- Email: [email protected]
Oral contraceptives and venous thromboembolic disease: The findings from database studies in the United Kingdom and Germany
Oral contraceptives and venous thromboembolic disease: The findings from database studies in the United Kingdom and Germany R. D. T. Farmer, PhD, FFPHM, and R. A. Lawrenson, MD, FAFPHM Guilford, England OBJECTIVE: Three research articles published in late 1995 and early 1996 suggested that oral contraceptives containing either of the newer progestogens (gestodene or desogestrel) could be associated with an increased risk of venous thromboembolism. During the months after the initial publications, the results have been scrutinized with great care and further studies have been published. The findings of 2 recent database studies, 1 in the United Kingdom and 1 in Germany, are presented in this article. PATTERNS OF USE: The average age of users of combined oral contraceptives in Germany was 27 years, compared with 26 years in the United Kingdom. In Germany the use of gestodene-based products was lower than that in the United Kingdom. In the United Kingdom the users of desogestrel with 20 µg ethinyl estradiol (Mercilon) were older than the users of desogestrel with 30 µg ethinyl estradiol (Marvelon). CRUDE INCIDENCE: The crude incidence of venous thromboembolism in the UK study was 4.1 cases/10,000 woman-y exposure to combined oral contraceptives. In Germany it was 4.2 cases/10,000 woman-y. In Germany the rates among users of second-generation combined oral contraceptives were higher than those among users of third-generation products. The reverse was the case in the United Kingdom. In the United Kingdom the crude incidence rates were higher for the 20 µg estrogen desogestrel product than for the 30 µg product. CASE-CONTROL ANALYSIS: The adjusted odds ratios in the UK study did not show significant increases for desogestrel or gestodene compared with levonorgestrel products. There were inconsistencies in the results among centers in the 2 international studies (the World Health Organization and Transnational studies). In both there was a consistent inverse dose-response relationship with estrogen in all centers. CONCLUSION: The limitations of the observational studies are such that the hypothesis that the newer progestogens are more likely to cause venous thromboembolism cannot be proved. (Am J Obstet Gynecol 1998;179:S78-86.)
Key words: Case-control studies, desogestrel, gestodene, oral contraceptives, venous thromboembolism
Venous thromboembolic disease, manifested either as peripheral thrombosis (eg, deep venous thrombosis in a limb) or as pulmonary embolism, is rare among young women. Its clinical manifestations are ubiquitous. Both deep venous thrombosis and pulmonary embolism can occur without symptoms and resolve spontaneously. Current noninvasive investigative techniques do not always allow deep venous thrombosis or pulmonary embolism to be either diagnosed or excluded with certainty, and fatalities, although rare (between 0.5% and 4%), are often unexpected. The diagnostic problems are highFrom the Department of Public Health and Epidemiology, Imperial College of Science, Technology, and Medicine, University of London. Reprint requests: Professor R. D. T. Farmer, European Institute of Health and Medical Sciences, University of Surrey, Stirling House Campus, Surrey Research Park, Guilford Surrey, GU2 5RF, England. Copyright © 1998 by Mosby, Inc. 0002-9378/98 $5.00 + 0 6/0/92634
S78
lighted by the observation from an autopsy study that although 83% of subjects were found to have a pulmonary embolism, only 19% had been thought to have had a deep venous thrombosis during life.1 The clinical manifestation and investigation of deep venous thrombosis and pulmonary embolism complicate the epidemiologic investigation of the disease. In particular, not all cases come to medical attention, fatality among untreated patients is not inevitable, and some affected persons die without manifesting symptoms during life. A number of thrombophilic disorders carry an increased risk of venous thrombosis; however, the presence of such a disorder does not make thrombosis inevitable, even under adverse circumstances. Deep venous thrombosis (with subsequent pulmonary embolism) has an increased frequency after surgery, trauma, immobilization, cancer, pregnancy, obesity, air travel, and (among women) use of oral contraceptives (OCs).2 The increased risk after
Farmer and Lawrenson S79
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
surgery is maintained for several weeks. In pregnancy the risk is greatest in the few weeks immediately before and after delivery. Other circumstances in which the risk of deep venous thrombosis is increased include occult cancers, some of the autoimmune diseases (eg, systemic lupus erythematosus), varicose veins, inflammatory bowel disease, hypothyroidism, and renal disease. Among women who use OCs the risk of venous thromboembolic disease appears to be related to the dose of estrogen.3-5 Since the early 1970s the dose of estrogen in combined OCs has been reduced, and preparations are now available with as little as 20 µg of estrogen. The reduction in dose of estrogen has been associated with reductions in both the absolute incidence of deep venous thrombosis and the incidence relative to nonusers of OCs.6 Gerstman et al4 investigated the influence of estrogen and progestogen potency on the risk of venous thromboembolism among women between 15 and 44 years old who were members of the Michigan Medicaid scheme. They found a positive association with estrogen potency but no relationship with progestogen potency. The low-potency (low-dose) estrogen OC products yielded an unadjusted incidence rate of venous thromboembolism of 5.3 cases/10,000 woman-y exposure, whereas those classified as high-potency products carried an incidence of 9.8 cases/10,000 woman-y. Despite the reassuring and consistent reductions in incidence of deep venous thrombosis and pulmonary embolism among women who use combined OCs, the World Health Organization (WHO) initiated a further study in the mid 1980s. The WHO investigation was an internationally based case-control study. Among the reasons for the investigation were that “most studies have been done in northern Europe and the USA, and it may not be appropriate to extrapolate the results to populations with different incidence rates of VTE disease and prevalence of associated risk factors” and that “part of the apparent variability in OC-associated risk estimates reported was due to difficulties in diagnosing DVT and PE accurately, particularly when the diagnosis was based mainly on clinical grounds.”7 The study was conducted in 21 centers across 17 countries in different regions of the world. One of the incidental findings was that combined OCs containing desogestrel, gestodene, or norgestimate (which were classified as third-generation OCs) were associated with a higher incidence of venous thromboembolism than were those containing levonorgestrel, norgestrel, or norgestrienone in combination with <35 µg ethinyl estradiol (classified as second-generation OCs).8 Some time after the start of the WHO8 study Spitzer et al9 began a multicenter, Europeanbased case-control study of the cardiovascular effects of OCs. Their study was prompted because the “results of a pharmacokinetic study of 22 women in Germany suggested that gestodene, but not desogestrel, may increase
the risk of vascular events.”9 The Spitzer et al9 study (known as the Transnational study) used a protocol similar to that of the WHO8 study but was specifically designed to compare the cardiovascular risks of combined OCs containing different progestogens. After the early release of the results of the WHO study relating to the comparative risks of OCs containing different progestogens,8 Jick et al10 investigated the risks of cardiovascular disease among women who used different OCs through the UK General Practice Research Data Base, a study referred to as the Boston Collaborative Drug Surveillance Program study. In October 1995 the UK Committee on Safety of Medicines reacted to the unpublished results of these 3 studies by issuing a formal warning to doctors regarding an apparent increased risk of venous thromboembolism associated with third-generation OCs.11 On their publication it was apparent that all 3 pivotal studies (WHO,8 Transnational,9 and BCDSP10 studies) showed the incidence of venous thromboembolic disease among women who used third-generation OCs to be higher than that among women who used second-generation products. Subsequently the interpretation of the results of the 3 studies has been criticized. In particular it has been suggested that they could have been affected by bias and confounding.12-16 In view of the importance of the issues it was decided to initiate further investigations. In this article the results of 2 studies, 1 from the United Kingdom and 1 from Germany, both of which were based on general practice computer-generated clinical databases, are presented. These studies were specifically designed to compare the risks of venous thromboembolism associated with OCs containing different progestogens. The main results of both have been published elsewhere.17,18 Methods UK study. The UK study used a clinical database generated by collating the computer-based records from about 140 general practices in the United Kingdom. The database (which is known as the UK MediPlus Database and is maintained by Intercontinental Medical Statistics London, United Kingdom) is similar in concept and execution to the General Practice Research Database that was used by Jick et al.10 There is no overlap between the populations in the 2 databases. The records used in the analysis cover a population of about 490,000 women between 15 and 44 years old. The period investigated (September 1991-September 1995) involved about 235,000 woman-years exposure to OCs. The information recorded in most of the medical records included the following: date of registration with the practice, year of birth, date of death, height, weight, blood pressure, smoking habits, record of each prescription, diagnosis at consultation, investigations, and hospital referrals. The first part of the study was designed to describe the way in which OCs were prescribed in UK general prac-
S80 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table I. Woman-years of prescriptions and average age of users in the UK and German MediPlus databases for combined OCs according to the major type of progestogen United Kingdom
Progestogen Desogestrel Gestodene Levonorgestrel Others
Germany
Prescription (1000 woman-y)
Proportion (%)
Average age (y)
Prescription (1000 woman-y)
Proportion (%)
Average age (y)
60.5 49.9 63.5
24.9 20.5 26.1
25.7 25.2 27.3
31.0 7.0 35.2 31.8
29.5 6.7 33.5 30.3
27.7 27.8 29.4 28.5
Woman-years of prescription are given in thousands; proportions are percentage of total OC use.
