Ethinylestradiol 0.03 mg/drospirenone 3 mg was approved for use by the U.S. Food and Drug Administration on May 11, 2001, as the oral contraceptive Yasmin (Berlex Laboratories, Montville, NJ). Its progestin component (drospirenone) is unique among oral contraceptives, possessing antimineralocorticoid activity with potassium-sparing diuretic effects similar to those of spironolactone. The drospirenone content in the commercially available contraceptive is equivalent in its potassium-sparing activity to a low dose (25 mg) of spironolactone.1,2
After marketing of ethinylestradiol/drospirenone, case reports of thromboembolism associated with its use appeared in the medical literature (Tiffany. Pulmonary embolism due to Yasmin [letter]. BMJ rapid response to van Grootheest. BMJ.com. July 12, 2003; Tartarone A, Gallucci G, Cammarota A, Romano G, Ardito R, Di Renzo N. Venous thromboembolism after high-dose chemotherapy in a patient with Hodgkin’s lymphoma receiving the new oral contraceptive ethinylestradiol and drospirenone [“Yasmine”] [letter]. Bone Marrow Transplant 2005;35:103),3 and a prescription event monitoring study suggested an elevated risk of thromboembolism in women taking Yasmin, with an incidence rate of 13.7 cases per 10,000 person-years.4 Physicians in Holland were warned against prescribing ethinylestradiol/drospirenone largely on the basis of case reports until additional data on the risk of thromboembolism relative to other oral contraceptives was available.5 Thromboembolism is a well-recognized complication of oral contraceptive use,6 and the newness of ethinylestradiol/drospirenone relative to other contraceptives rather than a differential effect on thromboembolism might have led to the case reports.
As a condition of approval, the manufacturer agreed to a postmarketing study of ethinylestradiol/drospirenone to evaluate its use and association with hyperkalemic complications possibly related to its potassium-sparing diuretic effects.7 This study of thromboembolism was conducted within the cohorts of ethinylestradiol/drospirenone and other oral contraceptive initiators used for hyperkalemia surveillance.
MATERIALS AND METHODS
We followed ethinylestradiol/drospirenone initiators and medically similar initiators of other oral contraceptives, identified within a proprietary research database built from electronically captured provider, facility, and pharmacy claims at UnitedHealthcare-affiliated health plans and large employer groups, with follow-up in the written medical records. The database complies with privacy guidelines specified by the Health Insurance Portability and Accountability Act. Data collection covered events occurring among ethinylestradiol/drospirenone users and the comparison cohort, as accrued through the first 36 months of ethinylestradiol/drospirenone marketing.
Figure 1 outlines the creation of the study population. Women aged 10–59 years who received a dispensing of ethinylestradiol/drospirenone after at least 6 months in the health plan without having had a previous dispensing for ethinylestradiol/drospirenone (ethinylestradiol/drospirenone initiators) between June 2001 and June 2004 were identified on a quarterly basis along with women who by the same criteria initiated another oral contraceptive in the same quarter. If a woman initiated more than one oral contraceptive within the same calendar quarter (a small fraction), we counted only the first, with the exception of ethinylestradiol/drospirenone initiation, which was counted regardless of the order in which it occurred. These ethinylestradiol/drospirenone and other oral contraceptive initiators were separated into groups according to date of initiation, with each group representing initiators within one calendar quarter. Using claims for medical services across the 6 months preceding initiation, we developed variables representing numerous aspects of their medical history, and we developed a prediction model (propensity score)8–11 of ethinylestradiol/drospirenone initiation. We matched a twofold larger group of other oral contraceptive initiators to the ethinylestradiol/drospirenone group within a fixed caliper (0.01) of the propensity score. This matching was conducted separately within each calendar quarter of initiation and had the effect of creating a close balance between the groups with respect to calendar time and all the medical history, treatment, and personal characteristics that went into the propensity model.12
Follow-up for events and accrued person-time for each woman began with entry into the cohort and continued until the earliest of end of enrollment in the health plan, end of follow-up on June 30, 2004, or 180 days after the last use of the initiating oral contraceptive.
