The risk of venous thromboembolism in women prescribed cyproterone acetate in combination with ethinyl estradiol: a nested cohort analysis and case–control study

H.E. Seaman1, C.S.de Vries and R.D.T. Farmer

Department of Pharmacoepidemiology, Postgraduate Medical School (University of Surrey), Stirling House, The Surrey Research Park, Guildford GU2 7DJ, UK

1 To whom correspondence should be addressed. e-mail: h.seaman{at}surrey.ac.uk


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
BACKGROUND: Cyproterone acetate combined with ethinyl estradiol (CPA/EE) is licensed in the UK for the treatment of women with acne and hirsutism and is also a treatment option for polycystic ovary syndrome (PCOS). Previous studies have demonstrated an increased risk of venous thromboembolism (VTE) associated with CPA/EE compared with conventional combined oral contraceptives (COCs). We believe the results of those studies may have been affected by residual confounding. METHODS: Using the General Practice Research Database we conducted a cohort analysis and case–control study nested within a population of women aged between 15 and 39 years with acne, hirsutism or PCOS to estimate the risk of VTE associated with CPA/EE. RESULTS: The age-adjusted incidence rate ratio for CPA/EE versus conventional COCs was 2.20 [95% confidence interval (CI) 1.35–3.58]. Using as the reference group women who were not using oral contraception, had no recent pregnancy or menopausal symptoms, the case–control analysis gave an adjusted odds ratio (ORadj) of 7.44 (95% CI 3.67–15.08) for CPA/EE use compared with an ORadj of 2.58 (95% CI 1.60–4.18) for use of conventional COCs. CONCLUSIONS: We have demonstrated an increased risk of VTE associated with the use of CPA/EE in women with acne, hirsutism or PCOS although residual confounding by indication cannot be excluded.

Key words: acne/cyproterone acetate/ethinyl estradiol/polycystic ovary syndrome/venous thrombosis


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Acne is generally associated with the onset of puberty and increased sebum production from the hair follicles on the face, chest and back. Propionibacterium acnes may proliferate within blocked pilosebaceous units and cause comedones and inflamed pustules. Sometimes acne is provoked by prolonged exposure to oral steroids. Conventional treatments for acne act by reducing sebum production, unblocking pores, delivering antibiotics or suppressing the effects of androgens. The anti-androgen cyproterone acetate (2 mg) combined with ethinyl estradiol (35 µg) (CPA/EE) is licensed in the UK for the treatment of women with severe acne that is refractory to prolonged oral antibiotic therapy and for moderately severe hirsutism. Although it is not licensed as an oral contraceptive (OC) in the UK, CPA/EE is sometimes prescribed for contraception. Acne and hirsutism are associated with polycystic ovary syndrome (PCOS) as are obesity, dyslipidaemia and insulin resistance, independent of obesity (Taylor, 1998Go; Solomon, 1999Go; Atiomo et al., 2000Go; Kelly et al., 2000Go). Thus women with PCOS have an intrinsically adverse cardiovascular risk profile, as demonstrated by a higher prevalence of cerebrovascular and cardiovascular morbidity (Wild et al., 2000Go) and a positive family history of thrombotic disease in women with PCOS (Atiomo et al., 2000Go).

Recent observational studies have suggested that CPA/EE may be associated with an increased risk of venous thromboembolism (VTE) compared with conventional combined oral contraceptives (COCs) (Farmer et al., 1999Go; Parkin et al., 2000Go; Vasilakis-Scaramozza and Jick, 2001Go; Lidegaard et al., 2002Go). The use of CPA/EE in the UK more than doubled following the ‘pill scare’ of 1995, possibly because women with androgenic skin conditions were switched to CPA/EE from COCs containing less androgenic progestogens (desogestrel and gestodene).

Because the VTE risk profile of women who have androgenic skin conditions (or PCOS) differs from that of other women of the same age, we conducted a cohort analysis and a case–control study to estimate the risk of VTE associated with CPA/EE in such women.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Our study used data from the General Practice Research Database (GPRD). The GPRD contains anonymous patient records for ~4% of the UK general population. Routine demographic information is recorded. Medical diagnoses are coded to an OXMIS (Oxford Medical Indexing System) or Read code. Prescribing information is ~95% complete (Walley and Mantgani, 1997Go). Other information relating to smoking habits, alcohol consumption, height and weight are incompletely recorded. The validity and utility of this database for studies of drug safety have been published elsewhere (Van Staa and Abenhaim, 1994Go; Hollowell, 1997Go; Walley and Mantgani, 1997Go; Lawrenson et al., 2000Go).

