| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Department of Internal Medicine (J.-P.B.), Université de Sherbrooke, Sherbrooke, Quebec J1H 5N4, Canada; and Departments of Biostatistics (D.K.M.), Internal Medicine (P.A.E., J.E.N.), and Obstetrics and Gynecology (J.E.N.), Medical College of Virginia, Virginia Commonwealth University, Richmond, Virginia 23298
Address all correspondence and requests for reprints to: Jean-Patrice Baillargeon, M.D., M.Sc., University of Sherbrooke, Endocrine Division, 3001 12th North Avenue, Sherbrooke, Quebec J1H 5N4, Canada. E-mail: jp.baillargeon{at}usherbrooke.ca.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Objective: We aimed to assess the risk of cardiovascular diseases associated with current use of low-dose combined OCs.
Data Sources: All studies published between January 1980 and October 2002 were searched using MEDLINE, BIOSIS, and Scientific Citations.
Study Selection: Original studies were selected independently by two investigators (J.P.B., P.A.E.) based on inclusion criteria: low-dose combined OC (<50 µg of ethinyl-estradiol); more than 10 cases in low-dose users; clear definition of cases; concurrent controls; and control for age. A third investigator (J.E.N.) adjudicated disagreements. From 2715 identified articles, 14 independent studies were included.
Data Extraction: All data were abstracted by one investigator (J.P.B.) in a systematic manner. Classification of OCs and types of exposure were directly abstracted from studies. Current use was defined as use at the time of the event or within 3 months. Only peer-reviewed studies with definition of events as definite or possible, based on prespecified criteria, were included.
Data Synthesis: The summary risk estimates associated with current use of low-dose OCs were 1.84 [95% confidence interval (CI) = 1.38, 2.44] for myocardial infarctions and 2.12 (95% CI = 1.56, 2.86) for ischemic strokes. The overall summary odds ratio for both outcomes was 2.01 (95% CI = 1.63, 2.48). Second generation OCs were associated with a significant increased risk of both myocardial infarction and ischemic stroke events [1.85 (95% CI = 1.03,3.32) and 2.54 (95% CI = 1.96,3.28), respectively]; and third-generation OCs, for ischemic stroke outcome only [2.03 (95% CI = 1.15,3.57)].
Conclusions: In conclusion, a rigorous meta-analysis of the literature suggests that current use of low-dose OCs significantly increases the risk of both cardiac and vascular arterial events, including a significant risk of vascular arterial complications with third generation OCs.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
50 µg) (5), raising concerns about their use. The risks of venous thromboembolism have been well established (4, 6), but the risks of cardiovascular diseases are still debated. Because a large-scale, long-term controlled trial to evaluate the risks of OCs cannot be performed, the best available evidence comes from cohort and case-control studies. Four case-control studies designed to assess the risk of myocardial infarction (MI) or ischemic stroke (IS) specifically in women using low-dose OCs (ethinyl-estradiol <50 µg, i.e. all second- and third-generation OCs) did not show a significant increase (7, 8, 9, 10). However, many other studies reported results on low-dose OCs only in subgroup analyses, often divided into second- and third-generation OCs, making it difficult to draw overall conclusions.
Previous epidemiological studies have shown increased risk of cardiovascular events associated with OCs in women with hypertension, smoking, or migraine (11, 12). Accordingly, these conditions are now considered contraindications to the use of OCs, particularly in women after age 35 (13, 14). However, OCs are still widely prescribed to other women who do not smoke or have hypertension or migraines but nonetheless have an increased risk for cardiovascular disease at baseline, e.g. women with the metabolic syndrome (15) or polycystic ovary syndrome (PCOS) (16). Even a small degree of increased risk for cardiovascular events associated with OCs in the general population might be greatly amplified in such women and be highly clinically relevant.
