Hospital standardized mortality ratios: sensitivity analyses on the impact of coding

Area of Science:

• Health Services Research
• Health Informatics
• Quality Improvement

Background:

• Hospital Standardized Mortality Ratios (HSMRs) are administrative database metrics for in-hospital deaths.
• HSMRs adjust for case mix but face criticism regarding sensitivity to clinical coding and data issues.
• Limited quantitative evidence exists on the impact of various adjustments on HSMRs.

Purpose of the Study:

• To assess the sensitivity of HSMRs to different adjustment methodologies.
• To quantify the impact of specific data and modeling variations on HSMR calculations.
• To understand how changes in HSMRs affect hospital performance assessment.

Main Methods:

• Conducted sensitivity analyses comparing regular HSMRs with variant calculations.
• Variations included patient-based measures, altered comorbidity adjustment, palliative care inclusion/exclusion, same-day discharge handling, and diagnosis group selection.
• Analyzed correlations and magnitude of HSMR differences across methods.

Main Results:

• Regular and variant HSMRs demonstrated high correlation (ρ>0.8), yet differences up to 10 points were frequent.
• Excluding palliative care significantly impacted two hospitals; excluding same-day live discharges had minimal effect.
• Capturing post-discharge deaths and using limited diagnosis groups showed the largest impacts.

Conclusions:

• While most HSMRs showed minor changes with tested adjustments, small-to-medium shifts were not rare.
• Even modest HSMR changes can reposition a significant minority of hospitals within funnel plot control limits.
• This highlights the importance of consistent methodology in HSMR calculation for reliable performance evaluation.