Effective management of chronic conditions such as type 2 diabetes requires long-term use of pharmacotherapy and frequent reevaluation of treatment decisions over the course of the patients’ illness. Patients and their clinicians not only must choose adequate medications among many options, but they also must decide when to initiate, switch, or intensify treatments and how to titrate medication to balance health benefits and harms. Randomized clinical trials and their standard data analyses, known as intention-to-treat (ITT) analyses, are widely accepted as the gold standard for generating reliable causal evidence to inform these pharmacotherapy decisions, but their advantage hinges on subjects adhering to the randomly assigned therapy. Nonadherence to prescribed medication is widespread, including in randomized trials. As a result, standard ITT analyses can produce misleading information. To control for medication nonadherence in randomized trials, analyses known as per-protocol (PP) analyses based on advanced causal and statistical methods have been developed. The utility of PP analyses extends beyond the artificial setting of randomized trials. When trial evidence is not available, these analyses are used with electronic data collected in real-world clinical settings to emulate conceptual trials for the purpose of informing treatment decisions. Such electronic heath care data (EHD) are used increasingly to generate evidence about the comparative benefits and safety of various pharmacotherapy options, even though they do not capture patients’ actual medication intake but only the date of drug dispensing and the quantities dispensed. To control for medication nonadherence in EHD-based pharmacotherapy studies, researchers have thus resorted to largely arbitrary, opaque, and untestable assumptions to map electronic pharmacy dispensing data in artificial measures of medication adherence. This practice raises concerns over the validity of causal inferences from PP analyses of EHD and can result in important loss of information that limits the evaluation of rare, long-term outcomes even with large amounts of EHD. In addition, current PP analyses were originally developed for studies with well-controlled data collection protocols, such as randomized trials, and are thus not well suited to EHD, for which information is collected at variable and irregularly spaced clinic visits. To address these limitations, we propose to develop generalized PP analyses that are adapted to EHD in order to improve the validity, precision, transparency, and reproducibility of causal evidence generated to inform the management of chronic conditions. Unlike current PP analyses that aim to control for medication nonadherence through the evaluation of presumed medication use, the proposed generalized PP analyses are based on the evaluation of known drug refill behaviors. Consequently, we also propose to use these generalized PP analyses to identify drug refill behaviors that if promoted or avoided can improve the health of patients who live with chronic conditions. To develop these analyses, we will seek the guidance of stakeholder partners who represent patients, clinicians, pharmacists, researchers, and healthcare systems to illustrate and ensure the practical relevance of the proposed analytic approaches with real EHD from two prior studies that were conducted to inform the management of type 2 diabetes.
*All proposed projects, including requested budgets and project periods, are approved subject to a programmatic and budget review by PCORI staff and the negotiation of a formal award contract.