Venous thromboembolic disease and combined oral contraceptives: A re-analysis of the MediPlus database

J.-C. Todd1, R. Lawrenson1,3, R.D.T. Farmer1, T.J. Williams1 and G.M. Leydon2

1 European Institute of Health and Medical Sciences, University of Surrey, Surrey Research Park, Guildford GU2 5RF, and 2 Cancer and Public Health Unit, Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
In October 1995 the Committee on Safety of Medicines advised UK doctors and pharmacists that oral contraceptives containing desogestrel and gestodene were associated with double the risk of venous thromboembolic events (VTE) compared to pills containing other progestogens. In 1997 data was analysed from the MediPlus database of UK general practitioner records, which reported odds ratios for desogestrel and gestodene lower than that for levonorgestrel. Here the results of a more stringent nested case control analysis on the MediPlus database are reported. The study was larger and cases were verified. A crude incidence of idiopathic VTE was found amongst users of combined oral contraceptives of 4.6 per 10 000 exposed women years. Using levonorgestrel 150 µg + ethinyloestradiol 30 µg as reference, non-significant odds ratios of 1.1 (0.5–2.6) for desogestrel 150 µg + ethinyloestradiol 30 µg and 1.1 (0.5–2.4) for gestodene 75 µg + ethinyloestradiol 30 µg were found. The results of this study show no significant difference in risk between different formulations of combined oral contraceptive.

Key words: case control/oral contraceptives/pill scare/study Mediplus database/venous thromboembolic disease


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
At the end of 1995 three different studies investigating the risk of venous thromboembolic disease, reported raised odds ratios associated with the so-called third-generation oral contraceptives containing desogestrel and gestodene relative to the older (mostly levonorgestrel-based) second-generation products. (Farley et al., 1995Go; Jick et al., 1995Go, Spitzer et al., 1996Go) The adjusted odds ratios reported varied between 1.5 and 4.8 for desogestrel and between 0.9 and 5.3 for gestodene. In response to these studies the Committee on Safety of Medicines issued new prescribing guidelines to UK doctors. (CSM 1995) Since the 1995 pill scare, the use and market share of levonorgestrel products has greatly increased, accompanied by a considerable fall in the use of desogestrel- and gestodene-based products. In 1997 the results of a nested case control study based on the MediPlus database of UK general practice records were published (Farmer et al., 1997Go). Low odds ratios were found for desogestrel of 0.9 (0.4–1.8) and gestodene 0.8 (0.4–1.9) when compared to levonorgestrel-based products. A higher odds ratio was associated with desogestrel + 20 µg ethinyloestradiol [OR 2.9 (0.9–10.0)] than desogestrel + 30 µg ethinyloestradiol [OR 0.6 (0.3–1.5)] when compared to levonorgestrel 150 µg + 30 µg ethinyloestradiol. Farmer et al. (1997), the Transnational Team, the World Health Organization (WHO), and Lewis et al. (1996) all cautioned that the studies were likely to be affected by bias and confounding (Farmer et al., 1997Go; Lewis et al., 1996Go). In this paper the results of a re-examination of the MediPlus database are reported. The study spanned a different and longer time frame and involved more cases than the previous MediPlus investigation. The aim of the study was to estimate the relative risks associated with distinct formulations of combined oral contraceptive (COC).


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The investigation was conducted using the UK MediPlus database. The study window ran from the beginning of 1992 to March 1997 which was the end of the database at the time of the study. The database comprised the records of over one million women from 151 practices. The database was searched for all women who had a record of a venous thromboembolic event (VTE). A second search identified women with a record of an anticoagulant prescription or an indication of anticoagulation therapy such as international normalized ratios (inr)/prothrombin times. A third search identified those women who had been prescribed a combined oral contraceptive at any time. The records of the women common to each of these searches were printed and examined manually, by two teams.

