Genetic factors, adherence to healthy lifestyle behaviors, and risk of bladder cancer

The UK Biobank is a large ongoing population-based prospective cohort of over 500 000 individuals aged 37–73 years. Between 2006 and 2010, eligible participants were invited to attend one of 22 assessment centers throughout England, Scotland, and Wales. During the baseline assessment, participants were asked to complete a touchscreen questionnaire and a brief interview to collect sociodemographic, lifestyle factors and health-related information. Participants also underwent a range of physical measurements and provided biological samples. More details of UK Biobank could be found elsewhere [24]. Each participant provided written informed consent before data collection. The UK Biobank study has been approved by the North West Multi-centre Research Ethics Committee.

In the present study, we included 488 169 participants with available genetic data and excluded participants of non-European ancestry or related individuals (n = 69 790), those with a prior cancer diagnosis (n = 31 237), those with incomplete data on lifestyle factors (n = 10 223), and those who subsequently withdrew from the study (n = 921). Figure S1 in the Supporting Information showed the flowchart of the study sample selection.

In the UK Biobank, participants were followed through records linkage to the Health and Social Care Information Centre (in England and Wales) and the National Health Service Central Register (in Scotland). These registrations recorded the diagnosis of cancer and cancer deaths using the 10th revision of the international classification of diseases (ICD-10) codes. The primary outcome of the study was incident bladder cancer (ICD 10 C67), or bladder cancer listed as the underlying cause of death on the death certificate.

At baseline, information on sociodemographic characteristics (age, sex), lifestyle habits (smoking status, physical activity, and dietary intake) and health and medical history, were obtained using a self-administered touchscreen questionnaire and nurse-led interviews. Physical activity was assessed using the validated Short International Physical Activity Questionnaire (IPAQ). Height, weight and waist circumference (WC), were measured by trained research staff using standardized procedures. More details of these measures can be found elsewhere [24].

We created a healthy lifestyle score (HLS) based on the new World Cancer Research Fund/American Institute of Cancer Research (WCRF/AICR) lifestyle score [25]. Specifically, the HLS was generated from the combination of four lifestyle factors, including body weight, smoking status, physical activity, and diet score (see Supporting Information Table S1). According to WCRF guidelines, each component was categorized into 3 groups (optimal, intermediate, poor). Body weight was classified according to an individual’s BMI and WC. Optimal weight was defined as 18.5 ≤ BMI < 25 kg/m² as well as WC < 94 cm in men and WC < 80 cm in women, poor weight as BMI ≥ 30 kg/m², or WC ≥ 102 cm in men and WC ≥ 88 cm in women, or intermediate (all other combinations). Smoking status was defined as optimal if participants never smoked or quit 10 years ago, intermediate if they were previous smokers, or as poor if they were current smokers. Physical activity was defined as optimal if individuals had ≥ 150 min/week per week moderate or ≥ 75 min per week vigorous or an equivalent combination, intermediate as 1– 149 min/week moderate or 1–74 min/week vigorous or 1–149 min/week mixed activity, and poor as not performing any moderate or vigorous activity. A diet score was built based on the sum of intake of fruits and vegetables, whole grains, red and processed meat, and was then categorized into 3 groups based on tertiles (optimal, intermediate, poor). Given that the greater consumption of tea could reduce the risk of bladder cancer based on WCRF/AICR report [26], we added the tea intake as a new component to the diet score. The healthy lifestyle was categorized into optimal (having at least 3 optimal lifestyle factors), poor (having at least 3 poor lifestyle factors), or intermediate (all other combinations).

We also constructed a weighted healthy lifestyle score based on the β coefficients of each lifestyle factor in the Cox model for bladder cancer risk as a sensitivity analysis. The weighted healthy lifestyle score was derived by summing the product of each lifestyle factor with its corresponding β coefficient, then dividing the sum by the total of the β coefficients, and subsequently categorizing it into three groups based on tertiles (optimal, intermediate, poor).

Details of the array design, genotyping process and quality control in the UK Biobank can be found elsewhere [27]. In brief, 488 170 participants had DNA samples assayed using two genotyping arrays sharing a 95% marker content, the UK Biobank Axiom array (825 927 markers) and the UK BiLEVE Axiom Array (807 411 markers in 49 934 participants) [27]. Genotype data was phased and imputed with SHAPEIT3 and IMPUTE3 based on reference panels of the Haplotype Reference Consortium or UK10K and 1000 Genomes phase 3. From the resulting dataset, we excluded participants with a mismatch between reported and genetic sex, sex chromosome aneuploidy, a high missingness or excess heterozygosity, participants of non-European ancestry, second-degree (or higher) related participants (kinship coefficient ≥ 0.088).

