Anxiety Symptoms Predict Subsequent Depressive Symptoms in Neurodivergent Youth: A 10-Year Longitudinal Study

The LINEUP study comprises three assessment waves, baseline (T1), 2-year follow-up (T2), and 10-year follow-up (T3). At each assessment, participants completed measures including self-reports of anxiety and depressive symptoms. Participants were recruited from child and adolescent psychiatric outpatient clinics at Innlandet Hospital Trust, Norway, upon consecutive referrals. All individuals aged between 8 and 17 years referred for assessment of autism or ADHD were invited to participate. The neurotypical comparison group was recruited through local schools and had to attend regular classes. All neurotypical youths had to screen negative for all present or lifetime psychological disorders in the semi-structured interview, the Kiddie-Schedule for Affective Disorders and Schizophrenia/Present and Lifetime version (Kiddie-SADS; Kaufman et al., 1997). Exclusion criteria for all participants were prematurity (< 36 weeks gestational age), having a disease affecting the central nervous system (e.g., epilepsy), or having IQ < 70. An additional exclusion criterion for the ADHD group was no history of stimulant treatment. This was because the baseline study focused on neuropsychological functioning and thus, a medication naïve sample was preferable for the study purpose of comparing the cognitive functioning of youth with ADHD to neurotypical youth prior to medication (Andersen et al., 2013; Skogli et al., 2014). Additional exclusion criteria for the neurotypical comparison group were dyslexia or head injury with loss of consciousness. The study was prospectively reviewed and approved by the Regional Committee for Medical Research Ethics in South-Eastern Norway (T1 and T2 ref. REK 6-2009-24; T3 ref. 2018/1611).

At baseline, 38 autistic youth, 85 youth diagnosed with ADHD, and 50 neurotypical youth were included. Demographic and clinical characteristics of all three groups are presented in Table 1. Unfortunately, no data was collected on the participants’ ethnicity. There were no significant differences in age or sex between the groups, but the groups differed in IQ, mothers’ education, and ADHD and autism symptomatology. The diagnostic assessment at T1 was based on a semi-structured clinical interview (Kiddie-SADS; Kaufman et al., 1997) conducted separately with the youths and their parents, and supplemented with information from the Autism Spectrum Screening Questionnaire (ASSQ; Ehlers et al., 1999), the ADHD Rating Scale IV (ARS-IV; DuPaul et al., 1998a, b), the Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001), and teacher reports on academic and social functioning. All measures have demonstrated good psychometric properties in terms of diagnostic specificity and sensitivity (e.g., DuPaul et al., 1998; Kaufman et al., 1997; Nøvik, 1999; Posserud et al., 2009). Diagnostic assessment was conducted by experienced clinical psychologists and educational therapists. The assessments and diagnostic decisions were supervised and reviewed by a clinical neuropsychologist specialized in neurodevelopmental disorders. Disagreements were discussed to arrive at a ‘best estimate’ DSM-IV consensus diagnosis. Eight participants met the criteria for both ADHD and autism and were included in the autistic group. Among the participants with ADHD, six also met the criteria for an anxiety disorder and three for a depressive disorder. Among the autistic participants, 10 also met the criteria for an anxiety disorder and two for a depressive disorder.

Between the assessments, neurodivergent participants received treatment as usual (TAU) at the child and adolescent psychiatric outpatient clinic or in their community health and social services. Among the participants with ADHD, 44 participants (54%) received stimulant treatment at T2, and 7 participants (12%) received stimulant treatment at T3. Among autistic participants, 3 (8%) participants received stimulant treatment at T1, 4 participants (11%) received stimulant treatment at T2, and 2 participants (8%) received stimulant treatment at T3. At T1, none of the neurodivergent participants received antidepressants. At T2, one participant with ADHD (1%) and one autistic participant (3%) received antidepressants. At T3, six participants with ADHD (10%) and two autistic participants (8%) received antidepressants.

