Contents lists available at ScienceDirect

Thrombosis Research

journal homepage: www.elsevier.com/locate/thromres

Full Length Article

Systematic review and meta-analysis of the association of combined oralcontraceptives on the risk of venous thromboembolism: The role of theprogestogen type and estrogen dose

Carina Oedingena,b,⁎, Stefan Scholzc, Oliver Razumb

a Institute of Epidemiology, Social Medicine and Health Systems Research, Hannover Medical School, Hannover, GermanybDepartment of Epidemiology and International Public Health, School of Public Health, Bielefeld University, Bielefeld, Germanyc Department of Health Economics and Health Management, School of Public Health, Bielefeld University, Bielefeld, Germany

A R T I C L E I N F O

Keywords:Venous thromboembolismCombined oral contraceptivesProgestogenEthinylestradiolMeta-analysis

A B S T R A C T

Introduction: Currently available combined oral contraceptives (COC) reportedly increase the risk of venousthromboembolism (VTE). We aimed to quantify this risk considering both progestogen type and estrogen dose.Materials and methods: PubMed, Embase and LIVIVO were searched for relevant publications until April 2017.Case-control and cohort studies including healthy women taking COC and assessing incident VTE as outcomewere selected. Adjusted relative risks (RR) with 95% confidence intervals (CI) derived from random effectsmodel using a generic inverse-variance approach are reported.Results: Overall, 1,359 references were identified and 17 studies were included in the meta-analysis. The pooledRR of VTE was associated with various COC, with the association depending on their respective estrogen doseand progestogen type. Compared to the reference, levonorgestrel with 30–40 μg ethinylestradiol, the overall riskof VTE was higher for all other COC. Preparations with desogestrel with 30–40 μg estrogen showed the highestrelative risk (RR: 1.46; 95% CI: 1.33–1.59), while RRs for drospirenone (30–40 μg ethinylestradiol) and deso-gestrel (30–40/20 μg ethinylestradiol) were lower. COC containing gestodene and cyproterone with 30–40 μgestrogen showed the lowest risk (RR: 1.27; 95% CI: 1.15–1.41 and RR: 1.29; 95% CI: 1.12–1.49, respectively).Conclusions: Compared to levonorgestrel with 30–40 μg ethinylestradiol, all COC showed a significantly in-creased VTE risk. The association varied depending on the progestogen type and the dose of estrogen. Our resultssuggest that the prescription of COC with the lowest possible dose of ethinylestradiol may help to avoid VTEcases among young, healthy women.

1. Introduction

Combined oral contraceptives (COC) are the most common andpreferred form of birth control globally, especially among youngwomen. Approximately 104 million sexually active women aged 15 to49 years worldwide use the “pill” [1], and its usage is particularlycommon in more developed nations [2].

Since the pill was introduced in the early 1960s, a large variety ofcombined estrogen and progestogen preparations became available thatdiffer by their constituents, particularly the type of progestogen and theestrogen dose [3]. Preparations currently in use are levonorgestrel ornorgestrel, also known as second generation progestogens, which arederivatives of testosterone with different degrees of androgenic andestrogenic activities. In so called third generation pills, the androgenicactivity was reduced and the chemical structure was modified in order

to produce progestogens like desogestrel or gestodene. Newest pro-gestogens with an antiandrogenic activity classified as a separate ca-tegory, sometimes also called fourth generation pills, are chlormadi-none acetate, cyproterone acetate or drospirenone (Table 1). Amongthese preparations, the dose of ethinylestradiol ranges from 20 to 50 μg[4,5].

The efficacy of a COC is quantified by the Pearl Index (PI), i.e. therate of unwanted pregnancies per 100 women-years, as tested in ran-domized control trials (RCT). With a PI of around 0.3, all COC pre-parations have a comparable efficacy and therefore they are highlyeffective for prevention of (unwanted) pregnancy [6]. However, theyhave rare but potentially serious side effects such as increased risk ofarterial events (specifically myocardial infarction and ischemic stroke)and venous thromboembolism (VTE), including deep vein thrombosis(DVT) and pulmonary embolism (PE). VTE is a rare disease with an

https://doi.org/10.1016/j.thromres.2018.03.005Received 1 November 2017; Received in revised form 7 March 2018; Accepted 9 March 2018

⁎ Corresponding author at: Institute of Epidemiology, Social Medicine and Health Systems Research, Hannover Medical School, Hannover, Germany.E-mail address: oedingen.carina@mh-hannover.de (C. Oedingen).

Thrombosis Research 165 (2018) 68–78

Available online 15 March 20180049-3848/ © 2018 Elsevier Ltd. All rights reserved.

T

estimated annual incidence of 1 to 2 per 10,000 women of reproductiveage, which increases with age [7]. Despite the low incidence amongwomen, the impact of COC on the risk is relevant on a population levelconsidering their widespread use globally [1,2].

Previous studies have shown an increased risk of VTE while takingCOC which differed depending on the type of progestogen and the doseof estrogen. The annual VTE incidence among COC users is estimated tobe 7 to 12 per 10,000 women of reproductive age [3,4]. Among COCpreparations currently on the market, a two- to threefold increase in therisk with COC containing desogestrel or gestodene (third generation) isestimated compared to those containing levonorgestrel (second gen-eration) [8–11]. The highest risk was ascribed to preparations withnewer progestogens, such as drospirenone or cyproterone [12–15].Among currently available preparations, greater risk of VTE has beenassociated with COC containing desogestrel or gestodene (third gen-eration), compared with those containing levonorgestrel (second gen-eration). A dose-related association of ethinylestradiol has been re-ported, whereby higher doses (= 50 μg) are associated with higher riskof VTE [16–18]. Overall, COC appear be associated with a four- tosixfold increase in risk of developing this condition compared to non-users [19], while the absolute risk remains small.

Critics argue that confounding, bias or both lead to erroneously lowrisk estimates [20–22]. Some studies, indicating no difference betweenthe risk of VTE with third and fourth generation pills compared to COCfrom the second generation, had a case-control design with a high po-tential for recall bias [23,24]. Little data on COC risk is available for thecontraceptives containing chlormadinone acetate and dienogest, mainlyused in Germany [25].

Nevertheless, in the light of the complete body of literature theassociation between COC use and VTE has long been recognized andconfirmed in various observational studies as well as in meta-analyses[3,8–15,26–29], but the evidence on the magnitude of the association isconflicting. Previous studies showed a considerable heterogeneity intheir respective populations and the drug dose, e.g. classified in gen-erations rather than actual preparations, women's age, presence ofhereditary genetic mutations, duration of COC use and study design.The dose of ethinylestradiol, which might increase the VTE risk, wasrarely considered or reported. The present analysis, however doesstratify for different doses of ethinylestradiol.

