Pregnancy and delivery is the single largest group of diagnoses, by cost, for employers providing health insurance benefits, accounting for approximately $30 billion in hospital bills in 2008.1 Avoiding unnecessary hospitalizations is a focus for insurers and employers of cost reduction, but hospital admissions for deliveries are inherently different from other admissions, because they cannot be avoided or substituted with other appropriate outpatient care. Although the majority of pregnancies are planned, many are not; in fact, the unintended pregnancy rate among employer-sponsored health insurance plans was as high as 29% in 2002.2 In this study, we use the term “employer,” because employer-sponsored health programs cover the majority of commercially insured lives, but the results are applicable to other forms of commercial insurance (eg, policies sold directly to individuals on exchanges or union-sponsored plans).
Unintended pregnancy is a major cost component for employer-sponsored health benefits; yet, the most recent cost estimates are based on surveys that were conducted almost 20 years ago (in the mid-1990s), and do not reflect the socioeconomic and healthcare changes that have occurred since that time.2 Furthermore, published analyses of the costs of unintended pregnancy have often focused on spending by public insurance programs, perhaps because approximately 66% of births resulting from unintended pregnancies are paid for by such programs.3
Analyses based on spending by public insurance programs suggest that implementing or expanding public policies to prevent unintended pregnancies has the potential to provide substantial savings to the public.4 In addition, findings from a recent literature review on the costs of pregnancy suggest that reducing unintended pregnancies may lower the overall economic burden of pregnancy on the US healthcare system.3 Given all these factors, we sought to update the cost to employers of unintended pregnancy with more current data.
The purpose of our study was to model pregnancy intention in a large commercial claims database and to provide aggregate estimates about the typical employer cost and healthcare resource utilization of deliveries associated with unintended pregnancy. An additional goal was to create a model that could be applied retrospectively to a health plan’s claims data to identify the risk that the pregnancies covered by the health plan were unintended. Such identification could be used as an aid for population health efforts to reduce unintended pregnancy.
We used 2009-2011 data from the Centers for Disease Control and Prevention’s Pregnancy Risk Assessment Monitoring System (PRAMS) database5 and 2010-2011 data from the Truven Health MarketScan Commercial Claims and Encounters Database6 (hereafter MarketScan) to estimate the probability of unintended pregnancy in a commercially insured population.
The 2009-2011 PRAMS database contains population-based data on maternal attitudes and experiences before, during, and shortly after a live birth from approximately 50,000 women in approximately 40 states.5 The PRAMS questionnaire has more than 300 questions, which focus on topics such as pregnancy intention, contraception use, patient demographic characteristics, health insurance status, and the general health of the mother before and during pregnancy. Researchers can gain access to the PRAMS database through the Centers for Disease Control and Prevention5 and through individual state health departments.
The MarketScan database, which includes medical and pharmacy claims for approximately 50 million commercially insured lives, contains International Classification of Diseases, Ninth Edition diagnosis codes, Current Procedural Terminology codes, National Drug Codes, identifiers of individuals associated with homogeneous benefit design groups, and identifiers that allow longitudinal studies of individuals.6 MarketScan is a proprietary database of Truven Health.
The PRAMS questionnaire contains 2 parts. Part 1 is a set of 56 core questions asked in all participating states. Part 2 is a set of standard state-specific questions. In our analysis, we used 7 of the core questions and 4 of the standard state-specific questions. The questions were chosen based on our ability to find responses from the PRAMS questionnaire (eg, the mother’s age) in the MarketScan claims data. Commercial claims databases, such as MarketScan, do not include most socioeconomic information (eg, income and race).
We identified pregnancy intention by core question 11 in the PRAMS questionnaire, which asks, “Thinking back to just before you got pregnant with your new baby, how did you feel about becoming pregnant?” We mapped the respondent answers to either intended or unintended categories.
Unintended pregnancy was defined as a mistimed (“I wanted to be pregnant later”) or unwanted (“I didn’t want to be pregnant then or at any time in the future”) pregnancy. Pregnancy intended at the time of conception (“I wanted to be pregnant then”) or earlier (“I wanted to be pregnant sooner”) was defined as intended. These definitions based on the PRAMS questionnaire are consistent with definitions found in other publications.7
In our analysis, we excluded women who were uninsured or covered by Medicaid, to remain consistent with the MarketScan database, which contains data only from employer-sponsored plans. We included only women with employer-sponsored health insurance plans who indicated that they were covered by health insurance from their job (or the job of their husband, partner, or parents), using respondent answers to PRAMS core question 2, which asks, “During the month before you got pregnant with your new baby, were you covered by any of these health insurance plans?” The age distribution of the survey respondents we considered closely resembled the age distribution of the study population in the MarketScan database, which we used to estimate the employer cost burden of unintended pregnancies.
