Main

Cognitive dysfunction is a concurrent feature of multiple mental illnesses among adults and youth. Patients’ performance on neuropsychological tests can be used to quantify cognitive dysfunction. These tests involve performing tasks and can provide objective measurements of cognitive domains such as set shifting, response inhibition, working memory, fluency, planning, verbal memory, nonverbal memory, processing speed, attention and visuospatial function1.

Regarding findings in adults, a 2021 systematic review of meta-analyses2 compared cognitive performance in 97 samples of individuals suffering from various psychiatric disorders to cognitive performance in healthy reference data. Results indicated an impairment in at least one cognitive domain in patients with autism spectrum disorder, bipolar disorder, eating disorders, depression, schizoaffective disorder, obsessive-compulsive disorder, personality disorders, post-traumatic stress disorder, schizophrenia, substance use disorders and Tourette’s syndrome, with effect sizes ranging from 0.4 to 0.6 (ref. 2). In particular, disorders characterized by psychotic symptoms, such as schizophrenia, showed the largest deficits in cognitive performance. This is in line with previous evidence showing that speed of processing, verbal learning and memory, visuospatial learning and memory, working memory, attention and problem solving all show medium to large impairments in patients with psychosis compared to healthy samples3,4,5,6,7,8,9,10,11,12,13,14,15. Conversely, eating disorders and substance use disorders showed the smallest cognitive deficits, although these were detectable across all examined domains2. This is consistent with previous findings showing that patients with anorexia nervosa (‘anorexia’ hereafter), at least in the acute state, perform worse than controls in particular in tasks involving attention, processing speed, memory and visuospatial construction16,17,18,19,20.

Overall, these results suggest that cognitive deficits are a transdiagnostic feature of mental illness, that is, a feature that cuts across traditional diagnostic boundaries. Therefore, they are not limited to disorders using cognitive dysfunction as part of their diagnostic criteria. So far, studies of clinical samples have largely investigated cognitive impairments in adults with mental illness, despite the majority of mental illnesses emerging in childhood and adolescence21,22,23,24. Mental illnesses are the leading cause of disability in children and adolescents, affecting an estimated 13.4% of youth worldwide25,26. Evidence suggests that cognitive deficits associated with mental illnesses begin to manifest before adulthood27,28,29,30,31,32. For example, in youth suffering from bipolar disorder, research using multiple assessments has shown moderate to large impairments in verbal learning, verbal memory, working memory, visual learning and visual memory33,34. In early onset schizophrenia, studies have found large deficits in general intellectual ability, processing speed, working memory, verbal memory and learning; medium deficits in rule discovery and perseveration, planning and problem solving; and minimal deficits in attention35. In youth with attention deficit and hyperactivity disorder (ADHD), performance on neurocognitive tests has consistently shown decreased overall cognitive ability, as well as worse performance across neuropsychological measures of learning, spanning memory and executive function36,37,38,39,40. In youth with anorexia, studies have found poorer nonverbal performance, altered attention to disorder-related stimuli, perceptual processing impairment in discriminating body images, weaknesses in central coherence, set-shifting weaknesses at low weight status, decision-making weaknesses and greater neural resources required for working memory41. Finally, studies in youth with major depressive disorder have found deficits in psychomotor speed, attention, memory and executive function42.

