Sex	Age	Language
Overall accuracy
11 years	0.250	−0.248	0.043	0.186	−0.399**
17 years	0.192	0.006	0.127	0.272*	−0.287**
Valence accuracy
11 years: Positive	−0.112	−0.053	−0.051	0.256*	−0.124
17 years: Positive	−0.070	0.139	−0.030	0.180	−0.091
11 years: Negative	0.290*	−0.264*	0.030	0.063	−0.438***
17 years: Negative	0.208	0.079	0.145	0.264*	−0.173
11 years: Neutral	0.277*	−0.179	0.124	0.121	−0.277*
17 years: Neutral	0.245	−0.163	0.142	0.177	−0.359**

Values are Pearson r. RME, Reading the Mind in the Eyes; SES, socioeconomic status assessed on the Hollingshead Index (Hollingshead, 2011).

^* p < 0.05; **p < 0.005; ***p < 0.001.

Group differences in Reading the Mind in the Eyes performance: stability and change

Within‐group performance across age

Accuracy on each of the RME measures was moderately correlated between the two ages (Table 5). Not surprisingly, test–retest reliability was higher for overall RME than for the valence measures, each of which were assessed on fewer items. Paired sample t‐tests showed that all groups improved with age on overall RME accuracy and when rating neutral RME stimuli (all ps < 0.001; Figure 2A,D). For positive stimuli (Figure 2B), accuracy increased for the FAS (p = 0.004) and control groups (p = 0.014) and marginally for the PFAS group. (p = 0.061). Accuracy for negative stimuli (Figure 2C) increased for all groups (all ps < 0.05) from 11 to 17 years, except for the FAS group (p = 0.213).

TABLE 5.

Test–retest reliability of the RME measures between the 11‐year and 17‐year follow‐up assessments (n = 62)

Overall RME accuracy	0.611**
Positive valence accuracy	0.475**
Negative valence accuracy	0.341*
Neutral valence accuracy	0.523**

Values are Pearson r.

RME, Reading the Mind in the Eyes.

^* p < 0.05; **p < 0.001.

FIGURE 2.

(A) Performance on overall RME accuracy showing a main effect for age, F(1, 58) = 93.93, p < 0.001, and diagnosis F(3, 58) = 12.49, p < 0.001, but no interaction effect for age by diagnosis, F(3, 58) = 0.50, p = 0.683. All groups showed significant increases in performance between 11 and 17 years. (B) Performance on positive valence accuracy showing a main effect for age, F(1, 58) = 25.84, p < 0.001 but not for diagnosis, F(3, 58) = 1.35, p = 0.267 and no age by diagnosis interaction, F(3, 58) = 1.31, p = 0.281. FAS and control groups showed significant increases in performance between 11 and 17 years. (C) Performance on negative valence accuracy showing a main effect for age, F(1, 58) = 32.09, p < 0.001, and diagnosis, F(3, 58) = 19.65, p < 0.001 but no age by diagnosis effect, F(3, 58) = 1.03, p = 0.385. PFAS, HE, and control groups showed significant increases in performance between 11 and 17 years. (D) Performance on neutral valence accuracy showing a main effect for age, F(1, 58) = 58.86, p < 0.001, and diagnosis, F(3, 58) = 6.91, p < 0.001, but no age by diagnosis effect, F(3, 58) = 0.23, p = 0.874. All groups showed significant increases in performance between 11 and 17 years

Between‐group differences across age

On overall RME accuracy, participants with FAS performed more poorly than all other groups at both ages (Table 3). The PFAS group performed more poorly than the HE and control groups at 11 years, but at 17 years, those differences fell short of statistical significance. For the negative and neutral stimuli, at 11 years the FAS and PFAS groups performed more poorly than the HE and control groups. At 17 years, the FAS group continued to perform more poorly than the HE and controls on the negative and neutral items, but accuracy of the PFAS group no longer differed from that of the controls. No significant group by age interactions was found on any RME outcomes.

Sex differences in performance accuracy

Independent sample t‐tests showed no sex differences on any of the RME accuracy measures for the sample as a whole (i.e., males vs. females; all ps > 0.20). When contrasting outcomes within each group across sex, we also found no differences for overall RME accuracy for any of the groups at either age (all ps > 0.10). Although valence‐dependent sex differences were not seen in the control group at 11 years (t = 0.79 p = 0.440), at 17 years, girls in the control group performed better than boys on positive RME accuracy (t = 2.39; p = 0.028). For negative stimuli at 11 years, girls from the HE group performed better than boys (t = −2.20, p = 0.040), but this sex difference was no longer evident at 17 years (p = 0.112). No other between‐group sex differences were found.