Because no rigorous studies exist assessing the cardiovascular disease risk of OCs in women with the metabolic syndrome or PCOS, we conducted a meta-analysis to rigorously review all pertinent studies estimating the risk of cardiac or vascular arterial events associated with the current use of low-dose combined OCs in the population-at-large. It is reasonable to assume, however, that an established cardiovascular risk associated with OCs in the general population likely underestimates that risk in women with the metabolic syndrome or PCOS.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
We reviewed the medical literature published since 1980, when low-dose combined OCs began to dominate the market. A search was conducted from January 1980 to October 2002 using MEDLINE, BIOSIS, and Scientific Citations and the following key words and subject terms: OCs (adverse effects, toxicity) and heart, coronary disease, coronary arteriosclerosis, cardiovascular diseases, myocardial ischemia, vascular diseases, stroke, cerebral infarction, or ischemia. Only studies published in peer-reviewed journals were considered. Bibliographies of pertinent original articles, reviews, and textbooks were searched for additional references.
All potential original articles were first screened based on their titles. Abstracts of selected articles were retrieved and screened again. Those subsequently selected, or those without available abstract, were reviewed in detail for inclusion criteria. The most recent and pertinent review articles were also obtained to search their bibliographies. Two investigators (J.-P.B., P.A.E.) independently screened and applied inclusion criteria to articles. At each step, a third investigator (J.E.N.) adjudicated disagreements.
We used the following inclusion criteria adapted from Gillum et al. (17): 1) specific data on low-dose combined OCs (OCs containing <50 µg ethinyl-estradiol) or second- or third-generation OCs users (as defined in each study); 2) greater than 10 cases of cardiac or vascular arterial events in low-dose OC users; 3) clear definition of IS, myocardial ischemia or MI, or other arterial events (see Data extraction below); 4) cohort or case-control designs with controls gathered within 2 yr of cases; 5) sufficient data provided to determine the odds ratio (OR) or relative risk (RR) and 95% confidence intervals (CIs) comparing OC users with nonusers; 6) controlled for age in study design or analysis; and 7) no later study fully reported the same data.
Data extraction
All data were abstracted by one investigator (J.-P.B.) in a systematic manner. The classification of OCs into second or third generation based on progestogen content, and the types of exposure, were directly abstracted from the selected studies. Definitions of current use of OCs varied from use at the time of the event to use within 3 months.
Although the literature was systematically reviewed for any cardiac or arterial events, studies fulfilling the inclusion criteria only assessed myocardial infarct or stroke outcomes. A definite diagnosis of IS was defined as one confirmed by computed tomography, magnetic resonance imaging, or angiography. A possible diagnosis of IS or cerebral thromboembolic attack (IS and transitory cerebral ischemic attack) was defined by a clinical diagnosis not confirmed by imaging. A definite diagnosis of MI was defined according to diagnostic criteria of the American Heart Association Council on Epidemiology (18), i.e. the presence of symptoms, elevated cardiac-enzyme levels, and electrocardiographic changes. A possible diagnosis of MI was defined by electrocardiographic changes alone.
Results were considered controlled for a potential confounder if the factor was used to match controls or included in multivariate analysis. Risk estimates abstracted were those that controlled for the greatest number of potential confounders by stratification or multivariate analysis.
Statistical analysis
None of the cohort studies met the inclusion criteria, primarily because they did not include sufficient women on low-dose OCs and, therefore, failed to observe greater than 10 cases of cardiac or vascular arterial events. Consequently, all included studies had case-control designs, and ORs were used to estimate the risks. Because the absolute risk of cardiovascular events is low in the population studied (see Discussion), the ORs approximate the RRs. The primary analysis is the meta-analysis of estimated risk for cardiac, vascular, or all cardiovascular events in current users of low-dose combined OCs.