Previously, we have added to this literature by using standardized cognitive tests to assess cognitive impairments in youth within individual diagnostic groups, including ADHD43,44, first onset psychosis45, anorexia17 and functional neurological disorder46. However, it is unclear how different cognitive domains are affected differently within diagnoses and how the dysfunctions of each cognitive domain in one mental illness compare transdiagnostically to dysfunctions of the same domain in other illnesses. In this Article, a unique focus is a new transdiagnostic evaluation comparing cognitive performance in patients aged 6–18 years diagnosed with ADHD, anorexia, first onset psychosis and functional neurological disorder with the cognitive performance of age-matched healthy controls. Prior findings are based on multiple different types of assessment batteries, but we assessed cognitive performance using the same standardized web-based cognitive battery in all groups1, thus enabling us to quantify cognitive deficits in each clinical diagnosis consistently. Also, by collecting the same battery of tests across diagnoses, we were able to directly compare the performance in each cognitive domain between every diagnosis pair and, vice versa, the performance between every cognitive domain pair in each diagnosis. We focused on the cognitive domains of attention, planning, response inhibition, verbal fluency, verbal memory, visual memory and working memory to align with previous adult literature. We hypothesized that youth suffering from the included diagnoses would display cognitive deficits compared to age-matched controls and that findings would mirror previous findings in adult clinical populations.

Results

A summary of the demographic characteristics of the participants is given in Table 1.

Table 1 Study demographics
Full size table

We used a linear mixed model to assess the impact of each diagnosis on performance in our cognitive domains of interest. Consistent with our hypotheses, all diagnoses showed an impairment in at least one cognitive function measure compared to healthy matched controls and each diagnosis showed a distinctive profile of cognitive dysfunction (cognitive domain × diagnosis interaction; F = 8.148, P < 0.001).

Figures 1 and 2 summarize the performance of each group of participants for each cognitive domain. Table 2 presents the results of all significant comparisons between the cognitive performance in each diagnosis and healthy controls. Table 3 presents the results of all significant comparisons between each pair of diagnoses within each cognitive domain (see Supplementary Table 1 for all comparisons, including nonsignificant ones). Supplementary Table 2 presents the results of all significant comparisons between each pair of cognitive domains within each diagnosis, and Supplementary Table 3 presents all comparisons, including nonsignificant ones.

Fig. 1: Cognitive performance of patients and controls across cognitive domains grouped by diagnosis.
figure 1

The bar plots show the effects estimated by the linear mixed model predicting cognitive performance. Mean predicted cognitive performance is expressed in standard deviations relative to a healthy norm (z scores), accounting for model covariates (data collection site) and repeated measures. Controls (N = 483), ADHD (N = 343), anorexia (N = 40), FND (N = 56) and first onset psychosis (N = 25) independent participants were examined over nine domains of cognitive function. Bars indicate mean effects, and whiskers indicate standard errors. FND, functional neurological disorder.

Full size image
Fig. 2: Cognitive performance of patients and controls across cognitive domains grouped by diagnosis.
figure 2

The bar plots show the estimate by the linear mixed model predicting cognitive performance. Mean predicted cognitive performance is expressed in standard deviations relative to a healthy norm (z scores), accounting for model covariates (data collection site) and repeated measures. Controls (N = 483), ADHD (N = 343), anorexia (N = 40), FND (N = 56) and first onset psychosis (N = 25) independent participants were examined over nine domains of cognitive function. Bars indicate mean effects, and whiskers indicate standard errors. FND, functional neurological disorder.

Full size image
Table 2 Comparisons of the mean cognitive performance of each diagnosis and that of healthy controls
Full size table
Table 3 Comparisons of the mean cognitive performance of each pair of diagnoses within each cognitive domain
Full size table

The following sections outline the cognitive deficits of each diagnosis in detail.

