Clinical risk factors for venous thromboembolus in users of the combined oral contraceptive pill

Abstract

Aims

To estimate the risk of venous thromboembolism among women prescribed the oral contraceptive pill who have acute clinical conditions such as lower limb fractures, compared with women with idiopathic venous thromboembolism.

Methods

A nested case-control analysis using the General Practice Research Database, January 1993 to December 1999 was carried out. The participants were women aged 15–39 years, prescribed third generation oral contraceptives (gestodene and desogestrel) or oral contraceptives containing levonorgestrel. The main outcome measures were odds ratios as a measure of the relative risk estimate for venous thromboembolism in women with clinical conditions that predispose to VTE.

Results

The adjusted relative risk estimate for venous thromboembolism among patients with the acute clinical conditions, compared with those without such illness, and adjusted for oral contraceptive use, was 17 (95% CI 6.5, 46).

Conclusions

This paper documents the strong independent association between certain acute clinical conditions and venous thromboembolism in women prescribed oral contraceptives. Failure to accurately identify and exclude such patients from a study of the effect of oral contraceptives on the risk of venous thromboembolism would result in an underestimate of the risk of venous thromboembolism associated with oral contraceptives.

Introduction

The increased risk of idiopathic venous thromboembolism (VTE) in current users of oral contraceptives (OC) is well documented in numerous studies [1–3]. In most of these studies women with acute clinical conditions, thought likely to be strong risk factors for VTE (i.e. recent lower limb fracture, surgery and cancer) [4] were excluded from the case series used in the final analysis to estimate risk of VTE associated with OCs. In these patients, the contribution of OC use to the overall risk of VTE was assumed to be minimal and inclusion of such cases in the analysis tends to modify the OC effect on risk of VTE towards the null [5].

In order to further quantify the effect of certain acute clinical conditions on the risk of VTE in current OC users, we analysed cases excluded from a study recently published by this group addressing the relation of OCs (third generation [gestodene and desogestrel] vs second generation [levonorgestrel]) with ‘idiopathic’ VTE. In that study, women with acute clinical illnesses occurring in the 3 months prior to the VTE were excluded [6].

The objective of the current study was to estimate the relative risk of VTE in patients with these acute clinical conditions that in the previous study resulted in exclusion, compared with patients with idiopathic VTE.

Methods

Data source

The GPRD has been described in many publications before. Over 300 general practices within the UK have participated in data collection since 1988. The quality and completeness of the data collection is regularly assessed. Practices failing to meet the high standards necessary for good quality research have been excluded. Validation studies have been published previously [7, 8].

We used a nested case control design [9] to estimate the relative risk of VTE associated with certain acute clinical illnesses such as lower limb fracture, excluded from our previous study because they were assumed to be strongly and independently associated with VTE.

From the GPRD we identified all women aged 15–39 years of age during the study period of January 1993 and December 1999, who were current users of oral contraceptives containing levonorgestrel, gestodene or desogestrel.

Study design

Blind to OC exposure, we identified all patients with a first time diagnosis of VTE during the study period. The date of the first recording of VTE in the computer records was taken as the index date. Current OC use was defined as a prescription for one of the two classes of study OCs that spanned the index date for that case. Patients who were pregnant or within 3 months post partum were excluded from the study.

As in the earlier study, two investigators independently assessed each potential case by reviewing the computer-generated records. A potential case was included in the study if there was evidence of hospitalization and anticoagulation. The occurrence of any of the conditions listed in Table 1, during the 3 months prior to the index date, were noted.

Table 1.

Type of proximate medical conditions considered to predispose to VTE. Frequency and timing of the events in relation to the index date among the cases and controls.

	Number of cases (Mean time prior to index date)	Number of controls (Mean time prior to index date)
Risk factors
Lower limb fracture	7 (14 days)	0 (–)
Surgery	6 (17 days)	0 (–)
Cancer	1 (–)	2 (–)
Arthroscopy	5 (8 days)	0 (–)
Ankle injury	2 (12 days)	2 (30 days)
Invasive endoscopy	1 (10 days)	2 (10 days)
Other*	4 (–)	0 (–)
No risk factors	106	767
Total	132	773

^*Include one case of each of the following: cystic fibrosis, acute cholecystitis, severe pelvic sepsis and a stab wound that cut the jugular vein resulting in intensive care treatment.

Up to six controls were matched by age ±1 year, practice and index date to each case. The date of VTE for a case was assigned as the index date for the matched controls. Controls were required to be current users of one of the study OCs on the index date. As for the cases, the same exclusion criteria were applied to controls and the occurrence of any of the clinical conditions listed in Table 1, during the 3 months prior to the index date, was noted.

Analysis

Using the Mantel-Haenszel method, and adjusting for type of OC use, we estimated the relative risk of VTE for current oral contraceptive users with the acute clinical conditions listed in Table 1. Analyses were carried out using SAS statistical software.

Results

From the population of current OC users aged 15–39 years; we identified 132 incident cases of VTE. We identified 26 patients in whom VTE occurred within 3 months of an acute clinical condition such as lower limb fracture, surgery or arthroscopy. [Table 1] We considered 106 patients to have ‘idiopathic’ VTE and these have been reported previously [6].