Women were accepted as cases if they were aged between 15 and 49, had a VTE (pulmonary embolism, deep-vein thrombosis, or venous thrombosis not otherwise specified) between 1992 and 1997 which was treated with an anticoagulant and they were exposed to a combined oral contraceptive (COC) on the day of their event. To focus on idiopathic VTE, women were excluded as cases if they had had a prior VTE event, were pregnant at the time of the event or, up to 6 weeks post-partum, they had had a termination of pregnancy, surgery requiring a general anaesthetic, suffered a major trauma to the lower limbs, were using other sex hormones concurrently with their COC, or had evidence of malignant disease within 6 weeks of the VTE event. Records were also only accepted if there were at least 6 months of medical history available prior to the event date. This time period was chosen primarily so that there was a reasonable length of medical history preceding the event with which to determine COC exposure on the event date, and also so that the first signs and symptoms of a VTE event could be recorded, and so that non-idiopathic events could be spotted and excluded. The issue of a prescription for any drug or product to a patient, was deemed to represent an active patient file and therefore only those women whose first prescription on record was at least 6 months prior to the event date were accepted as cases.

Cases occurring between 1994 and 1997 were used for validation of the VTE event. Due to cost restraints, not all events were validated. The more recent years of the study (post-1994) were used because it was more likely that women who had recently had an event would still be registered with the same practice. Intercontinental Medical Statistics (IMS) requested 57 of the contributing doctors to fill in a questionnaire and provide anonymized copies of hospital notes or letters relating to the VTE event. Not all cases could be validated as some practices had previously requested that they not be sent questionnaires. The questionnaire asked whether the patient had been admitted to hospital, whether they had been treated with an anticoagulant, whether there was evidence of a prior VTE, whether they were exposed to a COC at the time of the event, and if so, to which product. The type, date and accompanying test results of the VTE event were requested. Information was also requested regarding prior illness or pregnancy. As the controls were not validated in this way, however, the information collated on cases was only used to assess the validity of the VTE event. This created an extra variable, but no other case details were altered for use in the main analysis.

Up to four controls were matched to each case by practice and year of birth. Controls also had to be exposed to a COC on the event date of their matched case. Eighty-three per cent of cases had four matched controls. Controls were subject to exactly the same inclusion and exclusion criteria as cases. For all women, information regarding body mass index (BMI), blood pressure and smoking status nearest to the event date was abstracted. The COC product to which each woman was exposed and its duration of use, the existence of concurrent disease such as thyroid disease or asthma, and the number of non-oral contraceptive-related prescriptions issued in the 6 months prior to the event were all recorded.

All prescriptions for COC on the MediPlus database between 1992 and 1997 issued to all women aged between 15 and 49 at this time were identified. The number of cycles of each formulation was calculated for each of the study years, and crude incidence rates of VTE calculated for each COC formulation in the cohort. Adjustments were made to take account of cycles prescribed pre-January 1992 that would have been used during 1992, cycles that could only have been used after July 1997, and those pills that would not have been used because women had switched their contraceptive product.

Conditional logistic regressions were carried out on the data using STATA software (STATA version 5.0, STATA Corp., TX 77840, USA).


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Cohort study
Ninety-nine cases of treated VTE who were exposed to a combined oral contraceptive were identified. There was a total of 216 356 exposed women years on the database between 1992 and 1997 for all women aged between 15 and 49. This gave a crude incidence rate for VTE, of 4.6 per 10 000 exposed women years. The incidence rates across all years are shown in Table IGo.


View this table:
[in this window]
[in a new window]
 
Table I. Crude incidence rate for venous thromboembolic event (VTE) on the MediPlus database 1992–1997a
 
Variations in the type of COC used during the study period are shown in Figure 1Go.



View larger version (16K):
[in this window]
[in a new window]
 
Figure 1. Percentage change in combined oral contraceptive (COC) use by year for all women on the MediPlus database: principal formulations.

