Randomized trials of interventions for patients with critical, chronic, and serious illnesses often focus on reducing mortality and improving various nonmortality patient-centered outcomes, such as quality of life, measures of functional status, and the days that individuals are not in a hospital setting. A major challenge when assessing the impact of interventions on nonmortality patient-centered outcomes for patients with critical illnesses is that many of these trial participants die during the follow-up period of a trial. Randomization ensures that the groups of individuals in a trial will be similar at the start of a trial. However, deaths that occur after a trial starts can result in unclear or statistically biased results because the measures of nonmortality patient-centered outcomes cannot be fully assessed among those who die. Nonmortality outcomes that are missing due to death impair the information that can be derived from trials to improve patient-centered outcomes for patients with critical, chronic, and serious illnesses.
Our research team has shown that, in trials that randomize patients, a failure to account for this “missing data due to death” issue will cause biased, misleading, and ambiguous conclusions and trial interpretations. We have developed two broad solutions to deal with this issue: (1) endpoints that treat death as a value of the nonmortality endpoint, such as the worst possible value; and (2) statistical methods that can make informed comparisons using information that has been observed, along with certain assumptions about what happened to those who died.
However, our solutions have not yet been adapted to trials that randomize groups of individuals, or so-called cluster-randomized trials. When individuals are randomized in groups, such as the clinic, nursing home, or hospital where they received care, the assessment of the effect of an intervention is more complex. For example, the analysis must account for unmeasured similarities of patients in each setting, and shared exposures to the same doctors, providers, and other setting-specific factors.
The goal of this proposal is to develop and extend new statistical methods to analyze patient-centered outcome data missing due to death in cluster-randomized trials. In aim 1 of the study, we will develop and test new methods to deal with this type of missing data in cluster-randomized trials. In aim 2 we will compare these new methods with existing methods in thousands of hypothetical trials that we will generate to look similar to real-world cluster-randomized trials that have already been completed. In each simulated trial we will know the actual impact of an intervention and measure how long it takes for the software to complete an analysis with each method, and then how many times each method correctly or incorrectly identifies the true effect. Using this information, we can compare how valid and biased different methods would be in practice. We will also re-analyze 10 real cluster-randomized trials to examine how the choice of statistical method impacts the interpretation of existing trials. In aim 3 we will present all these results to 18 diverse stakeholders and end-users, including trialists, regulators, funders, journal editors, patients, and caregivers to identify desirable qualities (from the end-user perspective) for these methods, and then rank the methods by these qualities. We will then create combined technical and end-user guidance. In aim 4 we will focus on dissemination through tutorial and software development.
Thus, to summarize, we will create new statistical methods, user-friendly statistical software, and methodologic guidance that incorporates stakeholder views of desirable qualities of competing approaches to deal with patient-centered outcome data that are missing due to death in cluster-randomized trials. The methods and guidance we develop will improve the usability, uptake, and ultimate impact of cluster-randomized trials of critically and seriously ill patients with informatively missing patient-centered outcomes. The project deliverables will also enhance research in the medical sciences, social sciences, and other disciplines.