Effects of PAE and diagnostic group on executive function

Higher levels of PAE, measured in terms of AA/day, were associated with poorer Verbal Fluency at both ages and with poorer Cognitive Control at 17 years (Table 6). At 11 years, the FAS and PFAS groups performed more poorly than the HE group on Cognitive Control and Verbal Fluency. At 17 years, the FAS group performed more poorly than the control group on Verbal Fluency and marginally more poorly than the HE group, but the PFAS group no longer differed from the HE group. The PFAS group performed more poorly than the HE and control groups on Cognitive Control.

TABLE 6.

Performance measures of executive function by AA/day and FASD diagnostic group at the 11‐ and 17‐year follow‐up assessments

EF composites	oz AA/day	FAS (n = 8)	PFAS (n = 15)	HE (n = 19)	CTL (n = 20)	F
11‐year follow‐up
Working memory a	−0.232	−0.43 (1.00)	−0.41 (1.17)	0.39 (0.95)	0.11 (0.77)	2.57 †
Cognitive control b	−0.150	−0.48 (1.02)	−0.26 (0.70)	0.37 (0.44)	0.06 (0.60)	4.30**
Verbal fluency c	−0.296*	−0.54 (0.67)	−0.21 (0.50)	0.37 (1.0)	−0.00 (0.68)	3.25*
17‐year follow‐up
Working memory d	−0.213	−0.37 (0.50)	−0.29 (0.78)	0.15 (0.79)	0.20 (0.70)	2.25 †
Cognitive control e	−0.288*	−0.16 (0.78)	−0.44 (0.78)	0.21 (0.56)	0.24 (0.69)	3.59**
Verbal fluency f	−0.369**	−0.66 (0.86)	−0.05 (0.72)	−0.04 (0.65)	0.33 (0.89)	3.14*

Values are means (SD) for the total mean standard composite score, except for AA/day where values are Pearson r. AA, absolute alcohol; EF, executive function; FAS, fetal alcohol syndrome; PFAS, partial fetal alcohol syndrome; HE, heavily exposed nonsyndromal; CTL, controls.

^a FAS < HE, p = 0.048; PFAS < HE, p = 0.019.

^b FAS < HE, p = 0.003; FAS < CTL, p = 0.053; PFAS < HE, p = 0.007.

^c FAS < HE, p = 0.006; FAS < CTL, p = 0.095; PFAS < HE, p = 0.031.

^d FAS < HE, p = 0.093; FAS < CTL, p = 0.065; PFAS < HE, p = 0.086; PFAS < CTL, p = 0.055.

^e PFAS < HE, p = 0.008; PFAS < CTL, p = 0.005.

^f FAS < PFAS, p = 0.079; FAS < HE, p = 0.063; FAS < CTL, p = 0.004.

^† p < 0.10; *p < 0.05; **p < 0.005.

Mediation of the effects of FASD on Reading the Mind in the Eyes outcome by executive function

The path model examining mediation by executive function of the effect of FASD accounted for 41.3% of the variance in overall RME performance at 11 years, F(6, 55) = 8.16, p < 0.001, and 39.2% of the variance in overall RME at 17 years, F(6, 55) = 7.56, p < 0.001 (Table 7; Figure 3). The paths from the FAS indicator variable through Verbal Fluency to overall RME (i.e., the indirect effects) were significant at both 11 and 17 years, but the paths through Working Memory and Cognitive Control to RME were not significant at either age. None of the indirect effects from the PFAS and HE variables were significant at either age.

TABLE 7.

