520 Integrating patient voices to enhance patient-reported outcomes: A human-guided machine learning approach to analyzing health narratives

Objectives/Goals: To evaluate the extent to which characterization of unstructured patient-generated health data (UPGD) (e.g., health journals) with human-guided machine learning (HGML) improves model explainability and integrates patient voices among renal cancer patients in a randomized control trial (RCT). Methods/Study Population: Data were collected in an RCT (NCT00505310) assessing the benefits of an emotional expression intervention on renal cancer patients’ quality-of-life outcomes. Over a 10-day period, patients completed four journal-type written responses to structured prompts about their health lifestyle behaviors or deepest thoughts and feelings linked to cancer. I will 1) evaluate the extent to which a standard ML model versus an HGML model aligns with principles of trustworthy health AI when characterizing UPGD, 2) evaluate the extent to which UPGD captures condition-specific data complementary to PROMs, and 3) extract journey maps from narratives about patient experiences navigating cancer care. Results/Anticipated Results: Results will illustrate a patient-centered value research approach that centers patient perspectives, contextualizes heterogeneous factors influencing patient outcomes, and mitigates biases in the health research machine learning pipeline. Discussion/Significance of Impact: This research contributes to the development of evidence-based and empirically validated strategies for health researchers to incorporate explainable HGML tools that facilitate the inclusion of patients’ experiences, perspectives, needs, and priorities throughout the research process.