Risk Factors (Excluding Hormone Replacement Therapy) for Endometrial Hyperplasia: A Systematic Review

Endometrial Hyperplasia (EH) is a condition that is characterised by abnormal growth of the endometrium lining the uterus [1-3]. This condition is more prevalent among peri-menopausal and postmenopausal women [4]. While previously EH was classified into simple or complex EH, with or without atypia [2], the 2014 World Health Organisation classification simplifies this into EH without atypia and atypical hyperplasia [5]. Atypical hyperplasia is less common than other types, and results from observational studies suggest that it is the type which is more associated with the risk of progression to endometrial cancer [1-3]. The endometrial cancers which develop from EH are socalled type 1 endometrial cancers of endometrioid type [6].


Introduction
Endometrial Hyperplasia (EH) is a condition that is characterised by abnormal growth of the endometrium lining the uterus [1][2][3]. This condition is more prevalent among peri-menopausal and postmenopausal women [4]. While previously EH was classified into simple or complex EH, with or without atypia [2], the 2014 World Health Organisation classification simplifies this into EH without atypia and atypical hyperplasia [5]. Atypical hyperplasia is less common than other types, and results from observational studies suggest that it is the type which is more associated with the risk of progression to endometrial cancer [1][2][3]. The endometrial cancers which develop from EH are socalled type 1 endometrial cancers of endometrioid type [6]. enzyme aromatase [14]. Given the predominant role of hormones in the development of EH, a Cochrane review on hormone therapy in postmenopausal women which included 45 trials and 38,702 participants found that unopposed oestrogen is associated with an increased risk of EH with relative risks of 3.20 (95% CI 2.02 -5.26) and 10.09 (95% CI 4.90 -20.80) for moderate and high doses of oestrogen respectively, although this increased risk was not observed with low doses of hormone replacement therapy (HRT) use [13].
Similar to known risk factors for endometrial cancer [15], it is possible that demographic and modifiable factors such as age, parity, oral contraceptive use, body fatness, physical activity, smoking and co-morbidities may play an aetiological role in the development of EH [1,12,16]. The aim of this systematic review and meta-analyses is to quantify the association between risk factors (excluding HRT, since this is incorporated in a Cochrane review 13 and development of EH.

Search strategy
Three electronic databases namely MEDLINE (US National Library of Medicine, Bethesda, Maryland), EMBASE (Reed Elsevier PLC, Amsterdam, Netherlands), and Web of Science (Thomson Reuters, USA) were systematically searched from inception up to 30 June 2015 for relevant studies that included one or more keyword(s) or Medical Subject Heading from each of the following groups of terms: (i) endometrial hyperplasia, simple endometrial hyperplasia, complex endometrial hyperplasia, complex hyperplasia with atypia, simple hyperplasia with atypia, complex atypical endometrial hyperplasia, simple atypical endometrial hyperplasia; (ii) risk factor(s), causality, association, predisposing factor(s), predisposing factor(s), parity, obesity, history of diabetes, ethnicity, race, socio-economic status, occupation, education, oral contraceptive use, tamoxifen use, NSAID use, aspirin use, age at first birth, miscarriage history, age at menarche, alcohol, smoking, PCOS, polycystic ovarian syndrome, family history of cancer, personal history of cancer, medications, BMI, body mass index, waist circumference, body weight, diet, body fatness, waist-hip ratio, physical activity, use of fertility treatments.
Review articles and animal studies were excluded and no language restriction was applied.

Data extraction
Titles and abstracts for potentially relevant articles were independently screened by two of three reviewers (OS, LM and HC). Then two reviewers (OS and HC) independently screened full text articles for the remaining studies to identify relevant studies that meet the pre-set inclusion criteria for the systematic review:   (iv) Outcome: Risk of EH.
The full text of the remaining articles were independently screened by two reviewers (OS and HC) to identify relevant studies that meet the pre-set inclusion criteria for the systematic review. Articles reporting on less than 10 cases of EH were excluded from the review. Meetings were held between three reviewers (LJ, HC, OS) to resolve any discrepancies. The full protocol for this review can be found at http://www.crd. york.ac.uk/PROSPERO/display_record.asp?ID=CRD42015016569#. VRfpRJMe5ME. Relevant information about study design, number of cases, controls or cohort size, menopausal status of the study population, age, and method used to diagnose EH, control definition, method used to measure exposure and adjusted confounders were extracted from full text articles. The Newcastle Ottawa Scale coding manual was used to assess quality of each study. Some studies reported EH risk as part of a combined EH and endometrial cancer risk estimate, and were retained for inclusion in the systematic review, and sensitivity analysis conducted removing such studies from overall pooled estimates. Studies that compared risk between different types of EH, and not in comparison with a true control group that did not have EH, were excluded. Attempts were made to retrieve additional information where required from a number of authors via e-mail contact [17][18][19][20][21][22].