Some studies otherwise fulfilling our inclusion criteria did not report on the current use of low-dose OCs. These studies were not included in this meta-analysis, except for the subgroup analysis on exposure definition (see Table 3
). Our meta-analysis assessed only the risk of cardiovascular events in women currently on low-dose OCs because this is more pertinent for clinicians who have to recommend to their patients whether to start or discontinue OCs. Furthermore, mixing different exposure definitions in the meta-analysis increases the heterogeneity of the results and reduces their applicability to clinical situations. Note that, if multiple analyses were reported in an article, only those including more than 10 cases were considered (as per the inclusion criteria).
|
Seven of 14 studies (20, 21, 22, 23, 24, 25) reported ORs for second- and third-generation OCs but not for all low-dose OC users. In these cases, we combined the ORs for second- and third-generation OCs using the general variance-based method, a fixed-effects method (19). Because subgroup analyses in the same study population cannot be considered as independent random samples, a random-effects model would not be appropriate here. Variances were calculated from adjusted CIs.
Because of the heterogeneity issue with vascular outcome, a sensitivity analysis was performed to evaluate the stability of the summary OR by successively excluding each study, one at a time (17). Subgroup analyses to evaluate sources of heterogeneity and bias were stratified by study design characteristics. These analyses include only studies on current use of low-dose OCs, except for the subgroup analysis on exposure definition, which include all studies meeting our inclusion criteria for the systematic literature review. Regarding this subgroup analysis, one study assessing the risk of IS and reporting on current vs. noncurrent use and past vs. never use (9) was included in both strata (both analyses included >10 cases).
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
)
The systematic retrieval process to identify eligible studies is summarized in Fig. 1.
|
. Notably, we have reviewed systematically the literature for all cardiac and arterial events, but eligible studies only included MIs and ISs. For example, there was no study on peripheral artery disease with enough cases (see inclusion criteria above).
|
)
Only studies assessing the risk of cardiovascular events associated with current use of low-dose OCs were included in the primary analysis, that is, 14 independent studies published in 12 articles (8, 9, 20, 21, 22, 23, 24, 25, 26, 27, 29, 31). Thus, four studies reporting only on the risk associated with past or ever OC use were excluded from this primary analysis. Their ORs are presented in Table 2.
|
|
plots the ORs of each individual study, divided into MI (n = 7) and IS (n = 7) outcomes. The overall summary OR estimating the risk of both cardiovascular outcomes associated with the use of low-dose OCs was 2.01 (95% CI of 1.63, 2.48). The summary OR for MIs was 1.84 (95% CI of 1.38, 2.44), and the individual ORs ranged from 1.40–4.69. These ORs were homogeneous and compared current vs. noncurrent use, except for the study by Rosenberg et al. (8), which compared current vs. never use. Three of seven studies showed a significant increase in the risk estimates (23, 27).
The summary OR for ISs was 2.12 (95% CI of 1.56, 2.86), with a range between 1.18 and 3.63. Five of seven studies found a significant increase in the risk estimates (20, 22, 24, 25, 31). Because of heterogeneous ORs reported in the studies assessing IS, a sensitivity analysis was performed (Fig. 3): ORs ranged from 1.92–2.28, and the lower limits of the 95% CIs never crossed one (1.45–1.67).
|
) showed a symmetrical distribution around the overall summary OR of the individual study ORs, plotted against study sample sizes. This would be expected in the absence of publication bias (19).
|
)Second-generation OCs increased significantly the estimated risks of both cardiovascular outcomes. The risk estimate of MI associated with third-generation OCs was nonsignificant, but third-generation OCs still conferred a significantly increased IS risk.
Four independent studies reported on the risk of cardiovascular events with past use of low-dose OCs compared with never use, and they found a significant reduction of risk both for MI (28) and IS (7, 9) outcomes. Two studies reporting risk estimates of MI events associated with ever vs. never use and current vs. never use (8, 30) resulted in an homogeneous summary OR that was comparable with the summary OR for current vs. noncurrent use.
Most of the studies meeting the inclusion criteria were performed in European countries (20, 21, 22, 23, 24, 25, 26, 27, 29, 31). Those conducted in the United States (8, 9) tended to yield lower summary ORs for both outcomes, which were not statistically significant. The use of controls from the population (9, 20, 21, 25, 26, 29, 31) tended to result in lower risk estimates compared with hospital-based (8, 24, 27) or both (22, 23) types of controls. However, all risk estimates were significantly increased. Regardless of being controlled or not for a specific factor, the summary ORs were still significantly increased, except for two studies not controlled for smoking and diabetes mellitus that assessed the risk of MI events (21, 26), and two studies controlled for obesity that assessed the risk of IS events (9, 22).
