Association of Progestogens and Venous Thromboembolism Among Women of Reproductive Age.

The IBM® MarketScan® Commercial Claims and Encounters Databases contain routinely collected, de-identified data for individuals and their dependents with employer-based insurance and some non-group commercial health plans in the U.S. 26 Databases include enrollment records, inpatient and outpatient adjudicated medical claims, and pharmacy claims for filled prescriptions. Case-defining VTE diagnoses (DVT, PE, or both) were identified by International Classification of Diseases, Ninth Revision (ICD-9) or Tenth Revision (ICD-10) diagnosis codes from inpatient and outpatient claims ( Appendix 1 , available online at http://links.lww.com/AOG/C813 ). Cases were identified as women of reproductive age, defined as 15–49 years old, who were newly diagnosed with VTE from January 1, 2010 (when diagnosis codes first differentiated acute and chronic VTE) through October 8, 2018 (the most recently available data). 9 , 11 , 27 , 28 The index date for cases was defined as the inpatient date of admission or outpatient date of service. Inpatient diagnoses adjudicated at the time of discharge have been shown to be more accurate than outpatient diagnoses. 29 Outpatient diagnoses required corroboration with a new anticoagulation prescription within 42 days before or after the index date. 7 , 30 , 31 Finally, continuous enrollment in medical and pharmacy benefit coverage was required from one year before the index date through 84 days after the index date, to ascertain conditions for exclusion. Exclusions were applied to create a sample representative of cases with incident acute VTE who could have been exposed to progestogens at typical doses for contraception or treatment of benign menstrual disorders (e.g., endometriosis, abnormal uterine bleeding). We excluded cases with non-qualifying VTE diagnoses (e.g., chronic, pregnancy-related) or with any earlier anticoagulation prescription that could indicate recurrent VTE. Superficial venous thrombosis was evaluated as a covariate rather than an exclusion criterion. 2 , 19 We then excluded women diagnosed with conditions that can be treated with ultra-high dose progestogens (e.g., breast cancer, ductal carcinoma in situ, endometrial cancer, endometrial intraepithelial neoplasia or hyperplasia, and cachexia or acquired immune deficiency syndrome) or who had a megestrol acetate prescription within one year before the index date. Prior studies have already demonstrated increased odds of VTE and mortality in these settings. 16 , 32 , 33 Cases who were determined to be pregnant or postpartum based on claims within 42–84 days before or after the index date were then excluded to allow for an adequate washout period and for new progestogen exposure ( Appendix 1 , available online at http://links.lww.com/AOG/C813 ). Similarly, we excluded cases with exposure within 84 days before the index date to a medication containing estrogen. Controls were initially individually matched 50:1 to cases by risk set sampling with replacement based on year of birth and index date. The index date for controls was set to the index date of the matched case so that the surveillance periods overlapped exactly within each stratum. Additional confounders were incorporated as covariates instead of matching criteria as recommended by “Strengthening the Reporting of Observational Studies in Epidemiology” (STROBE). 34 Propensity score matching may have reduced bias from differences in measurable VTE risk factors between cases and controls. However, it would have prevented use of a single study population for multiple progestogen exposures and reduced statistical efficiency without eliminating residual confounding. Controls could not have any VTE diagnosis or anticoagulation prescriptions in the available records prior to the index date. This initial larger pool ensured adequate eligible controls remained after exclusions that were time dependent. After applying enrollment and exclusion criteria, excess controls for each case were randomly removed to reach a 5:1 ratio. Cases with inadequate matched controls were excluded from the sample. Exposures to four oral progestogens (NET, Prog, MPA, and NETA) were defined by filled outpatient prescriptions as proxies for actual use as prescribed. Exposures to three non-oral progestogens (LNG-IUD, Implant, and DMPA) were defined by pharmacy and medical claims ( Appendix 1 , available online at http://links.lww.com/AOG/C813 ). Claims data did not allow for direct ascertainment of indication for progestogen use or for identification of all contraindications for alternatives. Some progestogens were more likely used for contraception only (NET and Implant), treatment only (Prog, MPA, and NETA), or both (LNG-IUD and DMPA). Definitions of exposure characteristics ( Table 1 ) were generated for additional analyses of recency, dose, and chronicity. Categories reflected common prescribing practices, assumed that progestogen use nearest the index date was most important, and aligned with prior studies. 