All methods were carried out in accordance with relevant guidelines and regulations. We conducted a comparative risk assessment analysis, using as comparator a hypothetical scenario in which nobody would drink alcohol, to estimate the number of cancer deaths and DALYs that would be prevented.
Calculation of population attributable fraction
We calculated the population attributable fraction (PAF) of alcohol using the formula below [24].
$$PAF=\frac{\sum_{i=1 }^{n}{P}_{i}\left({RR}_{i}-1\right)}{\sum_{i=1}^{n}{P}_{i}\left({RR}_{i}-1\right)+1}$$
where n is the number of exposure levels; in our case three (light, moderate, and heavy drinking), P is the proportion of the population at level i of exposure and RRi is the relative risk of cancer at level i of exposure.
The PAF is the proportional reduction in disease or mortality that would occur if exposure to alcohol would be reduced to zero (i.e., no alcohol consumption), and thus is the proportion of disease/mortality that can be attributed to alcohol consumption. The main inputs to this formula are: 1) the etiological association (relative risk) between alcohol and site-specific cancer and 2) the current prevalence of alcohol consumption in Argentina.
Risk estimates alcohol and site-specific cancer
We derived relative risks (RR) for different levels of alcohol consumption with site-specific cancer risk from the most comprehensive high-quality meta-analysis to date, conducted by Bagnardi et al. [17]. In this meta-analysis, sex-specific relative risks were reported for various alcohol consumption levels: non-drinkers, ≤ 12.5 g/day, 12.6—50 g/day, and > 50 g/day, in which the unit represents grams of pure alcohol consumed per day. Non-drinkers (including former drinkers) were regarded the reference category. Statistically significant associations with at least one level of alcohol consumption were reported for six types of cancer: breast, esophagus, liver, oral cavity and pharynx, larynx, and colorectal cancer. When associations were not statistically significant, we assumed the relative risk to be 1. All analyses were adjusted for age and relevant confounders specific for each cancer site (Supplementary Table 1).
Prevalence of alcohol consumption
To estimate current prevalence of alcohol consumption levels we used data from the 4th national risk factor survey, held in 2018 in Argentina. This is a nationwide representative survey of 29,224 individuals aged > 18 years from the general population [25], data are publicly available from: https://www.indec.gob.ar/indec/web/Institucional-Indec-BasesDeDatos-2. The survey contained a beverage-specific quantity-frequency questionnaire on alcohol consumption. Participants had to report their habitual weekly or, when drinking less than weekly, monthly consumption of units of beer, wine, and spirits. A standard unit in the Americas in general contains approximately 14 g of alcohol [9]. We calculated the total alcohol consumption per person in g per day. To be consistent with the consumption levels used in the meta-analysis, we categorized the participants based on their daily alcohol consumption into 1) non-drinkers, 2) light drinkers (0.1 – 12.5 g/day), 3) moderate drinkers (12.6 – 50 g/day), and 4) heavy drinkers (> 50 g/day). Similar categories were used for men and women. We accounted for complex survey design by using sampling weights. We presented the prevalence of the aforementioned consumption categories stratified by age (categorized into 18–29 y, 30–39, 40–49 y, 50–59 y, 60–69 y, and > 70 y) and sex.
Calculation of deaths attributable to alcohol
We used vital registration data from the Directorate of Health Statistics and Information (DEIS) of the Argentine government to derive cause-specific mortality for the year 2018 [26], data are publicly available from https://www.argentina.gob.ar/salud/deis/datos. Data were coded according to the International Classification of Diseases, Tenth Revision (ICD-10). We selected the following codes: C50 (breast cancer), C18-20 (colorectal cancer), C15 (esophagus), C00-C14 (cancer of lip, oral, pharynx), C32 (larynx cancer), and C22 (liver cancer). We excluded data for which information on sex and/or age was missing (N = 2,309, 0.7%). We calculated the number of cancer-specific deaths per age category and for men and women separately. To obtain the number of cancer deaths that can be attributed to alcohol, we multiplied the disease-specific mortality numbers with the PAF of alcohol.
Calculation of DALYs attributable to alcohol
Disability-adjusted life years (DALYs) is a composite measure that accounts both for morbidity (years lived with disability (YLDs)) and premature mortality (years of life lost (YLLs)), to express health loss in the population. YLLs are estimated by multiplying the estimated number of deaths by age with life expectancy at that age. We multiplied the number of deaths per age category, as derived from the vital registration data, with the life expectancy for Argentina per age category according to estimations from the World Health Organization (WHO) [27]. We derived the YLDs from the Global Health Data exchange, using the Global Burden of Disease (GBD) results tool [28], available from http://ghdx.healthdata.org/gbd-results-tool. Following GBD methodology, YLDs were calculated by multiplying disease prevalence by a disability weight [29]. A disability weight quantifies the severity of health loss associated with that disease, ranging on a scale from 0 (perfect health) to 1 (death) [30]. Since cancer has different phases, for which different disability weights are appropriate (e.g., the controlled phase of cancer has a lower disability weight than the terminal phase of cancer), specific disability weights were multiplied with the prevalence of each of the following four phases, or sequelae: 1) diagnosis and primary therapy phase, 2) controlled phase, 3) metastatic phase, and 4) terminal phase. Finally, YLLs and YLDs were summed to obtain DALYs, stratified by sex and age category. We calculated the number of DALYs that can be attributed to alcohol consumption by multiplying it with the PAF of alcohol.
We quantified disease burden stratified by alcohol consumption category. Heavier drinkers often have a higher risk for developing cancer as compared to light-to-moderate drinkers, and some cancer types are associated with heavy drinking only. However, on a population level there are more moderate drinkers than heavy drinkers. Therefore, we calculated for which exposure level the alcohol-attributable cancer burden was highest. Additionally, we explored three hypothetical scenarios to analyze which changes in alcohol consumption levels would account for the greatest reduction in alcohol-attributable cancer burden: 1) heavy drinkers become moderate drinkers, 2) moderate drinkers become light drinkers, and 3) moderate drinkers become light drinkers and heavy drinkers become moderate drinkers.
For three additional types of cancer (melanoma, pancreas, and prostate cancer) the meta-analysis reported a possible association with alcohol consumption. More recent evidence supports this claim [31,32,33]. Therefore, as a sensitivity analysis, we additionally included numbers of alcohol-attributable deaths and DALYs for melanoma (ICD-10 code: C43), pancreas (C25), and prostate cancer (C61) to estimate what the additional alcohol-attributable burden of cancer would be if these associations were to be legitimate.