Is early age at menarche a risk factor for endometriosis? A systematic review and meta-analysis of case-control studies

Of the 40 studies identified from the database search (list available on request), only 13 fulfilled the predefined entry criteria (see Supplemental Table 2 , available online, for excluded studies). Five additional studies were identified from the reference list search. These 18 case-control studies, involving 3,805 cases of surgically diagnosed endometriosis and 9,526 controls, were published between 1986 and 2010 ( Table 1 ) (13–30) . The studies were conducted in the United States (n = 4), Italy (n = 4), Canada (n = 2), and the United Kingdom (n = 2); the other six studies were conducted in Australia, Belgium, People's Republic of China, Malaysia, and Spain. Women were generally 18–49 years old, although one study included women up to the age of 69 years (17) . The study population in five studies (15, 19, 20, 25, 28) comprised infertile women, although one of those studies (25) used fertile women as controls and another (20) enrolled women who had male factor infertility as controls. All but four studies (14, 20, 22, 30) prospectively recruited women with endometriosis and, although 11 studies defined early age at menarche as <12 years old, in four studies (17–19, 24) it was defined as ≤12 years old. Three studies expressed outcomes as means and medians without defining cutoff ages (21, 23, 30) . The Newcastle–Ottawa quality scores ranged from 4–8 and the mean score for all 18 studies was 5.56 (±SD 1.25). Effect sizes did not significantly vary with quality scores ( Supplemental Fig. 1 , available online). As shown in Table 2 , there was careful selection of cases in included studies, as only surgically confirmed cases were recruited and extent of disease was mostly described in detail. Cases were largely representative of source populations, reducing the risk of selection bias. In two studies, however, patients were reviewed retrospectively for inclusion, with some risk of bias in the case selection (20, 22) . Eight studies recruited hospital controls who were either healthy (13) or had diverse conditions such as gynecological diseases (23, 29) , acute illnesses (16, 26) , trauma (17) , infertility (20) , and live birth (25) . Studies either used community controls (14, 18, 24, 30) or controls sampled from the same cohort as cases (15, 18, 21, 22, 27, 28) . However, in only eight of the studies was endometriosis ruled out in controls at laparoscopy (15, 19–22, 27–29) . All except two of those eight studies were hospital patient-controlled studies, which sampled controls from the cohort of women undergoing laparoscopy (20, 29) . The overall performance of the included studies on comparability of participants was inadequate. Ten studies (14–17, 20, 24–26, 29, 30) adequately controlled for potential confounders, although only four of those controlled for the potential confounding effect of adult BMI (15, 20, 24, 30) . Two studies failed to control for any potential confounders, thereby limiting the comparability of the study groups (18, 21) . The overall performance of the included studies on assessment of exposure was poor. In six of the studies, it was not clear whether exposure ascertainment was blinded (16, 17, 21, 23, 26, 29) . Indeed, either the participants or the trained interviewers who collected exposure information in these studies may have been aware of participants' disease status at the time of interview. Effect sizes were positive (range, 0.05–0.56) in 13 of the 18 studies (i.e., early menarche associated with greater risk of endometriosis) (13–21, 24–26) . Five effect sizes (22, 23, 27–29) were negative (i.e., early menarche associated with reduced risk of endometriosis; range, −1.09 to −0.08). Effect sizes suggested statistically significantly greater risk of endometriosis with early menarche in four studies (18, 20, 21, 30) and significantly lesser risk in two studies (23, 27) . In meta-analysis, a pooled effect size of 0.10 (range, −0.01–0.21) was found ( Fig. 1 ), suggesting that women with endometriosis were 0.10 SDs of age (in years) younger than controls at menarche. This “small” (31) association between early age at menarche and the risk of endometriosis was, however, not statistically significant. An effect of this size, interpreted using the “Common Language Effect Size” approach of McGraw and Wong (32) , implies there is a 53% chance that a woman with endometriosis was younger at menarche than a woman without endometriosis if both individuals were chosen at random from a population. In random effects meta-analysis, a high amount of variation across included studies was explained by heterogeneity rather than chance (χ² = 61.92, df  = 17; P =.000; I² = 72.5%). The effect of this residual heterogeneity on the results was investigated in sensitivity analyses. As shown in the funnel plot in Supplemental Figure 2 , visual examination may suggest the presence of funnel plot asymmetry. However, Egger's method to test statistically for the presence of funnel plot asymmetry ( Supplemental Fig. 3 , available online) shows the regression line to have a positive slope, with no evidence for asymmetry ( t  = −1.31, P  = .21, 95% confidence interval [CI] −4.06–0.95). Iterative removal of primary studies from the meta-analysis suggested that two studies (23, 27) with a relatively small sample size may have disproportionately influenced the pooled effect size. After removing each of the other 16 studies, the pooled estimate ranged from 0.05–0.10 and remained nonsignificant. However, after removing these small negative studies, the pooled estimate was 0.15 (95% CI 0.10–0.21). Other sensitivity analyses were based on a priori stated characteristics of the populations and study designs. Five studies assessed infertile women only (15, 19, 20, 25, 28) , although one of those studies (25) compared them to fertile controls. When