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
A subgroup analysis based on type of progestogen showed that second-generation OCs increased 1.9-fold the risk of MI events and 2.5-fold the risk of IS events. Third-generation OCs significantly doubled the risk of IS outcome and increased nonsignificantly the risk of MI outcome by 28%.
A meta-analysis is dependent on the quality of the studies included. Because case-control studies are subject to multiple potential biases, we performed subgroup analyses stratified by study designs. Regarding the type of exposure, past use of OCs seemed to significantly protect against acute MI and IS (7, 9). This suggests that the risk of cardiovascular events associated with current use of low-dose OCs disappears after their discontinuation. It might also be possible that OCs have beneficial effects that persist after OCs discontinuation but that are overcome by acute detrimental effects during use of OCs.
American studies resulted in lower risk estimates than those of European or developing countries and did not achieve statistical significance. Because there was only one study for each outcome, the power of these analyses is reduced. However, the lower risk might be explained by fewer risk factors of cardiovascular disease in American women who receive OCs and, therefore, a lower baseline risk to develop an adverse arterial outcome, as suggested by Heinemann (32).
When only population-based controls were included in the study, the summary risk estimates were lower than studies using hospital-based controls. This suggests a significant effect of sampling bias. Because population-based controls are usually considered methodologically more desirable, this would suggest that current use of low-dose OCs in controls was underestimated using hospital-based data. However, it is still possible that exposure determination in cases was underestimated using population-based data.
Finally, to decrease the potential for confounding, we analyzed studies that controlled for important baseline cardiovascular risk factors in their design (e.g. hypertension and smoking) and found that these women still had a 2- to 3-fold increased risk for cardiovascular disease, with risk estimates higher than uncontrolled studies in general. Because these subgroup analyses did not find lower estimates when correcting for hypertension, smoking, dyslipidemia, or family history of MI or IS, they suggest that our overall risk estimates are not overestimated by including uncontrolled studies.
To our knowledge, there are only four other publications that applied the strict criteria of meta-analysis and assessed the risk of cardiac or arterial vascular outcomes associated with OCs (17, 33, 34, 35). However, our study differs substantively from these in that none of the previous studies was specifically designed to assess all cardiovascular arterial risks associated with the modern use of OCs, i.e. current use of low-dose formulations. Moreover, previous reviews included different types of exposures and/or high-dose OCs in their primary analysis, making it difficult to apply the conclusions for clinicians who have to start or continue low-dose OCs to new or current users. Finally and importantly, previous reviews performed subgroup analyses to answer our main objectives, which are not as robust as primary analyses, because they are not the focus of the initial systematic review of the literature and are usually underpowered.
One meta-analysis, published in 2000 (17), focused on the risk of IS with all dosage OCs and did not include two large, well-designed and recent studies (20, 25). Based on subgroup analyses, the authors also concluded that the risk of IS was significantly increased with the current use of all low-dose OCs (2.08; 95% CI of 1.55, 2.80) and second-generation OCs (3.05; 95% CI of 2.37, 3.91). Third-generation OCs did not reach statistical significance (2.11; 95% CI of 0.96, 4.64).
The second meta-analysis (33), published in 2002, assessed the risk of MI associated with the use of third-generation OCs only and showed a nonsignificant increase (1.13; 95% CI of 0.56, 1.92). However, the authors included estimated ORs from unpublished data and one study comparing OC usage in women who died from myocardial infarct to those who survived (36), which should not be an outcome of interest in such meta-analysis. Another meta-analysis, published in 2003 (34), evaluated only MI as an outcome and included high-dose OCs users. This meta-analysis reported that current OC users have an overall adjusted OR for MI of 2.48 (95% CI of 1.91, 3.22) compared with never users.