14 , 28 , 30 , 35 Users with more than one current or recent progestogen exposure were classified as having mixed use. The primary progestogen for analysis was selected as that with the highest systemic dose using a dose hierarchy by Bergendal et al. modified to incorporate non-contraceptive progestogens (from lowest to highest: LNG-IUD, NET, Implant, Prog, MPA, NETA, DMPA). 13 , 36 , 37 The hierarchy did not account for potential pharmacological differences in oral and non-oral administration or for dosing patterns (e.g., cyclic, continuous). Current and recent users of more than two progestogens were excluded from analyses as outliers. Covariates were identified using diagnosis and procedure codes from inpatient and outpatient claims during the one year prior to the index date ( Appendix 1 , available online at http://links.lww.com/AOG/C813 ). We included 16 VTE risk factors as pre-specified confounders based on prior literature: obesity (body mass index ≥30 mg/kg 2 ), history of smoking, atherosclerotic cardiovascular disease, hypertension, congestive heart failure, chronic kidney disease, rheumatoid arthritis, systemic lupus erythematous, inflammatory bowel disease, cancer (excluding non-melanoma skin cancer, and breast and uterine cancers based on study exclusion criteria), hemiplegia or paraplegia, hypercoagulable state, superficial venous thrombosis, varicose veins, history of surgery or lower extremity fracture within 180 days before the index date, and history of hospitalization within 180 days before the index date. 1 – 3 , 9 , 11 , 27 , 38 Descriptive statistics were calculated for case and control characteristics. Qualifying VTE diagnoses, pregnancy exclusions, and progestogen use were assessed by year to evaluate secular trends and similar ascertainment between ICD-9 and ICD-10 codes. For each of the seven progestogens of interest, the primary outcome was the odds of incident acute VTE comparing current users with non-current users of any progestogen ( Table 1 ). 28 , 30 , 35 Crude odds ratios were calculated by conditional logistic regression, adjusted for year of birth and index date by study design. The final multivariable model for each progestogen additionally adjusted for all 16 VTE risk factors previously described. Other studies have variably adjusted for VTE risk factors even when individual covariates fail to significantly modify the measure of association. 8 , 9 , 14 , 16 , 28 To aid in clinical application of the findings, estimates of the absolute risks of VTE were calculated as excess VTEs per 10,000 exposed women at both ends of the reproductive age range for all associations with an adjusted odds ratio of at least 2.0. 1 , 39 , 40 Secondary analyses evaluated the effect of varying dose and chronicity within exposure to individual progestogens ( Table 1 ), adjusted for the same covariates as the primary analysis. For each oral progestogen, weighted average daily dose was assessed as a categorical and continuous variable in conditional logistic regression models. The categorical model determined odds of VTE for low, standard, and high dose users compared with non-current users of any progestogen ( Table 1 ). The continuous model determined the change in odds of VTE for increasing weighted average daily dose. Units were set at half of the most commonly prescribed tablet of that progestogen per day. Users with more than one recent progestogen exposure and prescriptions with missing dose information or with quantities less than one were excluded from this analysis. Chronicity was assessed for all progestogens by operationalizing exposure as a four-level categorical variable combining current use and prior use variables ( Table 1 ) in conditional logistic regression models. For each progestogen, comparisons were made between new, chronic, former, and non-users. Finally, we planned two sensitivity analyses. To assess for misclassification bias from inclusion of false positive VTE diagnoses, the primary analysis was repeated after excluding inpatient cases without a corroborating anticoagulation prescription. The primary analysis was also repeated with an extended exposure time window comparing recent progestogen use to non-recent use to assess for misclassification from delayed use and to allow for comparisons to prior literature. 14 For all models, significance was set at P <.01 to allow for multiple comparisons in the primary analysis. Adjusted odds ratios were calculated with 99% confidence intervals. We considered secondary analyses to be hypothesis-generating. A pre-specified sample size calculation was performed with 90% power to detect an association with an odds ratio of at least 2.0 at a frequency of current use of each progestogen of at least 0.1% among controls. Reliable estimates of frequencies for non-contraceptive progestogens were not available. The entire available sample was used since it was less than the target of 24,252 cases. All analyses were performed using SAS version 9.4. This study using de-identified data was exempted from review by the University of Chicago Biological Sciences Division Institutional Review Board.