these five studies only were included in the meta-analysis, residual heterogeneity (measured by the I 2 ) decreased to 58% and the summary effect size increased to 0.11 (95% CI −0.06–0.29). Of the 18 studies included in this review, 2 enrolled only cases with minimal-to-mild endometriosis. Six studies included no information on the disease stage, eight included women with all revised American Fertility Society (AFS) stages, and 2 studies included women with stage III/IV disease. Of the eight studies that enrolled cases of all stages, three predominantly included moderate-to-severe cases (mean, 58% of all cases) and five predominantly included minimal-to-mild cases (mean, 67% of all cases). The pooled effect sizes for the studies with more minimal-to-mild and moderate-to-severe cases were 0.02 (95% CI −0.26–0.29) and 0.32 (95% CI 0.04–0.59), respectively. Heterogeneity remained high in all scenarios within this group of sensitivity analyses. Ten studies adequately controlled for important confounders, thus ensuring comparability of cases and controls (14–17, 20, 24–26, 29, 30) . When these studies only were included in the meta-analysis ( Fig. 2 ), the pooled effect size was 0.15 (95% CI 0.08–0.22) and I 2 was 0, suggesting that the probable source of the variation seen across the 18 included studies was the lack of comparability of cases and controls arising from variation in the adequacy of control for potential confounders.

In this meta-analysis of published case-control studies evaluating the association between age at menarche and endometriosis risk, we found a small, but not statistically significant, increase in risk of endometriosis with early age at menarche (defined as <12 years old). There was substantial heterogeneity across included studies over and above what would be explained by chance alone. Sensitivity analyses suggested that this heterogeneity was explained principally by variations in respect of control of potential confounders of the relationship between age at menarche and endometriosis in individual studies. Consequently, limiting meta-analysis to studies that controlled more rigorously for potential confounders eliminated heterogeneity and suggested that early age at menarche was significantly associated with a higher risk of endometriosis. Smaller studies are, on average, conducted and analyzed with less methodological rigor than larger studies and trials of lower quality also tend to show the larger effects (12) . In this meta-analysis, two small studies (one with a relatively large effect) caused the pooled effect size to tend toward the null value. Their exclusion yielded a larger pooled effect size that suggested that women who were younger at menarche have a significantly higher risk of endometriosis than those who were older. The inverse relationship between age at menarche and risk of endometriosis was reported previously in a prospective cohort study of fertile and infertile premenopausal women (33) . Our study, however, represents the first systematic attempt to review the literature on the relationship between age at menarche and endometriosis risk, and provides a quantitative estimate of the relationship, with careful attention given to understanding the sources of heterogeneity in included primary studies. It uses valid methods of data synthesis that overcome limitations commonly presented by primary studies reporting results as continuous and binary outcomes. The study highlights the effects that inadequacies in case-control design can have, and has particular relevance to the many other putative risk factors of endometriosis in the literature (5) . Well-designed case-control studies of nongenetic risk factors of endometriosis should enroll newly diagnosed cases, collect exposure information predating symptom onset, and use controls representative of the population from which cases are drawn (such as community controls or controls recruited consecutively from the same clinics as cases), with data collected on key confounding factors to allow for adjusted or matched analyses (34) . The study of newly diagnosed cases should mitigate the potential bias arising from a change in behavior upon awareness of disease status, although such changes may have occurred already from the time of symptom onset, which in a condition such as endometriosis often precedes diagnosis by many years (35) . Collecting information on exposure that predated symptom onset is therefore important, but as such information in case-control studies is collected at the time of diagnosis, differential recall between cases and controls may still produce biased results. It can be argued, however, that this is unlikely to be an important consideration for an exposure such as recall of age at menarche, unless patients are aware of the hypothesis. The characteristics of the study population in a case-control study of endometriosis are critical to the validity of its findings. To allow generalizability of results, cases should ideally be representative of the general population, but—owing to the lack of a noninvasive diagnostic tool—studies generally recruit as cases women scheduled for laparoscopic investigations to diagnose or rule out endometriosis. As infertility is often a reason for laparoscopy in these women, the frequency of infertile women in a population of cases is artificially raised by this selection mechanism (16) . Although this may sufficiently complicate interpretation of findings to warrant studying or analyzing fertile and infertile populations of women with endometriosis separately, the pooled effect size for studies of infertile women only did not differ from the reported pooled effect for other 12 studies (0.11, 95% CI −0.06 to −0.29 vs. 0.09, 95% CI −0.05 to −0.23). Similarly, in the cohort study by Missmer et al. (33) , the risk of endometriosis associated with early age at menarche did not significantly differ in infertile women