ADHD

Patients with ADHD performed worse than controls across all domains (see Table 2, Fig. 1 for model effects and Supplementary Fig. 1 for raw data). Deficits within verbal memory (mean difference 0.75, s.e.m. 0.07, t = 10.42, Pcorr < 0.001, d = 1.08) and sustained attention were the strongest (mean difference 0.73, s.e.m. 0.07, t = 10.14, Pcorr < 0.001, d = 1.048). The ADHD group was the only group to display significant dysfunction in psychomotor function compared to healthy controls (mean difference 0.42, s.e.m. 0.08, t = 5.41, Pcorr < 0.001, d = 0.598). Other domains negatively affected include executive function (mean difference 0.67, s.e.m. 0.07, t = 9.19, Pcorr < 0.001, d = 0.970), response inhibition (mean difference 0.63, s.e.m. 0.07, t = 8.94, Pcorr < 0.001, d = 0.91), cognitive flexibility (mean difference 0.42, s.e.m. 0.07, t = 5.87, Pcorr < 0.001, d = 0.599), decision speed (mean difference 0.45, s.e.m. 0.08, t = 5.69, Pcorr < 0.001, d = 0.644), information processing speed (mean difference 0.36, s.e.m. 0.07, t = 5.01, Pcorr < 0.001, d = 0.513) and working memory (mean difference 0.55, s.e.m. 0.07, t = 7.78, Pcorr < 0.001, d = 0.794).

Patients with ADHD performed significantly worse than patients with anorexia in tasks measuring information processing speed (mean difference −0.48, s.e.m. 0.16, t = −2.96, Pcorr = 0.026, d = −0.685) and worse than patients with functional neurological disorder in response inhibition tasks (mean difference −0.48, s.e.m. 0.16, t = −3.08, Pcorr = 0.018, d = −0.689; see Table 3, Fig. 2 for model effects and Supplementary Fig. 2 for raw data).

Anorexia

Patients with anorexia displayed dysfunction in five out of the nine total cognitive domains tested (~56%; see Table 2, Fig. 1 for model effects and Supplementary Fig. 1 for raw data). Performance was worst in tasks measuring response inhibition (mean difference 1.07, s.e.m. 0.18, t = 6.14, Pcorr < 0.001, d = 1.548) and verbal memory (mean difference 0.71, s.e.m. 0.16, t = 4.33, Pcorr < 0.001, d = 1.021). Other domains negatively affected include cognitive flexibility (mean difference 0.46, s.e.m. 0.16, t = 2.84, Pcorr = 0.037, d = 0.668), sustained attention (mean difference 0.51, s.e.m. 0.17, t = 2.90, Pcorr = 0.031, d = 0.73) and decision speed (mean difference 0.51, s.e.m. 0.18, t = 2.85, Pcorr = 0.036, d = 0.728).

Patients with anorexia performed significantly worse than the functional neurological group in tasks testing response inhibition (mean difference −0.92, s.e.m. 0.20, t = −4.56, Pcorr < 0.001, d = −1.327; Table 3 and Fig. 2).

First onset psychosis

Compared to their healthy counterparts, patients with first onset psychosis displayed significant dysfunction in the majority of cognitive domains tested (six out of nine, ~67%; see Table 2, Fig. 1 for model effects and Supplementary Fig. 1 for raw data). Patients with first onset psychosis displayed striking deficits in sustained attention (mean difference 1.88, s.e.m. 0.20, t = 9.38, Pcorr < 0.001, d = 2.716), verbal memory (mean difference 1.69, s.e.m. 0.19, t = 8.77, Pcorr < 0.001, d = 2.433) and response inhibition (mean difference 1.46, s.e.m. 0.20, t = 7.28, Pcorr < 0.001, d = 2.107). Other domains affected include cognitive flexibility (mean difference 1.03, s.e.m. 0.19, t = 5.37, Pcorr < 0.001, d = 1.489), information processing speed (mean difference 0.92, s.e.m. 0.20, t = 4.72, Pcorr < 0.001, d = 1.327) and working memory (mean difference 0.96, s.e.m. 0.19, t = 5.01, Pcorr < 0.001, d = 1.388).

Compared to other diagnostic groups, patients with first onset psychosis showed particularly pronounced overall cognitive dysfunction. The first onset psychosis group performed significantly worse compared to at least one other diagnostic group in almost every cognitive domain, with the exception of executive function. The most distinctive patterns of differences between patients with first onset psychosis and other diagnostic groups were seen in tasks measuring verbal memory and sustained attention (see Table 3, Fig. 2 for model effects and Supplementary Fig. 2 for raw data).