Out of 50,692 individuals with a qualifying VTE diagnosis and adequate continuous enrollment, 21,405 cases matched 1:5 by year of birth and index date to 107,025 controls remained after exclusions ( Figure 1 ). Characteristics of all qualifying acute VTE diagnoses by type and source are provided in Appendix 2 , available online at http://links.lww.com/AOG/C813 . We did not see abrupt changes in ascertainment of VTE or pregnancy diagnoses at the transition from ICD-9 to ICD-10 codes ( Appendix 3 , available online at http://links.lww.com/AOG/C813 ). Characteristics of cases and controls are provided in Table 2 . Sixty percent of the study population were aged 40–49 years old at the index date. Cases and controls were significantly different in all other covariates with 66.3% of cases having one or more VTE risk factors compared with 25.4% of controls ( P <.001). Frequency of current use of any progestogen was 4.1% and 2.4% for cases and controls, respectively ( P <.001), when combining single and mixed use. While current use of LNG-IUD, NET, and Implant increased during the study period, current use of other progestogens remained stable ( Appendix 3 , available online at http://links.lww.com/AOG/C813 ). Odds of VTE associated with current use of each progestogen compared with non-current use of any progestogen are provided in Figure 2 . Use of the higher dose progestogens demonstrated significantly increased adjusted odds of VTE (MPA, adjusted OR, 1.98 [99% CI,1.41–2.80]; NETA, adjusted OR, 3.00 [99% CI, 1.96–4.59]; DMPA, adjusted OR, 2.37 [99% CI, 1.95–2.88]), whereas low dose NET use was associated with significantly reduced adjusted odds of VTE (adjusted OR, 0.57 [99% CI, 0.43–0.76]). Use of other progestogens was not significantly different from non-use (LNG-IUD, adjusted OR, 0.72 [99% CI, 0.50–1.05]; Implant, adjusted OR, 1.09 [99% CI, 0.66–1.82]; Prog, adjusted OR, 1.07 [99% CI, 0.66–1.73]). All 16 VTE risk factors included in the multivariable models were independently associated with VTE. The conditional logistic regression model for NETA is provided as a representative example ( Appendix 2 , available online at http://links.lww.com/AOG/C813 ). Odds ratios for covariates included for adjustment differed by less than 10% between all models. In estimates of absolute risk (99% CI) for women aged 20–24 years old, current use of NETA was associated with 7 (3–13) excess VTEs per 10,000 exposed women, and current use of DMPA was associated with 5 (3–7) excess VTEs per 10,000 exposed women. For women aged 45–49 years old with a higher baseline VTE incidence, current use of NETA was associated with 12 (6–22) excess VTEs per 10,000 exposed women, and current use of DMPA was associated with 8 (6–11) excess VTEs per 10,000 exposed women. As defined in Table 1 , most oral progestogen users were exposed to low or standard weighted average daily dose (median, interquartile range): 97% of NET users (0.34 mg/day, 0.28–0.35), 83% of Prog users (94.1 mg/day, 53.3–168.6), 95% of MPA users (3.45 mg/day, 1.71–7.63), and 86% of NETA users (4.68 mg/day, 2.51–6.33). In models where progestogen exposure was represented by weighted average daily dose as a continuous variable ( Appendix 2 , available online at http://links.lww.com/AOG/C813 ), increasing NETA and MPA doses were directly proportional to adjusted odds of VTE: +22% (adjusted OR, 1.22 [99% CI, 1.05–1.41]) for each 2.5 mg/day increase in NETA and +19% (adjusted OR, 1.19 [99% CI, 1.05–1.34]) for each 5 mg/day increase in MPA. Increasing NET dose was inversely proportional to adjusted odds of VTE. We did not find an association between increasing Prog dose and VTE. Models of weighted average daily dose as a categorical variable ( Table 3 ) were limited by small sample sizes but reflected the results of the continuous model. Comparisons of chronicity by a 4-level categorical variable broadly reflected the associations in the primary analysis ( Appendix 2 , available online at http://links.lww.com/AOG/C813 ). In the sensitivity analysis excluding potentially false positive