How Accurate are Hospital Discharge Data for Evaluating Effectiveness of Care?

Jesse Green, Neil Wintfeld

Medical Care · 1993 · 185 citations · 31 references

Concepts

TL;DR

Demand for quality‑of‑care data has driven the publication of adjusted hospital mortality rates, yet the accuracy of the underlying databases is questionable and can mislead consumer‑oriented provider profiles. The study analyzes audit results to assess how coding errors affect expected death rates used in Medicare hospital mortality reports and calls for stronger data quality control. The authors audited a stratified random sample of discharge abstracts from Medicare‑aged California patients using the state health data agency’s review. The audit revealed widespread coding errors, including underreporting of comorbidities and nursing‑home transfers and miscoding of admission urgency, with hospital error rates varying significantly and producing measurement errors of 0.2 to 2.2 expected 30‑day deaths per 100 admissions.

Abstract

Demand for quality of care data has led to publication of adjusted mortality rates of hospitals and physicians. Yet the accuracy of databases used for this purpose is questionable, raising the possibility that consumer-oriented profiles of providers could be misleading. A stratified random sample of discharge abstract records of Medicare-aged patients hospitalized in California were audited by the state's health data agency. The results of the audit were analyzed to determine the effect that coding errors have on expected death rates assigned to hospitals as risk-adjustment measures in the annual Medicare hospital mortality report. Discharge abstracts of Medicare-aged patients contained many errors among coded risk factors used to calculate expected death rates. Comorbidities and transfers from nursing homes were seriously underreported and ‘urgency of admission’ was often miscoded. Hospitals differed with respect to error rates (P < 0.001); these varying levels of miscoding caused measurement errors that ranged in size from 0.2 to 2.2 expected 30-day deaths per 100 admissions (10th vs. 90th percentile). Detailed knowledge of the limitations of claims data can help analysts minimize the impact of coding error. Beyond this, quality control of data used for outcomes research needs strengthening.

References

31