Quantifying the economic burden of unintended pregnancies due to drug–drug interactions with hormonal contraceptives from the United States payer perspective

Key points for policy makers

Introduction

Unintended pregnancies are a major economic burden, costing the United States (US) government USD21 billion in 2010¹, and mainly result from contraceptive non-use or incorrect or inconsistent use of effective contraceptives². Several economic evaluations have established the cost-effectiveness of contraceptive use from a payer and societal perspective^3–5. Long-acting reversible contraceptives (LARC), which include copper intrauterine device (IUD), levonorgestrel intrauterine system (IUS), and etonorgestrel implants have been found to be the most cost-effective options across various time-horizons and geographies^3,6–16. Although uptake of LARCs in the USA has been increasing, adoption has been low compared with oral contraceptives (OC), with 12% of reproductive age women using IUDs and 3% using implants compared to 25.3% using OCs in 2014¹⁷. With over 9.5 million women in the USA and 151 million women worldwide currently using OCs, factors leading to contraceptive failure and consequent unintended pregnancies among women constitute an important public health issue^17–19. Non-adherence to OCs arising from forgetfulness, not filling a prescription, or experiencing side effects has been found to affect contraceptive efficacy^20,21. The extent to which incorrect or inconsistent use of a contraceptive results in pregnancies, is quantified in a term known as the typical-use failure rate, while pregnancy rates during perfect use of a contraceptive are known as perfect-use failure rates, the former being a metric of contraceptive effectiveness in the real world, and the latter quantifying efficacy in a trial or controlled setting². Perfect-use or typical-use failure rates are quantitatively represented as the percentage of women experiencing an unintended pregnancy within the first year of use²². Considering full adherence to OCs, another reason resulting in unintended pregnancies may be due to contraceptive failure from drug–drug interactions (DDIs).

Pharmacokinetic DDIs result with the co-administration of certain perpetrator drugs which may alter the systemic exposures of the OCs. Combined OCs consist of an estrogen component (e.g. ethinyl estradiol) and a progestin component (e.g. levonorgestrel, desogestrel, drospirenone) which are metabolized by cytochrome P450 enzymes in the liver to varying extents, depending upon their composition. Consequently, certain enzyme inducers may decrease the effectiveness of OCs and potentially result in unintended pregnancies, while certain enzyme inhibitors may increase plasma hormone concentrations, thereby increasing the incidence of adverse events such as thromboembolism²³. Drugs which interact with OCs and meant for long-term use include rifamycin antibiotics²⁴, antiepileptics²⁵, antiretrovirals²⁶ and psychotropic drugs²⁷, of which a few specific drugs have been classified by the CDC as Category 3, i.e., the risks outweigh the advantages of using the method²⁸. Unintended pregnancies that occur as a consequence of DDIs are important, given that they occur in women who are otherwise compliant to their contraceptive method but who may be unaware of the impact of concomitant drugs on the effectiveness of their contraception.

Given the lack of data on the clinical effect of DDIs, the US Food and Drug Administration (FDA) uses class labelling for product labels of hormonal contraceptives based on information known about estrogens and progestins. To address the dearth of epidemiological data regarding the real-world impact of DDIs on unintended pregnancies a recent analysis of a commercial claims database by Sarayani et al. estimated the contraceptive failure rates among users of concomitant enzyme-inducer and enzyme-neutral antiepileptic drugs²⁹. The paucity of data regarding the clinical impact of DDIs on OCs was recognized by the FDA which resulted in a public meeting in 2015 which brought together regulators, clinicians and representatives from the industry and academia to seek solutions³⁰. The FDA recently released a guidance document for industry on clinical drug interaction studies for drugs under development having interacting potential with OCs and recommends alternative forms of contraception for women at risk of DDIs available at https://www.fda.gov/media/134581/download and https://www.fda.gov/media/110050/download. However, women may fail to receive sufficient counseling by their physician to make a switch to alternative contraception options, which puts them at risk of the consequences of DDIs such as unintended pregnancies or adverse events^31,32. Currently, none of the economic evaluations address the economic impact of DDIs on hormonal contraceptives. The consequences of DDIs which may accrue substantial costs for the payer have not been addressed in economic evaluations. Therefore, the current study is a model-based analysis that aims to estimate the economic burden of unintended pregnancies due to DDIs with OCs in the USA from a payer perspective.