Finally, a more recent meta-analysis published in 2004 looked at the association between any exposure to all-dose OCs and the risk of IS and hemorrhagic stroke, which are more associated with hypertension than atherosclerosis. Subgroup analyses resulted in significantly increased risk estimates for low-dose OCs (1.93; 95% CI of 1.61, 2.31), second-generation OCs (2.43; 95% CI of 1.92, 3.09), and third-generation OCs (2.87; 95% CI of 1.84, 4.48). Notably, the subgroup analysis of four cohort studies did not show an increased risk (RR of 0.96; 95% CI of 0.76, 1.22); however, two studies observed few stroke outcomes (one and nine cases), one looked only at hemorrhagic strokes, two looked at all stroke types, and results were very heterogeneous (RR ranged from 0.36–2.4; P = 0.01).
In a large prospective study, Petitti et al. (37) determined the incidences of MI and IS in American women aged 15–44 yr, which were 5.0 in 100,000 women annually and 5.1 in 100,000 women annually, respectively. Assuming that 17.3% of these women were using OCs (1) and that they were all on low-dose formulations, our meta-analysis suggests that 12.7% of all MIs in this group, or 0.6 in 100,000 women per year, were attributable to current use of OCs, as well as 16.2% of all ISs, or 0.8 in 100,000 women per year. However, a significantly increased risk might be compensated by the important health and social benefits resulting from the prevention of unintended pregnancies (4, 17).
If OCs are prescribed to women with an increased absolute risk at baseline, or for reasons other than contraception, the benefits might not justify the risks. Many epidemiological studies have shown increased risks of cardiac events or stroke in women with hypertension, tobacco use, or migraine headache, especially in those aged more than 35 yr (11, 12), which are now established contraindications to the use of OCs (13, 14).
However, an issue that has received little attention is whether other conditions that are associated with increased cardiovascular risk at baseline, and are not currently taken into consideration by physicians when prescribing OCs, should also be considered relative contraindications to OC use. Such conditions would include the metabolic syndrome (15) and PCOS (a condition affecting 6–10% of young women) (16), because both insulin-resistant conditions have been associated with substantially increased risk of cardiovascular disease, as well as a high prevalence of various cardiovascular risk factors, such as reduced plasma high-density lipoprotein cholesterol, increased plasma triglycerides, glucose intolerance, and elevated C-reactive protein. It is prudent to assume that these women would be potentially more susceptible to any cardiovascular risk of OCs identified in the female population-at-large (15, 38, 39). Moreover, with regard to women with PCOS in particular, these women are also treated long-term with OCs, which substantially increases their exposure to the risks of OCs. Because there is an alternative to OCs, i.e. treatment with insulin-sensitizing drugs (39), a doubling in cardiovascular risk estimates in the general population of women, which is likely amplified in PCOS, might prompt physicians to reconsider the use of OCs as first-line therapy in PCOS women not requiring contraception. Studies assessing risks of arterial outcomes in these populations would be important to make recommendations, as done previously with hypertension, smoking, and migraine.
In conclusion, this rigorous meta-analysis of the pertinent literature suggests that low-dose OCs containing second- and third-generation progestogens significantly increase the risk of both cardiac and vascular arterial events. The increased risks seem less robust for the use of third-generation OCs, although the risk remains significant for IS outcome. Similarly, amplitude of these risks might be less if there was population-selection biases in hospital-based studies, although risks are still clinically and statistically significant when these studies are excluded. The modern use of low-dose OCs, limited to healthy women and restricted in time, should not increase noticeably the incidence of these adverse outcomes, which might be outweighed by the benefits of contraception. However, prolonged exposure to low-dose OCs in a population at higher risk may significantly increase the incidence of cardiovascular outcomes and prompt consideration of alternative therapeutic or contraceptive interventions.
|
Footnotes |
|---|
First Published Online April 6, 2005
Abbreviations: CI, Confidence interval; IS, ischemic stroke; MI, myocardial infarction; OC, oral contraceptive; OR, odds ratio; PCOS, polycystic ovary syndrome; RR, relative risk.
Received October 4, 2004.