Functional neurological disorder

Patients with functional neurological disorder displayed deficits in five out of the nine tested cognitive domains (~56%; see Table 2, Fig. 1 for model effects and Supplementary Fig. 1 for raw data). Deficits in executive function appeared to be an exceptional struggle for patients with functional neurological disorder (mean difference 1.13, s.e.m. 0.18, t = 6.45, Pcorr < 0.001, d = 1.627). Patients with functional neurological disorder also particularly underperformed in tasks challenging working memory (mean difference 0.89, s.e.m. 0.16, t = 5.54, Pcorr < 0.001, d = 1.285). Other domains affected include cognitive flexibility (mean difference 0.50, s.e.m. 0.16, t = 3.14, Pcorr = 0.015, d = 0.726), decision speed (mean difference 0.55, s.e.m. 0.16, t = 3.43, Pcorr = 0.006, d = 0.799) and information processing speed (mean difference 0.76, s.e.m. 0.16, t = 4.74, Pcorr < 0.001, d = 1.094).

Patients with functional neurological disorder displayed greater dysfunction than patients with anorexia in multiple domains. The functional neurological disorder group performed significantly worse than those with anorexia in tasks measuring executive function (mean difference 0.79, s.e.m. 0.21, t = 3.78, Pcorr = 0.002, d = 1.144), information processing (mean difference 0.88, s.e.m. 0.19, t = 4.54, Pcorr < 0.001, d = 1.267) and working memory (mean difference 0.74, s.e.m. 0.20, t = 3.80, Pcorr = 0.001, d = 1.068; see Table 3, Fig. 2 for model effects and Supplementary Fig. 2 for raw data).

Discussion

Previous investigations of adults have shown that cognitive dysfunction is a key feature of mental illness regardless of specific diagnosis. We aimed to extend these findings to children and adolescents suffering from ADHD, anorexia, first onset psychosis and functional neurological disorder. Consistent with our hypotheses, all clinical groups displayed impairments within at least one cognitive domain compared to age-matched healthy controls. Our findings suggest that cognitive dysfunction exists as a transdiagnostic feature of mental illness that can be identified with standardized cognitive testing as early as childhood and adolescence.

Each clinical group sampled displayed a distinct pattern of cognitive dysfunction. Children and adolescents with ADHD showed a diffuse pattern of differences in performance with healthy controls, while displaying especially exceptional impairment in verbal memory and sustained attention. This broad impairment found in our sample aligns with results in adult literature, which indicated impairments in response inhibition, working memory and executive function47,48,49,50.

Compared to ADHD, cognitive deficits in youth anorexia were more specific and limited to cognitive flexibility, sustained attention, decision speed, response inhibition and verbal memory. This is consistent with previous findings showing that patients with anorexia have smaller but still detectable performance deficits across several cognitive domains compared to controls2,16,17,18,19,20.

Youth with first onset psychosis displayed the most severe impairment of any clinical group in the sustained attention, verbal memory, response inhibition, cognitive flexibility, information processing speed and working memory domains. This pattern and intensity of dysfunction was consistent with what has been detected within adult samples2,3,4,5,6,7,8,9,10,11,12,13,14,15. Furthermore, it also matched what was found in other previous studies focusing on early onset psychosis in adolescents51,52,53,54,55,56,57. Importantly, evidence suggests that patients with youth onset schizophrenia have more severe cognitive deficits than those with late onset schizophrenia54 and that these cognitive deficits are relatively stable over time57,58,59. Thus, early detection of cognitive impairment through behavioral testing could be important to intervene on these impairments early on.

Cognitive dysfunction in youth with functional neurological disorder was limited to executive function, working memory, cognitive flexibility, decision speed and information processing speed. Executive dysfunction compared to healthy controls uniquely distinguished youth with functional neurological disorder from other diagnostic groups. Although sparse, adult literature also suggests impairments in working memory in these patients60. Limited evidence also shows that adults with functional neurological disorder also suffer from executive function deficits, particularly planning60.

