PJ Kaboli, Iowa City VAMC, VA National Quality Scholars Program and the Univ. of Iowa, Iowa City, IA; MJ Barnett, Iowa City VAMC, Iowa City, IA; SM Fuehrer, Cleveland VAMC, Cleveland, OH; GE Rosenthal, Iowa City VAMC, VA National Quality Scholars Program and the Univ of Iowa, Iowa City, IA

Objectives: Compare severity-adjusted hospital mortality in VA and private sector patients admitted to intensive care units (ICUs) and evaluate the potential bias introduced by differences in hospital length of stay (LOS).

Methods: We studied consecutive ICU patients admitted to a single large VA hospital (N=1,142) and 27 private sector hospitals (N=52,249) serving the same health care market during the period 1/94-12/95. Severity of illness was measured using APACHE III, a validated and widely used methodology based on abnormalities in 17 physiologic parameters during the first 24 hours of ICU care. We used two statistical methods to estimate the risk of death in VA patients, relative to private sector patients. Logistic regression provided the odds of in-hospital death in VA patients. Cox proportional hazards regression was used to account for potential differences in LOS and in the timing of death by censoring patients at the dime of discharge. All analysis adjusted for age, gender, comorbidity, admit diagnosis, admit source, and APACHE III score.

Results: VA patients were similar in age (62.5 vs. 62.9 years; p=.50), but were more likely to be male (98% vs. 52%; p<.001), and had higher APACHE III scores (54 vs. 50; p<.001). Unadjusted in-hospital mortality was higher in VA patients (14.5% vs. 12.0%; p=0.01), as was hospital LOS (28.3 vs. 11.3 days; p<.001). VA patients tended to die later in hospitalization. For example 33% of VA deaths occurred after 21 days compared to only 13% of private sector deaths (p<.001). In a logistic regression analysis, adjusting for admission severity of illness and other factors, the odds of death in VA and private sector patients were similar (OR, 1.16; 95%CI 0.93-1.44; p=.18). However, when the longer hospital LOS in VA patients was accounted for using proportional hazards regression, the hazard of death in the VA was actually lower (OR, 0.70; 95%CI 0.59-0.82; p<.001). This finding was consistent in stratified analyses in medical ICU patients (hazard ratio, 0.80; 95% CI 0.66-0.97; p=.02) and surgical ICU patients (hazard ratio, 0.46; 95%CI 0.33-0.65; p<.001).

Conclusions: If generalizable to VA hospitals in other health care markets, our results indicate that ICU mortality in VA hospitals may be similar or lower than in private sector hospitals. Our results also highlight the importance of using valid severity adjustment methods in comparing hospital mortality in VA patients who may be sicker at the time of admission. Moreover, our findings illustrate the need to consider potential differences in LOS when comparing hospital mortality in VA and private sector hospitals.

Impact: The current study represents the first formal comparison of ICU care in VA and private sector hospitals and provides insight into the relative quality of VA care. The findings also expose the potential bias of in-hospital mortality as a fixed endpoint when comparing outcomes across health care systems with different hospital utilization and discharge practices. In the absence of readily available data on vital status after hospital discharge, health services researchers should consider adopting analytical techniques that censor patients at the time of hospital discharge.