Introduction

Worldwide, endometrial cancer is the fifth most common cancer in women [1] and it represents one of the most frequent gynecologic malignancies [1]. In 2014, endometrial cancer is projected to develop in about 52,630 women in the United States (US) and an estimated 8590 US women will die from this cancer [2]. Each year, approximately 7406 new cases occur in the UK and 88,068 new cases occur in the European Union [3]. The most important risk factors for endometrial cancer are postmenopausal unopposed estrogen therapy, obesity, and nulliparity [4, 5]. In contrast, physical activity represents an important modifiable preventive factor for endometrial cancer [6]. Plausible biologic mechanisms linking increased physical activity to decreased endometrial cancer risk include decreased levels of sex steroids, insulin resistance, and chronic inflammation [7]. Physical activity may be directly involved in these biologic pathways or indirectly by reducing obesity.

Four publications have summarized the available epidemiologic evidence regarding physical activity in relation to risk of endometrial cancer [8–11]. Concertedly, findings from those studies suggest that physical activity is associated with reduced risk of developing endometrial cancer [8–11]. However, those investigations did not provide summary risk estimates for specific activity domains, such as household activity and walking, which greatly contribute to women’s total daily activity [12–14]. For example, it is not known whether walking for transportation better characterizes the aspect of physical activity that is relevant for protection against endometrial cancer than strenuous exercise performed during recreation. Because the hormonal milieu changes throughout the life course, it also remains to be evaluated whether the apparent protective effect of physical activity on endometrial cancer risk depends on specific time periods in life. Moreover, obesity is an important risk factor for endometrial cancer [15, 16] and is inversely associated with physical activity [17], but whether adiposity modifies the relation of physical activity to endometrial cancer or represents an intermediate factor linking physical activity to endometrial cancer remains unclear.

We conducted a comprehensive systematic review and meta-analysis with a specific focus on evaluating physical activity across various domains and intensities, age levels, and body mass index (BMI) groups.

Materials and methods

Literature search and inclusion criteria

In preparing the present meta-analysis, we followed the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines [18]. We conducted a comprehensive literature search in 28 databases (e.g., EMBASE, Medline, Cochrane database of Systematic Reviews, Database of Abstracts of Reviews of Effects, SciSearch, Social SciSearch, PSYNDEX, PsycINFO) up to October 2014. The following search terms were used: physical activity, motor activity, exercise, walking, sports, athletes, endurance training, physical fitness, sedentary (lifestyle), sedentariness, physical inactivity, sitting, motor inactivity, recreation. The search included the following terms for cancer: (endometrium; endometrial; endometrioid; uterus; uterine) cancer(s), carcinoma (s), adenocarcinoma(s), neoplasm (s), tumor (s), tumour (s), or sarcoma(s). In addition, we screened references lists from retrieved original articles to identify further potentially eligible studies. We imposed a restriction to articles on human studies and those published in English. The inclusion criteria were as follows: (1) the study was a cohort study, case–control study, or case–cohort study; (2) the study investigated the association between physical activity and endometrial cancer incidence; (3) the relative risk (RR), odds ratio (OR), or standardized incidence ratio (SIR) and corresponding 95 % confidence interval (CI) was provided or could be calculated; (4) age as a risk factor was taken into account. If studies were found to overlap, we included the study that provided the most comprehensive data.

Data extraction and study quality assessment

From each article, two authors (D.S. and M.F.L.) independently extracted the following information: name of the first author, year of publication, country where the study was performed, total number of individuals and cases, method of physical activity assessment, RRs, ORs, or SIRs with corresponding 95 % CIs comparing the highest with the lowest level of physical activity, and confounding factors that were adjusted for in the analysis. If a study provided unadjusted and adjusted effect measures, the most completely adjusted effect measure that also adjusted for age was used.

Quality assessment was conducted using the Newcastle-Ottawa-Scale (NOS), a validated scale that awards a maximum of nine points to each cohort study (four for quality of selection, two for comparability, and three for quality of outcome and adequacy of follow-up) or case–control study (four for quality of selection, two for comparability, and three for quality of exposure) [19]. We considered studies with a NOS score of ≥6 as high quality studies and those with a NOS score of <6 as low quality studies.

Statistical analysis

Main analysis

In the main analysis, we pooled the risk estimates comparing the highest versus the lowest categories of physical activity in relation to endometrial cancer, with the exception of one study that compared the highest versus the second lowest physical activity category [20]. Most studies provided data on recent physical activity. Thus, we prioritized risk estimates for recent physical activity in the main analysis. If data from more than one physical activity domain were available, we prioritized risk estimates in the following order: total physical activity, recreational activity, occupational activity. If physical activity was expressed using more than one metric (i.e., MET-hours per week, hours per week, times per week), we used MET-hours per week because it represents the most comprehensive physical activity measure, combining activity intensity, duration, and frequency [21].