and women without past or concurrent infertility. It has been suggested that moderate-to-severe, rather than minimal-to-mild endometriosis, represents progressive disease, as the latter may only be a transient phase in an ongoing process that often results in cytolysis of recently implanted endometrial cells (36) . We found in this review that on analysis of primary studies of moderate-to-severe disease in exclusion of studies of minimal-to-mild disease, the size and statistical significance of the association between early menarche and endometriosis increased. In light of this finding, we cautiously suggest that early menarche may be associated with the risk of moderate-to-severe, not minimal-to-mild, endometriosis. Ideally, a case-control study should initially define a source population precisely, from which cases and controls are then randomly sampled. In reality and with specific regard to endometriosis, this means that a source population should be defined explicitly, and should then generate the cases attending for care at a clinic, controls being also sampled randomly from that population. This explicit identification of a source population in endometriosis studies is, however, often unrealistic except in circumstances where a population registry can be compiled. Consequently, in most case-control studies of endometriosis, the source population is defined secondarily to the definition of a case-finding mechanism (e.g., voluntary attendance for care because of symptoms). This secondary definition of a source population on the basis of an identified case series complicates control selection as it is then difficult to demonstrate that controls are members of the same population as cases at the time of sampling. These difficulties notwithstanding, control selection needs to focus on endometriosis-free women who are representative of the population from which cases are drawn. This is especially difficult for endometriosis. Consequently, control women undergoing laparoscopy for sterilization are unlikely to be representative of the symptomatic population from which cases were drawn; indeed community or symptomatic hospital-based controls would be more representative (34) . Controls sampled from women with a negative laparoscopy (who are members of the same case series as women who had a positive laparoscopy), would ostensibly be representative of the source population if that population was explicitly defined before case selection, and not secondarily to case series identification. Otherwise, it may be difficult to establish that cases and controls identified through clinics for benign women's health symptoms are representative of the general population in terms of exposure profiles. Most community- and hospital-based controls in the primary studies in this review did not have endometriosis ruled out by laparoscopy, raising the possibility of disease misclassification. Furthermore, hospital-based controls should ideally not have conditions related to the exposure of interest. In one study (29) , some hospital controls had ovarian tumors, which have been linked positively with early menarche (37) . Misclassification and use of controls with exposure-related conditions also potentially alter the relationship between age at menarche and endometriosis risk. In addition to other important potential confounders, such as age and socioeconomic status, adult BMI confounds the relationship between early age at menarche and endometriosis risk, being inversely related to both early age at menarche and the risk of endometriosis (38) . Only 10 of the 18 studies adequately controlled for potential confounders. As indicated by the results of the sensitivity analyses, residual heterogeneity was due largely to the inclusion of studies with less rigorous control of confounding. The potential for misclassification of disease in the primary studies means that the actual pooled effect size found in our meta-analysis ought to be viewed with some caution. All cases were diagnosed through laparoscopy, which may not be fail proof as evidenced by the reported intraobserver and interobserver agreements for visualization of endometriotic lesions during the procedure (39) . The presence of disease misclassification would, however, have underestimated the relationship between age at menarche and endometriosis risk. Furthermore, age at menarche was self-reported in most included studies but the validity of age at menarche self-reported in middle age is only moderate compared with that recorded in adolescence (40) . The impact of potential recall bias is, however, unlikely to be significant as there is no evidence to suggest that recall might be differential between cases and controls. It should be noted that, although this review provides a quantitative measure of the relationship between early age at menarche and endometriosis risk, the pooled effect size, being a weighted standardized mean difference, may be more clinically meaningful if directly interpreted qualitatively, rather than quantitatively. This review concludes that early age at menarche is associated with a very modest increase in endometriosis risk when studies with better methodological quality adequate control of potential confounders are considered. It highlights the 1) need for well-designed studies incorporating collection of confounder information to explore other risk-factors that may be even more subject to bias, and 2) the need to understand the significance of these factors in the diagnosis of endometriosis and understanding of its etiology. Finally, it has been suggested that a history of earlier age at menarche may be used to guide diagnostic and therapeutic strategies if other symptoms point to endometriosis as a possible diagnosis (14) . The results of this meta-analysis, however, do not present strong evidence for the clinical utility of a history of early menarche in the evaluation of endometriosis.

