Designing a Fall Prediction Model for Inpatient Rehabilitation Facilities Using Readily Available Data

Abstract

ObjectiveTo create a fall risk assessment tool for inpatient rehabilitation facilities (IRFs) using available data and compare its predictive accuracy with that of the Morse Fall Scale (MFS). DesignWe conducted a secondary analysis from a retrospective cohort study. Using a nomogram that displayed the contributions of QI codes associated with falls in a multivariable logistic regression model, we created a novel fall risk assessment, the Inpatient Rehabilitation Fall Scale (IRF Scale). To compare the predictive accuracy of the IRF Scale and MFS, we used receiver operator characteristic (ROC) curve analysis. SettingWe included data from 4 IRFs owned and operated by Intermountain Health. ParticipantsData came from the medical records of 1699 patients. All participants were over the age of 14 and were admitted and discharged from 1 of the 4 sites between January 1 and December 31, 2020. InterventionsNot applicable. Main Outcome Measure(s)We assigned point values on the IRF Scale based on the adjusted associations of QI codes with falls. Using ROC curve analysis, we discovered an optimal cutoff score, sensitivity, specificity, and overall AUC of the IRF Scale and MFS. ResultsROC curve analysis revealed the optimal IRF Scale cutoff score of 22.4 yielded a sensitivity of 0.74 and a specificity of 0.63. With an AUC of 0.72, the IRF Scale demonstrated acceptable accuracy at identifying patients who fell in our retrospective cohort. ConclusionsBecause the IRF Scale uses readily available data, it minimizes staff assessment and outperforms the MFS at identifying patients who previously fell. Prospective research is needed to investigate if it can adequately identify in advance which patients will fall during their IRF stay.