This study was conducted using the preferred reporting items for systematic review and meta-analysis (PRISMA) standards (29). To find pertinent papers up to June 2022, we conducted a thorough search of the online databases of PubMed, Scopus, and ISI Web of Science. In the systematic search, we utilized both MeSH (medical subject heading terms) and non-MeSH terms (Supplementary Table 1). The publishing schedule and language of the pieces were both unrestricted. Following the thorough search, all of the results were imported into Endnote software before the screening process began. In Endnote, duplicate citations were eliminated. We also performed a web-based search in Google Scholar using the phrases “body mass index” and “cataract” in addition to the databases already listed. In order to ensure that we did not overlook any publications, we lastly checked the reference list of the chosen articles.

The following criteria were considered to select eligible studies in the screening step: (1) studies with prospective cohort designs, such as prospective cohort, nested case–control, and case-cohort studies; (2) studies on healthy adults (18 years); (3) studies measuring BMI to assess general obesity; (4) studies taking into account ARC or its subtypes, such as nuclear, cortical, or PSC as an outcome variable; and (5) studies reporting hazard ratio (HR), risk ratio (RR), and odds ratio (OR), with 95% confidence intervals for the association between obesity and ARC. If we found two papers that were published on a population, only the paper with higher quality or the most number of cases was included in our systematic review and meta-analysis. We disregarded retrospective, case–control, and cross-sectional studies as well as cohort studies that enrolled critically ill patients or people with chronic illnesses including diabetes mellitus and chronic kidney disease. Additionally, studies that did not report relative risks for the link between obesity and cataracts or lacked the necessary information to calculate these effect sizes were disregarded. Two independent reviewers chose the studies by taking the inclusion and exclusion criteria into account.

Two independent reviewers extracted data from each selected article and entered the data in an Excel-based form that was previously designed. Any discrepancies were discussed with a third reviewer in order to be rectified. Based on the form, the following information was extracted from each article: first author’s name, year of publication, cohort name, geographical region, characteristics of participants (age and gender), sample size, number of cases with ARC, follow-up period, methods used for the assessment of obesity and cataract, relative risk estimates, including ORs, RRs, and HRs for the link between obesity and ARC risk, and confounding variables adjusted in statistical analysis. If research did not offer the necessary estimations, we calculated them using conventional techniques.

We used the Newcastle Ottawa Scale (NOS), designed for prospective cohort studies, to assess the quality of included studies (30). Based on the NOS, each cohort study can get a maximum of nine points: four for the selection of participants, two for comparability, and three for the assessment of outcomes. In the current study, we categorized studies based on the median score of NOS in which studies with a score more than the median were considered high-quality ones.

We included the RRs, HRs, or ORs and 95% CIs reported for the association between obesity and ARC risk into the meta-analysis. These RRs were calculated based on the comparison between the highest versus lowest categories of BMI. However, it should be noted that some studies reported RRs of ARC per one-SD increment in BMI. To include in the meta-analysis, we converted the per SD increment risk estimates to the RRs for the comparison of the top versus bottom tertiles of BMI using the method suggested by Danesh et al. in which the log risk estimates reported for the comparison between the top and bottom tertiles of exposure variable are equivalent to 2.18 times of the log risk estimates for a 1-SD increase in that variable (31). This method assumes that the exposure is a normally distributed variable and that the association with disease risk is log-linear. To combine the RRs of ARC, a random-effects model was used. Random-effects models take into consideration different sources of uncertainties including within-study (sampling or estimation) error and between-studies variance (32, 33). To assess heterogeneity among studies, we used Cochran’s Q test and the I² statistic. For the I² statistic, we considered the I² values of >50% as high between-study heterogeneity (34). To find possible sources of heterogeneity, subgroup analyses were conducted. Publication bias was examined using Egger’s linear regression test for the associations with more than 10 effect sizes (35). In the case of substantial publication bias, the trim-and-fill method was used to detect the effect of possibly missing studies on the overall RR (36). To assess the dependency of overall ES on one study, sensitivity analysis was done using a random-effects model in which each study was excluded to examine the influence of that study on the overall estimate.