tice. This was done by identifying all prescriptions for OCs, together with the number of cycles prescribed and the age of the woman at the time of the prescription. The age distributions of the recipients of different types of OCs were compared. The cases of venous thromboembolism were then selected to calculate the population incidence rates. Potential case patients with venous thromboembolism were women who had a diagnosis of deep venous thrombosis or pulmonary embolism between September 1991 and September 1995 and had prescriptions for oral anticoagulants. Women who did not have a record of a prescription for an OC that would have covered the day on which she had the venous thromboembolic episode were excluded. Women with a history of venous thromboembolism before September 1991 were excluded and those with “nonfirst episodes” (venous thromboembolic episodes subsequent to a previously recorded episode before September 1991) during the period investigated were also excluded. The records of all the potential case patients were examined to establish whether any of the potential case patients had given birth within 38 weeks of the venous thromboembolic episode; those women who had given birth within this period were excluded as case patients. Women with a record of a recent surgical operation or trauma were also excluded. Patients who died but were not excluded for any of the preceding reasons were included as case patients. The crude incidence rates were calculated for different groups of OC products. Finally, a nested case-control study of venous thromboembolism was conducted. As many as 4 control subjects were matched with case patients with respect to the general practice with which they were registered and by exact year of birth. Women with a record of a past venous thromboembolism, those who were pregnant on the day the episode occurred in the case patient, and those who had a recent history of a surgical operation or trauma were not eligible to be selected as control subjects. All the control subjects were covered by a prescription for an OC on the event day of the case patient with whom they were matched. The data were modeled with conditional logistic regression and by imputing missing variables in
cases with incomplete records of body mass index. The following models were used: 1. In model 1 third-generation OCs are compared with second-generation products. This model is similar to a model used in the Transnational9 study. 2. In model 2 gestodene and desogestrel are compared with levonorgestrel. This model is comparable to that used in the WHO8 and Jick et al10 studies. 3. Model 3 is the same as model 2 except that the 2 products containing desogestrel (desogestrel with 20 µg ethinyl estradiol and desogestrel with 30 µg ethinyl estradiol) were separated. In all models adjustment was made for body mass index (as a continuous variable with imputation of missing values), change in the OC used within 3 months of the event or of the event day of the case patient for control subjects (as a binary variable), number of cycles used before the event, previous pregnancy (as a binary variable), concurrent illness (as a binary variable), and previous use of postcoital contraception (as a binary varable). Blood pressure and smoking were not included in the model because these data were not available for a high proportion of case patients and control subjects. The body mass index (weight in kilograms divided by the height in meters squared) was available for 65.1% of the case patients and 64.7% of the control subjects. For a further 14.1% of case patients and 9.0% of control subjects, categoric information on body mass was available (low, normal, or high). The weight alone was recorded for a further 5.6% of case patients and 7.1% of control subjects. No data were recorded for 20.2% of case patients and 14.1% of control subjects. Body mass is known to be an important risk factor for venous thromboembolism. To maximize the numbers of case patients and control subjects that could be used in the regression model, missing values were imputed from all available data. There is some evidence that recency of use could affect the probability of the occurrence of a venous thromboembolic episode19; the numbers of cycles used before the episode were therefore included in the model. It is known that pregnancy carries a relatively high risk of venous thromboembolic disease. If a woman
Farmer and Lawrenson S81
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Fig 1. Exposure (in 1000 woman-years) to OCs vesus age (in years) for the MediPlus databases from the United Kingdom and Germany, 1991-1995.
Fig 2. Age distribution of desogestrel with 30 µg ethinyl estradiol (Marvelon) compared with desogestrel plus 20 µg ethinyl estradiol (Mercilon).
had a history of an episode during pregnancy, it is unlikely that an OC would be prescribed after delivery. Previous pregnancy therefore is a potential marker for prescription bias and was included as a binary variable. Among the case patients there were women who had illnesses that could be associated with an increased risk of venous thromboembolism, for example systemic lupus erythematosus or asthma. To take this into account a binary variable indicating whether the woman had a chronic illness was included in the model. Previous use of postcoital contraception was included as an indicator of the woman’s compliance with contraceptive practice. German study. The German MediPlus database is similar in concept to the UK version, but the data are less complete and the population covered is smaller. The data were harvested from about 450 doctors and yielded 101,000 woman-years exposure to OCs between January 1991 and December 1995. In this study, case patients were defined as women who had a diagnosis of venous thromboembolism, evidence of long term treatment with an anticoagulant, no record of a previous venous thromboembolism, and exposure to an OC on the day of the event. Four control subjects were randomly selected from the whole database (there was no matching for practice). The control subjects were born in the same year as the case patients, were exposed to an OC on the event day of the matched case patient, and had no record of a previous venous thromboembolism. Case patients and control subjects with a record of pregnancy at the time of the event were excluded. The methods were otherwise similar.
Table II. Incidence rates of venous thromboembolism among users of combined OCs with low estrogen dose
Results Patterns of use. In the UK database there were 234,900 woman-years (calculated as the number of prescribed cycles of OCs divided by 13) of OC prescriptions between September 1991 and September 1995. The German database had 105,000 woman-years exposure. Although the distributions of prescriptions according to the age of the women were similar in the 2 countries (Fig 1), a slightly higher proportion of the prescriptions in Germany were for older women. The median age of the OC users in the United Kingdom was 26 years, whereas that in Germany
Source
Rate (events/10,000 woman-y)
Vessey et al5 Gerstman et al4 Jick et al10 Farmer and Preston22 Farmer et al17 Farmer et al18
3.50 4.20 2.30 3.00 4.10 4.20
was 27 years. Among the combined OCs the 3 main progestogens used in both countries were desogestrel, gestodene, and levonorgestrel. The woman-years of use and the average ages of the users of these 3 products are shown in Table I. Gestodene products were used less frequently in Germany than in the United Kingdom. The average ages of the users of all 3 types of products were higher in Germany than in the United Kingdom, but in both countries the users of products containing the newer progestogens (desogestrel and gestodene) were younger than the users of levonorgestrel. Products containing cyproterone acetate represented a remarkably high proportion of the total use (>6%) in Germany. Two products contained desogestrel, Marvelon (desogestrel plus 30 µg ethinyl estradiol) and Mercilon (desogestrel plus 20 µg ethinyl estradiol). In the United Kingdom there were notable differences in the age distributions of the users of these 2 products (Fig 2), with the distribution of Mercilon being shifted toward the older ages and having a relatively large tail. Among the Mercilon users 23.8% were >35 years old, compared with 6.3% among Marvelon users. Crude incidence. Eighty-three cases of venous thromboembolism were found in the UK database. This is the equivalent of 3.5 cases/10,000 woman-y. When progestogen-only products were excluded the rate was 4.1 cases/10,000 woman-y. In the German database there were 42 cases (4.2 cases/10,000 woman-y). These rates are comparable to those found in other studies, although they are higher than those reported by Jick et al10 in 1995 (Table II). The age-specific incidence rates for the UK
S82 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table III. Logistic regression models from the UK MediPlus study
Fig 3. Age-specific (in years) incidence rates of venous thromboembolic disease calculated from cases in the UK MediPlus database.
case patients are shown in Fig 3. Rates tended to rise exponentially with age. In Germany 27 of the case patients were using a second-generation product (mainly levonorgestrel) and 15 were using either gestodene or desogestrel products. In the United Kingdom 44 women were using desogestrel or gestodene products and 29 were using levonorgestrel products. The crude product-specific rates for Germany were 3.95 cases/10,000 woman-y for third-generation products and 4.03 cases/10,000 woman-y for second-generation products. The equivalent figures for the UK study were 4.96 and 3.10 cases/10,000 woman-y for third- and second-generation products, respectively. Among the case patients in the United Kingdom there were 19 exposed to desogestrel with 30 µg ethinyl estradiol (Marvelon) and 13 to desogestrel with 20 µg ethinyl estradiol (Mercilon). The unadjusted incidence rate of venous thromboembolism for Marvelon was 3.99 cases/ 10,000 woman-y exposure (95% confidence interval 2.46.22 cases/10,000 woman-y); for Mercilon it was 11.53 cases/10,000 woman-y (95% confidence interval 6.1419.71 cases/10,000 woman-y). Case-control analyses. The nested case-control study on the UK data was analyzed with 3 conditional logistic regression models. The first used second-generation products as the reference group, the second was restricted to levonorgestrel-based combined OCs, and the third used levonorgestrel products as the reference group but considered the 2 desogestrel products (Marvelon and Mercilon) as separate entities. In all cases adjustment was made for body mass index, change in the OC used within 3 months of the event, number of cycles used preceding the event, previous pregnancy, concurrent illness, and previous use of postcoital contraception. Because the control subjects were matched with the case patients by exact year of birth, adjustment for age was not required for age in any of the regression models. The key results are shown in Table III. The comparison of all third-generation OCs with all second-generation prod-
Reference group
Product
OR
95% CI
Second-generation combined OCs
Thirdgeneration Thirdgeneration except Mercilon Mercilon Other secondgeneration Gestodene Desogestrel Gestodene Desogestrel + 20 µg ethinyl estradiol Desogestrel + 30 µg ethinyl estradiol
1.34
0.7-2.4
1.18
0.7-2.2
3.49 0.42
1.2-10.1 0.1-2.1
0.87 0.84 0.95 2.93
0.4-1.8 0.4-1.9 0.4-2.1 0.9-10.0
0.64
0.6-1.5
Levonorgestrel (model 2) Levonorgestrel (model 3)
OR, Odds ratio; CI, confidence interval.
