Constructing inverse probability weights for institutional comparisons in healthcare

Area of Science:

• Health Services Research
• Biostatistics
• Health Economics

Background:

• Disease-specific quality indicators are crucial for comparing hospital care quality across structural, process, and outcome measures.
• Fair hospital comparisons necessitate adjustment for patient case-mix, including demographics, comorbidities, and disease progression, to control for confounding factors.
• Traditional standardization methods like direct and indirect standardization have limitations in complex comparative analyses.

Purpose of the Study:

• To explore inverse probability weighting as an alternative standardization method for comparing hospital quality of care.
• To propose and evaluate methods for handling challenges in inverse probability weighting, particularly with a large number of hospitals and small patient volumes.
• To provide tools for assessing the validity of inverse probability weights, including checks for positivity/overlap and covariate balance.

Main Methods:

• Utilized inverse probability weighting (IPW) by fitting multinomial logistic hospital assignment models to construct weights.
• Developed and applied methods to incorporate small hospital categories into the weighted analysis, addressing data sparsity.
• Employed metrics and visualizations to assess positivity (overlap) and covariate balance, crucial for valid IPW assumptions.

Main Results:

• Demonstrated the feasibility of using inverse probability weighting for hospital quality comparisons, even with numerous hospitals and small patient volumes.
• The proposed methods effectively managed small categories, enhancing the robustness of the standardization procedure.
• The running example using linked administrative data on kidney cancer surgery in Ontario illustrated the practical application and validation of the IPW approach.

Conclusions:

• Inverse probability weighting offers a robust alternative to traditional standardization methods for fair hospital quality comparisons.
• The proposed techniques enhance the applicability of IPW in real-world scenarios with complex data structures and numerous small providers.
• This methodology facilitates more accurate and reliable assessments of healthcare quality, ultimately benefiting patient care and health system management.