M F Shapiro1, R E Park, J Keesey, R H Brook. 1. Division of General Internal Medicine and Health Services Research, UCLA School of Medicine/GIM 90024.
Abstract
OBJECTIVE: This study investigated how mortality differences between groups of municipal versus voluntary hospitals are affected by case-mix adjustment methods. DATA SOURCES AND STUDY SETTING: We sampled about 10,000 random admissions from administrative data for patients hospitalized with each of six conditions in hospitals in New York City during 1984-1987. STUDY DESIGN: We developed logistic regression models adjusting for age and gender, for principal diagnosis, for "limited other diagnoses" (secondary diagnoses that were very unlikely to result from care received), for "full other diagnoses" (all secondary diagnoses irrespective of whether they might have been due to care received), for previous diagnoses, and for other variables. PRINCIPAL FINDINGS: For five of the six conditions, when the limited other diagnoses adjustment was used there was higher mortality in the municipal hospitals (p < .05), with 3.3 additional deaths/100 admissions for myocardial infarction, 1.2 for pneumonia, 8.3 for stroke, 2.8 for head trauma, and 0.8 for hip repair. However, when the full other diagnoses adjustment was used, differences remained significant only for stroke (4.3 additional deaths/100 admissions) and head trauma (1.3) (p < .05). CONCLUSIONS: Estimates of mortality differences between New York City municipal and voluntary hospitals are substantially affected by which secondary diagnoses are used in case-mix adjustment. Judgments of quality should not be based on administrative data unless models can be developed that validly capture level of sickness at admission.
OBJECTIVE: This study investigated how mortality differences between groups of municipal versus voluntary hospitals are affected by case-mix adjustment methods. DATA SOURCES AND STUDY SETTING: We sampled about 10,000 random admissions from administrative data for patients hospitalized with each of six conditions in hospitals in New York City during 1984-1987. STUDY DESIGN: We developed logistic regression models adjusting for age and gender, for principal diagnosis, for "limited other diagnoses" (secondary diagnoses that were very unlikely to result from care received), for "full other diagnoses" (all secondary diagnoses irrespective of whether they might have been due to care received), for previous diagnoses, and for other variables. PRINCIPAL FINDINGS: For five of the six conditions, when the limited other diagnoses adjustment was used there was higher mortality in the municipal hospitals (p < .05), with 3.3 additional deaths/100 admissions for myocardial infarction, 1.2 for pneumonia, 8.3 for stroke, 2.8 for head trauma, and 0.8 for hip repair. However, when the full other diagnoses adjustment was used, differences remained significant only for stroke (4.3 additional deaths/100 admissions) and head trauma (1.3) (p < .05). CONCLUSIONS: Estimates of mortality differences between New York City municipal and voluntary hospitals are substantially affected by which secondary diagnoses are used in case-mix adjustment. Judgments of quality should not be based on administrative data unless models can be developed that validly capture level of sickness at admission.