Composite Score of Healthy Lifestyle Factors and Risk of Hepatocellular Carcinoma: Findings from a Prospective Cohort Study

This article is featured in Highlights of This Issue, p. 243

Primary liver cancer was ranked the sixth most commonly diagnosed cancer and fourth leading cause of cancer-related deaths worldwide in 2018 (1). Hepatocellular carcinoma (HCC) accounts for 85%–90% of all primary liver cancers (2). The established major risk factors for HCC are chronic infection with hepatitis B virus (HBV) and/or hepatitis C virus (HCV), excessive alcohol use, and dietary exposure to aflatoxin (3, 4). Approximately two-third of HCC cases are attributable to hepatitis B and/or C infection (5). In the United States, the incidence rate of liver cancer has been increasing by 3%–4% per year since the mid-1970s and was three times higher in the 2012–2016 period than in the 1975–1979 period (6). With the implementation of effective universal hepatitis B vaccination program and curative therapy for hepatitis C, it is expected that noninfectious modifiable risk factors, including obesity, alcohol intake, cigarette smoking, unhealthy diet, and/or physical inactivity, will become increasingly important in the development of HCC.

Prior studies have shown that unhealthy lifestyle factors, including high intake of alcohol, higher body mass index (BMI), cigarette smoking, a diet with high red/processed meat and low fiber, and physical inactivity, are individually associated with higher risk of primary liver cancer or HCC (7–9). A recent analysis reported that longer duration of sleep was associated with moderately increased risk of primary liver cancer (10). Longer duration of sleep was also associated with an increased risk of nonalcoholic fatty liver disease (NAFLD), an emerging risk factor for HCC (11, 12). No study has examined the aggregated effect of these modifiable risk factors on the risk of developing HCC. Such data are important to support the development of comprehensive lifestyle modification programs for primary prevention of HCC. Therefore, we evaluated the association between the composite score of healthy lifestyle factors and the risk of HCC incidence in a population-based prospective cohort of more than 60,000 participants with up to 25 years of follow-up.

Data for this analysis were obtained from the Singapore Chinese Health Study, which was described in details elsewhere (13). Briefly, the Singapore Chinese Health Study is an ongoing population-based prospective cohort study that recruited 63,257 Chinese men and women, ages 45–74 years from two main dialect groups of Chinese in Singapore (i.e., Hokkiens and Cantonese) who resided in the government-built housing estates between April 1993 and December 1998. The Hokkiens and Cantonese were originated from the Fujian and Guangdong provinces in southern China, respectively. All study participants provided written informed consent. The Singapore Chinese Health Study is approved by the Institutional Review Boards of the National University of Singapore (Singapore) and the University of Pittsburgh (Pittsburgh, PA).

During the 1994–1999 period of recruitment, blood and urine samples were collected from a 3% random sample of the cohort. Between July 1999 and December 2003, all surviving study participants were reinterviewed by phone to update information on alcohol use, tobacco smoking, medical history, current physical activity, and body weight. Subsequently, they were asked to donate a blood (if declined, a mouthwash sample was collected instead) and urine sample for research use. Of all the subjects that we recontacted successfully, 28,346 subjects (∼57% of eligible participants) provided blood samples.

At baseline, study participants were interviewed at home by trained interviewers who used a structured questionnaire to collect information on demographics, body weight and height, lifetime use of tobacco (cigarettes and water-pipe), current physical activity, occupational exposure, medical history, family history of cancer, menstrual and reproductive history (for women only), and dietary habits during the past 12 months. For cigarette smoking, subjects were first asked their smoking status (i.e., never, current, and former smokers). Former and current smokers were further asked for information on: (i) age of starting and age of quitting smoking, (ii) number of cigarettes per day, and (iii) number of years of smoking (14). For physical activity, the questionnaire listed the following categories: never, 0.5–1, 2–3, 4–6, 7–10, 11–20, 21–30, and 31 hours or more per week, for each of three physical activity categories: (i) strenuous sports (i.e., jogging, bicycling on hills, tennis, squash, swimming laps, or aerobics), (ii) vigorous work (i.e., moving heavy furniture, loading or unloading trucks, shoveling, or equivalent manual labor), and (iii) moderate activities (i.e., brisk walking, bowling, bicycling on level ground, tai chi, and chi kung; ref. 15). Sleep duration was evaluated using the following question: “on average, during the last year, how many hours in a day did you sleep?” with a response choice of six predefined categories: ≤5, 6, 7, 8, 9, and ≥10 hours per day.