 
Four formulations accounted for 73% of all COC use across the years 1992–1997: levonorgestrel 150 µg + ethinyloestradiol 30 µg (Microgynon 30/Ovranette, 23%), desogestrel 150 µg + ethinyloestradiol 30 µg (Marvelon, 18%), gestodene 75 µg + ethinyloestradiol 30 µg (Femodene/Minulet, 19%), and triphasic levonorgestrel + ethinyloestradiol (Logynon/Trinordiol, 13%). There was a fall of 5.4% in exposed women years on the database after the Committee on Safety of Medicines announcement (1995 compared to 1996); however, there were more considerable shifts in the proportionate use of certain formulations.

Levonorgestrel 150 µg + 30 µg ethinyloestradiol accounted for over 40% of all COC use on the database in 1997 compared to 15% in 1994. The use of the so-called third-generation products fell from 55% in 1994 to 16% in 1997.

The crude incidence rates were calculated according to exposure for those formulations where there were more than five cases. (Table IIGo) The low-oestrogen desogestrel formulation has the highest crude rate for VTE (8.6), based on nine cases. The crude incidence rates for desogestrel 150 µg + 30 µg ethinyloestradiol and gestodene 75 µg + 30 µg ethinyloestradiol are also higher than that for levonorgestrel 150 µg + 30 µg ethinyloestradiol.


View this table:
[in this window]
[in a new window]
 
Table II. Cohort analysis: exposed women years (EWY) (15–49-year-olds) to combined oral contraceptives on the Mediplus database 1992–1997
 
The average age of use on the database for all the principal formulations across all years is between 25.7 (gestodene 75 µg + 30 µg ethinyloestradiol and desogestrel 150 µg + 30 µg ethinyloestradiol) and 28.8 years (desogestrel 150 µg + 20 µg ethinyloestradiol).

As shown in Figure 2Go, the rate of VTE increased with age. Over all 99 cases, the rate was 4.6 per 10 000 exposed women years. For 15–19-year-olds the rate was 3.7 per 10 000 exposed women years, rising to 9.7 amongst 40–44-year-olds and to 25.0 in the oldest age group, 45–49.



View larger version (18K):
[in this window]
[in a new window]
 
Figure 2. Crude incidence rate of venous thromboembolic events (VTE) by age per 10 000 exposed women years (EWY) on the MediPlus database.

 
Case control study
Three hundred and sixty-six controls were matched to the 99 cases. The level of COC exposure concordance amongst the case control sets was assessed to check for the possibility of `overmatching'. This was discounted as in only one set were all the controls using the same COC formulation as the case, over half the sets had just two women using the same COC formulation and half had no controls that were using the same formulation as the case on the event day. Forty-seven cases of pulmonary embolism, 45 cases of deep-vein thrombosis, and seven cases of venous thrombosis not otherwise specified were found. Conditional logistic regressions were carried out and odds ratios (OR) with 95% confidence intervals were calculated for individual COC formulations. Table IIIGo shows the OR where there were more than five cases of VTE using levonorgestrel 150 µg + 30 µg ethinyloestradiol as the reference category.


View this table:
[in this window]
[in a new window]
 
Table III. Unadjusted and adjusted odds ratios (OR) combined oral contraceptive (COC) formulation and confounding variables
 
Both crude OR and OR adjusted for BMI category, smoking status, diastolic blood pressure and non-oral-contraceptive (non-OC) prescriptions were calculated. Adjustment modestly reduced the OR for all the formulations. No significant difference between cases and controls was found with respect to the level of diastolic blood pressure or BMI category. There was a higher proportion of non-smoking controls than cases. Around 44% of cases were current smokers compared to 25% of controls. There were 13 and 14% with unrecorded smoking status for cases and controls respectively. Using non-smokers as the reference category, being a current smoker was clearly associated with a higher risk of VTE [OR 3.1 (1.7–5.5)]. The number of non-OC/non-pregnancy-related prescriptions issued to each woman in the 6 months prior to the event date as a general health proxy measure were also calculated. It was found that 45% of cases compared to 28% of controls had been issued three or more prescriptions. [OR 2.6 (1.4–4.9)]. However, no significant difference between any of the COC products compared to levonorgestrel 150 µg + 30 µg ethinyloestradiol was found in either the unadjusted or the adjusted analysis.