VTE diagnoses, 4,374 (35.6%) of 12,276 inpatient cases did not have a corroborating new anticoagulation prescription ( Appendix 2 , available online at http://links.lww.com/AOG/C813 ). Only the adjusted odds ratios for current use of NETA and DMPA were changed by more than 10% and both increased (adjusted OR, 3.34 [99% CI, 2.08–5.35], and adjusted OR, 2.66 [99% CI, 2.15–3.31], respectively). In the sensitivity analysis using the extended exposure window of recent use ( Appendix 2 , available online at http://links.lww.com/AOG/C813 ), all adjusted odds ratios decreased, but only the estimate for NETA decreased by more than 10% (adjusted OR, 2.65 [99% CI, 1.82–3.87]).

We found that the association of current use of progestogens and VTE was dependent on the type and dose of progestogen. Use of higher dose progestogens (MPA, NETA, and DMPA) was significantly associated with increased odds of VTE compared with non-use, whereas use of lower dose progestogens (LNG-IUD, NET, Implant, and Prog) was not significantly different from non-use or had reduced odds of VTE. The importance of progestogen type and dose is supported by the associations for LNG-IUD and DMPA at the extremes of systemic dose because both can be used for contraception and to treat menstrual disorders. 19 , 20 This study provides the first adequate assessments for use of the contraceptive Implant and non-contraceptive progestogens (Prog, MPA, and NETA). To do so, it used a larger sample size than the sum of all relevant studies identified in recent systematic reivews. 4 , 6 Compared with those studies, we performed the first exploration of varying progestogen dose and chronicity. 4 , 6 Modeling oral progestogen exposure by weighted average daily dose for NETA and MPA suggested a dose-response relationship with VTE, though the categorical dose model was limited by wide confidence intervals. In the chronicity model, we expected new users to demonstrate a stronger association with VTE than chronic users due to depletion of susceptibles, but these groups were only significant different for NET. 41 The effect size was as expected for NETA but much smaller for DMPA and MPA. We also expected associations with VTE would not persist after progestogen discontinuation. Former users and non-users were not significantly different for NETA and MPA, but they were significantly different for DMPA (possibly due to its long duration) and NET. Future research could be powered to evaluate these associations and incorporate dosing pattern. Progestogens as a class have not been associated with acute prothrombotic changes, but effects may vary by type. 33 , 42 – 45 NET and NETA are uniquely metabolized to ethinyl estradiol at 0.2–1% of the orally administered dose. 46 – 48 Approved dosing for NETA at 5–15 mg/day approximates at least 10–30 μg/day of oral ethinyl estradiol, typical contraceptive doses. Medroxyprogesterone acetate (MPA and DMPA) upregulates thrombin receptor expression through its potent glucocorticoid activity. 49 Clinical studies have found dose-dependent increases in VTE risk associated with glucocorticoid use. 50 , 51 Progesterone (Prog) and etonogestrel (Implant) also potentiate thrombin-induced procoagulant activity but have much lower affinity and activity at the glucocorticoid receptor. 37 , 52 , 53 Additional research is needed on progestogen-mediated prothrombotic effects at relevant doses and with novel types (e.g., dienogest). This study has several limitations. First, 35.6% of inpatient VTE diagnoses did not have a corroborating anticoagulation prescription, similar to other studies. 