Abstract
BACKGROUND: Studies examining the potential impact of mothers' health during pregnancy on the health of their offspring often rely on self-reported information gathered several years later. To assess the validity of this approach, we analysed data from a national case-control study of childhood cancer (diagnosed <15 years) that collected health information from both interviews and medical records. METHODS: Mothers' interview reports of infections and medications in pregnancy were compared with primary care records. Taking clinical diagnoses and prescriptions as the reference, sensitivity and specificity of maternal recall along with kappa coefficients of agreement were calculated. Differences in the odd ratios estimated using logistic regression for each information source were assessed using the proportional change in the odds ratio (OR). RESULTS: Mothers of 1624 cases and 2524 controls were interviewed ∼6 years (range 0-18 years) after their child's birth. Most drugs and infections were underreported; in general practitioner records, antibiotic prescriptions were nearly three times higher and infections >40% higher. Decreasing with increasing time since pregnancy, sensitivity was ⩽40% for most infections and all drugs except 'anti-epileptics and barbiturates' (sensitivity 80% among controls). ORs associated with individual drug/disease categories that were based on self-reported data varied from 26% lower to 26% higher than those based on medical records; reporting differences between mothers of cases and controls were not systematically in the same direction. CONCLUSIONS: The findings highlight the scale of under-reporting and poor validity of questionnaire-based studies conducted several years after pregnancy. Future research using prospectively collected data should be encouraged to minimize measurement errors.