We conducted 2 distinct analyses. First, we developed regression models from survey data that assigned the probability that each pregnancy was unintended. We used stepwise multinomial logistic regressions to select the variables from the PRAMS data that were significant at the 0.1 level for explaining pregnancy intention. The resulting models retroactively assigned the probability that a given delivery was the result of an unintended pregnancy. Then, we applied the regression model to claims data in MarketScan to estimate the unintended pregnancy rate in employer-sponsored health benefits and normalized the coefficients of the model to closely replicate the unintended pregnancy rate in the PRAMS data. Table 1 shows the unintended pregnancy rates by the mother’s age.
The explanatory variables in the PRAMS database were selected from survey questions that could be identified using information in typical claims and exposure data. This approach was taken so that model equations developed using PRAMS data could be applied to the MarketScan data. In subsequent steps, we further excluded explanatory variables based on odds ratios and clinical reasonability. The variables used in the regression were the mother’s age; prescription drug contraception use; existence of a previous live birth; number of dependents in the family; marital status at the time of conception; medical conditions (ie, hypertension, anemia, heart problems, seizures, thyroid problems, and mental health problems); the use of progesterone, Gestiva (trade name has since changed; generic hydroxyprogesterone caproate), or a 17-alpha hydroxyprogesterone shot; a previous miscarriage, fetal death, or stillbirth; and the state’s unintended pregnancy rate (low, medium, high).
The logistic regression models were applied to inpatient deliveries that were identified in the 2011 MarketScan database. For deliveries in 2011 that might have begun as pregnancies in 2010, we looked back to 2010 claims, as needed, to obtain the data (eg, existing medical conditions) required for the regression models. The probability that the pregnancy was unintended was determined by the results of the regression formulas.
We applied the results of the regression models to deliveries from the MarketScan database to develop the average cost of unintended pregnancies to employers. By applying the adjusted regression coefficients, each live delivery was assigned a probability of being unintended or intended. The average delivery costs weighted by pregnancy intention were developed by the cost component, which was done separately for unintended and intended pregnancies.
The average claim costs per delivery were developed for the following categories:
• Caesarean section delivery
• Vaginal delivery
• Newborn without complications
• Newborn with complications
• Delivery (professional)
• Inpatient visits
• All other delivery costs (between the maternity admission and the discharge dates).
Because some payer systems show costs separately for the baby and for the mother, we included all claims for babies that occurred within 30 days of the delivery date. To arrive at the average cost by pregnancy intention, the costs for these categories were weighted by the probability that the pregnancy was unintended.
We estimated that among employer-sponsored health insurance plans in the United States, 28.8% of pregnancies are unintended, and these pregnancies account for 27.4% of the employers’ maternity delivery costs. The model’s estimate of the portion of deliveries associated with unintended pregnancies is very close to that observed in the PRAMS database.
The probability that a pregnancy is unintended is generally much higher in younger women than in older women, as shown in Table 1. In women aged 15 to 19 years, the proportion of unintended pregnancies was 78%, which is almost 4 times higher than in women aged 35 to 39 years. In addition, we observed an increase in the unintended pregnancy rate in women aged 40 to 44 years and in women aged 45 to 49 years relative to that in women aged 35 to 39 years. This increase could potentially reflect pregnancies in women who might have already had the number of children they wanted and who did not intend to become pregnant at these relatively older ages.
Table 2 shows the average allowed costs per delivery in 2011, which are summarized by major categories of care for unintended and intended pregnancies. The allowed costs include amounts paid by the health plan and by the patient (eg, deductibles and coinsurance).
In per-member per-month (PMPM) terms, unintended pregnancies for a typical employer-sponsored insured population cost approximately $5 of a total monthly medical allowed amount of $359, including prescription drug coverage, based on our analysis of data in the MarketScan database. Table 3 compares the annual frequency of deliveries per 1000 women of childbearing age resulting from unintended and intended pregnancies, presents the monthly allowed costs of these deliveries (including patient cost-sharing), and contrasts these costs to the total monthly allowed medical cost for a typical population covered by employer-sponsored insurance.