These circumstances raise the need for a systematic review of theavailable literature in order to allow a combined assessment of pub-lished findings. Our aim was to systematically review the existing evi-dence on the associations of COC exposure containing desogestrel,gestodene, cyproterone and drospirenone as well as different estrogen

doses on the risk of VTE compared to COC with levonorgestrel amonghealthy women. A meta-analysis was performed based on the results ofthe systematic review. We selected observational studies and not RCTs,which are usually considered the highest level of evidence, becausethey focus primarily on pregnancy prevention i.e. efficacy as outcome.A VTE is a rare disease, which strongly increases with age and thereforethe observation period in a RCT design is too short to recognize (suf-ficient) cases of VTE [30,31]. We hypothesized that the newer COCwould show a higher risk for VTE than COC from the second generationcontaining levonorgestrel.

2. Materials and methods

2.1. Search strategy for studies

Case-control and cohort studies evaluating the risk of VTE in usersof COC and listed from their dates of inception to 09 April 2017 weresearched in the three databases PubMed, Embase and LIVIVO withoutlanguage restriction. The search was done with the support of a sci-entific librarian from the University of Bielefeld, Germany (detailedsearch strategy in appendix A).

In addition, we checked reference lists from retrieved literature forrelevant publications. Two researchers (CO and StS) independentlyscreened studies in a two-step process. First, abstracts and titles werescreened to identify potentially relevant studies; these were then read infull for more detail. Disagreements were resolved by consensus with thehelp of a third reviewer (OR).

2.2. Selection method

Inclusion criteria were: (a) original study; (b) case-control design orcohort design; (c) sufficient data to enable comparison of the risk be-tween users of different COC; (d) healthy participants without knownillnesses as gene mutations; (e) control for other pre-existing conditionslike diabetes mellitus or obesity; (f) COC prescribed with the main aimof preventing pregnancy; (g) comparison of COC use to a reference COC(for example, levonorgestrel with 30 μg of ethinylestradiol); and (h) aclear classification of progestogen type dependent on estrogen doses.

Exclusion criteria for original studies were: (a) RCTs, cross-sectionalstudies, reviews and meta-analyses, case reports and re-analyses; (b) noincident VTE; (c) women on postmenopausal hormone replacementtherapy; (d) women taking non-oral or progestogen-only contra-ceptives; (e) women with pre-existing conditions like venous throm-bosis recurrence and/or gene mutations; (f) no control for other pre-existing conditions like diabetes mellitus or obesity; (g) no comparisonbetween different COC or only COC use compared with non-use; and (h)COC without classifying the progestogen type and estrogen dose.

2.3. Outcomes

The primary outcome was fatal or non-fatal first venous thrombosisevent (DVT and/or PE). VTE was defined as a combined outcome of PE(ICD-10-I26.9) and/or DVT (ICD-10-I80.2) and diagnosed in the studiesas follows: (a) clinical evaluation by a physician; (b) diagnostic testslike serial impedance plethysmography or real-time B-mode ultrasoundsupplemented, if necessary, by contrast venography; (c) required tohave anticoagulant assessment; (d) a medical history, including use ofCOC. As meta-analyses of observational studies may be strongly af-fected by single-study biases and heterogeneity [32], we focused on thepotential impact of confounders. To be included, studies had to controlfor age, body mass index (BMI), smoking, pre-existing conditions likediabetes mellitus and the duration of COC exposure.

2.4. Data extraction and quality assessment

All data was extracted from published work by two independent

Table 1List of COC with common progestogen types.

Generation of progestogen(with ethinylestradiol)

Progestogen type Activity

First generation Lynestrenol EstrogenNorethisterone Androgen

Second generation Levonorgestrel AntiestrogenNorgestimate AndrogenNorgestrel

Third generation Desogestrel AntiestrogenGestodene Androgen

Fourth generation Chlormadinone acetate⁎ AntiestrogenCyproterone acetate⁎ AntiandrogenDienogestDrospirenone

Used abbreviations: levonorgestrel with 30–40 μg ethinylestradiol = LNG/EE30–40;desogestrel with 20 μg ethinylestradiol=DSG/EE20; desogestrel with 30–40 μgethinylestradiol =DSG/EE30–40; gestodene with 20 μg ethinylestradiol =GSD/EE20;gestodene with 30–40 μg ethinylestradiol=GSD/EE30–40; cyproterone with 30–40 μgethinylestradiol = CPA/EE30–40; drospirenone with 30–40 μg ethinylestradiol=DRSP/EE30–40.

⁎ Acetate forms have a longer half-life compared to other progestogens.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

69

researchers (CO and StS) who used the described inclusion and exclu-sion criteria to extract data and evaluate the quality of studies.Extracted data comprised study period, methods, patients' character-istics, COC use and VTE outcome. We followed the Meta-analysis ofObservational Studies in Epidemiology Statement (MOOSE) and theStrengthening the Reporting of Observational Studies in EpidemiologyStatement (STROBE) to assess the quality of cohort studies and case-control studies [33,34]. Both quality indicators were added to a com-bined measure (ranging from 0 to 12) consisting of three parts with fouritems each. A study was regarded to be of “low quality” if the score wasless or equal six points [35]. If the score was above that value, the studywas regarded to be of “high quality”. It was intended to exclude studieslow quality, but all studies reviewed were rated high quality (appendixB).

2.5. Statistical analyses

We analyzed the results of studies that compared the risk of VTEbetween levonorgestrel with 30 to 40 μg ethinylestradiol and deso-gestrel, gestodene, cyproterone and drospirenone with 30 to 40 μgethinylestradiol. Furthermore, we compared desogestrel and gestodenewith a 20 μg dose of estrogen because estrogen doses in these formulasvaried. To consider the dose of ethinylestradiol we stratified the ana-lyses in regard to the different dosages.

Some studies reported both frequencies of events and odds ratios(OR) or relative risks (RR), whereas others reported only absolute risks.We therefore performed a pooled analysis based on two-by-two tables.We calculated the meta-analysis adjusted RR by pooling adjusted OR

and RR from individual studies. The data was combined and weightedby a generic inverse-variance approach [36] and, in order to accountfor between-study variance, the random effects method was adopted[37]. This approach enables the inclusion of diverse estimation of riskwithin the same meta-analysis. As the OR accurately estimates the RRwhen risks of disease are small we used the same method for case-control studies and cohort studies. Where applicable, we also reportedORs (see appendix C). We used forest plots for graphical representationof the individual study point estimates and their associated 95% con-fidence interval (CI). We calculated the statistical analyses using thesoftware Review Manager (RevMan), version 5.3 developed by theCochrane Collaboration [38].

To calculate absolute risks per 10,000 women per year and esti-mated respective additional number of VTE with taking COC from thepooled adjusted RRs we used following formulae. Marginal differencesare caused by rounding errors:

(Absolute risk of VTE per 10,000 women per year taking levo-norgestrel with 30 to 40 μg ethinylestradiol) ∗ (pooled adjusted RR fromthe different COC)=Absolute risk of VTE per 10,000 women per yeartaking different COC.

(21.7/104 Mio. users in Western Europe/Worldwide per year) ∗ (7.6(absolute risk of levonorgestrel with 30 to 40 μg ethinylestradiol) /10,000) ∗ (pooled adjusted RR of different COC)= Estimated respectiveadditional number of VTE with taking COC.

