Alcohol consumption: context and association with mortality in Switzerland

The menuCH study is a cross-sectional population-based study conducted between January 2014 and February 2015 in ten centers across Switzerland [24, 25]. It included two 24-h dietary recalls (24HDR) and one questionnaire about sociodemographic, dietary, and lifestyle factors [25, 26]. The first 24HDR was conducted on-site in one of the centers, and the second one took place two to six weeks later by telephone [25].

In collaboration with the Federal Statistical Office (FSO), a target sample of 4,627,878 Swiss residents was drawn. The stratified, random sample included adults of 18–75 years of age, which represented both sexes, five age categories (18–29, 30–39, 40–49, 50–64, and ≥ 65 years old), the three main language regions in Switzerland (CH-German, CH-French, and CH-Italian), and the twelve most populated Swiss cantons of the seven major regions [25, 27].

From the source population consisting of 13,606 participants, 5496 participants were successfully contacted and eligible for the study [28]. Excluded from the study were 3410 non-responders and 29 participants who did not complete the dietary assessment [28]. The final study sample consisted of 2057 participants [28]. Detailed information on the study recruitment are presented in Online Resource Fig. S1.

The participants’ lifestyle and demographic factors were derived from the self-administered questionnaire [26]. The following variables were used in our study: sex (male, female), age (divided into eleven categories: 18–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–75 years old), Swiss language region (CH-German, CH-French, CH-Italian), education level (primary, secondary, tertiary), physical activity (low, moderate, high; based on the short-form International Physical Activity Questionnaire (IPAQ) definitions [29]), and smoking habits (never, former, current). The anthropometric factors of height, weight, and hip circumference were measured during the first 24HDR by trained interviewers following the WHO-MONICA protocol [25, 30]. The participants’ body weight and height were used to calculate their body mass index (BMI). For pregnant or lactating women, their self-reported weight before pregnancy was used to calculate the BMI. According to the World Health Organization (WHO) definitions, the participants were classified as ‘underweight’ (BMI < 18.5 kg/m²), ‘normal weight’ (18.5 kg/m² ≤ BMI < 25.0 kg/m²), ‘overweight’ (25.0 kg/m² ≤ BMI < 30.0 kg/m²) or ‘obese’ (BMI ≥ 30 kg/m²) [31].

The dietary data were assessed by two 24HDR, which were distributed over all weekdays and seasons [32]. The data were collected by trained dietitians who used the trilingual Swiss version (0.2014.02.27) of the automated software GloboDiet® (GD, formerly EPIC-Soft^®, IARC, Lyon, France [33, 34]), which was adapted by the Federal Food Safety and Veterinary Office, Bern, Switzerland. To facilitate the quantification of consumed amounts, a book with 119 series of six graduated portion-size pictures [35] and about 60 actual household measures were presented to the participants [28]. To ensure the quality of the collected data, the data were screened and cleaned according to the IARC’s guidelines using an updated version of GD^® (0.2015.09.28) [36]. Afterwards, the foods, recipes, and ingredients obtained by the GD^® software were linked using the matching tool FoodCASE (Premotec GmbH, Winterthur, Switzerland) to the most suitable item found in the Swiss Food Composition Database [37].

Each menuCH participant was categorized into one of six alcohol consumption groups considering the participant’s pure alcohol intake from alcoholic beverages in the 24HDR as well as information on general alcohol avoidance from the self-administered questionnaire. Participants, who did not consume alcoholic beverages in the 24HDR and reported alcohol avoidance were categorized as ‘abstainer’, whereas participants who did not report alcohol consumption in the 24 HDR but did not report alcohol avoidance were categorized as ‘safe_no’. Participants, who did consume alcoholic beverages in the 24HDR were categorized into four groups. On the one hand, the categorization was based on whether the participants consumed more pure alcohol during or outside mealtime (‘during’ and ‘outside’ drinkers, respectively). Since the 24HDR did not directly assess the context of drinking, the definition by Sieri et al. [38] was used: an alcoholic beverage was consumed outside mealtime if during that specific time of the day less than 10% of the total daily energy intake (excluding energy intake from alcoholic beverages) was consumed and the consumed amount of pure alcohol was at least 5 g. On the other hand, the categorization was based on whether the participant drank on average more than the maximum daily recommendation (women: 12 g pure alcohol; men: 24 g pure alcohol) proposed by the EKAL [19]. If the participants drank on average more than recommended, they were categorized as heavy drinker (‘heavy’) and otherwise categorized as safe drinker (‘safe’). Therefore, the resulting six alcohol consumption groups were: ‘abstainer’, ‘safe_no’, ‘safe_during', ‘safe_outside’, ‘heavy_during’, or ‘heavy_outside’. An overview of the six alcohol consumption groups is given in the Online Resource Table S1.

To investigate diet quality, the alternate healthy eating index (AHEI) was calculated for each participant for each of the two 24HDR interviews [32]. Then, the average AHEI was determined and used in the analyses of alcohol consumption data. The AHEI is based on eleven components, each having a score between 0 and 10 points [32]; these components are: vegetables, fruits, whole grains, sugar sweetened beverages, nuts, meat, trans fat, long chain omega-3 fatty acids, polyunsaturated fatty acids (PUFA), sodium, and alcohol. We excluded the component ‘alcohol’ since the effect of alcohol consumption should only be contained in the alcohol group variable. Therefore, in the present study, AHEI scores between 0 and 100 were possible, with higher scores indicating a healthier diet.