Statistical analysis
Statistical analyses were conducted with STATA version 13 (StataCorp, College Station, TX, USA). Unadjusted and maximally adjusted relative risk (RR) estimates and corresponding 95% confidence intervals (CI) were extracted from published articles where possible. Random-effects models were used to derive pooled RRs [23] and CI. It was decided a priori to perform meta-analyses where at least three studies had reported risk estimates for a particular risk factor. When applying these criteria we were able to conduct meta-analyses of EH risk comparing high versus low for body mass index (BMI), smoking, hypertension, and diabetes.
Sensitivity analyses were conducted for EH risk in relation to BMI and diabetes removing individual studies; this was not possible for other risk factors as too few studies reported these. Sub-group analysis was performed where possible for EH with or without atypia in relation to BMI, diabetes and hypertension. We also assessed heterogeneity of studies included in meta-analyses using the I2 statistic [24,25]; I2 values of 25%, 50% and 75% are typically interpreted as low, moderate and high heterogeneity respectively. We investigated the likelihood of publication bias using the Egger's test [26,27]. Combined RR were calculated before entry into final meta-analyses for studies that reported separate EH risk estimates only by different types of EH or different age categories. Specifically, one study reported separate EH risk estimates for complex EH and atypical EH [28].

Results
After application of our search strategy in the three databases, and removal of duplicates, a total of 2,890 titles and abstracts were reviewed in the first instance to determine potentially relevant studies for inclusion. After title and abstract review, 79 full text articles and abstracts were reviewed, and 15 full text articles were retained in the review ( Figure 1). Included articles assessed lifestyle factors, menstrual history, age, medical history, reproductive history and socio-economic factors and their relationship with risk of developing EH.

Other body fatness measures
Two studies have investigated other body fatness measures and EH risk, as summarised in Tables 1 and One study among premenopausal women with abnormal menstrual bleeding reported a significant 7-fold increased risk for complex atypical EH and endometrial carcinoma combined (OR 7.3, 95%CI 3.2-16.8), comparing body weight >90kg versus <90kg [35].
In a further study, the Quetelet index was reported to be significantly higher in postmenopausal EH cases, compared with controls, leading to an increased risk (OR 3.8, 95%CI 1.27-11.40) when comparing >2.9 versus ≤2.9. In contrast, a protective effect for premenopausal women was noted (OR 0.25, 95% CI 0.07-0.95) [36].

Smoking
Two population-based [28,37,38] and one hospital-based 12 casecontrol studies examined the relationship between smoking and risk of EH. Descriptions of study characteristics are shown in Table 4.
Results from these three studies were pooled to derive an unadjusted EH risk estimate of 0.98 (95%CI 0.64-1.49) with I2 45.7% for smokers compared with non-smokers. Adjusted pooled risk estimate was 0.88 (95%CI 0.66-1.17) with 0% heterogeneity for smokers when compared with non-smokers ( Figure 3).
The moderate heterogeneity among studies disappeared after adjusting for confounders. However, only one study adjusted for HRT, and two adjusted for BMI as shown in Table 4.

Physical activity
One Italian hospital-based case-control study 30 reported nonsignificant increased risk of EH (OR 1.38 95%CI 0.50-3.77) among women who reported high levels of physical activity (≥60 minutes 3 times/week) when compared with those who reported lower levels of physical activity (Tables 1 and 4).