Methods

Model design

The model assumes a cycle length of one month and a time horizon of one year, since even when women are trying to conceive, the monthly fecundity rate is only 20%³³ (Age and Fertility. A Guide for Patients. American Society for Reproductive Medicine. 2012. https://www.reproductivefacts.org/globalassets/rf/news-and-publications/bookletsfact-sheets/english-fact-sheets-and-info-booklets/Age_and_Fertility.pdf) Our model comprises of five states. All women begin in the “initial method” (non-pregnant) state, consisting of either “OC alone” or “OC plus an enzyme inducer,” or “OC plus enzyme neutral drug” and continue in this state until experiencing contraceptive failure resulting in unintended pregnancy. Consequently, women may further transition to “birth”, “induced abortion”, “spontaneous abortion” or an “ectopic pregnancy”, which are assumed to be absorbing states. “Unintended pregnancy” by itself is not considered as a true state in the model and its outcomes have transition probabilities equal to the product of probability of OC failure and probability of unintended pregnancy resulting in the outcome (e.g., probability of having a birth while on the pill is the product of pFailure and pBirth) (Figure 1). After experiencing one of these terminal outcomes, a woman is assumed to discontinue the “initial method” and does not enter the model again.

Figure 1. Model schematic.

pFailure: transition probability of method failure; pBirth: transition probability of birth; pInducedAbortion: transition probability of induced abortion; pSpontaneousAbortion: transition probability of spontaneous abortion; pEctopicPregnancy: transition probability of ectopic pregnancy.

Results

Scenario analysis

Three scenario analyses were conducted to test the robustness of the model results and impact on pregnancy outcomes (Figure 2) and cost (Figure 3).

Figure 2. Scenario analysis: Comparison of key health outcomes for each scenario.

Scenario 1. Increasing oral contraceptive failure rate; Scenario 2. Increased or Decreased Abortion Rate; Scenario 3. Age-specific patterns of pregnancy outcomes.

Figure 3. Scenario analysis: Comparison of key cost outcomes for each scenario.

Scenario 1. Increasing oral contraceptive failure rate; Scenario 2. Increased or Decreased Abortion Rate; Scenario 3. Age-specific patterns of pregnancy outcomes.

• Scenario 1: The model is highly sensitive to assumptions about the failure rates. Considering the typical-use failure rate resulted in 3.8-fold larger numbers of unintended pregnancies in the OC + enzyme-inducing drugs scenario as compared with the base-case scenario as shown in Figure 2a.

• Scenario 2: The model is sensitive to assumptions of abortion rates as shown in Figure 2b. The model predicted almost a 2-fold increase in number of unintended births when comparing the lowest and highest abortion rates.

• Scenario 3. The model was minimally sensitive to age-specific pregnancy outcomes as shown in Figure 2c.

