Development and Application of New Risk-Adjustment Models to Improve the Current Model for Hospital Standardized...

Area of Science:

• Health Services Research
• Medical Informatics
• Healthcare Quality Improvement

Background:

• The hospital standardized mortality ratio (HSMR) is a key performance indicator for hospitals.
• Current HSMR models in South Korea rely solely on administrative data.
• There is a need to assess the validity and potential improvements of HSMR models.

Purpose of the Study:

• To evaluate the validity of the existing HSMR risk-adjusted model in South Korea.
• To compare the performance of models incorporating clinical information against the current administrative data-based model.
• To determine if clinical data can enhance HSMR model accuracy.

Main Methods:

• Analysis of data from 53,976 inpatients.
• Comparison of logistic regression models: Model 1 (administrative data only) vs. models with clinical data (e.g., ASA score, APACHE II, SAPS 3, CCI, cancer stage).
• Model performance assessed using C-index values.

Main Results:

• A strong correlation exists between administrative and clinical data variables.
• The model with all clinical data (Model 2) showed a higher C-index (0.857) than the administrative-only model (Model 1, C-index=0.785).
• A refined model (Model 4) replacing some clinical scores with administrative proxies further improved the C-index to 0.863.

Conclusions:

• Adjusting HSMR models with clinical data significantly improves their validity and accuracy.
• The HSMR evaluation method remains valid even when some clinical information is substituted with administrative data.
• These findings support the integration of clinical data for more robust hospital performance assessment.