What's in a day? Determining intensive care unit length of stay

Intensive care unit (ICU) length of stay (LOS) and hospital LOS are common indices used to compare performance of hospitals and are yardsticks used in efforts to contain costs, yet there is no standardized method of quantitating this outcome variable. Attempts have been made to correct LOS according to disease severity. The aim of this study was to quantify and compare ICU LOS using four commonly used methods and to determine the relationship between severity of illness at admission as determined by the Acute Physiology and Chronic Health Evaluation (APACHE) II and APACHE III scoring systems and LOS.

Prospective, cohort study.

Medical and surgical ICUs of a community teaching hospital.

The demographic and clinical data of all patients admitted to the medical ICU and the surgical ICU during a 6-month period were recorded and stored in a computerized database. Coronary care unit boarders and cardiothoracic patients were excluded from analysis. The date and exact time of all admissions and discharges were abstracted from the patients' flowcharts and nurses' notes. The ICU LOS of all patients was calculated using four common methods: a) number of calendar days (LOS-calendar); b) midnight bed-occupancy days (LOS-midnight); c) exact LOS calculated in hours divided by 24 (LOS-exact); and d) the method described by Pollack and Ruttimann (LOS-Pollack). There were 1,004 admissions during the study period; of these, 254 were excluded from analysis (65 coronary care unit boarders and 189 cardiothoracic patients). Of the remaining 750 admissions, 391 were medical ICU patients and 359 were surgical ICU patients. Mean age was 64 +/- 18 yrs, with 420 (56%) male patients. The LOS-calendar differed significantly from the other three methods (p = .001). The LOS-midnight most closely approximated the LOS-exact. The mean (+/- SD) LOS-exact for the entire cohort of patients was 2.8 +/- 3.9 days, with a geometric mean of 1.6 days and a median of 1.4 days. An analysis of the data distribution showed many outliers with the plot markedly skewed to the right. Log transformation of the LOS-exact revealed a normal distribution. The APACHE II and APACHE III scores were significantly higher and the LOS-exact was nonsignificantly higher in the nonsurvivors. There was a poor correlation among the LOS-exact, log LOS-exact, LOS-exact of survivors, and LOS-exact below upper 95th percentile with the APACHE II and APACHE III scores.

We suggest that the LOS-midnight be used to record LOS when a hospital/ICU information system is unable to calculate the exact LOS in hours. Furthermore, because the LOS distribution is highly skewed, the geometric mean and median should be reported. Although APACHE II and APACHE III scores are predictive of group outcomes, they should not be used to predict or adjust for LOS.