Dietary intake was assessed using semiquantitative food frequency questionnaire (FFQ), which had been developed for and validated in the study population against a series of 24-hour dietary recall (24-HDR) interviews (16) and selected biomarkers studies (17, 18). The dietary questionnaire comprised 165 commonly consumed food and beverage items by Chinese Singaporeans. Study participants were asked how frequently they consumed the food or food groups in eight predefined categories, from “never or hardly ever” to “two or more times a day,” followed by predetermined portion sizes assisted with the illustration of food photos in an album. For alcohol consumption, each study participant was asked about his/her drinking frequency during the past year of four types of alcoholic beverages, beer, wine, Western hard liquor, and Chinese hard liquor, with a response choice of eight predefined categories for each type of beverage: never or hardly, once a month, 2–3 times a month, once a week, 2–3 times a week, 4–6 times a week, once a day, and ≥2 times a day. The portion size for each type of alcoholic beverage was defined as follows: (i) beer: ≤1 small bottle (375 mL), two small bottles or one large bottle (750 mL), and ≥3 large bottles; (ii) wine: ≤1 glass (118 mL), 2, 3, and ≥4 glasses; (iii) Western or Chinese hard liquor: ≤1 shot (30 mL), 2, 3, and ≥4 shots. Between April 1994 and March 1997, the FFQ was validated against two 24-HDRs, one on a weekday and the other on a weekend that was approximately 2 months apart, among a random sample of 810 participants of the Singapore Chinese Health Study. The correlation coefficients between the FFQ and 24-HDR for the majority of calorie-adjusted nutrients ranged from 0.24 to 0.79 (16).

The classification of the composite score of healthy lifestyle factors was based on their independent association with liver cancer or HCC (7–9, 19, 20). The lifestyle factors were alcohol consumption, cigarette smoking, dietary intake, and sleep duration, in addition to BMI. Alcohol consumption was categorized by the number of alcoholic drinks: one drink was defined as 375 mL of beer (13.6 grams of ethanol), 118 mL of wine (11.7 grams of ethanol), and 30 mL of Western or Chinese hard liquor (10.9 grams of ethanol; ref. 21). BMI was calculated as the current weight in kilograms divided by height in meters squared (kg/m²).

The previous studies have found an inverse association between dietary habits of Mediterranean populations and the risk of chronic diseases (22, 23). It originally included nine items (24). In our analysis, we included eight components: (i) vegetables, (ii) fruit and nuts, (iii) cereals, (iv) dairy, (v) legumes, (vi) fish, (vii) the ratio of monounsaturated fatty acid over saturated fatty acids, and (viii) meat and meat products after excluding alcohol (an independent risk factor; ref. 9). Each component was assigned a score of 0 or 1 according to the consumption level (below or above the study population–specific median) and healthy/unhealthy status of the component. The summed score of alternative Mediterranean diet (aMED) ranged from 0–8.

For BMI, we used the cutoff of 23 kg/m², following both World Health Organization and the Asia-Pacific recommendations (25, 26). Individuals with BMI < 23 kg/m² were assigned a lifestyle score of one, otherwise a score of zero. For alcohol consumption, zero was assigned to heavy drinker, defined as men who drank ≥15 drinks per week and women who drank ≥8 drinks per week, and one otherwise (i.e., to both nondrinker and nonheavy drinker group). For cigarette smoking, zero was assigned to heavy smokers (defined as those who started smoking before 15 years of age and smoked 13 or more cigarettes per day), one to the remaining light smokers, and two to never-smokers (27). For aMED, we divided all study participants into three groups on the basis of their aMED scores: zero was assigned to the lowest quartile of aMED score, one to quartiles 2 and 3, and two to quartile 4. For sleep duration, one was assigned to those with 6–8 hours of sleep per day and zero otherwise. The sum of the five individual scores constituted the overall combined lifestyle index score, ranging from 0 to 8, of which higher scores represented healthier lifestyle.