Table 4 represents the secondary objective of this study, using a regression model to identify the factors and associated magnitude that contribute to unintended pregnancies in the employee benefits population.
The most recent article on the cost to employers and commercial insurers of unintended pregnancy was published in 2002 and was based on maternal delivery dates that occurred between October 1, 1995, and March 31, 1996.2 This 2002 survey-based study by Green and colleagues reported that the unintended pregnancy rate among commercially insured lower-risk women was 29%.2 One purpose of our analysis was to update this outdated information for employers and commercial insurers, and to provide a tool for health plans to determine the extent of unintended pregnancies in their population.
Based on our analysis of the MarketScan database, we estimated that 28.8% of deliveries in employer-sponsored health insurance plans in 2011 were a result of unintended pregnancies, and that the direct cost associated with these deliveries, which was mostly borne by employers, was approximately $5 PMPM before cost-sharing (or approximately 1% of the typical employer’s annual spending on health benefits, including patient cost-sharing, as shown in Table 3). The $5 PMPM cost, in the authors’ experience, is comparable with the cost of other conditions that employers attempt to manage.8 These findings suggest an opportunity for better health management, and we argue that the target population for appropriate intervention can be easily identified through methods that we developed and presented in this article.
Much of the literature pertaining to unintended pregnancy involves the impact of contraception. In particular, the available estimates of the cost-effectiveness of public spending on contraception and family planning indicate that there is significant opportunity to reduce the taxpayers’ cost of maternal and newborn care that is associated with unintended pregnancy.9,10
Numerous studies describe how access to contraception varies as a result of cost or convenience.11,12 The American College of Obstetricians and Gynecologists recommends making oral contraceptives available over the counter to increase contraception access and use, and to possibly reduce unintended pregnancy rates.13 Cost barriers, such as out-of-pocket expenses, have also been identified as major deterrents to the use of contraception.14,15 More effective long-acting reversible contraception tends to be associated with higher upfront patient cost-sharing,11,12 but it is more cost-effective than short-acting reversible methods.16
The Affordable Care Act (ACA), which requires that commercial health plans cover certain women’s health medical services without cost-sharing, has ameliorated concerns over out-of-pocket expenses for contraception services.14 The federal Health Resources and Services Administration (an organization that is responsible for improving access to healthcare)17 used recommendations from the Institute of Medicine (an independent nonprofit organization that provides unbiased and authoritative advice to decision makers and the public) to establish the provision that women with reproductive capacity will have access to all US Food and Drug Administration–approved contraceptive methods, sterilization procedures, and patient education and counseling.14,17 This provision applies to all nongrandfathered plans starting with the plan year that began on or after August 1, 2012. Plans eligible to be grandfathered have been in existence since March 23, 2010 (and, in some cases, even earlier), and have not made significant benefit reductions or beneficiary cost increases since that time.14,18 Exceptions also exist for nonprofit religious organizations under certain circumstances.14
Because of the time lapse between the provision of health services (including contraception) and the availability of claims data, it was too early at the time of this study to tell whether the ACA’s no cost-sharing contraception mandate has had any effect on the use of contraception services or on the rate of unintended pregnancy. We believe this topic merits follow-up research.
Lower cost-sharing is associated with increased use of services,15 including preventive services. Despite this effect, there are well-known examples of the suboptimal use of other preventive services (eg, cancer screening).19 It is likely that contraceptive preventive services are similarly underutilized or misused,20 suggesting the need to further promote awareness of the availability of contraceptives via communication efforts by providers and health insurers.21
The financial case for employers to promote contraception as a preventive service has not been made, although the prevention of unintended pregnancies is a public health goal and benefit. Our findings suggest that employer promotion of contraception has the potential to reduce unintended pregnancies, as well as costs to employer-sponsored insurance. We argue that family planning fits well into broader wellness and health promotion efforts that are aimed at improving overall employee well-being and health cost-savings.
The high cost of unintended pregnancy for employers suggests the need for research that identifies the most effective patient management methods.22 Typical processes used for managing patients who are at risk for medical conditions include identifying at-risk patients; stratifying at-risk patients so that certain patients receive more intense or less intense outreach efforts; and offering patient education, reminders, or other services aimed at the patient’s clinical situation.
We acknowledge several limitations in our analysis, which include potential survey bias, our inability to capture socioeconomic status in the MarketScan database, potential mismatches between PRAMS survey-reported conditions and diagnosis codes in claims, and the exclusion of terminated pregnancies.
