Improving large-scale estimation and inference for profiling health care providers.

Area of Science:

• Health Services Research
• Biostatistics
• Health Informatics

Background:

• Provider profiling is crucial for healthcare quality monitoring, care coordination, and cost-effectiveness.
• Existing generalized linear models face computational challenges with increasing provider numbers.
• Current inferential methods struggle with small providers and extreme outcomes, leading to inaccurate small-sample approximations.

Purpose of the Study:

• To develop computationally efficient algorithms for large-scale provider profiling.
• To create an accurate inferential approach for detecting outlying provider performance, especially in small providers.
• To improve the analysis of healthcare quality using real-world data.

Main Methods:

• A serial blockwise inversion Newton algorithm utilizing the information matrix's block structure.
• A shared-memory divide-and-conquer algorithm to enhance computational efficiency.
• An exact test for provider effects using finite-sample distributions, including the Poisson-binomial distribution for binary outcomes.

Main Results:

• The proposed algorithms significantly reduce computational burden for large-scale provider profiling.
• The exact test provides more accurate inference for outlying performance, especially with small providers.
• Simulation studies confirm improved estimation and inference compared to existing methods.

Conclusions:

• The novel algorithms offer a computationally feasible solution for extensive provider profiling.
• The exact inferential test addresses limitations in detecting outlier performance in small providers.
• These advancements can enhance the monitoring of healthcare quality, exemplified by dialysis facility profiling.