Mediation of effects of FASD on overall RME by three aspects of executive function

	Direct effects: FASD to RME			Indirect effects: FASD through mediator to RME			Model summary
	FAS	PFAS	HE	FAS	PFAS	HE	R 2
11‐year overall RME							0.47***
Direct effects	−0.71***	−0.44**	−0.06
Indirect effects
Working memory				−0.06 (−0.90, 0.35)	−0.06 (−0.75, 0.37)	0.03 (−0.20, 0.55)
Cognitive control				0.37 (−0.18, 2.25)	0.22 (−0.09, 1.11)	−0.21 (−0.97, 0.07)
Verbal fluency				−0.81 (−2.37, −0.10)	−0.30 (−1.18, 0.21)	0.56 (−0.10, 1.92)
17‐year overall RME							0.45***
Direct effects	−0.45*	−0.10	0.10
Indirect effects
Working memory				0.04 (−0.95, 0.93)	0.03 (−0.74, 0.91)	0.00 (−0.38, 0.40)
Cognitive control				−0.35 (−2.55, 0.15)	−0.60 (−2.56, 0.30)	−0.03 (−0.76, 0.34)
Verbal fluency				−1.77 (−4.46, −0.25)	−0.68 (−2.34, 0.08)	−0.65 (−2.06, 0.05)

Values for direct effects are standardized regression coefficients. Values for indirect effects are raw regression coefficients (95% confidence intervals). Indirect effects are significant at p < 0.05 (shown in bold) if confidence interval does not include zero. R ² is the total variance explained by the direct and indirect effects. FASD, fetal alcohol spectrum disorders; RME, Reading the Mind in the Eyes; FAS, fetal alcohol syndrome; PFAS, partial fetal alcohol syndrome; HE, nonsyndromal heavily exposed.

^* p < 0.05; **p < 0.01; ***p < 0.001.

FIGURE 3.

Path model showing the mediation by Verbal Fluency of the effect between diagnostic group and overall RME outcome at 11 years (A) and 17 years (B). Values are standardized regression coefficients (β) that indicate by how many standard deviations the outcome changes for each unit change in the predictor. The control group was used as the reference group for the FAS, PFAS, and HE coefficients. RME, Reading the Mind in the Eyes; FAS, fetal alcohol syndrome; PFAS, partial fetal alcohol syndrome; HE, nonsyndromal heavily exposed.

DISCUSSION

This is the first longitudinal study to investigate PAE‐related effects on mentalizing ability from late childhood to adolescence. Overall RME accuracy improved with age for all participants, which is consistent with previous findings suggesting that the ability to recognize emotions from facial expressions increases with age (Fox, 2001; Herba & Phillips, ²⁰⁰⁴; Widen & Russell, ²⁰⁰⁷). However, the mentalizing impairment seen in children with FAS at 11 years persisted at 17 years, suggesting that these individuals have greater difficulty with this social‐cognitive domain than nonsyndromal individuals with heavy PAE and healthy controls. By contrast, the differences between the PFAS and control groups for overall RME and negative valence were no longer significant at 17 years, suggesting that their scores in late childhood represented a developmental delay rather than a persistent deficit.³

Fetal alcohol‐related effects on the RME differed by valence across age and between groups. The FAS group exhibited developmental delay in the identification of positive mental states—less accurate at 11 years but as accurate as the other groups by adolescence. These findings are consistent with research indicating that, among the basic emotions, “happy” is among the easiest facial emotions to identify (Kirita & Endo, 1995; Leppänen & Hietanen, ^{2003, 2004}) and is expected to mature first (Boyatzis et al., 1993; Camras & Allison, ¹⁹⁸⁵; Widen & Russell, ²⁰⁰³). By contrast, the FAS group did not improve with age on the identification of negative and neutral mental states, whereas the PFAS group no longer performed significantly more poorly than the controls groups by age 17. Hence, as for overall RME accuracy, there was developmental delay with “catch‐up” in the PFAS group but not in the FAS group.⁴

It is notable that the studies examining the valence of different RME stimuli in autism spectrum disorder found deficits in accuracy only on positive stimuli (Baribeau et al., 2015; Rueda et al., ²⁰¹⁵). These findings suggest that positive affect is particularly difficult to interpret in autism spectrum disorder; we did not detect this impairment in FASD. The improvement in the interpretation of positive affect during adolescence in the FAS group was not seen in the Stevens et al. (2017) study, which examined school age children and did not distinguish FAS from other forms of FASD.

The FAS and PFAS groups performed more poorly on the negative valence items than the HE and control groups at 11 years, and the FAS group continued to perform more poorly on those items than the PFAS, HE, and controls at 17 years. These findings indicate a persistent deficit in the ability to identify negative emotions in the FAS group and are consistent with findings from our 12‐year affective appraisal neuroimaging study (Lindinger et al., 2021), in which children with FAS and PFAS showed reduced neuronal activation in two regions involved in identifying angry (compared with happy) faces relative to controls. Difficulty in the appraisal of negative affect is likely to result in inappropriate responses to another's behavior, based on false inferences being made following the misidentification of that person's facial affect. For example, several studies have shown that children who make more hostile attributions regarding peer intent when presented with both benign and ambiguous scenarios are more likely to engage in aggressive behaviors (Dodge, 1980; Orobio de Castro et al., ²⁰⁰²).