ucts yielded a nonsignificant odds ratio of 1.34 (95% confidence interval 0.74-2.39). For Marvelon the odds ratio was 1.18 (95% confidence interval 0.66-2.17), and for Mercilon the odds ratio was 3.49 (95% confidence interval 1.21-10.12). The second model, which used levonorgestrel-based combined OCs as the reference group, is directly comparable to the models used in the Transnational,9 WHO,8 and BCDSP10 studies.10 It showed a nonsignificant odds ratio of 0.42 for nonlevonorgestrel second-generation products and nonsignificant odds ratios <1 for both gestodene and desogestrel. This analysis indicates that the nonlevonorgestrel second-generation products had a lower, but not significantly so, incidence of venous thromboembolism than did the levonorgestrel products. Moreover, it is consistent with the interpretation of the apparent elevation in risk of third-generation products compared with all second-generation OCs that was found in the first model as caused by the inclusion of nonlevonorgestrel products among the second-generation products. The third model showed that desogestrel with 20 µg of ethinyl estradiol had a higher, although not statistically significantly so, risk than did desogestrel with 30 µg ethinyl estradiol. The German database did not have the body mass index and other variables recorded with sufficient consistency to allow simultaneous adjustment for confounding variables in the same way as was done for the UK data. There were 42 case patients and 168 control subjects. Exposures to second-generation products were 64.3% of the case patients and 53.0% of the control subjects; the equivalent figures for third-generation products were 35.7% and 38.1%, respectively. Exposure to other types of combined OCs was 8.9% among the control subjects. The
Farmer and Lawrenson S83
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Table IV. Previous health care use among the case patients and control subjects in the German Mediplus study Case patients
Control subjects
Type
Consultations
Prescriptions
Consultations
Prescriptions
Significance
Oral contraception Psychologic and social Physical complaints
3.07 0.29 13.57
1.83 0.97 11.71
3.35 0.27 5.44
2.25 2.49 7.46
NS NS P < .0001
Numbers represent average numbers of consultations and prescriptions per person. NS, Not significant.
unadjusted odds ratio of venous thromboembolism among users of third-generation products compared with users of second-generation products was 0.77 (95% confidence interval 0.38-1.57). Despite the fact that it was not possible to adjust this odds ratio for possible confounding, it is unlikely that adjustment would result in an odds ratio that was either significant or greater than unity. In the German study the number of consultations in the months preceding the event were counted for each of the case patients and control subjects. The consultations were divided into 3 types: psychologic problems, physical complaints, and oral contraception. The average numbers of consultations, and of related prescriptions, for each group are set out in Table IV. There was no difference between the case patients and control subjects with respect to OC consultations or prescriptions, nor was there one for consultations or prescriptions for psychologic complaints. However, the case patients had a significantly higher average number of consultations for physical illnesses than did the control subjects (13.57 compared with 5.44; P < .0001). This finding is consistent with the hypothesis that either the case patients had other physical illnesses before the episode or there were prodromal symptoms of the impending venous thromboembolism. Comment Observational studies of rare events are always difficult to execute and interpret. The possibility of uncontrolled and uncontrollable bias or confounding always exists and often cannot be identified. At first sight the results of the 2 MediPlus studies appear to be at variance with those of the 3 earlier studies,8-10 all of which could be taken to imply odds ratios on the order of 2. However, differences among these studies and inconsistencies within studies may explain some of the apparent conflicts in the results. Table V sets out the key findings from the studies. The WHO8 study showed an elevated odds ratio for levonorgestrel versus nonuse of 3.4 (95% confidence interval 2.5-4.7) for pooled data from all centers; for pooled data for gestodene and desogestrel the odds ratio was 9.4 (95% confidence interval 5.6-15.6). The odds ratio for levonorgestrel products was low compared with odds ratios from other studies that used similar case definition criteria; in contrast, the odds ratio for desogestrel and gestodene combined versus nonuse is not out of line
Table V. Summary of results from recent studies8-10 Study Levonorgestrel versus nonuse WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) Third-generation versus nonuse WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) Gestodene versus levonorgestrel WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) BCDSP MediPlus (United Kingdom) Desogestrel vs levonorgestrel WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) BCDSP MediPlus (United Kingdom)
Exposed
OR
95% CI
97 40
4.0 3.1
2.7-5.8 1.7-5.5
36
8.5
3.5-20.3
37 52
2.5 3.3
1.5-4.0 2.0-5.7
23 48
20.7 6.8
7.9-53.7 3.5-13.4
43
11.5
4.6-28.8
98 29
4.4 6.7
3.0-5.6 3.4-13.0
16 20
5.3 2.0
1.8-15.5 0.8-4.7
16
0.9
0.3-2.8
45 10
1.4 2.6
0.9-2.3 1.0-7.2
22 22
2.1 0.9
1.0-4.4 0.4-1.8
7 28
5.2 2.2
2.0-13.7 1.1-4.9
27
1.8
0.7-4.8
53 25
1.6 1.5
1.0-2.5 0.8-3.1
30 32
2.2 0.8
1.1-4.4 0.4-1.9
OR, Odds ratio; CI, confidence interval.
with previous findings. There are several possible explanations for these findings. It could be that OCs, with the exception of those containing gestodene or desogestrel, became safer between the earlier investigations and those most recently published. Alternatively, the differences between the earlier investigations and the WHO8 study could be explained by differences in methodology or differences in the exposed populations. In contrast,
S84 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table VI. Inverse dose-response relationships with dose of estrogen from recent studies8-10 Study
Reference
Case patients
WHO Transnational BCDSP MediPlus UK
Nonusers LNG LNG LNG
8 13 4 13
LNG
9
WHO
OR
95% CI
Desogestrel + 20 µg EE 38.2 4.5-325 2.8 1.3-6.5 2.7 NA 2.9 0.9-10.0 Cyproterone + 35 µg EE 5.1 1.3-20.3
Case patients
27 32 26 19 9
OR
95% CI
Desogestrel + 30 µg EE 7.6 3.9-14.7 1.5 0.9-2.5 1.9 NA 0.6 0.3-1.5 Cyproterone + 50 µg EE 1.3 0.5-3.8
OR, Odds ratio; CI, confidence interval; EE, ethinyl estradiol; LNG, levonorgestrel; NA, not available in the publication.
the odds ratio from the UK center (Oxford) of the WHO8 study for the third-generation products versus nonuse is more in line with previous finding relating to combined OCs. For the comparison of the risks associated with second-generation OCs with those associated with third-generation OCs, data from the non-Oxford centers are difficult to interpret because the use of thirdgeneration OCs was so low in those centers. Thus 26 of the 28 control subjects exposed to desogestrel and 22 of the 28 control subjects exposed to gestodene were from the Oxford center; in contrast, 76 of the 203 control subjects exposed to levonorgestrel were from Oxford. It follows that the data from the Oxford center are probably more relevant in this issue. It is possible to compare the odds ratios for desogestrel and gestodene separately with levonorgestrel in each of the 3 pivotal studies and the UK MediPlus study. For gestodene the odds ratios vary between 0.9 and 2.6 (excluding the WHO8 case patients from non-Oxford centers); only the German arm of the Transnational9 study and the BCDSP10 study yielded significant results. In Germany the use of gestodene-containing combined OCs fell to a low level after the alarm caused by the publication of the pharmacokinetic study in 1989,20,21 and the odds ratio was therefore calculated from only 10 case patients. For desogestrel the odds ratios varied between 0.8 and 2.2; among all, except the non-Oxford WHO8 centers, 3 analyses (WHO8 Oxford with hospital control subjects, the BCDSP10 study, and the UK centers from the Transnational9 study) produced statistically significant results. In the UK MediPlus study body mass index was treated as a continuous variable. It was converted into a categoric variable in the other studies. The WHO8 and Jick et al10 studies used 3 body mass index categories (<20, 20-25, and >25), whereas the Transnational9 study used 4 categories (<20, 20-25, 25-30, and >30). In the WHO8 study the mean body mass indices of case patients and control subjects were as follows: case patients from non-Oxford centers, 25.1; case patients from Oxford centers, 25.2; control subjects from non-Oxford centers, 23.7; control subjects from Oxford hospitals, 23.6; and control subjects from Oxford general practice, 23.1. It follows that the highest category of body mass index in this study was
less than the mean body mass index for the case patients and only slightly greater than the mean body mass index for the control subjects. High body mass index (obesity) is an important risk factor for venous thromboembolism under any circumstances, and it is possible that there was inadequate adjustment in the regression model. Jick et al10 did not publish the mean body mass indices of the case patients and control subjects, but 37% of the case patients and 18% of the control subjects were in the highest of their categories, compared with 7% and 11% in the lowest category. In the Transnational9 study 16.4% of the case patients had a body mass index ≥30, compared with 11.2% of the hospital control subjects and 8.6% of the community control subjects. All studies indicate that desogestrel combined with 20 µg ethinyl estradiol carries a higher risk than does desogestrel combined with 30 µg ethinyl estradiol. The WHO8 study provides further evidence of an inverse dose response relationship for estrogen, with a higher odds ratio for cyproterone with 35 µg estrogen than for cyproterone with 50 µg. All previous work had shown a positive dose-response relationship with estrogen and no relationship between the potency (dose) of the progestogen and the incidence of venous thromboembolism. In view of this, the estimate of the odds ratio for all desogestrel products should be viewed with caution. Overall the MediPlus studies are consistent with all the other investigations except the BCDSP10 study. Taken together, they do not indicate a consistent increase in risk associated with either of the third-generation OCs. The key findings from the WHO8 study with respect to thirdgeneration products are from the Oxford region, which contributed to the global study 80% of the case patients and 93% of the control subjects for desogestrel and 56% of the case patients and 79% of the control subjects for gestodene. For gestodene the odds ratio with community control subjects was 0.9 (95% confidence interval 0.32.3); that with hospital control subjects was 1.4 (95% confidence interval 0.9-2.3). For desogestrel use and community control subjects the odds ratio was 1.8 (95% confidence interval 0.7-4.8). The only significant elevation in odds ratio was for desogestrel use with hospital control subjects (2.2, 95% confidence interval 1.1-4.9).
