Literature search

Using the search terms described earlier, we screened 489 articles from PubMed, 954 from EMBASE and 354 from Web of Science. A total of 1219 articles were reviewed through titles and abstracts after excluding 578 duplicates. Subsequently, 1175 articles were removed because they did not meet the inclusion criteria, and 44 articles were selected for full-text review. Of the 44 articles, 22 were eliminated based on our eligibility criteria: 13 articles did not report ORs with corresponding CIs, and we could not calculate the ORs using existing data; 5 articles did not report the outcome of interest; 4 articles used similar population with the most recent papers included. Finally, 22 articles met our inclusion criteria with 24 studies involving 325 899 participants and 19 593 cases were included in this meta-analysis.^12–33 Figure 1 shows the flow chart with our detailed methods for studies’ selection.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow diagram of literature search.

Baseline characteristics

For all the studies included, the years of publication ranged from 1998 through 2019. As for study design, 14 were case–control studies,^{12 15 17–19 22 24 26 29–33} 4 cohort studies^{13 20 23 27} and 6 cross-sectional studies.^{14 16 21 25 28} Looking at study region, 10 studies were performed in North America,^{12 13 16 20 23–25 30} 1 in Africa,¹⁴ 9 in Asia^{15 17–19 22 26 27 32 33} and 4 in Europe.^{21 28 29 31} With respect to adjustment for confounders, 18 studies were adjusted^{12 13 15 18–25 27 28 30–32} and 6 unadjusted.^{14 16 17 26 29 33} When evaluating study quality, 17 studies were deemed to be high quality,^{12 13 18–28 30 31} and 7 low quality.^{14–16 18 29 32 33}

For reports of menstrual status, 19 studies included premenopausal women,^{12–24 26 27 29 31 33} 5 studies contained a small number of postmenopausal participants^{25 28 30 32}; as for the assessment approaches of UFs, 17 studies used ultrasound and/or surgery,^{12–15 17–20 22 23 25 26 29 31–33} 5 were based on self-report^{16 24 28 30} and 2 reported clinical diagnosis.^{21 27} Regarding obesity, 23 studies focused on BMI,^{12–17 19–33} 4 on weight change since age 18,^{13 20 23 27} 5 on waist-to-hip ratio^{13 20 22 26 32} and 4 on waist circumference.^{18 20 22 32} The basic information for the included articles and quality assessment scores are presented in online supplemental file 3.

Association of obesity with the risk of UFs

We found that there was a positive association between obesity and the risk of UFs (OR, 1.19; 95% CI, 1.09 to 1.29; p=0.00; figure 2), through combining ORs of risk of UFs for the highest obesity degree defined in original articles. For the highest versus reference category of BMI, the pooled OR of risk of UFs was 1.19 (95% CI 1.09 to 1.31). For the waist-to-hip ratio, the pooled OR was 1.95 (95% CI 1.23 to 3.08) when comparing the highest waist-to-hip ratio to reference waist-to-hip ratio. For weight change since age 18, the pooled OR risk of UFs was 1.26 (95% CI 1.12 to 1.42) when comparing the highest weight change to reference weight change over time. We found a similar trend for waist circumference with a pooled OR risk of UFs of 1.47 (95% CI 0.95 to 2.28) when comparing the high waist circumference to the reference category. These results are shown in table 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Forest plot of included studies. OR (95% CI). The size of the grey box is positively proportional to the weight assigned to each study, which is inversely proportional to the SE of the OR. The horizontal lines represent the 95% CI.

Subgroup meta-analysis

To explore possible sources of between-study heterogeneity, subgroup analyses were conducted using potential confounders including study design, study population, adjustment for covariates, study quality and assessment methods of UFs. As shown in table 1, the pooled ORs for case–control, cohort and cross-sectional studies were 1.12 (0.99 to 1.27), 1.17 (1.10 to 1.25) and 1.53 (1.12 to 2.09), respectively. The pooled ORs for studies by region were 1.24 (1.10 to 1.40) in North America, 1.30 (1.09 to 1.55) in Asia, 0.95 (0.72 to 1.26) in Europe and 3.54 (1.81 to 6.89) in Africa. The pooled ORs for ultrasound and/or surgery 1.18 (1.07 to 1.31), self-reported 1.14 (0.90 to 1.45) and clinical diagnosis methods 1.29 (1.09 to 1.53) were within a similar range. However, we found a decrease in the estimates between the unadjusted 1.73 (1.01 to 2.96) and adjusted 1.16 (1.06 to 1.27) covariates of the pooled ORs.

