In an extensive 8-year study published in Cell Reports Medicine conducted by Koa Health, researchers closely monitored 59,750 de-identified patient records, with a primary goal of predicting mental health crises occurring within a 28-day window. The results of this study provide critical insights into the power of combining different types of data for predictive analytics in mental healthcare. Findings reveal that utilizing both structured electronic health records (EHR) and narrative clinical notes significantly improves the accuracy of mental health crisis relapse prediction. The main author of the article disseminated by Cell Reports Medicine is Roger Garriga, researcher of the Department of Information and Communication Technologies (DTIC) at Pompeu Fabra University and Lead Data Scientist at Koa Health.

A mental health crisis is defined as situations in which patients are unable to function effectively in the community or when there is a risk of hurting themselves or others. Mental health crises continue to present a significant challenge to healthcare systems worldwide. With a consistent rise in demand for mental health services and increasingly limited resources, it's essential to find and leverage effective predictive methods. This research, conducted over a substantial timeframe, highlights the importance of incorporating diverse types of data to enhance predictions. The combination of structured and quantitative data, such as demographics and diagnosis, with qualitative data from clinical notes that provide the subjective context of a patient's mental health status, significantly strengthens the accuracy of predicting mental health crisis relapses.

More accurate crisis prediction enables providers and healthcare systems to drive positive outcomes, optimize resource allocation, and facilitate timely interventions, saving lives by helping ensure people struggling don’t slip through the cracks.

Roger Garriga (DTIC UPF and Koa Health): "Predictive models offer the opportunity to enhance clinical decision-making and enable timely interventions geared toward prevention, thereby transitioning from the current reactive care paradigm to a proactive approach"

Roger Garriga, researcher of the Department of Information and Communication Technologies (DTIC) at UPF and Lead Data Scientist at Koa Health, said: “Mental health crises can profoundly impact the mental and social wellbeing of individuals who experience them. Predictive models offer the opportunity to enhance clinical decision-making and enable timely interventions geared toward prevention, thereby transitioning from the current reactive care paradigm to a proactive approach. Our study demonstrates the effectiveness of clinical notes in predicting mental health crises and underscores the benefits of effectively combining these notes with structured electronic health records, leading to improved outcomes.” Garriga is member of the Artificial Intelligence and Machine Learning group at DTIC-UPF.

Dr. Oliver Harrison, CEO of Koa Health, said, “Access to mental health care remains a huge challenge worldwide. Too often, mental health services are reactive—managing crisis events in the Emergency Room, rather than proactive—detecting problems early and avoiding crises. Building on our paper in Nature Medicine last year, this research shows that analyzing a combination of clinical notes and structured health records can create robust predictions, offering the hope of preventing these crises and avoiding human suffering and very high cost.”

Dr. João Guerreiro, Senior Applied Scientist at Koa Health, said, “There is a clear lack of standardization in clinical note-taking. In order to address it, our study introduces methods that tackle the challenges presented by the diversity of doctors’ personal note-taking styles and the varying amounts of clinical notes per patient. Our research highlights the importance of choosing an appropriate machine learning method to account for the biases that arise due to patients with more severe crises having a greater volume of notes, as well as the challenge of combining structured and unstructured data sources.”

Dr. Aleksandar Matic, Research and Development Director at Koa Health, said, “Although structured EHRs (electronic health records) are becoming a standard, clinical notes remain a dominant data source in most healthcare systems. Our study marks a crucial milestone, urging the integration of both data types for preventative interventions. By blending these models, we unlock the potential to revolutionize clinical decision-making and improve health outcomes in mental health crises.”

Reference article:

Roger Garriga, Teodora Sandra Buda, João Guerreiro, Jesús Omaña Iglesias, Iñaki Estella Aguerri, Aleksandar Matić. Combining clinical notes with structured electronic health records enhances the prediction of mental health crises. Cell Reports Medicine, 2023, 101260, ISSN 2666-3791, https://doi.org/10.1016/j.xcrm.2023.101260.