To explore substantial heterogeneity between studies, we tested thehypothesis that the differences between the reported RR were only dueto random error around the true RR using the χ2 test and I2 statistic.Results were considered heterogeneous when homogeneity was

Fig. 1. Study flow diagram.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

70

unlikely (p < 0.10). To determine the stability of the overall RR esti-mate, we performed a sensitivity analysis in which each study wasgradually eliminated. Additionally, we performed subgroup analyseswith comparison between second vs. third and fourth generation pills aswell as case-control and cohort studies. For the results of the subgroupanalyses see appendix D–E.

3. Results

3.1. Search results and characteristics of the included studies

The search strategy retrieved 2,905 references, of which 1,546 du-plicates were excluded. After screening title, abstract and full-text, 17studies represented in 16 publications met the inclusion criteria andwere included in meta-analysis (Fig. 1). Of the included studies, 11were case-control studies [8–9,16–17,19,39–43] and 6 were cohortstudies [10–11,18,44–46]. They were published between 1995 and2016, and most were performed in Europe (4 in the UK and in theNetherlands, respectively; 3 in Denmark; 1 in France; 3 in the USA; 1 inIsrael and 1 transnational study covering Europe, North and SouthAmerica). The 6 cohort studies represented a total of 25,040,169women-years, while the 11 case-control studies included> 68,069women. Table 2 summarizes the main study characteristics.

3.2. Synthesis of results

All investigated COC showed higher pooled adjusted estimates ofrisk compared to the reference levonorgestrel in the respective meta-analysis. The adjusted RR estimate was highest in users of DSG/EE30–40 and lowest in users of GSD/EE20. For desogestrel and gesto-dene, a dose-response association was observed, with higher dosesbeing associated with higher venous thrombosis risk.

In the synthesis of 14 case-control studies and cohort studies com-paring the risk of VTE associated with DSG/EE30–40 versus LNG/EE30–40 the pooled adjusted RR was 1.46 (95% CI: 1.33–1.59), and in11 case-control studies and cohort studies 1.39 (95% CI: 1.16–1.67) forthose containing DSG/EE20 (Figs. 2 and 3). When choosing OR as thecombined outcome, the pooled adjusted measure was 1.72 (95% CI:1.59–1.85).

Compared with LNG/EE30–40, the risk in women taking GSD/EE30–40 is increased 1.27-fold (95% CI: 1.15–1.41), and for GSD/EE201.18-fold (95% CI: 0.93–1.64). As shown in Figs. 4 and 5, these meta-analyses contained 12 case-control studies and cohort studies for thecomparison GSD/EE30–40 vs. LNG/EE30–40 but only 6 studies forGSD/EE20. That may be the reason, why solely COC containing ges-todene with 20 μg estrogen did not significantly increase the risk ofVTE. However, the results were similar to the previous associations. TheOR differed little for the comparisons of GSD/EE20 vs. LNG/EE30–40(1.20; 95% CI: 0.90–1.58), but was higher for GSD/EE30–40 vs. LNG/EE30–40 (1.38; 95% CI: 1.22–1.56).

The pooled adjusted results from fourth generation progestogenscomparing VTE in women using COC containing cyproterone or dros-pirenone versus levonorgestrel with 30–40 μg ethinylestradiol showed ahigh risk in DRSP/EE30–40 (RR: 1.40; 95% CI: 1.26–1.56/OR: 1.54.;95% CI: 1.34–1.77) compared to the reference COC of LNG/EE30–40.Users with CPA/EE30–40 (RR: 1.29; 95% CI: 1.12–1.49/OR: 1.43; 95%CI: 1.15–1.79) have a slightly higher risk as the reference (Figs. 6 and7). The risk of VTE for DRSP/EE30–40 was similar to the risk for DSG/EE20, whereas the risk for CPA/EE30–40 is comparable to the risk forGSD/EE30–40. Also, the synthesis combined 6 respectively 8 case-control studies and cohort studies for CPA/EE30–40 respectively DRSP/EE30–40.

The estimated risks of VTE per 10,000 women per year, according tothe use of COC and type of progestogen calculated from the RR valuesobtained in the present meta-analysis are shown in Table 3.

3.3. Heterogeneity and publication bias

For all comparisons, the I2 was significant (p < 0.03), except forCPA/EE30–40 where the I2 showed no heterogeneity (p=0.16).However, the funnel plots displaying RR of the individual comparisonsversus the logarithm of their standard error (SE) appeared slightlyskewed to the right for all analyses. This suggests the presence ofpublication bias. Fig. 8 shows the exemplary funnel plot for DSG/EE20vs. LNG/EE30–40 (for the other funnel plots see appendix F). The di-rection of such a bias indicates that the actual risk of VTE might belower than the present estimates in the meta-analyses. Nevertheless, theestimated CI under the random effects model are quite narrow and theresults are statistically significant. Furthermore, the results were robustunder the fixed effect model.

4. Discussion

We performed a systematic review and meta-analysis based on 17studies represented in 16 publications. The results lead us to reject thenull hypothesis that the risk of VTE between different COC would be thesame. We found that all investigated COC with newer formulas have ahigher risk compared with the reference COC containing levonorgestrelfrom the second generation. Among these newer preparations, thehighest elevations in VTE risk are observed in DSG/EE30–40 (46%), inDRSP/EE30–40 (40%), and in DSG/EE20 (39%). COC containing GSD/EE30–40 and CPA/EE30–40 showed an increase in risk of nearly 30%.Solely GSD/EE20 did not significantly increase the risk of VTE.However, this investigation examined the potential association of dif-ferent COC on the pathogenesis of VTE. Thereby, the association de-pends not only on the progestogen type used, but also on the dose ofethinylestradiol.

Our findings are consistent with those of a previous network meta-analysis from the Cochrane Collaboration [27]. This method sum-marizes data from direct and indirect comparisons and therefore allowscomparison between types of progestogen where no sufficient data isavailable, like dienogest or chlormadinone acetate. The disadvantage isthat higher degrees of heterogeneity and inconsistency are possible andas a result, the CI and p-values are not as precise as in a conventionalpairwise meta-analysis. Compared to levonorgestrel with 30 μg ethi-nylestradiol, their results showed that the VTE risk is increased 1.8-fold(95% CI: 1.4–2.2) and 1.5-fold (95% CI: 1.5–1.2-2.0), respectively, forusers of desogestrel or gestodene, and 1.6-fold (95% CI: 1.1–2.2 and1.2–2.1) for users of pills with cyproterone and drospirenone with thesame estrogen dose. The comparison with a stratified ethinylestradiol of20 μg in reference with levonorgestrel and 30 μg ethinylestradiol withdesogestrel have a slightly elevated risk (RR: 1.4; 95% CI: 1.0–1.9),while gestodene with 20 μg estrogen shows the same risk as the re-ference. However, the calculated RR are a little higher and the CIbroader than in this analysis (our RR and CI all lie within their ranges).

