Body Mass Index and Cancer Risk: An Umbrella Review of Meta-Analyses of Observational Studies

Abstract Increasing evidence indicates that obesity is a risk factor for various tumors. We aimed to clarify the evidence for an association between body mass index (BMI) and cancer risk based on existing systematic reviews and meta-analyses. Eighteen studies were included in this umbrella review after searching PubMed, Embase and Web of science. The results revealed that underweight was inversely associated with the incidence of brain tumors and positively related to the risk of esophageal and lung cancer. Overweight enhances the incidence of brain tumors, kidney cancer, endometrial cancer, ovarian cancer, multiple myeloma, bladder cancer and liver cancer. Obesity was related to the increased incidence of brain tumors, cervical cancer, kidney cancer, endometrial cancer, esophageal cancer, gastric cancer, ovarian cancer, multiple myeloma, gallbladder cancer, bladder cancer, colorectal cancer, liver cancer, thyroid cancer and Hodgkin’s lymphoma. Moreover, dose-response analysis was conducted by 10 studies, and the results demonstrated that each 5 Kg/m2 increase in BMI was associated with a 1.01- to 1.13-fold increased risk of general brain tumors, multiple myeloma, bladder cancer, pancreatic cancer, breast cancer, and non-Hodgkin’s lymphoma. Every 1 Kg/m2 increase in BMI was linked to 6% and 4% increases in the risk of kidney cancer and gallbladder cancer, respectively.


Introduction
Excess body fatness, a growing public health problem worldwide, is most commonly measured by body mass index (BMI), which is used to classify overweight (BMI ≥ 25 kg/m 2 ) and obesity (BMI ≥ 30 kg/m 2 ) in adults. According to the data given by the World Health Organization (WHO), more than 1.9 billion people who were 18 years and older were overweight in 2016, which accounted for 39% of adults. Moreover, over 650 million adults of these people were obese, which accounted for 13% of adults. Overweight and obesity have been regarded to be related to many chronic diseases including hypertension, hypercholesterolemia, insulin resistance, type 2 diabetes, cardiovascular disease, osteoarthritis, kidney failure and liver disease (1)(2)(3). Furthermore, overweight and obesity were also reported to be risk factors for many cancers (4,5).
Cancer is a group of diseases which caused by the tumor-like transformation of normal cells under selective stress. During this process, any change has the potential to promote the conversion of normal cells to tumor cells (6). Cell growth and proliferation must be coordinated with the presence of sufficient nutrients to support macromolecular synthesis. Obesity, a state of overnutrition, disrupts that balance, resulting in cellular growth factor signaling pathways being activated for a long time and increasing the risk of tumor conversion (7). Renehan et al. performed a meta-analysis of BMI and incidence of cancer, which included 221 datasets from 141 articles and involved 282137 incident cases. The results revealed that an increase in BMI was strongly related to the incidence of many cancers in males and females (8). The WCRF (World Cancer Research Fund) has also presented several common cancers related to obesity including esophageal adenocarcinoma and gallbladder, liver, pancreatic, renal, colorectal, advanced prostate, ovarian, endometrial, and postmenopausal breast cancers. The factors causing tumors are diverse. Among these factors, lifestyle and environment were associated with 90% to 95% of cancer cases (9), of which 14% to 20% were caused by obesity (10). A previous study reported that 3.6% of all new tumors in the world are due to obesity, and during these new cancer cases, colon cancer, postmenopausal breast cancer and uterine cancer might account for more than 60% (11).
Currently, many systematic reviews and meta-analyses have reported associations between BMI and various cancers. However, to our knowledge, there has been no article extracting data from published systematic reviews and meta-analyses related to BMI and the incidence of various cancers to be reported until now. Therefore, we performed an umbrella review to assess the quality of evidence and the extent of possible bias as well as systematically evaluate the associations between BMI and the incidence of multiple cancers, which will help us better understand the influence of BMI on the risk of different cancers.

