Alcohol consumption and risks of more than 200 diseases in Chinese men

Study population

Details of the CKB study design and methods have been previously reported⁵⁵. Briefly, 512,724 adults aged 30–79 years were recruited from ten geographically diverse (five rural and five urban) areas across China during 2004–2008. At local study assessment clinics, trained health workers administered a laptop-based questionnaire recording sociodemographic factors, lifestyle (for example alcohol drinking, smoking, diet and physical activity) and medical history; undertook physical measurements (for example blood pressure and anthropometry); and collected a blood sample for long-term storage. Two resurveys of ~5% randomly selected surviving participants were subsequently conducted in 2008 and 2013–2014 using similar procedures.

Ethics approval

Ethical approval was obtained from the Ethical Review Committee of the Chinese Centre for Disease Control and Prevention (Beijing, China, 005/2004) and the Oxford Tropical Research Ethics Committee, University of Oxford (UK, 025-04). All participants provided written informed consent.

Assessment of alcohol consumption

Detailed questionnaire assessment of alcohol consumption has been described previously^3,17,56. In the baseline questionnaire, participants were asked how often they had drunk alcohol during the past 12 months (never or almost never, occasionally, only at certain seasons, every month but less than weekly or usually at least once a week). Those who had not drunk alcohol at least weekly in the past 12 months were asked whether there was a period of at least a year before that when they had drunk some alcohol at least once a week. Based on their past and current drinking history, participants were classified into: non-drinkers (had never drunk alcohol in the past year and had not drunk in most weeks in the past); ex-drinkers (had not drunk alcohol in most weeks in the past year but had done so in the past); occasional drinkers (had drunk alcohol but less than weekly in the past year and had not drunk alcohol in most weeks in the past); and current drinkers (had drunk alcohol on a weekly basis (regularly) in the past year).

Current drinkers were asked further questions about their drinking patterns, including frequency, beverage type (beer, grape wine, rice wine, weak spirits with <40% alcohol content and strong spirits with ≥40% alcohol content) and amount consumed on a typical drinking day, mealtime drinking habits, age started drinking in most week and their experience of flushing or dizziness after drinking.

Alcohol intake level was estimated based on the reported frequency (taken as the median of the reported frequency intervals; 1.5 for 1–2 d per week, 4 for 3–5 d per week, 6.5 for 6–7 d per week), beverage type and amount consumed, assuming the following alcohol content by volume (v/v) typically seen in China: beer 4%, grape wine 12%, rice wine 15%, weak spirits 38% and strong spirits 53%⁵⁷. Among current drinkers, men were grouped into four consumption categories (<140, 140–279, 280–419 and 420+ g per week) and women into three categories (<70, 70–139 and 140+ g per week), broadly based on the recommended cutoffs for alcohol categories by the WHO⁵⁸ and national drinking guidelines. Heavy episodic drinking was defined as consuming >60 g of alcohol on a typical drinking occasion for men and >40 g per occasion for women⁵⁸. Drinking outside of meals was defined as usually drinking between or after meals or having no regular patterns (versus usually drinking with meals). Duration of drinking was derived by the difference in years between age at baseline and age started drinking.

Ex-drinkers were asked how long (in years) ago they had stopped drinking in most weeks. Ex-drinkers were grouped with current drinkers as ‘ever-regular drinkers’.

Follow-up for mortality and morbidity

The vital status of participants was obtained periodically from local death registries, supplemented by annual active confirmation through local residential, health insurance and administrative records. Additional information on morbidity was collected through linkage with disease registries (for cancer, stroke, IHD and diabetes) and the national health insurance system, which record any episodes of hospitalization and almost has universal coverage. All events were coded with ICD-10 codes, blinded to the baseline information. By 1 January 2019, 56,550 (11%) participants had died, 311,338 (61%) were ever hospitalized, but only 4,028 (<1%) were lost to follow-up.

