University of Utah College of Nursing, 10 S 2000 E, Salt Lake City, UT, 84112, USA.
Department of Computer Science, Boise State University, 777 W Main Street, Boise, ID, 83704, USA.
BMC Med Inform Decis Mak. 2021 Jan 6;21(1):12. doi: 10.1186/s12911-020-01371-z.
Hospital-acquired pressure injuries (HAPrIs) are areas of damage to the skin occurring among 5-10% of surgical intensive care unit (ICU) patients. HAPrIs are mostly preventable; however, prevention may require measures not feasible for every patient because of the cost or intensity of nursing care. Therefore, recommended standards of practice include HAPrI risk assessment at routine intervals. However, no HAPrI risk-prediction tools demonstrate adequate predictive validity in the ICU population. The purpose of the current study was to develop and compare models predicting HAPrIs among surgical ICU patients using electronic health record (EHR) data.
In this retrospective cohort study, we obtained data for patients admitted to the surgical ICU or cardiovascular surgical ICU between 2014 and 2018 via query of our institution's EHR. We developed predictive models utilizing three sets of variables: (1) variables obtained during routine care + the Braden Scale (a pressure-injury risk-assessment scale); (2) routine care only; and (3) a parsimonious set of five routine-care variables chosen based on availability from an EHR and data warehouse perspective. Aiming to select the best model for predicting HAPrIs, we split each data set into standard 80:20 train:test sets and applied five classification algorithms. We performed this process on each of the three data sets, evaluating model performance based on continuous performance on the receiver operating characteristic curve and the F score.
Among 5,101 patients included in analysis, 333 (6.5%) developed a HAPrI. F scores of the five classification algorithms proved to be a valuable evaluation metric for model performance considering the class imbalance. Models developed with the parsimonious data set had comparable F scores to those developed with the larger set of predictor variables.
Results from this study show the feasibility of using EHR data for accurately predicting HAPrIs and that good performance can be found with a small group of easily accessible predictor variables. Future study is needed to test the models in an external sample.
医院获得性压力性损伤（HAPrIs）是发生在 5-10%手术重症监护病房（ICU）患者中的皮肤损伤区域。HAPrIs 大多是可以预防的；然而，由于护理成本或强度，预防措施可能不适用于每个患者。因此，推荐的实践标准包括在常规间隔对 HAPrI 风险进行评估。然而，在 ICU 人群中，没有 HAPrI 风险预测工具显示出足够的预测有效性。本研究的目的是利用电子病历（EHR）数据开发和比较预测手术 ICU 患者 HAPrIs 的模型。
在这项回顾性队列研究中，我们通过查询机构的 EHR，获取了 2014 年至 2018 年期间入住手术 ICU 或心血管外科 ICU 的患者的数据。我们利用三组变量开发了预测模型：（1）常规护理期间获得的变量+Braden 量表（一种压力性损伤风险评估量表）；（2）仅常规护理；以及（3）根据 EHR 和数据仓库的可用性选择的五个常规护理变量的简化集。为了选择预测 HAPrIs 的最佳模型，我们将每个数据集分为标准的 80:20 训练：测试集，并应用了五种分类算法。我们对这三组数据分别进行了处理，根据接收者操作特征曲线和 F 分数的连续性能来评估模型性能。
在纳入分析的 5101 名患者中，有 333 名（6.5%）发生了 HAPrI。五种分类算法的 F 分数被证明是评估模型性能的有价值的指标，考虑到类别不平衡。与使用较大的预测变量集开发的模型相比，使用简化数据集开发的模型具有相当的 F 分数。
本研究结果表明，使用 EHR 数据准确预测 HAPrIs 是可行的，并且可以通过一小组易于访问的预测变量获得良好的性能。需要进一步的研究来在外部样本中测试这些模型。
