Estimating the Beginning of Pregnancy in German Claims Data: Development of an Algorithm With a Focus on the Expected Delivery Date

Abstract

Background: Estimating the beginning of pregnancy is crucial when studying drug safety in pregnancy, but important information in this regard, such as the last menstrual period (LMP), is generally not recorded in claims databases. The beginning of pregnancy is therefore usually estimated by subtracting a median length of pregnancy from the date of birth. Due to the variability in pregnancy lengths, this might result in non-negligible errors. German claims data may offer the possibility to estimate the beginning of pregnancy more precisely based on the expected delivery date (EDD) which can be coded once or more often during a pregnancy. Purpose: To estimate the beginning of pregnancy in German claims data focusing on the potential of the expected delivery date (EDD). Methods: We included data of all pregnancies in women aged 12–50 years ending in a live birth between 2006 and 2015 identified in the German Pharmacoepidemiological Research Database (GePaRD). We assessed the number of coded EDDs per pregnancy and the concordance if ≥ 2 EDDs were coded. We estimated the beginning of pregnancy by subtracting 280 days from the EDD or the most frequent EDD (in case of discordant EDDs). To examine plausibility, we determined the distribution of pregnancy lengths and assessed whether the gestational age at which prenatal examinations were coded was plausible. For pregnancies without EDD, the beginning was estimated by subtracting the respective observed median lengths of pregnancy for preterm births, term births, and births after due date from the actual dates of birth. Results: In 82.4% of pregnancies, at least one EDD was available (thereof 6.1% with only one EDD and 80.9% with ≥ 2 EDDs that were all concordant). The maximal difference between discordant EDDs was in median 5 days (interquartile range: 3–7 days). Based on the EDD, the median length of pregnancy was 276 days for term births and in 84.7% of pregnancies the second antibody screening test was performed in the recommended interval ± 2 weeks. In pregnancies without EDD the respective proportion was 84.9%. Conclusions: By using the EDD, the beginning of pregnancy can plausibly be estimated in German claims data.

Characterization of pregnancies exposed to St. John’s wort and their outcomes: A claims data analysis

“…There is also information on the date of both prescription and dispensation. Regarding research on drug safety during pregnancy based on GePaRD, different algorithms to identify pregnancies including the date of the outcome and to classify pregnancy outcomes 34,35 , to estimate the beginning of pregnancy 36 , and to link mothers with their offspring 37 have been developed.…”

“…Using the published algorithms [34][35][36][37] , all pregnancies ending between 2006 and 2016 with at least one dispensation of a reimbursable preparation of St. John's wort between beginning and end of pregnancy were included. Accordingly, there had to be a defined end of pregnancy in the data, i.e., an outcome such as life birth or induced abortion.…”

Estimated date of delivery with electronic medical records by a hybrid GBDT-GRU model

“…Fung et al 23 developed a machine learning approach based on ultrasound-derived and fetal biometric data to estimate gestational age and delivery date, but this article did not mention the type of ML. Schink et al 24 developed an algorithm to estimate the beginning of pregnancy in German claims data focusing on the potential of the expected delivery date. Torres et al 25 designed a system to calculate the gestational age and delivery date.…”

Estimated Date of Delivery with Electronic Medical Records by a Hybrid GBDT-GRU Model

“…Schink T at al. [16] developed an algorithm to estimate the beginning of pregnancy in German claims data focusing on the potential of the expected delivery date. Torres M T et al [17] designed a system to calculate the gestational age and delivery date.…”