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Because of the inverse dose-response relationship with the estrogen, the odds ratio estimates for desogestrel must be viewed with caution. Most of the case patients with venous thromboembolism associated with third-generation OCs in the Transnational9 study were from the United Kingdom. The German part of the study, although interesting, is not directly comparable to the UK part because in Germany the market share for gestodene was falling during the period of the investigation, whereas in the United Kingdom it was rising. Neither the odds ratio for desogestrel (1.6) nor that for gestodene (1.4) was significant. Again, the odds ratio for desogestrel must be regarded with caution because of the inverse dose-response relationship with estrogen. The UK MediPlus study did not show an increase in odds ratio for either desogestrel or gestodene. The German MediPlus study did not yield an increased odds ratio for third-generation OCs. We turn now to some of the details of the studies that may explain some of the residual differences in results. The WHO8 and BCDSP10 studies excluded fatalities, and this may have had an effect on the results. It is particularly important to note that pulmonary embolism is classically associated with sudden, unexpected death. In the UK MediPlus study and in the Transnational9 studies, deaths were included. There were differences between the studies in case definition. In the Transnational,9 WHO,8 and Jick et al10 studies, objective evidence of the disease was sought through hospital records. In contrast, the 2 MediPlus studies relied on the diagnosis in the medical records combined with evidence of long-term treatment with anticoagulants. Both sets of criteria are problematic. Venous thromboembolism is difficult to diagnose even with the most sophisticated imaging techniques. Sometimes the disease is wrongly excluded; it may also be incorrectly diagnosed. It follows that the criteria used in the Transnational,9 WHO,8 and Jick et al10 studies may have resulted in the exclusion of some true case patients; the low incidence reported in the Jick et al10 study is consistent with this hypothesis. It is also possible that the criteria used in the MediPlus studies could have resulted in the inclusion of inappropriate case patients. Had that been the case, it would only affect the odds ratio estimates if there were a differential misdiagnosis according to the type of OC being used by the case patients. The WHO8 and Transnational9 studies relied on hospital case patients, thereby excluding cases that were diagnosed and treated on an outpatient basis. In the UK MediPlus study there were case patients who were not admitted to hospital at any time during the illness. Neither the WHO8 study nor the Transnational9 study provides evidence that hospital admission is inevitable in cases of venous thromboembolism in all the centers involved in their investigation. Differential hospital referrals among
Farmer and Lawrenson S85
users of different OCs or women of different ages could have introduced bias into the studies. In all case-control studies the source and selection of control subjects is crucial. The WHO8 and Transnational9 studies both used a combination of hospital and community control subjects. In their initial article, the Transnational team9 did not separate the odds ratio derived from the community control subjects from that derived from the hospital control subjects; however, this information was subsequently made available. The WHO8 study distinguished between the 2 types of control subjects for the Oxford region (United Kingdom). The problem with hospital control subjects when investigating the effects of OCs is that hospital admission for reasons other than those associated with reproduction is rare among young women (<5%/y in the United Kingdom). Whether those admitted to the hospital can be regarded as representative of the population from which the case patients were drawn is thus open to question. The MediPlus UK study and the Jick et al10 study both suffer from the limitation that the general practice databases do not include prescriptions for OCs issued through family planning and other clinics outside the general practice system. In the United Kingdom about 20% of all OCs are prescribed in such clinics. In the MediPlus study only exposures (both among case patients and control subjects) that were recorded in the database were taken into account. Jick et al10 made further inquiries of the general practitioners regarding exposure among the case patients but not the control subjects. It follows that in the MediPlus study some of the case patients and control subjects would have been wrongly classified as not exposed. Because the investigation was limited to a comparison of the risks associated with different types of OC and nonexposed case patients and control subjects were excluded, this should not have introduced a major bias. The Jick et al10 study could be biased because the data on the case patients were not collected in the same way as those on the control subjects. The UK MediPlus study differs from the other investigations in 2 other major respects: (1) by the matching of control subjects with case patients by exact year of birth and (2) by including a binary variable representing concurrent disease within the regression model. Age is an important variable in the investigation of OCs because it is associated with a number of other variables that may affect the probability of venous thromboembolism or use of an OC (eg, which OC is being used, whether there had been a previous pregnancy, the number of previous pregnancies, whether there had been a previous venous thromboembolism). Each of these variables change rapidly with age. Between 15 and 19 years the proportion of women having ever used an OC changes from <5% to as much as 60%, depending on the area in which the in-
S86 Farmer and Lawrenson
vestigation is done. At the other end of the scale, the proportion of women who have been sterilized changes rapidly between 30 and 35 years; again, the actual figures depend on the area in which the investigation is taking place. Age matching in 5-year bands thus may distort the results of studies. The extent to which such matching does cause distortion is a matter of speculation, because none of the investigators used both types of matching. Adjustment for age within the model, which was used by some of the investigators, also has problems because it involves making assumptions regarding the change in risk with age. Venous thromboembolism is associated with a range of other problems and illnesses, most of which are rare. Because they are rare it is unlikely that any single disease, exposure, or syndrome would affect the regression model. In the MediPlus study a composite binary variable representing the presence of a chronic illness that was not among the exclusion criteria was incorporated into the regression model. This variable made a significant contribution to the total model. It is clear that none of the studies is perfect, and all have important differences in design and execution. Taken together, the studies do not indicate consistent, statistically significant differences in the risk of venous thromboembolism among combined OCs containing different progestogens. With an underlying risk on the order of 3 cases/10,000 woman-y exposure, a case fatality rate <2%, and little residual morbidity among survivors, the small differences in odds ratios (which vary above and below unity) cannot be regarded as clinically significant. The inverse odds ratio according to estrogen dose associated with both desogestrel and cyproterone are not biologically plausible and must cast doubt on the completeness of the analytic models that were used in all the studies.
September 1998 Am J Obstet Gynecol
5.
6.
7.
8.
9.
10.
11.
12.
13.
14. 15.
16. 17.
18. REFERENCES
1. Sandler DA, Martin JF. Autopsy proven pulmonary embolism in patients: are we detecting enough deep vein thrombosis? J R Soc Med 1989;82:203-5. 2. Goldhaber SZ. Epidemiology of pulmonary embolism and deep vein thrombosis. In: Bloom AL, Forbes CD, Thomas DP, Tuddenham EG, editors. Haemostasis and thrombosis. 3rd ed. London: Churchill Livingstone; 1994. 3. Inman WH, Vessey MP, Westerholm B, Engelund A. Thromboembolic disease and the steroidal content of oral contraceptives: a report to the Committee on Safety of Drugs. BMJ 1970;2:203-9. 4. Gerstman BB, Piper JM, Freiman JP, Tomita DK, Kennedy DL, Ferguson WJ, et al. Oral contraceptive oestrogen and progestin
19. 20.
21.
22.
potencies and the incidence of deep venous thromboembolism. Int J Epidemiol 1990;19:4:931-6. Vessey MP, Mant D, Smith A, Yeates D. Oral contraceptives and venous thromboembolism: findings in a large prospective study. BMJ (Clin Res Ed) 1986;292:526. Piper JM, Kennedy DL. Oral contraceptives in the United States: trends in content and potency. Int J Epidemiol 1987; 16:215-21. World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraception. Venous thromboembolic disease and combined oral contraceptives: results of international multicentre case-control study. Lancet 1995;346:1575-82. World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraception. Effect of different progestagens in low oestrogen oral contraceptives on venous thromboembolic disease. Lancet 1995;346:1582-8. Spitzer WO, Lewis MA, Heinemann LA, Thorogood M, MacRae KD. Third generation oral contraceptives and risk of venous thromboembolic disorders: an international case-control study. BMJ 1996;312:83-8. Jick H, Jick SS, Gurewich V, Myers MW, Vasilakis C. Risk of idiopathic cardiovascular death and nonfatal venous thromboembolism in women using oral contraceptives with differing progestagen components. Lancet 1995;346:1589-93. Committee on Safety of Medicines. Combined oral contraceptives and thromboembolism: letter to doctors and pharmacists in the UK. Committee on Safety of Medicines; 1995 Oct. Lewis MA, Heinemann LA, MacRae KD, Bruooacher R, Spitzer WO. The increased risk of venous thromboembolism and the use of third generation progestagens: the role of bias in observational research. Contraception 1996;54:5-13. Westhoff CL. Oral contraceptives and venous thromboembolism: should epidemiologic associations drive clinical decision making? Contraception 1996;54:1-3. Farmer RD, Lawrenson RA. Utilization patterns of oral contraceptives in UK general practice. Contraception 1996;53:211-5. van Lunsen HW. Recent oral contraceptive use patterns in four European countries: evidence for selective prescribing of oral contraceptives containing third generation progestagens. Eur J Contracept Reprod Health Care 1996;1:39-45. Walker AM. Confounding by indication. Epidemiology 1996; 7:335-6. Farmer RD, Lawrenson RA, Thompson CR, Kennedy JG, Hambleton IR. Population based study of risk of venous thromboembolism associated with various oral contraceptives. Lancet 1997;349:83-8. Farmer RD, Todd JC, Lewis MA, MacRae KD, Williams TJ. The risks of venous thromboembolic disease amongst German women using oral contraceptives: a database study. Contraception 1998;57:67-70. Poulter NR, Farley TM, Chang CL, Marmot MG, Meirik O. Safety of combined oral contraceptive pills. Lancet 1996;347:547. Kuhl H, Jung-Hoffman C, Heidt F. Alterations in the serum levels of gestodene and SHBG during 12 cycles of treatment with 30 micrograms of ethinylestradiol and 75 micrograms gestodene. Contraception 1988;38:477-86. Jung-Hoffman C, Kuhl H. Interaction with the pharmacokinetics of ethinylestradiol and progestagens contained in OCs. Contraception 1989;40:299-312. Farmer RD, Preston TD. The risk of thromboembolism associated with low oestrogen oral contraceptives. J Obstet Gynaecol 1995;15:195-200.
