Associating the risk of three urinary cancers with obesity and overweight: an overview with evidence mapping of systematic reviews

This overview was registered by the PROSPERO (International Prospective Register of Systematic Reviews), number CRD42019119459. This overview was conducted following the Preferred Reporting Items for OoSRs (PRIO-harms) checklist.

Two authors independently searched the PubMed, Cochrane Library, and EMBASE databases in March 2019. The search was limited to articles written in English and used the following search terms: BMI, obesity, cancer, carcinoma, neoplasm, meta-analysis, and SRs. A detailed description of the search strategy is presented in the supplementary material. In addition, we conducted supplementary retrieval of all included references and updated the search in November 2020 to enable a comprehensive search.

Studies were included if (1) SRs or meta-analysis associated obesity and overweight with incidence and mortality of aforementioned urinary cancer, (2) SRs or meta-analysis associated increase in BMI with incidence and mortality of aforementioned urinary cancer, (3) articles were published in English, and (4) latest article was included when SRs or meta-analysis had been updated. Studies were excluded if (1) they were only abstracts and/or letters; (2) SRs and meta-analysis examined the association between BMI increase and prognosis, survival, or recurrence of urinary cancers; (3) protocols of SRs and meta-analysis or methodological articles; and (4) SRs without meta-analysis.

The retrieved articles were first imported into the EndNote X7 software, then titles and abstracts independently selected by two reviewers. The reviewers thereafter retrieved full texts of potentially eligible studies and independently subjected them to the aforementioned criteria (J.Y.S and X.N.M.). Any disagreement was discussed with a third reviewer. The two reviewers independently extracted the following characteristic from each study: first author’s name, year of publication, funding, number of reference test, name of the database, country of the first author, the epidemiological study design (case-control or cohort), number of cases, features of the urinary cancers, summary effects between BMI and cancer risk (at 95% CI), and the number of included studies.

The quality assessment of included SRs was independently performed by two review authors (Y.G. and J.Y.S.) according to the predefined criteria. Disagreements regarding by-item and overall rating of quality were resolved by consensus or third-party adjudication (P.W. or J.H.T.). The methodological quality was assessed using the assessment of multiple systematic reviews-2 (AMSTAR-2) tool, which is a reliable methodological quality tool applicable to SRs of randomized and/or non-randomized studies with good agreement, construct validity, and feasibility. The AMSTAR-2 contains 16 items, among which seven are critical domains. Each item was responded to “yes” (item/question fully addressed), “no” (item/question not addressed), or “partial” (item/question not fully addressed). The overall confidence of the quality of each SR was classified as high, moderate, low, or critically low according to the critical and non-critical domains [13, 14].

To assess the reporting quality of included SRs and meta-analyses, we used the PRISMA checklist, which comprises 27 items. To show the degree of compliance, total PRISMA scores were calculated by summing 1 point for “yes” (total confirmed), 0.5 points for “partial” (partial confirmed) and “cannot answer” (limited information), and 0 points for “no” (non-compliance) [15]. The SR and meta-analysis were regarded as major flaws if PRISMA scores were below 15 points, minor flaws if they recorded 15.0–21.0 points, and minimal flaws if > 21.0 points were recorded [14, 16]. Quality assessment of the included SRs and meta-analyses was independently performed by two authors (J.Y.S. and Y.G.), and any disagreements between them were discussed with a third author (J.H.T.).

We created a bubble plot with Microsoft Excel 2016 (Microsoft Corp., Redmond, WA, www.​microsoft.​com) to present the compliance of AMSTAR-2 and PRISMA of the included SRs. The evidence map displayed information in four dimensions: (a) the X-axis displayed different items of AMSTAR-2 or PRISMA; (b) the Y-axis represented the number of each item of AMSTAR-2 or PRISMA defined as “yes,” “partial,” and “no”; (c) the bubble size represented the compliance of each item of AMSTAR-2 or PRISMA defined as “yes,” “partial,” and “no”; (d) the bubble color indicated the items of AMSTAR-2 or PRISMA defined as “yes,” “partial,” and “no.”

