Criteria for considering studies for this review
Types of studies
Myocardial infarction and ischemic stroke are potential side effects of combined oral contraceptive use. Previous research has suggested that results of observational studies are credible when studying side effects of medication Vandenbroucke 2004. This was supported by a recent meta-analysis that showed no difference between the risk of side-effects assessed in meta-analyses of experimental data and the risk of side-effects assessed in meta-analyses of observational data Golder 2011. Therefore, for the current review, observational studies with a cohort, case-control or nested case-control design will be included. In addition, if available, data from randomized controlled trials will be analyzed.
Types of participants
The participants are women in the reproductive age group (18-50 years) who use combined oral contraceptives and women in the reproductive age group who do not use combined oral contraceptives. Studies on women using postmenopausal hormone therapy, non-oral contraceptives or progestagen only contraceptives will be excluded.
Types of interventions
The risk of myocardial infarction and ischemic stroke will be compared between combined oral contraception users and non-users. Both previous combined oral contraception users and never users will be considered to be non-users. In addition we will compare previous users to non-users. The risk of myocardial infarction and ischemic stroke will be assessed for different types of combined oral contraceptive preparations. Combined oral contraceptives will be categorized according to their generation (progestagens), and also according to the dose of estrogen and the type of progestagen used.
Types of outcome measures
The outcome measures of interest are objectively diagnosed fatal or non-fatal first myocardial infarction or ischemic stroke. A myocardial infarction will be classified as objectively confirmed if diagnosed based on a medical examination and pain assessment combined with an electrocardiogram (ECG), serum cardiac biomarkers or other specified strict diagnostic criteria of myocardial infarction, or by autopsy examination. Ischemic stroke will be classified as objectively confirmed if a sudden onset focal neurological deficit is diagnosed on the basis of a medical history and neurological examination combined with brain imaging, or by autopsy examination.
The risk of developing either myocardial infarction or ischemic stroke in combined oral contraceptive users compared with non-users will be quantified by obtaining or computing crude odds ratios, risk ratios, rate ratios and the appropriate variances from each study. These estimates will be combined to an overall relative risk. In case of zero events in a single cell of a 2x2 table, all cells of that 2x2 table will be inflated by 0.5 in order to obtain a valid risk estimate.
The primary outcome is fatal or non-fatal myocardial infarction or ischemic stroke.
Search methods for identification of studies
The search will be created in association with an expert librarian (C Manion, Cochrane Collaboration).
The following databases will be searched: the Cochrane Database of Systematic Reviews (1998 to search date), DARE (1995 to search date), MEDLINE (1966 to search date), EMBASE (1980 to search date), Web of Science (1900 to search date), CINAHL (1982 to search date), Academic search Premier (1997 to search date), and ScienceDirect (1995 to search date). The study search will be performed without language restrictions.
Searching other resources
The references of the selected studies and of reviews will additionally be checked in case any relevant studies are not captured by our search strategy.
Data collection and analysis
The study results will be used to compare the relative risk of myocardial infarction and ischemic stroke between users and non-users of combined oral contraceptives, and between women using different types and doses of combined oral contraceptives. Most studies include women without oral contraception or women who use preparations containing levonorgestrel with 30mcg of ethinylestradiol as a reference group. Therefore, not all combined oral contraceptive preparations will have been directly compared in the literature. To solve this problem, we will perform a network meta-analysis White 2012. This type of analysis allows two preparations to be indirectly compared via a common comparator. The evidence from both direct and indirect comparisons is then combined in a single analysis.
Selection of studies
In order to select relevant studies for this review, the title and abstract of all studies retrieved from the search strategy will be independently evaluated by two reviewers (REJR, FMH). This will be done using standard piloted forms and specific inclusion and exclusion criteria. Any disagreements will be resolved by consensus. A third author (OMD) will be consulted if disagreement persists.
Data extraction and management
Data will be independently extracted by two reviewers (REJR and FMH) using standard piloted forms. Any disagreements will be resolved by consensus. If disagreements persist, a third author (OMD) will be consulted.
Assessment of risk of bias in included studies
Our risk of bias assessment was equipped for observational studies and adapted from the Newcaste Ottowa scale. The risk of bias in the included observational studies will be examined based on four aspects that may affect the association between the exposure (combined oral contraceptives) and the outcome (myocardial infarction and ischemic stroke).
Firstly, exposure to combined oral contraception had to be confirmed through a prescription base in order for the risk of bias to be classified as ‘low’. Other, less objective, methods such as interviews and questionnaires were classified as a high risk of bias, as research has shown that women have difficulty accurately recalling the type of preparations they used Nischan 1993; Norell 1998.