Data from previous studies suggest that executive function plays an important role in higher‐order theory of mind in both middle childhood and early adolescence (Im‐Bolter et al., 2016; Joseph & Tager‐Flusberg, ²⁰⁰⁴; Kouklari et al., ²⁰¹⁸). Although deficits in working memory and cognitive control were seen in the FAS group, we found that only Verbal Fluency mediated the effects of PAE on RME in that group at both ages, suggesting that semantic (category) fluency and lexical (letter) fluency are the aspects of executive function that impact most directly on the interpretation of facial emotions in individuals with FAS. Verbal fluency may be important for higher order theory of mind because, as semantic language skills develop and become increasingly sophisticated, they provide the means to formulate and represent the unobservable mental states that require processing during social cognition (Im‐Bolter et al., 2016). Semantic language continues to develop into adulthood, providing a vehicle to represent increasingly nuanced and sophisticated intentions, beliefs, and emotions.

There have been inconsistent findings regarding sex differences on the RME with some indication that adult females perform better than adult males (Baron‐Cohen et al., 2015; Guariglia et al., ²⁰¹⁵; Vellante et al., ²⁰¹³). No sex differences were seen on overall RME accuracy in our sample of socioeconomically disadvantaged children and adolescents. In the FAS and PFAS groups, who demonstrated impairment and delay in the processing of negative stimuli, sex differences were also not apparent. This absence of sex differences in a clinical population is consistent with findings showing no significant differences on overall RME accuracy between male and female adults with autism (Baron‐Cohen et al., 2015).

Limitations

Participants with FAS were slightly younger than the HE group, which may have contributed to their poorer performance. However, the PFAS and control groups were also younger than the HE group at both ages, and their performance improved with age in comparison to the other groups. Given the small sample size, the findings warrant replication in a larger sample. However, in the mediation analyses, the sample size problem is obviated to some extent by the use of bootstrap‐derived confidence intervals to calculate statistical significance, an approach that does not rely on large sample asymptotic results normally required to estimate measurement error. The current research was conducted with a socio‐environmentally disadvantaged South African cohort and the effects, therefore, need to be confirmed in other populations. Research on the degree to which social context might modify these effects is also warranted. Future studies are also needed to study these social cognitive outcomes in adults with FAS to see if they eventually catch up.

CONCLUSIONS

Our findings provide empirical evidence supporting clinical reports and the DSM‐5 proposed criteria for ND‐PAE that individuals with FASD may show “impairment in social communication and interaction.” These findings also suggest that PAE‐related facial emotion attribution errors are found particularly in the processing of negative emotional expressions, such as anger, anxiety and disbelief, increasing the risk of misinterpretation of other people's feelings and intentions in emotionally fraught social interactions. Our data provide the first evidence that developmental trajectories for interpreting positive and negative mental states differ in individuals with FAS and PFAS, with the FAS group improving in their interpretation of positively valenced items between school age and adolescence; the PFAS group, improving on negatively valenced items. The degree to which these PAE‐related developmental challenges are attributable to deficits at a basic level of facial emotion processing and/or at a more complex higher‐order level of social cognitive functioning is not clear and warrants investigation. Our mediation data examining the role of executive function in RME suggest that, for individuals with FAS, an intervention focusing on verbal fluency might be effective for remediating impairment in the ability to evaluate and interpret the mental states and intentions of others. Given their greater difficulty in identifying negatively valenced facial expressions, children and adolescents with FAS and PFAS might also benefit from social skills interventions that have been shown to reduce hostile attribution bias in response to benign and ambiguous scenarios (Keil et al., 2010).

CONFLICT OF INTEREST

The authors declare that they have no conflict of interest.

Lindinger, N.M. , Jacobson, J.L. , Dodge, N.C. , Malcolm‐Smith, S. , Molteno, C.D. , Meintjes, E.M. & Jacobson, S.W. (2022) Stability and change in the interpretation of facial emotions in fetal alcohol spectrum disorders from childhood to adolescence. Alcoholism: Clinical and Experimental Research, 46, 1268–1281. Available from: 10.1111/acer.14851