Incident HCC cases and deaths were identified by annual linkage analysis of all surviving cohort participants with the national databases of the Singapore Cancer Registry and the Singapore Birth and Death Registry, respectively. The International Classification of Diseases-Oncology 2nd Edition Codes C22.0 were used to determine HCC cases (28). To date, only 56 (<0.1%) of the entire cohort participants are known to be lost to follow-up due to migration out of Singapore. Thus, the ascertainment of incident cancer cases and deaths among the cohort participants was virtually complete. As of December 2015, with an average 17.7 years of follow-up, 561 study participants developed HCC among 61,321 participants who were free of cancer at enrollment after excluding 1,936 participants with a history of cancer at baseline.

We also constructed a case–control study of HCC within the Singapore Chinese Health Study. All 197 incident HCC cases diagnosed before 31 December 2015 who donated a baseline blood sample were eligible for this study. For each case, we randomly selected 2–3 control subjects who donated a baseline blood sample and were alive and free of cancer at the date of diagnosis of the index case. Controls were individually matched to the index case by age at enrollment (±2 years old), gender, dialect groups (i.e., Hokkien and Cantonese), and date of baseline blood collection (± 6 months).

Serum samples of all subjects included in the case–control study of HCC were tested for serologic markers of HBV and HCV exposures using standard assays as described previously (9, 20, 29). The case/control status of the test samples was blinded to laboratory personnel. Briefly, the presence of hepatitis B surface antigen (HBsAg) was determined on the first 302 samples (80 cases vs. 222 controls) using a commercially available test kit (AUSRIA, Abbott Laboratories). Negative samples were further tested for the presence of antibodies to hepatitis core antigen (anti-HBc) and antibodies to hepatitis surface antigen (anti-HBs) using standardized test kit (Corab and Ausab, Abbott Laboratories). For the remaining 360 serum samples (117 cases vs. 243 controls), only the HBsAg status was determined using the same assay. We also tested for the presence of antibodies to the HCV (anti-HCV) on the first 302 samples (80 cases and 222 controls) using ELISA (version 2.0) Test Kit (Ortho Diagnostic System) and confirmation of positive samples was performed using RIBA (version 2.0) Test Kit (Chiron). Given the relatively low prevalence of anti-HCV in our study population (1.4%), we stopped testing anti-HCV for the remaining HCC cases and controls.

Means and SDs were calculated for continuous variables, while counts and proportions were computed for categorical variables. The t test and χ² test were used to compare the distributions of continuous and categorical variables, respectively, between cases and noncases. ANOVA was used for the comparisons across different groups of the composite score of healthy lifestyle factors. Person-years at risk for each participant was calculated from the date of interview at entry into the cohort to the date of HCC diagnosis, death, migration out of Singapore, or 31 December 2015, whichever occurred first.

The Cox proportional hazard regression method was used to evaluate the association between the individual and composite score of healthy lifestyle factors and the risk of developing HCC. HRs and their corresponding 95% confidence intervals (CI) of HCC associated with higher individual and composite score of healthy lifestyle factors were derived from the Cox proportional hazard regression models. Proportional hazard assumption was tested using Schoenfeld residuals test, and no violation was found. An ordinal variable was used for the linear trend test for the index score of combined lifestyle factors and HCC risk. Potential confounders included in the multivariable Cox proportional hazards models were age, gender, dialect group (i.e., Hokkien or Cantonese), level of education (i.e., no formal education, primary school, secondary, or higher education), year of enrollment (i.e., 1993–1995 and 1996–1998), history of type 2 diabetes (yes vs. no), and total energy intake (Kcal/day). We also performed stratified analysis by gender and sensitivity analysis by excluding HCC cases and person-years observed within the first 2 years after the baseline interview. Additional sensitivity analyses were also performed for composite score of healthy lifestyle factors without sleep and aMED variables.

We used conditional logistic regression method in analysis for the entire dataset of the nested case–control study to evaluate the association between composite score of healthy lifestyle factors and the risk of HCC, with additional adjustment for HBsAg and anti-HCV serologic statuses. For those with missing anti-HCV (117 cases and 243 controls), a separate indicator variable was created and included in the logistic models. In the stratified analysis by HBsAg and anti-HCV statuses, unconditional logistic regression method was used with the inclusion of all matching factors (i.e., age, sex, dialect group, and year of enrollment) because the matched pairs were broken.

All statistical analyses were conducted using the computing software SAS version 9.4 (SAS Institute Inc.). All P values were two-sided and 0.05 was used as a threshold of statistical significance.