Deficits in cognitive flexibility appeared to be the most stable dysfunction across diagnostic groups, as all four clinical groups sampled performed significantly worse than their healthy counterparts. Within the cognitive flexibility domain, first onset psychosis displayed a significant difference in performance compared to patients with ADHD.

Our findings support the assertion that cognitive impairment is a pervasive feature of mental illness, even in diagnoses that do not include cognitive impairment as a core diagnostic criterion (such as anorexia). Importantly, our findings also demonstrate that cognitive impairment in mental illness can be detected as early as childhood or adolescence with standardized computerized testing.

These results additionally present important future research and clinical implications. First, the transdiagnostic nature of overall cognitive dysfunction but unique patterns of specific impairments among diagnoses suggests that objective measurement of cognition can potentially serve as a helpful tool for mental health clinicians for early detection and differentiation of mental illnesses in youth. This is especially important considering our results derived from youth as young as 6 years of age. Traditionally, many mental health diagnostic criteria rely on patients’ ability to recognize their dysfunction or emotions and be able to explain them in detail. This reliance on self-reported symptoms presents unique barriers for youth mental health diagnostics due to developmental realities. Future research can build on this work by further investigating cognitive signatures in other widely experienced psychiatric disorders in youth such as anxiety and depression. This study also suggests that cognitive dysfunction could be a target for treatment using strategies such as cognitive remediation that has been identified as effective in disorders such as schizophrenia61.

Our team’s prior work suggests that up to 27% of adult patients suffering from depression display cognitive impairments measured by the same tests used in the current study62. These patients are less likely to remit with standard antidepressant treatment and are likely to continue experiencing cognitive impairments despite any improvement in mood symptoms. Future studies should investigate the relationship of cognitive dysfunction and treatment outcomes among various youth mental health diagnoses. Another important future direction would be to investigate whether findings generalize to other widely used cognitive batteries (for example, ref. 63). Finally, these findings could be expanded by investigating the neuroimaging correlates of cognitive dysfunction in children and adolescents with mental illness64,65. For example, impaired segregation of task-positive and task-negative brain networks has been found in ADHD and could explain the diffuse performance deficit in this disorder66. In first onset (as well as chronic) psychosis, efficiency of the frontoparietal network and cingulo-opercular networks was associated with worse performance on standardized tests of general cognition67. Cognitive deficits in anorexia have been found to be associated with disruption of connectivity between the thalamus and prefrontal cortex68. Children and adolescents with functional neurological disorder showed altered connectivity in several brain networks, but to the best of our knowledge, how this relates to cognitive performance has not yet been investigated. Extending these neuroimaging insights could help advance future precision approaches for early treatment of developmental and mental health disorders.

The current study is subject to limitations. First, as referenced earlier, the sample analyzed features of only four clinical diagnoses, since these were the only ones available in the BrainNet database, namely ADHD, anorexia, first onset psychosis and functional neurological disorder. Future studies should extend investigations of cognitive impairments to other mental illnesses common in childhood and adolescence, such as mood (for example, bipolar disorder and major depressive disorder) and anxiety disorders. As every diagnostic group in our sample featured some degree of cognitive impairment, we anticipate that future studies investigating other diagnoses will produce similar results. Also, even though our dataset allowed us to compare results across several diagnoses, the sample sizes for each were heterogeneous. In particular, future studies investigating cognitive performance in anorexia, first onset psychosis and functional neurological disorder using larger samples than ours might be able to detect subtle performance deficits in these diagnoses that we could not detect. Next, the current study was not able to consider the impact of social determinants relevant to brain health such as education access and quality, economic stability and social and community context due to inconsistent reporting of relevant data from the included protocols. Future studies should make the effort to collect and examine these important demographic factors related to brain development and overall brain health. Additional limitations include the cross-sectional nature of the current results. This is particularly pertinent for disorders such as anorexia and first onset psychosis that often worsen over time. Future endeavors should include longitudinal analyses to evaluate the effect of illness duration on cognition. Finally, another limitation was that we could not consider the potential confounding effects of treatment in analyses. Future studies should account for the potential impact of treatments on cognitive performance by clinical diagnosis and assess if these treatments improve cognitive deficits.