This review was restricted to published research articles that compared age at menarche in women with surgically confirmed endometriosis and those without endometriosis. These studies were identified in two main ways: 1) Medline and Embase databases were searched through the National Library for Health from 1980 to 2011 for all published case-control studies examining the relationship of early age at menarche and the risk of endometriosis. The search was conducted by two of the authors and was limited to human studies published in the English language, and 2) the reference lists of identified publications were also searched in an iterative manner for relevant studies and authors of primary articles contacted where clarity was needed about data in primary studies. For the database searches, the search terms “case-control studies,” “epidemiologic determinants,” “menarche,” “risk factors,” and “endometriosis” were used as a combination of free text and thesaurus terms (see Supplemental Table 1 , available online, for search syntax). Included studies had to 1) be case-control studies involving women with surgically confirmed endometriosis, as the condition can only be diagnosed reliably at surgery, 2) have examined the relationship between endometriosis risk and early age at menarche as a primary or secondary outcome of interest, and 3) have clearly described criteria for the selection of controls. Important details on design, methods, and results of primary studies were extracted from appropriate articles and summarized. Early menarche is often defined as menarche before the age of 12 years (≤11 years old), but some investigators base definition on menarche at ≤12 years. In this review, studies with either definition of early menarche were included. The quality of primary studies was assessed using the Newcastle-Ottawa scale, a validated tool for assessing the quality of observational and nonrandomized studies (7) . The scale uses a star system to evaluate observational studies on three criteria: participant selection, comparability of study groups, and assessment of exposure. Key in the assessment of the comparability of study groups is the extent to which potential confounders are controlled for. All results were expressed in terms of an “effect size” of the difference in mean age at menarche of women with and without endometriosis. Most studies expressed their findings as odds ratios of early menarche in women with endometriosis compared to controls. For these studies, odds ratios were converted directly to effect sizes using the approach described by Chinn (8) . For one study in which results were expressed as the mean ages at menarche in cases and controls, an effect size calculator worksheet was used to derive an effect size from the means and the pooled SDs (9) . For another study that expressed the outcome as a median and range, the mean ± SD was estimated using the approach of Hozo et al. (10) . Effect sizes were used in random effects meta-analysis of DerSimonian and Laird (11) in Stata program (version 11). The impact of heterogeneity between studies was assessed by calculating the I 2 . To determine whether any one study unduly influenced the pooled effect size (small study effects), a sensitivity analysis was conducted by recalculating the pooled effect size after deleting each study, one at a time. To explore the presence of publication bias, a funnel plot was produced and the approach by Egger et al. (12) was used to test the significance of funnel plot asymmetry. The latter involves regression of the standard normal deviate of each effect size on the inverse of its standard error (precision). The regression line should have a positive slope and an intercept of zero in the absence of bias. Further sensitivity and subgroup analyses considered studies by important population and study characteristics. These included: 1) the category of women studied—infertile women versus both fertile and infertile women, 2) the stage of endometriosis studied, 3) the approach to case recruitment—prospective or otherwise, 4) the cutoff age for early age at menarche—≤11 versus ≤12 years, and 5) controlling for important potential confounders, principally body mass index (BMI).