Specialty ICUs are believed to improve mortality and shorten length of stay because the limited number of diagnoses enables staff to be more knowledgeable about the diseases and their clinical courses. Similarly, the diversity of diagnoses in general PICUs would also be expected to effect mortality and length of stay in the same manner. We assessed the effect of diagnostic diversity on severity-adjusted mortality and length of stay in 32 PICUs.

Methods: Data from consecutive admissions to a diverse group of 32 PICUs (n= 11,165, deaths = 543) included PRISM III, outcome, length of stay, and descriptive information. Expected mortality risk and length of stay were computed with validated severity and diagnosis adjusted models. Diagnoses were classified by the system and etiology of the primary reason for PICU admission, resulting in 21 categories with at least 2% of all admissions. Diagnostic diversity for each PICU was measured by the logarithmically weighted average of the proportions of patients in the diagnostic categories(entropy). This measure was normalized to be between 0 (all patients in one category) and 1 (patients uniformly distributed over all categories). Analyses at the patient and at the institution levels were performed. At the patient level, diagnostic diversity was included as a predictor variable into regression models for risk of mortality and length of stay (LOS). At the institution level, a linear regression analysis of either the institutional standardized mortality ratios (SMR = observed/predicted mortality rate), or the institutional observed/predicted LOS ratios versus diagnostic diversity was performed.

Results: Both, at the patient and institution level, the SMR was not associated with diagnostic diversity (p >.45, and p >.47). However, the LOS ratio was positively associated with diversity. PICUs with less diagnostic diversity had reduced length of stay (p < 0.0001, and p = 0.05).

Conclusion: PICUs with fewer diagnostic entities (e.g. specialty PICUs) discharge patients relatively earlier. However, there is no evidence for reduced severity-adjusted mortality rates.