At T1 and T2, we assessed anxiety symptoms with self-reports on the Revised Youth’s Manifest Anxiety Scale, second edition (Reynolds & Richmond, 1985). The RCMAS-2 comprises 49 items rated on a “yes” or “no” scale that can be summed up to a total anxiety score. At T3, we assessed anxiety symptoms with self-reports on the Adult Self-Report (ASR; Achenbach & Rescorla, 2003). The DSM-oriented anxiety subscale of the ASR comprises 9 items rated on a scale from 0 (absent), 1 (occurs sometimes), to 2 (occurs often). The change in measure from T2 to T3 was because data on the RCMAS-2, as a child-scale, was not collected at the 10-year follow-up, and thus was unavailable. The RCMAS-2 and the ASR have demonstrated good internal consistency (α ≥ 0.80 and α ~ 0.85, respectively) and concurrent validity in the form of strong correlations (r ≥.61) with other measures of anxiety in previous research on general population samples and samples of youth with anxiety (e.g., de Vries et al., 2020; Etkin et al., 2021; Guerrero et al., 2020). The RCMAS has also shown acceptable psychometric properties among neurodivergent youth specifically, in the form of good internal consistency (α = 0.88) and sensitivity to change following cognitive-behavioral treatment for co-occurring anxiety disorders (Chalfant et al., 2007; Mazefsky et al., 2011).

At all assessments, we assessed depressive symptoms with self-reports on the Short Mood and Feelings Questionnaire (SMFQ; Angold et al., 1995). The SMFQ comprises 13 items rated on a scale from 1 (not true) to 3 (true) and can be summed up to a total depressive symptoms score. The SMFQ has demonstrated good internal consistency (α = 0.87), a unifactorial structure, and good sensitivity (1.00) and specificity (0.79) among adolescents in a juvenile detention center and a unifactorial structure and good discrimination in an item response theory analysis among general population children (Kuo et al., 2005; Sharp et al., 2006). The SMFQ has also shown acceptable psychometric properties among neurodivergent youth specifically, in the form of good internal consistency (α = 0.85-0.87) and convergent validity with other measures of internalizing problems (Jarbin et al., 2020; Mazefsky et al., 2014).

We used analysis of variance (ANOVA) and chi-square test of independence to test for selective attrition based on baseline characteristics. To test the relationships between anxiety symptoms and depressive symptoms over time, we conducted multiple-group structural cross-lagged regression models (Selig & Little, 2012) in JASP version 0.19 (JASP Team, 2024). The models were estimated using robust maximum likelihood (MLR) to account for nonnormal distribution of data (Kline, 2023). The following effects were modeled: (1) Cross-lagged effects between anxiety and depressive symptoms at different times (e.g., anxiety at T1 as a predictor of depressive symptoms at T2 and depression at T1 as a predictor of anxiety symptoms at T2), while controlling for (2) autoregressive effects within each variable (e.g., depressive symptoms at T1 as predictor of depressive symptoms at T2), capturing the extent to which individual differences in symptoms are stable over time (i.e., rank order stability), and (3) concurrent associations between the variables at each assessment wave. Thus, the cross-lagged effects were the parameters of main interest. The cross-lagged pathways represented associations between anxiety symptoms and depressive symptoms from T1 to T2 and T2 to T3, allowing us to test both unidirectional and bidirectional relationships between the two variables. Lastly, due to selective attrition based on IQ, we included IQ as a covariate in the model by regressing IQ onto T1 anxiety and depressive symptoms.

We handled attrition by using full information maximum likelihood (FIML). Model fit was determined using the following indices: non-significant chi-square test of model fit, root mean square error approximation (RMSEA, should be ≤ 0.08), comparative fit index (CFI, should be ≥ 0.96), and standardized root mean square residual (SRMR, should be ≤ 0.09) (Alhija, 2010; Mueller & Hancock, 2001). We first tested a model where all effects (i.e., the cross-lagged and autoregressive effects and concurrent associations) were constrained across groups (i.e., neurodivergent youth and comparison youth). Thereafter, we tested whether the model fit improved when removing equality constraints of the cross-lagged effects and estimating them separately for neurodivergent and comparison youth. To interpret the effect size of the cross-lagged effects, we used the benchmark values established by Orth et al. (2024) with 0.03, 0.07, and 0.12 corresponding to a small, medium, and large effect, respectively. We expected large effect sizes, based on the effect sizes of 0.32 to 0.49 for the cross-lagged effects reported by Schiltz et al. (2023).We run a post-hoc power analysis using semPower (Moshagen & Bader, 2024). The power analysis (α = 0.05, df = 31, p = 24) indicated that we had adequate power (0.94) to detect misspecifications corresponding to a model with less than acceptable fit (RMSEA ≥ 0.08).

