Background/Aims: A common algorithm using delivery date and preterm birth diagnosis codes recorded in health plan databases to define gestational length had not been rigorously validated.
Methods: We identified live born deliveries among women aged 15–45 years in 2001–2007 within 8 HMORN health plans participating in the Medication Exposure in Pregnancy Risk Evaluation Program. We compared the descriptive statistics between the gestational length derived from an algorithm based upon health plan data and that obtained from the birth certificates (“gold standard”). We calculated the proportions of deliveries for which the two gestational length estimates differ by 0, ±1– 7, ±8–14, ±15–21, ±22–28, or more than ±28 days, stratified by health plan data-defined term/preterm delivery status. We compared the prenatal exposure status to fluoxetine (representing medications used chronically) and amoxicillin (representing medications used for acute conditions) defined using the algorithm-derived gestational length versus the gold-standard gestational length.
Results: The study population comprised 225,384 deliveries. The mean algorithm-derived gestational length was lower than the mean obtained from the birth certificates among singleton deliveries (267.9±8.3 vs. 273.5±14.3 days) but not among multiple-gestation deliveries (253.9±19.0 vs. 252.6±23.1 days). The difference between the two gestational length estimates was within ±7 days in 46% (103,506/225,384) of all deliveries, 45% (91,964/206,492) of term deliveries, and 61% (11,542/18,892) of preterm deliveries. The corresponding proportions for a difference within ±14 days were 77% in both term (159,101/206,492) and preterm (14,565/18,892) deliveries. Among women with continuous enrollment and pharmacy benefit from 100 days before pregnancy through delivery, 2.7% (3,912/146,173) used fluoxetine and 14.0% (20,510/146,173) used amoxicillin during pregnancy based on the gold-standard gestational length. The algorithm-derived prenatal exposure to fluoxetine had sensitivity and positive predictive value (PPV) both above 96% and specificity and negative predictive value (NPV) close to 100%. The accuracy was slightly lower for amoxicillin, with the sensitivity and PPV above 90%, and specificity and NPV above 99%.
Discussion: A gestational length algorithm based upon health plan data correctly classified the prenatal exposure status of drugs intended for chronic and short-term use in most deliveries, but exposure misclassification might be higher for a drug typically prescribed for acute conditions.