(12) Development of an algorithm to identify contraceptive exposure in health insurance claims data

Background: Worldwide, almost 1 billion women and their partners use contraceptive products. Questions regarding safety and efficacy of contraceptives in real world settings remain partially unanswered. Real world data, such as health insurance claims data, can help to fill evidence gaps, but identifying contraceptive exposure in this setting remains challenging. Therefore, we developed and assessed an algorithm to identify contraceptive use episodes in a large claims database.

Methods: We used peer reviewed literature, health insurance billing guidances and code books to identify relevant procedures and national drug codes to capture hormonal contraceptive and intrauterine device (IUD) exposure episodes. For short-acting reversible contraceptives, defined as pills, patches, injectables, and vaginal rings, we used the dispensing date and days of supply to determine the episodes start and end date. For long-acting reversible contraceptives (LARC), defined as implants or IUDs, we used insertion date and removal or insertion as start and end date. For LARC episodes without end date we imputed the median duration of complete episodes after stratification by the maximum approved use duration (MAUD) of the contraceptive products. To evaluate the algorithm, we identified women aged 12-55 years from MarketScan® claims data (2005-2019). We required women to have continuous enrollment in the previous 4 years and the year of analysis to examine annual utilization trends. For LARCs, we compared the time from insertion to pregnancy via Kaplan-Meier curves across imputed and non-imputed episodes for each specific LARC product. To capture pregnancy (which relies on measurement of pregnancy outcomes), we required continuous enrollment until the end of MAUD plus 9 months. We excluded women with a previous claim for infertility, and censored at conception date, end of contraceptive episode, claim for infertility, or end of study period.

Results: We identified 1,816,416 eligible women in 2009 and 2,157,888 in 2019. Contraceptive exposure among women increased from 19.3% in 2009 to 28.5% in 2019, with the most distinct increase attributable to IUDs (from 2.0% to 6.6%) and the least for rings & patches (1.2% to 1.3%). Differences in the pregnancy rates across imputed and non-imputed LARC users were comparable in the first two years after insertion. At the end of follow-up, the pregnancy rates differed for implants with a MAUD of 3 years (1.8% (95%CI: 1.6%-1.9%) vs. 5.3% (95%CI: 4.7%-5.8%)) and IUDs with a MAUD of 5 years (7.8% (95%CI: 7.5%-8.1%) and 4.9% (95%CI: 4.8%-5.1%)). The differences remained comparable for IUDs with a MAUD of 3 years (5.1% (95%CI: 4.0%-6.2%) vs. 3.7% (95%CI: 3.2%-4.3%) and 10 years (9.5% (95%CI: 7.2%-11.8%) vs. 10.8% (95%CI: 9.2%-12.4%)) until the end of follow-up.

Conclusion: We developed an algorithm to identify contraceptive exposure in a commercial healthcare claim database and observed an increased utilization of contraceptive products in a cohort of commercially insured women from 2009-2019. Our median imputation approach achieved comparable pregnancy rates across imputed and non-imputed episodes in the first two years after insertion, but depending on the specific LARC type, differences slightly increased after the first two years. Further studies are required to validate the algorithm.