Risk estimates were interpreted as relative risk estimates (RRi) and were log-transformed to log(RRi). Standard errors of the log-transformed relative risks log(RRi) were defined as si = di/1.96, where di represented the maximum of [log(upper 95 % CI bound of RRi)-log(RRi)] and [(log(RRi)-log(lower 95 % CI bound of RRi)]. To meta-analyze those log-values, we employed random effects models to allow for study heterogeneity [22].

The Q-statistic was used to test for between-study heterogeneity and the I2-statistic was used to quantify the proportion of the total variation due to heterogeneity [22]. Potential publication bias was assessed by visual inspection of a funnel plot and by using Egger’s regression test [23] and Begg’s rank correlation test [24].

To investigate the robustness of our main analysis, we performed a sensitivity analysis in which we excluded one study at a time from the initial meta-analysis to assess whether a particular study may have influenced the summary risk estimate.

Stratified analyses

In stratified analyses, we investigated high versus low levels of physical activity of different domains in relation to endometrial cancer using a random-effects model. Specifically, we considered recreational activity, occupational activity, household activity, and walking (walking/biking for transportation, walking for recreation, and walking without specification). One study [25] investigated walking for transportation and three studies [26–28] examined walking/biking for transportation. We assigned studies to the category walking/biking for transportation that defined walking as walking for transportation [25], walking/bicycling to school or work [26], number of years walked or biked to work most days [27], or walking/bicycling (mainly for transportation) [28]. One study provided two separate risk estimates for walking for transportation and biking for transportation, of which we included the risk estimate for walking for transportation in our meta-analysis [25]. The category walking for recreation included three studies [29–31] that examined walking/hiking and one study [32] that investigated walking/biking during recreation. Three studies reported risk estimates for walking [33], walking for exercise, pleasure, or transportation [34], and walking/biking to work, shopping, and/or walking the dog [32], which were defined as unspecified walking in the present meta-analysis.

Moreover, we explored different intensities of physical activity (light, moderate to vigorous, and vigorous) in relation to risk of endometrial cancer. We further investigated whether the association between physical activity and endometrial cancer varied across different time periods in life (childhood/adolescence, young adulthood/midlife, older age). To evaluate whether BMI modifies the physical activity and endometrial cancer relation, we summarized physical activity risk estimates for women with a BMI < 25 kg/m2 and those with a BMI ≥ 25 kg/m2. Using random-effects meta-regression, we further summarized studies that adjusted for adiposity and those that did not adjust for adiposity to investigate whether the physical activity and endometrial cancer relation is mediated by regulation of body weight. In addition, we assessed potential heterogeneity of the physical activity and endometrial cancer relation according to geographic location, number of participants, number of cases, study quality, and adjustments for parity, oral contraceptive use, and hormonal replacement therapy.

We calculated the P for difference across strata using meta-regression comparing the model including the stratification variable as explanatory variable with the null model not including any explanatory variables.

Dose–response meta-analysis

Because studies used different measures of physical activity, we conducted an additional analysis that was restricted to studies that used MET-hours per week as physical activity measure. To account for variability in the range of MET-hour levels in the individual studies, we further performed analyses summarizing studies that provided RRs for approximately 3–8, 9–20, and >20 MET-hours as compared with <3 MET-hours of physical activity per week.

Finally, we performed a non-linear dose–response meta-analysis of recreational physical activity expressed in MET-hours per week in relation to endometrial cancer using second degree fractional polynomials [35]. We excluded two studies [28, 36] from the dose–response meta-analysis because in those studies, MET-hours calculations were based on the combination of all activities and inactivities, including time spent sitting and sleeping.

All statistical analyses were conducted using the R-packages ‘metafor’ [37] and ‘mvmeta’ [38]. The analyses were two tailed and a P < 0.05 was considered statistically significant.

Results

Literature search and description of the studies

Supplementary Figure 1 shows details of the literature search strategy and study selection. A total of 2636 publications were retrieved from the electronic literature search databases and one article was identified by manual search. After removal of 889 duplicate articles identified in different databases, 1748 remained for evaluation. After screening titles and abstracts, 1703 articles were excluded that were not original articles related to physical activity and endometrial cancer incidence. A total of 45 articles remained for full review, of which twelve were excluded because they provided information from overlapping studies [10, 39], combined different exposures or different outcomes [40–44], compared different stages of endometrial cancer with each other [45], did not adjust for age [46], or lacked appropriate data to calculate risk estimates [47, 48]. One study was excluded that used individuals with prevalent cancer as controls [49]. The remaining 33 articles met the pre-specified inclusion criteria and were included in our meta-analysis [20, 25–34, 36, 50–70].

