Analysis of diagnosis instability in electronic health records reveals diverse disease trajectories of severe mental illness

BackgroundElectronic health record (EHR) databases, increasingly available in low- and middle-income countries (LMIC), provide an opportunity to study transdiagnostic features of serious mental illness (SMI) and delineate illness trajectories using clinical data. AimsCharacterize transdiagnostic features and diagnostic trajectories of SMI using structured and unstructured data from an EHR database in an LMIC institution. MethodsWe conducted a retrospective cohort study using EHR data from 2005-2022 at Clinica San Juan de Dios Manizales, a specialized mental health facility in Caldas, Colombia. We included 22,447 patients treated for schizophrenia (SCZ), bipolar disorder (BD), severe or recurrent major depressive disorder (MDD). We extracted diagnostic codes, clinical notes, and healthcare use data from the EHR database. Using natural language processing, we analyzed the frequency of suicidality and psychosis across SMI diagnoses. Using the diagnostic trajectories, we studied patterns of diagnostic switching and accumulation of comorbidities. Mixed-effect logistic regression was used to assess factors influencing diagnostic stability. ResultsHigh frequencies of suicidality and psychosis were observed across diagnoses of SCZ, BD, and MDD. Most SMI patients (64%) received multiple diagnoses over time, including switches between primary SMI diagnoses (19%), diagnostic comorbidities (30%), or both (15%). Predictors of diagnostic switching included mentions of delusions in clinical notes (OR=1.50, p=2e-18), prior diagnostic switching (OR=4.02, p=3e-250), and time in treatment, independent of age (log of visit number; OR=0.56, p=5e-66). Over 80% of patients reached diagnostic stability within six years of their first record. ConclusionsThis study demonstrates that integrating structured and unstructured EHR data can reveal clinically relevant, transdiagnostic patterns in SMI, including early predictors of disease trajectories. Our findings underscore the potential of EHR-based tools to aid etiological research and the development of personalized treatment strategies, particularly in LMIC.