Participants of this study were selected from the UK Biobank, which is a large community-based cohort of over 500,000 participants from the United Kingdom [15]. Detailed study methodology has been reported previously [15]. In brief, participants aged 40 to 69 years who were registered with the National Health Service (NHS) and lived within 25 miles of any of the 22 assessment centres were invited to join the study at baseline between 2006 and 2010. The UK Biobank was conducted in accordance with the principles of the Declaration of Helsinki, and ethics approval was granted by the National Information Governance Board for Health and Social Care and the NHS North West Multicenter Research Ethics Committee (REC reference: 16/NW/0274). All participants provided informed consent through electronic signature at recruitment. The UK Biobank has been established as an open‐access resource and is globally accessible to approved researchers and scientists undertaking research to improve public health [16]. The present study was conducted under application number 62443 of the UK Biobank resource. The participants selection flowchart is shown in Fig. 1.

Study flowchart of population selection from the UK Biobank.

A web-based 24 h dietary assessment tool, the validated Oxford WebQ, was used for dietary intake assessment in a subgroup of participants between 2009 and 2012 [17]. Collection of dietary data from the food frequency questionnaire in UK Biobank had been validated previously [18]. The consumption levels of F&V were categorized into five groups, with cut-off points based on the distribution of intake frequency [19]. Types of F&V intake were also linked to incident cataract. The questionnaire was administered online and only participants who finished at least one of the five questionnaires were included in the current analysis. We further excluded those without self-reported eye health data or with cataract at baseline.

The amount of each food consumed was calculated by multiplying the assigned portion size of each food by the quantity consumed. Energy intake was calculated by multiplying the quantity of consumption of each food by energy of the portion (as taken from McCance and Widdowson’s The Composition of Foods and its supplements) and then summing this across all food items [17].

Basal metabolic rate was estimated using the Henry equation [20]. Participants deemed to have under-reporting (defined as total energy intake <1.1 × basal metabolic rate) or over-reporting (defined as >2.5 × basal metabolic rate) of total energy intake were further excluded from the analysis.

Cataract during the follow-up was defined by self-report or hospital inpatient records, using codes for International Classification of Diseases (ICD) numbers of H250, H251, H252, H258, H259, H261, H262, H263, H264, H268, H269, H280, H281, H282; ICD9: 366, 3661, 3662, 3663, 3664, 3665, 3668, 3669. In addition, we used surgical procedures (OPCS4) to identify cataract events (codes: C71.2 or C75.1). The earliest recorded code date was used as the onset date of cataract. Person-years were calculated from the date of baseline assessment to the date of onset cataract, date of death, or the end of follow-up (December 31, 2020, for England and Wales and January 18, 2021, for Scotland), whichever came first.

Participants answered a detailed touch-screen questionnaire which included information regarding their age, gender, ethnicity, education, household income, history of disease and surgery, use of vitamin supplement (yes/no), as well as lifestyle factors, including sleep duration (hours/day), alcohol drinking (never/previous/current) and smoking status (never/former/current). Physical activity (PA) was assessed using the short form International Physical Activity Questionnaire [21], and the metabolic equivalent (MET)- minutes/week of PA was calculated based on their answers to time spent on walking, moderate PA and vigorous PA. Weight was measured with the BV-418 MA body composition analyser (Tanita), and height was measured in a barefoot standing position using a Seca 202 height measure. Body mass index was calculated based on measured weight (kg) divided by measured height (m) squared. Depression was recorded during the interview with a research nurse. Blood cholesterols, including triglycerides, high-density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C), were measured by direct enzymatic methods ((Konelab, Thermo Fisher Scientific, Waltham, Massachusetts), and Glycosylated haemoglobin, Type A1C (HbA1c)) was measured using high-performance liquid chromatography on a Bio-Rad Variant II Turbo.

Ethnicity was categorized into three groups (whites, non-whites and unknown) and education was categorized into four groups (0–5 years, 6–12 years, ≥13 years and missing). Household income was also divided into seven subgroups, including <18000, 18000–30999, 31000–51999, 52000–100000, >100000, unknown and not answered.

One-way ANOVA and chi-square test for categorical variables was used to examine the difference of baseline characteristics among participants with different quintiles (Q) of fruit and vegetable intake. Cox Proportional Regression models were used to estimate the risk for incident cataract associated with vegetable and fruit intake. Model 1 was adjusted for age and gender; Model 2 was adjusted for Model 1 plus ethnicity, education, household income, total energy intake, vitamin supplement intake, alcohol consumption, physical activity, smoking, and sleep duration; Model 3 was adjusted for Model 2 plus BMI, HDL-C, LDL-C, triglycerides, HbA1c, hypertension, and depression; Model 4 was adjusted for Model 3 plus vitamin D and medications for lipids, blood pressure, or glucose lowering. Cox Proportional Regression models were used to test whether the association between vegetable/fruit intake and incident cataract was moderated by age, gender, education, smoking, obesity, hypertension, diabetes, or depression. The analysis was adjusted for age, gender, ethnicity, education, household income, total energy intake, alcohol consumption, physical activity, smoking, sleep duration, BMI, HDL-C, LDL-C, triglycerides, HbA1c, hypertension, and depression. All analyses were completed using the SAS software package (version 9.4; SAS Institute, Cary, NC, USA). A two-tailed P-value of <0.05 was used as the level of statistical significance.