Medical history Diabetes
Five studies evaluated the relationship between diabetes and EH risk. Three were hospital-based case-control [12,30,34], one was a population-based case-control,28 and one was a prospective cohort study [39]. Characteristics of the studies are shown in Table 5.
Three studies [12,28,38] were included in meta-analysis in order to derive unadjusted pooled EH risk estimate for diabetic versus   Figure  4. Egger's test showed no significant evidence of publication bias (p=0.34).
While risk estimates remained non-significant for the most part of sensitivity analyses, a significant positive association was observed when the study by Epplein et al. [28] was excluded. Heterogeneity however ranged from low to moderate throughout. Sub-group analysis by EH type showed non-significant positive association between EH without atypia and diabetes (RR 1.32, 95%CI 0.31-5.70) ( Table 2). Two studies reported adjusting for BMI while none adjusted for HRT use as shown in Table 5. 1 hospital-based case-control, 1 cross-sectional study n=1 study reported a non-significant reduced risk of non-atypical EH among post-menopausal women when compared with pre-menopausal women but non-significant increased risk of atypical EH was reported for postmenopausal women compared with pre-menopausal women. n=1 study reported a significant reduced risk of complex amongst post-menopausal women in comparison with pre-and peri-menopausal women.

Hypertension
Four studies [12,28,30,39] reported on hypertension in relation to EH risk. Characteristics of the studies are shown in Table 5.
Sub-group analyses showed non-significant positive associations between atypical EH (RR 1.92 95%CI 0.57-6.53) and EH without atypia (RR 1.17 95%CI 0.39-3.45) ( Table 2) and hypertension status; only the latter showed reduced heterogeneity. Two studies adjusted for BMI while one reported adjusting for HRT as shown in Table 5.

Family History of cancer
Women with a family history of endometrial cancer were reported by Ricci et al. [12] to have around 20 percent reduced risk of developing EH , although not statistically significant (OR 0.8 95% CI 0.2 -2.6). Another study found women with abnormal bleeding who had a family history of colon cancer or endometrial cancer to be more likely to develop endometrial cancer or complex EH with atypia (OR 9.1, 95%CI 2.2 -37.1;OR 5.8, 1.1 -28.6, respectively) [35]. (Tables 1 and 5). None of the studies reported adjusting for BMI or HRT use as shown in Table  5.

Oral contraceptive use
Two studies [12,40] examined the relationship between oral contraceptive use and the risk of developing EH (Table 6). One population-based case-control study reported a reduced risk of complex and atypical EH among women who had used OC 6 months before presenting with abnormal bleeding (OR 0.2, 95%CI 0.0-0.6) after   adjusting for BMI. However, a non-significant increased risk of EH was found in another study among women who had ever used OC versus those who had never used OC (OR 1.6, 95%CI 0.9-2.8) [12] ( Table 1).
The authors further assessed OC use and EH risk by duration of use and consistently found non-significant increased risks when they compared women who had used OC for more than 5 years, 13-60 months, 12 years or less with never users (OR 1.2, 95%CI 0.4-3.4;OR 1.4, 95%CI 0.5-3.6; and OR 2.0, 0.9-4.3, respectively). It should however be noted that BMI or HRT use was not adjusted for in the latter as shown in Table 6.

Parity
When comparing women who had given birth to two or more babies with nulliparous women, Ricci et al. [12] found an almost significant 2-fold (OR 1.8, 95%CI 0.9-3.6) increase in risk of complex EH (Table  1) after adjusting for age and education [12]. In contrast, Epplein et al. [28], found significant reduced risk (OR 0.29, 95% CI 0.07-0.51) of EH among women who had given birth to three or more babies when compared with nulliparous women, after adjusting for BMI [28].
Two studies which compared nulliparous women with multiparous women found a significant increased risk of EH in nulliparous women (OR 3.7, 95%CI 1.2-10.9) 35 and (OR 2.8, 95% CI 1.3-6.1) (after adjusting for prior use of oestrogen) [41], respectively. Meta-analysis was not performed for these four studies, summarised in Table 6 Table   6:

Menopausal status
Two hospital-based case-control studies [12,29] evaluated menopausal status in relation to EH risk (Table 6). One study among Chinese women 29 found that postmenopausal women were less likely to develop EH without atypia (OR 0.65, 95%CI 0.17-2.50) but they were more likely to develop EH with atypia (OR 2.40, 95%CI 0. 43-13.27) when compared with premenopausal women, although these estimates did not achieve statistical significance. Similarly, another study 12 also reported a significant reduced risk of complex EH among postmenopausal women in comparison with pre-and perimenopausal women (OR 0.2, 95%CI 0.1-0.5). The authors also found a non-significant 20% increased risk of complex EH among women who reported menopause at ≥53 years versus <50 years at menopause [12] (Table 1). One of the studies reported adjusting for BMI and HRT use as shown in Table 6. One further study suggested that polycystic

Other factors Age
Two hospital-based case-control studies reported risk estimates for age with regards to EH [12,41]. Characteristics and results from these studies are shown in Tables 1 and 7 . One study reported a significant increased risk of EH or EC amongst women ≥70 years old versus women 49-59 years old after adjusting for confounders including prior use of oestrogen, and another study reported a non-significant decreased risk of EH amongst women ≥65 years old versus <45 years old. Metaanalyses were not conducted because only two studies [12,41] provided risk estimates for age in relation to development of EH, reports from the studies were mixed ( Table 1). One of the studies adjusted for quetelet index, a measure of body fatness as shown in .

