Modelling associations between neurocognition and functional course in young people with emerging mental disorders: a longitudinal cohort study

Neurocognitive impairment is commonly associated with functional disability in established depressive, bipolar and psychotic disorders. However, little is known about the longer-term functional implications of these impairments in early phase transdiagnostic cohorts. We aimed to examine associations between neurocognition and functioning at baseline and over time. We used mixed effects models to investigate associations between neurocognitive test scores and longitudinal social and occupational functioning (“Social and Occupational Functioning Assessment Scale”) at 1–7 timepoints over five-years in 767 individuals accessing youth mental health services. Analyses were adjusted for age, sex, premorbid IQ, and symptom severity. Lower baseline functioning was associated with male sex (coefficient −3.78, 95% CI −5.22 to −2.34 p < 0.001), poorer verbal memory (coefficient 0.90, 95% CI 0.42 to 1.38, p < 0.001), more severe depressive (coefficient −0.28, 95% CI −0.41 to −0.15, p < 0.001), negative (coefficient −0.49, 95% CI −0.74 to −0.25, p < 0.001), and positive symptoms (coefficient −0.25, 95% CI −0.41 to −0.09, p = 0.002) and lower premorbid IQ (coefficient 0.13, 95% CI 0.07 to 0.19, p < 0.001). The rate of change in functioning over time varied among patients depending on their sex (male; coefficient 0.73, 95% CI 0.49 to 0.98, p < 0.001) and baseline level of cognitive flexibility (coefficient 0.14, 95% CI 0.06 to 0.22, p < 0.001), such that patients with the lowest scores had the least improvement in functioning. Impaired cognitive flexibility is common and may represent a meaningful and transdiagnostic target for cognitive remediation in youth mental health settings. Future studies should pilot cognitive remediation targeting cognitive flexibility while monitoring changes in functioning.


Introduction
Reducing the burden of disability attributable to mental disorders is a global health priority. Mental, neurological, and substance use disorders are the world's leading contributors to years lived with disability and the thirdranked cause of disability-adjusted life years 1,2 . This disability burden is particularly heavy for young people. For example, disability-adjusted life years related to common mental disorders reach their peak between ages 10-29 years 2 , and depression, bipolar disorder, and schizophrenia are three of the four most burdensome conditions in those aged 10-24 years 3 . As 50% of mental disorders emerge before the middle-teens and 75% by the midtwenties 4 , it is likely that the burden of disability in early and later adulthood represents an extension of problems originating premorbidly or in early phases of illness. These early phases are characterized by changes in social behaviors and difficulties participating in work and study 5 , both of which are reflected in the high rates of functional impairment at presentation to youth-specific services 6,7 . As functional trajectories are heterogeneous and often persistently impaired 6 , there is a critical need for identification of factors associated with functional course which may guide interventions to those at risk of enduring and lifelong functional disability 8 .
Despite clinical remission, many individuals with mental disorders fail to reach their expected or premorbid levels of social and occupational functioning, suggesting the presence of enduring factors which limit functional recovery. One such factor is neurocognitive impairment, which is common from early in the course of depressive 9 , bipolar 10 , and psychotic disorders 10,11 . While there is good evidence to suggest that neurocognitive impairments represent an enduring and trait-like feature of neurodevelopmental disorders such as schizophrenia 12 , the state-trait distinction is less clear for other mental disorders. However, some evidence suggests that attentional and other executive impairments commonly persist despite remission of symptoms in major depression 9,13 , while domains such as memory, verbal fluency, and processing speed are more strongly influenced by mood state 9,13 . Likewise, impairments in memory, processing speed, and executive functions are common in individuals with bipolar disorder across mood episodes 14,15 , notwithstanding common inter-episode syndromal and subsyndromal symptoms 16 .
The functional consequences of neurocognitive impairments in people with mental disorders are increasingly clear. Research over the last two decades has demonstrated that neurocognition is a strong and prospective determinant of functioning in schizophrenia 17 , with impairments in domains including processing speed, executive functions, and memory 18 limiting patients' capacity to acquire, retain, and relearn skills required for adaptive functioning 17 . More recently, similar impacts have been appreciated for a wider range of mental disorders. Longitudinal studies of people with bipolar disorder have demonstrated associations between impairments in executive functions 14,19 , processing speed 14,20 , and verbal learning and memory 14,19,20 and poorer social and occupational outcome. While a less developed literature, poorer memory and executive functions have also been linked to worse follow-up social and occupational outcome in major depression 21,22 .
Importantly, there is a developing notion that neurocognition may represent a continuum cutting across diagnostic boundaries 10,23 , with the National Institute of Mental Health's Research Domain Criteria endorsing a dimensional framework to the study of neurocognition in mental disorders 24 . Our group has demonstrated the utility of a transdiagnostic approach to examining neurocognition and functioning across the major mental disorders. We have reported strong cross-sectional 25 and longitudinal 23,26 relationships between general neurocognition and social and occupational functioning in a cohort of young people presenting to mental health services with a range of mood, anxiety, and psychotic syndromes. Moreover, we have shown that changes in neurocognition map onto changes in functioning when statistically adjusting for diagnosis and symptom severity, supporting a meaningful and robust link between neurocognition and functioning across mental disorders 27 . Limited work however has examined relationships between specific neurocognitive domains and the course of functioning in young transdiagnostic samples.
Accordingly, we aimed to test several questions regarding the links between neurocognition and functioning and their broader implications across mental disorders in a cohort of adolescents and young adults accessing mental health services. First, we aimed to examine associations between neurocognitive test scores across nine domains and functioning at baseline and change in functioning over time. Second, we aimed to determine whether associations between neurocognition and functioning at baseline and change in functioning over time would be robust to adjustment for confounding factors (age, sex, premorbid IQ, and symptom severity). Based on our work 23,26 and the wider literature 19,20,28 , we hypothesized that baseline executive functions, processing speed, and verbal learning and memory would be uniquely associated with baseline functioning and change in functioning longitudinally.

