Predicting Conversion From Unipolar Depression to Bipolar Disorder and Schizophrenia: A 10-Year Retrospective Cohort...

Background: The initial stages of bipolar disorder (BD) and schizophrenia (SCZ) often exhibit depressive symptoms and syndromes, leading to potential misdiagnosis and treatment for unipolar depression (UD). However, no consensus exists on individualized and time-varying intervenable conversion predictors for both BD and SCZ. Methods: This study examined the rate of true conversion from UD to BD and SCZ, considering factors such as sex, family history of mental illness, psychotic features, recurrent depression, and treatment patterns. The objective was to develop predictive models for short-, medium-, and long-term risk stratification for BD/SCZ conversion. Data were extracted from electronic medical records (EMRs) between January 2009 and December 2020 in a large academic medical center-based health system in China. Participants included 12,182 depressive inpatients without previous or comorbid diagnoses of BD and SCZ. The outcome measure was a subsequent admission record with a diagnostic code reflective of BD or SCZ. Four machine-learning algorithms using sociodemographic, clinical, laboratory, vital signs, symptoms, and treatment features were applied to predict this outcome. Explainable methodologies, specifically SHapley Additive exPlanations (SHAP) and Break Down, were employed to analyze the contribution of each individual feature. Results: Among 12,182 individuals, 344 (2.82%) and 64 (0.53%) received a subsequent diagnosis of BD and SCZ, respectively. Higher risk factors for BD progression included being female, having severe depression, being prescribed mood stabilizers, β receptor blockers (e.g., metoprolol tartrate and propranolol hydrochloride), and antipsychotics (e.g., sulpiride and quetiapine). Higher risk factors for SCZ progression included being male, exhibiting psychotic symptoms, being prescribed antipsychotics (e.g., risperidone and sulpiride), antiside effects drugs (e.g., trihexyphenidyl and hemp seed pill), and undergoing psychotherapy. Individuals with a family history of mental illness were particularly susceptible to conversion to BD and SCZ. Conclusions: The model's performance on the test dataset declined over time, with area under the curve (AUC) values for predicting BD conversion decreasing from 0.771 in 1 year to 0.749 in 2 years and 0.733 in 7 years, and for SCZ conversion, from 0.866 in 1 year to 0.829 in 3 years and 0.752 in 7 years. A key finding is that individuals with refractory (particularly psychotic) UD had an elevated risk of transitioning to BD and SCZ, with social-demographic factors, lifestyle behaviors, vital signs, and blood markers becoming significant risk factors over follow-up. Upon further validation, these models could provide clinicians with dynamic information regarding a patient's risk of disease conversion.