Reliability of Maternal Self-Report of Medical Conditions and Obstetric Interventions

Background:  Pregnancy and perinatal complications have been implicated as risk factors in the etiology of autism spectrum disorders (ASD).   Since acquisition and abstraction of data from medical records is both expensive and labor-intensive, most retrospective studies rely on maternal self-report of such events.

Objectives:  We compared the reliability of maternal report of peri-gestational medical conditions at time points near to when conditions would have occurred and again two to five years later.

Methods:  Ninety-four participants in the high-risk autism spectrum disorders cohort MARBLES (Markers of Autism Risk – Learning Early Signs) underwent structured telephone interviews during and immediately following pregnancy and again between two and five years later.  Kappa statistics measured agreement between self-report during the peri-gestational period and recall two to five years later.

Results:  While agreement between self-report of pre-pregnancy conditions; including diabetes, hypothyroidism and polycystic ovarian syndrome; over time was superb with kappas ranging from 0.85 to 1.00, recall of chronic hypertension was poor to fair (kappa 0.18, 95% CI -0.04, 0.40).  Agreement of maternal recollection of vaginal bleeding and thyroid dysfunction during pregnancy over time was mediocre, each with a kappa of 0.58;  more women reported having had vaginal bleeding or a thyroid disorder when asked during and just after pregnancy compared with recall years later.  Women were also far more likely to report having had preeclampsia when interviewed within a few months of the diagnosis (kappa 0.26, 95% CI -0.18, 0.70).  There was substantial agreement between reports over time of gestational diabetes and preterm labor, each with a kappa 0.82, though women interviewed years later tended to believe that their gestational diabetes was diagnosed earlier than women reporting within months of the diagnosis.  Finally, agreement of maternal recall of amnioscentesis was better than that of chorionic villus sampling (kappa 0.61, 95% CI 0.39, 0.82 vs 0.26 95% CI -0.18, 0.70), perhaps reflecting confusion about the difference between the two procedures.

Conclusions:  Studies attempting to link gestational conditions and obstetric interventions to the subsequent health and development of the offspring require careful attention to exposure assessment.   Agreement of repeated measures of maternal recall regarding medical diagnoses and procedures included in this study varied by the condition under study, and women were more likely to remember less common events when asked close to the time the diagnosis or procedure would have occurred.  This potential for exposure misclassification must be taken into consideration when designing and evaluating studies using self-report of pregnancy events.