Differences in Long-Term Patient Outcomes with Disease-Modifying Versus Symptomatic Treatments for Alzheimer’s Disease (Ad)

Abstract

treatment, prior to launch using clinical trial (CT) data and pre-launch observational data. Methods: This study utilized data from CTs for fingolimod, and administrative claims data from IMS PharMetrics Plus, 2007-2013 constructed between fingolimod (pre-launch) and other disease-modifying therapies (DMTs) given CT data and pre-launch observational data. We assessed the projected relapse probability (RP) of fingolimod, as if it were available to treat the entire pre-2010 RW MS population. This was achieved in a two-step modeling process: (1) we selected a population of MS patients as a reference population and characterized this population in the pre-2010 claims data by estimating the joint distribution of the covariates, and (2) we then modeled projections for the RP using CT data standardized by RW population characteristics. Using a machine learning platform, Reverse Engineering and Forward Simulation (REFSTM), we built an ensemble of predictive models to create weights that were used to standardize the CT population to the pre-2010 RW MS population. The method was further validated using post-2010 fingolimod claims data. Results: The RW MS population was older (7,471 patients; mean age= 42 years, SD= 8.6) than the CT population (243 fingolimod patients; mean age= 37 years, SD= 7.9). The projected RP in the RW setting was 0.12-0.24 and was similar to that in the CT setting (0.11-0.22). ConClusions: We developed a causal methodology that estimated the RP among MS patients treated with fingolimod from CT data by standardizing to a reference RW pre-launch population. Our E2E methodology is generalizable and thus proposes a framework for translation of intervention efficacy data into estimates of intervention effectiveness.