Variability in duration of stay in pediatric intensive care units: a multiinstitutional study.

OBJECTIVE

Development of a statistical model to predict length of stay (LOS) in a pediatric intensive care unit (PICU) that adjusts for patient-related risk factors at admission.

DESIGN

Randomized selection of sites by cluster sampling from a 1989 national survey of all hospitals with PICUs, stratified for four quality-of-care factors into 16 clusters (size, presence of an intensive care specialist, medical school affiliation, coordination of care). The data collection was prospective in the selected units.

PATIENTS

5415 consecutive medical, surgical, or emergency admissions to 16 PICUs.

MEASUREMENTS

Patients: Pediatric Risk of Mortality (PRISM) score for the initial 24 hours, admission diagnosis classified into system and cause of the primary dysfunction, operative status, preadmission care, critical care modalities required during the first 24 hours, age, sex, PICU length of stay, and outcome. PICU sites: admission volume, coordination of care, presence of an intensivist, presence of residents, and number of pediatric ICU and pediatric hospital beds.

METHODS

Log-logistic regression analysis of LOS on patient-related and institution-related factors.

RESULTS

Significant (p < 0.05) patient-related predictors of LOS included PRISM, 10 diagnostic groups, 3 preadmission factors (operative status, inpatient/outpatient, previous PICU admission), and first-day use of mechanical ventilation. The ratio of observed to predicted LOS varied among PICUs from 0.83 to 1.25, with three PICUs displaying significantly (p < 0.05) shorter and three PICUs longer LOS. The PICU factors associated (p < 0.05) with shorter (5% to 11%) LOS were presence of an intensivist, presence of residents, and coordination of care, whereas an increased ratio of PICU to hospital beds was associated with longer (p < 0.05) LOS. Medical school affiliation, admission volume, number of pediatric hospital beds, and PICU mortality rates did not have statistically significant effects on LOS when adjusted for patient conditions.

CONCLUSIONS

The predictor can be used to adjust LOS in PICUs for patient-related risk factors, enabling the comparison of resource utilization among different institutions. Organizational factors known to foster team-oriented care are associated with shorter LOS, whereas increased relative PICU size may pose an incentive to keep PICU beds occupied longer.