During 1995 to 2009 1
254 Danish women aged 15-49 were identified, corresponding to 17
718 women years of observation. The study period from January 2001 to December 2009 included 1
130 women and 9
925 observation years. Among these women 455
421 (31.7%) had never used hormonal contraception and 980
709 (68.3%) were ever users of some kind of hormonal contraception.
After exclusions and censoring owing to pregnancy (n=403
972 or 486
037 women years); ovarian stimulation (n=74
823 or 460
454 women years); previous cardiovascular disease including venous thromboembolism (n=31
252 or 135
828 women years); cancer (n=21
080 or 135
828 women years); coagulation disturbances (n=5122 or 19
258 women years); hysterectomy, bilateral oophorectomy, or sterilisation (n=146
019 or 760
449 women years); censoring after three years of using a hormone releasing intrauterine device (n=48
875 or 164
270 women years); and one month exclusions at switch of oral contraceptive use (n=252
968 or 32
598 women years), 1
120 women were included in the statistical analysis, contributing 8
290 women years of observation, with 4307 first time venous thromboembolic events recorded.
The venous thromboembolic events were distributed, with 82 (1.9%) women having cerebral venous thrombosis, 2738 (63.6%) deep venous thrombosis only, 1130 (26.2%) pulmonary embolism (with or without deep venous thrombosis), 55 (1.3%) portal thrombosis, 15 (0.4%) cava thrombosis, 4 (0.1%) thrombosis of a kidney vein, and 283 (6.6%) unspecified deep vein thrombosis.
Of the 4307 venous thromboembolic events, 61 occurred in women using hormonal contraceptives with so little exposure time and so few venous thromboembolic events that we did not calculate estimates.
The adjusted relative risk increased 6.8-fold from the youngest to the oldest women, and by 41% over the study period (5.1% per year), and was reduced by 51% with increasing length of education (table 1).
Table 1 Characteristics of current users and non-users of combined oral contraceptives and adjusted relative risk of venous thromboembolism according to age, calendar year, and length of education
Relative risk according to progestogen type and oestrogen dose
Table 2 shows the absolute and relative risks of venous thromboembolism in current users of combined oral contraceptives with different types of progestogens and varying doses of oestrogen. The incidence rate of venous thromboembolism in non-users of combined oral contraceptives was 3.7 per 10
000 women years. Compared with non-users, the relative risk of venous thromboembolism in current users of oral contraceptives with levonorgestrel and 30 μg ethinylestradiol was 2.19 (95% confidence interval 1.74 to 2.75) and with levonorgestrel phasic 30-40 μg ethinylestradiol was 2.28 (1.85 to 2.83). The relative risk of venous thromboembolism in current users of oral contraceptives with 30 μg ethinylestradiol combined with desogestrel was 4.21 (3.63 to 4.87), with gestodene was 4.23 (3.87 to 4.63), and with drospirenone was 4.47 (3.91 to 5.11). The corresponding estimates for oral contraceptives with the same progestogens but 20 μg ethinylestradiol were 3.26 (2.88 to 3.69), 3.50 (3.09 to 3.97), and 4.84 (3.19 to 7.33). Progestogen only products conferred no increased risk of venous thromboembolism, whether taken as low dose norethisterone pills, as desogestrel only pills, or in the form of hormone releasing intrauterine devices.
Table 2 Exposure time, number of events of venous thromboembolism, crude incidence per 10 000 user years, and adjusted relative risk of venous thromboembolism in current users of different oral contraceptives and hormone releasing intrauterine device (more ...)
The relative risk of venous thromboembolism from using oral contraceptives with norethisterone, levonorgestrel, desogestrel, or gestodene decreased with decreasing oestrogen dose, whereas no difference was apparent between oral contraceptives with drospirenone and either 30 μg ethinylestradiol or 20 μg ethinylestradiol. Oral contraceptives containing drospirenone and 20 μg ethinylestradiol were launched in Denmark in 2006.