The study population was drawn from a cohort of women who had ever had a diagnosis of acne, hirsutism or PCOS. Women with PCOS were identified as those with an unequivocal diagnosis of PCOS or the identification of polycystic ovaries by ultrasound examination, and also any women with three or more markers for PCOS (acne, hirsutism/alopecia, obesity, anovulation/infertility, amenorrhoea/ oligomenorrhoea or endocrinological measures). From that population we selected women who at any time between January 1, 1992 and December 31, 1998 were aged 15–39 years. Women who had a VTE (pulmonary embolism or deep vein thrombosis) confirmed by evidence of anticoagulant therapy were identified and the ‘event date’ taken as the date on which the first symptoms were recorded. These symptoms included haemoptysis, shortness of breath, chest pain and swelling or redness of a limb. Prescriptions for conventional OCs and CPA/EE were identified and used to map drug use to each day in the study period (Farmer et al., 1999Go). We were thus able to ascertain for each case her OC exposure status on the VTE event date. Total exposure to CPA/EE and conventional OC products in terms of exposed women years was used to calculate rates of VTE stratified by age.

For the case–control study, up to seven controls were randomly selected from the study population matched to the case by event (index) date, general practice and year of birth. The GPRD records for cases and controls were reviewed and relevant medical and drug exposure information recorded. Women with chronic illness, which included diabetes, asthma, systemic lupus erythematosus, Crohn’s disease/ulcerative colitis and thyroid and renal disease, were identified. The use of conventional OCs and CPA/EE amongst controls was described with reference to the index date as before. The period of uninterrupted exposure prior to the event/index date for each case and control exposed to a conventional COC or CPA/EE was calculated. Also, because it appeared that the use of CPA/EE in the UK increased considerably following the ‘pill scare’ of 1995 (Seaman et al., 2003Go), we partitioned the study population according to whether the event/index date was prior to or after 1995.

Differences in the length of time cases and controls were registered with practices contributing to the GPRD could have biased our observations with regard to assignation of acne status. In order to address this potential confounding we performed a sub-group analysis including only those cases and controls with at least 2 years (760 days) registration after the index date and at least 1 years (365 days) registration prior to the index date.

Conditional logistic regression models were built to provide risk estimates for VTE associated with CPA/EE adjusted for potential confounding.


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
A total of 179 women had a VTE during the study period; none were fatal. Of these, 23 (13%) were exposed to CPA/EE on the index date giving a crude VTE incidence rate of 8.05 cases/104 exposed women years. The rate of VTE increased with increasing age. The age-adjusted incidence rate ratio for CPA/EE versus conventional COCs was 2.20 [95% confidence interval (CI) 1.35–3.58] (Table I).


View this table:
[in this window]
[in a new window]
 
Table I. Crude rates and incidence rate ratio (IRR) for VTE (1992–1998)
 
The 179 cases were matched to 1076 controls. Characteristics of cases and controls are shown in Table II. Cases were more likely to have a history of VTE and to have visited the general practitioner more frequently in the 6 months prior to the index date. Pregnancy, especially the intra/postnatal period or post termination of pregnancy (TOP), chronic illness, recent surgery or trauma and use of oral steroids were all significantly associated with VTE. Using as the reference group women who were not using COCs and who had no recent pregnancy or menopausal symptoms, exposure to conventional COCs gave a crude OR of 2.57 (95% CI 1.69–3.91) compared with 6.86 (95% CI 3.71–12.69) for CPA/EE use. Cases and controls were similar in terms of body mass index (BMI) and smoking status. Amongst women in our study population who had been using conventional COCs on the index date, 90% had been using a conventional COC containing the same or a smaller dose of ethinyl estradiol as contained in CPA/EE (35 µg). The remaining 10% had been using triphasic products.


