Intervention
The Delaware Contraceptive Access Now (DelCAN) initiative was launched in 2015 by Upstream USA and the state of Delaware, with the goal of ensuring same-day access to all contraceptive methods, and to LARCs in particular, regardless of a patient’s ability to pay. DelCAN ran from 2015 through early 2020 and encompassed several components focused on postpartum patients (Choi et al., 2020). In particular, Medicaid payment policy was altered to allow hospitals to obtain reimbursement for IPP LARC devices. Similar payment mandates were not issued for the privately insured. Hospitals obtained clinical and business operations training which was meant to address the billing and coding challenges that have been encountered in other states implementing IPP payment reform (Fuerst et al., 2021; Lacy et al., 2020). The largest delivery hospital in the state, responsible for approximately 50% of deliveries, had a strong champion provider and began offering IPP LARC services in 2015, while other hospitals joined on a rolling basis between 2016 and 2018.
In addition to hospital-based reforms, DelCAN also included outpatient reforms that may have increased access to contraceptives for postpartum patients. From February 2016 to March 2020, Upstream USA provided LARC training for medical providers and technical assistance to support staff in publicly funded Title X clinics, and in the largest private outpatient clinics in Delaware. Starting in 2015 and continuing through 2019, the Delaware Division of Public Health funded the acquisition of LARC devices, which enabled Title X clinics to provide these methods at no cost to women of any income level, therefore eliminating the sliding scale co-insurance fee that was previously in place for women above the poverty line. Finally, the public awareness campaign “Be Your Own Baby” was fielded from May 2017 to October 2018 to help patients find healthcare providers who could provide free or low cost same-day contraceptives.
To evaluate changes in postpartum LARC use after DelCAN, our analytical design offers three innovations relative to studies assessing similar outcomes after a payment policy change (Liberty et al., 2020; Smith et al., 2021; Steenland et al., 2021). First, we use survey-based, state-representative measurement that assesses LARC use among all postpartum women, not just LARC insertion rates among those enrolled in Medicaid. Second, we use comparison states to account for national trends in LARC use. Third, we assess heterogeneity in LARC use change by Medicaid enrollment. We offer details on each of these points below.
Data Sources
We used data from the 2012 to 2017 Pregnancy Risk Assessment Monitoring System (PRAMS). The PRAMS is an annual, state-level probability survey representative of women who gave birth in each participating state and year. The PRAMS methodology and protocol have received approval from the Institutional Review Boards (IRB) of the Centers for Disease Control and Prevention and of participating states (Shulman et al., 2018). The University of Maryland IRB determined our analyses to be exempt from review. Our analysis was entirely based on survey and administrative data and did not use clinical study or patient data. Throughout the paper, we will sometimes refer to the DelCAN intervention as the “treatment” to adhere to the conventional language used in the difference-in-differences literature (Wing et al., 2018).
Our study period includes 3 years in the pre-treatment period (2012–2014), and 3 years in the treatment period (2015–2017). We included the 15 comparison states for which PRAMS data was available for all years in the study period: Alaska, Illinois, Massachusetts, Maryland, Maine, Missouri, New Jersey, New Mexico, Oklahoma, Pennsylvania, Utah, Washington, Wisconsin, West Virginia, and Wyoming. Colorado and South Carolina were excluded because they also implemented interventions to promote LARC use during the observed years (Liberty et al., 2020; Ricketts et al., 2014). We restricted our analytical sample to the 105,795 women who completed the survey between 2 and 9 months postpartum, who were neither pregnant nor trying to get pregnant at the time of the survey, and who were sexually active. We further excluded a weighted 7% who had missing values in any of the individual variables used in our models, for a final analytical sample of 93,285 women, with 4815 being from Delaware and 88,470 being from the 15 comparison states. The observation numbers reported above and in tables correspond to unweighted frequencies.