Secondly, the diagnosis of a myocardial infarction or ischemic stroke had to be ascertained by objective measures. Studies in which myocardial infarction had been diagnosed on the basis of an electrocardiogram (ECG), serum cardiac biomarkers or other specified strict diagnostic criteria of myocardial infarction, or by autopsy examination, were classified as having a low risk of bias. For ischemic stroke these criteria were a neurological examination combined with brain imaging or other specified strict diagnostic criteria of ischemic stroke, or autopsy examination.
In cohort studies, loss to follow up can lead to biased risk estimates. We classified studies with <10% loss to follow up as having a low risk of bias.
Finally, in case-control selection, the selection of controls affects the validity of the results. We classified case-control studies including controls from the source population of the cases (i.e. controls from the same neighbourhood as the cases who would most likely have been admitted to the same hospital as the cases if they had developed arterial thrombosis) as low risk of bias Grimes 2005.
As myocardial infarction and ischemic stroke are side effects of using combined oral contraceptives, and it is unethical to perform a randomized controlled trial for side effects alone, we do not anticipate finding any randomized controlled trials on this topic. However, if such studies are found, the risk of bias will be assessed according to recommended principles Higgins 2011. No aggregate risk of bias score was used, as this is generally discouraged Juni 1999.
The risk of bias will be assessed independently by two reviewers (REJR and FMH) using a standard piloted form. Both reviewers are trained in Clinical Epidemiology and study methodology. Any persistent disagreement will be resolved by consensus or discussion with a third author (OMD). The risk of bias assessment will not be used to accept or reject studies.
Measures of treatment effect
Effect measures will be extracted from all studies. Effect measures can either be odds ratios (from case-control studies), risk ratios or hazard ratios (from cohort studies or randomized controlled trials). The accompanying 95% confidence intervals will be extracted or recalculated based on standard errors or p-values. Where possible, risk estimates adjusted for age and calendar time will be obtained in order to reduce the risk of bias. If such risk estimates are not available, results from adjusted models will be presented as reported on by the authors. In case no adjustments have been made, the crude risk estimates will be presented. It is expected that in case-control studies adjustment for age and calendar time is dealt with by design (matching).
Unit of analysis issues
Current use of combined oral contraceptives, stratified according to the dose of ethinylestradiol and the type of progestagen, will be analyzed in women without a history of myocardial infarction or ischemic stroke. The effect of previous combined oral contraception use on the risk of cardiovascular disease will additionally be studied if data are available.
Dealing with missing data
Only participants with complete data on exposure to combined oral contraceptives and the outcomes myocardial infarction and ischemic stroke will be included.
Assessment of heterogeneity
In order to assess the heterogeneity of study results, we will compare the standard deviation/ variance of the effect between studies. In order to obtain valid effect measures, strata with no events will be inflated to 0.5. For results from standard meta-analytic techniques statistical measures of heterogeneity will be presented (Chi2 test and I2 statistics). For the network meta-analysis, the consistency of the network will be checked White 2012.
Assessment of reporting biases
Reporting biases (e.g. publication bias) will be investigated using a funnel plot.3
Data from studies with similar designs, interventions and outcome measures will be combined. Results will be organized by the generation of the combined oral contraceptive preparation, as well as by the type and dose of estrogen and progestagen used. The risk of bias will be assessed by a an adapted version of the Newcaste Ottowa scale as described in the ‘Assessment of risk of bias in included studies’ section.
An extension of frequentist random-effects models for mixed multiple treatment comparisons will be used to analyze the data. Network meta-analysis will be performed with the mvmeta command for STATA as described by White et al White 2012. If data on the same stratum are provided by more than one study, the consistency of the results will be evaluated through interaction analysis. An interaction term will be added to the model in order to estimate the difference in results from direct and indirect evidence. All potential interactions will be tested in an overall test to determine whether there are any inconsistencies in our network meta-analysis. Inconsistencies will only be checked when there is more than one study comparing the same groups
Subgroup analysis and investigation of heterogeneity
The outcomes myocardial infarction and ischemic stroke will be pooled as well as analyzed separately. The standard deviation/ variance of the effect between studies will be compared in order to assess the heterogeneity of study results, as described in the section ‘Assessment of heterogeneity’.
Sensitivity analyses will be performed per study design, per funding source (whether studies are industry sponsored or not) and according to the risk of bias.