| Description |
Hospital-acquired pressure injuries (PI) are localized areas of damage to the skin, underlying tissue, or both, as a result of pressure. Critical-care patients represent a highly specialized patient population, and currently available risk-assessment scales, such as the Braden scale, tend to identify most critical-care patients as being "at risk" for pressure injuries, and therefore are of limited clinical utility. The purpose of this dissertation was to (a) conduct a systematic review of the literature to identify independent risk factors for pressure injuries, (b) use longitudinal analysis to identify the hazards of developing a pressure injury based on changing Braden Scale total and subscale scores, and (c) develop a PI prediction model. We conducted our systematic review based on standardized criteria and developed a tool for quality assessment based on a literature search and input from experts. Mobility/activity, age, and vasopressor infusion emerged as important risk factors, whereas results from other risk categories were mixed. For the Braden scale analysis and the predictive model we used electronic health record cases (N=6,376). We employed time-dependent Cox regression to determine the hazards of developing a pressure injuries based on the Braden scale subscale scores. With the exception of the friction and shear subscales, patients of all ages with midrange Braden scale scores were more likely to develop pressure injuries than their counterparts with higher risk scores. We developed a predictive model using random forest analysis. The model, an ensemble classifier, was composed of 500 decision trees, each using a random subset of 4 of 20 clinical features. The area under the receiver operating characteristic curve was 0.9 for the outcome >category 1 pressure injuries and 0.87 for the outcome >category 2 pressure injuries. The most important variables in our model in descending order based on the mean decrease in accuracy were longer surgical duration, lower hemoglobin, higher creatinine, older age, higher glucose, lower body mass index, lower albumin, and higher lactate. Due to our model's relatively strong performance, it may be useful for directing preventive interventions that are not feasible for every patient due to cost. |