Descriptive data from studies included in the present meta-analysis are shown in Table 1. We included 18 cohort studies, one case–cohort study, and 14 case–control studies, yielding a total number of 2,219,151 participants and 19,558 endometrial cancer cases. Physical activity was assessed by self-administered questionnaire in 16 studies, by interview in ten studies, and in four studies physical activity assessment was based on job titles. In three studies, total or recreational physical activity was assessed by interviews or self-administered questionnaires and occupational physical activity was assessed by job titles. The number of adjustment factors in the models ranged from one to 18. Twenty studies showed a quality score equal to or greater than six points and thirteen studies had a quality score of less than six points (Table 1).

Table 1 Characteristics of the 33 included studies of physical activity and endometrial cancer
Full size table

Main analysis

As shown in Fig. 1, the highest compared with the lowest physical activity category revealed a summary RR of endometrial cancer of 0.80 (95 % CI 0.75–0.85). No evidence for heterogeneity between studies was observed (I2 = 6.5 %, \(P_{heterogeneity} = 0.38\)). Further, no publication bias was indicated by the funnel plot, Egger’s regression test (P = 0.06), or Begg’s rank correlation test (P = 0.25). In sensitivity analyses, omission of one study at a time did not materially alter the results.

Fig. 1
figure 1

Forest plot corresponding to the random effects meta-analysis summarizing the relation of physical activity and risk of endometrial cancer across case–control studies and cohort studies; *case–cohort study; RR relative risk; CI confidence interval

Full size image

To investigate whether variability in the underlying physical activity measure may have influenced our results, we meta-analyzed nine studies that provided physical activity measures expressed as MET-hours per week. We obtained a RR of 0.80 (95 % CI 0.70–0.92) comparing the highest versus lowest level of overall physical activity (recreational physical activity only: RR 0.85; 95 % CI 0.74–0.97; total physical activity only: RR 0.68; 95 % CI 0.50–0.93). We further summarized studies that used relatively uniform categories of MET-hours per week of recreational physical activity and obtained RRs of 0.94 (95 % CI 0.74–1.20), 0.79 (95 % CI 0.64–0.98), and 0.87 (95 % CI 0.71–1.06) for approximately 3–8, 9–20, and greater than 20 MET-hours as compared with less than 3 MET-hours of physical activity per week (Supplementary Figure 2). Within the range of 0 to approximately 40 MET-hours per week of recreational physical activity, we observed a non-linear inverse dose–response relation for recreational physical activity with endometrial cancer risk (\( P_{non - linearity} < 0.05 \)), which indicated a 5 % reduced risk in endometrial cancer for those engaging in 12 MET-hours per week of recreational physical activity compared to those not engaging in regular recreational physical activity (RR 0.95; 95 % CI 0.91–0.99).

Stratified analyses

Investigating the association between physical activity and endometrial cancer risk by study design showed a RR of 0.84 (95 % CI 0.78–0.91) for cohort studies and a RR of 0.72 (95 % CI 0.64–0.80) for case–control studies (\(P_{difference} = 0.03\), Fig. 1). Our analyses of physical activity domain included 22 studies of recreational activity, 19 studies of occupational activity, and seven studies of household activity. Comparisons of high versus low levels of recreational activity, occupational activity, and household activity resulted in summary RRs of 0.84 (95 % CI 0.78–0.91), 0.81 (95 % CI 0.75–0.87), and 0.70 (95 % CI 0.47–1.02), respectively (Fig. 2). Ten studies explored the relation between walking or walking in combination with hiking or biking and endometrial cancer risk. One study [32] provided two risk estimates: one for walking/biking for transportation and one for walking/biking during recreation. When we considered the risk estimate for walking/biking for transportation from that study [32], the summary RR for all ten studies was 0.82 (95 % CI 0.69–0.97; Fig. 2). Inclusion of the risk estimate for walking/biking during recreation from that study [32] did not materially change the summary RR (RR 0.85; 95 % CI 0.73–0.99). The overall P value for difference according to activity domain was 0.88 (Fig. 2).

