Low dose oestrogen combined oral contraception and risk of pulmonary embolism, stroke, and myocardial infarction in five million French women: cohort study

Comparison with other studies

Most authors have observed a dose-effect relation with oestrogen, higher doses being associated with an increased risk of venous thromboembolism. Lidegaard and colleagues12 showed that desogestrel or gestodene combined with 20 µg of oestrogen was associated with 23% and 17% lower relative risks of venous thromboembolism compared with the same progestogens combined with 30 µg of oestrogen. Van Hylckama and colleagues7 reported a higher risk of venous thromboembolism for oral contraceptives containing 50 µg of oestrogen compared with 30 µg of oestrogen (odds ratio 1.9, 1.1 to 3.4), and no statistically significant difference for risk with oral contraceptives containing 20 µg of oestrogen (0.8, 0.5 to 1.2). A similar trend has been shown for oral contraceptives with higher doses of oestrogen (50-100 µg).35 36 37

In the present study, after adjustment for the type of progestogen, we observed a statistically significant reduction in the risk of pulmonary embolism by about 25% with the 20 µg dose compared with 30-40 µg dose. A similar tendency was found for risk of venous thromboembolism with desogestrel and gestodene, but not with levonorgestrel.15 Finally, the analysis by Lidegaard and colleagues31 showed that oral contraceptives with 20 µg, 30-40 µg, and 50 µg of oestrogen were associated with relative risks of myocardial infarction of 1.40 (95% confidence interval 1.07 to 1.81), 1.88 (1.66 to 2.13), and 3.73 (2.78 to 5.00), respectively (P<0.001 for trend). We also observed a statistically significantly lower risk for the 20 µg dose compared with 30-40 µg doses (adjusted relative risk 0.56, 95% confidence interval 0.39 to 0.79), in line with the results reported previously.31

Similarly, the relative risk of pulmonary embolism was 2.16 (1.93 to 2.41) for desogestrel and 1.63 (1.34 to 1.97) for gestodene compared with levonorgestrel. These relative risks were similar to those reported for risk of venous thromboembolism by Lidegaard and colleagues (2.2, 1.7 to 2.3 and 2.1, 1.6 to 2.8)12; by van Hylckama and colleagues (2.0, 1.4 to 2.8 and 1.6, 1.0 to 2.4)7; in the meta-analysis by Stegeman and colleagues (1.8, 1.4 to 2.2 and 1.5, 1.2 to 2.0)15; and by Vinogradova and colleagues18 (1.8, 1.5 to 2.1 and 1.5, 1.2 to 1.9).

The effect of age on the events of interest in this study is in line with those reported in the literature, supporting the validity of our reported results. For example, the 3.6-fold increased risk of venous thromboembolism with age (45-49 years v 15-19 years) calculated in by Lidegaard and colleagues8 was close to the risk of pulmonary embolism calculated in our study (4.1-fold). Similarly, the risks of ischaemic stroke and myocardial infarction with age (45-49 years v 15-19 years) were the same in the two studies: 20 and 100, according to Lidegaard and colleagues31 and 21 and 106 in our study.

Strengths and limitations of this study

These results were shown with a high statistical power based on 5.4 million women years of combined oral contraceptive use, allowing us to observe more than 3200 serious events (pulmonary embolism, ischaemic stroke, or myocardial infarction) in this study. However, the number of events was low for some subgroups, particularly the number of myocardial infarctions among women treated with gestodene and levonorgestrel with 20 µg of oestrogen. These results therefore need to be interpreted cautiously, and further studies are necessary to confirm the present results. The follow-up period was also relatively short (maximum of two years and three months), which limits the bias associated with changing diagnostic procedures and health system management of the three events of interest, and eliminated the potential time trend bias. However, certain diseases may take longer to manifest, particularly ischaemic stroke and myocardial infarction. The availability of levonorgestrel with 20 µg of oestrogen on the French market in April 2010, with more than 650 000 women years in our cohort, allowed us to calculate arterial and venous thromboembolic risk under real conditions of use. A possible indication bias was limited by adopting September 2012 as the end date for inclusion in the study, owing to the publication of a practice guideline for doctors in October and November 2012.38

This study was based on two databases (SNIIRAM and PMSI) with completely independent data collection. The search for a possible increased risk of hospital admission among women reimbursed for oral contraceptives was therefore performed by using both health insurance reimbursement data related to pharmacy purchases of oral contraceptives and inhospital diagnoses.