Relative risk by validity of diagnosis
The venous thromboembolic events were stratified into confirmed (anticoagulation therapy recorded in the national registry of medicinal products) and unconfirmed (table 3). Of the 4246 events diagnosed among non-users of hormonal contraception or among users of products included in this study, 2847 (67.1%) were confirmed and 1399 (32.9%) had no or less than four weeks’ anticoagulation therapy recorded in the registry. The relative risks of venous thromboembolism were generally twofold to threefold higher in the confirmed group than the unconfirmed group. Thus in the confirmed group the relative risk of venous thrombolism with use of oral contraceptives with levonorgestrel increased to around 3, and for oral contraceptives with desogestrel, gestodene, drospirenone, or cyproterone and 30 μg ethinylestradiol increased to at least 6.
Table 3 Relative risk of venous thromboembolism among current users of oral contraceptives and hormone releasing intrauterine device according to certainty of diagnosis of venous thromboembolism, with non-users of hormonal contraception as reference (more ...)
Progestogen only products had relative risk estimates below unity compared with non-users in both the confirmed and the unconfirmed groups.
The rate ratio between the estimates in the confirmed and unconfirmed groups was highest for oral contraceptives with desogestrel and lowest for those with norethisterone (table 3).
The proportion of confirmed events for specific oral contraceptives varied from 64% to 84%, and ranged from 72% to 78% for those with levonorgestrel, norgestimate, gestodene, and drospirenone and from 76% to 84% for those with desogestrel.
Table 4 shows the rate ratio estimates between different product types. In the confirmed group, oral contraceptives with desogestrel, gestodene, or drospirenone conferred at least twice the risk of venous thromboembolism compared with oral contraceptives with levonorgestrel, and the rate ratio between oral contraceptives with drospirenone and those with desogestrel or gestodene was 1.01 (0.86 to 1.18). The corresponding rate ratios in the unconfirmed group were generally lower. The comparison between oral contraceptives with drospirenone and those with levonorgestrel was thus 1.78 (1.21 to 2.60), or 16% lower than the 2.12 (1.68 to 2.66) in the confirmed group. The rate ratio between these two product groups for all venous thromboembolic events was 2.00 (1.64 to 2.43), not far off the estimate in the confirmed group.
Table 4 Rate ratios of venous thromboembolism between users of combined oral contraceptives with different progestogens according to certainty of diagnosis of venous thromboembolism
Relative risk adjusted for differences in length of use
To account for differences in the distribution of lengths of use between the groups, analyses were done in which the rate ratios with oral contraceptives containing levonorgestrel and 30 µg ethinylestradiol as reference were adjusted for differences in length of use and restricted to confirmed events (table 5). The rate ratio estimates were slightly reduced for the newest products, reflecting a relatively higher proportion of short term users in these groups. The overall results, however, were unchanged, and the rate ratio between oral contraceptives with drospirenone compared with those containing levonorgestrel was still 2.09 (1.55 to 2.82). Table 6 displays detailed results according to length of use and specific combinations of progestogen types and oestrogen dose.
Table 5 Rate ratio of confirmed venous thromboembolism between different combined oral contraceptives with adjustment for length of use
Table 6 Relative risk of venous thromboembolism in current users of combined oral contraceptives according to length of use and with non-users of hormonal contraception as reference
Relative risk through different steps in exposure line formation
In preliminary analyses, the influence of different steps in the exposure line formation was investigated. In the raw exposure lines no gap filling or prolongation of exposure was realised. The adjusted rate ratio between oral contraceptives with drospirenone and 30 μg ethinylestradiol and those with levonorgestrel and 30-40 μg ethinylestradiol was 2.2 (1.7 to 2.8), and between oral contraceptives with 30 μg ethinylestradiol and drospirenone versus oral contraceptives with 30 µg ethinylestradiol and desogestrel or gestodene the rate ratio was 1.1 (0.9 to 1.3).
In the gap corrected dataset these rate ratio estimates were unchanged, as they were in the dataset for switch corrected exposure lines. For this reason the analyses were done with all allocation rules applied (see web appendix 2).
