Describing the performance of U.S. hospitals by applying big data analytics

Area of Science:

• Health Services Research
• Health Informatics
• Machine Learning in Healthcare

Background:

• Public reporting of hospital performance is crucial for quality improvement.
• Characterizing hospital performance is complex due to numerous quality measures.
• Existing systems often aggregate measures, potentially obscuring nuanced performance differences.

Purpose of the Study:

• To develop a novel method for characterizing hospital performance using multiple quality measures.
• To identify common patterns and variations in hospital performance profiles.
• To compare the insights from the novel method with existing hospital rating systems.

Main Methods:

• Developed a semi-supervised machine learning algorithm.
• Applied the algorithm to publicly available quality measures for 1,614 U.S. hospitals.
• Utilized graph analytics for visualization and quantitative characterization of hospital performance.

Main Results:

• Identified distinct clusters of hospitals with shared performance profiles.
• Revealed that top-ranked hospitals by existing systems exhibit diverse performance strengths.
• Demonstrated that the novel approach uncovers nuanced performance differences not apparent in composite scores.

Conclusions:

• A novel graph analytics and machine learning approach provides a more nuanced understanding of hospital performance.
• Current composite hospital rating systems may oversimplify performance by masking diverse excellence.
• This method offers a valuable tool for quality improvement and comparative analysis of healthcare providers.