Objective: To test if there was significant variation of mean adjusted duration of respiratory support (RS) at the site level among Australian and New Zealand pediatric intensive care units (PICUs) and to determine whether the pattern of variation differed from the site-level pattern of variation in length of stay (LoS).
Design: Separate outcome prediction models for estimating PICU LoS and duration of RS were constructed, using patient data collected between 2002 and 2007. Gamma regression was used to model LoS for all admissions included in the study population, and log normal regression was used to model duration of RS for the subset of patients receiving RS. For both models, case-mix adjustment was achieved by entering patient risk factors as fixed effects, and the PICU or site of care was entered as a random effect.
Setting: Data for 31,358 admissions were collected from nine specialist PICUs in Australia and New Zealand.
Measurements and main results: Average risk-adjusted duration of RS and LoS for each PICU. There was significant unit-level variation in the adjusted mean LoS and duration of RS among PICUs in Australia and New Zealand. One site had a mean duration of RS that was significantly longer than expected, whereas two sites had a mean duration of RS that was significantly shorter than expected at the 95% level. Unit-level variation in duration of RS is consistent with unit-level variation in LoS for six PICUs and significantly different in two units.
Conclusion: There is unit-level variation in LoS and duration of RS, not accounted for by case-mix. Concurrent analysis of unit-level variation in LoS and duration of RS can help to identify differences in discharge practice and provide direction for improvements in clinical or administrative efficiency.