Human ethics
The study and consent procedure were approved by the University of Sydney Human Research Ethics Committee (project numbers 2012/1626 and 2012/1631) and conducted in accordance with the revised Declaration of Helsinki. All participants aged 16 and older provided written informed consent and parental or guardian consent was obtained for participants aged under 16 years.

Participants
Participants were drawn from a cohort of 6743 consecutive referrals (aged 12-30) presenting to youth mental health clinics at the Brain and Mind Center in Sydney, Australia, who were recruited to a research register of adolescents and young adults with mental disorders between 2008-2018. These clinics (e.g., headspace) aim to provide youth-friendly and highly accessible early intervention services for young people with emerging mental and substance use disorders 29 . Headspace consists of an integrated mix of primary-level services and more specialized services (e.g., drug and alcohol) and primarily attracts young people with a wide range of mental health problems (typically anxiety, mood and/or psychotic syndromes). All participants were receiving ongoing clinicianbased case management and relevant social, psychological and/or medical treatments as part of standard care, which may have involved contact with a psychiatrist, psychologist, occupational therapist, support worker or hospitalization for those whose need exceeded the capacity of the primary care services.

Eligibility criteria
Inclusion criteria for this study were: (a) a baseline neurocognitive assessment with the majority of test scores available; (b) aged 12-30 at the neurocognitive assessment; (c) an available proforma assessment (see below) within three-months of the neurocognitive assessment; and (d) willing and able to provide written informed consent (or parental/guardian consent was obtained). Exclusion criteria were: (i) history of neurological disease; (ii) medical illness known to impact brain function (e.g., cancer, epilepsy); (iii) electroconvulsive therapy in threemonths prior to neurocognitive assessment; (iv) clinicallyevident intellectual disability; and/or (v) insufficient understanding of the English language to allow participation in verbal assessments or testing.

Data collection (baseline)
A subset of the wider cohort participated in detailed clinical and neurocognitive assessments between 2008-2015. A board-certified neuropsychologist, research psychologist or supervised doctoral student administered a neurocognitive battery with the domains chosen on the basis of sound validity and reliability 30 , relevance to the diagnoses under study 9,11,31 , and overlap with instruments used in the Measurement and Treatment Research to Improve Cognition in Schizophrenia (MATRICS) initiative 32 . The following domains were assessed: processing speed (Trail Making Test, part A) 33 , cognitive flexibility (Trail Making Test, part B) 33 , verbal learning (sum of trials 1-5 of the Rey Auditory Verbal Learning Test; RAVLT) 34 , verbal memory (20-min delayed recall of the RAVLT) 34 , sustained attention (A' Prime subtest of the Rapid Visual Information Processing Test) 35 , set-shifting (Intra-Extra Dimensional Set Shift) 35 , visuospatial learning (Paired Associates Learning Task) 35 and working memory (Spatial Span Task) 35 . Premorbid IQ was estimated using the Wechsler Test of Adult Reading 36 or the Wide Range Achievement Test 37 (for participants younger than 16). Neurocognitive scores were standardized to age-matched and sex-matched norms (z-scores) using established norms 38,39 . To limit the impact of extreme scores and minimize data transformation, zscores were curtailed at a maximum of ±5.0, with fewer than 3% of scores curtailed for each test. Symptom type and severity were determined using the Brief Psychiatric Rating Scale, with four dimensions empirically derived (depressive, negative, positive, and manic) 40 .