Key words: Case-control studies, desogestrel, gestodene, oral contraceptives, venous thromboembolism
Venous thromboembolic disease, manifested either as peripheral thrombosis (eg, deep venous thrombosis in a limb) or as pulmonary embolism, is rare among young women. Its clinical manifestations are ubiquitous. Both deep venous thrombosis and pulmonary embolism can occur without symptoms and resolve spontaneously. Current noninvasive investigative techniques do not always allow deep venous thrombosis or pulmonary embolism to be either diagnosed or excluded with certainty, and fatalities, although rare (between 0.5% and 4%), are often unexpected. The diagnostic problems are highFrom the Department of Public Health and Epidemiology, Imperial College of Science, Technology, and Medicine, University of London. Reprint requests: Professor R. D. T. Farmer, European Institute of Health and Medical Sciences, University of Surrey, Stirling House Campus, Surrey Research Park, Guilford Surrey, GU2 5RF, England. Copyright © 1998 by Mosby, Inc. 0002-9378/98 $5.00 + 0 6/0/92634
S78
lighted by the observation from an autopsy study that although 83% of subjects were found to have a pulmonary embolism, only 19% had been thought to have had a deep venous thrombosis during life.1 The clinical manifestation and investigation of deep venous thrombosis and pulmonary embolism complicate the epidemiologic investigation of the disease. In particular, not all cases come to medical attention, fatality among untreated patients is not inevitable, and some affected persons die without manifesting symptoms during life. A number of thrombophilic disorders carry an increased risk of venous thrombosis; however, the presence of such a disorder does not make thrombosis inevitable, even under adverse circumstances. Deep venous thrombosis (with subsequent pulmonary embolism) has an increased frequency after surgery, trauma, immobilization, cancer, pregnancy, obesity, air travel, and (among women) use of oral contraceptives (OCs).2 The increased risk after
Farmer and Lawrenson S79
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
surgery is maintained for several weeks. In pregnancy the risk is greatest in the few weeks immediately before and after delivery. Other circumstances in which the risk of deep venous thrombosis is increased include occult cancers, some of the autoimmune diseases (eg, systemic lupus erythematosus), varicose veins, inflammatory bowel disease, hypothyroidism, and renal disease. Among women who use OCs the risk of venous thromboembolic disease appears to be related to the dose of estrogen.3-5 Since the early 1970s the dose of estrogen in combined OCs has been reduced, and preparations are now available with as little as 20 µg of estrogen. The reduction in dose of estrogen has been associated with reductions in both the absolute incidence of deep venous thrombosis and the incidence relative to nonusers of OCs.6 Gerstman et al4 investigated the influence of estrogen and progestogen potency on the risk of venous thromboembolism among women between 15 and 44 years old who were members of the Michigan Medicaid scheme. They found a positive association with estrogen potency but no relationship with progestogen potency. The low-potency (low-dose) estrogen OC products yielded an unadjusted incidence rate of venous thromboembolism of 5.3 cases/10,000 woman-y exposure, whereas those classified as high-potency products carried an incidence of 9.8 cases/10,000 woman-y. Despite the reassuring and consistent reductions in incidence of deep venous thrombosis and pulmonary embolism among women who use combined OCs, the World Health Organization (WHO) initiated a further study in the mid 1980s. The WHO investigation was an internationally based case-control study. Among the reasons for the investigation were that “most studies have been done in northern Europe and the USA, and it may not be appropriate to extrapolate the results to populations with different incidence rates of VTE disease and prevalence of associated risk factors” and that “part of the apparent variability in OC-associated risk estimates reported was due to difficulties in diagnosing DVT and PE accurately, particularly when the diagnosis was based mainly on clinical grounds.”7 The study was conducted in 21 centers across 17 countries in different regions of the world. One of the incidental findings was that combined OCs containing desogestrel, gestodene, or norgestimate (which were classified as third-generation OCs) were associated with a higher incidence of venous thromboembolism than were those containing levonorgestrel, norgestrel, or norgestrienone in combination with <35 µg ethinyl estradiol (classified as second-generation OCs).8 Some time after the start of the WHO8 study Spitzer et al9 began a multicenter, Europeanbased case-control study of the cardiovascular effects of OCs. Their study was prompted because the “results of a pharmacokinetic study of 22 women in Germany suggested that gestodene, but not desogestrel, may increase
the risk of vascular events.”9 The Spitzer et al9 study (known as the Transnational study) used a protocol similar to that of the WHO8 study but was specifically designed to compare the cardiovascular risks of combined OCs containing different progestogens. After the early release of the results of the WHO study relating to the comparative risks of OCs containing different progestogens,8 Jick et al10 investigated the risks of cardiovascular disease among women who used different OCs through the UK General Practice Research Data Base, a study referred to as the Boston Collaborative Drug Surveillance Program study. In October 1995 the UK Committee on Safety of Medicines reacted to the unpublished results of these 3 studies by issuing a formal warning to doctors regarding an apparent increased risk of venous thromboembolism associated with third-generation OCs.11 On their publication it was apparent that all 3 pivotal studies (WHO,8 Transnational,9 and BCDSP10 studies) showed the incidence of venous thromboembolic disease among women who used third-generation OCs to be higher than that among women who used second-generation products. Subsequently the interpretation of the results of the 3 studies has been criticized. In particular it has been suggested that they could have been affected by bias and confounding.12-16 In view of the importance of the issues it was decided to initiate further investigations. In this article the results of 2 studies, 1 from the United Kingdom and 1 from Germany, both of which were based on general practice computer-generated clinical databases, are presented. These studies were specifically designed to compare the risks of venous thromboembolism associated with OCs containing different progestogens. The main results of both have been published elsewhere.17,18 Methods UK study. The UK study used a clinical database generated by collating the computer-based records from about 140 general practices in the United Kingdom. The database (which is known as the UK MediPlus Database and is maintained by Intercontinental Medical Statistics London, United Kingdom) is similar in concept and execution to the General Practice Research Database that was used by Jick et al.10 There is no overlap between the populations in the 2 databases. The records used in the analysis cover a population of about 490,000 women between 15 and 44 years old. The period investigated (September 1991-September 1995) involved about 235,000 woman-years exposure to OCs. The information recorded in most of the medical records included the following: date of registration with the practice, year of birth, date of death, height, weight, blood pressure, smoking habits, record of each prescription, diagnosis at consultation, investigations, and hospital referrals. The first part of the study was designed to describe the way in which OCs were prescribed in UK general prac-
S80 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table I. Woman-years of prescriptions and average age of users in the UK and German MediPlus databases for combined OCs according to the major type of progestogen United Kingdom
Progestogen Desogestrel Gestodene Levonorgestrel Others
Germany
Prescription (1000 woman-y)
Proportion (%)
Average age (y)
Prescription (1000 woman-y)
Proportion (%)
Average age (y)
60.5 49.9 63.5
24.9 20.5 26.1
25.7 25.2 27.3
31.0 7.0 35.2 31.8
29.5 6.7 33.5 30.3
27.7 27.8 29.4 28.5
Woman-years of prescription are given in thousands; proportions are percentage of total OC use.
