Clinical Promise of Brain-Phenotype Modeling

Greene, Abigail S.; Constable, R. Todd

doi:10.1001/jamapsychiatry.2023.1419

Cited by 7 publications

(3 citation statements)

References 112 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These broad definitions may also explain the models’ lack of localization. More specific phenotypes will likely improve a model’s interpretability (Enkavi et al, 2019; Greene and Constable, 2023). We also see strong cross-dataset predictions for individual measures, so testing this hypothesis is plausible for future work.…”

Section: Discussionmentioning

confidence: 99%

Brain-phenotype predictions can survive across diverse real-world data

Adkinson,

Rosenblatt,

Dadashkarimi

et al. 2024

Preprint

View full text Add to dashboard Cite

Recent work suggests that machine learning models predicting psychiatric treatment outcomes based on clinical data may fail when applied to unharmonized samples. Neuroimaging predictive models offer the opportunity to incorporate neurobiological information, which may be more robust to dataset shifts. Yet, among the minority of neuroimaging studies that undertake any form of external validation, there is a notable lack of attention to generalization across dataset-specific idiosyncrasies. Research settings, by design, remove the between-site variations that real-world and, eventually, clinical applications demand. Here, we rigorously test the ability of a range of predictive models to generalize across three diverse, unharmonized samples: the Philadelphia Neurodevelopmental Cohort (n=1291), the Healthy Brain Network (n=1110), and the Human Connectome Project in Development (n=428). These datasets have high inter-dataset heterogeneity, encompassing substantial variations in age distribution, sex, racial and ethnic minority representation, recruitment geography, clinical symptom burdens, fMRI tasks, sequences, and behavioral measures. We demonstrate that reproducible and generalizable brain-behavior associations can be realized across diverse dataset features with sample sizes in the hundreds. Results indicate the potential of functional connectivity-based predictive models to be robust despite substantial inter-dataset variability. Notably, for the HCPD and HBN datasets, the best predictions were not from training and testing in the same dataset (i.e., cross-validation) but across datasets. This result suggests that training on diverse data may improve prediction in specific cases. Overall, this work provides a critical foundation for future work evaluating the generalizability of neuroimaging predictive models in real-world scenarios and clinical settings.

show abstract

Section: Discussionmentioning

confidence: 99%

Brain-phenotype predictions can survive across diverse real-world data

Adkinson,

Rosenblatt,

Dadashkarimi

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…There is increasing recognition in psychiatry that precision phenotyping is necessary both for advancing diagnosis/treatment, as well as for progress in psychiatric research (Redish et al, 2016;Friston et al, 2017;Anticevic and Murray, 2017;Tiego et al, 2023;Karvelis et al, 2023;Greene and Constable, 2023). This is due to the great heterogeneity observed in psychiatric conditions, and the reduced statistical power caused by lumping groups of subjects by diagnostic category.…”

Section: Applications To Precision Psychiatrymentioning

confidence: 99%

“…In addition, identifying qualitatively distinct groups will aid in neuroimaging, psychopharmacology and genetics research Greene and Constable (2023). Identifying sub-groups allows one to maximize the statistical power and identify effects using these modalities (Tiego et al, 2023).…”

Section: Applications To Precision Psychiatrymentioning

confidence: 99%

Attention and learning strategies reveal distinct profiles of psychiatric traits

Pettine,

Tseng,

Yang

et al. 2024

Preprint

View full text Add to dashboard Cite

Humans presented with the same problem in the same environment commonly adopt wildly different strategies for attention and learning. Indeed, psychiatric conditions are defined by qualitative differences in behavior. However, most tasks measure an individual's deviation from a single expected strategy rather than the utilization of distinct strategies. Measuring diverse strategies is especially important for psychiatry, were conditions are defined by qualitatively distinct patterns of behavior. We paired psychiatric trait questionnaires with a context generalization task whose metrics for goal-directed attention and short-term memory identify qualitatively distinct strategies. Questionnaires assessed for traits associated with ASD, attention-deficit/hyperactivity disorder (ADHD), obsessive compulsive disorder (OCD), depression, schizotypy and psychosis. The subject population recruited online was matched for self-reported sex, and the sample was enriched with those reporting a formal diagnosis of ASD. 744 subjects completed the first session of the task, and 584 returned after four to six weeks to complete the second session. We found that a strategy dominated by goal-directed attention was associated with a profile of reduced trait scores relative to other subjects across all measures. During the second session, this strategy was particularly pronounced for those with reduced ADHD traits. In contrast, a strategy of attending to features based on frequency was associated with a profile of increased trait scores relative to other subjects, particularly traits for ASD and OCD. During the second session, this strategy was again associated with elevated traits, particularly those for ASD and ADHD traits. These results provide insight into the relationship between psychiatric traits and qualitatively distinct attention and learning strategies.

show abstract