On 2016 Jun 08, David C. Norris commented:

The abstract above must be appreciated in context of the Jun 2014 Erratum, which reports findings of a post-publication review by 3 independent reviewers assigned by Health Affairs. Those findings nullify the abstract’s key statistical claim:

Reviewers found that the authors’ use of a cluster-robust variance estimator was inappropriate in light of the small number of clusters (two) used in this analysis, because variances estimated by this method are biased downward toward zero as the number of clusters diminishes—a problem that had not been detected in the initial reviews of the paper. As a consequence, the standard errors reported in the article are understated, and the reported finding of a significant difference in mortality between the experimental and control groups is not supported by evidence presented in the article.

On 2016 Jun 08, David C. Norris commented:

The abstract above must be appreciated in context of the Jun 2014 Erratum, which reports findings of a post-publication review by 3 independent reviewers assigned by Health Affairs. Those findings nullify the abstract’s key statistical claim:

Reviewers found that the authors’ use of a cluster-robust variance estimator was inappropriate in light of the small number of clusters (two) used in this analysis, because variances estimated by this method are biased downward toward zero as the number of clusters diminishes—a problem that had not been detected in the initial reviews of the paper. As a consequence, the standard errors reported in the article are understated, and the reported finding of a significant difference in mortality between the experimental and control groups is not supported by evidence presented in the article.