Identifying potentially ineffective care in a community hospital

To develop a formula to predict mortality for intensive care unit patients between day 5 in an intensive care unit and 100 days after hospital discharge from a community hospital.

Retrospective 1-yr derivation study with validation on a subsequent year's intensive care unit population.

An 850-bed, not-for-profit community hospital with three adult intensive care units, including medical-surgical, cardiac-medical, and cardiac-surgical units.

The development patient set included 4045 consecutive adult admissions to the intensive care unit between July 1995 and June 1996. The validation sample consisted of 4084 admissions between July 1996 and June 1997.

During the first year, 100-day posthospital discharge mortality was predicted by the combination Acute Physiology and Chronic Health Evaluation (APACHE) III predicted mortality on day 5 of >0.92 or the product of day 1 and day 5 APACHE predicted mortality of >0.40, with an increase in the APACHE predicted mortality from day 1 to day 5 of >0.10. Specificity in the development cohort was 0.99, sensitivity was 0.30, and positive predictive value was 0.95. The second-year validation study demonstrated a specificity, sensitivity, and positive predictive value of 0.98, 0.29, and 0.91, respectively, when applying the model to the validation sample.

By using readily available APACHE III data, we were able to identify patients at high risk of dying between intensive care unit day 5 and 100 days after discharge. Although the low sensitivity limits the number of patients for whom death at 100 days is predicted, the high specificity and positive predictive value suggests this information may provide useful information for families and physicians. If these formulas can be validated in diverse institutional settings, decisions regarding short- and long-term outcomes may be improved by using objective survival predictions from two time points.