The mean (± SD) age at cancer diagnosis of the 561 HCC cases included in this study was 70.8 (± 8.4) years. The median time interval from the date of baseline interview to the diagnosis of HCC was 11.4 years (range, 0.03–22.2 years).

We observed lower proportions of secondary school or higher education and aMED score in quartile 4 among cases than noncases. However, there was no significant difference in sleep duration between HCC cases and noncases. Compared with those in the lowest composite score, those in the highest score were on average approximately 3 years younger at recruitment, and more likely to be women, to attain higher level of education, or to engage in physical exercise, but less likely to be Hokkien ethnic group and to have a history of diabetes (all P < 0.0001; Table 1). Among HCC cases, there were higher proportions of men, Hokkien ethnicity, BMI ≥ 23 kg/m², cigarette smokers, alcohol drinkers with ≥7 drinks per week, aMED score in quartile 1, or those with a history of type 2 diabetes compared with non-HCC individuals (Supplementary Table S1).

Significantly lower HR of HCC was associated with BMI < 23 kg/m², never smoking, nonheavy drinking, or highest 25th percentile of aMED score. We combined light drinkers (≤14 drinks/week for men and ≤7 drinks/week for women) with nondrinkers due to their similar risk for HCC. Compared with heavy drinkers, HRs (95% CIs) for HCC in light drinkers and nondrinkers were 0.37 (0.24–0.57) and 0.42 (0.28–0.63), respectively. A daily sleep of 6–8 hours was associated with a statistically nonsignificant 10% decrease in risk of HCC (Table 2).

Higher composite score of the five healthy lifestyle factors was significantly associated with lower risk of HCC in a dose-dependent manner (P_trend < 0.0001). Compared with the score of 0–4, the HRs (95% CIs) of HCC for those with the scores of 5, 6, 7, and 8 were 0.67 (0.52–0.85), 0.61 (0.48–0.77), 0.49 (0.37–0.65), and 0.13 (0.06–0.30), respectively, after adjustment for age, gender, history of diabetes, daily energy intake, and other potential confounders. The increment per composite score was associated with a statistically significant 22% reduction in risk of HCC. This inverse association was present consistently and comparable in both men and women (Table 3).

We determined HBsAg status for 197 HCC cases and 465 matched controls. In addition, 80 HCC cases and 222 controls also were tested for anti-HCV status. Among tested HCC cases, 41.3% were positive for HBsAg and 5% were positive for anti-HCV. The corresponding figures in controls were 3.2% and 1.4%.

In the case–control analysis, with additional adjustment for chronic infection with HBV and/or HCV, we found a similar inverse association between the composite score of healthy lifestyle factors and HCC risk overall (Table 4). Among HBsAg-negative and anti-HCV–negative/unknown subjects, HR of HCC decreased by more than 60% for those with the highest composite score compared with the lowest score category of healthy lifestyle factors (OR, 0.38; 95% CI, 0.19–0.79; P_trend = 0.001). Given the small sample size for HBsAg- or anti-HCV–positive individuals, the risk estimates for HCC associated with composite score of healthy lifestyle factors varied greatly and were not informative (Table 4).

We conducted several sensitivity analyses. First, we excluded HCC cases and person-years within the first 2 years of observation and found that the results were similar as those based on the entire cohort. Compared with the composite score of 0–4, the multivariable-adjusted HRs (95% CIs) of HCC for the composite scores of 5, 6, 7, and 8 were 0.66 (0.51–0.85), 0.62 (0.48–0.79), 0.50 (0.37–0.62), and 0.14 (0.06–0.32), respectively (P_trend < 0.0001; Supplementary Table S2). Then, we excluded sleep duration from the composite score because it alone was not statistically significantly associated with HCC risk (Table 2). The association between the composite score without sleep duration and HCC risk weakened, but remained statistically significant; HRs (95% CIs) for new scores 5, 6, and 7 were 0.65 (0.53–0.80), 0.57 (0.45–0.71), and 0.28 (0.17–0.43), respectively, compared with new score 0–4 (P_trend < 0.0001). We also conducted similar analysis by removing aMED from the composite score and yielded similar inverse, but weakened association; HRs (95% CIs) for scores 5 and 6 were 0.62 (0.51–0.76) and 0.46 (0.36–0.58), respectively, compared with score 0–4 (P_trend < 0.0001).

