Risk of venous thromboembolism from oral contraceptives containing gestodene and desogestrel versus levonorgestrel: a meta-analysis and formal sensitivity analysis

Abstract

Controversy exists regarding whether oral contraceptives (OCs) containing desogestrel and gestodene are associated with an increased risk of venous thromboembolism (VTE) versus OCs containing levonorgestrel. We were interested in synthesizing the available data, exploring explanations for mixed results, and characterizing the degree of uncontrolled confounding that could have produced a spurious association.

We performed a meta-analysis and formal sensitivity analysis of studies that examined the relative risk of VTE for desogestrel and gestodene versus levonorgestrel. Twelve studies, all observational, were included. The summary relative risk (95% CI) was 1.7 (1.3–2.1; heterogeneity p = 0.09). If real, the incremental risk of VTE would be about 11 per 100,000 women per year. An association was present when accounting for duration of use and when restricted to the first year of use in new users. However, in the sensitivity analysis, the association abated in many, but not all, scenarios in which an unmeasured confounding factor increased the risk of VTE three to fivefold and in nearly all examined scenarios in which the factor increased the risk 10-fold.

The summary relative risk of 1.7 does not appear to be caused by depletion of susceptibles, but is sensitive to a modest degree of unmeasured confounding. Whether such confounding occurred is unknown. However, given this sensitivity, this issue probably cannot be settled unequivocally with observational data. In the absence of a definitive answer, this apparent increased risk, together with its uncertainty and small magnitude and its important consequences, should be considered when selecting an OC for a given woman.

Introduction

Oral contraceptives (OCs) have been used since the 1960s and are currently taken by approximately 100 million women worldwide [1]. Desogestrel- and gestodene-containing OCs (sometimes called third generation OCs) were introduced in Europe in 1981 and 1987, respectively [2], [3]. In 1995, data began to emerge from observational studies suggesting that desogestrel and gestodene may be associated with a higher risk of pulmonary embolism and deep vein thrombosis (collectively referred to as venous thromboembolism or VTE) than were OCs containing levonorgestrel with comparable amounts of ethinyl estradiol (EE). However, as additional data and re-analyses emerged, these results were not always replicated.

The likelihood of a cause-effect relationship has been the subject of protracted debate. As summarized elsewhere [4], [5], [6], the most plausible arguments that the observed associations are spurious take two related forms. The first is that the association is due to unmeasured confounding. Confounding is the presence of a factor that is associated both with the exposure of interest (in this case, the selection of an OC containing desogestrel or gestodene rather than levonorgestrel) and with the outcome of interest (in this case, VTE). To have affected the study results, a confounder would need to have been inadequately adjusted for in the studies showing an association. For example, if third generation OCs were preferentially prescribed to women who, because of factors not adequately adjusted for, were at an increased risk of VTE, then these agents would be associated with VTE even if there were no cause-effect relationship. This concern is often referred to as “selection bias,” “prescription bias,” “confounding by indication,” or “channeling.” [4], [5], [6], [7], [8]. The degree of unmeasured confounding that could have produced the observed associations has not been formally characterized.

A second argument that the observed association is spurious involves a phenomenon known as “depletion of susceptibles.” Depletion of susceptibles occurs when the risk of an outcome appears to decline over time because those at highest risk for the event experience it early, leaving fewer high-risk individuals remaining in the population at later time periods. According to this argument, because desogestrel and gestodene were introduced into the market more recently than levonorgestrel, any given group of desogestrel and gestodene users is likely to contain a higher proportion of new starters of OCs than any given group of levonorgestrel users, and thus to contain a higher proportion of individuals with an elevated risk for VTE, because the “susceptibles” have not yet been depleted. Therefore, according to this argument, the observed associations may have resulted from the comparison of new users of desogestrel and gestodene with longer-term users of levonorgestrel. Thus, the “depletion of susceptibles” concern can be re-stated as failure to adequately account for duration of oral contraception use.

We set out to assemble all of the available data that assess the relative risk of VTE from desogestrel and gestodene versus levonorgestrel, to perform a quantitative synthesis, and to explore explanations for any variability in study results. In addition, we wished to characterize the degree of confounding that could have produced the observed association if, in truth, there were no cause-effect relationship. This would permit future discussion to focus on the plausibility of a given degree of confounding. We therefore performed a meta-analysis and formal sensitivity analysis of studies that examined the relative risk of VTE for OCs containing desogestrel and gestodene versus levonorgestrel.