Data collection (longitudinal)
A standardized clinical proforma was used to retrospectively extract demographic, clinical, and functioning data from clinical and research files across eight predetermined timepoints (baseline, 3 months, 6 months, 1 year, 2 years, 3 years, 4 years, and 5 years) 41 . A "timelast-seen" entry was also recorded; however, this was not included in the current study. The proforma captures information at each timepoint regarding the current presentation and illness course, including: (a) demographics; (b) socio-occupational functioning; (c) clinical presentation (including clinical diagnosis according to DSM-5 42 ); (d) self-harm and suicidal thoughts and behaviors; (e) alcohol and other substance use; and (f) physical health comorbidities.
The proforma provided the primary outcome measure of socio-occupational functioning as assessed by a trained clinician using the Social and Occupational Functioning Assessment Scale (SOFAS). The SOFAS is a 100-point scale (with higher scores denoting better functioning) which improves on other measures of global functioning in its instruction to the rater to avoid confounding the rating with symptoms. A score of 60-70 is indicative of moderate difficulty in social, occupational, or school functioning. The SOFAS is widely used and has good construct validity 43 , inter-rater reliability 43 , and predictive validity 44 .
As neurocognition was the primary baseline predictor for this study, we used the nearest proforma assessment occurring within a three-month interval of the neurocognitive assessment as the participants' baseline timepoint (T1), with remaining proforma timepoints recoded if necessary. As we allowed a three-month interval for recoding, we subsequently excluded the three-month proforma timepoint from further analysis. The number of proforma assessments at each timepoint and the number of participants with one or more proforma assessments over time are presented in Supplementary Tables 1, 2, respectively.

Statistical analyses
Analyses were conducted in RStudio (version 1.0.143). Linear mixed-effects models with random-intercepts were constructed using the "lme4" package (version 1.1-18-1 45 ). Full-information maximum-likelihood estimation was used to handle missing follow-up data (as loss to follow-up was uncontrolled). The mixed-effects framework is recommended for longitudinal designs as it tolerates: (a) repeated-measures within participants (i.e., non-independence); (b) unbalanced assessment intervals; and (c) missing follow-up data. The continuous SOFAS rating at each timepoint represented the outcome variable, and participants could contribute one or multiple assessments over time (i.e., assessments nested within participants) (see Supplementary Table 2). To model associations with the rate of change in SOFAS over time, a "Time" variable was used which represented the timepoint of each assessment and was linearly coded. All baseline predictor variables were continuous (except for sex).
The literature describing relationships between neurocognitive test performance and aspects of functioning in major mental disorders report associations between a large variety of neurocognitive tests and various measures of global functioning and specific subdomains of functioning (e.g., relationship impairment, work impairment, and independent living) 14,[18][19][20][21][22] . Relatedly, most studies have focussed on specific diagnostic groups (e.g., schizophrenia) and there is very limited research regarding specific neurocognition-functioning associations in earlyphase, transdiagnostic cohorts. Accordingly, we chose to use a data-driven, backward elimination statistical approach to identify associations between individual neurocognitive test scores and social and occupational functioning in our cohort. Modeling proceeded in three stages. First, we examined unadjusted associations between all baseline predictors and variation in SOFAS scores at baseline as well as the rate of change in SOFAS over time. Second, we examined associations between all baseline predictors and variation in baseline SOFAS scores, using backward elimination to iteratively remove the least significant variable until only significant predictors remained (α = 0.05). Third, we examined associations between the rate of change in SOFAS longitudinally and all predictor variables that had significant associations with variation in SOFAS at baseline, using backward elimination to reduce the full model.
Normality of residuals was inspected with Q-Q plots, with an approximate normal distribution evident. Multicollinearity was assessed with the variation inflation factor (VIF), with no predictors exceeding a VIF of 3.0. Parameter-specific p-values were calculated using Satterthwaite's approximation for degrees of freedom in the "lmerTest" package (version 1.0 46 ). Missing baseline neurocognitive and clinical data were imputed using multiple imputation by chained equations in the "mice" package (version 3.3.0 47 ). Missing data patterns were consistent with a missing-at-random mechanism and fewer than 10% of each neurocognitive domain and fewer than 12% of BPRS scores were missing (see Supplementary Table 3 for numbers and proportions of missing values for each predictor variable). Following recommendations, we multiply imputed 100 datasets using predictive mean matching (which makes use of all available data), modeled each imputed dataset separately, and pooled the coefficients, test statistics, and p-values [47][48][49] .  I.B.H.). The funders of this study had no involvement in the: study design; collection, analysis and reporting of the data; writing of the report; or decision to submit the paper for publication.

