|
Citation
|
Milosevic V, Linkens A, Winkens B, Hurkens KPGM, Wong D, van Oijen BPC, van der Kuy HM, Mestres-Gonzalvo C. BMJ Open 2021; 11(5): e042941. |
|
Abstract
|
OBJECTIVES: To develop (part I) and validate (part II) an electronic fall risk clinical rule (CR) to identify nursing home residents (NH-residents) at risk for a fall incident.
DESIGN: Observational, retrospective case-control study. SETTING: Nursing homes. PARTICIPANTS: A total of 1668 (824 in part I, 844 in part II) NH-residents from the Netherlands were included. Data of participants from part I were excluded in part II. PRIMARY AND SECONDARY OUTCOME MEASURES: Development and validation of a fall risk CR in NH-residents. Logistic regression analysis was conducted to identify the fall risk-variables in part I. With these, three CRs were developed (ie, at the day of the fall incident and 3 days and 5 days prior to the fall incident). The overall prediction quality of the CRs were assessed using the area under the receiver operating characteristics (AUROC), and a cut-off value was determined for the predicted risk ensuring a sensitivity ≥0.85. Finally, one CR was chosen and validated in part II using a new retrospective data set.
RESULTS: Eleven fall risk-variables were identified in part I. The AUROCs of the three CRs form part I were similar: the AUROC for models I, II and III were 0.714 (95% CI: 0.679 to 0.748), 0.715 (95% CI: 0.680 to 0.750) and 0.709 (95% CI: 0.674 to 0.744), respectively. Model III (ie, 5 days prior to the fall incident) was chosen for validation in part II. The validated AUROC of the CR, obtained in part II, was 0.603 (95% CI: 0.565 to 0.641) with a sensitivity of 83.41% (95% CI: 79.44% to 86.76%) and a specificity of 27.25% (95% CI 23.11% to 31.81%).
CONCLUSION: Medication data and resident characteristics alone are not sufficient enough to develop a successful CR with a high sensitivity and specificity to predict fall risk in NH-residents. TRIAL REGISTRATION NUMBER: Not available.
Language: en |