NLM logo

National Information Center on Health Services Research and Health Care Technology (NICHSR)

HSRProj (Health Services Research Projects in Progress)

Information about ongoing health services research and public health projects

Sensitivity analysis tools for clinical trials with missing data
Investigator (PI): Scharfstein, Daniel
Performing Organization (PO): (Current): Johns Hopkins University, Johns Hopkins Bloomberg School of Public Health, Department of Biostatistics / (410) 955-3067
Supporting Agency (SA): Patient-Centered Outcomes Research Institute (PCORI)
Initial Year: 2014
Final Year: 2019
Record Source/Award ID: PCORI/ME-1303-6016
Funding: Total Award Amount: $553,858
Award Type: Contract
Award Information: PCORI: More information and project results (when completed)
Abstract: Missing outcome data are a widespread problem in clinical trials, including trials with patient-centered outcomes. While unbiased estimates can be obtained from trials with no missing data, this is no longer true when data are missing on some patients. The essential problem is that inference about treatment effects relies on unverifiable assumptions about the nature of the mechanism that generates the missing data, leading to concerns about the validity and robustness of trial results. A recent National Research Council (NRC) report recommends that examining sensitivity to the assumptions about the missing data mechanism should be a mandatory component of reporting. PCORI Methodology Standard MD-5 echoes this recommendation. While chapter 5 of the NRC report outlines a general framework for conducting global sensitivity analysis (like stress testing), there are two major problems with existing methods that have limited their usefulness: (1) they have not been implemented in software packages; and (2) they do not adequately address non-monotone missing data patterns (i.e., patients provide data irregularly). The NRC report recognizes the development of software that supports coherent missing data analyses as a high priority and highlights non-monotone missing data as one of the important areas in which progress is particularly needed. This project aims to address these problems by (a) creating unified and coherent methods for global sensitivity analysis of clinical trials with monotone and non-monotone missing data; (b) developing free, open source, and reproducible software in SAS and R to implement the methods; (c) demonstrating the methods and software using clinical trial data with patient-centered outcomes; and (d) disseminating the methods and software. A website will facilitate distribution of software, documentation, datasets, case studies, videos, etc. and allow communication with our user-base. A monograph will be written and short-courses presented at major stakeholder conferences. A highly diverse and talented advisory board will help ensure that our methods and software meet the needs of stakeholders and are broadly disseminated. With software, PCOR researchers will be able to conduct and report the results of sensitivity analyses of clinical trials with missing data. In this way, patients, their caregivers, regulators, and policy makers can more adequately judge the robustness of the inferences from these trials to assumptions about the missing data mechanism. This will lead to greater or less confidence in the study results, which may impact health care decisions. We expect that PCOR researchers, knowing that the results of their studies will be subjected to a "stress test," will place greater emphasis on minimizing missing data through better trial design and conduct. More on this study: Scharfstein D, Rotnitzky A, Abraham M, McDermott A, Chaisson R, Geiter L. On the analysis of tuberculosis studies with intermittent missing sputum data. Ann Appl Stat. 2015;9(4):2215-2236.
MeSH Terms:
  • Algorithms
  • * Clinical Trials as Topic
  • Data Collection
  • Health Policy
  • Humans
  • Models, Statistical
  • Outcome Assessment, Health Care
  • Patient-Centered Care
  • Program Development
  • Programming Languages
  • Research Design
  • Software
  • Statistics as Topic
  • United States
Country: United States
State: Maryland
Zip Code: 21205
UI: 20143591
Project Status: Completed
Record History: ('2019: Project extended to 2019 ',) ('2018: Project extended to 2018',)