Socio-economic status
Two studies [12,36] examined education and income in relation to EH risk (Table 7). In one study, a positive association was observed among women who had ≥12 compared with <7 years of education (OR 2.8 95% CI 1.70-4.80).
Similarly, women who earned ≥$30,000 were found to have higher chances of developing EH when they were compared with women who earned less than $30,000 in a Canadian study ( Table 1). The observed association was significant for premenopausal women (OR 1.85, 95%CI 1.16 -2.96) but not for postmenopausal women (OR 1.15, 95%CI 0.78 -1.69) [36]. Neither of the studies reported adjusting for HRT use or BMI as shown in Table 7.

Main findings
In this novel systematic review of risk factors for EH (excluding hormone replacement therapies), meta-analyses suggested a significant positive association between increased BMI and risk of EH; no significant associations were detected between smoking, hypertension or diabetes and EH risk in pooled analyses of a limited number of studies. However, there was paucity of high quality, consistent evidence for the aforementioned and other factors in the review. There was also inadequate adjustment for relevant confounders, namely HRT and BMI, in some of the included studies.
The importance of pooling risk estimates is demonstrated by the expected finding that higher BMI is positively associated with EH compared with lower BMI, considering that only three out of six studies which reported a positive association between BMI and EH risk showed statistical significance. EH is an oestrogen-driven disease. From a biologically plausible viewpoint, it is well known that oestrogen can be synthesized from adipose tissue, this increases the level of circulating oestrogen which in turn stimulates growth of the endometrium. Reduction in high heterogeneity which occurred when the study of simple EH 30 was excluded during sensitivity analyses suggests that the relationship between BMI and EH may differ according to the presence of atypia. Due to the role of body fatness in the development of EH and endometrial cancer, it is important for women to maintain a healthy weight [15].
Pooled analysis of studies that investigated hypertension showed non-significant positive association between hypertension and EH risk. Some authors have reported that hypertension is positively associated with EH 40 or endometrial cancer [42]. However, this association was found among overweight or obese women compared with lean women [42], this observation should therefore be viewed with caution as it is likely to be confounded, considering the association between obesity and hypertension. Hypertension has previously been linked to insulin-like growth factor 1 (IGF-1), and measures of body fatness such as waist-hip ratio and obesity were reported to be higher among hypertensive patients than controls [43,44]. IGF-1 is known to be related to cell growth and cancer progression [45].
We found a non-significant increased EH risk among diabetic versus non-diabetic women. Although the mechanism for a potential association between diabetes and EH is not very clear, diabetes has been linked to IGF-1 [46]. Low levels of IGF-1were found to be positively associated with diabetes after adjusting for confounders including BMI [46]. In a rat model, Type 1 diabetes was also been shown to induce EH development, potentially mediated by oestrogen receptor alpha and p16 expression [47]. Several authors have reported overweight/obesity as one of the most important modifiable risk factors for diabetes [48,49]. Despite the well-known relationship between obesity and diabetes, few of the studies included in our meta-analysis adequately adjusted for this confounder.
Meta-analysis of three studies showed no association between tobacco smoking and risk of EH. An earlier literature review suggested that smoking has an anti-oestrogenic effect, which can reduce the rate of androgen-oestrogen conversion [50]. Smoking has also been linked to early menopause [51][52][53]. Women who undergo menopause early are less exposed to oestrogen than women who are older at menopause. However, smoking has been consistently linked to the development of many neoplastic conditions and is certainly not advised [54,55]. It is plausible that known carcinogenic effects of smoking may be countered by the aforementioned anti-oestrogenic effect, explaining the observed null association for tobacco smoking and EH risk.
One study reported a non-significant increased EH risk for women with self-reported higher levels of physical activity compared to those who reported lower levels of physical activity. However, as with all selfreported information of desirable lifestyle factors, this result should be interpreted with caution. Physical activity has previously been shown to be protective against endometrial carcinoma, given that physical activity may modulate metabolism, and excretion of endogenous sex hormones such as oestrogen which is also known to be responsible for development of EH [56]. Interestingly the previously described EH diabetic rat model did observe a significant reduction in oestrogenreceptor alpha and p16 expression for those rats undertaking aerobic exercise [48].
Contrasting results were reported for parity and EH risk by individual studies included in this review, although the majority reported protective effects of child-bearing for EH risk. Nulliparity is known to be associated with an increased risk of endometrial cancer [57] -a possible mechanism for this is that during pregnancy, a woman is exposed to larger amounts of progesterone as opposed to oestrogen. Contrasting reports were also observed for the two studies investigating OC use and EH risk. It should however be noted that OC usage has consistently been found to reduce EC risk among users when compared with non-users [58,59]. Biologically, this is related to the low dose of oestrogen in relation to progestin contained in OC, which inhibits endometrial proliferation [60,61]. A significant decreased risk of complex EH was reported among postmenopausal versus pre-and peri-menopausal in an Italian study. Conversely, findings from a further study included in our review suggesting an increased risk of atypical EH among postmenopausal versus premenopausal women, which may indicate that HRT use, a well-known risk factor for EH, has more of a propensity to invoke atypical than non-atypical EH. We however did not assess use of HRT in this review due to an earlier Cochrane review which assessed the effects of different hormone therapy regimens on the postmenopausal endometrium. The reviewers found unopposed oestrogen to be associated with increased risk of all types of EH at all doses, in line with the existing literature. Although the reviewers did not perform subgroup analysis for the different types of EH, they found no difference in the risk of EH in women who took low dose oestrogen combined with progestogen compared with controls who took placebo [13].
It is notable that one study reported an increased risk of complex EH in women with higher versus lower level of education while another reported the same association amongst women with higher versus lower income. Measures of social class have been implicated in the development of neoplasms due to the differing medical attention seeking behaviour of the different groups [62,63].
Finally, two studies in the review suggest a link between family history of endometrial or colon cancer in relation to EH risk. This points to a shared genetic or environmental risk factor in EH development. Families with a history of Lynch syndrome have been found to have between 1.5-3 fold increased risk of developing endometrial cancer [64]. Indeed endometrial cancer is more common than colonic cancer in patients with Lynch syndrome.