Thus, COC containing desogestrel with 30–40 μg ethinylestradiolshowed the highest risk of VTE. These formulas have a risk equivalentto that of COC with drospirenone and 30–40 μg ethinylestradiol. Otheranalyses found that drospirenone alone has the highest risk [11–14].We did not find the VTE risks between the progestogen-estrogen com-binations from the third and fourth generation to be varied as much asit is often assumed [12–15]. Besides, our findings suggest that cypro-terone with 30–40 μg ethinylestradiol has the same VTE risk like ges-todene with the same dosage of estrogen. The Cochrane Collaborationconcluded that the risks of cyproterone acetate and drospirenone areidentical. This can be attributed to the work of Vinogradova et al. [43],which was published in 2015 and investigated two analyses in differentdatabases with a total of 52,596 women, including 637 users of pillscontaining cyproterone. Our meta-analysis was the first to include thisstudy. It is possible that previous meta-analyses, especially from theCochrane Collaboration, would have had an insufficient number of suchusers, which may have led to an overestimation of risk.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

71

Table2

Cha

racteristics

ofinclud

edstud

ies.

Autho

ran

dye

arStud

yde

sign

Databa

se/o

bserva

tion

period

Participan

tsEx

position

Prim

aryou

tcom

eAdjustedforco

nfou

nder

Cou

ntry

Com

bine

dscore

Bird

etal.[

18]

Coh

ort

stud

yIM

SLifelin

k™05

/200

1–12

/200

943

2,17

8wom

en/2

63,902

wom

enye

ars

age:

18–4

6y

diag

nosis:

antico

agulation/

ICD-9

LNG/E

E30–

40;

DRSP

/EE3

0–40

354VTE

(LNG:1

18/1

32,681

;DRSP

:23

6/13

1,22

1)

Prop

ensity

score

USA

9

Bloe

men

kamp

etal.[

8]Case-

control

stud

y

Leiden

Thrombo

philiastud

y01

/198

8–12

/199

212

6cases/15

9co

ntrols

(pop

ulationba

sed)

age:

15–4

9y

diag

nosis:

antico

agulation

LNG/E

E30–

40;D

SG/

EE30

–40

103VTE

(LNG:2

0/38

;DSG

:37/

52)

Age

;fam

ilyhistory;

prev

iously

preg

nanc

y;factor

VLe

iden

-mutation

Nethe

rlan

ds8

Bloe

men

kamp

etal.[

39]

Case-

control

stud

y

2stud

ycenters(A

cade

mic

Med

ical

Cen

terof

theUnive

rsityof

Amsterda

m&Amsterda

mTh

rombo

sis

Servicean

dLa

boratory

forGen

eral

Practition

ers)

09/1

982–

10/1

995

185cases/59

1co

ntrols

(hospitalba

sed)

age:

15–4

9y

diag

nosis:

diag

nostic

tests

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE

30–4

0

134VTE

(LNG:1

8/46

;DSG

:28/

58;

GSD

:5/9

)

Age

;fam

ilyhistory;

duration

ofuse;

stud

ycenter;

prev

iously

preg

nanc

y;pre-

existing

cond

itions

Nethe

rlan

ds9

Gronich

etal.[

44]

Coh

ort

stud

yClalit

HealthSe

rvices

01/2

002–

12/2

008

329,99

5wom

en/

819,74

9wom

enye

ars

age:

12– 5

0y

diag

nosis:

ICD-9

2.&

3.ge

neration

;DRSP

/EE

30–4

0

506VTE

(2.g

eneration:

23/3

3,18

7;3.

gene

ration

:38

4/65

1,45

5;DRSP

:99/

114,79

7)

Age

;BMI;sm

oking;

duration

ofuse;

pre-existing

cond

itions

Israel

9

Herings

etal.[

45]

Coh

ort

stud

yPH

ARMO

system

01/1

986–

12/1

995

450,00

0wom

en/

54,939

wom

enye

ars

age:

15–4

9y

diag

nosis:

ICD-9

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE

30–4

0

33VTE

(LNG:6

/24,95

3;DSG

:22/

25,005

;GSD

:5/4

,982

)

Age

;durationof

use

Nethe

rlan

ds8

Jick

etal.[

17]

Nested

case-

control

stud

y

Gen

eral

Practice

ResearchDatab

ase(G

PRD)

01/1

991–

11/1

994

75cases/30

0co

ntrols

(pop

ulationba

sed)

age:<

40y

diag

nosis:

diag

nostic

tests;

antico

agulation

LNG/E

E30–

40;D

SG/

EE30

–40;

GSD

/EE3

0–40

75VTE

(LNG:2

3/16

4;DSG

:30/

121;

GSD

:22/

90)

Age

;BMI;sm

oking;

duration

ofuse

UK

10

Jick

etal.[

40]

Nested

case-

control

stud

y

PharMetrics

01/2

000–

03/2

005

281cases/1,05

5co

ntrols

(pop

ulationba

sed)

age:

15–3

9y

diag

nosis:

antico

agulation;

ICD-9

LNG/E

E30–

40;D

SG/

EE30

–40

157VTE

(LNG:7

0/38

6;DSG

:87/

315)

Durationof

use;

pre-existing

cond

itions;co

nsultation

san

d/or

emerge

ncyroom

USA

9

Jick/ Herna

ndez

[41]

Nested

case-

control

stud

y

PharMetrics

01/2

002–

12/2

008

186cases/68

1co

ntrols

(pop

ulationba

sed)

age:

15–4

4y

diag

nosis:

diag

nostic

tests;

antico

agulation;

ICD-9

LNG/E

E20;

LNG/E

E30–

40;

DRSP

/EE3

0–40

186VTE

(LNG:6

5/43

3;DRSP

:121

/43

4)

Age

;durationof

use

USA

9

Lide

gaardet

al.[

16]

Case-

control

stud

y

Nationa

lPa

tien

tReg

ister

01/1

994–

12/1

998

987cases/4,05

4co

ntrols

(pop

ulationba

sed)

age:

15–4

4y

diag

nosis:

diag

nostic

tests;

antico

agulation

LNG/E

E30–

40;D

SG/

EE30

–40;

GSD

/EE2

0;GSD

/EE3

0–40

;CPA

/EE

30–4

0

907VTE

(LNG:9

8/29

6;DSG

:121

/340

;GSD

:212

/72

8)

Age

;BMI;family

history;

duration

ofuse;

prev

iously

preg

nanc

y;pre-existing

cond

itions;ed

ucationye

ars

Den

mark

11

Lide

gaardet

al.[

46]

Coh

ort

stud

yNationa

lDan

ish

Reg

ister

01/1

995–

12/2

005

10,447

,373

wom

enye

ars

age:

15–4

9y

diag

nosis:

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE2

0;GSD

/EE3

0–40

;CPA

/EE

30–4

0;DRSP

/EE3

0–40

3,96

9VTE

(LNG:2

38/4

11,099

;DSG

:44

2/67

6,10

6;GSD

:928

/1,33

2,15

7;CPA

:90/

Pre-existing

cond

itions;

educ

ationye

ars

Den

mark

10

(con

tinuedon

next

page)