View this table:
[in this window]
[in a new window]
 
Table II. Characteristics of cases and controls
 
Information on duration of exposure in those women exposed to a conventional COC or CPA/EE is presented in Table III. In women exposed to a conventional COC, cases were significantly more likely than controls to be exposed for <=6 cycles prior to the index date. That relationship was not demonstrated in women exposed to CPA/EE. A sub-group analysis of CPA/EE-exposed cases and controls who, according to the information available, had not been exposed to a conventional COC at any time prior to the index date included 13 cases (six exposed for <=6 cycles and seven exposed for >6 cycles) and 16 controls (six exposed for <=6 cycles and 10 exposed for >6 cycles). With reference to an exposure of >6 cycles, the age-adjusted OR for VTE was 1.36 (95% CI 0.26–7.17).


View this table:
[in this window]
[in a new window]
 
Table III. Duration of exposure in women exposed to a conventional COC or CPA/EE
 
Stepwise conditional logistic regression was used to identify variables for the final conditional logistic model (Table IV). The strongest predictors of VTE were recent surgery or trauma and the intra/postnatal period or post-TOP. Using as the reference group women who were not using oral contraception and who had no recent pregnancy or menopausal symptoms, use of CPA/EE gave an adjusted odds ratio (ORadj) of 7.44 (95% CI 3.67–15.08) compared with an ORadj of 2.58 (95% CI 1.60–4.18) for users of conventional COCs.


View this table:
[in this window]
[in a new window]
 
Table IV. Conditional multivariate logistic regression (number of observations = 1255)
 
Our sub-group analysis identified 82 cases and 457 matched controls with at least 365 days registration time prior to the index date and 760 days registration time after the index date. A total of 77 cases (94%) had at least four matched controls. Stepwise logistic regression was used to identify variables for sub-group conditional logistic regression analysis (Table V). Recent surgery or trauma, a prior VTE, the intra- and postnatal period of pregnancy or post-TOP and increasing numbers of consultation with the general practitioner remained significantly associated with VTE. Exposure to conventional COCs gave a significantly raised OR of 2.26 (95% CI 1.13–4.56) and the OR for exposure to CPA/EE was significantly raised to 3.87 (1.13–13.30). Using exposure to conventional COCs as the reference category, further analysis revealed that the point estimates for CPA/EE and conventional COCs were not significantly different [OR 1.71; (95% CI 0.49–5.96)].


View this table:
[in this window]
[in a new window]
 
Table V. Conditional multivariate logistic regression restricted to cases and controls with at least 2 years GPRD registration after the index date and at least 1 year prior to the index (number of observations = 539)
 
There were 86 cases of VTE prior to 1995 and 62 cases after 1995. Conditional logistic regression analysis revealed no difference in the risk estimates for VTE associated with CPA/EE prior to, or after, the 1995 ‘pill scare’ (OR 7.22; 95% CI 2.10–24.82 versus 7.27; 95% CI 2.53–20.87).


    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Our study population comprised a group of women who were likely to have an increased risk of VTE. That risk appeared to be amplified by exposure to CPA/EE. Previous work on CPA/EE and VTE risk has compared the risk in users of CPA/EE with that of conventional COC users (Farmer et al., 1999Go; Parkin et al., 2000Go; Vasilakis-Scaramozza and Jick, 2001Go; Lidegaard et al., 2002Go) although in only one paper was the risk estimate adjusted for acne or PCOS (Vasilakis-Scaramozza and Jick, 2001Go). In most countries, CPA/EE is not licensed as an oral contraceptive. Thus, we believe it is appropriate to consider the adverse event risk profile associated with CPA/EE alongside that associated with other acne therapies. The adverse reactions associated with topical acne treatments include local inflammation and occasionally photosensitivity. The tetracyclines have been associated with vestibular disturbances, hyperpigmentation and photosensitivity and, more seriously, lupus-like reactions and hepatitis. The treatment options for women whose acne proves non-responsive to topical or oral antibiotic treatment include CPA/EE and oral retinoids, although in the UK the use of isotretinoin requires specialist dermatological supervision (British National Formulary, 2001Go). Isotretinoin is teratogenic and in women of childbearing age concomitant use of a combined estrogen-progesterone oral contraceptive is recommended. The potential side effects of isotretinoin use include hypervitaminosis A-like symptoms, benign intracranial hypertension, alopecia, liver disease, elevated serum cholesterol levels, photosensitivity and pancreatitis (British National Formulary, 2001Go). More recently spontaneous reports linking the use of isotretinoin with severe depression and suicide attempts have been cited (Wysowski et al., 2001Go) although there is currently no epidemiological evidence to support a causal association (Jick et al., 2000Go).

