Kudos to Tracey Yap, associate professor, and her entire team for the submission of their subcontract proposal to the University of Utah on their Department of Defense (DOD) application entitled: “Early Pressure Injury Prediction among Critical Care Patients: A Machine-Learning Approach.” This proposal is for a four-year period with a start date of September 29, 2021.

The FY20 PRMRP Topic Area of this application is pressure injuries (PrIs, formerly called pressure ulcers). Preventing hospital-acquired pressure injuries (HAPrIs) is critically important in the Military Health System and Veterans Health Administration (VHA) populations because Operation Iraqi Freedom/ Operation Inherent Resolve and Operation Enduring Freedom/ Operation Freedom’s Sentinel veterans are now surviving previously fatal injuries, and those survivors are at risk for a HAPrI.

Background

Most HAPrI are preventable, but, given the continued prevalence, prevention is better served using risk stratification and associated clinical decision support. Risk stratification is particularly important in the Intensive Care Unit (ICU) because current risk assessment tools, such as the widely-used Braden Scale, classify most ICU patients as ‘high risk’ of PrI development. When all patients are ‘high risk,’ it is not possible for nurses to identify highest risk patients and differentiate among patients for care planning. Our research team previously developed HAPrI risk prediction models using data from University of Utah Hospital patients.

Rationale

Accurate HAPrI risk prediction among ICU patients will enable nurses and other members of the interdisciplinary health care team to apply risk-based preventive interventions before a HAPrI occurs. Early and accurate risk prediction will also enable nurses to identify incipient pressure injuries at the earliest, reversible stage (Stage1).

Objective

To produce a validated HAPrI risk-prediction model suitable for future deployment in multiple electronic health records (EHRs).

Specific Aims

Specific Aim 1: Validate and calibrate our previously developed HAPrI prediction models using data from a national sample of Veteran ICU patients.

• Rationale: External validation is necessary to establish validity. VHA patients may be substantively different from the University of Utah patients, and heterogeneity among facilities may require model adaptation to achieve accurate performance.
• Approach: We will calibrate our previously developed models, developing adjustments as necessary for population, data-availability, and facility differences, to provide accurate predictions using EHR data obtained from the VHA’s corporate data warehouse.
• Expected outcome: Externally validated HAPrI prediction models, calibrated to data from a national sample of VHA patients, adjusted by facility if necessary, that can be used in the future to implement HAPrI risk prediction and associated real-time clinical-decision support within the VHA EHR to improve HAPrI prevention.

Specific Aim 2: Develop a longitudinal model to predict HAPrI risk after ICU admission and throughout the ICU stay, incorporating additional dimensions of time, prior conditions, and prior events.

• Rationale: A longitudinal prediction approach is necessary because ICU patients’ latent HAPrI risk fluctuates and can be observed through clinical events and measures, which are recorded in the EHR at time-varying intervals.
• Approach: We will use data from the VHA’s CDW and prepare the predictor variables for analysis using data-appropriate methods to account for varying time intervals between data points (e.g., vital signs). We will apply longitudinal machine learning (ML) prediction methods that use predictions from multiple stacked learning algorithms.
• Expected Outcome: A longitudinal prediction model suitable for providing real-time HAPrI risk prediction throughout the ICU stay, allowing nurses to prioritize time and resources to those at the highest risk.