Knowledge discovery modeling for building a clinical decision support system: Prevention of hospital-acquired pressure ulcers.

摘要

Pressure ulcers continue to be a nationwide healthcare issue. According to evidence-based practice guidelines, the identification of at-risk individuals using risk assessment tools is a crucial and integral part of preventing this adverse health outcome. Given the need for further refinement of existing risk assessment methods, this study was designed to answer the question, "What is the best model to support clinicians' decision-making in predicting hospital-acquired pressure ulcers?" The purpose of this study was to (a) obtain a better understanding of contributing factors to pressure ulcer development; and (b) examine the best predictive model to identify at-risk patients admitted to a hospital setting. A one-to-one case control study was conducted on a pre-existing dataset created from electronic patient records. As an analytic approach, the knowledge discovery in databases (KDD) process was employed using univariate, multivariate statistical analyses and decision tree induction techniques. The best model and predictors identified from ten subsets of the pre-existing dataset were evaluated using ten additional validation datasets. The best components for predicting pressure ulcer development consisted of eight. Five predictors were routinely collected through electronic patient records---the need for a nurse's accompaniment, edema of cardiovascular system, a foley catheter, nutrition consult, and use of wheelchairs. The remaining three predictors were derived from the Braden subscales---activity, friction/shear, and sensory perception. Entering these eight predictors into the logistic regression model yielded high performance, showing a sensitivity of 92%, a specificity of 67%,and the area under the ROC curve of 89%. This study reveals that patients with impaired mobility need individualized interventions based on their specific risk factors as well as a reduction of external irritation such as friction shear. In order to facilitate evidence-based decision-making for practitioners, however, further advanced, systematic strategies are required in the area of pressure ulcer prevention. As a solution, integrating the best predictive model into electronic health record systems is recommended. Gaps between research and practice will then be reduced and thus, the quality of care for pressure ulcer prevention will be improved. The relationships among pressure ulcer predictions, preventive measures provided, incidence, and cost-benefit ratios will also be clarified.