Of the 173 original participants, 168 participated at T2, giving an overall retention rate of 97.1% (see Table 2 for retention rates in each group). At T3, 127 of the original participants were re-assessed, giving an overall retention rate of 73.4%. Of the 46 original participants not re-assessed at T3, we were unable to locate 9 of them (4 from the ADHD group, 3 from the autistic group, 2 from the TD group) and 37 declined further participation (20 from the ADHD group, 9 from the autistic group, 8 from the TD group). We examined differences between those who participated at T3 and those dropping out in terms of age, sex, intellectual functioning, mothers’ educational level, and anxiety and depressive symptoms. Those who dropped out had significantly lower baseline IQ compared to those who were retained (p =.006).

See Table 3 for descriptive statistics on anxiety and depressive symptoms across the three assessment points. The first model with equality constraints for all effects exhibited unacceptable model fit (χ2 (35) = 41.82, p =.199, RMSEA = 0.09, CFI = 0.93, SRMR = 0.14). We modified the model by removing the equality constraints on the parameters of main interest, the cross-lagged effects, and let them be estimated separately for each group. The second model, where the cross-lagged paths were allowed to vary across group, showed acceptable model fit (χ2 (31) = 24.84, p =.775, RMSEA = 0.04, CFI = 0.99, SRMR = 0.10).

Figure 1 illustrates the final models. Among neurodivergent youth, anxiety symptoms at T1 (β = 0.50, SE = 0.10, Z = 4.89, p <.001, 95% CI [0.30, 0.71]) and T2 (β = 0.43, SE = 0.12, Z = 3.54, p <.001, 95% CI [0.19, 0.66]) significantly predicted subsequent depressive symptoms at T2 and T3, respectively, with large effect sizes. These results are consistent with the prodromal perspective which suggests that impairing anxiety may precede the onset of greater depressive symptoms over time in children and adolescents. Among neurotypical youth, all cross-lagged effects were non-significant. Across groups, there were significant and strong concurrent associations between anxiety and depressive symptoms at all assessment waves (r =.48 to 0.62, p <.001). From T1 to T2, a significant autoregressive effect was found for anxiety symptoms (β = 0.50, SE = 0.09, Z = 5.47, p <.001, 95% CI [0.32, 0.67]). However, when the time gap increased between T2 and T3, the autoregressive effect for anxiety symptoms was no longer significant (β = 0.23, SE = 0.12, Z = 1.83, p =.067, 95% CI [-0.02, 0.47]). This suggests, not surprisingly, that the rank-order stability in anxiety symptoms is stronger over short (i.e., two years) than longer (i.e., eight years) time spans.

As a supplementary analysis, we examined whether the results held up when dividing the neurodivergent youth into two separate groups, autistic youth and youth with ADHD. The results showed that the cross-lagged effects from anxiety symptoms to subsequent depressive symptoms were present in both groups (see Fig. 2). Thus, the main findings held up when the two groups were examined separately.

The current findings have several clinical implications. First and foremost, the findings suggest that preventing and addressing anxiety symptoms in clinical work with neurodivergent youth may be indicated to avoid negative developmental cascades. Building on the model by Wood and Gadow (2010) and the neurodiversity framework (Sonuga-Barke, 2023), one approach could be the implementation of programs that aim to remove societal barriers for neurodivergent youth and increase acceptance of their differences. Such programs might prevent anxiety symptoms from developing by reducing the number of aversive everyday situations neurodivergent youth encounter. When anxiety symptoms emerge, anxiety treatment adapted to the needs of neurodivergent youth (e.g., Wood et al., 2020, 2021) may be important to prevent the development of depressive symptoms. A recent study of cognitive-behavioral treatment for youth with anxiety disorders found that the anxiety symptoms trajectory predicted the depressive symptoms trajectory, i.e., when anxiety symptoms increased or decreased, depressive symptoms developed accordingly (Fjermestad et al., 2024). Similarly, Wood et al. (2020) found that autism-adapted anxiety treatment was associated with less depressive and general internalizing symptoms compared to standardized anxiety treatment or treatment as usual. Thus, adequate treatment of anxiety in neurodivergent youth may lessen the development of depressive symptoms among those with underlying risk for internalizing difficulties.