Relative risk in different sub-periods
Another exploratory step in the analysis was to assess rate ratio estimates in three sub-periods. A non-significant tendency was for lower rate ratios for oral contraceptives with drospirenone compared with those containing levonorgestrel in the last period, but for the period 2001-9 the adjusted rate ratio between oral contraceptives with drospirenone and 30 μg ethinylestradiol compared with those containing levonorgestrel and 30-40 μg ethinylestradiol was 2.00 (1.64 to 2.43), and for the sub-period 2001-5 was 2.16 (1.65 to 2.83). Similar results were found when oral contraceptives with other progestogens were compared with those containing levonorgestrel (see web extra appendix 3). Consequently, subsequent analyses were done for the whole period 2001-9.
Results for different exposure categories
Sensitivity analyses were also done according to different user categories, including successively first starters only, then starters and new users, then including restarters, and finally including switchers. Starters had slightly higher rate ratios between users of oral contraceptives with drospirenone compared with those containing levonorgestrel of 2.69 (1.76 to 4.10) than estimates including the other categories, where the same rate ratios were between 1.96 (1.57 to 2.44) and 2.05 (1.56 to 2.70). See web extra appendix 4 for details.
Different reference groups
A third methodological issue was the oestrogen component in the levonorgestrel products used as reference. The rate ratio of venous thromboembolism between users of oral contraceptives with levonorgestrel and 30 μg ethinylestradiol and with levonorgestrel and 30-40 μg ethinylestradiol including phasic products did not differ significantly in any of the sub-periods. About half of women years using oral contraceptives with levonorgestrel contained 30 μg ethinylestradiol, the other half phasic products 30-40 μg ethinylestradiol. For the period 2001-9, the rate ratio between oral contraceptives with drospirenone and 30 μg ethinylestradiol and all levonorgestrel products with 30-40 μg ethinylestradiol was 2.00 (1.64 to 2.43) and with only levonorgestrel and 30 μg ethinylestradiol was 2.04 (1.58 to 2.63). Accordingly, all users of oral contraceptives with levonorgestrel and 30 μg or 30-40 μg ethinylestradiol were chosen as reference group. For rate ratio comparisons with specifically drospirenone, however, estimates with both 30 μg ethinylestradiol and all levonorgestrel users were calculated.
Among women with confirmed venous thromboembolism, 33 (1.2%) had major surgery in the four weeks before the admission for venous thromboembolism. The results were similar with and without exclusion of women with recent surgery. Thus the rate ratio between oral contraceptives with drospirenone and 30 µg ethinylestradiol compared with those containing levonorgestrel was 2.18 (1.62 to 2.94) with these events included and 2.13 (1.58 to 2.87) without.
Chart evaluation of venous thromboembolism events
Of 200 evaluated hospital charts, 148 (74%) venous thromboembolic events were confirmed and 52 unconfirmed. Except for two women with distal limb thrombosis who were not offered anticoagulation therapy, the remaining 146 confirmed events were in women who had received anticoagulation therapy. However, two unconfirmed events were in women who had received anticoagulation therapy; one for a recent venous thromboembolism, which was not excluded because it was coded at the primary admission (before actual admission) with a superficial venous thrombosis diagnosis and therefore not excluded as previous venous thromboembolism. The other woman was treated for connective tissue disease. All 200 evaluated patients coded as having venous thromboembolism had clinical symptoms at admission.
Of the 200 validated events, 148 (74.0%) women had received anticoagulation therapy according to the medical charts. Of these, 133 (89.9%) were recorded in the national registry of medicinal products as having had anticoagulation therapy, suggesting that about 10% received treatment for free from the hospitals, and therefore were not recorded in the registry.
Among the 52 women without information on anticoagulation therapy in the medical charts, four (7.7%) were recorded in the registry as having received anticoagulation therapy. This can occur when treatment starts after discharge from the department to which the women were primarily admitted—that is, initiated from a coagulation laboratory just after discharge from the department. If these four events were added to the confirmed events in the sample of 200 women, the confirmed proportion increased to 152 of 200, or 76.0%.