Fig. 2
figure 2

Forest plot corresponding to the random effects meta-analysis summarizing the relation of physical activity and risk of endometrial cancer across different activity domains; *Summary effect estimate based on studies related to walking, including walking/biking for transportation, walking for recreation, and unspecified walking; RR relative risk; CI confidence interval

Full size image

Restricting the analysis of walking to studies that explored walking/biking for transportation, a summary RR of 0.70 (95 % CI 0.58–0.85) was obtained. Walking for recreation (RR 0.94; 95 % CI 0.76–1.17) and walking without specification (RR 0.88; 95 % CI 0.52–1.50) were unrelated to risk of endometrial cancer. The P value for difference according to the different categories of walking was 0.13.

Two studies provided information on light intensity activity, eight studies reported on moderate to vigorous intensity activity, and eight studies considered vigorous intensity activity. As shown in Fig. 3, the pooled risk estimates for the associations between high versus low levels of light intensity, moderate to vigorous intensity, and vigorous intensity activity were 0.65 (95 % CI 0.49–0.86), 0.83 (95 % CI 0.71–0.96), and 0.80 (95 % CI 0.72–0.90), respectively (\(P_{difference} = 0.35\)).

Fig. 3
figure 3

Forest plot corresponding to the random effects meta-analysis summarizing the relation of physical activity and risk of endometrial cancer across different activity intensities; RR relative risk; CI confidence interval

Full size image

We next examined whether the association between physical activity and endometrial cancer differed across time periods in life. Random effect models yielded summary RRs of 0.94 (95 % CI 0.82–1.08), 0.77 (95 % CI 0.58–1.01), and 0.69 (95 % CI 0.37–1.28) for high versus low physical activity performed during childhood/adolescence, young adulthood/midlife, and older age, respectively (\(P_{difference} = 0.51\), Table 2).

Table 2 Summary risk estimates from random effects models of physical activity in relation to endometrial cancer, stratified by selected potential effect modifying factors
Full size table

We noted a statistically significant inverse relation between physical activity and endometrial cancer in postmenopausal women (RR 0.81; 95 % CI 0.67–0.97), whereas the relation was statistically non-significant in premenopausal women (RR 0.74; 95 % CI 0.49–1.13), although the difference according to menopausal status was not statistically significant (\(P_{difference} = 0. 7 8\)).

Seven studies provided risk estimates of physical activity in relation to endometrial cancer risk according to standard BMI categories. High versus low physical activity was unrelated to endometrial cancer in women with a BMI < 25 kg/m2 (RR 0.97; 95 % CI 0.84–1.13), whereas it was inversely associated with endometrial cancer in women with a BMI ≥ 25 kg/m2 (RR 0.69; 95 % CI 0.52–0.91; \(P_{difference} = 0.0 7\)) (Fig. 4). To explore whether adiposity mediates the association between physical activity and endometrial cancer risk, we compared studies that adjusted for adiposity with studies that did not adjust for adiposity. We observed no difference in summary risk estimates according to adjustment for adiposity (\(P_{difference} = 0.39\)).

Fig. 4
figure 4

Summary risk estimates from random effects models of physical activity in relation to endometrial cancer, stratified by BMI; BMI body mass index, RR relative risk; CI confidence interval

Full size image

Other potential effect modifying factors had no impact on the association between physical activity and endometrial cancer, including geographic location, number of study participants, number of cases, study quality, and adjustments for parity, oral contraceptive use, or hormone replacement therapy (all \(P_{difference} > 0.05\), Table 2).

Discussion

The present meta-analysis revealed a 20 % reduction in risk of endometrial cancer with high versus low levels of physical activity. The apparent beneficial effect of physical activity on endometrial cancer was statistically significant for recreational activity, occupational activity, and walking/biking for transportation. It was also statistically significant for different physical activity intensities, including activities of light, moderate to vigorous, and vigorous intensities. Taken together, these findings suggest that a broad range of physical activity participation protects against endometrial cancer.

A notable finding of our meta-analysis is that even physical activity of light to moderate intensity, such as walking/biking for transportation, showed an apparent beneficial effect on endometrial cancer risk. By comparison, walking during recreation and walking without further specification was unrelated to endometrial cancer. Walking represents an important physical activity domain as it comprises light physical activity, which contributes to a substantial amount of women’s total daily activity [12–14]. It can be easily implemented in the daily routines and requires no specific facilities. One reason for the difference in associations between walking for transportation, walking for recreation, and unspecified walking is potential imprecision in measuring recreational or unspecified walking, whereas participants may be able to more accurately recall walking for transportation specifically. Moreover, three out of four studies on walking for transportation combined walking and biking, the latter of which may consist of more intense or more regular physical activity, leading to a more pronounced inverse association. In contrast, three out of four studies on recreational walking combined walking and hiking, the latter of which may comprise irregular activity and include interruptions in physical activity and breaks. Because overall walking may consist of walking for transportation, recreational walking, and short bouts of walking in the household or at the workplace, future studies should be designed to gather more precise information about the specific walking domain to uncover potential beneficial effects of walking on endometrial cancer risk.

