An online tool to assess the potential impact of missing outcome data on the estimates of treatment effect of trials

Article type
Authors
Akl E1, Agarwal N2, Guyatt G3
1Department of Medicine, State University of New York at Buffalo, NY, USA
2State University of New York at Buffalo, NY, USA
3Department of Clinical Epidemiology and Biostatistics, Hamilton, ON, Canada
Abstract
Background: Systematic reviewers commonly deal with RCTs for which relevant outcome data are missing. The Cochrane Handbook recommends reviewers to assess 'how sensitive results are to reasonable changes in the assumptions that are made’.

Objective: To develop an online tool to assess the potential impact of reasonable assumptions about missing data for binary outcomes on the estimates of treatment effect of RCTs.

Methods: We developed the tool as an online interactive Java application. It conducts the sensitivity analysis at the RCT level. For each trial arm (i.e., the intervention and control), the user enters the number of participants randomized, the number of observed events, and the number of participants for whom outcome data is missing. For the RCT under consideration, the tool provides the effect estimate as relative risk with its 95% confidence interval based on each set of assumptions being tested.

Results: The tool is available freely online at: http://apps.medinnovations.us/. It allows the testing of the following assumptions about the outcomes of participants for whom data is missing: (1)none had an event; (2)all had an event; (3)all in the treatment group had an event while none in the control group had one (worst case scenario); (4)none in the treatment group had an event while all in the control group had one (best case scenario); (5)the event incidence among participants with missing data (relative to observed participants) varies by ratios specified by the user for the intervention and control groups separately. It also provides the results for a complete case analysis (e.g., excluding participants for whom data I missing from the analysis).

Conclusions: The tool allows systematic reviewers to conduct sensitivity analyses using the assumptions that are most plausible for the specific question of their review.