Search Strategy
The umbrella review was designed and conducted in strict accordance with the guidelines developed by the working group of Joanna Briggs Institute (12). Three databases, including PubMed, Embase, and Web of Science, were systematically searched. The search algorithm used the following terms: (BMI OR Body Mass Index OR underweight OR overweight OR obese OR obesity) And (cancer OR tumor OR carcinoma OR neoplasm OR malignancy) And (systematic review OR meta-analysis). We also conducted a manual screen of reference lists cited in all included articles.

Selection Criteria
BMI is calculated by dividing weight by the square of height. Based on the WHO classification, BMI is divided into underweight (<18.5 kg/m 2 ), normal weight (18.5-24.9 kg/m 2 ), overweight (25-29.9 kg/m 2 ), and obese (≥30 kg/m 2 ). A systematic review and meta-analysis of BMI and various cancer risks were included regardless of the sex, race and region of the participants. If more than one study reported the association between BMI and one kind of cancer risk, we will include the most recent one that has more participants. If a subgroup analysis was conducted by a meta-analysis based on the study design (case-control and cohort studies), we included the results of cohort studies when the number of included studies was more than three; otherwise, we included the results of case-control studies. However, studies reporting BMI and other cancer outcomes including survival, mortality, prognosis, recurrence and so on, were excluded. Systematic reviews without meta-analyses were also excluded.

Data Extraction
Data extraction was conducted independently by two authors (HJC and MKK). If there was a discrepancy, a third author (QD) made the final decision. Data extracted from eligible articles included 1) cancer outcomes, 2) category of exposure, 3) first author's name, 4) publication year, 5) number of case and total participants, 6) meta-analysis metric, 7) summary effect size (OR, odds ratio; RR, relative risk; HR, hazard ratio) and 95% confidence intervals, 8) number of included studies, 9) study design (cohort, case-control), 10) fixed or random effect model, 11) heterogeneity, 12) publication bias (Egger's test), and 13) statistical significance.

Methodological and Evidence Quality Assessment
The methodological quality of the included systematic review and meta-analysis was evaluated by the AMSTAR, a reliable, valid and widely used measurement tool based on 11 questions (13,14). The strength of evidence was assessed by using the GRADE (Grading of Recommendations, Assessment, Development and Evaluation), which classified the evidence as "very low", "low", "moderate" and "high" quality based on the assessment of risk of bias, inconsistency, indirectness, imprecision, publication bias and so on (15).

Data Analysis
We extracted the summary effect size with its 95% confidence interval (95% CI) and effects model reported in each meta-analysis if available. To assess the heterogeneity among studies, we used the I 2 statistic and Cochran's Q test. The heterogeneity was considered to be substantial or considerable if I 2 was more than 50% or 75%, respectively (16,17). Publication bias was assessed by calculating an estimate through Egger's regression test (18). A P value < 0.10 was considered to be significant for Egger's test. The outcome of the dose-response meta-analysis was also extracted if data were available in the included articles.

Literature Review
First, we retrieved 643 articles by searching the database. Second, 358 studies relevant to BMI and cancer were remained after deduplication. Then, after browsing the titles and abstracts, 265 articles were excluded. Next, 75 studies were also excluded after screening the full text. The selection process and reasons for exclusion are presented in a flow diagram ( Figure 1).
Finally, a total of 18 studies with 19 cancer outcomes involved seven physiological systems were regarded as eligible for the umbrella review ( Figure 2).

Characteristics of the Included Meta-Analyses
The characteristics of 43 meta-analyses of 19 types of tumors are presented in the Table 1. Among these, three meta-analyses were about the relationship between underweight and cancer risk, and 14 meta-analyses reported the relationship between overweight and cancer risk. The association between obesity and cancer risk was investigated by 16 meta-analyses. Moreover, dose-response meta-analysis was conducted by 10 studies. According to statistical significance, we concluded that the results of 37 meta-analyses were significant association, while the results of eight meta-analyses did not have a significant association.
The association between obesity and thyroid cancer was examined in 32 studies, and a meta-analysis of the included studies demonstrated that obesity could increase the risk of thyroid cancer by 33% (RR: 1.33, 95% CI: 1.24-1.42) compared to people with normal BMI categories. Moreover, obese men and women were both significantly at risk of thyroid cancer according to subgroup analysis by sex (34). When the risk estimates from 12 cohort studies of BMI and breast cancer incidence were combined, a 5 kg/m 2 increment in BMI was associated with a 2% (RR: 1.02, 95% CI: 1.01-1.04) increased risk of breast cancer (35). Based on a random effect model, compared to normal weight people, the estimated RR of Hodgkin's lymphoma was 0.97 for overweight (RR: 0.97, 95% CI: 0.85-1.12) and 1.41 for obese (RR: 1.41, 95% CI: 1.14-1.75), which indicated that the risk of Hodgkin's lymphoma increased by 41% for obesity, while no significant statistical significance was detected between overweight and risk of Hodgkin's lymphoma (36). The dose-response meta-analysis of 16 prospective cohort studies showed that every 5 kg/m 2 increase in BMI corresponded to a 7% (RR: 1.07, 95% CI: 1.04-1.10) increased risk of in non-Hodgkin's lymphoma (36). The results of dose-response meta-analysis were presented in the Figure 3.