Odds ratios were also calculated for COC products aggregated according to their progestogen component, as shown in Table IVGo. Levonorgestrel was used as the reference hormone and adjustment was made for the same confounding variables as in the COC formulation analysis. The same variables of three or more non-OC prescriptions [2.7 (1.5 –5.0)] and current smoking status [3.2 (1.8–5.7)] yielded significant OR. Neither the adjusted nor unadjusted (shown in Table IVGo) OR were significant.


View this table:
[in this window]
[in a new window]
 
Table IV. Adjusted odds ratios (OR) by progestogen
 
Of all cases, 72 had events occurring between 1994 and 1997. Fifteen practices had requested that they be sent no questionnaires; therefore further information was requested for 57 of these women. A response was received from 41 (72%). There was one fatality which was verified by a death certificate. For each of the remaining 40 cases the GP confirmed that the patient was treated with an anticoagulant and was taking a combined oral contraceptive at the time of the event. Thirty-nine of the 40 cases (98%) had been admitted to hospital; however, on the computer record only 26 of the 40 (65%) had an entry noting hospital admission. It was also found that whereas 24 of 41 had no record of prior pregnancy on the computer database, according to the GP only 13 were nulliparous. A total of 30 cases were verified by both the GP and hospital notes confirming that a test for VTE had been carried out and was positive.

The data were re-analysed using only those 30 cases where the event was clearly confirmed. OR were calculated for the subset of verified cases, by progestogen (rather than COC formulation) as the number of cases involved were greatly reduced. The adjusted odds ratio for desogestrel compared to levonorgestrel for the verified cases was 1.8 (0.4–7.6) compared to 1.8 (0.8–4.2) for all cases in the period 1994 to 1997. For gestodene it was 2.4 (0.5–12.5) for verified cases compared to 2.3 (1.0–5.3) for all cases.


    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The results of this study largely accord with those of the UK arms of previous studies (Farley et al., 1995Go; Spitzer et al., 1996Go; Farmer et al., 1997Go) However, in the case-control study, the VTE cases included 47% with pulmonary embolism, which is considerably higher than the hospital-based WHO case-control study (25%) (Farley et al., 1995Go), or the transnational study (Spitzer et al., 1996Go). In the current study, cases were identified on the basis of anticoagulant therapy and a clinical diagnosis, compared with proof of positive imaging of the event in the WHO and transnational studies. Thus it is possible that some pulmonary embolisms were excluded from the latter studies because of the difficulty of proving a pulmonary embolism by most imaging techniques.

This and the previous MediPlus investigation differed in a number of ways. The study window ran from 1992 to March 1997 in the current study and from 1991 to 1995 previously. Some practices had left the database and others were newly recruited. Data from 148 practices in the previous study and 151 in the current study were available. Some patient records were no longer available on the database and some new ones were. There were also methodological differences. In the current study, exposure to a COC on the event day was calculated by the length of uninterrupted COC prescriptions plus 14 days (as women would rarely start a new prescription on the day it was issued). This 14-day period was not used in the first study. The 14-day rule is likely to have had a minimal effect of perhaps capturing one or two extra cases. A record of prothrombin time testing or international normalized ratios (inr) were also considered evidence of anticoagulant treatment, which was not the case in the first study. Also cases were only accepted if they had 6 months of historical data, i.e. the event date occurred at least 6 months after the first prescription for any drug/product on record. Overall there were 85 cases in the first study and 99 cases in this study, 61 of whom occurred in both.

The crude risk of VTE across all COC was estimated at 4.6 per 10 000 women years, compared to 4.1 in the first MediPlus study. Other studies have reported rates of 4.3 and 4.2. (Gerstman et al., 1990Go; Vessey et al., 1986Go). All these figures are higher than the rate of around one per 10 000 exposed women years estimated by Farley (Farley et al., 1998Go) and 2.3 found by Jick (Jick et al., 1995Go).