9 , 11 , 28 , 29 The sensitivity analysis confirmed that such misclassification bias shifted the associations towards unity, supporting the primary findings. Second, exposure to oral progestogens was determined by filled prescriptions without confirmation of actual use. Third, though using a nationwide sample with inpatient and outpatient cases improved generalizability, associations in this study may not apply to individuals without private insurance or outside of the U.S. Fourth, the data source did not contain objective clinical data (e.g., weight, family history), and under-ascertainment of covariates cannot be differentiated from absence of a condition. However, obesity and smoking status have not demonstrated large confounding effects in the association of hormone use and VTE. 9 , 14 , 28 Fifth, these data were not collected for research purposes and were subject to misclassification, such as through conversion of validated ICD-9 codes to ICD-10 codes. Finally, these associations may have been confounded by indication due to contraindications for alternatives (i.e., estrogen) or conditions being treated. 4 , 19 Because progestogens are used by women with VTE risk factors, the associations for MPA, NETA, and DMPA may remain inflated after adjustment for measured confounders. Future studies on individual progestogens could construct more balanced populations using propensity score matching. This study design also did not permit direct comparisons of VTE risk with users of combined hormonal contraception because of differential bias in those populations (e.g., selection bias, healthy user bias). Our findings for LNG-IUD, NET, and DMPA aligned closely with prior studies that controlled for indications for use and other confounders not available in the data source, adding to the external validity of the study. 7 , 11 , 14 , 15 , 54 We also assumed that benign conditions treated by progestogens were not causally linked to VTE. Menstrual disorders were not found to have confounding effects in one of the largest studies on hormone use and VTE in women of reproductive age. 9 Acknowledging that residual confounding cannot be excluded, our study design benefited from a large sample size, a well-suited data source, rigorous classification, and sensitivity analyses of bias. While a dose-response relationship between progestogens and VTE remains uncertain, we recommend that clinicians consider progestogen type and dose to balance benefits, risks, and alternatives, which differ for contraceptive and non-contraceptive indications.

Venous thromboembolism (VTE), defined as deep venous thrombosis, pulmonary embolism, or both, is associated with severe morbidity and mortality and is increasing in the U.S. 1 For women with significant VTE risk factors in which estrogen use is contraindicated, progestogen-only medications (progestogens) have been recommended as alternative forms of contraception and treatments for benign menstrual disorders. 2 , 3 However, the assumption that progestogens are safer may not hold for all types and doses. Twenty-two studies have assessed the association between progestogen use and VTE, but most were judged to be of poor to fair quality and limited by small sample sizes. 4 – 12 In their systematic review, Tepper et al. described a potential dose-response relationship based on a hierarchy by Bergendal et al. that reflected systemic effects (e.g., ovulation inhibition). 4 , 13 The available evidence suggests that use of the very low dose levonorgestrel intrauterine device (LNG-IUD), low dose norethindrone (NET), or the moderate dose etonogestrel subcutaneous contraceptive implant (Implant) does not increase VTE risk. However, a recent meta-analysis found that use of the high dose injectable contraceptive depot medroxyprogesterone acetate (DMPA) was associated with a 2.6-fold increase in odds of VTE. 7 , 12 , 14 , 15 Non-contraceptive progestogens (e.g., oral norethindrone acetate [NETA], oral medroxyprogesterone acetate [MPA], and oral progesterone [Prog]), have also been associated with a 2.4- to 5.9-fold increase in odds of VTE in four small studies. 10 , 16 – 18 In contrast to clinical guidelines that have questioned studies on DMPA use, some authors have urged caution when prescribing higher dose progestogens to individuals at increased VTE risk. 2 , 4 , 10 , 16 , 19 – 25 This nested matched case-control study aimed to evaluate associations between current use of seven progestogens and incident acute VTE, compared with non-use of any progestogen.