AHRQ Grant HS021854: Related Publication Summaries

Improving Medicare's Hospital Compare mortality model

Silber JH, Satopää VA, Mukherjee N, Rockova V, Wang W, Hill AS, Even-Shoshan O, Rosenbaum PR, George EI.
Health Services Research 2016 Jun;51 Suppl 2:1229-47.

The researchers sought to improve the predictions provided by Medicare's Hospital Compare (HC) to facilitate better informed decisions regarding hospital choice by the public. Their model produces very different predictions from the current HC model, with higher predicted mortality rates at hospitals with lower volume and worse characteristics. The expanded model would advise patients against seeking treatment at the smallest hospitals with worse technology and staffing.

Mortality rate estimation and standardization for public reporting: Medicare’s Hospital Compare

George EI, Rockova V, Rosenbaum PR, Satopaa VA, Silber JH.
Journal of the American Statistical Association 2017; 112(519):933-947.

The authors calibrated Bayesian recommendation systems by checking, out of sample, whether predictions aggregate to give correct general advice derived from another sample. Their process leads to substantial revisions in the Hospital Compare model for acute myocardial infarction mortality. They found that indirect standardization, as currently used by Hospital Compare, fails to adequately control for differences in patient risk factors and systematically underestimates mortality rates at the low volume hospitals. They proposed direct standardization instead.

Fast Bayesian factor analysis via automatic rotations to sparsity

Rockova V, George EI.
Journal of the American Statistical Association 2016; 111(516):1608-1622.

The authors bridged rotational post-hoc transformations and regularization methods with a unifying Bayesian framework, deploying intermediate factor rotations, outputting a solution path indexed by a sequence of spike-and-slab priors. They concluded that the potential of the proposed procedure is demonstrated on both simulated and real high-dimensional data, which would render posterior simulation impractical.

Negotiating multicollinearity with spike-and-slab priors

Rockova V, George EL.
Metronomics 2014 Aug; 1:72(2):217-229.

The authors showed how adding a spike-and-slab prior mitigates the presence of multicollinearity by filtering the likelihood surface into a posterior distribution that allocates the relevant likelihood information to each of the subset model modes. They considered three EM algorithms and compared the regions of convergence. They found that deterministic annealing versions of the EMVS algorithm are seen to substantially mitigate this multimodality, and a single simple running example is used for illustration throughout.

Bayesion Penalty Mixing: The Case of a Non-separable Penalty

Ročková V., George E.I. (2016).
In: Frigessi A., Bühlmann P., Glad I., Langaas M., Richardson S., Vannucci M. (eds) Statistical Analysis for High-

The authors estimated sparse high-dimensional normal means by studying two examples of fully Bayes penalties: the fully Bayes LASSO and the fully Bayes Spike-and-Slab LASSO. They discussed discrepancies and highlighted the benefits of the two-group prior variant. They developed an Appell function apparatus for coping with adaptive selection thresholds and showed that the fully Bayes treatment of a complexity parameter is tantamount to oracle hyperparameter choice for sparse normal mean estimation.

Bayesian estimation of sparse signals with a continuous spike-and-slab prior

Ročková, Veronika.
Annals of Statistics 46 (2018), no. 1, 401-437. doi:10.1214/17-AOS1554.

The authors introduced the family of Spike-and-Slab LASSO (SS-LASSO) priors, which form a continuum between the Laplace prior and the point-mass spike-and-slab prior. They established several frequentist properties of SS-LASSO priors, contrasting them with these two limiting cases. They concluded that the SS-LASSO priors, despite being continuous, possess similar optimality properties as the "theoretically ideal" point-mass mixtures.