Predicting inpatient falls using a shift-based, pre-fall model of interdependency factors: patient, organization, and nurse.

The purpose of this study was to illustrate a methodology for preventing patient falls on a medical surgical unit (MSU) in a small hospital, by applying a shift-based, centric model focusing on system interdependencies (patient, organization, nurse) and primary factors for risk for fall. The research question ask: Are there patient, organization and nurse centric factors associated with falls that can predict a patient’s risk for falling on this MSU.
An exploratory, quantitative study with a retrospective review of 32 patient falls on a Medical Surgical unit was performed.
As data for each patient was collected across shifts (2 shifts on the day prior to the fall, 2 shifts on the day of the fall and 2 shifts on the day after the fall), there was the potential for 192 observations. Accounting for patients that fell on either their admission shift or their discharge shift, the total number of observations was reduced to 155, of which, 84 observations were pre-fall. Data was entered into SPPS.
Chi -square test of independence was performed for each of the 220 independent variables (patient, organization and nurse) with patient falls. Standard logistic regression and generalized linear mixed model logistic regression was applied to identify variables in combination that predict falls and non-falls and identified 5 predictive variables.
Predictive accuracy of the predictive model was calculated using area under the curve (AUC) of the receiver operating characteristics (ROC).
The pre-fall predictive model classified falls and non-falls correctly at 89.3% with a high level of predictive accuracy (area under the curve of .956) and high model quality of 91%.
This study adds to the current body of knowledge to prevent inpatient falls by predicting population specific variables for falls on a MSU. The applied predictive model methodology is generalizable, although the results are not, as this was a small population in one small community hospital.
With the advent of machine learning, the use of large amounts of data, such as 220 variables or more, can be used to make timely decisions to help predict nurse, patient and organization variables impacting falls.

本研究旨在阐述一种在小型医院中，针对医疗外科病房（MSU）预防患者跌倒的方法论。该方法论基于以班次为中心的模型，专注于系统间的相互依存性（患者、组织、护士）以及跌倒风险的首要因素。研究问题提出：是否存在与跌倒相关的患者中心、组织中心及护士中心因素，能够预测患者在MSU中跌倒的风险。通过一项探索性、定量研究，对32起患者在医疗外科病房的跌倒事件进行了回顾性审查。由于每个患者的数据收集跨越了班次（包括跌倒前一天的两个班次、跌倒当天和跌倒后一天的两个班次），因此潜在的观察次数为192次。考虑到部分患者在入院或出院班次跌倒，总观察次数减少至155次，其中84次为跌倒前的观察数据。数据被录入SPSS系统。对与患者跌倒相关的220个独立变量（患者、组织及护士）进行了卡方检验，并运用标准逻辑回归和广义线性混合模型逻辑回归来识别预测跌倒和非跌倒的变量组合，共识别出5个预测变量。使用接受者操作特征曲线（ROC）下的面积（AUC）来计算预测模型的预测准确度。跌倒前的预测模型正确分类跌倒和非跌倒的比例为89.3%，具有高度的预测准确度（曲线下面积为0.956）和91%的高模型质量。本研究丰富了当前关于预防住院患者跌倒的知识体系，所应用的预测模型方法论具有可推广性，尽管其结果并不具有普遍性，因为这是一家小型社区医院的少量患者群体。随着机器学习的兴起，使用大量数据（如220个或更多变量）可以及时作出决策，帮助预测影响跌倒的护士、患者和组织变量。