In previous published studies, a PRS for bladder cancer was built based on the 15 SNPs identified in previous GWAS among individuals of European ancestry [7, 8, 18, 28]. Using a similar approach, we searched and reviewed the previous published GWAS of bladder cancer to identify susceptibility SNPs. A total of 16 SNPs with minor allele frequency (MAF) ≥ 0.01 and P < 5 × 10⁻⁸ were used to construct the PRS in our study (Supporting Information Table S2). The corresponding effect estimates were extracted from the largest studies accordingly. For risk variants that were not available in UK Biobank data, the proxy SNPs (r² > 0.8, reference population: EUR) were identified using the LDlink (https://​ldlink.​nci.​nih.​gov/​) [29]. We excluded SNPs identified exclusively from non-European populations, with allele mismatches or in high linkage disequilibrium (LD; r² ≥ 0.2), and palindromic SNPs with MAF ≥ 0.45 [28, 30].

PRS for each study participant was constructed by summing the product of the number of risk alleles (0, 1, and 2) for each SNP and its respective weight [30]. The PRS was then categorized into low (the bottom quintile of PRS), intermediate (quintiles 2–4), and high (the top quintile).

We calculated person-years from the recruitment date to the date of the first diagnosis of any cancer, death, or the last date of follow-up (31 December 2020), whichever came first. Multivariable Cox proportional hazards models were fitted to evaluate the effects of genetic factors, lifestyle factors, and the combination of genetic and lifestyle factors on bladder cancer risk, by calculating HRs and 95% confidence intervals (CI). To evaluate the extent to which adhering to a healthy lifestyle might mitigate the underlying bladder cancer risk owing to genetic susceptibility, we created joint categories of the genetic and lifestyle factors categories, with the high-risk group (i.e., with high genetic risk and poor lifestyle) as the reference group. Cox models were adjusted for age, sex, family history of cancer, socioeconomic status, and the first 10 principal components of ancestry. The interaction between the lifestyle and genetic risk score was tested by including an interaction term in the regression model. We further calculated absolute risk difference over a 10-year period, and the number needed to adopt an optimal lifestyle to prevent one bladder cancer case in 10 years, based on the method described by Altman D.G and Andersen P.K [31]. We also calculated the population-attributable fraction to estimate the proportion of bladder cancer cases that would be prevented if the whole population were in the optimal lifestyle category.

Given that there is considerable gender difference in the incidence of bladder cancer, we repeated all analyses separately by sex. We conducted several sensitivity analyses to test the robustness of the primary results. Firstly, to minimize the possibility of the reverse causality bias resulting from lifestyle changes caused by undiagnosed cancer, we excluded events that occurred within the first 2 years of follow-up. Secondly, we limited our analyses to the participants of white British descent. Thirdly, we used the subdistribution hazard model approach proposed by Fine and Gray to account for the competing risk of death, and diagnosis of any other cancer (except for non-melanoma skin cancer). Fourthly, we created a weighted healthy lifestyle score to control for varied effect estimates of associations between different lifestyle factors and bladder cancer. Fifth, we developed a new lifestyle score without smoking status to identify whether smoking status might drive the associations between lifestyle with bladder cancer.

As shown in the Fig. 1 and Table 2, higher PRS was significantly associated with an increased risk of bladder cancer (HR 1.20, 95% CI 1.13–1.29 per SD increment). Compared with those with low PRS (the lowest quintile of PRS), participants with intermediate (quintiles 2 to 4) and high PRS (the highest quintile) had a higher risk of incident bladder cancer (HR 1.29, 95% CI 1.07–1.56; HR 1.63 95% CI 1.32–2.02, respectively). The association was unchanged after additional adjustment for lifestyle factors, indicating that genetic risk was independently associated with bladder cancer risk regardless of lifestyle factors. The similar pattern of results was noted when regrouping the PRS into tertiles or quartiles (Supporting Information Table S4), or restricting the analysis to participants of white British descent, or in the competing risk analysis using Fine-Gray subdistribution hazard model (Supporting Information Table S5).