We have demonstrated an increased risk of VTE associated with the use of CPA/EE in women with acne, hirsutism or PCOS. The estrogen component of the CPA/EE formulation in current use and used by women in our study barely exceeds that of the majority of conventional COCs. To our knowledge no published research has associated CPA with VTE and as far as we are aware there is no biological plausibility for any such relationship. The ‘depletion of susceptibles’ principle maintains that patients who remain on a drug are those who can tolerate it whilst those who cannot are selected out of the population at risk (Yola and Abenhaim, 1994Go). Thus risk per unit of time depends both on history of prior exposure and on duration of exposure. Whilst mindful of the limitations imposed by censoring of data held on the database (a patient’s medical and drug exposure experience prior to or after periods of registration with a contributing practice are unknown) and the small numbers of exposed cases, we found that, compared with long-term use (>6 cycles), there seemed to be an increased risk for VTE in short term users (<=6 cycles) of conventional COCs but not users of CPA/EE. Thus we did not find evidence to support the possibility that the increased VTE risk in women using CPA/EE was associated with depletion of susceptibles, even when we restricted the analysis to CPA/EE-exposed cases and controls who had no evidence of prior oral contraceptive use. The growth of the GPRD in longitudinal data will allow us to explore further the duration of exposure issue in future work.

The observation that the use of CPA/EE in the UK increased considerably following the ‘pill scare’ of 1995 should be considered in an assessment of VTE risk in women prescribed CPA/EE. As part of the analysis, therefore, the data were partitioned according to whether the VTE occurred prior to or after 1995—conditional logistic regression analysis revealed no difference in the risk estimates for VTE associated with CPA/EE before or after the ‘pill scare’.

Women with an acne diagnosis at any time in the medical records contributed to our study population. Thus it was possible that if the time registered with practices contributing to the GPRD differed for cases and controls a bias could be introduced into the analysis with regard to acne status. We have described an ‘acne period’ for patients with acne diagnoses on the GPRD (Seaman et al., 2003Go). That theoretical period began 2 years prior to a patient’s first acne diagnosis and ended 1 year after the last acne diagnosis. Based on that reasoning, to reliably assign acne status to cases and controls we performed a sub-group analysis including only those cases and controls with at least 2 years (760 days) registration after the index date and at least 1 years (365 days) registration prior to the index date. The results of the sub-group conditional logistic regression indicated a statistically significant association between VTE and exposure to CPA/EE (OR 3.87) and exposure to conventional COCs (OR 2.26). Thus it appears that at least part of the significantly elevated risk estimate for VTE in women prescribed CPA/EE may be accounted for by residual confounding with regard to acne status. The difference in risk between CPA/EE and conventional COCs was not significant, unlikely to be associated with ethinyl estradiol dose and could be a chance finding. The age-adjusted IRR for VTE associated with CPA/EE use versus use of conventional COCs was 2.2 and after adjustment in the sub-group case–control analysis this ratio was reduced to 1.7. Thus if complete adjustment for confounding could be achieved it is possible that no difference in risk between conventional COCs and CPA/EE would remain.

Our study population comprised women with acne, hirsutism or PCOS. It is possible that the morbidity associated with more severe symptoms in such women may include an elevated VTE risk. We cannot determine disease severity using routinely collected data such as that available on the GPRD and further work is planned to explore the underlying risk profiles of CPA/EE users.

In conclusion, it is likely that confounding by disease severity is an important component of the elevated VTE risk associated with CPA/EE exposure in women with acne, hirsutism or PCOS. Our findings suggest that CPA/EE remains a viable treatment option for women with hirsutism or acne which has proved unresponsive to other drugs. In addition the findings of this study suggest that, compared with other estrogen-containing products of a similar dose, CPA/EE does not increase a woman’s baseline VTE risk. Careful consideration should be given to the risks associated with CPA/EE versus other treatment options.


    Acknowledgements
 
This study was supported by an unconditional research grant from Schering AG. The company was not involved in the investigations at any point. The investigation was approved by the Scientific and Ethical Advisory Group (SEAG) for the GPRD.


