Using population death rate to predict rate of admissions to the intensive care unit

Objective: To determine whether the intensive care unit (ICU) admission rates of new health plan members can be predicted by the mortality of non-ICU-treated members. Design: Retrospective study of health records. Patients: Five sequential cohorts of new health plan members (298,974 members) seen at any of three tertiary care medical centers of a health maintenance organization in northern California who joined the health plan during the first quarter of 1994, 1995, 1996, 1997, or 1998 and retained membership for greater than or equal to1 yr. Setting: Three medical centers in northern California. Interventions: None. Measurements and Main Results: We measured rates of ICU admission, death, and mechanical ventilation among cohort members. ICU admission rate varied between hospitals and over time but was predicted by non-ICU mortality-ICU admission rate = 0.83 x non-ICU mortality-and was linear throughout its range. In no hospital or time period was a higher mortality associated with fewer ICU admissions. Seventeen percent of population deaths occurred among ICU patients and did not differ among medical centers. Conclusions: A single linear equation predicted ICU admission rate from death rate of non-ICU-treated patients among cohorts of new members. ICU admission rates can be predicted from a measure of population illness burden, such as the mortality of non-ICU-treated patients. It may be possible to extend this analysis to other hospitals and health care systems to evaluate the adequacy of ICU services provided.