Introduction

Epidemiology is the study of the distribution and determinants of health related states or events within a specified population,¹ its purpose being to inform decisions about the control of health problems. This population perspective usually results in epidemiologists being interested in looking at overall, average effects within the groups under investigation. In a randomised trial this would be the average effect (usually benefit) of a treatment, and in an observational study it would be the average effect (often risk) associated with a factor. Subgroup analyses looking at effects in individuals with specific characteristics tend to be treated with circumspection, even in systematic reviews, in which the quantity of data is greater.² Clinicians, on the other hand, are not interested in average effects; they need information about specific risks and benefits faced by the individual patients consulting them. Difficulties arise when clinicians try to use epidemiological data to guide clinical decisions.

View larger version (96K): [in this window] [in a new window]   Full details of references w1-w43 appear on our website

In October 1995 the Medicine Control Agency of the United Kingdom announced that new epidemiological data indicated that users of certain brands of combined oral contraceptives might have a higher risk of venous thromboembolism than women using other types of combined contraceptive pill. At the same time, tentative evidence was emerging that the contraceptive pills associated with an increased risk of venous thrombosis might have a lower risk of myocardial infarction. In the following few weeks, thousands of users of combined oral contraceptives attended their doctors for information about their particular risk of cardiovascular disease. For most women, the specific clinical question that needed an answer was: "What is the risk of cardiovascular disease among users of currently available, low dose, combined oral contraceptives who are aged less than 35 years, do not smoke, and do not have a medical condition known to increase the risk of vascular disease?" In order to see if this question could be answered by available epidemiological data, we reviewed all published studies of cardiovascular disease in users of combined oral contraceptives, adopting an approach similar to that of a systematic reviewer.

	Methods

Identification of studies
We identified studies from papers in the extensive library of reprints held by the Royal College of General Practitioners' Manchester Research Unit, by searching the reference lists of each paper, and by conducting a computerised literature search of Medline. We considered papers that provided a risk estimate for current use of combined oral contraception (or with sufficient raw data to enable us to calculate it). We were not interested in studies which examined only former users since it is generally agreed that the risk of cardiovascular disease associated with combined oral contraceptives is confined to current users.³ We used only the most recently published report from each study unless an earlier paper contained information that was not available in the later report.

Selection of studies
We both assessed the identified papers using the following inclusion criteria.

• Did the study examine currently available combined oral contraceptives?
• Did the study examine apparently healthy women?
• Did the study compare users with non-users?

• Did the studies of stroke and myocardial infarction collect information about smoking?
• Were data for healthy women presented specifically for low dose formulations?

	Results

Our search identified 74 papers about the relation between current use of combined oral contraceptives and cardiovascular disease, some of which reported on more than one vascular outcome. Thus, 23 reported on venous thromboembolism,^{4-10 w1-w16} 22 on ischaemic stroke,^{11-16 w1-w4 w17-w28} 13 on haemorrhagic stroke or subarachnoid haemorrhage, ^{12 16-20 w1 w12 w18 w22 w23 w25 w27} 13 on all stroke, ^{12 17 21 w9 w13 w18 w26 w29-w33} and 33 on myocardial infarction. ^{17 22-34 w1-w3 w6 w9 w11-w13 w23 w34-w43}

Selection criteria

Did study examine currently available combined oral contraceptives?
During the past three decades there have been major changes in the composition of combined oral contraceptives and the characteristics of women using them.³⁵ In view of these changes, we decided to include only studies that completed data collection after 1980 unless an earlier study supplied data about the risk associated with low dose products (that is, those containing <50 µg oestrogen). Of the 74 studies, 28 failed to meet this criterion. ^{16-19 22-26 w1-w10 w18 w23 w30 w31 w34-w38} Another study was conducted between 1980 and 1982, but 65% of its periods of observation related to use of combined oral contraceptives containing 50 µg oestrogen, and the authors did not provide separate risk estimates for lower dose preparations.^w11 We therefore also excluded this paper.

Did study examine apparently healthy women?
Sometimes it was difficult to determine the characteristics of subjects in a study. If a study of venous thromboembolism was reported to have excluded events that occurred in women with a history of this problem or during or soon after surgery or pregnancy, we assumed that it studied an apparently healthy group of women. Failure to make these exclusions in any analyses resulted in our rejecting five papers from our assessment of the risk of venous thromboembolism. ^{6 7 w12-w14} Similarly, we rejected 15 studies of stroke for failing to exclude events occurring in women with a history of stroke, other arterial disease, hypertension, or diabetes mellitus, ^{12-15 20 w17 w19 w20 w22 w24-w29} and we rejected nine studies of myocardial infarction for failing to exclude events occurring in women with a history of this condition, other arterial disease, hypertension, or diabetes mellitus. ^{28-30 32 33 w12 w40-w42}

Did study compare users with non-users?
Four studies were excluded because they lacked an adequate control group.^{21 w15 w21 w32} Another paper provided estimates of the risk of venous thromboembolism in users of different combined oral contraceptives but did not have a comparison group of non-users.^w16 Since it was uninformative about whether users of these preparations have a different risk to that of non-users, we rejected it.