tice. This was done by identifying all prescriptions for OCs, together with the number of cycles prescribed and the age of the woman at the time of the prescription. The age distributions of the recipients of different types of OCs were compared. The cases of venous thromboembolism were then selected to calculate the population incidence rates. Potential case patients with venous thromboembolism were women who had a diagnosis of deep venous thrombosis or pulmonary embolism between September 1991 and September 1995 and had prescriptions for oral anticoagulants. Women who did not have a record of a prescription for an OC that would have covered the day on which she had the venous thromboembolic episode were excluded. Women with a history of venous thromboembolism before September 1991 were excluded and those with “nonfirst episodes” (venous thromboembolic episodes subsequent to a previously recorded episode before September 1991) during the period investigated were also excluded. The records of all the potential case patients were examined to establish whether any of the potential case patients had given birth within 38 weeks of the venous thromboembolic episode; those women who had given birth within this period were excluded as case patients. Women with a record of a recent surgical operation or trauma were also excluded. Patients who died but were not excluded for any of the preceding reasons were included as case patients. The crude incidence rates were calculated for different groups of OC products. Finally, a nested case-control study of venous thromboembolism was conducted. As many as 4 control subjects were matched with case patients with respect to the general practice with which they were registered and by exact year of birth. Women with a record of a past venous thromboembolism, those who were pregnant on the day the episode occurred in the case patient, and those who had a recent history of a surgical operation or trauma were not eligible to be selected as control subjects. All the control subjects were covered by a prescription for an OC on the event day of the case patient with whom they were matched. The data were modeled with conditional logistic regression and by imputing missing variables in
cases with incomplete records of body mass index. The following models were used: 1. In model 1 third-generation OCs are compared with second-generation products. This model is similar to a model used in the Transnational9 study. 2. In model 2 gestodene and desogestrel are compared with levonorgestrel. This model is comparable to that used in the WHO8 and Jick et al10 studies. 3. Model 3 is the same as model 2 except that the 2 products containing desogestrel (desogestrel with 20 µg ethinyl estradiol and desogestrel with 30 µg ethinyl estradiol) were separated. In all models adjustment was made for body mass index (as a continuous variable with imputation of missing values), change in the OC used within 3 months of the event or of the event day of the case patient for control subjects (as a binary variable), number of cycles used before the event, previous pregnancy (as a binary variable), concurrent illness (as a binary variable), and previous use of postcoital contraception (as a binary varable). Blood pressure and smoking were not included in the model because these data were not available for a high proportion of case patients and control subjects. The body mass index (weight in kilograms divided by the height in meters squared) was available for 65.1% of the case patients and 64.7% of the control subjects. For a further 14.1% of case patients and 9.0% of control subjects, categoric information on body mass was available (low, normal, or high). The weight alone was recorded for a further 5.6% of case patients and 7.1% of control subjects. No data were recorded for 20.2% of case patients and 14.1% of control subjects. Body mass is known to be an important risk factor for venous thromboembolism. To maximize the numbers of case patients and control subjects that could be used in the regression model, missing values were imputed from all available data. There is some evidence that recency of use could affect the probability of the occurrence of a venous thromboembolic episode19; the numbers of cycles used before the episode were therefore included in the model. It is known that pregnancy carries a relatively high risk of venous thromboembolic disease. If a woman
Farmer and Lawrenson S81
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Fig 1. Exposure (in 1000 woman-years) to OCs vesus age (in years) for the MediPlus databases from the United Kingdom and Germany, 1991-1995.
Fig 2. Age distribution of desogestrel with 30 µg ethinyl estradiol (Marvelon) compared with desogestrel plus 20 µg ethinyl estradiol (Mercilon).
had a history of an episode during pregnancy, it is unlikely that an OC would be prescribed after delivery. Previous pregnancy therefore is a potential marker for prescription bias and was included as a binary variable. Among the case patients there were women who had illnesses that could be associated with an increased risk of venous thromboembolism, for example systemic lupus erythematosus or asthma. To take this into account a binary variable indicating whether the woman had a chronic illness was included in the model. Previous use of postcoital contraception was included as an indicator of the woman’s compliance with contraceptive practice. German study. The German MediPlus database is similar in concept to the UK version, but the data are less complete and the population covered is smaller. The data were harvested from about 450 doctors and yielded 101,000 woman-years exposure to OCs between January 1991 and December 1995. In this study, case patients were defined as women who had a diagnosis of venous thromboembolism, evidence of long term treatment with an anticoagulant, no record of a previous venous thromboembolism, and exposure to an OC on the day of the event. Four control subjects were randomly selected from the whole database (there was no matching for practice). The control subjects were born in the same year as the case patients, were exposed to an OC on the event day of the matched case patient, and had no record of a previous venous thromboembolism. Case patients and control subjects with a record of pregnancy at the time of the event were excluded. The methods were otherwise similar.
Table II. Incidence rates of venous thromboembolism among users of combined OCs with low estrogen dose
Results Patterns of use. In the UK database there were 234,900 woman-years (calculated as the number of prescribed cycles of OCs divided by 13) of OC prescriptions between September 1991 and September 1995. The German database had 105,000 woman-years exposure. Although the distributions of prescriptions according to the age of the women were similar in the 2 countries (Fig 1), a slightly higher proportion of the prescriptions in Germany were for older women. The median age of the OC users in the United Kingdom was 26 years, whereas that in Germany
Source
Rate (events/10,000 woman-y)
Vessey et al5 Gerstman et al4 Jick et al10 Farmer and Preston22 Farmer et al17 Farmer et al18
3.50 4.20 2.30 3.00 4.10 4.20
was 27 years. Among the combined OCs the 3 main progestogens used in both countries were desogestrel, gestodene, and levonorgestrel. The woman-years of use and the average ages of the users of these 3 products are shown in Table I. Gestodene products were used less frequently in Germany than in the United Kingdom. The average ages of the users of all 3 types of products were higher in Germany than in the United Kingdom, but in both countries the users of products containing the newer progestogens (desogestrel and gestodene) were younger than the users of levonorgestrel. Products containing cyproterone acetate represented a remarkably high proportion of the total use (>6%) in Germany. Two products contained desogestrel, Marvelon (desogestrel plus 30 µg ethinyl estradiol) and Mercilon (desogestrel plus 20 µg ethinyl estradiol). In the United Kingdom there were notable differences in the age distributions of the users of these 2 products (Fig 2), with the distribution of Mercilon being shifted toward the older ages and having a relatively large tail. Among the Mercilon users 23.8% were >35 years old, compared with 6.3% among Marvelon users. Crude incidence. Eighty-three cases of venous thromboembolism were found in the UK database. This is the equivalent of 3.5 cases/10,000 woman-y. When progestogen-only products were excluded the rate was 4.1 cases/10,000 woman-y. In the German database there were 42 cases (4.2 cases/10,000 woman-y). These rates are comparable to those found in other studies, although they are higher than those reported by Jick et al10 in 1995 (Table II). The age-specific incidence rates for the UK
S82 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table III. Logistic regression models from the UK MediPlus study
Fig 3. Age-specific (in years) incidence rates of venous thromboembolic disease calculated from cases in the UK MediPlus database.
case patients are shown in Fig 3. Rates tended to rise exponentially with age. In Germany 27 of the case patients were using a second-generation product (mainly levonorgestrel) and 15 were using either gestodene or desogestrel products. In the United Kingdom 44 women were using desogestrel or gestodene products and 29 were using levonorgestrel products. The crude product-specific rates for Germany were 3.95 cases/10,000 woman-y for third-generation products and 4.03 cases/10,000 woman-y for second-generation products. The equivalent figures for the UK study were 4.96 and 3.10 cases/10,000 woman-y for third- and second-generation products, respectively. Among the case patients in the United Kingdom there were 19 exposed to desogestrel with 30 µg ethinyl estradiol (Marvelon) and 13 to desogestrel with 20 µg ethinyl estradiol (Mercilon). The unadjusted incidence rate of venous thromboembolism for Marvelon was 3.99 cases/ 10,000 woman-y exposure (95% confidence interval 2.46.22 cases/10,000 woman-y); for Mercilon it was 11.53 cases/10,000 woman-y (95% confidence interval 6.1419.71 cases/10,000 woman-y). Case-control analyses. The nested case-control study on the UK data was analyzed with 3 conditional logistic regression models. The first used second-generation products as the reference group, the second was restricted to levonorgestrel-based combined OCs, and the third used levonorgestrel products as the reference group but considered the 2 desogestrel products (Marvelon and Mercilon) as separate entities. In all cases adjustment was made for body mass index, change in the OC used within 3 months of the event, number of cycles used preceding the event, previous pregnancy, concurrent illness, and previous use of postcoital contraception. Because the control subjects were matched with the case patients by exact year of birth, adjustment for age was not required for age in any of the regression models. The key results are shown in Table III. The comparison of all third-generation OCs with all second-generation prod-
Reference group
Product
OR
95% CI
Second-generation combined OCs
Thirdgeneration Thirdgeneration except Mercilon Mercilon Other secondgeneration Gestodene Desogestrel Gestodene Desogestrel + 20 µg ethinyl estradiol Desogestrel + 30 µg ethinyl estradiol
1.34
0.7-2.4
1.18
0.7-2.2
3.49 0.42
1.2-10.1 0.1-2.1
0.87 0.84 0.95 2.93
0.4-1.8 0.4-1.9 0.4-2.1 0.9-10.0
0.64
0.6-1.5
Levonorgestrel (model 2) Levonorgestrel (model 3)
OR, Odds ratio; CI, confidence interval.
ucts yielded a nonsignificant odds ratio of 1.34 (95% confidence interval 0.74-2.39). For Marvelon the odds ratio was 1.18 (95% confidence interval 0.66-2.17), and for Mercilon the odds ratio was 3.49 (95% confidence interval 1.21-10.12). The second model, which used levonorgestrel-based combined OCs as the reference group, is directly comparable to the models used in the Transnational,9 WHO,8 and BCDSP10 studies.10 It showed a nonsignificant odds ratio of 0.42 for nonlevonorgestrel second-generation products and nonsignificant odds ratios <1 for both gestodene and desogestrel. This analysis indicates that the nonlevonorgestrel second-generation products had a lower, but not significantly so, incidence of venous thromboembolism than did the levonorgestrel products. Moreover, it is consistent with the interpretation of the apparent elevation in risk of third-generation products compared with all second-generation OCs that was found in the first model as caused by the inclusion of nonlevonorgestrel products among the second-generation products. The third model showed that desogestrel with 20 µg of ethinyl estradiol had a higher, although not statistically significantly so, risk than did desogestrel with 30 µg ethinyl estradiol. The German database did not have the body mass index and other variables recorded with sufficient consistency to allow simultaneous adjustment for confounding variables in the same way as was done for the UK data. There were 42 case patients and 168 control subjects. Exposures to second-generation products were 64.3% of the case patients and 53.0% of the control subjects; the equivalent figures for third-generation products were 35.7% and 38.1%, respectively. Exposure to other types of combined OCs was 8.9% among the control subjects. The
Farmer and Lawrenson S83
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Table IV. Previous health care use among the case patients and control subjects in the German Mediplus study Case patients
Control subjects
Type
Consultations
Prescriptions
Consultations
Prescriptions
Significance
Oral contraception Psychologic and social Physical complaints
3.07 0.29 13.57
1.83 0.97 11.71
3.35 0.27 5.44
2.25 2.49 7.46
NS NS P < .0001
Numbers represent average numbers of consultations and prescriptions per person. NS, Not significant.
