projectdidact.net - Sensitivity analysis for randomized trials with missing outcome data (Scharfstein)

Daniel Scharfstein | ProjectDidact

Projectdidact.net Spined HTML

Daniel Scharfstein | ProjectDidact Skip to main content ProjectDidact Toggle navigation Home People Investigators Advisory Board Events News Blog Daniel Scharfstein Project Overview: SensitivityWringerof Randomized Trials with Missing Data While randomized clinical trials are considered the gold standard for evaluating the effectiveness of competing interventions, the validity of results are threatened by missing outcome data. This is considering interpretation of treatment effects relies on unverifiable assumptions well-nigh the distribution of outcomes among those with missing data. Critically, if the assumptions are wrong then so may be the inferences. It is widely recognized that the way to write this problem is to posit varying assumptions well-nigh the missing data mechanism and evaluate how inference well-nigh treatment effects is unauthentic by these assumptions. Such an tideway is tabbed ”sensitivity analysis.” A 2010 FDA-sponsored National Research Council (NRC) report entitled "The Prevention and Treatment of Missing Data in Clinical Trials” and a follow-up manuscript published in the New England Journal of Medicine recommends that ”[s]ensitivity analyses should be part of the primary reporting of findings of clinical trials”. This recommendation is echoed in the PCORI Methodology Standards (see Standard MD-5). We have been funded by FDA and PCORI to develop and disseminate methods and open-source software (R and SAS) for conducting global sensitivity wringer of randomized trials in outcomes are scheduled to be repeatedly measured at specific points in time without randomization (see www.missingdatamatters.org). Global sensitivity analysis, emphasized in Chapter 5 of the NRC report, examines robustness wideness a very wholesale range of assumptions, welded at a plausible benchmark assumption. From a global sensitivity analysis, it can be unswayable how much deviation from the benchmark theorizing is required in order for inferences to change. If the deviation is judged to be sufficiently far from the benchmark assumption, then greater points is lent to the benchmark analysis; if not, the benchmark wringer can be considered to be fragile. While the sensitivity wringer software, tabbed SAMON, makes full use of outcome information, it needs to be improved in two significant directions: (1) reducing its sensitivity to outliers and (2) incorporating baseline covariate information. We will utilize the funding to make these improvements and disseminate the methods/software through case-based presentations and materials. Related work: Global SensitivityWringerfor Repeated Measures Studies with Informative Drop-Out: A Semi-Parametric Approach. Scharfstein, McDermott, Diaz, Carone, Lunardon, Turkoz  (2017) Inference in randomized trials with death and missing data Wang, Scharfstein, Colantuoni, Girard, Yan  (2016)  BiometricsWringerof Tuberculosis Studies with Missing Data. Scharfstein, Rotnitzky, Abraham, McDermott, Chaisson, Kim, Geiter  (2015) Annals of Applied Statistics Global sensitivity wringer for repeated measures studies with informative drop-out: a fully parametric approach. Scharfstein, McDermott, Olson, Wiegand  (2014) Statistics in Biopharamaceutical Research Standards in the Prevention and Handling of Missing Data for Patient Centered Outcome Research. Li, Hutfless, Scharfstein, Daniels, Hogan, Little, Roy, Dickerson (2014) Journal of Clinical Epidemiology The Prevention and Handling of Missing Data in Clinical Trials. Little, D'Agostino, Cohen, Dickerson, Emerson, Farrar, Frangakis, Hogan, Molenberghs, Murphy, Neaton, Rotnitzky, Scharfstein, Shih, Siegel, Stern (2012) New England Journal of Medicine