Dose–response analysis for BMI, weight change since age 18, waist-to-hip ratio and waist circumference

A two-stage random-effect dose–response meta-analysis for BMI, weight change since age 18, waist-to-hip ratio and waist circumference was performed (results are shown in figure 3). In the first step, a restricted cubic spline model with four knots at the 3rd, 35th, 65th and 95th percentiles was constructed; in the second step, study-specific ORs and corresponding CIs were combined through a multivariate random-effect model using restricted maximum likelihood.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Dose–response analysis between obesity and the risk of UFs. (A) Dose–response analysis between BMI and riskof UFs, (B) Dose–response analysis between weight change since age 18 and the risk of UFs, (C) Dose–response analysis between waist-to-hip ratio and the risk of UFs, (D) Dose–response analysis between waist circumference and the risk of UFs. BMI, body mass index; UFs, uterine fibroids. The solid line and the long dash line represent the estimated ORs and corresponding 95% CIs.

The results of our dose–response analyses indicated that there was a non-linear association between obesity and the prevalence of UFs. For BMI, we found an inverse J-shaped pattern based on 14 studies.^{12–14 16 20–23 25–27 32 33} For the weight change since age 18, we found a parabola pattern based on four studies.^{13 20 23 27} When we evaluated the waist-to-hip ratio, we found a relatively horizontal line pattern based on five studies.^{13 20 22 26 32} Finally, for waist circumference, we found a flat right-skewed bell curve based on three studies.^{20 22 32}

Sensitivity analysis and evaluation of publication bias

Through removing one study at a time, the pooled ORs (95% CIs) ranged from 1.161 (1.068 to 1.262) to 1.243 (1.106 to 1.397), suggesting that the combined effect estimates were relatively robust (see online supplemental file 1). Additionally, obvious asymmetry was not found in funnel plots (see online supplemental file 2), indicating that no obvious publication bias was found in these studies.

Positive association between obesity and the risk of UFs

There are several observational studies assessing the association between obesity and the risk/prevalence of UFs. Interestingly, the findings of these studies are highly variable and inconsistent. Perhaps one of the reasons for wide discrepancies between studies can be the relatively small sample size and insufficient analytical power to determine the association in some of the studies. Our meta-analysis to evaluate the association of obesity with the risk of UFs presents an integrated evaluation of current studies’ findings. In this meta-analysis, 22 articles with 24 studies involving 325 899 participants and 19 593 cases were evaluated. Our pooled result showed a positive association between the measures of obesity and the prevalence/risk of UFs (OR, 1.19; 95% CI, 1.09 to 1.29; p=0.00). This finding suggests there is room for an intervention aiming at reducing the incidence of UFs in premenopausal women, especially when the pathogenesis of UFs is still unclear.

For BMI and weight change since age 18, this meta-analysis suggested that participants classified with obesity had 17% and 26% higher risk of UFs than participants without obesity. Additionally, we found a non-linear dose–response for BMI, weight change since age 18 and waist circumference in association with risk of UFs. Yet, our results might underestimate the overall correlation between obesity and the risk of UFs, since the pooled ORs reported in table 1 were calculated using the high level of obesity. In addition, the combined OR of waist-to-hip ratio with the risk/prevalence of UFs was relatively higher, but the inclusion of Sun’s study³² skewed our result away from the null. Furthermore, dose–response analysis also showed that the correlation of waist-to-hip ratio with UFs was not strong.

Latent mechanisms to explain the association between obesity and the risk of UFs

Although the underlying mechanisms implicated in obesity associated with an increased risk of UFs remain unclear, several theoretical pathways for this association have been proposed. One of these pathways suggests an increased conversion of adrenal androgens to oestrone by adipose tissue promotes the incidence of UFs.^39–41 As an endocrine organ, adipose tissue is responsible for the peripheral conversion of circulating androgens to oestrone. Characterised by an increased proliferation of smooth muscle cells and an overproduction of extracellular matrix (ECM), UFs are known to be oestrogen- and progesterone-dependent and have an increased sensitivity to sex steroid hormones. Thus, increased body fat may result in an overproduction of oestrogens, with enhanced cell proliferation of UFs.^{40 41}

Another mechanism could be increased oestrogen and progesterone or decreased production of SHBG is able to advance the occurrence of UFs. Increased cholesterol biosynthesis in overweight women could promote oestrogen and progesterone production and then induce UFs.⁴² Simvastatin, a 3-hydroxy-3-methyl-glutaryl-CoA reductase inhibitor, is reported to have a differential effect on leiomyoma and myometrial cells at concentrations regularly achieved clinically, including breakdown of the leiomyoma ECM and induction of the apoptotic pathway.^{40 42} Increased unbound circulating oestrogen levels are common in overweight women due to reduced hepatic production of SHBG. Consequently, circulating oestrogens may stimulate cell proliferation of UFs.⁴³