Stratified analyses were performed to assess the heterogeneity in the composite score–HCC risk association by selected factors, including diabetes status and BMI. The association between the composite score and HCC was more apparent in participants without history of diabetes (P_trend < 0.0001) than in those with history of diabetes (P_trend = 0.0005); however, the difference between the two was not statistically significant (P_{heterogeneity} = 0.278). The association was more apparent in participants with BMI < 23 kg/m² (P_trend < 0.0001) than in those with BMI ≥ 23 kg/m² (P_trend = 0.06), with a statistical significance (P_{heterogeneity} = 0.004) in heterogeneity between the two subgroups (Supplementary Table S3).

In this analysis, in a population-based prospective cohort study of more than 60,000 individuals with an average follow-up of 17.7 years, we found that higher composite score of healthy lifestyle factors, including BMI, smoking, alcohol consumption, Mediterranean-style diet, and sleep duration, was associated with decreased risk of HCC in a dose-dependent manner. This inverse relationship was robust and present in both men and women and in those with at least 2 or more years of follow-up. The observed inverse association was confirmed among individuals with negative HBsAg and negative anti-HCV, suggesting that the effect of these lifestyle factors on HCC risk was independent of chronic infection with HBV and HCV, major risk factors for HCC.

This study showed lower risk of HCC for individuals who adhered to a healthier lifestyle, including adhering to heathier diet based on Mediterranean dietary habits (i.e., eating more vegetables, fruit, and nuts/legumes, and fish and fatty foods), maintaining a normal range of BMI, moderate or no alcohol consumption, never smoking cigarettes, and regular sleep duration of 6–8 hours a day. Compared with the lowest score, individuals with the highest composite score of healthy lifestyle factors had reduced by 85% in the risk of developing HCC overall, and by 90% in those without chronic infection with HBV.

Previously, we reported an inverse association between a similarly constructed composite score of healthy lifestyle factors and risk of colon and gastric cancers (30, 31). Accordingly, Odegaard and colleagues (30) found that this same composite score was significantly associated with lower risk of colon cancer, but not with risk of rectal cancer. A similar inverse association was observed for gastric cancer in our study population (31).

Several prior studies provide supportive evidence for an inverse association between the individual healthy lifestyle factors and HCC risk. A meta-analysis including 26 prospective cohort studies with a total of 9 million study participants, including 16 Asian studies, four U.S. studies, and six European studies, showed that both overweight and obesity were associated with higher risk of primary liver cancer (32). Recently, an analysis on registry data of more than 1.2 million Swedish men enlisted for conscription between 1969 and 1996 reported that overweight and obese men were at higher risk of severe liver disease (i.e., end-stage liver disease, HCC, or death from any liver disease) or gallbladder or bile duct carcinoma (33). Obesity is an underlying cause for the development and progression of NAFLD, which includes nonalcoholic steatohepatitis and cirrhosis, a precondition for HCC (4).

Alcohol abuse is an important risk factor for HCC. A recent meta-analysis of 19 prospective cohort studies from Asia (15 cohorts), Europe (two cohorts), and the United States (two cohorts) showed that the pooled RRs were 0.91 (95% CI, 0.81–1.02) for moderate drinking (1–2 drinks/day) and 1.16 (95% CI, 1.01–1.34) for heavy drinking (≥3 drinks/day) compared with nondrinking (34). Excessive alcohol consumption, potentially via carcinogenicity of acetaldehyde, the first metabolite of alcohol (35), is associated with alcoholic cirrhosis, which increases the risk of HCC (36). However, the association between moderate drinking and HCC risk is inconsistent. Our study showed that non- or moderate drinkers (≤14 drinks/week for male or ≤7 drinks/week for female) had a statistically significant 59% lower risk of HCC than heavy drinkers. Among non-Asians in Los Angeles, we have previously reported that there was no increased risk of HCC for consumption of 1–4 drinks per day (37). There may be different alcohol threshold for risk of HCC among different race/ethnic groups, especially at the lower end of the alcohol consumption level with the risk of HCC in Asian compared with non-Asian populations. More studies are warranted to clarify this observation.