Outcome measures

To enable a ‘phenome-wide’ investigation, all recorded diseases and injuries (referred to as ‘diseases’ for simplicity) coded by three-character ICD-10 codes were reviewed. ICD-10 codes were combined (where appropriate) based on disease characteristics and their potential relationships with alcohol consumption^4,8,10,59. Disease end points were curated based on diseases considered to be causally impacted by alcohol by the WHO^4,59 and major diseases previously shown to be related to alcohol in CKB and other large prospective cohort studies^8,10, while retaining maximal granularity. Diseases with at least 80 cases recorded during follow-up among current drinkers, separately by sex, were analyzed individually to capture a wide range of specific conditions while ensuring reasonable statistical power (around 60–80% power to detect a HR of 2.00 per 280 g per week higher usual alcohol intake at P < 0.01 and P < 0.05, respectively). Within each ICD-10 chapter, diseases with <80 events were grouped into a ‘less-common’ category. Several ICD-10 chapters considered not directly relevant in this population (for example perinatal-origin diseases (chapter XVI) and congenital conditions (XVII); pregnancy-related diseases (XV) in men) were excluded.

Major diseases defined by the WHO as likely to be causally related with alcohol consumption⁴, including several cancers (mouth and throat, esophagus, colon-rectum, liver and female breast), diabetes mellitus, IHD, stroke, liver cirrhosis and external causes, were also selected a priori for detailed analyses of associations with drinking patterns (daily drinking, heavy episodic drinking, mealtime habit, spirit drinking and drinking duration). Similarly, diseases that were significantly and adversely associated with alcohol in the ‘phenome-wide’ investigations (either with ever-regular versus occasional drinking or in dose–response associations with amounts consumed) were further categorized as ‘CKB WHO alcohol-related diseases’ and ‘CKB new alcohol-associated diseases’ respectively for genetic investigation of causality. Detailed outcome classifications are reported in Supplementary Table 7.

Genotyping and alcohol genetic instruments

The two East Asian genetic variants (ALDH2–rs671 and ADH1B–rs1229984) were genotyped in 168,050 participants (151,347 randomly selected, 16,703 selected as part of nested case–control studies of CVD and chronic obstructive pulmonary disease, which were only included in analyses of relevant outcomes; Supplementary Fig. 1) using Affymetrix Axiom (n = 100,396) or custom Illumina GoldenGate (n = 93,125) arrays at BGI (Shenzhen, China), with some overlap between them. Among 25,471 participants genotyped with both arrays, the concordance was >99.9% for both variants. Where discordant, genotypes obtained from the Affymetrix Axiom array were used.

The genetic instrument for alcohol was derived from ALDH2–rs671 and ADH1B–rs1229984 and ten study areas from the random genotyped subset of male participants to avoid potential selection bias, using a previously developed method in CKB¹⁹. Briefly, nine genotype combinations were defined based on the genotypes for each of the two variants (each AA, AG or GG). As alcohol use varies greatly by study area, among men, mean alcohol intake was calculated for each of these nine genotype across ten study areas (that is a total of 90 genotype-area combinations) to reflect a wide range of alcohol consumption, assigning an intake of 5 g per week to occasional drinkers and excluding ex-drinkers from the calculation. Ex-drinkers were excluded from the calculation of mean alcohol intake as their baseline intake did not reflect their long-term intake; nevertheless, they were included in subsequent genetic analyses once they had been assigned a genetic group. These 90 combinations were then grouped into six categories (C1–C6) according to their corresponding mean intake values, at cutoff points of 10, 25, 50, 100 and 150 g per week, selected to facilitate investigation of the causal effects of alcohol across a wide range of mean alcohol intakes while allowing adequate sample size in each category for reliable comparisons. In this way participants (including ex-drinkers) were classified only based on their genotypes and study area, but not on individual self-reported drinking patterns. Comparisons of these six genetic categories can, where analyses are stratified by area, be used to estimate the genotypic effects on disease risks.

To facilitate the comparison of genotypic effects between sexes (pleiotropic effects), women were classified into the same six categories as men based on their genotypes and study area, regardless of female alcohol intake. This allowed comparison of genotypic effects between men (where genotype were strongly associated with alcohol intake) and women (where alcohol intake was low in all genotypic categories) (Supplementary Tables 10 and 11).

