Defining maternal obesity in studies of birth outcomes: Comparing ICD-9 codes at delivery and measures on the birth certificate.

BackgroundUsing ICD-9 codes underestimates the prevalence of obesity in adults; however, the validity of these codes in studies of pregnancy-related outcomes is not known.ObjectivesTo compare classification of maternal obesity based on ICD-9 codes in hospital discharge records versus data from birth certificates in the same women, examine predictors of agreement, and assess how associations between obesity and two birth outcomes differ by source of weight data.MethodsThis population-based study included 2 329 145 California births between 2007 and 2012. We compared data on obesity from childbirth hospital discharge records (ICD-9 codes for obesity) and birth certificates (pre-pregnancy body mass index (BMI) calculated from weight and height) and identified predictors of agreement between the two sources. Logistic regression models assessed whether the two definitions of obesity resulted in different estimates of the associations of obesity with caesarean birth and large-for-gestational age.ResultsOverall, 464 754 women (20.0%) had obesity based on their pre-pregnancy BMI while only 100 002 (4.3%) had an obesity-related ICD-9 code. The sensitivity of ICD-9-based obesity was low at 16.2%; however, obesity codes were highly specific at 98.7%, with a negative predictive value of 82.5% and a positive predictive value of 75.2%. Among women with obesity identified by the birth certificate, those with pre-pregnancy and pregnancy-related complications (eg diabetes and hypertension) were more likely to have an obesity-related diagnosis in their delivery hospital discharge record. Using ICD-9 codes overestimated the association of obesity with caesarean birth and newborn large-for-gestational age.ConclusionsICD-9 codes in childbirth discharge records captured only one in five women with pre-pregnancy obesity. Sensitivity varied by maternal characteristics and conditions. This misclassification resulted in bias when examining the association of obesity and p