Cigarette smoking has also been reported as an independent risk factor for HCC. In a meta-analysis of 38 cohort studies and 58 case–control studies, Lee and colleagues (38) reported that the adjusted pooled RR estimates were 1.51 (95% CI, 1.37–1.67) for current smokers and 1.12 (95% CI, 0.78–1.60) for former smokers compared with never smokers. Our results are consistent with these previous studies (38). Tobacco smoke contains a myriad of chemical compounds that are carcinogenic or toxic to the liver (39). For example, N-nitrosodimethylamine, which is abundant in cigarette smoke, can induce liver tumor in rodents (40).

aMED was found to be associated with lower risk of HCC across different populations. In the United States, four prospective cohort studies (41–43), all reported an inverse association between aMED and risk of HCC, with a 25%–39% lower risk of HCC with highest adherence to aMED. These results are consistent with our findings (i.e., 22% risk reduction of HCC for the highest quartile compared with the lowest quartile), as shown in Table 2.

Recent studies have demonstrated an association between sleep duration and the risk of liver cancer (10) or NAFLD (11, 12). Mechanistically, the circadian clock is believed to play a role in mitochondrial dysfunction (44) and in the regulation of hepatic triglyceride accumulation, oxidative stress, and inflammation (45); all of which may contribute to the development and progression of NAFLD. In wild-type mice, chronic circadian misalignment led to the disruption of circadian clock and induced spontaneous NAFLD-related HCC independent of dietary exposure, exogenous genotoxic stress, and/or germline gene mutation (46).

Strengths of this study include prospective study design, a comprehensive questionnaire collecting multiple lifestyle factors and detailed dietary information before the HCC occurrence, large sample size allowing for robust estimates of HCC risk, use of a composite score of healthy lifestyle factors that comprehensively characterized an individual person's profile, long-term and nearly complete follow-up that reduced potential bias due to loss of follow-up or the reverse impact of disease on the exposure–HCC risk association, and known HBsAg and anti-HCV serology on a sizable subset of HCC cases and the controls presenting the entire cohort. This robust study design allowed us to examine these lifestyle factors on HCC risk among HBsAg-negative and anti-HCV–positive individuals only, which ruled out the potential confounding effect of both HBV and HCV infections on the observed association.

One potential limitation of our study was the misclassification of lifestyle factors because their assessment at baseline might not reflect the level of long-term exposure due to the change of lifestyle after baseline assessment. As in any observational study, such misclassification is usually nondifferential to both HCC cases and non-HCC participants. Nondifferential misclassification usually results in the underestimation of HRs toward null. Thus, the significant association between composite score of healthy lifestyle factors and HCC risk observed in this study may be underestimated, but the conclusion is valid.

In summary, this study shows a strong, dose-dependent association between the composite score of healthy lifestyle factors and lower risk of HCC development. The overall results, especially among HBsAg-negative and anti-HCV–negative individuals, support the development of a prevention strategy including multiple lifestyle factors that would result in significant reduction in HCC incidence in general populations. This finding highlights the importance of promoting healthy living for primary prevention of HCC, even in a population with a relatively high prevalence of HBV infection and high background risk of HCC. Our findings are pertinent to the populations in the United States and other populations where nonviral factors are projected to have increasingly greater impact on HCC incidence and mortality.

J.-M. Yuan reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

H.N. Luu: Conceptualization, formal analysis, supervision, validation, investigation, methodology, writing–original draft, writing–review and editing. J. Behari: Formal analysis, investigation, methodology, writing–review and editing. G.B.-B. Goh: Investigation, methodology, writing–review and editing. R. Wang: Data curation, software, formal analysis, validation, writing–review and editing. A. Jin: Data curation, software, formal analysis, validation. C.E. Thomas: Software, formal analysis, writing–review and editing. J.C. Clemente: Investigation, methodology, writing–review and editing. A.O. Odegaard: Validation, methodology, writing–review and editing. W.-P. Koh: Conceptualization, resources, formal analysis, supervision, validation, methodology, writing–review and editing. J.-M. Yuan: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.

The authors thank Siew-Hong Low of the National University of Singapore for supervising the field work of the Singapore Chinese Health Study. They also thank the Singapore Cancer Registry for the identification of incident cancer cases among participants of the Singapore Chinese Health Study. The Singapore Chinese Health Study was supported by the NIH (grant nos. R01 CA144034 and UM1 CA182876). W.-P. Koh was supported by the National Medical Research Council, Singapore (NMRC/CSA/0055/2013). H.N. Luu was partially supported by the University of Pittsburgh Medical Center Hillman Cancer Center start-up grant.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.