To identify thromboembolic events, we searched the insurance claims data of the cohorts during follow-up for a broadly inclusive list of diagnoses, procedures, and drugs possibly related to thromboembolism. Medical records were sought for all patients whose insurance claims sequence was possibly consistent with the occurrence of thromboembolism. Trained abstractors recorded pertinent information from the patient medical record using a standardized abstraction instrument created for this study. Of the medical records sought for 125 potential thromboembolic events, we obtained 109 (90.1%), with similar fractions in the ethinylestradiol/drospirenone group and the other oral contraceptive group. All abstraction forms were formally reviewed by a clinician blinded to oral contraceptive use status for determination of whether a thromboembolic event had occurred and to ascertain the date of onset of the condition. All use of data complied with Health Insurance Portability and Accountability Act requirements, and the study was carried out under approval from the New England Institutional Review Board and Privacy Board.
Propensity scores were estimated using logistic regression, with the outcome being initiation of ethinylestradiol/drospirenone or another oral contraceptive and predictors being derived from a woman’s history of claims preceding initiation. Variables were selected for the propensity score model through a combination of a priori and empiric considerations. Some of the variables included in the propensity score were specified in the study protocol based on expected association with ethinylestradiol/drospirenone, and some were included on an empirical basis as characteristics that were found to differ between ethinylestradiol/drospirenone and other oral contraceptive initiators. Analysis for thromboembolism was based on both as-matched (similar to intent-to-treat) and as-treated (similar to time-on-drug) paradigms. For the as-matched analysis, all women were assigned to the initially matched cohort (ethinylestradiol/drospirenone or other oral contraceptive), regardless of subsequent changes in contraceptive use. Proportional hazards models were used to provide estimates of the relative incidence of thromboembolism in the compared groups. In the as-treated analysis, we used pharmacy dispensing claims to continuously reclassify oral contraceptive exposure during follow-up. Current use included the day of dispensing and continued through the period that corresponded to the number of days of contraception supplied in 28-day intervals. With each new dispensing, a patient continued or reinitiated her current-use status. Recent use was classified from the last date of current use through the following 60 days. Past use included all days from the conclusion of recent therapy through 180 days after end of current use. We used multivariable Poisson regression analysis of incidence, incorporating age, calendar time, health plan, history of oral contraceptive use, health services consumption, and chronic medical conditions identified in the baseline 6 months as adjustment variables.13 Because matching in this study was performed on the basis of exposure rather than outcome (as is the case with matched case-control studies), the independent association of variables that were part of the propensity score could be assessed.13
In addition to the proportional hazards, and Poisson regression models, for each study outcome we calculated incidence rates, rate ratios, and associated 95% confidence intervals. Calculations were performed in SAS 8.2 (SAS Institute Inc, Cary, NC) and Stata 6.0 (Stata Corporation, College Station, TX).
The propensity models were moderately predictive of ethinylestradiol/drospirenone initiation, and it was possible to match nearly all ethinylestradiol/drospirenone initiators (more than 95%) to two initiators of a comparison oral contraceptive. The final cohorts included 22,429 ethinylestradiol/drospirenone initiators matched to 44,858 other oral contraceptive initiators. The average follow-up was 7.8 months (14,081 woman-years) among ethinylestradiol/drospirenone initiators and 7.5 months (27,575 woman-years) among the comparators. Tables 1 and 2 illustrate the similarity of the groups with respect to selected baseline characteristics after matching. The characteristics shown in Tables 1 and 2 were chosen for their prevalence or pertinence to the outcome. There were more than 100 other characteristics that were included in propensity score estimation, and none was meaningfully different between the study groups.
In the as-matched analysis, there were 18 confirmed instances of thromboembolism among the ethinylestradiol/drospirenone initiators, for an incidence rate of 1.3 per 1,000 woman-years, and 39 among other oral contraceptive initiators, for an incidence rate of 1.4 per 1,000 woman years (relative risk 0.9, 95% confidence interval [CI] 0.5–1.6). Thromboembolism subgroups, deep vein thrombosis, and pulmonary embolism also occurred with equal frequency between ethinylestradiol/drospirenone and other oral contraceptive initiators (Table 3). The absolute incidence rates we observed could be translated into clinical practice as follows: a clinician can expect to observe one case of thromboembolism among 769 women over the course of 1 year if they were prescribed ethinylestradiol/drospirenone oral contraceptives and one case of thromboembolism among 714 women over the course of 1 year if they were prescribed another oral contraceptive. With this difference in incidences, to observe one additional case of thromboembolism, 9,286 women would have to be prescribed oral contraceptives (number needed to harm).