Heterogeneity of the Included Meta-Analyses
Three meta-analyses reporting the relationship between underweight and cancer risk presented low levels of heterogeneity (I 2 < 25%). The association between overweight and cancer risk was reported by 14 meta-analyses, among which low level heterogeneity (I 2 < 25%) was reported by six meta-analyses and moderate-to-high level heterogeneity (I 2 25%-75%) was also reported by six meta-analyses, while high level heterogeneity (I 2 > 75%) was reported by two meta-analyses. Among 15 meta-analyses related to obesity, low levels of heterogeneity (I 2 < 25%) were shown by seven meta-analyses, and moderate-to-high levels of heterogeneity (I 2 25%-75%) were shown by seven meta-analyses, while only one presented high levels of heterogeneity (I 2 > 75%). In approximately, 10 dose-response meta-analyses, two dose-response meta-analyses (I 2 < 25%) showed low levels of heterogeneity, and moderate-to-high levels of heterogeneity were presented in 4 dose-response meta-analyses (I 2 25%-75%), while high levels of heterogeneity were only reported in one dose-response meta-analysis (I 2 > 75%). However, the I 2 statistic could not be found in 3 dose-response meta-analyses.

Publication Bias of the Included Meta-Analyses
In three meta-analyses about underweight, all detected a significant publication bias. Among 14 meta-analyses about overweight, two meta-analyses reporting kidney cancer and gastric cancer detected a significant publication bias, and eight meta-analyses did not, while the data of publication bias were not available for four meta-analyses. Regarding 15 meta-analyses about obesity, two meta-analyses involving gallbladder cancer and thyroid cancer had significant publication bias and 10 meta-analyses did not have, however, the data of publication bias were not available for three meta-analyses. In approximately 10 dose-response meta-analyses, significant publication bias was found only in one dose-response meta-analysis and not in 6 dose-response meta-analyses, while the data of publication bias were not available for three dose-response meta-analyses.

AMSTAR Assessment and GRADE Classification
The AMSTAR was used to assess the methodological quality of all included studies (supplementary Table  S1), and the results showed that the median AMSTAR score was 7 (range 5-9; IQR 7-8) ( Table 2). The AMSTAR score of twenty meta-analyses (47%) was more than 7, and it was less than 7 in twenty-three meta-analyses (53%).
The GRADE system was used to assess the quality of evidence of all outcomes (supplementary Table S2). The quality of most evidence evaluated by the GRADE score was classified as low or very low quality because the studies included in our umbrella review were mainly cohort and case-control studies (Table 2), which resulted in a serious risk of bias.

