Program Director/Principal Investigator (Last, First, Middle): Scharfstein, Daniel, Oscar Project Summary/Abstract Missing outcome data threaten the validity of randomized clinical trials because inference about treatment effects then necessarily relies on untestable assumptions, which wrongly stated can lead to incorrect conclusions. While it is widely recognized that evaluating the sensitivity of trial results to assumptions about the missing data mech- anism should be a mandatory component of reporting, rigorous sensitivity analyses are not routinely reported. Likely explanations include inadequate knowledge translation by statistical methodologists to both principal in- vestigators and their statistical collaborators as well as lack of software. Substance use disorder clinical trials are known to suffer from high rates of missing data. Unlike regulatory trials where missing data are primarily the result of premature study withdrawal, individuals in substance use disorder trials tend to intermittently skip their scheduled outcome assessments. This produces an explosion of ?non-monotone? missing data patterns that makes sensitivity analysis methodologically and computationally chal- lenging. There has been relatively little research on sensitivity analysis procedures for analyzing such data and the procedures that have been developed are anchored to assumptions that are problematic. Thus, investigators are faced with challenging analytic barriers and the conclusions they draw from their trials may be ?awed. In this three-year proposal, we will reanalyze 29 clinical trials conducted by NIDA's Clinical Trials Network (CTN), and made publicly available on NIDA's DataShare website, to evaluate their robustness to missing data assump- tions through rigorous sensitivity analysis. Since adequate tools for conducting sensitivity analysis of studies with highly non-monotone missing data patterns do not yet exist, we plan to develop, implement and dissemi- nate (through journal articles, short courses and webinars) an innovative sensitivity analysis methodology and open-source, user-friendly software to evaluate the robustness, to missing data assumptions, of trials in which binary outcomes (e.g., substance use) are scheduled to be repeatedly collected at ?xed points in time after ran- domization and participants intermittently skip their scheduled assessments. Our tool will be developed by an interdisciplinary team of biostatisticians and substance use disorder treatment experts, with input from an advi- sory board comprised of highly regarded statistical experts and leading scientists in the substance use disorder community. Through reanalysis of the NIDA's CTN trials using our tool, we will be better able to understand the impact of missing data assumptions on the evaluation of the studied interventions. Additionally, demonstrat- ing the importance and utility of our tool to our advisory board and to the substance use disorder community more broadly stands to increase the likelihood of adoption. Finally, the development, testing, and dissemination of this innovative statistical tool can serve as a template for other scienti?c domains, making ?stress-testing? to untestable missing data assumptions a more routine component of scienti?c reporting.
Scharfstein, Daniel, Oscar Project Narrative We will develop, implement and disseminate an innovative sensitivity analysis methodology and open-source, user-friendly software tool to evaluate the robustness to missing data assumptions of clinical trials in which binary outcomes are collected at ?xed points over time but where participants intermittently miss their scheduled assess- ments, as frequently occurs in addiction research. We will utilize this tool to reanalyze data from 29 substance abuse clinical trials conducted by NIDA's Clinical Trials Network and available publicly on NIDA's DataShare web- site. As a result, we will be better able to understand the impact of missing data assumptions on the evaluation of the studied interventions and, by demonstrating the importance and utility of our tool to an advisory board comprised of study investigators and to the substance use disorder community more broadly, we aim to make ?stress-testing? to untestable missing data assumptions a more routine part of scienti?c reporting.
