OBJECTIVES: Polypharmacy is a growing health concern that is prevalent among patients with multiple chronic conditions. However, the term “polypharmacy” is inconsistently defined in the literature. We seek to measure the relationship between prescription medication count and self-reported health, and to determine if an optimal cut-point for count of medications exists to define polypharmacy. METHODS: Retrospective cohort study of adults 18+ using the 2014-2015 Medical Expenditure Panel Survey (n=8,995). The main exposure variable was prescription count defined as the number of unique therapeutic prescription drug classes taken during the baseline year. The outcome was functional (physical and mental) health captured through the Short Form 12 (SF-12). To identify inflection points in the data, we employed locally weighted smoothing (LOESS) in scatterplots. We employed Generalized Additive Models (GAM), allowing modeling non-linearity of the functional health trajectory across prescription count, and controlled for age, race, sex, insurance status, and multi-morbidity. Population-level weights were applied to provide national estimates. RESULTS: 69% of US adults are prescribed at least one drug in 2014, with 55%, 26%, and 13% taking two, five, and eight unique therapeutic drugs, respectively. Prescription medication count was inversely associated with physical health (β =-0.37 , 95% CI:[-0.46, -0.26]), but not mental health (β = -0.06, 95% CI: [-0.16, 0.05]). Loess curves show a mostly linear relationship between health and prescription count, with no clear inflection points. We identify that a cubic GAM model reflects the trajectory (p=0.05), and explains 22.8% of total variance of functional health at follow-up. CONCLUSIONS: Higher number of prescription count is associated with poorer health as expected. However, the data do not show any obvious break point that would justify operationalizing polypharmacy at a specific number of medications (e.g. 5 or more medications). The association varies greatly by underlying personal contextual factors (e.g. age, race-ethnicity).