Principal Findings and Interpretation
The results revealed that underweight was inversely associated with the incidence of brain tumors and positively related to the risk of esophageal and lung cancer. Overweight would enhance the incidence of brain tumors, kidney cancer, endometrial cancer, ovarian cancer, multiple myeloma, bladder cancer and liver cancer. Obesity was related to the increased incidence of brain tumors, cervical cancer, kidney cancer, endometrial cancer, esophageal cancer, gastric cancer, ovarian cancer, multiple myeloma, gallbladder cancer, bladder cancer, colorectal cancer, liver cancer, thyroid cancer and Hodgkin's lymphoma. Moreover, dose-response analysis was conducted by 10 studies, and the results demonstrated that per 5Kg/m 2 increment of BMI was associated with a 1.01-to 1.13-fold increased risk of brain tumors, multiple myeloma, bladder cancer, pancreatic cancer, breast cancer, and non-Hodgkin's lymphoma. Every 1Kg/m 2 increase in BMI was linked to a 6% increase in the risk of kidney cancer and a 4% increase in the risk of gallbladder cancer.
Many potential mechanisms exist to explain the relationship between increment of BMI and the risk of various cancers. First of all, inflammation might be a bridge between obesity and tumors. Obesity itself is a chronic inflammatory condition (37). The initiation of obesity-associated inflammation is related to metabolic processes, and excessive nutrient consumption may be the major contributor (38). This kind of inflammation mainly exists in white adipose tissue, a special metabolic tissue made up of lipocytes (38). Adipose tissues have been regarded as one of endocrine organs that can secrete a class of substances called adipokines, including growth factors, hormones, cytokines and inflammatory factors. Adipokines can take part in a variety of biological processes in the body involving inflammatory reactions, immune responses, glucose metabolism and insulin sensitivity (39). Adipokines involved in the inflammatory response process include interleukin 1β, 6 and 8, tumor necrosis factor α (TNF-α), C-reactive protein (CRP), monocyte chemotactic protein 1 (MCP-1) and transforming growth factor β (TGF-β) (39,40). Excessive adipose tissues will also cause ischemia and hypoxia in tissues as well as cell death and the formation of crown-like structures (CLS), which are biomarker of inflammation (41). Furthermore, too many nutrients and obesity could activate the metabolic signaling pathways mediated by nuclear factor κ B (NF-κB), protein kinase R and c-Jun N-terminal kinase (JNK), which could result in a low-level inflammatory response in the body (42,43). Epidemiological studies and experimental data have demonstrated the association between chronic inflammation and cancer (44). The fact that anti-inflammatory therapies play a role in tumor treatment is further evidence of this relationship (45). Chronic infections with Helicobacter pylori, human papillomavirus and hepatitis virus are associated with gastric cancer, cervical cancer and liver cancer, respectively (41). Chronic inflammatory responses are involved in the initiation, proliferation and progression of cancer. On the one hand, a sustained inflammatory response can leave cells in oxidative stress for long periods of time, resulting in DNA and protein damage, inhibition of apoptosis and activation of proto-oncogenes, which are related to the occurrence and development of tumors (46). Oxidative stress refers to a state of imbalance between oxidation and antioxidant action in the body. Oxidation action is caused by the production of free radicals, including reactive oxygen species and reactive nitrogen species (ROS and RNS). Antioxidant action refers to the elimination of free radicals by protective mechanisms (47). Under normal circumstances, the oxidative and antioxidant effects in the body are in balance, and free radicals at this time have important signaling and physiological functions in human metabolism (48). However, when the balance is upset, the excess free radicals become toxic, reacting uncontrollably with lipids, proteins, and DNA in the body, causing damage to cells and tissues (46). Excess ROS and damage to cellular structures lead to activation of different kinases and transcription factors such as NF-κB or AP-1, which are key mediators leading to inflammatory response (49). These mediators promote the expression of pro-inflammation proteins to recruit and activate immunes cells, which can in turn promote a massive release and accumulation of free radicals, resulting in oxidative stress (50). Oxidative stress participates in the whole process of tumorigenesis and development, causes gene mutations and DNA structure changes in the initial stage of tumors, hinders intercellular communication and changes the second messenger system in the tumor development stage, resulting in increased cell proliferation or decreased apoptosis of mutant cells (51).