Information on reimbursement of medicinal products is exhaustively collected on a regular basis by entering the specific bar code for each medicinal product reimbursed. Patients are identified by their personal smart card, and daily teletransmission by pharmacists to the French national health insurance network avoided the recall or recording bias often associated with retrospective case-control studies.

Data on hospital stay is systematically recorded in PMSI (French hospital discharge database), as each hospital doctor in France is required to complete this information after a patient is admitted to hospital.39 Doctors specialised in these hospital codes collect, validate, and transmit all structured and coded medical information from the hospital to paying and controlling bodies. PMSI is used to determine the cost of each hospital stay. Retrospective linking between the two databases, based on independent data collection, therefore theoretically excludes the possibility of a differential observation bias on hospital management diagnoses of pulmonary embolism or arterial thromboembolic events according to outpatient prescribed and dispensed use of oral contraceptives. Validation of cases is essential in this type of observational study. A validation study of cases of pulmonary embolism in 25 French hospitals found that the sensitivity of the ICD-10 diagnostic code for pulmonary embolism in the PMSI was 88.9% (95% confidence interval 85.6 to 92.2).40 Our analysis of anticoagulant treatment on discharge from hospital (89% of cases filled at least two prescriptions, with an average of more than five prescriptions, for anticoagulants during the four months after hospital admission) confirmed the validity of this criterion. The authors of a validation study of cases of stroke based on the PMSI compared with the stroke registry reported a sensitivity of 77.1% (95% confidence interval 74.2 to 80) and a positive predictive value of 69.2.41 However, an improvement was observed over time, with a sensitivity of 82.9% and positive predictive value of 81.2% in 2008, the last year of the study. A classic difficulty in observational studies on stroke concerns interpretation of the I64 code (stroke, not specified as haemorrhage or infarction). This code made a limited contribution to the present study, as it represented only 102 of the 1046 ischaemic strokes compared with 944 of I63 codes (with the exception of I63.6).

The initial management of myocardial infarction has a specific code in the French hospital discharge database based on ICD-10 to distinguish initial management from a prevalent case by means of a mandatory additional digit. As this code has a major impact on the cost of hospital stay, the initial management of myocardial infarction seems to be associated with a relatively low risk of error, especially in young participants. We also observed that more than 94% of women in this study received outpatient post-infarction treatment comprising at least two treatment groups from among statins, β blockers, angiotensin converting enzyme inhibitors/sartans, and platelet aggregation inhibitors.

However, our study has several limitations. The first concerns the absence of the start date for use of combined oral contraceptives and the concept of new user or switch from another product not reimbursed by French national health insurance. Indeed, the incidence of venous thromboembolism is increased during the initial phase of oral contraceptives use, particularly during the first three months and the first year.12 An adjustment for the length of oral contraceptives use would be necessary to take this variable into account. Lidegaard and colleagues, who performed this adjustment, observed a slight reduction in the relative risk for new products, but the global results remained unchanged. If this factor had a major impact, it would have primarily concerned levonorgestrel with 20 µg oestrogen, the product most recently released onto the French market, for which the risk of pulmonary embolism would possibly be overestimated, although we observed a lower risk than with the other drug combinations. The frequently reported argument that third generation oral contraceptives such as gestodene, desogestrel that have been released onto the market more recently and therefore correspond to newer users, would lead to bias,42 can no longer be justified more than 25 years after the release of these oral contraceptives in France. The relative stability of the market shares of the various oral contraceptives other than levonorgestrel with 20 µg oestrogen in France between July 2010 and September 2012 is an argument in favour of the validity of our estimates. Another element concerns the analysis in the post-partum setting, in which all women of the cohort initiated use of oral contraceptives after the pregnancy. The results of this subgroup analysis were similar to those of the main analysis for desogestrel compared with levonorgestrel as reference.

