We propose to quantify biases associated with categorizing psychopharmacological adherence data, develop methods for using continuous adherence data, and test those methods using multiple modern and innovative longitudinal models to understand how changes in adherence affect changes in clinical outcomes among adults with schizophrenia and bipolar disorders. While adherence to psychopharmacological medications is thought to improve multiple outcomes, rates of adherence are not only low, but also fluctuate substantially over time. Given the importance of adherence in effecting positive clinical outcomes and reducing system costs, adherence data must be analyzed correctly. However, current research is negatively affected by two issues: unjustified categorization of continuous measures of adherence (a problem recently noted by ISPOR) and failing to utilize innovative statistical methods to model longitudinal and intensive longitudinal adherence data. Thus, a systematic study is needed to examine the most effective ways to analyze adherence data across a range of outcomes and conditions. We propose to use two secondary data sources-the NIMH-funded Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) data and Florida Medicaid data-to accomplish three sequential Aims:
Aim 1. Quantify parameter estimate biases resulting from categorizing adherence data via a simulation study and develop and test parametric and non- parametric methods for modeling continuous adherence data across multiple theoretical and empirical distributions;
Aim 2. Apply the methods developed and tested in Aim 1 by fitting latent growth models, growth mixture models, and dynamical systems analysis models to continuous adherence data from both the CATIE study and Florida claims data to examine patterns and inter-individual and intra-individual variability in adherence trajectories;
and Aim 3. Extend Aim 2 models to explore the predictive effects of changes in adherence on changes in longitudinal outcomes, such as psychotic symptoms, arrest and hospitalization, and substance use. Our proposal builds in part on our study team's NIH-funded grants and successful collaborations in applying modern and innovative statistical models to important mental health services- related issues. Our proposed R03 will strongly complement recent efforts undertaken by the NIH Adherence Network and NIMH's efforts to improve data-based decision models through improved systems science. These efforts will not only improve the field's understanding of how adherence relates to multiple outcomes for adults with schizophrenia and bipolar disorders, but also has the potential to affect adherence research across all NIH institutes.
This research is one important step toward better understanding the role of psychopharmacological adherence among adults with serious mental illness (SMI) and improving illness trajectories for this vulnerable population. Specifically, this grant will quantify biases resulting from unjustified categorization of continuous adherence data and will develop and test methods for using continuous adherence data in research settings. In addition, we will examine the utility of numerous modern and innovative approaches to evaluating the longitudinal relationship between psychopharmacological adherence (including patterns of intra-individual variations in adherence) and relevant outcomes among adults with SMI.
| Tueller, Stephen J; Deboeck, Pascal R; Van Dorn, Richard A (2016) Getting less of what you want: reductions in statistical power and increased bias when categorizing medication adherence data. J Behav Med 39:969-980 |