On the other hand, some inflammatory factors not only play a key role in the inflammatory responses but also promote cancer occurrence. An animal experiment has showed that overexpression of IL-1β could result in gastric carcinoma in mice without Helicobacter infection (52). IL-6 has also been reported to promote tumorigenesis by meditating the gp130/JAK/STAT3 signaling pathway (53). Second, the relationship between being overweight and obesity and increasing cancer risk could be explained by an imbalance between leptin and adiponectin. Produced by white adipose tissue, adiponectin was reported to play a significant protective role in carcinogenesis (54). On the one hand, adiponectin can activate the receptor-meditated signaling pathway to have a direct effect on tumor cells. It works mainly through three known receptors including adipoR1, adipoR2 and T-cadherin (55,56). On the other hand, it has anti-inflammatory properties, regulating insulin sensitivity, influencing angiogenesis and inhibiting the growth, proliferation, invasion, and metastasis of tumor cells (57). In addition to being influenced by genetic factors, diet and physical activity, the level of adiponectin was also affected by obesity. In obese people, circulatory adiponectin levels decrease, which may result in the systematic chronic inflammation and increase the risk of various cancers. Numerous studies have demonstrated that low adiponectin levels are related to an increased risk of endometrial cancer, postmenopausal breast cancer and colon cancer (58)(59)(60). Leptin is also an adipokine, whose level in the circulatory system is positively associated with the total body fat (61). In contrast to adiponectin, leptin has been demonstrated to promote inflammatory response and angiogenesis, inhibit apoptosis and stimulate cell growth, migration, and invasion, which might play a significant role in the development and risk of cancers (62)(63)(64). Previous studies have shown that elevated serum leptin was correlated with an increased risk of endometrial cancer and breast cancer (65,66).
Third, insulin resistance/hyperinsulinemia is one of the characteristics of obesity (67). The level of adiponectin decreases in obesity, which could result in insulin resistance, because adiponectin could stimulate AMPK (AMP-activated protein kinase) phosphorylation and activation to enhance insulin sensitivity (68). Increased inflammatory factors such as IL-6 and TNF-α in obese people could also lead to insulin resistance by activating the JNK (c-jun amino-terminal kinase) and IKK-β (IκB kinase-β)/ NF-κB (nuclear factor-κB) pathways (69). Insulin is not only a metabolic hormone but also a growth factor that can promote mitosis, especially on malignant cells with overexpressed insulin receptor (IR) (70). The biological effect of insulin is achieved by binding to IRs, including IR-A and IR-B subtypes and the activation of the former subtype will have a stronger mitogenic effect (71). Previous studies have reported that people with diabetes, obesity and other characteristics of insulin resistance/hyperinsulinemia have a higher risk of endometrial cancer, early gastric cancer and breast cancer (72)(73)(74). Increased insulin could also decrease the level of IGF-1-binding proteins synthesized by the liver; therefore, free IGF-1 will increase (70,75). IGF-1 has been reported to be a significant mediator of the effect of growth hormone, which could promote the proliferation and differentiation of cells as well as inhibit cell apoptosis. Moreover, the activation of IGF-1R can induce malignant transformation, and the transformation will stop when the expression of IGF-1R is suppressed by disrupting the IGF-1R gene (76,77). Many prospective studies have shown a possible association between high levels of IGF-1 in circulation and an increased risk of multiple cancers, including prostate cancer, colorectal cancer and breast cancer (78)(79)(80).
In addition, the potential biological mechanisms linking overweight/obesity and cancer also comprise the level of sex hormones, ectopic fat deposition and changes in the tumor microenvironment (81,82).

Strengths and Weaknesses of the Study
Umbrella review has become increasingly popular, especially in the last 2 years, and it has been regarded as one of highest levels of evidence synthesis because it includes meta-analyses with a relatively large number of subjects. To the best of our knowledge, this study is the first umbrella review investigating BMI and cancer risk. However, there were also several limitations. First of all, the studies included in the meta-analyses of our umbrella review were mainly cohort and case-control studies and lacked experimental studies, which will have an impact on the overall quality of the study. Second, this umbrella review was a comprehensive evaluation of existing systematic reviews and meta-analyses about BMI and cancer risk. Consequently, some related studies that have not been published before our search will not be included in this study. Third, among the 42 meta-analyses, five meta-analyses reported significant publication bias, and the data about publication bias were not available for 10 meta-analyses. Last but not least, the risk of tumor occurrence is the result of many factors, which might exert an impact on associations between BMI and cancer incidence.

Conclusion
In conclusion, the evidence presented in this umbrella review showed that overweight and obesity were associated with an increased risk of most cancer outcomes. We recommend that it is better to maintain BMI within the normal range to prevent the occurrence of tumors.