Article

Reweighted Mahalanobis distance matching for cluster-randomized trials with missing data.

VA Tennessee Valley Geriatric Research Education Clinical Center (GRECC), Nashville, TN, USA.
Pharmacoepidemiology and Drug Safety (Impact Factor: 3.17). 05/2012; 21 Suppl 2:148-54. DOI: 10.1002/pds.3260
Source: PubMed

ABSTRACT This paper introduces an improved tool for designing matched-pairs randomized trials. The tool allows the incorporation of clinical and other knowledge regarding the relative importance of variables used in matching and allows for multiple types of missing data. The method is illustrated in the context of a cluster-randomized trial. A Web application and an R package are introduced to implement the method and incorporate recent advances in the area.
Reweighted Mahalanobis distance (RMD) matching incorporates user-specified weights and imputed values for missing data. Weight may be assigned to missingness indicators to match on missingness patterns. Three examples are presented, using real data from a cohort of 90 Veterans Health Administration sites that had at least 100 incident metformin users in 2007. Matching is utilized to balance seven factors aggregated at the site level. Covariate balance is assessed for 10,000 randomizations under each strategy: simple randomization, matched randomization using the Mahalanobis distance, and matched randomization using the RMD.
The RMD matching achieved better balance than simple randomization or MD randomization. In the first example, simple and MD randomization resulted in a 10% chance of seeing an absolute mean difference of greater than 26% in the percent of nonwhite patients per site; the RMD dramatically reduced that to 6%. The RMD achieved significant improvement over simple randomization even with as much as 20% of the data missing.
Reweighted Mahalanobis distance matching provides an easy-to-use tool that incorporates user knowledge and missing data.

Full-text

Available from: Christianne L Roumie, Nov 05, 2014
0 Followers
 · 
121 Views
  • [Show abstract] [Hide abstract]
    ABSTRACT: In many experimental situations, researchers have information on a number of covariates prior to randomization. This information can be used to balance treatment assignment with respect to these covariates as well as in the analysis of the outcome data. In this paper, we investigate the use of propensity scores in both of these roles. We also introduce a randomization procedure in which the balance of all measured covariates is approximately indexed by the variance of the empirical propensity scores and randomization is restricted to those permutations with the least variable propensity scores. This procedure is compared with recently proposed methods in terms of resulting covariate balance and estimation efficiency. Properties of the estimators resulting from each procedure are compared with estimates which incorporate the propensity score in the analysis stage. Simulation results show that analytical adjustment for the propensity score yields results on par with those obtained through restricted randomization procedures and can be used in conjunction with such procedures to further improve inferential efficiency. Copyright © 2014 John Wiley & Sons, Ltd.
    Statistics in Medicine 11/2014; 34(4). DOI:10.1002/sim.6361 · 2.04 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Background No randomized trials have compared survival outcomes for men with localized prostate cancer (PC) being treated with radical prostatectomy (RP) or external beam radiotherapy (EBRT). The goal of the study, therefore, was to estimate the association of RP (compared with EBRT) with overall and PC mortality.Methods We analyzed an observational cohort from the population-based Prostate Cancer Outcomes Study, which included men aged 55 to 74 years diagnosed with localized PC between October 1994 and October 1995 who underwent either RP (n = 1164) or EBRT (n = 491) within 1 year of diagnosis. Patients were followed until death or study end (December 31, 2010). Overall and disease-specific mortality were assessed with multivariable survival analysis, with propensity scores to adjust for potential treatment selection confounders (demographics, comorbidities, and tumor characteristics). All statistical tests were two-sided.ResultsAfter 15 years of follow-up, there were 568 deaths, including 104 from PC. RP was associated with statistically significant advantages for overall (hazard ratio [HR] = 0.60, 95% confidence interval [CI] = 0.53 to 0.70, P <.0001.) and disease-specific mortality (HR = 0.35, 95% CI = 0.26 to 0.49, P <.0001.). Mortality benefits for RP were also observed within treatment propensity quintiles, when subjects were pair-matched on propensity scores, and in subgroup analyses based on age, tumor characteristics, and comorbidity.Conclusions Population-based observational data on men diagnosed with localized PC in the mid-1990s suggest a mortality benefit associated with RP vs EBRT. Possible explanations include residual selection bias or a true survival advantage. Results might be less applicable for men facing treatment decisions today.
    CancerSpectrum Knowledge Environment 04/2013; 105(10). DOI:10.1093/jnci/djt059 · 15.16 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In randomized trials, pair-matching is an intuitive design strategy to protect study validity and to potentially increase study power. In a common design, candidate units are identified, and their baseline characteristics used to create the best n/2 matched pairs. Within the resulting pairs, the intervention is randomized, and the outcomes measured at the end of follow-up. We consider this design to be adaptive, because the construction of the matched pairs depends on the baseline covariates of all candidate units. As a consequence, the observed data cannot be considered as n/2 independent, identically distributed pairs of units, as common practice assumes. Instead, the observed data consist of n dependent units. This paper explores the consequences of adaptive pair-matching in randomized trials for estimation of the average treatment effect, conditional the baseline covariates of the n study units. By avoiding estimation of the covariate distribution, estimators of this conditional effect will often be more precise than estimators of the marginal effect. We contrast the unadjusted estimator with targeted minimum loss based estimation and show substantial efficiency gains from matching and further gains with adjustment. This work is motivated by the Sustainable East Africa Research in Community Health study, an ongoing community randomized trial to evaluate the impact of immediate and streamlined antiretroviral therapy on HIV incidence in rural East Africa. Copyright © 2014 John Wiley & Sons, Ltd.
    Statistics in Medicine 11/2014; 34(6). DOI:10.1002/sim.6380 · 2.04 Impact Factor