Three principal contrasts to the first MediPlus investigation were found: (i) the OR for desogestrel and gestodene were higher than that for levonorgestrel; (ii) in the analysis by COC formulation, the low-oestrogen product (desogestrel 150 µg + 20 µg ethinyloestradiol) had a lower unadjusted OR of 1.3 compared to 1.5 for the desogestrel 150 µg + 30 µg ethinyloestradiol product, and the OR were the same (1.1) for both formulations after adjustment; (iii) The triphasic levonorgestrel formulation had a lower crude incidence rate of VTE and a lower OR than monophasic levonorgestrel.

The most likely explanation for the increased OR for desogestrel and gestodene compared to levonorgestrel is that when COC are aggregated according to their progestogen component, very different formulations are combined in one group. Levonorgestrel as the reference group combines levonorgestrel 150 µg + 30 µg ethinyloestradiol, which accounts for 61% of all levonorgestrel use on the database, with levonorgestrel 250 µg + 30 µg ethinyloestradiol (5%) and the triphasic levonorgestrel formulation. (34%). Desogestrel combines two formulations with 20 µg ethinyloestradiol (21% of desogestrel use) and 30 µg ethinyloestradiol (79%). Ninety-seven per cent of gestodene products used were the monophasic formulation. Levonorgestrel 250 µg + 30 µg ethinyloestradiol and the triphasic formulation are associated with lower OR [0.3 (0.0–2.8) and 0.7 (0.3–1.6)] when compared to levonorgestrel 150 µg + 30 µg ethinyloestradiol, and are used by women with different characteristics. Aggregated levonorgestrel as the reference category therefore lowers the baseline risk of levonorgestrel and subsequently increases the OR of the comparison progestogens. It is not thought appropriate to aggregate disparate formulations in this way or similarly into categories such as third and second generation, when their patterns of utilization, prescribing, and time of marketing differ.

In the previous MediPlus and transnational studies, desogestrel 150 µg + 20 µg ethinyloestradiol was found to have a higher odds ratio than desogestrel 150 µg + 30 µg ethinyloestradiol, which was ascribed to a high probability of prescribing bias. Nine cases of VTE used desogestrel 150 µg + 20 µg ethinyloestradiol (Mercilon) compared to 13 in the previous study. The rate of VTE per 10 000 exposed women years was higher at 8.6 than any other formulation, and is likely to be a reflection of the age of the users. In both MediPlus studies, 77% were over the age of 35. However, whereas in the previous study 54% were overweight (BMI 25+), just three women (33%) in this study were found to be overweight, and none was obese (BMI 30+). Overall the OR for BMI increased with increasing BMI, but were not found to be statistically significant in any of the analyses. It is further possible that as this study included 18 months of extra data spanning 1995–1997, that desogestrel 150 µg + 20 µg ethinyloestradiol has had time to stabilize at its baseline risk level since marketing in April 1988.

OR were adjusted for smoking status, non-OC prescriptions, diastolic blood pressure, and BMI category. The only variables which had a significant effect were smoking status and non-OC prescriptions issued in the 6 months prior to the event date (general health proxy). Increasing age (Figure 2Go) was found to be associated with both an increased risk of VTE and differential prescribing (Table IIGo), which stresses the importance of matching by exact year of birth.

A criticism of the first MediPlus study was that cases were not validated, that is, case status relied solely on the general practice computer recorded diagnosis of VTE (Jick et al., 1997Go, Poulter et al., 1997Go, Vandenbroucke et al., 1997Go). IMS had not previously requested GP response in this way, and in retrospect the length of the questionnaire probably contributed to the poor response rate. Information was returned on 41 out of 57 possible cases with events occurring between 1994 and 1997. Thirty-nine women were admitted to hospital, yet only 26 of these (65%) actually had a record of hospital admission on the computer database. This is a similar proportion to the total number of cases in the previous MediPlus study, where there was a database record of hospital admission in 75% of cases. This supports our previous assertion that a record of hospital admission on a GP database is not an appropriate way of identifying cases of venous thromboembolic disease, because it is likely to lead to an under-recording of cases. Hospital admission on the database was not a requirement in any of our studies for case-inclusion. The database proved remarkably accurate, however, with regard to the diagnostic and prescribing data, although there was an under-recording of historical data such as prior pregnancy. Prior pregnancy is an important potential confounding variable, which shows that some important variables cannot always be adjusted for adequately in database studies. In common with previous studies, the investigation was observational, with no randomization of exposure and therefore cannot be free from bias and confounding.