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

72

Table2(con

tinued)

Autho

ran

dye

arStud

yde

sign

Databa

se/o

bserva

tion

period

Participan

tsEx

position

Prim

aryou

tcom

eAdjustedforco

nfou

nder

Cou

ntry

Com

bine

dscore

diag

nostic

tests;

antico

agulation;

ICD-10

126,68

7;DRSP

:103

/13

1,54

1)Lide

gaardet

al.[

11]

Coh

ort

stud

yNationa

lDan

ish

Reg

ister

01/2

001–

12/2

009

1,29

6,12

0wom

en/

8,01

0,29

0wom

enye

ars

age:

15–4

9y

diag

nosis:

diag

nostic

tests;

antico

agulation

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE2

0;GSD

/EE3

0–40

;CPA

/EE

30–4

0;DRSP

/EE3

0–40

4,08

7VTE

(LNG:1

09/1

27,942

;DSG

:52

3/64

1,23

1;GSD

:1.059

/1,14

0,47

3;CPA

:109

/12

0,93

4;DRSP

:289

/30

9,91

4)

Age

;durationof

use;

educ

ationye

ars

Den

mark

10

Todd

etal.[

42]

Nested

case-

control

stud

y

Med

iPlusDatab

ase

01/1

992–

03/1

997

99cases/36

6co

ntrols

(pop

ulationba

sed)

age:

15–4

9y

diag

nosis:

antico

agulation

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

75VTE

(LNG:2

2/10

1;DSG

:32/

118;

GSD

:21/

92)

BMI;sm

oking;

pre-existing

cond

itions

UK

9

vanHylckam

aVlie

get

al.[

19]

Case-

control

stud

y

MEG

Astud

y(m

ultipleen

vironm

entalan

dge

netic

assessmen

tof

risk

factorsforve

nous

thrombo

sis-stud

y)03

/199

9–09

/200

4

1,52

4cases/1,76

0co

ntrols

(pop

ulationba

sed)

age:<

50y

diag

nosis:

diag

nostic

tests;

antico

agulation

LNG/E

E30–

40;D

SG/

EE30

–40;

GSD

/EE3

0–40

;CPA

/EE3

0–40

;DRSP

/EE

30–4

0

1,47

8VTE

(LNG:4

85/8

58;D

SG:2

89/

397;

GSD

:119

/186

;CPA

:12

5/18

7;DRSP

:19/

33)

Age

;BMI;da

teof

inclusion

Nethe

rlan

ds8

Vinog

rado

vaet

al.

[43]

(1)

Nested

case-

control

stud

y

Clin

ical

Practice

ResearchDatab

ase

(CPR

D)

01/2

001–

12/2

013

5,06

2cases/19

,638

controls

age:

15–4

9y

diag

nosis:

REA

Dco

des

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE2

0;GSD

/EE3

0–40

;CPA

/EE

30–4

0;DRSP

/EE3

0–40

1,04

1VTE

(LNG:5

21/1

,972

;DSG

:16

5/39

3;GSD

:78/

227;

CPA

:138

/330

;DRSP

:139

/33

9)

BMI;sm

oking;

alco

holuse;

race;p

re-existingco

nditions

UK

10

Vinog

rado

vaet

al.

[43]

(2)

Nested

case-

control

stud

y

QResearchPrim

aryCareDatab

ase

01/2

001–

12/2

013

5,55

0cases/22

,396

controls

age:

15–4

9y

diag

nosis:

REA

Dco

des

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GSD

/EE2

0;GSD

/EE3

0–40

;CPA

/EE

30–4

0;DRSP

/EE3

0–40

1,04

0VTE

(LNG:5

40/1

,951

;DSG

:16

3/42

5;GSD

:115

/297

;CPA

:120

/307

;DRSP

:102

/27

2)

BMI;sm

oking;

alco

holuse;

race;p

re-existingco

nditions

UK

10

Weillet

al.[

10]

Coh

ort

stud

yFren

chna

tion

alhe

alth

insuranc

eda

taba

se(SNIIRAM)&

Fren

chho

spital

discha

rgeda

taba

se(PMSI)

07/2

010–

09/2

012

4,94

5,08

8wom

en/

5,44

3,91

6wom

enye

ars

age:

15–4

9y

diag

nosis:

ICD-10

LNG/E

E30–

40;D

SG/E

E20;

DSG

/EE3

0–40

;GDS/

EE20

;GDS/

EE30

–40

1,75

4VTE

(LNG:1

,048

/3,751

,356

;DSG

:589

/1,256

,531

;GDS:

117/

296,41

6)

Age

;soc

io-eco

nomic

status;

consultation

s;sm

oking;

pre-

existing

cond

itions

Fran

ce9

WHO

[9]

Case-

control

stud

y

10stud

ycentersin

9co

untries

02/1

989–

01/1

993

769cases/2,22

5co

ntrols

(1,979

hospital

controls

&24

6co

mmun

ityco

ntrols)

age:

20–4

4y

diag

nosis:

antico

agulation

LNG/E

E30–

40;D

SG/

EE30

–40;

GSD

/EE3

0–40

605VTE

(LNG:1

37/3

40;D

SG:3

5/63

;GSD

:36/

64)

BMI;sm

oking;

alco

holuse;

pre-existing

cond

itions;

prev

iously

preg

nanc

y

Brazil,

Chile,

Colom

bia,

German

y,Hon

gKon

g,Hun

gary,

Jamaica,T

haila

nd&

UK

8

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

73

We assume that our pooling of 30 and 40 μg ethinylestradiol in re-lation to the sole consideration of 30 μg in the network meta-analysis ofthe Cochrane Collaboration led to marginal differences in the calculatedRR. With a different approach we would have had fewer studies for thesynthesis. Furthermore, the network meta-analysis investigated just theprogestogen norgestimate from the second generation as an additionalcomparison. We did not include it in our analysis since this progestogenis rarely used, especially in France and Germany [10,25]. Progestogenslike dienogest or chlormadinone acetate could also not be consideredbecause of few studies and insufficient data.

Other meta-analyses have evaluated the risk of VTE comparingcontraceptive use [4,26,28,29]. The authors compared non-users and/or levonorgestrel often with the progestogen type only and not with theestrogen dose (rarely also with a dose of< 50 μg). Usually, the pro-gestogen types are summarized by generations. In our opinion, thismethod does not allow a detailed comparison and classification of therisk profile between the different preparations of COC. The presentanalysis showed that a focus on the generations or the progestogen typewithout the consideration of ethinylestradiol may overestimate the truerisk: The risks for either the single progestogens or the whole genera-tion are overestimated compared to levonorgestrel, especially if womenuse pills with higher dosed ethinylestradiol. However, it remains un-certain which dosage of estrogen is most frequent among users.

