Comparative effectiveness research (CER) holds significant promise to improve health care quality. Nevertheless, it faces some significant methodological challenges in fulfilling this promise. Ongoing research addresses some of these challenges. However, the current CER framework does not directly meet the stated goal of CER, i.e., identifying """"""""which interventions are most effective for which patient under specific circumstances."""""""" To do so requires a shift of research paradigm from estimating population average treatment effects to estimating individual treatment effects. We propose to move the field of CER toward individualized CER (iCER) and directly address the need of clinicians and patients at the clinical decision point. The potential use for iCER is most prominent for chronic illnesses, such as major depressive disorder (MDD), that have significant heterogeneous treatment responses. Despite of 50 years of experience with many treatment options for MDD, information on their relative effectiveness for individual patients is still lacking. Guided by the goal of CER, we will conceptualize an iCER statistical modeling framework in mental health research. Our methodology is grounded in research on time-varying covariates and dynamic discrete choice models with longitudinal data;and it is conceptualized using the potential outcomes causal inference framework. Specifically, our models will (1) incorporate multiple treatment options and the patient's treatment preferences, (2) allow individual unobserved heterogeneity, and (3) generate practically useful predictive measures of benefits and harms of treatment alternatives at the individual level. We will develop models for continuous, discrete, and time-to-event outcomes. Empirically, we will use the NIMH funded Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial and the AHRQ supported Medical Expenditure Panel Survey (MEPS) data to evaluate our methods. We will incorporate the typical features of mental health intervention studies and observational studies, such as multiple outcomes, nonrandom dropout, censoring, missing item-responses and self-selection. By anticipating potential complications in model building we will maximize the usefulness of our framework. Ultimately, our research aims at generating useful input information for clinical decisions. We envision using parameters estimated from our models to improve the flexibility and individuality of computerized decision support tools in the future.
The completion of the proposed work will individualize comparative effectiveness research (CER). It will assist clinicians in making decisions based on the best evidence that incorporates the patient's preferences and balances benefits and harms. The new methods will also increase the usefulness of existing evidence from the traditional CER and thus significantly increase the return on investment by NIH.