We observed inverse relations between of physical activity and endometrial cancer risk in both pre- and postmenopausal women, although the association did not reach statistical significance in premenopausal women. Physical activity may reduce endometrial cancer risk by decreasing estrogens after menopause directly or indirectly through reducing peripheral adipose tissue [71, 72]. In postmenopausal women, excess estrogen is the main determinant of endometrial cancer because ovarian production of both estrogen and progesterone ceases and estrogens are mainly produced by conversion from androgens in the adipose tissue [73, 74].

In premenopausal women, obesity is associated with conditions that are related to increased endometrial cancer risk, including chronic anovulation, reduced luteal phase progesterone levels, irregular menstrual periods, early menarche, and delayed menopause [8].

We noted a statistically significant inverse association between physical activity and endometrial cancer in overweight/obese women, whereas the relation was null in normal weight women. One likely explanation for a more beneficial effect of physical activity in overweight/obese than normal weight women is that physical activity counterbalances unfavorable effects of obesity on endometrial cancer risk, such as elevated estradiol levels [75] and lower SHBG concentrations [76, 77]. Likewise, physical activity has been shown to improve insulin sensitivity [78, 79], alter the insulin-like growth factor (IGF) axis [80] reduce pro-inflammatory mediators [81, 82], and increase anti-inflammatory mediators [7], conditions that show greater imbalance in overweight and obese individuals than in normal weight persons [83–86] and impact endometrial tumor development [7]. Another possibility is greater residual confounding by BMI in the overweight/obese group than the normal weight group.

It appears worth mentioning that the most studies included in the meta-analysis herein used BMI as a measure of adiposity, which is an imperfect measure of adiposity because it also accounts for lean body mass. Future studies should consider using measures that differentiate between fat mass and lean mass, such as dual energy x-ray absorptiometry (DEXA) or magnetic resonance imaging (MRI) to better clarify whether obesity modifies the physical activity and endometrial cancer relation.

A major strength of our study is that it represents the most comprehensive meta-analysis of physical activity in relation to endometrial cancer risk to date. Our large sample size provided substantial statistical power and permitted extensive sub-analyses, including stratification and meta-regression to explore potential effect modifying factors. Particular attention was paid to examining whether physical activity protects against endometrial cancer across multiple activity domains, intensities, and time periods in life. Moreover, we employed a study quality score that evaluated potential selection bias, misclassification, and confounding. Reassuringly, we observed no variation in risk estimates between high and low quality studies. We also conducted a broad set of sensitivity analyses to confirm the robustness of our results.

One possible shortcoming of our meta-analysis is that we were unable to differentiate between endometrial tumor subtypes because such data were unavailable in the underlying studies. However, one study observed no difference in the relations of physical activity when restricting the endometrial cancer case definition to participants with type I endometrial carcinomas [57]. A further potential limitation is that a determination of the precise nature of the association between physical activity and endometrial cancer may have been hampered by the heterogeneous measures of physical activity and associated misclassification of the exposure across studies. However, we conducted an additional analysis by summarizing data comparing high versus low categories of MET-hours of physical activity; results were similar to our main findings. Our analysis relating increasing MET-hours of physical activity to endometrial cancer risk yielded no clear pattern but the summary risk estimates from that analysis were based on only four studies.

Four previous studies summarized the relations of physical activity to endometrial cancer and observed 18–30 % reductions in endometrial cancer risk with high versus low physical activity levels [8–11]. Our meta-analysis differs from those previous studies [8–11] in a number of important aspects. First, our analysis included 10,022 additional cases compared with the most recent meta-analysis [9]. Second, our study is the first meta-analysis to examine walking in relation to endometrial cancer risk. Third, we provided summary risk estimates for physical activity intensity in relation to endometrial cancer. Fourth, we meta-analyzed the relation of physical activity to endometrial cancer risk across different time periods in life. Finally, we quantified the relation between physical activity and endometrial cancer according to level of BMI.

In summary, our meta-analysis shows a 20 % decrease in risk of endometrial cancer with high versus low levels of physical activity. Reductions in endometrial cancer risk were observed for a broad range of activity domains, including recreational physical activity, occupational physical activity, and walking/biking for transportation, and for different intensities, including light, moderate to vigorous, and vigorous activities. Particular protection from endometrial cancer through physical activity participation was found for women who are overweight or obese, who carry an elevated risk for developing endometrial cancer. Future epidemiologic studies should include physical activity assessments designed to discern which specific durations and frequencies of physical activity are most relevant for protection against endometrial cancer.