Methods

Participants

The data used in the present work were downloaded from BrainNet, a large database for mental health research, which combines data from a number of separate studies (access to BrainNet can be requested69). In particular, the current sample was recruited through clinical services associated with the University of Sydney, University of Adelaide and University of Melbourne and through advertising and self-referral. Inclusion criteria were age between 6 and 18 years old; diagnosis of ADHD, anorexia, first onset psychosis and functional neurological disorder (clinical groups) or no psychiatric diagnosis (controls); capacity to undergo a computerized test; normal (or corrected to normal) vision; and ability to use a keyboard. No data from minors with other diagnoses were available in BrainNet; therefore, we focused on ADHD, anorexia, first onset psychosis and functional neurological disorder. The only exception to this was one single minor with post-traumatic stress disorder and two matched controls from the same site, which we removed from our analyses. In this sample, with ‘first onset psychosis’ we mean minors who had experienced their first episode of a psychosis with onset before 18 years. Participants were excluded if they had a developmental delay (assessed by intelligence quotient testing), a known medical condition such as epilepsy, a history of head trauma causing a loss of consciousness for 10 min or more or substance dependence. Each included dataset and corresponding study was approved by the Sydney West Area Health Service’s and Children’s Hospital at Westmead’s human research ethics committees. Approval numbers are as follows: psychosis sample and corresponding healthy controls (‘the first episode project’), WSAHS HREC 98/12/3.3(682); ADHD sample and corresponding healthy controls, HREC 2006/12/4.5(2502); functional neurological disorder and corresponding healthy controls, HREC/2006/8/4.15(2415); and anorexia and corresponding health controls, HS/TG HREC2003/12/4.14 (1785). All participants and parents/guardians, as appropriate, signed and dated an approved informed consent form. Where participants consented, their data were made available for open sharing in BrainNet. All research complies with the Code of Ethics of the World Medical Association (Declaration of Helsinki).

Diagnosis of clinical groups was confirmed by consensus from clinicians using the following criteria and scales. For ADHD, Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders fifth edition (DSM-IV) and Conners’ Rating Scale70 were used; for details see ref. 43. For anorexia, Structured Clinical Interview (DSM-IV) and physical conditions were used; for details see ref. 17. For first onset psychosis, Structured Clinical Interview (DSM-IV) and Positive and Negative Syndrome Scale71 were used; for details see ref. 45. For functional neurological disorder, Structured Clinical Interview (DSM-IV) was used; for details see ref. 46. For each clinical group, an age- and sex-matched sample of healthy individuals was recruited as a control group (Table 1).

Race, ethnicity and other socially relevant groupings were not included in the current analysis. This is because the data sourced for this study were archival data and race, ethnicity and other socially relevant groupings were not present in all available data.

