THE EFFECT OF MANAGED CARE ON ICU LENGTH OF STAY - IMPLICATIONS FOR MEDICARE

Objective.-To determine whether insurance status (managed care vs trad itional commercial and Medicare) influences resource consumption (as m easured by length of stay [LOS]) in the intensive care unit (ICU). Des ign.-Retrospective analysis of the 1992 Massachusetts state hospital d ischarge database, using prospectively developed and validated risk-st ratification models. Setting.-All nonfederal hospitals in Massachusett s. Subjects.-Of all adult hospitalizations where an ICU stay was incur red (n=104 270), we selected those covered by 1 of 4 payer groups (n=8 8 050): (1) commercial fee-for-service (patients aged <65 years); (2) commercial managed care (patients aged <65 years); (3) traditional Med icare (patients aged greater than or equal to 65 years); and (4) Medic are-sponsored managed care (patients aged greater than or equal to 65 years). Main Outcome Measure.-Mean ICU LOS. Analysis.-The ICU LOS regr ession models were constructed using split-halves validation to adjust for differences in age, sex, severity of illness, diagnosis, discharg e status, and payer. Separate models were constructed for those younge r than 65 years and those aged 65 years or older. Robustness of the mo dels was explored using goodness of fit and correlation. The effect of payer on hospital mortality was also explored using logistic regressi on. Observed minus predicted mean ICU LOS and mortality rates were cor related with managed care penetration at the hospital level. Results.- The ICU LOS models performed well (R(2)=0.84 and R(L)(2), [likelihood ratio statistic]=0.92 for the development set, and R(2)=0.83 and R(L)( 2)=0.89 for the validation set). Significant covariables affecting LOS included age, severity of principal illness, comorbidity, reason for admission, and discharge status (P<.001 for each). Among the cohort yo unger than 65 years (n=27 805), although unadjusted mean ICU LOS was s horter (2.9 vs 3.43 days; P<.05) for those covered by managed care org anizations, payer status had no independent effect on ICU LOS (P=.48). Among those older than 65 years, there was neither a difference in un adjusted ICU LOS (3.94 vs 3.88 days; P greater than or equal to.05) no r an independent effect of payer on ICU LOS (P=.35). Unadjusted mortal ity was lower among managed care patients (3.9% vs 5.1% in patients ag ed < 65 years [P<.05] and 8.7% vs 12.1% in patients aged greater than or equal to 65 years [P<.05]). Age, severity of principal diagnosis, c omorbidity, and reason for admission significantly influenced mortalit y (P<.001), After controlling for these factors with the mortality mod el (R(L)(2)=0.92 and 0.89, C statistic [12 df]=8.45 and 17.58, and P=. 75 and .13 [where a large P reflects good agreement] for the developme nt and validation sets, respectively), payer continued to have a small but significant effect on mortality (odds ratios ranging from 1.67 at 0.1% probability of death to 1.11 at 30% probability of death). Manag ed care penetration among the commercially insured varied across hospi tals (n=82) from 0% to 68%. There was no correlation between managed c are penetration and either ICU LOS (R(2)=0.04; P=.09) or mortality (R( 2)=0.0; P=.88). Conclusions.-Though patients covered under managed car e consume fewer ICU resources, this appears to be primarily attributab le to a difference in patient-related factors. Thus, as managed care c ase mix changes in the future to include sicker and older patients, th e initial advantages of reduced resource consumption may diminish.