In the light of our findings, the recommendations of the EuropeanMedicines Agency's Pharmacovigilance Risk Assessment Committeemight need to be revised, which states that the VTE risk for all COC

with a low dose ethinylestradiol under 50 μg is comparable [3].Further, deviations from previous findings might be attributable to

different inclusion and exclusion criteria. In our analysis we integratedonly studies which observed solely healthy women without hereditarythrombophilia and incident cases of VTE, leading to a smaller numberof included studies compared to the meta-analyses of de Bastos et al.[27] and Kemmeren et al. [28]. Both meta-analyses included the studyof Martinelli and colleagues [48] which included besides healthywomen also participants with APC resistance (factor V Leiden muta-tion) and prothrombin G20210A (factor II mutation). While the authorsadjusted for these conditions, they did not control for confounderslinked to them. Nightingale et al. [49] showed that women with genemutations have a lower general health status compared to healthywomen so that the VTE risk is inherently higher. Likewise, the analysisby the Cochrane Collaboration included the study of Lewis et al. [50]although there is no clear description if incident or prevalent cases areincluded. Including these studies could have led to overestimated RRsin the other meta-analyses.

4.1. Strengths and limitations

The main limitation of our meta-analysis of observational studies,and the main problem in conducting meta-analyses in general, is be-tween-study heterogeneity, which is due to different designs, sam-plings, settings, methods and classifications of COC. We as well foundheterogeneity and possible publication bias. When we conducted

Fig. 2. Forest-Plot DSG/EE30–40 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

Fig. 3. Forest-Plot DSG/EE20 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

74

Fig. 4. Forest-Plot GSD/EE30–40 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

Fig. 5. Forest-Plot GSD/EE20 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

Fig. 6. Forest-Plot DRSP/EE30–40 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

Fig. 7. Forest-Plot CPA/EE30–40 vs. LNG/EE30–40 and VTE. Asterisks mark case-control studies.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

75

sensitivity analyses in which single studies were excluded, calculatedthe synthesis once more under a fixed effect model as well as separatedfor case-control studies and cohort studies, we found no substantialdifference in the overall estimates of risk (appendix D–E).

Afterwards, we conducted a meta-analysis pooling two differenteffect sizes: the OR and RR. This approach was feasible because VTE is arare disease with an incidence of< 10%, whereby the OR could beapproximated well by the RR and vice-versa. Only 17 studies met theinclusion criteria, some reporting OR and others RR. To obtain com-parable results between the two effect sizes, the risk of VTE develop-ment should be identical over the whole observation period. Althoughwe controlled for the women's age, this assumption cannot be com-pletely fulfilled. However, there was no substantially difference in theseparate analyses for the respective study design. Marginal differenceswere found when calculated either RR or OR (appendix C).

Bias and confounding may have influenced our results; age in par-ticular is a potential confounder for the association between COC useand occurrence of VTE. Because levonorgestrel is on the market sincethe 1970s, women using these progestogens are presumably older thanwomen on third or fourth generation progestogens. Thus, the calculatedRR for the progestogens could have been underestimated comparedwith levonorgestrel from the second generation.

There is no generally accepted approach to classify COC. To avoidinformation- and misclassification bias, we included only studies whichgave detailed information about the type of progestogen and the dose ofestrogen. Additionally, the diagnoses of COC and of VTE respectivelywere confirmed by physicians and not self-reported by the patients[27].

We found that the dose of ethinylestradiol had an influence on VTE

risk. Preparations available on the market have different doses of es-trogen as well as progestogen, but no analyses have so far taken intoaccount the dose of progestogen. Equally, the progestogens dienogestand chlormadinone acetate, mainly used in Germany, could not beconsidered because of few studies. These circumstances raise the needfor research in this field.

The precision of the results is dependent on the quality of includedstudies and the sample size. All studies included in our analysis hadhigh quality with an average of 9 out of 12 possible total points (seeappendix B). The 11 case-control studies had a sample size of 19,758women and the 6 cohort studies of 23,595,640 women-years.

5. Conclusions

The investigated COC showed a significantly increased risk of VTE.The association varied depending on the progestogen type as well as thedose of ethinylestradiol. Hence, the COC with the lowest possible doseof estrogen, levonorgestrel with 20 μg or 30 to 40 μg ethinylestradiol,should be prescribed to minimize the VTE cases among young, healthywomen.

In addition, to avoid misunderstanding and to value risks realistic, itis important to report the calculated results into their equivalence inabsolute risks of VTE. According to our results, the absolute risks arevery similar between the different COC, but in view of the fact that inWestern Europe alone are nearly 21.7 million and worldwide 104million users, the estimated numbers of VTE are high although theabsolute risks tend to be small. So absolute risks can give adequateinformation about the currently available preparations for users, phy-sicians, especially gynecologists, and health insurances. This helps to

Table 3Estimated risk of VTE.

Risk of the development of VTE for a one-year period

Type of COC Pooled adjusted RR(95% CI)

Absolute risk of VTE per 10,000 women peryear (95% CI)

Estimated number of VTE with takingCOC per year⁎

Western Europe/worldwide

Additional number of VTE with taking COCper year⁎

Western Europe/worldwide

LNG/EE30–40 1 7.6 16,492/79,040 –DSG/EE30–40 1.46 (1.33–1.59) 11.1 (10.1–12.1) 24,078/115,398 7,586/36,358DSG/EE20 1.39 (1.16–1.67) 10.6 (8.8–12.7) 22,924/109,866 6,432/30,826GSD/EE30–40 1.27 (1.15–1.41) 9.7 (8.7–10.7) 20,945/100,381 4,453/21,341GSD/EE20 1.18 (0.93–1.49) 9.0 (7.1–11.3) 19,461/93,267 2,969/14,227CPA/EE30–40 1.29 (1.12–1.49) 9.8 (8.5–11.3) 21,275/101,962 4,783/22,922DRSP/EE30–40 1.40 (1.26–1.56) 10.6 (9.6–11.9) 23,089/110,656 6,597/31,616

⁎ With 21.7 million users of COC in Western Europe (France, Germany, Italy, Spain and UK) as well as 104 million users of COC worldwide [2,47].

Fig. 8. Funnel-Plot DSG/EE20 vs. LNG/EE30–40 and VTE.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

76

advise users individually and comprehensibly about potential side ef-fects and to take into account a woman's baseline risk as well as hercharacteristics and preferences.

Supplementary data to this article can be found online at https://doi.org/10.1016/j.thromres.2018.03.005.

Abbreviations

BMI body-mass-indexCI confidence intervalCOC combined oral contraceptiveCPA/EE30–40 cyproterone with 30–40 μg ethinylestradiolDRSP/EE30–40 drospirenone with 30–40 μg ethinylestradiolDSG/EE20 desogestrel with 20 μg ethinylestradiolDSG/EE30–40 desogestrel with 30–40 μg ethinylestradiolDVT deep vein thrombosisGSD/EE20 gestodene with 20 μg ethinylestradiolGSD/EE30–40 gestodene with 30–40 μg ethinylestradiolLNG/EE30–40 levonorgestrel with 30–40 μg ethinylestradiolMOOSE META-analysis of Observational Studies in Epidemiology

StatementOR odds ratioPE pulmonary embolismPI pearl indexRCT randomized control trialRR relative riskSTROBE Strengthening the Reporting of Observational Studies in

Epidemiology StatementVTE venous thromboembolism

Conflicts of interest

The authors report no conflict of interest with this study.

