A Fused Multi-Channel Prediction Model of Pressure Injury for Adult Hospitalized Patients Using Machine Learning Experimental Design رسالة دكتوراة

Abstract

Background: Pressure injuries (PI) are increasing worldwide, and there has been no significant improvement in preventing it. Traditional risk assessment tools are widely used to identify a patient at risk of developing a PI. However, these tools fail to identify valuable risk factors. This study aims to construct a fused multi-channel prediction model of PIs in adult hospitalized patients using machine learning algorithms (MLA). Methods: A multi-phase quantitative approach involves case-control and experimental designs was used to construct a fused multi-channel prediction model of pressure injuries using MLA. The dataset was collected retrospectively between March / 2022 and August / 2023 from the electronic medical records of three private hospitals in Palestine for patients admitted to the hospitals without pressure injuries on the admission day and screened by the Braden scale. The total number of patients 49,500. A balanced dataset was utilized with a total number of 1,110 patients (80% training and 20% testing). Four models were developed and each model recruited eight MLA have been trained and validated with 5-fold cross-validation technique. Performance metrics were used to evaluate the models, and the best model was selected. Results: The balance dataset consists of 1110 patients, including all hospital-acquired pressure injury (HAPI) patients (555 patients) and a random sample of patients without hospital-acquired pressure injury (non-HAPI) patients (555 patients). The performance metrics among the four models showed excellent performance. The best model was random forest, in which accuracy was 0.962, precision was 0.942, the recall was 0.922, F1 was 0.931, area under curve (AUC) was 0.922, false positive rate (FPR) was 0.155, and true positive rate (TPR) was 0.782. Finally, the predictive factors were age, moisture, activity, length of stay (LOS), systolic blood pressure (Systolic BP), and albumin. Conclusion: A novel fused multi-channel prediction models of pressure injury were developed from different datasets, which helps the nurses identify the patients who have the risk of pressure injury earlier, improve the patient satisfaction that affects the quality of nursing care, and promotes patient safety. Furthermore, this study found that some of the predictive factors in the developed model for predicting pressure injuries were not included in the traditional tools, such as age, LOS, Systolic BP, and albumin.

Keywords: Pressure injury, prediction model, nursing care, machine learning, patient safety, quality of care.