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Atiomo, W.U., Fox, R., Condon, J.E., Shaw, S., Friend, J., Prentice, A.G. and Wilkin, T.J. (2000) Raised plasminogen activator inhibitor-1 (PAI-1) is not an independent risk factor in the polycystic ovary syndrome (PCOS). Clin. Endocrinol., 52, 487–492.[CrossRef][ISI][Medline]

British National Formulary (2001) British Medical Association and Royal Pharmaceutical Society of Great Britain. British National Formulary, 42. BMA and The Pharmaceutical Society of Great Britain, London, UK.

Farmer, R.D.T., Lawrenson, R.A., Todd, J-C., Williams, T.J. and MacRae, K. (1999) Oral contraceptives and venous thromboembolic disease. Analysis of the UK General Practice Research Database and the UK MediPlus Database. Hum. Reprod. Update, 5, 688–706.[Abstract/Free Full Text]

Hollowell, J. (1997) The General Practice Research Database: quality of morbidity data. Popul. Trends, 87, 36–40.[Medline]

Jick, S.S., Kremers, H.M. and Vasilakis-Scaramozza, C. (2000) Isotretinoin use and risk of depression, psychotic symptoms, suicide, and attempted suicide. Arch. Dermatol., 136, 1231–1236.[Abstract/Free Full Text]

Kelly, C.J., Connell, J.M., Cameron, I.T., Gould, G.W. and Lyall, H. (2000) The long term health consequences of polycystic ovary syndrome. Br. J. Obstet. Gynecol., 107, 1327–1338.[ISI]

Lawrenson, R., Todd, J.C., Leydon, G.M., Williams, T.J. and Farmer, R.D. (2000) Validation of the diagnosis of venous thromboembolism in general practice database studies. Br. J. Clin. Pharmacol., 49, 591–596.[CrossRef]

Lidegaard, O., Edstrom, B. and Kreiner, S. (2002) Oral contraceptives and venous thromboembolism: a five-year national case-control study. Contraception, 65, 187–196.[CrossRef][ISI][Medline]

Parkin, L., Skegg, D.C., Wilson, M., Herbison, G.P. and Paul, C. (2000) Oral contraceptives and fatal pulmonary embolism. Lancet, 355, 2133–2134.[CrossRef][ISI][Medline]

Seaman, H.E., de Vries, C.S. and Farmer, R.D.T. (2003) Differences in the use of combined oral contraceptives amongst women with and without acne. Hum. Reprod., 18, 515–521.[Abstract/Free Full Text]

Solomon, C.G. (1999) The epidemiology of polycystic ovary syndrome. Prevalence and associated disease risks. J. Endocrinol. Metab. Clin. North Am., 28, 247–263.[ISI][Medline]

Taylor, A.E. (1998) Understanding the underlying metabolic abnormalities of polycystic ovary syndrome and their implications. Am. J. Obstet. Gynecol., 179, S94-S100.

Van Staa, T. and Abenhaim, L. (1994) The quality of information on a UK database of primary care records: a study of hospitalizations due to hypoglycaemia and other conditions. Pharmacoepidemiol. Drug Safety, 3, 15–21.

Vasilakis-Scaramozza, C. and Jick, H. (2001) Risk of venous thromboembolism with cyproterone or levonorgestrel contraceptives. Lancet, 358, 1427–1429.[CrossRef][ISI][Medline]

Walley, T. and Mantgani, A. (1997) The UK General Practice Research Database. Lancet, 350, 1097–1099.[CrossRef][ISI][Medline]

Wild, S., Pierpoint, T., McKeigue, P. and Jacobs, H. (2000) Cardiovascular disease in women with polycystic ovary syndrome at long-term follow-up: a retrospective cohort study. Clin. Endocrinol., 52, 595–600.[CrossRef][ISI][Medline]

Wysowski, D.K., Pitts, M. and Beitz, J. (2001) An analysis of reports of depression and suicide in patients treated with isotretinoin. J. Am. Acad. Dermatol., 45, 515–519.[CrossRef][ISI][Medline]

Yola, M. and Abenhaim, L. (1994) Evidence of the depletion of susceptibles effect in non-experimental pharmacoepidemiologic research. J. Clin. Epidemiol., 47, 731–737.[ISI][Medline]

Submitted on September 3, 2002; accepted on November 21, 2002.