Neurocognitive assessments

Cognitive performance was assessed using a standardized, computerized test battery, ‘IntegNeuro’, which has established norms across nine decades of the healthy lifespan, test–retest reliability72, construct validity with respect to traditional neuropsychological batteries and brain measures73, and utility in distinguishing cognitive impairments in psychiatric groups1,74. IntegNeuro assessed nine cognitive domains (and tests): executive function (Maze), response inhibition (Go-No-Go), cognitive flexibility (Stroop), verbal memory (Verbal Learning and Memory), sustained attention (Continuous Performance Test), decision speed (Choice Reaction Time), psychomotor function (Motor Tapping), information processing speed (Switching of Attention) and working memory (Digit Span). All tests were visual. See Supplementary Table 4 for a description of these tests. IntegNeuro provides performance measures for each of the tests named above, expressed on the basis of their deviation from a reference healthy population and matched by sex and age. A score for each cognitive domain was then obtained by averaging performance on each test within each of the nine domains (Supplementary Table 4). Scores >5 s.d. from the norm were considered outliers and removed (number of values removed: executive function, 11, 1.16%; response inhibition, 16, 1.69%; cognitive flexibility, 0, 0%; verbal memory, 0, 0%; sustained attention, 1, 0.11%; decision speed, 7, 0.74%; psychomotor function, 15, 1.58%; information processing speed, 1, 0.11%; and working memory, 0, 0%).

Statistical analysis

All analyses were conducted in RStudio 2022.07.2, R version 4.1.3, running on macOS Ventura 13.2. Code for these analyses is available at ref. 75.

Mixed linear model

We modeled the cognitive domain scores for each participant using a mixed linear model to take into account the fact that performance in all nine cognitive domains was measured in each participant. The model independent variables were diagnosis (factor with levels: control, ADHD, anorexia, first onset psychosis and functional neurological disorder), cognitive domain (factor with levels: sustained attention, cognitive flexibility, decision speed, executive function, information processing speed, psychomotor response speed, response inhibition, verbal memory and working memory) and site (factor with levels: Adelaide, Melbourne and Sydney); one random intercept was fit for every participant, and the dependent variable was the IntegNeuro performance score.

In mathematical notation, the model can be written as follows:

$$\begin{array}{l}{{\rm{Performance}}}_{ijk}={\beta }_{0}+{u}_{i}+{\beta }_{1}\times {{\rm{Diagnosis}}}_{ijk}+{\beta }_{2}\times {{\rm{Domain}}}_{ijk}\\ \qquad\qquad\qquad\quad+\,{\beta }_{3}\times ({{\rm{Diagnosis}}}_{ijk}\times {{\rm{Domain}}}_{ijk})+{\beta }_{4}\times{\mathrm{Site}}_{{ijk}}+{{\epsilon}}_{{ijk}}\end{array}.$$

In this formula:

  • β0 is the average intercept across all subjects.

  • ui is the random intercept for the ith subject.

  • β1, β2, β3 and β4 are the fixed effect coefficients for diagnosis, domain, the diagnosis × domain interaction and site, respectively.

  • Diagnosisijk is the value of the predictor variable ‘diagnosis’ for the ith subject, measured for the jth domain at the kth site.

  • Domainijk is the value of the predictor variable ‘domain’ for the ith subject, measured for the jth domain at the kth site.

  • Siteijk is the value of the predictor variable ‘site’ for the ith subject, measured for the jth domain at the kth site.

  • ϵijk is the residual error for the ith subject, measured for the jth domain at the kth site, assumed to follow a normal distribution ϵ ∼ N(0, σ2).

After fitting our mixed linear model, we used F tests to assess the significance of the main effects of all our predictors, as well as of the cognitive domain × diagnosis interaction, to test whether different diagnoses had different profiles of cognitive dysfunction across domains. We considered significant tests for which P < 0.05. We followed up the significant cognitive domain × diagnosis interaction by comparing the performance in each cognitive domain between each pair of diagnoses using t-tests. These tests were adjusted for multiple comparisons by using Tukey’s method and were considered significant if Pcorr < 0.05.

Power simulation

To determine our power to detect a significant cognitive domain × diagnosis interaction using our mixed linear model given our sample size, we conducted a simulation (see Supplementary Methods for details). The simulation showed that, given a mean difference in cognitive performance between clinical participants and controls of at least 0.3 and given a number of dysfunctional cognitive domains in each clinical group between 2 and 6, using an α = 0.05, we would achieve >80% power to detect our interaction of interest (Supplementary Fig. 3).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.