Funding sources

This research did not receive any specific grant from fundingagencies in the public, commercial, or non-for-profit sectors.

References

[1] Earth Policy Institute, World Modern Contraceptive Prevalence by Method, LatestYear (Data from U.N. Population Division), http://www.earth-policy.org/data_center/C21, (2011) , Accessed date: 17 September 2017.

[2] United Nations, Department of Economic and Social Affairs, World ContraceptivePatterns 2013, United Nations, 2013.

[3] European Medicines Agency (EMA), Assessment Report for Combined HormonalContraceptives Containing Medicinal Products, Procedure No: EMEA/H/A-31/1356, http://www.ema.europa.eu/docs/en_GB/document_library/Referrals_document/Combined_hormonal_contraceptives/Recommendation_provided_by_Pharmacovigilance_Risk_Assessment_Committee/WC500160272.pdf, (2014) ,Accessed date: 17 September 2017.

[4] G. Plu-Bureau, L. Maitrot-Mantelet, J. Hugon-Rodin, M. Canonico, Hormonal con-traceptives and venous thromboembolism: an epidemiological update, Best Pract.Res. Clin. Endocrinol. Metab. 27 (1) (2013) 25–34.

[5] T.A. Lawrie, F.M. Helmerhorst, N.K. Maitra, R. Kulier, K. Bloemenkamp,A.M. Gulmezoglu, Types of progestogens in combined oral contraception: effec-tiveness and side-effects, Cochrane Database Syst. Rev. (5) (2011) (CD004861).

[6] I. Wiegratz, C.J. Thaler, Hormonal contraception—what kind, when, and forwhom? Dtsch Arztebl Int 108 (28–29) (2011) 495–506.

[7] I.A. Naess, S.C. Christiansen, P. Romundstad, S.C. Cannegieter, F.R. Rosendaal,J. Hammerstrom, Incidence and mortality of venous thrombosis: a population-basedstudy, J. Thromb. Haemost. 5 (4) (2007) 692–699.

[8] K.W. Bloemenkamp, F.R. Rosendaal, F.M. Helmerhorst, H.R. Buller,J.P. Vandenbroucke, Enhancement by factor V Leiden mutation of risk of deep-veinthrombosis associated with oral contraceptives containing a third-generation pro-gestagen, Lancet 346 (8990) (1995) 1593–1596.

[9] World Health Organization Collaborative Study of Cardiovascular Disease andSteroid Hormone Contraception, Effect of different progestagens in low oestrogenoral contraceptives on venous thromboembolic disease, Lancet 346 (8990) (1995)1582–1588.

[10] A. Weill, M. Dalichampt, F. Raguideau, P. Ricordeau, P.-O. Blotière, J. Rudant,F. Alla, M. Zureik, Low dose oestrogen combined oral contraception and risk of

pulmonary embolism, stroke, and myocardial infarction in five million Frenchwomen: cohort study, BMJ 353 (2016) i2002.

[11] O. Lidegaard, L.H. Nielsen, C.W. Skovlund, F.E. Skjeldestad, E. Lokkegaard, Risk ofvenous thromboembolism from use of oral contraceptives containing differentprogestogens and oestrogen doses: Danish cohort study, 2001–9, BMJ 343 (2011)d6423.

[12] L. Parkin, K. Sharples, R.K. Hernandez, S.S. Jick, Risk of venous thromboembolismin users of oral contraceptives containing drospirenone or levonorgestrel: nestedcase-control study based on UK General Practice Research Database, BMJ 342(2011) d2139.

[13] S. Sidney, T.C. Cheetham, F.A. Connell, R. Ouellet-Hellstrom, D.J. Graham,D. Davis, M. Sorel, C.P. Quesenberry Jr., W.O. Cooper, Recent combined hormonalcontraceptives (CHCs) and the risk of thromboembolism and other cardiovascularevents in new users, Contraception 87 (1) (2013) 93–100.

[14] J. Dinger, A. Assmann, S. Mohner, T.D. Minh, Risk of venous thromboembolism andthe use of dienogest- and drospirenone-containing oral contraceptives: results froma German case-control study, J. Fam. Plann. Reprod. Health Care 36 (3) (2010)123–129.

[15] R.D. Farmer, R.A. Lawrenson, J.C. Todd, T.J. Williams, K.D. MacRae, F. Tyrer,G.M. Leydon, A comparison of the risks of venous thromboembolic disease in as-sociation with different combined oral contraceptives, Br. J. Clin. Pharmacol. 49 (6)(2000) 580–590.

[16] O. Lidegaard, B. Edstrom, S. Kreiner, Oral contraceptives and venous thromboem-bolism: a five-year national case-control study, Contraception 65 (3) (2002)187–196.

[17] H. Jick, S.S. Jick, V. Gurewich, M.W. Myers, C. Vasilakis, Risk of idiopathic car-diovascular death and nonfatal venous thromboembolism in women using oralcontraceptives with differing progestagen components, Lancet 346 (8990) (1995)1589–1593.

[18] S.T. Bird, J.A. Delaney, M. Etminan, J.M. Brophy, A.G. Hartzema, Drospirenone andnon-fatal venous thromboembolism: is there a risk difference by dosage of ethinyl-estradiol? J. Thromb. Haemost. 11 (6) (2013) 1059–1068.

[19] A. van Hycklama Vlieg, F.M. Helmerhorst, J.P. Vandenbroucke, C.J.M. Doggen,F.R. Rosendaal, The venous thrombotic risk of oral contraceptives, effects of oes-trogen dose and progestogen type: results of the MEGA case-control study, BMJ 339(2009) 1–8.

[20] R.D. Farmer, R.A. Lawrenson, C.R. Thompson, J.G. Kennedy, I.R. Hambleton,Population-based study of risk of venous thromboembolism associated with variousoral contraceptives, Lancet 349 (1997) 83–88.

[21] S. Suissa, L. Blais, W.O. Spitzer, J. Cusson, M. Lewis, L. Heinemann, First-time use ofnewer oral contraceptives and the risk of venous thromboembolism, Contraception56 (1997) 141–146.

[22] W.O. Spitzer, The aftermath of a pill scare: regression to reassurance, Hum. Reprod.Update 5 (1999) 736–745.

[23] L.A. Heinemann, J.C. Dinger, A. Assmann, T.D. Minh, Use of oral contraceptivescontaining gestodene and risk of venous thromboembolism: outlook 10 years afterthe third-generation "pill scare", Contraception 81 (5) (2010) 401–407.

[24] J. Dinger, A. Assmann, S. Mohner, T.D. Minh, Risk of venous thromboembolism andthe use of dienogest- and drospirenone-containing oral contraceptives: results froma German case-control study, J. Fam. Plann. Reprod. Health Care 36 (3) (2010)123–129.