unadjusted odds ratio of venous thromboembolism among users of third-generation products compared with users of second-generation products was 0.77 (95% confidence interval 0.38-1.57). Despite the fact that it was not possible to adjust this odds ratio for possible confounding, it is unlikely that adjustment would result in an odds ratio that was either significant or greater than unity. In the German study the number of consultations in the months preceding the event were counted for each of the case patients and control subjects. The consultations were divided into 3 types: psychologic problems, physical complaints, and oral contraception. The average numbers of consultations, and of related prescriptions, for each group are set out in Table IV. There was no difference between the case patients and control subjects with respect to OC consultations or prescriptions, nor was there one for consultations or prescriptions for psychologic complaints. However, the case patients had a significantly higher average number of consultations for physical illnesses than did the control subjects (13.57 compared with 5.44; P < .0001). This finding is consistent with the hypothesis that either the case patients had other physical illnesses before the episode or there were prodromal symptoms of the impending venous thromboembolism. Comment Observational studies of rare events are always difficult to execute and interpret. The possibility of uncontrolled and uncontrollable bias or confounding always exists and often cannot be identified. At first sight the results of the 2 MediPlus studies appear to be at variance with those of the 3 earlier studies,8-10 all of which could be taken to imply odds ratios on the order of 2. However, differences among these studies and inconsistencies within studies may explain some of the apparent conflicts in the results. Table V sets out the key findings from the studies. The WHO8 study showed an elevated odds ratio for levonorgestrel versus nonuse of 3.4 (95% confidence interval 2.5-4.7) for pooled data from all centers; for pooled data for gestodene and desogestrel the odds ratio was 9.4 (95% confidence interval 5.6-15.6). The odds ratio for levonorgestrel products was low compared with odds ratios from other studies that used similar case definition criteria; in contrast, the odds ratio for desogestrel and gestodene combined versus nonuse is not out of line
Table V. Summary of results from recent studies8-10 Study Levonorgestrel versus nonuse WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) Third-generation versus nonuse WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) Gestodene versus levonorgestrel WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) BCDSP MediPlus (United Kingdom) Desogestrel vs levonorgestrel WHO non-Oxford centers WHO Oxford (hospital control subjects) WHO Oxford (community control subjects) Transnational (UK centers) Transnational (German centers) BCDSP MediPlus (United Kingdom)
Exposed
OR
95% CI
97 40
4.0 3.1
2.7-5.8 1.7-5.5
36
8.5
3.5-20.3
37 52
2.5 3.3
1.5-4.0 2.0-5.7
23 48
20.7 6.8
7.9-53.7 3.5-13.4
43
11.5
4.6-28.8
98 29
4.4 6.7
3.0-5.6 3.4-13.0
16 20
5.3 2.0
1.8-15.5 0.8-4.7
16
0.9
0.3-2.8
45 10
1.4 2.6
0.9-2.3 1.0-7.2
22 22
2.1 0.9
1.0-4.4 0.4-1.8
7 28
5.2 2.2
2.0-13.7 1.1-4.9
27
1.8
0.7-4.8
53 25
1.6 1.5
1.0-2.5 0.8-3.1
30 32
2.2 0.8
1.1-4.4 0.4-1.9
OR, Odds ratio; CI, confidence interval.
with previous findings. There are several possible explanations for these findings. It could be that OCs, with the exception of those containing gestodene or desogestrel, became safer between the earlier investigations and those most recently published. Alternatively, the differences between the earlier investigations and the WHO8 study could be explained by differences in methodology or differences in the exposed populations. In contrast,
S84 Farmer and Lawrenson
September 1998 Am J Obstet Gynecol
Table VI. Inverse dose-response relationships with dose of estrogen from recent studies8-10 Study
Reference
Case patients
WHO Transnational BCDSP MediPlus UK
Nonusers LNG LNG LNG
8 13 4 13
LNG
9
WHO
OR
95% CI
Desogestrel + 20 µg EE 38.2 4.5-325 2.8 1.3-6.5 2.7 NA 2.9 0.9-10.0 Cyproterone + 35 µg EE 5.1 1.3-20.3
Case patients
27 32 26 19 9
OR
95% CI
Desogestrel + 30 µg EE 7.6 3.9-14.7 1.5 0.9-2.5 1.9 NA 0.6 0.3-1.5 Cyproterone + 50 µg EE 1.3 0.5-3.8
OR, Odds ratio; CI, confidence interval; EE, ethinyl estradiol; LNG, levonorgestrel; NA, not available in the publication.
the odds ratio from the UK center (Oxford) of the WHO8 study for the third-generation products versus nonuse is more in line with previous finding relating to combined OCs. For the comparison of the risks associated with second-generation OCs with those associated with third-generation OCs, data from the non-Oxford centers are difficult to interpret because the use of thirdgeneration OCs was so low in those centers. Thus 26 of the 28 control subjects exposed to desogestrel and 22 of the 28 control subjects exposed to gestodene were from the Oxford center; in contrast, 76 of the 203 control subjects exposed to levonorgestrel were from Oxford. It follows that the data from the Oxford center are probably more relevant in this issue. It is possible to compare the odds ratios for desogestrel and gestodene separately with levonorgestrel in each of the 3 pivotal studies and the UK MediPlus study. For gestodene the odds ratios vary between 0.9 and 2.6 (excluding the WHO8 case patients from non-Oxford centers); only the German arm of the Transnational9 study and the BCDSP10 study yielded significant results. In Germany the use of gestodene-containing combined OCs fell to a low level after the alarm caused by the publication of the pharmacokinetic study in 1989,20,21 and the odds ratio was therefore calculated from only 10 case patients. For desogestrel the odds ratios varied between 0.8 and 2.2; among all, except the non-Oxford WHO8 centers, 3 analyses (WHO8 Oxford with hospital control subjects, the BCDSP10 study, and the UK centers from the Transnational9 study) produced statistically significant results. In the UK MediPlus study body mass index was treated as a continuous variable. It was converted into a categoric variable in the other studies. The WHO8 and Jick et al10 studies used 3 body mass index categories (<20, 20-25, and >25), whereas the Transnational9 study used 4 categories (<20, 20-25, 25-30, and >30). In the WHO8 study the mean body mass indices of case patients and control subjects were as follows: case patients from non-Oxford centers, 25.1; case patients from Oxford centers, 25.2; control subjects from non-Oxford centers, 23.7; control subjects from Oxford hospitals, 23.6; and control subjects from Oxford general practice, 23.1. It follows that the highest category of body mass index in this study was
less than the mean body mass index for the case patients and only slightly greater than the mean body mass index for the control subjects. High body mass index (obesity) is an important risk factor for venous thromboembolism under any circumstances, and it is possible that there was inadequate adjustment in the regression model. Jick et al10 did not publish the mean body mass indices of the case patients and control subjects, but 37% of the case patients and 18% of the control subjects were in the highest of their categories, compared with 7% and 11% in the lowest category. In the Transnational9 study 16.4% of the case patients had a body mass index ≥30, compared with 11.2% of the hospital control subjects and 8.6% of the community control subjects. All studies indicate that desogestrel combined with 20 µg ethinyl estradiol carries a higher risk than does desogestrel combined with 30 µg ethinyl estradiol. The WHO8 study provides further evidence of an inverse dose response relationship for estrogen, with a higher odds ratio for cyproterone with 35 µg estrogen than for cyproterone with 50 µg. All previous work had shown a positive dose-response relationship with estrogen and no relationship between the potency (dose) of the progestogen and the incidence of venous thromboembolism. In view of this, the estimate of the odds ratio for all desogestrel products should be viewed with caution. Overall the MediPlus studies are consistent with all the other investigations except the BCDSP10 study. Taken together, they do not indicate a consistent increase in risk associated with either of the third-generation OCs. The key findings from the WHO8 study with respect to thirdgeneration products are from the Oxford region, which contributed to the global study 80% of the case patients and 93% of the control subjects for desogestrel and 56% of the case patients and 79% of the control subjects for gestodene. For gestodene the odds ratio with community control subjects was 0.9 (95% confidence interval 0.32.3); that with hospital control subjects was 1.4 (95% confidence interval 0.9-2.3). For desogestrel use and community control subjects the odds ratio was 1.8 (95% confidence interval 0.7-4.8). The only significant elevation in odds ratio was for desogestrel use with hospital control subjects (2.2, 95% confidence interval 1.1-4.9).
