Outcome-dependent sampling in cluster-correlated data settings with application to hospital profiling

Area of Science:

• Health Services Research
• Biostatistics
• Health Economics

Background:

• Hospital readmission rates are critical quality indicators and influence healthcare reimbursement.
• Current models adjust for case mix but lack socio-economic status (SES) data, hindering healthcare disparity analysis.
• Collecting comprehensive SES data poses a significant burden on hospitals.

Purpose of the Study:

• To propose an efficient sampling strategy for incorporating SES data into hospital readmission analyses.
• To enable more accurate identification and mitigation of healthcare disparities.
• To reduce the data collection burden on hospitals.

Main Methods:

• Utilized a cluster-stratified case-control design for outcome-dependent sampling of patients.
• Employed pseudo-maximum-likelihood estimation with inverse probability weights for generalized linear mixed models.
• Applied the method to Medicare data for congestive heart failure readmissions across numerous hospitals.

Main Results:

• The proposed cluster-stratified case-control sampling is efficient for estimating fixed and random effects in generalized linear mixed models.
• This approach effectively adjusts for unobserved or costly-to-collect covariates like SES.
• The framework mitigates disparities in identifying underperforming hospitals compared to analyses lacking SES adjustment.

Conclusions:

• Outcome-dependent sampling offers a viable solution for integrating SES data into readmission analyses.
• The proposed methodology enhances the accuracy of hospital quality assessments and supports efforts to reduce healthcare disparities.
• This approach provides a practical means to improve healthcare quality measurement and policy.