Development and validation of electronic health record-based, machine learning algorithms to predict quality of life among family practice patients.

Health-related quality of life (HRQol) is a crucial dimension of care outcomes. Many HRQoL measures exist, but methodological and implementation challenges impede primary care (PC) use. We aim to develop and evaluate a novel machine learning (ML) algorithm that predicts binary risk levels among PC patients by combining validated elements from existing measures with demographic data from patient electronic health records (eHR) to increase predictive accuracy while reducing prospectively-collected data required to generate valid risk estimates. Self-report questions from previously validated QoL surveys were collected from PC patients and combined with their demographic and social determinant (SD) data to form a 53-question item bank from which ML chose the most predictive elements. For algorithm development, 375 observations were allocated to training (n = 301, 80%) or test partitions (n = 74, 20%). Questions that asked participants to rate how happy or satisfied they have been with their lives and how easy or hard their emotional health makes work/school showed a good ability to classify participants' mental QoL (98% max balanced accuracy). Questions that asked participants to rate how easy or hard it is to do activities such as walking or climbing stairs and how much pain limits their everyday activities showed ability to classify physical QoL (94% max balanced accuracy). No demographic or SD factors were significantly predictive. Supervised machine learning can inform QoL measurements to reduce data collection, simplify scoring, and allow for meaningful use by clinicians. Results from the current study show that a reduced 4-question model may predict QoL almost as well as a full-length 40-question measure.