Sample characteristics
Participants were drawn from a cohort of 6743 young people who were recruited to a research register 50 . Of these, 2767 participants had an available baseline clinical proforma assessment, and a total of 767 participants met all eligibility criteria and were included in the final analysis.
Sample characteristics are presented in Table 1 Table 2).
Presenting primary mental health diagnoses are shown in Table 2. The majority of patients presented with a primary mood (depressive or bipolar) or anxiety disorder (N = 523/767, 68.2%). Level of symptoms on the BPRS were in the "very mild" to "mild" range across the four dimensions (depressive, negative, positive, and manic).

Unadjusted associations with baseline functioning and change in functioning over time
We first modeled associations between all baseline predictors and variation in SOFAS at baseline and variation in the rate of SOFAS change longitudinally. As presented in Supplementary Table 5, all variables (except for age at baseline) in the unadjusted models were significantly associated with baseline functioning. Significant positive associations with baseline functioning were observed for all nine neurocognitive domains and premorbid IQ, while significant negative associations were observed for male sex and depressive, negative, positive, and manic symptoms. There were significant and positive associations with the rate of SOFAS change longitudinally for cognitive flexibility, verbal learning, verbal memory, working memory, processing speed, male sex, baseline age, and depressive, positive, and manic symptoms.
Associations with baseline functioning and change in functioning over time, adjusted for socio-demographics and type and severity of symptoms We next used backward elimination to reduce the full model including all predictor variables down to a final model in which only variables significantly associated with SOFAS at baseline remained. This model is presented in Table 3. There were positive associations with baseline functioning for verbal memory (coefficient 0.82, 95% CI 0.33 to 1.32, p = 0.001), cognitive flexibility (coefficient 0.65, 95% CI 0.25 to 1.06, p = 0.002), and premorbid IQ (coefficient 0.12, 95% CI 0.06 to 0.18, p < 0.001), and negative associations for male sex (coefficient −1.87, 95% CI   In a second model also including associations with the rate of SOFAS change over time, there were associations with baseline functioning for verbal memory (coefficient 0.90, 95% CI 0.42 to 1.38, p < 0.001), premorbid IQ (coefficient 0.13, 95% CI 0.07 to 0.19, p < 0.001), male sex (coefficient −3.78, 95% CI −5.22 to −2.34, p < 0.001) and depressive (coefficient −0.28, 95% CI −0.41 to −0.15, p < 0.001), negative (coefficient −0.49, 95% CI −0.74 to −0.25, p < 0.001), and positive symptoms (coefficient −0.25, 95% CI −0.41 to −0.09, p = 0.002). The rate of change in functioning over time varied among patients depending on their sex (male; coefficient 0.73, 95% CI 0.49 to 0.98, p < 0.001), indicating that males had a greater rate of improvement in functioning than females, and the baseline level of cognitive flexibility (coefficient 0.14, 95% CI 0.06 to 0.22, p < 0.001), indicating that patients with the lowest scores had the least improvement in functioning.

