Statistical Modeling of Longitudinal Data with Multiple Structural Change Points

Key health policy reforms introduce structural “change points” in longitudinal data, the effects of which may vary across subgroups. The specification and presentation of statistical models that test for changes in slope occurring at or after such “change points” can be extremely important in understanding the differential impact of policy changes across subpopulations. A difficulty arises when additional change points are theorized to have a large effect on overall trends but little differential effect on subgroups. As an illustration, the establishment of fixed Medicare payment rates for outpatient dialysis in January 2011 and the erythropoiesis-stimulating agent (ESA) safety-related label change in July 2011 may have affected anemia management among dialysis patients. Using 2009-13 Medicare claims, we fitted mixed model regression with piecewise linear splines to investigate how trends in hemoglobin levels and ESA doses varied based on facility payer mix, which was hypothesized to have greater financial implications for independent facilities vs. facilities affiliated with large dialysis organizations (LDOs). The choice of splines was both informed by key time points in policy implementation and data driven. We found evidence of declining average hemoglobin for facilities with lower levels of employer insurance (EI) vs. facilities with higher levels of EI following payment reform among independent facilities, but not among LDO facilities (p=0.005 for independent, p=0.292 for LDO). This study applies structural change points to study the effects of Medicare dialysis payment reform and illustrates a novel way of graphically presenting results that is readily accessible to a broad audience.