We also observed a significant association between lifestyle score and a reduced bladder cancer risk (Fig. 1 & Table 2). Participants with an intermediate and optimal lifestyle had 35% and 51% reductions in risk of bladder cancer compared with those with poor lifestyle (HR 0.65, 95% CI 0.53–0.80; HR 0.49, 95% CI 0.39–0.61, respectively), and the estimated HR was similar after additional adjustment for PRS. Among individual lifestyle components, body weight, smoking and diet were significantly associated with the risk of bladder cancer, whereas physical activity was not associated with the risk (Table 3). And we also found that smoking status contributed most to the risk of incident bladder cancer.

Figure 2 showed a combined effect of genetic variants and lifestyle factors on the risk of bladder cancer. The risk of bladder cancer increased as both genetic risk and poor lifestyle-related risk increased (Supporting Information Figure S2). Participants with a high genetic risk and a poor lifestyle had 3.6-fold elevated risk of bladder cancer compared with those with a low genetic risk and an optimal lifestyle (HR 3.63, 95% CI 2.23 -5.91). We did not find sufficient evidence of a significant multiplicative interaction or an additive interaction between genetic and lifestyle factors on bladder cancer (P = 0.837, Supporting Information Table S6). Additionally, 19.8% of new-onset bladder cancer events during follow-up would be prevented if all individuals adopted an optimal lifestyle (Supporting Information Table S7).

In further analyses stratified by PRS categories with a poor lifestyle as the reference group, we found that an optimal lifestyle was associated with a lower risk of bladder cancer, regardless of PRS (Table 4). Within each genetic risk stratum, those with an intermediate or optimal lifestyle had a nearly 50% reduction in risk of bladder cancer as compared to participants with a poor lifestyle. For example, among participants with high PRS, participants with intermediate or optimal lifestyle had 38% and 49% lower risk of bladder cancer compared with those with poor lifestyle (HR 0.62, 95% CI 0.41–0.94; HR 0.51, 95% CI 0.33–0.80, respectively). Each 18 participants with high genetic risk who converted their lifestyle from poor to optimal may prevent 1 bladder cancer case in 10 years.

The observed associations of PRS and lifestyle with bladder cancer were consistent in men and women, although some observed associations were not significant and unreliable due to the relatively low number of cases (interaction P > 0.10, Supporting Information Figure S3-S4 &Table S8-10). And the joint effect of genetic and lifestyle factors on bladder cancer risk was not modified by smoking and diabetes (Supporting Information Table S11-12). These results remain essentially unchanged in a series of sensitivity analyses by excluding the incident bladder cancer identified during the first 2 years of follow-up, restricting the analysis to participants of white British descent, in the competing risk analysis using Fine-Gray subdistribution hazard model, or using a weighted lifestyle score (Supporting Information Table S13-15). In addition, the pattern of associations based on the healthy lifestyle scores without smoking status were similar (Supporting Information Figure S5). We also conducted an additional analysis to evaluate the joint effect of genetic and lifestyle factors on the early- and late-onset of bladder cancer risk, and found that the joint effect of genetic and lifestyle factors was more evident in the risk of bladder cancer diagnosed in adults < 60 years of age compared to that diagnosed in adults > 65 years (Supporting Information table S16).

The main strength was that this study was based on a well-designed prospective cohort of over 370 000 participants with genetic data and 880 events during up to 14.8 years of follow-up, providing sufficient statistical power to explore the effect of the combination of genetic risk and lifestyle on the bladder cancer in detail. In addition, several robust sensitivity analyses further strengthened the validity of our findings.

Our study also has several limitations. Firstly, some of lifestyle factors were self-reported, which may introduce recall bias and misclassification errors in assessing lifestyle risk levels. However, misclassification errors seemed more likely to drive associations towards the null. Secondly, the lifestyle factors were measured only once at baseline, so we were unable to evaluate the long-term effects of lifestyle behaviors. Thirdly, in the present study, the HLS did not include all components of the WCRF/AICR cancer prevention recommendations because information on the consumption of fast foods and sugar-sweetened drinks was not collected in the UK Biobank at baseline. Fourthly, given that those susceptibility SNPs were identified among people of European descent and our study population restricted to volunteers of European descent, the generalizability of our finding to populations with distinct ancestry should be further examined in future studies. Lastly, due to the observational nature of the study, we cannot establish a causal association between the lifestyle behaviors and bladder cancer risk. Although we have controlled for known potential sources of bias in our analyses, the possibility of unmeasured confounding (e.g., health literacy, socioeconomic status) and reverse bias cannot be fully excluded. However, the findings remained robust after excluding bladder cancer cases occurred within the first 2 years, suggesting that the reverse bias would be likely to be minimal.