Reut Apel

Have you ever thought about how life events, such as changes in family status or work, can impact your health and whether your doctor is aware of them and takes them into consideration in your treatment? In the realm of healthcare, patient-reported information often holds crucial insights into individual health experiences and needs. However, a significant challenge arises when such information, particularly non-medical life events, remains confined within the conversational context between the patient and the doctor without being documented in official medical records. This lack of documentation creates a gap in the continuity of care, hindering effective decision-making and personalized interventions.

This lecture will discuss how we addressed this challenge at K Health by leveraging Large Language Models (LLMs) to extract patient life events from unstructured chat data. Our approach involves structuring the output to include both structured data and relevant free text, capturing the nuances of patient-reported information. However, evaluating such models poses unique challenges. One key challenge is the lack of labeled data, which complicates the evaluation process. Additionally, we face the task of structuring the data while preserving the richness of patient-reported information. Furthermore, there is a need to balance between extracting correct life events without overwhelming doctors with irrelevant information, while ensuring that all relevant events are captured to provide a comprehensive view of the patient’s health history.

Overcoming these challenges is crucial in empowering healthcare providers with comprehensive patient insights beyond conventional medical records, thereby enhancing continuity of care and supporting informed decision-making in clinical practice.