[25] G. Glaeske, P. Thürmann, Pillenreport, Ein Statusbericht zu Oralen Kontrazeptiva,https://www.tk.de/centaurus/servlet/contentblob/771128/Datei/67490/Pillenreport_2015.pdf, (2015) , Accessed date: 17 September 2017.

[26] L. Manzoli, C. De Vito, C. Marzuillo, A. Boccia, P. Villari, Oral contraceptives andvenous thromboembolism: a systematic review and meta-analysis, Drug Saf. 35 (3)(2012) 191–205.

[27] M. de Bastos, B.H. Stegeman, F.R. Rosendaal, A. Van Hylckama Vlieg,F.M. Helmerhorst, T. Stijnen, O.M. Dekkers, Combined oral contraceptives: venousthrombosis, Cochrane Database Syst. Rev. 3 (2014) (CD010813).

[28] J.M. Kemmeren, A. Algra, D.E. Grobbee, Third generation oral contraceptives andrisk of venous thrombosis: meta-analysis, BMJ 323 (7305) (2001) 131–134.

[29] F. Martinez, I. Ramirez, E. Perez-Campos, K. Latorre, I. Lete, Venous and pulmonarythromboembolism and combined hormonal contraceptives. Systematic review andmeta-analysis, Eur J Contracept Reprod Health Care 17 (1) (2012) 7–29.

[30] C. Westhoff, A.M. Kaunitz, T. Korver, W. Sommer, L. Bahamondes, P. Darney,C. Verhoeven, Efficacy, safety, and tolerability of a monophasic oral contraceptivecontaining nomegestrol acetate and 17β-estradiol: a randomized controlled trial,Obstet. Gynecol. 119 (5) (2012) 989–999.

[31] D. Mansour, C. Verhoeven, W. Sommer, E. Weisberg, S. Taneepanichskul,G.B. Melis, I. Sundström-Poromaa, T. Korver, Efficacy and tolerability of a mono-phasic combined oral contraceptive containing nomegestrol acetate and 17β-oes-tradiol in a 24/4 regimen, in comparison to an oral contraceptive containingethinylestradiol and drospirenone in a 21/7 regimen, Eur J Contracept ReprodHealth Care 16 (2011) 430–443.

[32] M. Borenstein, L.V. Hedges, J.T. Higgins, H.R. Rothstein, Introduction to Meta-Analysis, Wiley, Chichester, 2009.

[33] D.F. Stroup, J.A. Berlin, S.C. Morton, I. Olkin, G.D. Williamson, D. Rennie,D. Moher, B.J. Becker, T.A. Sipe, S.B. Thacker, Meta-analysis of observational stu-dies in epidemiology, a proposal for reporting, JAMA 283 (15) (2000) 2008–2012.

[34] E. von Elm, D.A. Altman, M. Egger, S.J. Pocock, P.C. Gøtzsche, J.P. Vandenbroucke,The strengthening the reporting of observational studies in epidemiology (STROBE)statement: guidelines for reporting observational studies, Int. J. Surg. 12 (2014)1495–1499.

[35] G. La Torre, G. Chiaradia, F. Gianfagna, A. De Laurentis, S. Boccia, W. Ricciardi,Quality assessment in meta-analysis, IJPH 3 (2) (2006) 44–50.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

77

[36] J.J. Deeks, J.P.T. Higgins, D.A. Altman, Chapter 9: analysing data and undertakingmeta-analyses, in: J.P.T. Higgins, S. Green (Eds.), Cochrane Handbook forSystematic Reviews of Interventions, Version 5.1.0 [updated March 2011], TheCochrane Collaboration, 2011.

[37] R. DerSimonian, N. Laird, Meta-analysis in clinical trials, Control. Clin. Trials 7(1986) 177–188.

[38] ReviewManager (RevMan), [Computer program] Version 5.3, The Nordic CochraneCenter, The Cochrane Collaboration, Copenhagen, 2014.

[39] K.W. Bloemenkamp, F.R. Rosendaal, H.R. Buller, F.M. Helmerhorst, L.P. Colly,J.P. Vandenbroucke, Risk of venous thrombosis with use of current low-dose oralcontraceptives is not explained by diagnostic suspicion and referral bias, Arch.Intern. Med. 159 (1) (1999) 65–70.

[40] S.S. Jick, J.A. Kaye, S. Russmann, H. Jick, Risk of nonfatal venous thromboembo-lism with oral contraceptives containing norgestimate or desogestrel compared withoral contraceptives containing levonorgestrel, Contraception 73 (6) (2006)566–570.

[41] S.S. Jick, R.K. Hernandez, Risk of non-fatal venous thromboembolism in womenusing oral contraceptives containing drospirenone compared with women usingoral contraceptives containing levonorgestrel: case-control study using UnitedStates claims data, BMJ 342 (2011) d2151.

[42] J. Todd, R. Lawrenson, R.D. Farmer, T.J. Williams, G.M. Leydon, Venous throm-boembolic disease and combined oral contraceptives: a re-analysis of the MediPlusdatabase, Hum. Reprod. 14 (6) (1999) 1500–1505.

[43] Y. Vinogradova, C. Coupland, J. Hippisley-Cox, Use of combined oral contraceptives

and risk of venous thromboembolism: nested case-control studies using theQResearch and CPRD databases, BMJ 350 (2015) h2135.

[44] N. Gronich, I. Lavi, G. Rennert, Higher risk of venous thrombosis associated withdrospirenone-containing oral contraceptives: a population-based cohort study,CMAJ 183 (18) (2011) E1319–25.

[45] R.M. Herings, J. Urquhart, H.G. Leufkens, Venous thromboembolism among newusers of different oral contraceptives, Lancet 354 (9173) (1999) 127–128.

[46] O. Lidegaard, E. Lokkegaard, A.L. Svendsen, C. Agger, Hormonal contraception andrisk of venous thromboembolism: national follow-up study, BMJ 339 (2009) b2890.

[47] S.O. Skouby, Contraceptive use and behavior in the 21st century: a comprehensivestudy across five European countries, Eur J Contracept Reprod Health Care 15(Sup2) (2010) S42–S53.

[48] I. Martinelli, E. Taioli, P. Bucciarelli, S. Akhavan, P.M. Mannucci, Interaction be-tween the G20210A mutation of the prothrombin gene and oral contraceptive use indeep vein thrombosis, Arterioscler. Thromb. Vasc. Biol. 19 (3) (1999) 700–703.

[49] A.L. Nightingale, R.A. Lawrenson, E.L. Simpson, T.J. Williams, K.D. MacRae,R.D.T. Farmer, The effects of age, body mass index, smoking and general health onthe risk of venous thromboembolism in users of combined oral contraceptives, Eur JContracept Reprod Health Care 5 (4) (2000) 265–274.

[50] M.A. Lewis, L.A. Heinemann, K.D. MacRae, R. Bruppacher, W.O. Spitzer, The in-creased risk of venous thromboembolism and the use of third generation proges-togens: role of bias in observational research, the transnational research group onoral contraceptives and the health of young woman, Contraception 54 (1) (1996)5–13.

C. Oedingen et al. Thrombosis Research 165 (2018) 68–78

78