Strengths and limitations
This is the first systematic review examining the risk factors (excluding HRT use) for EH, the review has a number of strengths and limitations. A major strength of this review was the evaluation of three databases and the robust methodology and adherence to a previously published protocol. This included the strict exclusion of several studies that included simple EH cases in control groups. Although several studies that were included reported on EH combined with other outcomes such as benign endometrial polyps and/or carcinoma, we were careful to consider those separately in our interpretations, as their risk estimates would be distorted. We were also able to perform subgroup analyses for atypical and non-atypical EH in relation to BMI, diabetes and hypertension. Importantly, our collective assessment of EH risk factors has highlighted the general paucity of data available for this condition, but does suggest that EH could be potentially prevented through maintenance of normal body weight.
Limitations of this systematic review largely relate to insufficient data which would have been helpful in increasing precision of risk estimates for the different risk factors evaluated. Pooled risk estimates could not be derived for important risk factors such as age, parity and menopausal status due to insufficient number of studies (<3) providing risk estimates, in accordance with our protocol. Also, many hospitalbased studies were included, which limits applicability of results to the general population. In addition, very few studies adjusted for HRT use and BMI in their statistical models. HRT use is known to play a significant role in the development of EH, its use at all doses in the treatment of menopausal symptoms has been found to be associated with an increased risk of EH [65]. It is still plausible that additional risk factors may exist for EH and could be identified in future, high-quality studies.
The relationship between a high BMI and EH cannot be overemphasized as is shown in this review. Notably, no studies evaluated nutrition or dietary factors in relation to EH risk, even though several aspects of diet, for example coffee and high glycaemic load intake, have been associated with endometrial cancer risk [15]. We hope that this review stimulates further work in this area in an effort to identify more modifiable, preventative factors for EH.

Conclusions
In conclusion, body fatness was found to be associated with an increased risk of EH, therefore women should be encouraged to maintain a normal body weight. No significant associations were detected for other factors and EH risk. However, relatively few studies have been conducted and further aetiological studies which might help identify other non-modifiable risk factors for EH are warranted.