Volume 179, Number 3, Part 2 Am J Obstet Gynecol
Because of the inverse dose-response relationship with the estrogen, the odds ratio estimates for desogestrel must be viewed with caution. Most of the case patients with venous thromboembolism associated with third-generation OCs in the Transnational9 study were from the United Kingdom. The German part of the study, although interesting, is not directly comparable to the UK part because in Germany the market share for gestodene was falling during the period of the investigation, whereas in the United Kingdom it was rising. Neither the odds ratio for desogestrel (1.6) nor that for gestodene (1.4) was significant. Again, the odds ratio for desogestrel must be regarded with caution because of the inverse dose-response relationship with estrogen. The UK MediPlus study did not show an increase in odds ratio for either desogestrel or gestodene. The German MediPlus study did not yield an increased odds ratio for third-generation OCs. We turn now to some of the details of the studies that may explain some of the residual differences in results. The WHO8 and BCDSP10 studies excluded fatalities, and this may have had an effect on the results. It is particularly important to note that pulmonary embolism is classically associated with sudden, unexpected death. In the UK MediPlus study and in the Transnational9 studies, deaths were included. There were differences between the studies in case definition. In the Transnational,9 WHO,8 and Jick et al10 studies, objective evidence of the disease was sought through hospital records. In contrast, the 2 MediPlus studies relied on the diagnosis in the medical records combined with evidence of long-term treatment with anticoagulants. Both sets of criteria are problematic. Venous thromboembolism is difficult to diagnose even with the most sophisticated imaging techniques. Sometimes the disease is wrongly excluded; it may also be incorrectly diagnosed. It follows that the criteria used in the Transnational,9 WHO,8 and Jick et al10 studies may have resulted in the exclusion of some true case patients; the low incidence reported in the Jick et al10 study is consistent with this hypothesis. It is also possible that the criteria used in the MediPlus studies could have resulted in the inclusion of inappropriate case patients. Had that been the case, it would only affect the odds ratio estimates if there were a differential misdiagnosis according to the type of OC being used by the case patients. The WHO8 and Transnational9 studies relied on hospital case patients, thereby excluding cases that were diagnosed and treated on an outpatient basis. In the UK MediPlus study there were case patients who were not admitted to hospital at any time during the illness. Neither the WHO8 study nor the Transnational9 study provides evidence that hospital admission is inevitable in cases of venous thromboembolism in all the centers involved in their investigation. Differential hospital referrals among
Farmer and Lawrenson S85
users of different OCs or women of different ages could have introduced bias into the studies. In all case-control studies the source and selection of control subjects is crucial. The WHO8 and Transnational9 studies both used a combination of hospital and community control subjects. In their initial article, the Transnational team9 did not separate the odds ratio derived from the community control subjects from that derived from the hospital control subjects; however, this information was subsequently made available. The WHO8 study distinguished between the 2 types of control subjects for the Oxford region (United Kingdom). The problem with hospital control subjects when investigating the effects of OCs is that hospital admission for reasons other than those associated with reproduction is rare among young women (<5%/y in the United Kingdom). Whether those admitted to the hospital can be regarded as representative of the population from which the case patients were drawn is thus open to question. The MediPlus UK study and the Jick et al10 study both suffer from the limitation that the general practice databases do not include prescriptions for OCs issued through family planning and other clinics outside the general practice system. In the United Kingdom about 20% of all OCs are prescribed in such clinics. In the MediPlus study only exposures (both among case patients and control subjects) that were recorded in the database were taken into account. Jick et al10 made further inquiries of the general practitioners regarding exposure among the case patients but not the control subjects. It follows that in the MediPlus study some of the case patients and control subjects would have been wrongly classified as not exposed. Because the investigation was limited to a comparison of the risks associated with different types of OC and nonexposed case patients and control subjects were excluded, this should not have introduced a major bias. The Jick et al10 study could be biased because the data on the case patients were not collected in the same way as those on the control subjects. The UK MediPlus study differs from the other investigations in 2 other major respects: (1) by the matching of control subjects with case patients by exact year of birth and (2) by including a binary variable representing concurrent disease within the regression model. Age is an important variable in the investigation of OCs because it is associated with a number of other variables that may affect the probability of venous thromboembolism or use of an OC (eg, which OC is being used, whether there had been a previous pregnancy, the number of previous pregnancies, whether there had been a previous venous thromboembolism). Each of these variables change rapidly with age. Between 15 and 19 years the proportion of women having ever used an OC changes from <5% to as much as 60%, depending on the area in which the in-
S86 Farmer and Lawrenson
vestigation is done. At the other end of the scale, the proportion of women who have been sterilized changes rapidly between 30 and 35 years; again, the actual figures depend on the area in which the investigation is taking place. Age matching in 5-year bands thus may distort the results of studies. The extent to which such matching does cause distortion is a matter of speculation, because none of the investigators used both types of matching. Adjustment for age within the model, which was used by some of the investigators, also has problems because it involves making assumptions regarding the change in risk with age. Venous thromboembolism is associated with a range of other problems and illnesses, most of which are rare. Because they are rare it is unlikely that any single disease, exposure, or syndrome would affect the regression model. In the MediPlus study a composite binary variable representing the presence of a chronic illness that was not among the exclusion criteria was incorporated into the regression model. This variable made a significant contribution to the total model. It is clear that none of the studies is perfect, and all have important differences in design and execution. Taken together, the studies do not indicate consistent, statistically significant differences in the risk of venous thromboembolism among combined OCs containing different progestogens. With an underlying risk on the order of 3 cases/10,000 woman-y exposure, a case fatality rate <2%, and little residual morbidity among survivors, the small differences in odds ratios (which vary above and below unity) cannot be regarded as clinically significant. The inverse odds ratio according to estrogen dose associated with both desogestrel and cyproterone are not biologically plausible and must cast doubt on the completeness of the analytic models that were used in all the studies.
September 1998 Am J Obstet Gynecol
5.
6.
7.
8.
9.
10.
11.
12.
13.
14. 15.
16. 17.
18. REFERENCES
1. Sandler DA, Martin JF. Autopsy proven pulmonary embolism in patients: are we detecting enough deep vein thrombosis? J R Soc Med 1989;82:203-5. 2. Goldhaber SZ. Epidemiology of pulmonary embolism and deep vein thrombosis. In: Bloom AL, Forbes CD, Thomas DP, Tuddenham EG, editors. Haemostasis and thrombosis. 3rd ed. London: Churchill Livingstone; 1994. 3. Inman WH, Vessey MP, Westerholm B, Engelund A. Thromboembolic disease and the steroidal content of oral contraceptives: a report to the Committee on Safety of Drugs. BMJ 1970;2:203-9. 4. Gerstman BB, Piper JM, Freiman JP, Tomita DK, Kennedy DL, Ferguson WJ, et al. Oral contraceptive oestrogen and progestin
19. 20.
21.
22.
potencies and the incidence of deep venous thromboembolism. Int J Epidemiol 1990;19:4:931-6. Vessey MP, Mant D, Smith A, Yeates D. Oral contraceptives and venous thromboembolism: findings in a large prospective study. BMJ (Clin Res Ed) 1986;292:526. Piper JM, Kennedy DL. Oral contraceptives in the United States: trends in content and potency. Int J Epidemiol 1987; 16:215-21. World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraception. Venous thromboembolic disease and combined oral contraceptives: results of international multicentre case-control study. Lancet 1995;346:1575-82. World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraception. Effect of different progestagens in low oestrogen oral contraceptives on venous thromboembolic disease. Lancet 1995;346:1582-8. Spitzer WO, Lewis MA, Heinemann LA, Thorogood M, MacRae KD. Third generation oral contraceptives and risk of venous thromboembolic disorders: an international case-control study. BMJ 1996;312:83-8. Jick H, Jick SS, Gurewich V, Myers MW, Vasilakis C. Risk of idiopathic cardiovascular death and nonfatal venous thromboembolism in women using oral contraceptives with differing progestagen components. Lancet 1995;346:1589-93. Committee on Safety of Medicines. Combined oral contraceptives and thromboembolism: letter to doctors and pharmacists in the UK. Committee on Safety of Medicines; 1995 Oct. Lewis MA, Heinemann LA, MacRae KD, Bruooacher R, Spitzer WO. The increased risk of venous thromboembolism and the use of third generation progestagens: the role of bias in observational research. Contraception 1996;54:5-13. Westhoff CL. Oral contraceptives and venous thromboembolism: should epidemiologic associations drive clinical decision making? Contraception 1996;54:1-3. Farmer RD, Lawrenson RA. Utilization patterns of oral contraceptives in UK general practice. Contraception 1996;53:211-5. van Lunsen HW. Recent oral contraceptive use patterns in four European countries: evidence for selective prescribing of oral contraceptives containing third generation progestagens. Eur J Contracept Reprod Health Care 1996;1:39-45. Walker AM. Confounding by indication. Epidemiology 1996; 7:335-6. Farmer RD, Lawrenson RA, Thompson CR, Kennedy JG, Hambleton IR. Population based study of risk of venous thromboembolism associated with various oral contraceptives. Lancet 1997;349:83-8. Farmer RD, Todd JC, Lewis MA, MacRae KD, Williams TJ. The risks of venous thromboembolic disease amongst German women using oral contraceptives: a database study. Contraception 1998;57:67-70. Poulter NR, Farley TM, Chang CL, Marmot MG, Meirik O. Safety of combined oral contraceptive pills. Lancet 1996;347:547. Kuhl H, Jung-Hoffman C, Heidt F. Alterations in the serum levels of gestodene and SHBG during 12 cycles of treatment with 30 micrograms of ethinylestradiol and 75 micrograms gestodene. Contraception 1988;38:477-86. Jung-Hoffman C, Kuhl H. Interaction with the pharmacokinetics of ethinylestradiol and progestagens contained in OCs. Contraception 1989;40:299-312. Farmer RD, Preston TD. The risk of thromboembolism associated with low oestrogen oral contraceptives. J Obstet Gynaecol 1995;15:195-200.