Most person-time and outcome events occurred during current exposure to the initially matched study drug, with 14 thromboembolic events among 9,587 woman-years of current ethinylestradiol/drospirenone exposure and 30 thromboembolic events among 19,899 woman-years of other oral contraceptive current exposure, for a rate ratio of 1.0, 95% CI 0.5–1.9 (Table 4).14 The Poisson regression analysis identified age 40 years and above and aggregated preexisting chronic medical conditions (diabetes, hypertension, history of myocardial infarction, and arrhythmia), as well as prior thromboembolism, as having elevated incidence rates of thromboembolism during follow-up, but suggested no difference in incidence between users of ethinylestradiol/drospirenone and the other oral contraceptives. A separate analysis of incidence rates and rate ratios stratified by duration of follow-up yielded essentially the same results, indicating that there was not a different association at different stages of follow-up (no exposure by time interaction).
There was no difference between ethinylestradiol/drospirenone initiators and initiators of other oral contraceptives in risk of thromboembolism, and this finding did not vary with accounting for changes in therapy during follow-up (as-treated analysis), when we calculated rate ratios of events by duration of follow-up rather than person-time of exposure, or when we eliminated small numbers of high-risk women from the cohorts. Prior thromboembolism, other chronic medical conditions, and age were all associated with elevated risks for thromboembolism, indicating that the study, which did not indicate any difference between ethinylestradiol/drospirenone and other oral contraceptives, was sufficiently sensitive to pick up known medical risk factors for thromboembolism.
The matching of the cohorts (ethinylestradiol/drospirenone and other oral contraceptive initiators) was based on propensity scores estimated from claims data in the 6-month baseline period leading up to the date of cohort entry. The propensity score matching achieved close balance on the patient characteristics and patterns of health care use that were available in the automated claims data, but could not guarantee balance between the matched cohorts with respect to factors that are not part of the claims database, such as smoking and body mass index. We conducted a validation study, not reported here, in which we sought data on further variables (especially smoking and body mass index) that may confound the relationship between ethinylestradiol/drospirenone use and risk of thromboembolism by examining the medical records of a random sample of study subjects (Eng PM, Seeger JD, Loughlin J, Clifford CR, Mentor S, Walker AM. An internal validation study with case-cohort analysis to address unmeasured confounding in a large claims-based cohort study of oral contraceptive initiation and risk of venous thromboembolism [unpublished study]). There were no systematic differences in these characteristics derived from medical records between ethinylestradiol/drospirenone initiators and the matched initiators of other oral contraceptives, so the estimate of the relative effect of ethinylestradiol/drospirenone and other oral contraceptives on thromboembolism did not change, even accounting for smoking or body mass index.
Recent studies evaluating the effect of ethinylestradiol/drospirenone on hemostatic parameters have found almost no different effect relative to other oral contraceptives. In one of these studies, ethinylestradiol/drospirenone was compared to ethinylestradiol/desogestrel in 75 healthy volunteers.15 After six cycles of treatment, changes in fibrinogen, factor VII, thrombin-antithrombin complex, antithrombin, protein C, protein S, tissue-type plasminogen activator, plasminogen activator inhibitor, prothrombin, partial thromboplastin time, and D-Dimer were indistinguishable between the compared groups. Another study of 60 women randomized to receive ethinylestradiol/drospirenone or ethinylestradiol/desogestrel found no difference in the changes in 10 markers of hemostatic potential through seven cycles of follow-up.16 Both groups of women exhibited procoagulatory changes in the hemostatic markers, but there was no differential effect. It is unlikely that a difference in thromboembolism between ethinylestradiol/drospirenone and other oral contraceptives would manifest itself clinically when there was no difference between the drospirenone-containing oral contraceptive and the nondrospirenone oral contraceptive with respect to any of these more sensitive markers of hemostasis. A different mechanism for an effect of drospirenone on hemostasis than any of these known markers of hemostasis would have to be hypothesized, and the fact that our study, which followed women in actual practice, did not find any difference between ethinylestradiol/drospirenone and other oral contraceptives with respect to thromboembolism indicates that the lack of differential effect on the known markers of hemostasis translates into lack of effect on clinical thromboembolism.