Statistical analysis

Given the extremely low alcohol use among women^3,17, the analyses were conducted separately by sex but focused chiefly on men. All CKB participants and the genotyped subset with genomic principal components (PCs; derived from genome-wide genotyping array data and were informative for CKB population structure)⁶⁰ were included in conventional and genetic analyses, respectively (Supplementary Fig. 1). Means and percentages of baseline characteristics were calculated by self-reported alcohol consumption patterns and by genotype categories, adjusted for age (in 10-year intervals), ten study areas and (for genetic analysis) genomic PCs⁶⁰ to control for differences in genetic distribution due to population stratification, as appropriate.

For conventional observational analyses, Cox proportional hazard models were used to estimate HRs for individual diseases associated with different alcohol consumption categories (in three broad categories: occasional drinkers, ever-regular drinkers, non-drinkers; and in 6–7 detailed categories: occasional drinkers, ex-drinkers, non-drinkers, 3–4 further current drinker groups defined by alcohol intake level) and among current drinkers with continuous levels of alcohol intake (per 280 g per week in men, per 100 g per week in women) or with categories of alcohol intake (<140, 140–279, 280–419 and 420+ g per week in men; <70, 70–139 and 140+ g per week in women). The Cox models were stratified by age at risk (5-year groups between 35–84 years) and ten areas and adjusted for education (four groups: no formal school, primary school, middle or high school and technical school/college or above) and smoking status (six groups in men: never, occasional, ex-regular, current <15, current 15–24, current ≥25 cigarettes equivalent per day; four groups in women: never, occasional, ex-regular and current). Smoking data have been previously validated against exhaled carbon monoxide⁶¹. Competing risks from all-cause mortality for disease events were handled by censoring participants at death from any cause to estimate cause-specific HRs comparing event rates in participants who were alive and free of the disease of interest⁶². To reduce biases from residual confounding and uncontrolled reverse causation related to the choice of using non-drinkers (for example sick-quitter effect, pre-existing poor health or social disadvantages leading to alcohol cessation or abstinence) as the reference group^3,29, we used occasional drinkers as the reference group, together with separate dose–response analyses among current drinkers. To account for within-person variation of alcohol intake over the follow-up period, repeat alcohol measures for participants who attended the two resurveys were used to estimate usual alcohol intake (Supplementary Table 1) and correct for regression dilution bias^9,63. The shapes of dose–response associations between alcohol and disease risks were assessed among current drinkers by plotting the HRs of predefined baseline consumption categories against the corresponding mean usual alcohol intake. Log HR estimates and the corresponding standard errors for baseline alcohol intake, modeled as a continuous variable, were divided by the regression dilution ratio (0.53 for both men and women; calculated using the McMahon–Peto method⁶⁴) to obtain estimated HRs per 280 g per week higher usual alcohol intake among male current drinkers and HRs per 100 g per week among female current drinkers. For analyses involving drinking patterns, additional adjustments were conducted for total alcohol intake (continuous) and baseline age (continuous; for drinking duration analysis) where appropriate.

Sensitivity analyses were performed by (1) additional adjustments for further covariates (household income (<10,000, 10,000–19,999, 20,000–34,999 and ≥35,000 yuan per year), fresh fruit intake (4–7 d per week and ≤3 d per week), physical activity (continuous, in metabolic equivalent of task per hour per day), body mass index (<22, 22–24.9, 25–26.9 and ≥27 kg m²); and (2) excluding individuals with poor self-reported health or previous major chronic diseases (including self-reported coronary heart diseases, stroke, transient ischemic attack, tuberculosis, emphysema or bronchitis, liver cirrhosis or chronic hepatitis, peptic ulcer, gallstone or gallbladder disease, kidney disease, rheumatoid arthritis, cancer and diabetes) at baseline. For all aggregated end points (for example CKB WHO alcohol-related, CKB new alcohol-associated and all morbidity), subgroup analyses were conducted by baseline age (<55, 55–64 and ≥65 years), area (urban and rural), education (primary school or below, middle school, high school or above), household income (<10,000, 10,000–19,999 and 20,000+ yuan per year) and smoking status (ever-regular and never-regular), with heterogeneity or trend assessed by chi-squared tests⁶⁵. HRs for diseases associated with years of stopping among ex-drinkers compared to occasional drinkers were also estimated.