The pairwise meta-analysis was performed with the data of pooled RRs (at 95% CI) from included SRs and meta-analysis (HRs and ORs equivalent to RR) records using the fixed-effects model or random-effects model [17]. A condition was considered normal if a BMI of 18.5–24.9 kg/m² was recorded, overweight for 25–29.9 kg/m², and obesity for BMI ≥ 30 kg/m². We also analyzed each 1 kg/m² and 5 kg/m² increase in BMI according to previous protocols [18, 19]. In addition, we assessed the heterogeneity between studies using the I² statistic [20]. Specifically, we adopted the fixed-effect model when the I² value was less than 50%, and the random-effect model for an I² value was greater than 50% [21]. We carried out subgroup analyses according to the sex, location, history of smoking, age, family history of cancer, duration of follow up, different types of study (e.g., cohort study or case-control study), physical activity, methods of BMI measured, history of hypertension, history of alcohol drinking if data were sufficient, and if the definition of the subgroup is different, only the high methodological quality results were included. Furthermore, we analyzed the indirect comparisons of the outcomes across the meta-analyses. Statistical analyses were performed using the STATA software (version 12.0, College Station, TX, USA), with values that had P ≤ 0.05 considered statistically significant.

Our search resulted in a total of 733 articles, with 221, 232, and 275 identified from PubMed, EMBASE, and Cochrane Library, respectively. Additional five were retrieved from reference lists or other sources. A screening flow chart of the studies is presented in Fig. 1. After retrieval, a total of 248 duplicates were removed. A review of full-text articles resulted in the exclusion of 14 studies, with 31 SRs and meta-analyses meeting our criteria. Of these, 7 described mortality, 22 reported incidence, and 2 were studies combining mortality and incidence. A detailed description of these characteristics is outlined in Table 1.

Characteristics of included SRs and meta-analyses included in the current study are outlined in Table 1 [22‐52]. Among the 31 records that met our inclusion criteria, 4 SRs reported bladder cancer cases [22‐24, 40], 9 SRs and meta-analyses focused on kidney cancer [25‐30, 47‐49], 11 SRs and meta-analyses described prostate cancer [31‐37, 42, 45], and 7 meta-analyses studied more than one type of urinary cancers [38, 39, 41, 43, 44, 46, 50]. The included SRs and meta-analyses were published between 2001 and 2018, with only one record retrieved from Chinese databases [22]. One record had co-first authors [24], whereas the median number of authors was 4.5 and ranged from 2 to 10. In addition, five articles did not report sources of funding [26‐30], five others provided full search strategy [29, 39, 40, 43, 46], and only one was registered with PROSPERO [40].

Thirty-one SRs were available for quality assessment. Considering the overall methodological quality, all the SRs were rated as critically low quality. Specifically, of the sixteen individual items, one was fully reported (research questions and inclusion criteria for the review include the components of PICO). However, none of the SRs explicitly stated that the review methods were established before the conduct of the review and justified significant deviations from the protocol, and report on the sources of funding for the studies included in the review (Fig. 2). In general, 72.41% of the articles were of moderate quality (Table 1). The total of PRISMA points across the included studies was 21.5 (12.5–26) (Fig. 3). Particularly, four items were fully reported, whereas only one [40] provided the protocol’s registration number. In addition, three meta-analyses [27, 28, 44] had low reporting quality (PRISMA score < 15.0), 11 [22, 23, 25, 26, 32, 33, 35, 36, 38, 45, 48] were moderate (15.5–21.0), and 17 articles [24, 29‐31, 34, 37, 39‐43, 46, 47, 49‐52] were of high quality (21.5–26.0) (Table 1).

A summary of the effect of each 5 kg/m² BMI increase on mortality rates resulting from urinary cancers is provided in Table 6. Specifically, mortality rates of bladder [39] (1.05, 95% CI 1–1.11), kidney [39] (1.21, 95% CI 1.14–1.29), and prostate [32, 33, 39, 42, 45] (1.15, 95% CI 1.11–1.2) cancers were all associated with BMI increase. Zhang et al. [30] reported a strong association between mortality rates from kidney cancer with obesity (RR = 1.71, 95% CI 1.27–2) and overweight (RR = 1.19, 95% CI 1.05–1.35), whereas Zhong et al. [37] observed that prostate cancer-related mortality was strongly associated with per 5 kg/m² BMI increase (HR = 1.15, 95% CI 1.07–1.23). Conversely, each 5 kg/m² BMI increment showed no effect on all-cause mortality in prostate cancer patients.