All PRAMS respondents were asked “Are you or your husband or partner doing anything now to keep from getting pregnant?,” and if they answered affirmatively, they were asked “What kind of birth control are you or your husband or partner using now to keep from getting pregnant?” Our outcome of interest is a binary variable (1 = yes, 0 = no) that identifies women who were using an implant or an IUD when they were surveyed. Respondents who were using any other contraceptive method, or no method at all, where coded in the reference category. Intervention status was measured by a binary variable that was equal to 1 if the respondent lived in Delaware, and 0 if she lived in any of the 15 comparison states.
The PRAMS asked women “During the month before you got pregnant with your new baby, what kind of health insurance did you have?” and “During your most recent pregnancy, what kind of health insurance did you have for your prenatal care?” Using the responses to these questions, we classified women as enrolled in Medicaid before or during pregnancy versus not enrolled. Women who become eligible for Medicaid during pregnancy have coverage for at least 60 days postpartum (Ranji et al., 2019). Those who were eligible for Medicaid coverage based on their pre-pregnancy income are likely to continue to be covered in the postpartum months.
Individual-level control variables from the PRAMS data included in our analysis were women’s age, race and ethnicity, and marital status, and characteristics of the recent birth, including pregnancy intention, birth order, birth weight, whether the birth was vaginal, and the infant’s age in months. State-level characteristics were added from other sources. We merged state-year data on the number of Federal Qualified Health Centers and rural health clinics per 100,000 women aged 15–50, obtained from the Health Resources & Services Administration (HRSA, 2020). We linked aggregate measures of sociodemographic conditions of women aged 15–50 for each state and year, obtained from the American Community Survey (Ruggles et al., 2019). These measures included the percentages of women living in poverty and with health insurance coverage. Finally, we merged Medicaid income eligibility thresholds for parents and pregnant women per state-year, obtained from the Kaiser Family Foundation (KFF, 2020). Medicaid eligibility rules helped to account for potential confounding from the Affordable Care Act’s Medicaid expansion.
Statistical Analysis
We used difference-in-differences linear probability models (Wing et al., 2018) to estimate the associations of DelCAN with postpartum LARC use. The models compared changes in postpartum LARC use in Delaware, from before DelCAN to the first 3 years of DelCAN, to changes over the same period in the comparison states. The underlying assumption of the approach was that changes in the comparison group represent what would have happened in Delaware had DelCAN not been implemented. Because DelCAN was designed to influence LARC use through different mechanisms among women who were covered by Medicaid and women who were not, we estimated these models first for all women and then separately by Medicaid-coverage status.
All models included state and year fixed-effects, to which we added individual- and state-level time-varying controls. State fixed-effects account for sources of unobserved heterogeneity that may vary across states, but not over time, such as attitudes, cultural traits, or geographic characteristics. In contrast, year fixed effects capture unobserved confounders that vary over time, but are common to all states, such as macroeconomic trends.
Because only Medicaid-covered women were impacted by the IPP payment reform, and because those with and without Medicaid may have been affected by the outpatient components differently, we investigated if associations varied by Medicaid status. To do so, we pooled Medicaid and non-Medicaid women and included a triple interaction of Delaware, Medicaid status, and post-period indicators. This model also included individual controls, and Medicaid by state, Medicaid by year, and state by year interactions. We did not include state-level time-varying controls in these pooled models. In exploratory analyses, we estimated variations of these models in which we excluded state by year interactions and included state, year, and individual and state controls, all interacted by Medicaid, and obtained similar results.
We present results for two versions of each of the models described above. One in which time-relative to the intervention was measured as a binary pre-DelCAN to within-DelCAN (“pre-post”) variable, and one in which each year after the onset of DelCAN was allowed to have its own effect. The later of these approaches evaluated whether the effect of DelCAN was growing over time, coincident with increments in program components implemented over the 3 years (Choi et al., 2020).
Difference-in-differences models rely on the assumption that the trends in the comparison states accurately represent what would have occurred in Delaware had DelCAN not been implemented. Although this assumption cannot be directly tested, we follow the conventional approach (Ryan et al., 2015) and assess its plausibility by presenting a comparison of unadjusted trends in postpartum LARC use before and after the intervention started. All analyses were conducted in Stata 15. All our models adjusted for the complex survey design of PRAMS, using the Stata commands indicated in the survey documentation.