In contrast, to our knowledge the duration of use had no impact on the risk of myocardial infarction and ischaemic stroke—the other two events of interest in the study.

Among the limitations related to the study design, we failed to take into account women who may have died from a venous or arterial thromboembolic event before admission to hospital. This limitation could introduce a bias if women experiencing a thromboembolic event while taking a combined oral contraceptive with 20 µg of oestrogen died more rapidly and before admission to hospital compared with women taking one with 30-40 µg of oestrogen, which seems unlikely.

Potential for confounding by indication is a another limitation of the present study, as we did not take into account some factors known to increase the risk of venous or arterial thrombosis and which may therefore influence a doctor’s choice of oral contraceptives, such as body mass index and family predisposition. However, doctors preferentially prescribed levonorgestrel to the most economically underprivileged women and women with diabetes, two factors associated with higher body mass index. In the present study these two factors would underestimate the risk of pulmonary embolism associated with the other progestogens such as desogestrel and gestodene. The same argument could apply to women with a family predisposition, who would be prescribed oral contraceptives associated with a lower risk. Data on smoking, a major risk factor for arterial thrombosis, but with a low impact on the risk of venous thromboembolism,43 were only partially available and limited to situations of medically assisted smoking cessation. Again, the population most concerned by smoking was that using levonorgestrel, an argument in favour of underestimation of the risk associated with the other oral contraceptive combinations. Furthermore, analyses adjusted for smoking in previous published studies did not modify, or only marginally modified, the results.7 31

Finally, another limitation of this study was the failure to include oral contraceptives not reimbursed by French national health insurance, such as drospirenone for which discordant result have been previously reported.7 8 9 10 11 12 18 44 Based on data derived from declarations of annual sales by drug companies to the French National Agency for Medicines and Health Products Safety, we estimated that our study sample comprised about two thirds of all women receiving a third generation oral contraceptive and all women receiving a first or second generation oral contraceptive. Limiting our study to women using oral contraceptives reimbursed by French national health insurance has the advantage of improving the comparability of women within the sample, as it has been largely shown in the literature that the choice of method of contraception is related to women’s socioeconomic status as well as financial imperatives.45

Lastly, only pulmonary embolism was considered as an outcome of deep vein thrombosis, although use of oral contraceptives increases the risk of all forms of deep vein thrombosis.8 Indeed, acute diagnoses can be determined only for hospital related events in the present databases, and the frequency of deep vein thrombosis, of which not all cases involve hospital admission, would have been seriously underestimated.

Conclusions

Women using an oral contraceptive with 20 µg of oestrogen had a lower risk of pulmonary embolism and serious arterial thromboembolic events than women using an oral contraceptive with 30-40 µg oestrogen. We also found that use of an oral contraceptive containing levonorgestrel was associated with a 50% lower risk of pulmonary embolism compared with using an oral contraceptive with a third generation progestogen. The combination of levonorgestrel and 20 µg of oestrogen was the combination associated with the lowest incidence of serious adverse events. Under real conditions of use, among 100 000 women using a combined oral contraceptive for one year an estimated 33 will experience pulmonary embolism, 19 ischaemic stroke, and seven myocardial infarction. For each woman, it is necessary to determine the most appropriate contraceptive method in terms of acceptability and risk. A risk reduction strategy should be based on limitation of use of oral contraceptives associated with the highest risk to decrease the incidence of pulmonary embolism46 and arterial thromboembolic events among young women.