Seven hundred and seventy-three controls were identified from the base population. Among the controls, only six patients were identified as having any of the conditions in Table 1 during the 3 months prior to the index date of the matched case. [Table 1]

The mean age of the patients with idiopathic VTE was 28.7 years and with VTE associated with the acute medical illnesses included in this study was 28.8 years. Among the controls the mean age was 28.4 years. The distribution of age and OC use among cases is shown in Table 2. Among cases without risk factors, users of third generation OCs predominated (64 of 106, 60.4%). For cases with risk factors the pattern was reversed, with fewer third generation users (10 of 26, 38.5%).

Table 2.

The distribution of age and OC use among cases and controls.

	Case
	No risk factor (%)	Risk factor (%)
Age (years)
15–19	10 (9.4)	0 (0)
20–24	18 (7.0)	8 (30.8)
25–29	29 (27.4)	7 (26.9)
30–34	30 (28.3)	10 (38.5)
35–39	19 (17.9)	1 (3.8)
Mean age (years)	28.7	28.8
OC type
Levonorgestrel	42 (39.6)	16 (61.5)
Third generation	64 (60.4)	10 (38.5)
Total	106	26

From the matched case control analysis, we found an estimated relative risk for VTE among patients with the acute clinical conditions listed in Table 1, compared with those without such illness, and adjusted for OC use, of 17 (95% CI 6.5, 46). [Table 3].

Table 3.

Estimated relative risk for VTE in patients with an acute medical event during the 3 months prior to the index date.

	Cases (n = 132)	Controls (n = 773)	Adjusted odds ratio (95% CI)*
No risk factor	106	767	1.0
Risk factor	26	6	17 (6.5, 46)
Total	132	773

^*Controls matched on age, index date, general practice. Estimates adjusted for type of oral contraceptive used.

Discussion

The recognition of an increased risk of VTE and vascular disease associated with the OC pill has resulted in selective prescribing by doctors [1]. We studied only women who were current users of OCs, and who were therefore a highly select, relatively healthy, subset of the population. The types of proximate risk factor for VTE identified in these women were acute clinical conditions that would not have been predicted by the prescribing physicians. Such acute clinical conditions were found commonly in the cases (20%) and in less than 1% of the controls. Fracture of the lower limb, arthroscopy and recent surgery accounted for 70% of the acute clinical conditions identified in the cases.

The General Practice Research Database (GPRD) provides high quality, prospectively gathered medical and drug prescription information [10]. With careful evaluation of the computer generated patient records for every case and control, we were able to identified proximate medical risk factors for VTE.

For individual risk factors, numbers were too small to calculate meaningful relative risks. As a group, adjusting for type of OC prescribed, the estimated relative risk of VTE in a patient with one of the proximate risk factors studied, compared to those with no clinical risk factors, was 17 (95% CI 6.5, 46). The distribution of OC use among the cases with clinical risk factors differed from that among ‘idiopathic’ cases of VTE. Among cases with clinical risk factors for VTE, OC use resembled that of the controls with a preponderance of levonorgesterol. In contrast, among the ‘idiopathic’ cases of VTE 60.4% of patients were prescribed third generation OCs. This difference in OC use among cases with the clinical risk factors and those without clinical risk factors cannot be explained by prescribing practice as the risk factors are acute and unpredictable in nature; rather the difference occurs because of the increased risk for VTE associated with third generation OCs.

In two earlier studies by this group, comparing levonorgestrel and third generation OCs, patients with the risk factors described in Table 1 were excluded [5, 6]. The current study describes results for the patients excluded from the second (2000) study [6] because of clinical conditions thought to be associated with a high risk of VTE. Given the differences in OC distribution and the large, independent relative risk associated with the acute medical conditions studied, exclusion of such patients from our previous studies was appropriate. While the impact on the relative risk was small in this highly select group of generally healthy young women prescribed the OC, there is a bias towards the null. In the current study we confined our investigations to young (15–39 years old) women who were current users of OCs and were therefore, a select group of relatively healthy young women. These results are likely to differ from the findings of studies comparing OC users with nonusers and involving older populations or different drug exposures, where the range of types of clinical risk factors are likely to be greater and they are likely to be found more frequently. Stolley et al. [11] described a much greater distortion towards the null if they included cases with risk factors for VTE when studying association of VTE with OC use because their study included an older population and compared OC users to nonusers. The proportion of patients with clinical risk factors for VTE among the population Stolley et al. studied was much higher and therefore the impact on the relative risk estimate was greater (RR 7.2 among women without risk factors was reduced to 1.9 if the women with risk factors were included).

When studying the often subtle effects of a drug on a disease it is important to try to ensure comparability between the cases and controls for all aspects except the drug exposure of interest. A failure to exclude patients with strong clinical risk factors for the disease of interest will tend to mask the effect of drug exposure.

In summary, this paper documents the strong independent association between certain acute clinical conditions and VTE in women prescribed OCs. Failure to accurately identify and exclude such patients from a study of the effect of OC on the risk of VTE would result in an underestimate of the risk of VTE associated with OCs. The extent of this underestimation is largely dependent on the characteristics of the base population selected for study.