OBJECTIVES: This study investigated the agreement between self-reported and medical record assessment of exposure to violence and the impact of misclassification on the estimation of the association between exposure to violence and infant death and very low birthweight. METHODS: The study population consisted of women who participated in two case-control studies on infant death and very low birthweight. There were 254 pairs of interviews and medical record reviews available for comparison. RESULTS: A total of 153 women (60.2%) reported ever being exposed to violence--92 (60.1%) based on the interview only, 18 (11.8%) based on the medical record only, and 43 (28.1%) in both sources. The sensitivity of the violence variables was low, ranging from 16.9% to 31.9% and kappa statistics showed poor agreement. Lower rates of all types of violence were found through the medical record than through the interview. CONCLUSION: Prevalence of violence based on medical record alone had a high degree of misclassification and some odds ratios were biased toward the null. Studies in which violence is an exposure, outcome, or confounder must use participant interviews in order to gather accurate information. A combination of sources may be the most accurate.
OBJECTIVES: This study investigated the agreement between self-reported and medical record assessment of exposure to violence and the impact of misclassification on the estimation of the association between exposure to violence and infantdeath and very low birthweight. METHODS: The study population consisted of women who participated in two case-control studies on infantdeath and very low birthweight. There were 254 pairs of interviews and medical record reviews available for comparison. RESULTS: A total of 153 women (60.2%) reported ever being exposed to violence--92 (60.1%) based on the interview only, 18 (11.8%) based on the medical record only, and 43 (28.1%) in both sources. The sensitivity of the violence variables was low, ranging from 16.9% to 31.9% and kappa statistics showed poor agreement. Lower rates of all types of violence were found through the medical record than through the interview. CONCLUSION: Prevalence of violence based on medical record alone had a high degree of misclassification and some odds ratios were biased toward the null. Studies in which violence is an exposure, outcome, or confounder must use participant interviews in order to gather accurate information. A combination of sources may be the most accurate.