The OR of the validated cases were compared to all cases occurring between 1994 and 1997. There was no difference between the OR, although the confidence intervals were wider, due to the reduced number of cases. It would seem reasonable to suppose that the verified cases have similar characteristics to the complete set of identified cases. The OR were higher, however, using cases in the period 1994–1997 than when all cases occurring between 1992 and 1997 were used. It is likely that the `pill scare' in 1995 distorts the analysis, making the results difficult to interpret.

This study does not support the hypothesis that there are significant differences in risk between different COC formulations.


    Acknowledgments
 
The study was funded by an unconditional grant from NV Organon and Schering AG.


    Notes
 
3 To whom correspondence should be addressed Back


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Committee on Safety of Medicines. (1995) Combined Oral Contraceptives and Thromboembolism. Committee on Safety of Medicines, London.

Farley, T.M.M., Meirik, O., Chang, I. et al. (1995) 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, 346, 1582–1588.[ISI][Medline]

Farley, T.M.M., Meirik, O., Chang, C., and Poulter, N.R. (1998) Combined oral contraceptives, smoking, and cardiovascular risk. J. Epidemiol Community Health, 52, 775–785[Abstract]

Farmer, R.D.T., Lawrenson, R.A., Thompson, C.R. et al. (1997) Population-based study of risk of venous thromboembolism associated with various oral contraceptives. Lancet, 349, 83–88.[ISI][Medline]

Gerstman, B.B., Piper, J.M., Freiman, J.P. et al. (1990) Oral contraceptive oestrogen and progestin potencies and the incidence of venous thromboembolism. Int. J. Epidemiol., 19, 931–936.[Abstract]

Jick, H., Jick, S.S., Gurewich, V. et al. (1995) Risk of idiopathic cardiovascular death and nonfatal venous thromboembolism in women using oral contraceptives with differing progestagen components. Lancet, 346, 1589–1593.[ISI][Medline]

Jick, H., Jick, S.S., Myers, M.W. and Vasilakis, C. (1997) Third-generation oral contraceptives and venous thrombosis [letter; comment]. Lancet, 349, 731–732.

Lewis, M.A., Heinemann, L.A.J., MacRae, K.D. et al. (1996) The increased risk of venous thromboembolism and the use of third generation progestogens: role of bias in observational research. Contraception, 54, 5–13.[ISI][Medline]

Poulter, N.R., Chang, C.L., Marmot, M. et al. (1997) Third-generation oral contraceptives and venous thrombosis [letter; comment]. Lancet, 349, 732.[ISI][Medline]

Spitzer, W.O., Lewis, M.A., Heinemann, L.A. et al. (1996) Third generation oral contraceptives and risk of venous thromboembolic disorders: an international case-control study. Transnational Research Group on Oral Contraceptives and the Health of Young Women. Br. Med. J., 312, 83–88.[Abstract/Free Full Text]

Vandenbroucke, J.P., Helmerhorst, F.M., Bloemenkamp, K.W. and Rosendaal, F.R. (1997) Third-generation oral contraceptives and venous thrombosis [letter; comment]. Lancet, 349, 731.

Vessey, M., Mant, D., Smith, A. and Yeates, D. (1986) Oral contraceptives and venous thromboembolism: findings in a large prospective study. Br. Med. J. Clin. Res. Ed., 292, 526.[Medline]

Submitted on November 20, 1998; accepted on March 8, 1999.