Alessandra Camassa

Background: Mental health systems face escalating demand that exceeds clinician capacity, making accurate severity-based triage a critical bottleneck. Severity assessment guides treatment intensity, resource allocation, and risk management, yet most clinically relevant information remains embedded in unstructured electronic health record (EHR) narratives, limiting its utility for scalable decision support. Objectives: This study evaluates whether a single large language model (LLM) can autonomously extract clinical factors from psychiatric EHR narratives, derive predictive weights from those factors, and use the resulting structured representation to predict clinician-implied severity at scale. Methods: From a Mayo Clinic repository of more than 2.7 million encounters, 15,000 de-identified psychiatric notes were sampled into a 5,000-patient discovery cohort and a 10,000-patient replication cohort. The same LLM (Llama 3 8B Instruct) extracted 17 background clinical factors and 3 treatment-action factors from each note. Severity reference labels were derived from the treatment-action factors using pre-specified clinical criteria. The LLM independently derived two factor-weight dictionaries from the discovery cohort: one capturing risk-oriented predictors of severe presentations and one capturing protective predictors. Five weighting conditions were then evaluated against the severity labels: the two LLM-derived dictionaries, two controls (LLM-derived variables with randomized weights; clinically irrelevant variables with arbitrary weights), and an unweighted zero-shot baseline. Performance was assessed across 928 valid iterations in the replication cohort. Results: LLM-derived structured conditions significantly outperformed all controls and the baseline, with statistically equivalent performance between the two structured conditions. Improvements in precision and recall were balanced, indicating gains in discriminative capacity rather than threshold shifts. The variables and weights the LLM derived as predictors of severe presentations aligned closely with established clinical determinants of psychiatric severity. Conclusion: A single LLM can derive clinically meaningful factor weights from unstructured EHR narratives and use them to predict psychiatric severity at scale, supporting a viable path toward interpretable, scalable triage in resource-constrained mental health systems.

Graphene-based transistors (gSGFETs) enabled stable full-band brain recordings for 5+ months, allowing precise brain state identification and prediction, which is critical both in brain science and neurology.

Slow oscillations are an emergent activity of the cerebral cortex network consisting of alternating periods of activity (Up states) and silence (Down states). Up states are periods of persistent cortical activity that share properties with that of underlying wakefulness. However, the occurrence of Down states is almost invariably associated with unconsciousness, both in animal models and clinical studies. Down states have been attributed relevant functions, such as being a resetting mechanism or breaking causal interactions between cortical areas. But what do Down states consist of? Here, we explored in detail the network dynamics (e.g., synchronization and phase) during these silent periodsin vivo(male mice),in vitro(ferrets, either sex), andin silico, investigating various experimental conditions that modulate them: anesthesia levels, excitability (electric fields), and excitation/inhibition balance. We identified metastability as two complementary phases composing such quiescence states: a highly synchronized “deterministic” period followed by a low-synchronization “stochastic” period. The balance between these two phases determines the dynamical properties of the resulting rhythm, as well as the responsiveness to incoming inputs or refractoriness. We propose detailed Up and Down state cycle dynamics that bridge cortical properties emerging at the mesoscale with their underlying mechanisms at the microscale, providing a key to understanding unconscious states.SIGNIFICANCE STATEMENTThe cerebral cortex expresses slow oscillations consisting of Up (active) and Down (silent) states. Such activity emerges not only in slow wave sleep, but also under anesthesia and in brain lesions. Down states functionally disconnect the network, and are associated with unconsciousness. Based on a large collection of data, novel data analysis approaches and computational modeling, we thoroughly investigate the nature of Down states. We identify two phases: a highly synchronized “deterministic” period, followed by a low-synchronization “stochastic” period. The balance between these two phases determines the dynamic properties of the resulting rhythm and responsiveness to incoming inputs. This finding reconciles different theories of slow rhythm generation and provides clues about how the brain switches from conscious to unconscious brain states.