Development and Validation of Risk Prediction Tools for Pressure Injury Occurrence: An Umbrella Review

OBJECTIVES: Pressure injuries (PIs) significantly impact global healthcare systems. Stratifying patients based on their risk of developing PIs enables targeted preventive measures for those at greatest risk. There is a plethora of risk assessment scales and prediction models available. Our objective was to identify and delineate current risk prediction tools, elaborating on their components, and the approaches used in their development and validation.

METHODS: The umbrella review was conducted in compliance with Cochrane guidance. Searches for pertinent systematic reviews of risk prediction tools for PI were carried out across multiple databases (MEDLINE, Embase, CINAHL, EPISTEMONIKOS, Google Scholar), and through reference list checking. Risk of bias was evaluated using a modified version of AMSTAR-2 criteria. Every review encompassed in our study contributed to forming a complete collection of risk prediction tools.

RESULTS: Our search yielded five systematic reviews focused on the development and validation of risk prediction tools for PIs. A further 16 reviews evaluated the prognostic accuracy (sensitivity and specificity) of these tools and 10 examined clinical utility (impact on patient management and PI incidence). Among the reviews on model development and validation, four exclusively covered machine learning models. Only one review provided information on external validation, and it was also the sole review to report on model performance metrics (specifically Area under the ROC curve). Where quality assessment was conducted (3 of 5 reviews), the majority of prediction tools were judged to have a high risk of bias, with none categorized as low risk.

CONCLUSIONS: Available tools fall short of contemporary benchmarks for the development and/or reporting of risk prediction models. A significant proportion of tool lacks external validation. There is a necessity for following standardized and meticulous methodologies in the development and validation of risk prediction models to enable valid risk stratification, targeted care and improved patient outcomes.