Methodological and reporting quality of meta-analysis and SRs are crucial to public health and clinical decision-making. In the current study, we used the AMSTAR-2 checklist to assess the methodological quality. The methodological quality of the SRs was classified as critically low, and the adherence rate of individual items was suboptimal. Research protocols help to increase the transparency of the review methods and avoid bias in outcome reporting. The previous study has shown that prospective registration could effectively improve the overall methodological and reporting quality of systematic reviews. However, none of the SRs provided an explicit statement that the review methods were established before the conduct of the review and clarified the significant deviations from the protocol and reported on the sources of funding for the studies included in the review. 80.6% of SRs did not account for RoB in primary studies when interpreting/discussing the results of the review. 80.6% SRs did not explain the selection of the study design for inclusion and provide a list of exclusion studies and justify the exclusions PRISMA results showed that 9.6% of the included studies had serious, 35.4% had minor, and 54.8% had minimal reporting flaws. The scientific quality of individual studies affected the findings of this meta-analysis [54], although some of the articles did not report methods for assessing the risk of bias of different studies. For example, Cohen et al. [54] suggested searching Chinese databases may yield substantial additional clinical evidence, and recommended by the Chinese Biomedical Database for retrieval of SR literature. However, most SRs and meta-analyses are limited to English. In addition, some items did not have good performance, such as a structured summary, risk of bias, search, results of individual studies, and especially the protocol registration. Currently, the PRISMA checklist provides excellent guidelines for developing a high-quality SR [55], although improvement of the reporting quality is still needed.

Our comprehensive analysis revealed a strong association between kidney cancer and obesity, overweight, and BMI increase. On the other hand, a mild relationship was observed between the incidence of prostate and bladder cancers with obesity, overweight, and BMI increase, in line with Renehan et al.’s work [43].

We detected heterogeneity among the SRs and meta-analyses included in this study. To further demonstrate the relationships between the risk of urinary cancer and obesity, we performed subgroup analysis targeting major potential confounders such as gender, age, methods of assessing BMI, geographic location, physical activity, and family history of cancer. The results revealed a stronger association between the incidence of kidney and prostate cancers and BMI increase (5 kg/m²) in Asian populations compared to European and North American groups. This association was also evident between overweight and kidney cancer, although this was different from previous studies [39, 43]. The association between the incidence of kidney cancer and obesity, overweight, and BMI increase (5 kg/m²) was more influential in women than in men. However, we did not find convincing evidence to support the existence of differences across genders with regard to the relationship between BMI increase and incidence of bladder cancer. Age (> 50 years) was a confounding factor in the relationship between obesity and the prevalence of bladder cancer, although this trend was reversed in people with kidney cancer. Additionally, alcohol consumption was a confounding risk factor in the association between obesity and the incidence of bladder cancer.

Our results revealed a mild association between mortality rate in prostate and bladder cancers with increased BMI (5 kg/m²), whereas kidney cancer had a strong relationship, consistent with previous reports [39]. Increased BMI (5 kg/m²) also showed a strong association with mortality rates from prostate cancer, although this impact was not significant in all-cause mortality of prostate cancer. Although previous reports have confirmed that small sample size studies do not affect the results in the absence of obvious methodological deficiencies and marked differences among studies, it is still possible that a small sample size may affect our assessment of the association between increased BMI and urinary cancers [56].

Our study had several limitations. Firstly, our search criteria were limited to SRs and meta-analyses written in English, which may lead to language bias. Some studies have reported good-quality SRs published using other languages [57, 58]. Secondly, our findings including assessments of certainty of evidence are based on the information provided by the authors of the reviews, and we have not retrieved or evaluated data from any primary studies. Thirdly, we comprehensively performed subgroup analyses; however, it was unclear how original systematic reviews were classified for variables such as BMI self-reported and physical activity [39], and we did not perform all subgroup analyses to evaluate the impact of other factors on the incidence of three kinds of urinary cancers as prior defined due to limited data. Fourth, while it could be expected that there would be some overlap of primary articles within included SRs, we have not systematically explored these overlaps. Consequently, this may lead to inaccuracies in the reporting of data such as the numbers of participants and primary studies and may contribute to “double counting” of data within reported meta-analyses. Fifth, due to limited data, we did not perform subgroup analyses to evaluate the impact of BMI increase on mortality of urinary cancers.