Determining antenatal medicine exposures in South African women: a comparison of three methods of ascertainment

Background In the absence of clinical trials, data on the safety of medicine exposures in pregnancy are dependent on observational studies conducted after the agent has been licensed for use. This requires an accurate history of antenatal medicine use to determine potential risks. Medication use is commonly determined by self-report, clinician records, and electronic pharmacy data; different data sources may be more informative for different types of medication and resources may differ by setting. We compared three methods to determine antenatal medicine use (self-report, clinician records and electronic pharmacy dispensing records [EDR]) in women attending antenatal care at a primary care facility in Cape Town, South Africa in a setting with high HIV prevalence. Methods Structured, interview-administered questionnaires recorded self-reported medicine use. Data were collected from clinician records and EDR on the same participants. We determined agreement between these data sources using Cohen’s kappa and, lacking a gold standard, used Latent Class Analysis to estimate sensitivity, specificity and positive predictive value (PPV) for each data source. Results Between 55% and 89% of 967 women had any medicine use documented depending on the data source (median number of medicines/participant = 5 [IQR 3–6]). Agreement between the datasets was poor regardless of class except for antiretroviral therapy (ART; kappa 0.6–0.71). Overall, agreement was better between the EDR and self-report than with either dataset and the clinician records. Sensitivity and PPV were higher for self-report and the EDR and were similar for the two. Self-report was the best source for over-the-counter, traditional and complementary medicines; clinician records for vaccines and supplements; and EDR for chronic medicines. Conclusions Medicine use in pregnancy was common and no single data source included all the medicines used. ART was the most consistently reported across all three datasets but otherwise agreement between them was poor and dependent on class. Using a single data collection method will under-estimate medicine use in pregnancy and the choice of data source should be guided by the class of the agents being investigated