Since the highest and lowest categories of BMI were different across the included studies, we performed a dose–response meta-analysis to determine the RR of ARC at different levels of BMI. We applied the method described by Crippa et al. to do a dose–response meta-analysis (37). In this method, the number of participants and cases of cataract and also the RR of ARC in each category of exposure (BMI) were required. In each category of BMI, we considered the median or mean amount of BMI as the corresponding RR of ARC. For studies that reported BMI as ranges, we estimated the midpoint in each category by calculating the mean of the lower and upper bound. When the highest or lowest category was open-ended, the length of the open-ended interval was assumed to be the same as that of the adjacent interval. We conducted the one-stage dose–response meta-analysis using restricted maximum likelihood estimation to assess linear and non-linear associations (37). This method estimates the study-specific slopes and combines them to obtain an overall average slope in a single stage, and is a more precise, flexible, and efficient method than the traditional two-stage method. Statistical analyses were conducted using STATA version 14.0. p < 0.05 was considered statistically significant for all tests, including Cochran’s Q test.

In our initial search, we found 3,777 articles among the online databases, of them, 750 papers were duplicated. After excluding duplicate papers, we screened the remaining articles (n = 3,027) and disqualified any research that failed to fulfill the inclusion requirements (n = 2,997) (Supplementary Figure 1). After full-text reviews, nine articles were excluded because of being conducted on patients who underwent kidney transplantation or those with chronic diseases (n = 2) (38, 39), having a case–control or cross-sectional design (40–42), assessing waist circumference, weight or weight changes rather than BMI (28, 43, 44), and reporting incomplete data (45). Also, we excluded the study of Yuan et al. because they genetically predicted BMI (46). In addition, we found three different articles from the Physicians’ Health Study (21, 47, 48), two different papers from the UK Biobank (12, 49), and two different publications from the Blue Mountains Eye Study (22, 50). Since these publications evaluated similar associations, only the study with the highest quality or the greatest number of cases was considered for each dataset (12, 21, 22) and the duplicate papers were excluded (47–50). Moreover, two articles were published on the Beaver Dam Eye Study; however, both assessed different exposure and outcome variables in terms of BMI and cataract, and therefore, both were included (14, 16). After these exclusions, 16 articles containing 16 prospective cohort studies were included in the current systematic review and meta-analysis (10–25): 10 articles assessed BMI and risk of ARC (10–14, 21–25), 10 articles evaluated BMI and risk of nuclear cataract (13–15, 17–23), 6 papers assessed BMI and risk of cortical cataract (13, 16, 17, 20–22), and 7 publications assessed BMI and risk of PSC (13, 16, 17, 20–23). The flowchart of study selection is shown in Supplementary Figure 1.

Characteristics of prospective studies included in the current systematic review and meta-analysis are shown in Table 1. The sample size of these studies ranged from 372 to 1,312,051 participants. In total, these studies recruited 1,607,125 participants with an age range of ≥40 years. In addition, during follow-up periods ranging between 4 and 28 years, 103,897 cases of ARC, 4870 cases of nuclear cataract, 1,611 cases of cortical cataract, and 1,603 cases of PSC were detected. The included articles were published between 1998 and 2016. Among the 16 articles, one article included only males (21), two articles performed analysis on only females (12, 19), and the remaining articles included both genders in statistical analysis (10, 11, 13, 15–18, 20, 22) or presented gender-stratified risk estimates (14, 23–25). In terms of geographical region, included studies were conducted in the US (11, 13, 14, 16, 18–21, 23, 24), Europe (10, 12, 15), Asia (17, 25), and Australia (22). In 7 articles, researchers measured weight and height using a standard protocol for calculating BMI (13, 15, 17–20, 22), while in 9 studies, self-reported weight and height were used (10–12, 14, 16, 21, 23–25). Regarding outcome assessment, researchers performed a direct examination for cataract diagnosis in four articles (13, 17–19) and used data from medical records/registries in 3 articles (10, 12, 15). Among the remaining articles, self-reported data were used for cataract assessment. Of the 16 included articles, 13 papers presented adjusted risk estimates for the association between BMI and cataract risk (10–16, 19, 21–25). Some important confounding variables including age (n = 12), smoking (n = 9), and having diabetes mellitus (n = 8) were adjusted in these studies. By considering the median NOS score of 7, 13 articles, of the 16 papers, had high quality or low risk of bias in most components of NOS (10–16, 19–24) (Supplementary Table 2).

Of the 10 articles that assessed the association between BMI and risk of ARC, 6 papers showed a significant positive association (11, 12, 21, 23–25) and others did not find any significant association. In addition, two articles, among the 10 publications on the link between BMI and risk of nuclear cataract, indicated a significant positive association (19, 21), while the remaining articles reported a non-significant association. None of the studies that examined the association between BMI and risk of cortical cataract revealed a significant association. In terms of BMI and risk of PSC, a significant positive association was reported in 3 articles (16, 21, 23) of the 7 publications.

In this section, we included all studies that were evaluated in the systematic review. Below, findings from the meta-analysis were reported for ARC and its subtypes.