Methodology

Data Sources

• CMS 2024 100% LDS Inpatient SAF: diagnoses with present-on-admission (POA) flags, ICD-10-PCS procedures, dates, discharge status, DRG, and de-identified patient descriptors (age category, race, SSA county).
• CMS Care Compare (Hospital General Information): hospital name, address, type, ownership — joined on the CMS Certification Number (CCN).
• AHRQ QI Patient Safety Indicators v2024 ICD-10-CM/PCS value sets (official specification documents).
• AHRQ Elixhauser Comorbidity Software Refined v2026.1 for comorbidity identification.
• CDC/ATSDR Social Vulnerability Index 2022 (county) for socioeconomic context.

Surgical Cohorts

Eight procedure cohorts are defined by ICD-10-PCS codes: CABG, PCI/stenting, hip replacement, knee replacement, hip-fracture fixation, spinal fusion, cholecystectomy, and colorectal resection (814,105 index admissions across 3,520 hospitals).

Risk Adjustment

• Comorbidities: AHRQ Elixhauser Refined v2026.1 flags, counted only from secondary diagnoses that are present on admission (POA ∉ {N, U}) so a complication is never miscounted as a pre-existing condition. Burden is summarized with the van Walraven (2009) weighted score.
• Other factors: age category, emergent admission, inter-hospital transfer, ESRD/disability entitlement, DRG weight, and diagnosis count.

Complications (AHRQ Patient Safety Indicators)

Hospital-acquired complications use the official AHRQ PSI v2024 numerator value sets, applied to secondary diagnoses not present on admission (POA = N/U):

• PSI 09 perioperative hemorrhage/hematoma · PSI 10 postoperative AKI requiring dialysis · PSI 11 postoperative respiratory failure · PSI 12 perioperative PE/DVT · PSI 13 postoperative sepsis · PSI 14 wound dehiscence.
• Other outcomes: in-hospital + 30-day-readmission mortality (composite), 30-day all-cause readmission, and extended length of stay (> procedure 75th percentile).

Risk-Standardized Rates

• For each procedure × outcome we fit a hierarchical (random-intercept) logistic regression, which shrinks low-volume hospitals toward the average and improves reliability (Ash/Normand 2012; Dimick & Staiger 2009).
• The risk-standardized rate = (predicted ÷ expected) × national rate, with a 95% interval. Hospitals whose interval falls entirely below / above the national rate are flagged Better / Worse; otherwise No different — rather than ranked on a bare point estimate.
• Rates are reported for hospitals with ≥ 25 cases (CMS-style public-reporting threshold).

Socioeconomic Context

• Patient county (SSA → FIPS) is linked to the CDC SVI 2022 deprivation ranking. Outcomes are stratified by deprivation quartile (CMS/NQF-preferred) rather than adjusted away.
• A sensitivity analysis compares hospital rankings with vs. without SVI adjustment (ranking-shift). Race is reported for disparity stratification only and is never used as a risk adjuster.

Privacy & Suppression Rules

• Hospitals with fewer than 11 Medicare cases have all detailed metrics hidden.
• Surgeon case counts shown only for high-volume surgeons (top ~10% at that hospital).
• Procedure tables only list procedures with more than 10 cases.

Important Limitations

• We apply the AHRQ PSI numerator value sets and POA logic, but do not run the licensed AHRQ QI (WinQI) denominator/risk engine; risk adjustment uses our own hierarchical model.
• The 2024 LDS does not contain patient sex or dual-eligibility (Medicare-Medicaid) status; deprivation is measured at the county level (ecological).
• Medicare fee-for-service inpatient only: excludes younger, privately insured, and Medicare-Advantage patients.
• ICD-10 coding and POA reporting can under- or over-capture complications.
• Comparisons are made within this Medicare population, not all U.S. patients.