Discussion

The current analysis aimed to estimate the cost of UP attributed to DDIs with OCs for a cohort of 1,000 women considering various scenarios of contraceptive failure rates, namely perfect use with OC alone and in the presence of enzyme-inducer and enzyme-neutral drugs, estimates of which were obtained from a published real-world analysis. Given the large number of women who rely on OCs for their contraceptive needs in the USA and worldwide, even small incidences of these unintended pregnancies due to DDIs may result in substantial cost burden from a payer perspective. Therefore, interventions that seek to reduce care fragmentation and improve care coordination between primary care physicians and specialists will improve patient awareness and result in lower incidence of unintended pregnancy and adverse events because of DDIs. While clinical decision support systems embedded in electronic health record and pharmacy management software are designed to alert the physician or pharmacist of the potential of DDIs, these warnings are often overridden with rates as high as 90% due to “alert fatigue”^57,58. In a recent US study of 255 pharmacies in the Chicago area, investigators found that about 52% of them dispensed medications without warnings about potential drug interactions, attributed to the emphasis on speed of filling prescriptions over quality among several retail pharmacies⁵⁹. Several approaches to reduce alert fatigue, such as classification of DDI’s based on priority are currently being evaluated to avert potential patient safety issues related to clinically significant drug interactions⁶⁰. Therefore, despite the presence of systems designed to identify and flag DDIs, several of these alerts can proceed unnoticed, potentially resulting in dangerous consequences to patients. In this context, we believe our study highlights the economic burden of one such important drug interaction and emphasizes on the need to provide additional safeguards in the drug prescribing and dispensing process. Dispensing OCs, whether from a retail pharmacy, online or through Planned Parenthood health centers should carefully consider the medication history, and offer sufficient counseling to women to ensure they are using contraception that is appropriate for their individual needs.

This is the first model-based analysis estimating the costs of DDIs with hormonal contraceptives. There have been only a few studies describing health services utilization costs associated with DDIs with antiretroviral therapy and opioids using real world data^61–66. We considered a range of contraceptive failure rates, since the extent of enzyme induction and hence loss of efficacy of the OC can be dose dependent and drug specific, e.g., strong, moderate and weak CYP3A inducers decrease exposure of CYP3A substrates to different extents⁶⁷. We conducted the current analysis from a US payer perspective, the wide variation in the cost of the UP outcomes is reflective of costs accrued by public and private payers which were obtained from the published literature. The OC failure rates with enzyme-inducing and enzyme-neutral medications were calculated, assuming that drug exposure commenced at the prescription dispensing date and concluded on the last day of the pharmacy-entered dispensed days’ supply²⁹. However, this assumes women are fully adherent to their OC and likely underestimates failure rates for women who skip pills or discontinue use of their OC before the end of their pill supply. To account for this, we conducted a sensitivity analysis where we increased the OC-alone failure rate from the perfect (0.3%) to typical use (7.2%) failure rate and adjusted the concomitant enzyme-inducing and enzyme-neutral OC failure rates upward by a factor of 6.9% (7.2%–0.3%).

For each strategy within the total cohort, we assumed a constant failure rate which may not reflect real world experience due to heterogeneity in prescription medication and variation in DDI potential. The costs accrued due to adverse events with OCs which might be exacerbated by DDIs was not considered, as the extent to which adverse-event prevalence is altered in women taking interacting drugs is currently not well understood. Several costs were not considered in the current analysis, thus resulting in a likely underestimation of the true costs of DDIs. The direct medical costs of downstream events (e.g., teratogenicity), events related to the reduced efficacy of the perpetrator drug, additional indirect costs due to work productivity losses, costs to social welfare programs and intangible costs were beyond the scope of the current analysis. The model also did not consider additional costs accrued due to OC discontinuation and switching. The impact of inter-individual variability on the clinical effect of DDIs by several pharmaceutical factors (dose, timing of drug administration, frequency, and duration of the specific interacting medication(s)) and non-pharmaceutical factors (timing of sexual intercourse with respect to menstrual cycle, use of back-up contraceptives) were not explored in the current analysis (https://www.fda.gov/media/134581/download).

Conclusions

The current study quantifies the consequences of a complex clinical pharmacology problem of DDIs with OCs which has large public health implications from an economic point-of-view. This question is currently being addressed by partners in academia and industry resulting in the integration of interdisciplinary fields of pharmacokinetic and pharmacodynamic modeling, pharmacoepidemiology and pharmacoeconomics^30,68,69. This cost represents a potentially avoidable and unrecognized economic burden to the payer, thus building a case for developing policy for population health decision making to optimize contraceptive prescribing and usage among women at risk of DDIs. Future iterations of this model can go beyond the current empiric estimation to incorporate variability in the clinical pharmacology in patients and consider additional costs and health states that may be expected due to DDIs.

Data (and software) availability

Consent

Not applicable. Study was based on aggregated data available from the literature.