Although our findings do not directly address questions about the effect of ethinylestradiol/drospirenone in older age women (especially as drospirenone is used as part of hormone replacement therapy) or where ethinylestradiol/drospirenone is used for extended therapy (120 days), the consistency of our finding of a lack of difference in clinical venous thromboembolism (VTE) between ethinylestradiol/drospirenone users and other oral contraceptive users, with a similar lack of differences in sensitive markers of thromboembolism or even of hyperkalemia, an outcome that might be expected to have a greater differential effect,17 indicates that our findings likely extend beyond the population directly observed, because the underlying biologic mechanism for VTE development is unlikely to differ in older women or with extended treatment durations. Thus, even though our study cohorts were derived from an insured population, these results likely extend to an uninsured population because it is probable that the underlying biologic mechanism for the null association we observed between ethinylestradiol/drospirenone and VTE also operates in an uninsured population.
Strengths of this study include the integrated database of health services claims within which all billable health care interactions are recorded and the fact that clinical VTE outcomes were observed and compared, rather than surrogate measures of VTE, such as markers of hemostasis. Within the research database used for this study, capture of study outcomes is certain because any clinically meaningful thromboembolism will result in billable health care services. However the sensitivity offered by the health care claims database for an outcome like thromboembolism is tempered by a potential for falsely identifying work-ups for thromboembolism as actual thromboembolic events. Our requirement that a case of thromboembolism be confirmed through review of medical records before incorporation in our study meant that the study outcome was clinically meaningful and served to protect against false-positive identification, and our nearly complete access to underlying medical records suggests that the increase in specificity obtained through medical record review did not come at the expense of sensitivity. In this study of young, relatively healthy oral contraceptive initiators, we did not find any evidence of increased thromboembolism risk among ethinylestradiol/drospirenone initiators compared with other oral contraceptive initiators.
1. Heinemann LA, Dinger J. Safety of a new oral contraceptive containing drospirenone. Drug Saf 2004;27:1001–18.
2. Krattenmacher R. Drospirenone: pharmacology and pharmacokinetics of a unique progestogen. Contraception 2000;62:29–38.
3. van Grootheest K, Vrieling T. Thromboembolism associated with the new contraceptive Yasmin. BMJ 2003;326:257.
4. Pearce HM, Layton D, Wilton LV, Shakir S. Deep vein thrombosis and pulmonary embolism reported in the Prescription Event Monitoring Study of Yasmin. Br J Clin Pharmacol 2005;60:98–102.
5. Sheldon T. Dutch GPs warned against new contraceptive pill. BMJ 2002;324:869.
6. Blann AD, Lip GY. Venous thromboembolism. BMJ 2006;332:215–9.
7. U.S. Food and Drug Administration. Postmarketing study commitments. Available at: http://www.fda.gov/cder/pmc/default.htm
. Retrieved June 29, 2007.
8. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41–55.
9. Joffe MM, Rosenbaum PR. Invited commentary: propensity scores. Am J Epidemiol 1999;150:327–33.
10. Rubin DB. Estimating causal effects from large data sets using propensity scores. Ann Intern Med 1997;127:757–63.
11. Rubin DB, Thomas N. Matching using estimated propensity scores: relating theory to practice. Biometrics 1996;52:249–64.
12. Seeger JD, Williams PL, Walker AM. An application of propensity score matching using claims data. Pharmacoepidemiol Drug Saf 2005;14:465–76.
13. Rothman KJ, Greenland S, editors. Modern epidemiology. 2nd ed. Philadelphia (PA): Lippincott-Raven; 1998.
14. Daly L. Simple SAS macros for the calculation of exact binomial and Poisson confidence limits. Comput Biol Med 1992;22:351–61.
15. Klipping C, Marr J. Effects of two combined oral contraceptives containing ethinyl estradiol 20 μg combined with either drospirenone or desogestrel on lipids, hemostatic parameters and carbohydrate metabolism. Contraception 2005;71:409–16.
16. Kluft C, Endrikat J, Mulder SM, Gerlinger C, Heithecker R. A prospective study on the effects on hemostasis of two oral contraceptives containing drospirenone in combination with either 30 or 20 microg ethinyl estradiol and a reference containing desogestrel and 30 microg ethinyl estradiol. Contraception 2006;73:336–43.
17. Preston RA, White WB, Pitt B, Bakris G, Norris PM, Hanes V. Effects of drospirenone/17-β estradiol on blood pressure and potassium balance in hypertensive postmenopausal women. Am J Hypertens 2005;18:797–804.