A third potential pathway suggests excessive fat accumulation contributes to secretion of adipokines and inflammatory cytokines. Adipose tissue is now recognised not only a reservoir for energy but an immune organ. In the context of obesity caused by increased food intakes and/or decreased physical activities, the development of insulin resistance (IR) is considered to be initiated by inflammation of adipose tissue.^{44 45} These adipokines and inflammatory cytokines then activate key pathways related to inflammation, proliferation, autophage and mitosis, and subsequently induce the onset of UFs.^{46 47}

UFs are also growth factors dependent, with the abundance of fibrotic connective tissue and ECM components as one important feature. Transforming growth factor (TGF)-family as one of the most important regulators of the fibrosis processes may contribute to UFs through Smad pathway, PI3K/Akt/mTOR pathway, the Ras/Raf/MEK/ERK signalling cascade and focal adhesion kinase pathway.^{48 49} Tumour necrosis factor alpha secreted by adipocytes enhances the proliferation of UFs by functioning through NF-κB, JNKs and p38-MAPKs.^{46 47 50 51}

The metabolic syndrome which usually co-exists with obesity can result in the presence of UFs and be another pathway that needs to be considered. Hyperglycaemia, along with IR, has been observed to increase unbound circulating sex steroid hormones and to promote UFs’ cell growth by altering the tyrosine kinase signal pathway.¹⁷ However, the association between UFs and IR still needs further elucidation. Hypertension might promote the onset of UFs through vasoactive peptides (eg, TGF-β) stimulating uterine smooth muscle cells’ proliferation and vascular remodelling.¹⁸ Dyslipidemia, induced by abnormal lipid metabolism, is partly regulated by oestrogens, thus related to oestrogen-dependent UFs.²²

Unhealthy lifestyles related to obesity, including reduced physical activity, diet poor in fruits and vegetables, may also increase the risk of UFs.¹⁵ One possible reason can be irregular physical activity is responsible for decreased SHBG levels and increased insulin and sex hormone levels. These proposed relationships need to be further explored in the future.^{17 18 22}

Potential explanations for the non-linear dose–response

Potential explanations could be as follows for our finding of a non-linear pattern of association between obesity and the risk/prevalence of UFs. Thin or very obese women usually have decreased menstrual cycling compared to normal-weight women,⁵² which reduces the risk of UFs through lowering levels of circulating oestrogens and progesterone. In addition, visceral fat, compared to subcutaneous fat, is easier to promote the production of proinflammatory cytokines and is associated with higher IR and hyperinsulinaemia. All of these factors are able to induce the onset of UFs.²⁹

Examining possible confounding factors through subgroup analysis

In a meta-analysis, between-study heterogeneity is inescapable. We examined some of the main possible causes that could result in high between-study heterogeneity. Generally, study design, study quality, study population, the region in which the study was conducted, adjustment for covariates, the diagnostic methods of disease and the estimation methods of exposure may be sources of between-study heterogeneity. We performed subgroup analyses to minimise these sources of variation, but our between-study heterogeneity did not decrease significantly among most subgroups. Fortunately, our sensitivity analysis indicated that the pooled ORs were not substantially influenced by none of the studies except for combined ORs for waist-to-hip ratio, suggesting that the overall outcome was relatively robust.

Strengths and limitations

This meta-analysis has several strengths. First, 24 studies were included, which provided a large sample size to evaluate our hypotheses. Second, the possible confounder effects were controlled in most of the original studies included in our study when examining the effect of obesity on the occurrence of UFs. Third, the summary ORs were stable when sensitivity analysis was conducted by removing one study at a time (leave-one-out approach). Fourth, to assess and ascertain the association of obesity with the risk of UFs, different measures were used, which include the highest versus reference category of BMI, weight change since age 18, waist-to-hip ratio, waist circumference and dose–response analyses.

There were several limitations in this meta-analysis. First, in spite of a series of subgroup analyses, I² values of between-study heterogeneity did not reduce substantially, which indicated that there must be some unaccounted potential background confounding factors. To minimise the influence of between-study heterogeneity, we constructed a random-effect model. Second, covariates were inconsistent though most studies included performed adjustment for confounders. Third, some studies included contained a small number of perimenopausal and even postmenopausal women, who were unable to be removed from all participants according to the demographic characteristics of original studies. Fourth, race was indicated to be strongly associated with UFs by previous epidemiological studies.^{53 54} However, we could not carry out subgroup analysis stratified by race and/or ethnicity since most original articles did not provide race characteristic of participants except for whites and African-Americans mentioned in some studies. Additionally, information bias might be a concern, since we collected data on weight change since age 18 and self-reported UFs. Given all the aforementioned limitations, our results may be prone to underestimation of our pooled ORs or perhaps shifting our findings towards the null.