Did studies of stroke and myocardial infarction collect information about smoking?
This was a particular concern because smoking is an important confounder of the relation between the risk of arterial disease and the use of combined oral contraceptives. We rejected one paper that used data from a cohort of women in the group health cooperative of Puget Sound and was unable to collect any information about smoking.^w40

Were data presented specifically for low dose formulations?
Four papers provided risk estimates for healthy women using combined oral contraceptives of any dose. ^{6 31 34 w44} None reported separate results for use of low dose preparations and so were unable to address our question.

Eligible studies
After excluding the rejected studies, we were left with seven papers; five relating to risks of venous thrombosis, ^{4 5 8-10} one to risks of stroke,¹¹ and one to risks of myocardial infarction.²⁷

Venous thrombosis
There was reasonable evidence that currently available combined oral contraceptives are associated with an increased risk of venous thrombosis in healthy users (table). The more recent studies included many events, so the risk estimates had reasonably tight 95% confidence intervals. The World Health Organisation's study reported separate risk estimates for venous thromboembolism in healthy users living in different geographical areas (Europe and developing countries) and in separate age groups (<35 and 35).⁸ Information was also given about the risk in women who smoked, who were overweight, and who had a history of hypertension in pregnancy, but only as an overall risk among users of any type of combined contraceptive pill, not separately for low dose preparations. Three studies provided separate risk estimates associated with low dose formulations containing specific progestogens.^8-10

Two cohort studies provided data that could be used to estimate the incidence of venous thromboembolic disease in healthy users of low dose combined oral contraceptives and in non-users. ^{5 9} Vessey et al found that the crude incidence of possible, probable, or certain deep vein thrombosis or pulmonary embolism was 12.2/10⁵ woman-years in non-users (never and past users combined) compared with 39.4/10⁵ woman-years in users of low dose preparations.⁵ Jick et al estimated that the crude incidence of venous thrombosis was 3.8/10⁵ woman-years in past users and 23.2/10⁵ woman-years in current users of any low dose combined contraceptive pill.⁹

Stroke
The one eligible study of ischaemic stroke found a significant near doubling of risk of disease among current users of combined oral contraceptives compared with never users.¹¹ Although data about smoking were collected, the author did not provide separate risk estimates for healthy women using low dose combined contraceptives who smoked and for those who did not smoke. Even this study, therefore, did not answer our specific clinical question.

Myocardial infarction
The eligible study of myocardial infarction included only eight women who were using a low dose combined oral contraceptive at the time of their infarction.²⁷ Furthermore, separate risk estimates were not provided for users who smoked and those who did not. This meant that we did not find any studies of myocardial infarction which addressed our specific clinical question.

	Discussion

Our literature search found many studies of the risk of vascular disease in current users of combined oral contraceptives. Much of the information, however, was concerned with the effects of combined oral contraceptives which are no longer available, or was derived from studies with serious methodological problems. Many of the studies used statistical techniques such as multivariate analysis to control for the effects of factors (confounders) that might be alternative explanations for a study's findings. For instance, many studies adjusted for differences in the proportion of smokers among users and non-users of oral contraceptives. In effect, these adjustments level the epidemiological playing field so that the real effects of combined oral contraceptives can be determined, but at a cost of losing information about the effects of the adjusting factor (in this case smoking) among contraceptive users.

In order to determine the effects of oral contraceptives in women with particular characteristics, populations need to be divided into their various subgroups. This can be done when designing a study, by defining which women will be recruited from the population pool of contraceptive users into the study (for example, only healthy users). Alternatively, it can be done at the time of analysis by stratifying the study population into users with different characteristics.

In statistical terms stratification is less efficient than multivariate analysis, but it does allow the effects of oral contraceptives to be observed in different users. With reanalysis using stratification, one of the rejected papers might have provided more information about the risk of venous thromboembolism among healthy users of low dose combined oral contraceptives,⁷ five papers might have given more information on the risk of stroke, ^{12-15 20} and five might have given more information on the risk of myocardial infarction. ^{28-30 32 33} Another three papers examined healthy women but did not provide specific risk estimates for low dose combined oral contraceptives, even though these preparations were used by almost all the users of combined oral contraceptives in the studies. ^{6 31 34}

Our clinical example suggests that clinicians who wish to confine themselves to studies conducted in populations that closely represent their practice population will rarely find many studies to guide their practice. A similar situation exists with much experimental research. The paucity of information requires clinicians to extrapolate results from study populations that do not closely match their practice population. Such extrapolations usually imply an equal distribution of risk across the population; an assumption which may be wrong. For example, in several of the studies reviewed the risk of myocardial infarction or stroke was found to be concentrated in users of combined oral contraceptives with other risk factors for vascular disease, notably smoking, ^{12 14 16 17 20-26 30 32} and users with a history of hypertension. ^{11 12 14 16 18-20 24 30} This means that the estimate of overall risk for arterial disease among all users of combined oral contraceptives grossly exaggerates the risk among healthy non-smoking users.

In order to avoid making any assumptions about the distribution of risk, we need empirical data about the risk of cardiovascular disease in healthy users of combined oral contraceptives. This has been difficult to obtain because cardiovascular disease is uncommon in young women. Thus, even studies with large catchment areas and prolonged periods for recruitment have difficulty recruiting a large number of current users experiencing vascular problems, particularly arterial ones. At present, however, little of the epidemiological data about the risk of cardiovascular disease in users of combined oral contraceptives is of use to clinicians, although more could become available if the some data were reanalysed. This situation is unlikely to be unique to oral contraceptives.