The strengths of the current study include the controlled longitudinal design with a high retention rate 10 years after baseline assessment. Furthermore, the inclusion of autistic youth as well as those diagnosed with ADHD allows for the inclusion of a relatively large sample of youth with highly related manifestations of developmental diversity. Nonetheless, there are several limitations.

First, even though our sample size of neurodivergent youth was similar to that in the previous study by Schiltz et al. (2023; N = 130), the sample sizes of the individual groups are modest, and our findings should be considered preliminary. To preserve statistical power, we collapsed neurodivergent youth (i.e., autistic youth and youth with ADHD) into one group for the main analyses. This prevents us from saying anything about possible differences between the two groups of neurodivergent youth. As the study by Schiltz et al. (2023) also included several neurodevelopmental phenomena in the analyzed sample, there is still a question of whether different neurodevelopmental phenomena may differ in terms of the developmental relationship between anxiety and depressive symptoms. However, our supplementary analysis indicated that the main results held up separately in the autistic group and the group of youth with ADHD.

Second, although our sample size was similar to that of Schiltz et al. (2023), power may still be an issue. There are examples of cross-lagged regression analyses in the literature with samples as small as N = 54 (see Orth et al., 2024 for a recent review). However, the recommended sample sizes for structural regression models vary, but a common recommendation is N of ~ 200 (Baribeau et al., 2022). Our total sample size was just below the recommendation, and more considerably so if considering the neurodivergent and neurotypical youth as two separate samples. However, simpler models and larger effect sizes necessitate smaller samples than more complex models involving the estimation of latent variable parameters and subtle effects (Baribeau et al., 2022; Kline, 2023). In the current study, we used a relatively simple model comprising two manifest variables measured on three occasions, and we expected (and found) large effect sizes based on previous research (Schiltz et al., 2023). Thus, even though our findings should be considered preliminary, and replications are warranted, our findings contribute to a currently small part of the literature examining the developmental associations between anxiety and depressive symptoms among neurodivergent individuals.

Third, there was evidence of selective attrition due to lower IQ, which should be considered when interpreting our findings. At the 10-year follow-up, the sample comprised a subgroup that was relatively intellectually able. However, to control for this selective attrition, IQ was added as a covariate in the analyses. Fourth, our sample was clinically referred and may not be generalizable to the full population of neurodivergent youth. Thus, in conclusion, more research is needed to understand exactly how anxiety symptoms influence depressive symptoms in neurodivergent youth. A better understanding of this relationship may inform program development aimed at supporting the mental health and well-being of neurodivergent young people.

Fifth, our sample of youth with ADHD was stimulant naïve at baseline, and a stimulant naïve sample at the mean age of 11 years may differ from other clinical samples of youth with ADHD. For example, a recent systematic review suggested that a later age of diagnosis is associated with fewer co-occurring conditions (Rocco et al., 2021). Our sample also comprised more females (46%) than what is common in clinical samples (Faraone et al., 2015). This may be due to females being more likely to be diagnosed at a later age and less likely to be prescribed stimulants (Martin, 2024). However, a later diagnosis of females with ADHD may be partially due to diagnostic overshadowing, where females are more likely to be diagnosed with co-occurring anxiety and depression prior to receiving an ADHD diagnosis (see Martin, 2024 for a recent discussion). Also, although we lack data on the ethnicity of our participants, one should keep in mind that the sample was recruited from a rural part of Norway where the population is predominantly European-White. Thus, the particularities of our sample should be kept in mind when drawing conclusions and generalizations from our findings.

Lastly, we were limited by the available data. We were not able to examine possible explanatory mechanisms of the relationship between anxiety and depressive symptoms, such as avoidance, excessive worrying, and social aspects. Future studies should include a range of possible explanatory mechanisms to elucidate exactly how anxiety symptoms may contribute to subsequent depressive symptoms. Furthermore, in the future, it would be interesting to use the tripartite model of anxiety and depression (Anderson & Hope, 2008; Clark & Watson, 1991) and examine the temporal relationships between general negative affect, physiological hyperarousal, and lack of positive affect.