Combined Oral Contraceptives and Venous Thrombosis Risk
Combined Oral Contraceptives and Venous Thrombosis Risk
Publications of interest were observational studies (cohort or (nested) case-control studies) that included healthy women using combined oral contraceptives. The primary outcome of interest was a fatal or non-fatal first event of venous thrombosis with the main focus on deep venous thrombosis or pulmonary embolism. Publications with a minimum of 10 events in total were eligible.
The following databases were searched: PubMed (984 articles retrieved), Embase (1,339), Web of Science (306), Cochrane (57), Cumulative Index to Nursing and Allied Health Literature (122), Academic Search Premier (197), and ScienceDirect (105). Our search terms consisted of MeSH headings and subheadings, text words, and word variations for "combined oral contraceptive," "estrogens," "progestogens," and "venous thromboembolism." This search strategy was amended for each database. We searched each database from inception until 22 April 2013 (date of final search). No language restriction was applied. In addition to database searches, we checked references of potential interesting publications.
A standard form was used to select publications. Two investigators (BHS, MdB) independently assessed publications for eligibility. Titles and abstracts were screened and if deemed potentially relevant, full text articles were retrieved. Any disagreements between the investigators were discussed and if necessary, a third reviewer (OMD) was asked to resolve disagreements. In case of multiple publications from the same study, the publication with the most updated or the most inclusive data was included. Web Appendix 1 shows details of the search strategy.
Two investigators (BHS and MdB) independently extracted data using a standard form. Data were extracted on type of combined oral contraceptive (dose and type of oestrogen and progestogen), crude numbers for exposure and outcome, crude and adjusted risk estimates, and variables adjusted for in the analysis. In case of incomplete data for dose or type of oestrogen or progestogen, authors were approached for extra information. In total, 10 authors were sent an email on 25 July 2012 and if necessary, a reminder on 20 August 2012. Of these authors, 80% replied to our emails. If provided, data for the absolute thrombosis risk in non-users were extracted from the cohort studies. For sensitivity analyses, data for the funding source and first time use were abstracted.
Risk of bias assessment was based on design features that could potentially bias the association between exposure and outcome. We assessed adequacy of exposure (oral contraceptive) and outcome (venous thrombosis) measurement, loss to follow-up (cohort studies), and adequacy of control selection (case-control studies). Women are more likely to remember that they used oral contraceptives than what specific preparation they used. Therefore, assessment of the type of combined oral contraceptive through an interview or questionnaire was classified as high risk of bias, and information from a prescription database as low risk. Only 25-33% of patients presenting with clinical symptoms suggestive of venous thrombosis are objectively diagnosed with venous thrombosis. Therefore, studies with objective confirmation in all patients were judged as low risk of bias. Venous thrombosis was considered objectively confirmed when a deep venous thrombosis was diagnosed by plethysmography, ultrasound examination, computed tomography, or venography; or when pulmonary embolism was diagnosed by ventilation-perfusion scanning, spiral computed tomography, or pulmonary angiography. Less than 10% loss to follow-up was considered to represent a low risk of bias. For case-control studies, controls selected from hospital populations was considered to confer a high risk of bias.
There is no generally accepted way to classify combined oral contraceptives according to generations of progestogens. In this analysis, the most common classification system was used, which is in line with biological properties per group and is reflected in their effects on levels of sex hormone binding globulin. For the aim of the present network meta-analysis per generation of progestogen, the following progestogens were classified as first generation; lynestrenol and norethisterone. Norgestrel and levonorgestrel were categorised as second generation progestogens; and desogestrel, gestodene, and norgestimate were classified as third generation progestogens. This classification was irrespective of ethinylestradiol dose. Publications reporting on generations according to another classification were included. To assess the influence of combining different classifications, we performed an analysis restricted to studies using the above described classification.
Many different combined oral contraceptives are available. We selected 10 frequently prescribed oral contraceptives for the network meta-analysis:
We categorised 20LNG, 30LNG, and 50LNG as second generation progestogens, and 20GSD, 30GSD, 20DSG, 30DSG, and 35NRG as third generation progestogens. 35CPA and 30DRSP were not used in this classification by generations.
A network meta-analysis was conducted per generation of progestogen in a combined oral contraceptive and per selected oral contraceptive preparation. We used an extension of frequentist random effects models for mixed multiple treatment comparisons. The network meta-analysis was performed with the mvmeta command for Stata, as described by White and colleagues. We used crude data from a 2×2 table in the analysis. Odds ratios, risk ratios, or rate ratios and appropriate variances were computed and combined in the analysis leading to an overall relative risk. For publications with zero events in one cell of a 2×2 table, all cells of that 2×2 table were inflated by adding 0.5. If more than one study provided data for the same stratum (that is, data for generations of progestogen or on specific contraceptive preparations), we checked consistency of the results. An interaction term was added to the model to estimate the difference in results from direct and indirect evidence. All potential interactions were tested in an overall test to determine whether there were any inconsistencies in our network meta-analysis.
The following sensitivity analyses were planned: per study design, per funding source (whether industry sponsored or not), within first time users, and according to risk of bias. All statistical analyses were performed with Stata, version 12.0 (Statacorp LP).
Methods
Search Strategy and Selection Criteria
Publications of interest were observational studies (cohort or (nested) case-control studies) that included healthy women using combined oral contraceptives. The primary outcome of interest was a fatal or non-fatal first event of venous thrombosis with the main focus on deep venous thrombosis or pulmonary embolism. Publications with a minimum of 10 events in total were eligible.
