Transdiagnostic dimensions of psychopathology explain individuals’ unique deviations from normative neurodevelopment in brain structure

Abstract

Psychopathology is rooted in neurodevelopment. However, clinical and biological heterogeneity, together with a focus on case-control approaches, have made it difficult to link dimensions of psychopathology to abnormalities of neurodevelopment. Here, using the Philadelphia Neurodevelopmental Cohort, we built normative models of cortical volume and tested whether deviations from these models better predicted psychiatric symptoms compared to raw cortical volume. Specifically, drawing on the p-factor hypothesis, we distilled 117 clinical symptom measures into six orthogonal psychopathology dimensions: overall psychopathology, anxious-misery, externalizing disorders, fear, positive psychosis symptoms, and negative psychosis symptoms. We found that multivariate patterns of deviations yielded improved out-of-sample prediction of psychopathology dimensions compared to multivariate patterns of raw cortical volume. We also found that correlations between overall psychopathology and deviations in ventromedial prefrontal, inferior temporal, and dorsal anterior cingulate cortices were stronger than those observed for specific dimensions of psychopathology (e.g., anxious-misery). Notably, these same regions are consistently implicated in a range of putatively distinct disorders. Finally, we performed conventional case-control comparisons of deviations in a group of individuals with depression and a group with attention-deficit hyperactivity disorder (ADHD). We observed spatially overlapping effects between these groups that diminished when controlling for overall psychopathology. Together, our results suggest that modeling cortical brain features as deviations from normative neurodevelopment improves prediction of psychiatric symptoms in out-of-sample testing, and that p-factor models of psychopathology may assist in separating biomarkers that are disorder-general from those that are disorder-specific.

Capturing developmental dynamics within a transdiagnostic framework: Challenges and promises

“…The approach has been used to map the heterogeneous phenotypes of schizophrenia, bipolar disorder, ADHD, and autism (Wolfers et al, 2018(Wolfers et al, , 2020Zabihi et al, 2019). Applying this approach to transdiagnostic dimensions of psychopathology, recent work has further shown that modelling cortical brain features as deviations from normative neurodevelopment improves prediction of psychiatric symptoms in out-of-sample testing (Parkes, Moore, Calkins, Cook, et al, 2021). Importantly, more generally deviations can also take the form of acceleration or delay of 'typical' development, which can be explained by a variety of factors including socioeconomic status (Tooley et al, 2021).…”

Proportional intracranial volume correction differentially biases behavioral predictions across neuroanatomical features and populations

“…The rapidly growing use of predictive modelling in neuroimaging to map brain-behavior relationships has yielded numerous important advances in recent years. Studies have investigated how preprocessing (Li et al 2019), data transformation (Parkes et al 2021), predictive algorithms (He, Kong, et al 2020), neuroimaging features (Dhamala, Jamison, Jaywant, Dennis, et al 2021;Greene et al 2018), model translation , parcellation choices (Dhamala, Jamison, Jaywant, Dennis, et al 2021) 2021), but also generalizable across datasets (Scheinost et al 2019).…”

“…The rapidly growing use of predictive modelling in neuroimaging to map brain-behavior relationships has yielded numerous important advances in recent years. Studies have investigated how preprocessing (Li et al 2019), data transformation (Parkes et al 2021), predictive algorithms (He, Kong, et al 2020), neuroimaging features (Dhamala, Jamison, Jaywant, Dennis, et al 2021; Greene et al 2018), model translation (He, An, et al 2020), parcellation choices (Dhamala, Jamison, Jaywant, Dennis, et al 2021), sample sizes (Marek et al 2020), and phenotype selection (Chen et al 2020) can influence neuroimaging-based predictions of individual behaviors. Unfortunately, these studies have in large part relied on single datasets of healthy young adults to train and evaluate model performance, even though it is becoming increasingly evident that models must be not only replicable and reliable within a dataset (Tian and Zalesky 2021), but also generalizable across datasets (Scheinost et al 2019).…”

“…All models were evaluated on whether they performed better than chance using null distributions of performance as previously described (Dhamala, Jamison, Jaywant, Dennis, et al 2021; Dhamala, Jamison, Jaywant, and Kuceyeski 2021; Parkes et al 2021). For each predictive model, the behavioral score was randomly permuted 10,000 times.…”

Atypical functional connectivity of the left fusiform gyrus in infants at familial risk for developmental dyslexia

“…Moreover, a distribution of the classification performance was further computed for each seed by subtracting the classification accuracies of the permutation tests from those derived from the true labels within the same iteration. Upon this distribution, a 99% confidence interval (CI) was calculated to see if the lower CI bound was above 0, indicating better-than-chance-level classification performance (see a similar approach in [53]). Finally, for the seed-specific FC patterns with classification results significantly outperforming the permutation tests (i.e., p corrected < 0.5) and demonstrating a positive CI, the effect size was further measured using Cohen’s d , where larger magnitudes indicate stronger classification performance.…”