Accepted March 24, 2005.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  B. A. Frempong, M. Ricks, S. Sen, and A. E. Sumner Effect of Low-Dose Oral Contraceptives on Metabolic Risk Factors in African-American Women J. Clin. Endocrinol. Metab., June 1, 2008; 93(6): 2097 - 2103. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M Petri Sex hormones and systemic lupus erythematosus Lupus, May 1, 2008; 17(5): 412 - 415. [Abstract] [PDF]
|
|
|
|
 |
|
 J. V. Johnson, J. Lowell, G. J. Badger, J. Rosing, S. Tchaikovski, and M. Cushman Effects of Oral and Transdermal Hormonal Contraception on Vascular Risk Markers: A Randomized Controlled Trial Obstet. Gynecol., February 1, 2008; 111(2): 278 - 284. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. E. Nestler Metformin for the Treatment of the Polycystic Ovary Syndrome N. Engl. J. Med., January 3, 2008; 358(1): 47 - 54. [Full Text] [PDF]
|
|
|
|
|
|
E. A. Lawson, K. K. Miller, V. A. Mathur, M. Misra, E. Meenaghan, D. B. Herzog, and A. Klibanski Hormonal and Nutritional Effects on Cardiovascular Risk Markers in Young Women J. Clin. Endocrinol. Metab., August 1, 2007; 92(8): 3089 - 3094. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z. T. Bloomgarden Gut Hormones, Obesity, Polycystic Ovarian Syndrome, Malignancy, and Lipodystrophy Syndromes Diabetes Care, July 1, 2007; 30(7): 1934 - 1939. [Full Text] [PDF]
|
|
|
|
|
|
M. Luque-Ramirez, F. Alvarez-Blasco, J. I. Botella-Carretero, E. Martinez-Bermejo, M. A. Lasuncion, and H. F. Escobar-Morreale Comparison of Ethinyl-Estradiol Plus Cyproterone Acetate Versus Metformin Effects on Classic Metabolic Cardiovascular Risk Factors in Women with the Polycystic Ovary Syndrome J. Clin. Endocrinol. Metab., July 1, 2007; 92(7): 2453 - 2461. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. Damm, E. R. Mathiesen, K. R. Petersen, and S. Kjos Contraception After Gestational Diabetes Diabetes Care, July 1, 2007; 30(Supplement_2): S236 - S241. [Full Text] [PDF]
|
|
|
|
 |
|
 B. N. Torgrimson, J. R. Meendering, P. F. Kaplan, and C. T. Minson Endothelial function across an oral contraceptive cycle in women using levonorgestrel and ethinyl estradiol Am J Physiol Heart Circ Physiol, June 1, 2007; 292(6): H2874 - H2880. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A Pezzini, M Grassi, L Iacoviello, E Del Zotto, S Archetti, A Giossi, and A Padovani Inherited thrombophilia and stratification of ischaemic stroke risk among users of oral contraceptives J. Neurol. Neurosurg. Psychiatry, March 1, 2007; 78(3): 271 - 276. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Nader and E. Diamanti-Kandarakis Polycystic ovary syndrome, oral contraceptives and metabolic issues: new perspectives and a unifying hypothesis Hum. Reprod., February 1, 2007; 22(2): 317 - 322. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. L. Turgeon, M. C. Carr, P. M. Maki, M. E. Mendelsohn, and P. M. Wise Complex Actions of Sex Steroids in Adipose Tissue, the Cardiovascular System, and Brain: Insights from Basic Science and Clinical Studies Endocr. Rev., October 1, 2006; 27(6): 575 - 605. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Endocrinology | Endocrine Reviews | J. Clin. End. & Metab. |
| Molecular Endocrinology | Recent Prog. Horm. Res. | All Endocrine Journals |
, In these studies, the ORs were presented separately for second- and third-generation OCs, and they were combined using fixed-effect method to determine the ORs for all low-dose OCs. 
, Results reported for developing countries only. 
, Results reported for European countries only. ¦, Significant heterogeneity in the study results using a general variance-based method.