Discussion
This study is the first to model unique associations between neurocognitive test scores and longer-term social and occupational functioning in a transdiagnostic clinical cohort of adolescents and young adults accessing youth mental health services. Of note, we observed a novel link between scores on a measure of "cognitive flexibility" (TMT-B) and the rate of improvement in social and occupational functioning over time, which was statistically independent of socio-demographics and level of symptom severity. This approach aligns with the dimensional framework endorsed by the National Institute of Mental Health's Research Domain Criteria initiative 24 , and importantly extends diagnosis-specific links between executive functions and socio-occupational functioning 17,19,21 to a broader transdiagnostic context. Our results expand the evidence base to suggest that cognitive flexibility may represent a meaningful and transdiagnostic target for cognitive remediation protocols in youth mental health settings.
Consistent with previous studies, we observed specific associations between baseline functioning and scores on measures of verbal memory and cognitive flexibility, clinically significant impairments of which were common and experienced by 20.6% and 23.0% of the cohort, respectively. The mechanisms underlying the associations between these domains and functioning in a transdiagnostic context are not well understood but may involve both direct (e.g., difficulty remembering instructions or inflexible decision-making) and indirect effects (e.g., mediated by social cognition or self-efficacy), as observed in schizophrenia 51,52 . Critically, cognitive flexibility also had a robust association with the rate of improvement in functioning longitudinally, such that impaired flexibility was associated with a lower rate of functional recovery over time in contact with clinical services. The independence from level of symptom severity provides clues to an enduring executive impairment linked to functioning. The measure of cognitive flexibility used in this study-the TMT-B-is thought to index higher-order skills such as the ability to flexibly switch between different task demands, in addition to other lower-order abilities such as visual search and processing speed 53 . In general populations, greater cognitive flexibility predicts a range of favorable outcomes across the life course, including better reading ability in children 54 , trait resilience to emotional events in adults 55 , and better health-related quality of life in older adults 56 . Moreover, studies in patients with mental disorders such as bipolar disorder have reported associations between impairments on the TMT-B and poorer functioning cross-sectionally and longitudinally 14,57 .
Neuroimaging studies in healthy adults have revealed a distributed network of frontoparietal regions supporting cognitive flexibility 58 , a number of which are commonly altered in individuals with mental disorders. Thus, the link between cognitive flexibility and functioning observed in general populations may be amplified in individuals who have a mental disorder and neurocognitive impairment, as in our cohort wherein almost onequarter had a clinically significant impairment in cognitive flexibility.
The link between cognitive flexibility and functional improvement may have important treatment implications. Cognitive remediation is increasingly being incorporated into treatment plans for individuals with mental disorders, with evidence that real-world functional gains are greatest when cognitive training is combined with supplemental functional skills training 59 or other vocational interventions. Moreover, some preliminary animal modeling suggests that the adolescent brain may be better able to learn from cognitive training 60 as a function of the unique neurobiology of adolescence (e.g., reward hypersensitivity). Further, accumulating evidence suggests that cognitive remediation in early phases of illness may yield greater than when applied in chronic phases 61,62 . Our results suggest that cognitive flexibility may represent a meaningful and transdiagnostic target for cognitive remediation, which may be enhanced when offered to young people early in the course of illness alongside other interventions targeting social and occupational functioning.

Limitations
Several limitations need mention. First, we relied on a baseline neurocognitive assessment, which is less informative than tracking neurocognitive and functional change dynamically over time. Second, we relied on a single-item index of functioning, potentially missing specific associations with sub-domains of functioning (e.g., relating interpersonally vs. vocational performance). Third, the age range studied spans a dynamic phase of neurocognitive development. Age-related test heterogeneity may therefore have obscured age-specific effects, and our results may not be generalizable to all age groups. However, more than 80% of the sample was aged 15-25, and some research suggests that while cognitive flexibility peaks in early adulthood, it is relatively mature by later childhood 63 . Fourth, as a result of the naturalistic design of this cohort study, sample attrition over time was uncontrolled and may have biased our model estimates. For example, the number of participants retained at the 5year follow-up timepoint was limited, and it is possible that those remaining in care for longer durations have more severe illnesses which require greater attention from clinical services. Unfortunately, we did not collect data regarding specific patterns of treatment usage (e.g., number of sessions with a psychologist). However, the naturalistic design of this study may in fact better reflect the real-world patterns of service usage and functioning. Fifth, studies in schizophrenia consistently report statistical mediation of the path from neurocognition to functional outcome by several factors which were unmeasured here, including social cognition and intrinsic motivation 51 -they are likely relevant to other major mental disorders. Finally, cognitive flexibility and setshifting are related neurocognitive functions, but we did not observe an association between functioning and setshifting in our final model. One possible explanation for this discrepancy is that the test used to measure cognitive flexibility (TMT-B) additionally recruits functions including visual search, processing speed, and working memory and may therefore be inherently more difficult 53 .

Conclusions
In summary, we demonstrate for the first time a robust association between performance on a measure of cognitive flexibility and the rate of functional recovery over time in a transdiagnostic cohort of adolescents and young adults. Our results may have particular relevance for young people accessing broadly-based youth mental health services for whom impairments in cognitive flexibility may represent a treatment target for cognitive remediation in isolation or alongside functional interventions. Future studies should attempt to replicate our observations and determine the efficacy of cognitive remediation or functional interventions in individuals with impaired cognitive flexibility.