In the PRAMS database, the survey responses pertaining to pregnancy intention are self-reported; intention is inherently subjective, and our results may have been affected by respondent bias. Nonetheless, the rates of unintended pregnancy in the PRAMS database, overall and specific by demographic characteristics, are consistent with those noted by other reputable sources, such as the Guttmacher Institute and the Centers for Disease Control and Prevention.23,24
None of the known major risk factors associated with unintended pregnancy (eg, minority race, fewer years of education, and lower income)25 is available in typical commercial claims data. We therefore attempted to capture some of the differences in socioeconomic factors by adding a regional variable to the regression analysis based on the mother’s state of residence. We ranked states according to their unintended pregnancy rate for mothers with employer-sponsored health insurance, as defined earlier. We established 3 categories to reflect the aggregate unintended pregnancy rates, and then assigned each state to 1 of 3 categories (low, medium, or high), with an equal number of states in each category. Although our goal was to reflect race or income regional differences, we did not expect this simple approach to capture the full range of socioeconomic influences.
Although the MarketScan database is representative of the national aggregate employee and dependent population that is typically covered by employer-sponsored health insurance with regard to size, geography, and health benefits, the conclusions of this study may vary from the experience of any given employer-sponsored health plan.
Finally, the way in which we identified the medical conditions and events used in our model differed by database. In the PRAMS database, medical conditions and events are self-reported, whereas in the MarketScan database, they are based on diagnoses and procedure codes that appear in the administrative claims data. We also recognized the potential for varying degrees of accuracy or underreporting or overreporting in the 2 databases. For example, in the PRAMS database, a woman may have self-reported elevated blood pressure, whereas in the MarketScan database, a woman with hypertension was identified based on the diagnosis codes in her medical claims.
Pregnancies that ended in abortion or miscarriage are not included in the PRAMS database, because this database contains only data on live birth deliveries. We were therefore able to estimate the probabilities and cost burdens of live births only.
In our analysis, we used the 2009-2011 PRAMS database and the 2010-2011 MarketScan database to demonstrate that a significant proportion of pregnancies recently paid for by employer-sponsored health insurance programs are unintended. We also developed a regression model that can be applied to readily available medical claims databases to retrospectively quantify the cost and aggregate risk for unintended pregnancies, in particular, employer-sponsored health insurance benefit populations.
The cost-effectiveness of contraceptive methods has been widely demonstrated elsewhere.9,10 Population health management efforts for other conditions are popular with employers. The ability to identify women who are at high risk for an unintended pregnancy, as demonstrated in this analysis, is one important step toward promoting proved methods to address unintended pregnancies in the employee benefits population.
When this study was conducted, it was too early to assess whether the ACA’s no cost-sharing contraception mandate has had any effect on the use of contraception services or on the rate of unintended pregnancy. We believe this topic merits follow-up research.
This work was supported in part by the Eunice Kennedy Shriver National Institute of Child Health and Human Development grant for Infrastructure for Population Research at Princeton University, Grant R24HD047879.
This study was funded by Bayer Pharmaceuticals.
Author Disclosure Statement
Ms Dieguez and Mr Pyenson received actuarial consulting fees for this research from Bayer Pharmaceuticals; Dr Law is an employee of Bayer Pharmaceuticals; Dr Lynen is an employee and stockholder of Bayer HealthCare; Dr Trussell reported no conflicts of interest.
1. Wier LM, Andrews RM. The national hospital bill: the most expensive conditions by payer, 2008. Statistical brief #107. March 2011. In: Healthcare Cost and Utilization Project (HCUP) Statistical Briefs. Rockville, MD: Agency for Healthcare Research and Quality (US). www.ncbi.nlm.nih.gov/books/NBK53976/. Accessed November 3, 2014.
2. Green DC, Gazmararian JA, Mahoney LD, Davis NA. Unintended pregnancy in a commercially insured population. Matern Child Health J. 2002;6:181-187.
3. Sonfield A, Kost K, Gold RB, Finer LB. The public costs of births resulting from unintended pregnancies: national and state-level estimates. Perspect Sex Reprod Health. 2011;43:94-102.
4. Monea E, Thomas A. Unintended pregnancy and taxpayer spending. Perspect Sex Reprod Health. 2011;43:88-93.
5. Centers for Disease Control and Prevention. What is PRAMS? Updated May 14, 2014. www.cdc.gov/prams/index.htm. Accessed November 3, 2014.
6. Truven Health Analytics. MarketScan databases and tools: better understand health economics and treatment outcomes. http://truvenhealth.com/