Validation of a diagnostic protocol used to identify intimate partner violence in the emergency department setting

To assess the internal validity of a diagnostic protocol developed to facilitate the identification of women with intimate partner violence (IPV)-related injuries.

Using a cross-sectional study design, we enrolled a sample of female subjects presenting to the emergency department for treatment of injuries with non-verifiable etiologies. The study sample was divided randomly into index and validation sets. The index set was used to develop the diagnostic protocol, and the validation set was used to assess the protocol's internal validity. The predictor study variable was risk for IPV-related injury (ie, high or low). The outcome variable was self-report of injury etiology (IPV or other etiology). Appropriate univariate, bivariate, and multivariate statistics were computed, including estimates of sensitivity, specificity, and relative risk. Goodness of fit of the diagnostic protocol was estimated with the Hosmer-Lemeshow statistic. For all analyses, P < or = .05 was considered statistically significant.

The index and validation samples were composed of 200 and 100 women, respectively. In the index sample, subjects categorized at high risk of IPV-related injuries were statistically associated with an increased risk for self-report of IPV-related injury (relative risk, 25.2; 95% confidence interval, 10.6-59.5 [P < .05]; sensitivity, 90.2%; specificity, 96.4%; positive predictive value, 90.1%; negative predictive value, 96.4%). The agreement between the predicted and actual observations showed excellent agreement in the index and validation samples (P = .999, Hosmer-Lemeshow chi(2)).

The proposed diagnostic protocol effectively stratifies risk for IPV-related injuries with good internal validity.