The following databases were searched: PubMed (984 articles retrieved), Embase (1,339), Web of Science (306), Cochrane (57), Cumulative Index to Nursing and Allied Health Literature (122), Academic Search Premier (197), and ScienceDirect (105). Our search terms consisted of MeSH headings and subheadings, text words, and word variations for "combined oral contraceptive," "estrogens," "progestogens," and "venous thromboembolism." This search strategy was amended for each database. We searched each database from inception until 22 April 2013 (date of final search). No language restriction was applied. In addition to database searches, we checked references of potential interesting publications.
A standard form was used to select publications. Two investigators (BHS, MdB) independently assessed publications for eligibility. Titles and abstracts were screened and if deemed potentially relevant, full text articles were retrieved. Any disagreements between the investigators were discussed and if necessary, a third reviewer (OMD) was asked to resolve disagreements. In case of multiple publications from the same study, the publication with the most updated or the most inclusive data was included. Web Appendix 1 shows details of the search strategy.
Data Collection
Two investigators (BHS and MdB) independently extracted data using a standard form. Data were extracted on type of combined oral contraceptive (dose and type of oestrogen and progestogen), crude numbers for exposure and outcome, crude and adjusted risk estimates, and variables adjusted for in the analysis. In case of incomplete data for dose or type of oestrogen or progestogen, authors were approached for extra information. In total, 10 authors were sent an email on 25 July 2012 and if necessary, a reminder on 20 August 2012. Of these authors, 80% replied to our emails. If provided, data for the absolute thrombosis risk in non-users were extracted from the cohort studies. For sensitivity analyses, data for the funding source and first time use were abstracted.
Risk of bias assessment was based on design features that could potentially bias the association between exposure and outcome. We assessed adequacy of exposure (oral contraceptive) and outcome (venous thrombosis) measurement, loss to follow-up (cohort studies), and adequacy of control selection (case-control studies). Women are more likely to remember that they used oral contraceptives than what specific preparation they used. Therefore, assessment of the type of combined oral contraceptive through an interview or questionnaire was classified as high risk of bias, and information from a prescription database as low risk. Only 25-33% of patients presenting with clinical symptoms suggestive of venous thrombosis are objectively diagnosed with venous thrombosis. Therefore, studies with objective confirmation in all patients were judged as low risk of bias. Venous thrombosis was considered objectively confirmed when a deep venous thrombosis was diagnosed by plethysmography, ultrasound examination, computed tomography, or venography; or when pulmonary embolism was diagnosed by ventilation-perfusion scanning, spiral computed tomography, or pulmonary angiography. Less than 10% loss to follow-up was considered to represent a low risk of bias. For case-control studies, controls selected from hospital populations was considered to confer a high risk of bias.
Classification of Type of Combined Oral Contraceptive
There is no generally accepted way to classify combined oral contraceptives according to generations of progestogens. In this analysis, the most common classification system was used, which is in line with biological properties per group and is reflected in their effects on levels of sex hormone binding globulin. For the aim of the present network meta-analysis per generation of progestogen, the following progestogens were classified as first generation; lynestrenol and norethisterone. Norgestrel and levonorgestrel were categorised as second generation progestogens; and desogestrel, gestodene, and norgestimate were classified as third generation progestogens. This classification was irrespective of ethinylestradiol dose. Publications reporting on generations according to another classification were included. To assess the influence of combining different classifications, we performed an analysis restricted to studies using the above described classification.
Many different combined oral contraceptives are available. We selected 10 frequently prescribed oral contraceptives for the network meta-analysis:
20 μg ethinylestradiol with levonorgestrel (20LNG)
30 μg ethinylestradiol with levonorgestrel (30LNG)
50 μg ethinylestradiol with levonorgestrel (50LNG)
20 μg ethinylestradiol with gestodene (20GSD)
30 μg ethinylestradiol with gestodene (30GSD)
20 μg ethinylestradiol with desogestrel (20DSG)
30 μg ethinylestradiol with desogestrel (30DSG)
35 μg ethinylestradiol with norgestimate (35NRG)
35 μg ethinylestradiol with cyproterone acetate (35CPA)
30 μg ethinylestradiol with drospirenone (30DRSP)
We categorised 20LNG, 30LNG, and 50LNG as second generation progestogens, and 20GSD, 30GSD, 20DSG, 30DSG, and 35NRG as third generation progestogens. 35CPA and 30DRSP were not used in this classification by generations.
Statistical Analysis
A network meta-analysis was conducted per generation of progestogen in a combined oral contraceptive and per selected oral contraceptive preparation. We used an extension of frequentist random effects models for mixed multiple treatment comparisons. The network meta-analysis was performed with the mvmeta command for Stata, as described by White and colleagues. We used crude data from a 2×2 table in the analysis. Odds ratios, risk ratios, or rate ratios and appropriate variances were computed and combined in the analysis leading to an overall relative risk. For publications with zero events in one cell of a 2×2 table, all cells of that 2×2 table were inflated by adding 0.5. If more than one study provided data for the same stratum (that is, data for generations of progestogen or on specific contraceptive preparations), we checked consistency of the results. An interaction term was added to the model to estimate the difference in results from direct and indirect evidence. All potential interactions were tested in an overall test to determine whether there were any inconsistencies in our network meta-analysis.
The following sensitivity analyses were planned: per study design, per funding source (whether industry sponsored or not), within first time users, and according to risk of bias. All statistical analyses were performed with Stata, version 12.0 (Statacorp LP).