Hospital falls: development of a predictive model for clinical practice

Abstract

A retrospective case-control study related to falls was conducted at an 1,120-bed acute care tertiary hospital. The case (fall) sample consisted of 102 falls and 236 control (nonfall) charts during a 1-month period. An instrument developed by Hendrich (1988) was modified for use in the study. Demographic data and risk factors were recorded. Descriptive statistics included risk factor percentages for each sample and the corresponding univariate relative risks. Logistic regression was used to develop a multivariate risk factor model with seven risk factors. The significant risk factors were recent history of falls, depression, altered elimination patterns, dizziness or vertigo, primary cancer diagnosis, confusion, and altered mobility. The adjusted relative risks were converted to risk points to be used to assess a patient's level of fall risk. Within the data set, a sensitivity of 77% (79 of 102) and specificity of 72% (169 of 236) were calculated. The model was cross-validated in a 1987 data set with a sensitivity of 83% (59 of 71) and specificity of 66% (106 of 161).