Background: Injurious fall is a serious hospital-acquired condition. Screening tools for injurious falls in hospitalized patients have received limited evaluation.
Objective: To compare operating characteristics of a succinct screening tool for injurious falls, the University of Pittsburgh Medical Center (UPMC) screening tool (based on mobility, fall history, and nursing judgment), with the ABCS injurious fall screening tool (based on Age, Bone, Coagulation, and recent Surgery).
Design: Case control study.
Methods: Hospitalized patients with injurious falls were identified from the UPMC adverse events database for 2007–2008 (N = 43). Controls (n = 86) matched for age, location, and period of fall event were selected from the hospital’s administrative database. Tools were evaluated independently by 2 screeners using electronic charts. Interrater agreement, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and c-statistics for the screening tools were calculated.
Results: Case and control groups were similar in age, sex, and race. Interrater agreement was 71% for ABCS and 72% for UPMC screens. ABCS and UPMC screens had sensitivity of 60.5% (95% CI, 52.0%–68.9%) and 62.8% (95% CI, 54.5%–71.1%), specificity of 41.9% (95% CI, 33.4%–50.4%) and 58.1% (95% CI 49.6%–66.7%), and c-statistics of 51.2% and 59.3%, respectively. With a 33% prevalence of injurious fall, the PPV was 34.2%, and NPV was 67.9% for ABCS, and the PPV was 42.9%, and NPV was 75.8% for UPMC. Operating characteristics were not statistically significantly different, although the UPMC screen was 8% more accurate in predicting injurious falls and had a lower false-positive rate (44.2% versus 65.1%).
Conclusions: Compared with the ABCS screen, the UPMC screen is a simple, practical tool. Prospective studies are needed to establish the UPMC tool’s predictive value in hospital practices with lower rates of injurious falls. In general, better screening tools for injurious falls should be developed to meet quality standards.