In genetic analyses, Cox regression, stratified by age at risk and study area and adjusted for 11 genomic PCs⁶⁰, were used to estimate HRs for major alcohol-related diseases associated with the six genetic categories (C1–C6). Log HRs were plotted against the genotype-predicted mean male alcohol intake in the six categories. To control for potential confounding by population structure, similar analyses were repeated within each study area using age-at-risk-stratified and genomic PC-adjusted Cox models. A line of best fit was fitted through the log HRs against genotype-predicted mean male alcohol intake in the genetic categories present in the corresponding study area, using meta-regression. These within-area slopes (each reflecting purely genotypic effects) were combined by inverse-variance-weighted meta-analysis to yield the overall area-stratified genotypic associations, which controlled for any potential bias resulted from variations due to population structure, summarized as HR per 280 g per week higher genotype-predicted mean male alcohol intake. For total morbidity and aggregated alcohol-associated outcomes, sensitivity analyses were performed by (1) using age-at-risk- and area-stratified and genomic PC-adjusted Cox models to estimate HR per 280 g per week (area-adjusted genotypic associations); and (2) using a two-stage least-squares approach⁶⁶.

Genotypic analyses in women were conducted not to assess the health effects of alcohol in women, but to investigate the extent to which the genotypes studied in men had pleiotropic effects (genotypic effects not mediated by drinking patterns). As few women consumed alcohol, any genotypic effects of the six genetic categories that are mediated by drinking alcohol should be much smaller in women than in men, but any other pleiotropic genotypic effects should be similar in both sexes. Hence, among women, we used the same genetic categories as in men and related the genotypic effects in women to the mean male alcohol intake in these six categories, which allows comparisons of genetic findings by sex and assessment of potential pleiotropy. To further remove the small genotypic effects on alcohol use in women (Supplementary Tables 10 and 11), we restricted the genetic analyses to female non-drinkers in sensitivity analyses.

The genotypic associations of individual genetic variants (rs671, rs1229984; GG versus AG genotype) with alcohol-related disease risks were also assessed using a similar area-stratified approach.

The proportional hazards assumption was tested using scaled Schoenfeld residuals for the pre-specified major diseases (no clear evidence of violation was found). For analyses involving more than two exposure categories, the floating absolute risks were used to estimate group-specific 95% CIs for all categories including the reference group^9,19,67. All P values were two-sided. Statistical significance (at the 5% level) was evaluated using both FDR-adjusted P values applied within ICD-10 chapters to correct for multiple testing in the ‘phenome-wide’ investigation^68,69,70 and conventional P values for hypothesis testing for observational analyses of WHO alcohol-related diseases, analyses of drinking patterns and genetic analyses.

To assess the cumulative burden of alcohol consumption, the total number of hospitalizations were estimated for ever-regular versus occasional drinkers using the mean cumulative count, which does not assume independence between hospitalizations and all-cause mortality^71,72,73. All analyses used R software (v.4.0.5).

Ethics and inclusion statement

In accordance with the Nature Portfolio journals’ editorial policies, the research has included local researchers from China throughout the research process, including study design, study implementation, data ownership and authorship. The roles and responsibilities were agreed among collaborators ahead of the research and capacity-building plans, including data collection and study implementation skills for local researchers, were discussed and delivered. This research is locally relevant to the studied country and included local collaborative partners in all aspects of the study, thus, will provide local and regional organizations with epidemiological evidence on the health impacts of alcohol consumption to inform public health policies.

This research was not restricted nor prohibited in the setting of the researchers. The study was approved by local ethics review committee. The research raised no risks related to stigmatization, incrimination, discrimination, animal welfare, the environment, health, safety, security or other personal or biorisks. No biological materials, cultural artifacts or associated traditional knowledge has been transferred out of the country. In preparing the manuscript, the authors have reviewed and cited local and regional relevant studies.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.