Purpose: To assess the generalisability of published clinical risk predictive models for invasive candidiasis in ICU patients.
Methods: The performance characteristics of published clinical risk factor-only and Candida colonisation-only predictive models for invasive candidiasis were assessed in a multicentre cohort of Australian ICU patients. Clinical risk factors and Candida colonisation parameters were collected prospectively from patients.
Results: The two clinical risk factor-only predictive models applied to an Australian patient cohort (n = 615) performed less well than in published studies involving derivation populations. Model performance characteristics improved when Candida colonisation parameters were added post-hoc.
Conclusions: Risk predictive models should factor in both clinical risk factors and Candida colonisation parameters. Integrating these models into therapeutic algorithms first requires external validation in different patient populations and settings.