The Swiss population census data and the mortality data were provided by the FSO [39]. For both databases, only residents between 18 and 75 years old were included to ensure the same age range as in the menuCH study. All-cause mortality and cause-specific mortality determined by the final cause of death (encoded using the 10th revision of the international classification of diseases (ICD-10) [40]) were investigated. The following cause-specific mortalities were investigated: CVD (ICD-10: I00-I99), all-cancer (ICD-10: C00-C97, D32-D33, and D37-D48), colorectal cancer (ICD-10: C18-C21), liver cancer (ICD-10: C22), upper aero-digestive tract (UADT) cancer (all organs and tissues of the respiratory tract, upper part of the digestive tract, and the upper esophagus (ICD-10: C00-C15 and C32), but excluding the stomach), breast cancer (only in women; ICD-10: C50), diabetes (ICD-10: E10-E14.9).

In addition, based on evidence for alcohol-related carcinogenic effects on human organs [17, 18], the following eight cancer sites were combined into one group: colorectal, liver, UADT, breast, prostate (ICD-10: C61), pancreatic (ICD-10: C25), urinary tract (ICD-10: C67-C68), stomach (ICD-10: C16) [41, 42].

Mortality ratios standardized for sex, age, and year of death (SMR) were calculated at the district level by dividing the number of observed deaths by the number of expected deaths in the overall Swiss population, based on the Swiss population census data. The latter was determined using an indirect method based on the standardized Swiss population mortality rates. The district of each menuCH participant was determined using their postal code and data provided by the FSO dated on the 1st of January 2019 [43, 44].

Since not all Swiss inhabitants had the same probability to be included in the menuCH study sample, the analyses of the participants’ data of the final sample were weighted based on sex, age, marital status, major living region, nationality, household size, weekday, and season of the 24HDR day [27].

Descriptive statistics (absolute numbers, median, interquartile range, and percentages) were used to characterize the study population. Additionally, descriptive maps were used to show the geographic distribution of chronic diseases district-level SMR. The association between alcohol consumption and mortality was investigated by modeling mortality rates, fitting generalized linear regression models. Negative binomial regression models were used to handle overdispersion (for all-cause, CVD, all-cancer, and breast cancer mortality) and Quasipoisson models to handle underdispersion (for colorectal cancer, liver cancer, UADT cancer, and diabetes mortality). For each of the menuCH participants, the total number of observed deaths in the participant’s sex, age, and district category was used as outcome variable. The explanatory variable was the alcohol consumption group and the participant’s sex, age (continuous variable), smoking status, physical activity level, BMI group, education level, and average AHEI were included as further covariates. The log of the total number of residents stratified by the participant’s sex, age, and district category was included as offset term.

Missing values for physical activity level (n = 524), education level (n = 3), and smoking category (n = 4) (Table 1) were imputed using multivariate imputation by chained equations (MICE) [45]. The results of the 30 imputed data sets were pooled using Rubin’s rule [46]. In general, the results obtained with the imputed data sets were similar to the results of the complete case analyses. Therefore, only the results based on the imputed data sets are presented and used for further analyses.

The districts were defined as neighboring districts based on a first order neighborhood structure with rook contiguity. Additionally, the neighbors’ data were weighted by taking the inverse of the total number of neighbors of the corresponding district. The residuals of the regression models were investigated for spatial autocorrelation at the district level using global and local Moran’s I. The global Moran’s I statistic is an indicator for the existence and degree of spatial autocorrelation [47]. The statistic can range from -1 indicating spatial dispersion up to + 1 indicating spatial cluster-building [47]. A one-sided P value based on the Z-score was calculated [48]. To check the robustness, a one-sided P value based on 1000 Monte Carlo (MC) simulations was calculated additionally. Local Moran’s I were checked for significance based on a permutation test (n = 1000). No correction for multiple testing was included, since the number of Monte Carlo simulations determined the lower limit of the P value [49]. Local Moran’s I values were visualized using local indicators of spatial autocorrelation (LISA) cluster maps.

If evidence for spatial autocorrelation was detected, an integrated nested Laplace approximation (INLA) model was fitted. The structured spatial component was a Besag model [50] and the unstructured spatial component was an iid model (random noise). The default LogGamma prior distribution (shape = 1; rate = 0.00005) was used for both components. The results for each imputed data set were pooled by calculating the average of the estimates.

The analyses were performed in GeoDa (version 1.14.0) and in the R programming language (version 4.1.0; [51]). In R, the packages popEpi (version 0.4.8) and Epi (version 2.44) were used to calculate the SMR, mice (version 3.13.0) to impute missing values, survey (version 4.1.1) and DescTools (version 0.99.42) to conduct weighted analyses, MASS (version 7.3.54) to fit generalized linear regression models, spdep (version 1.1.8) and rgeos (version 0.5.5) to conduct spatial analyses, ggplot2 (version 3.3.5), ggsn (version 0.5.0), and sf (version 1.0.2) for creating figures, and INLA (version 21.11.22) to set up INLA models. For all analyses, the statistical significance level was set to 0.05.

The menuCH participants’ baseline characteristics stratified by alcohol consumption group are shown in Table 1. The largest group were the occasional drinkers (safe_no) with 678 participants and the two smallest were the two outside mealtime groups with 117 and 115 participants, respectively. All six alcohol consumption groups were characterized by differences across the variables investigated in comparison to the overall study population, e.g., abstainers and occasional drinkers were more likely to be never smokers, whereas during mealtime drinkers were more likely to be current or former smokers and outside mealtime drinking was more common among younger and during mealtime drinking among older participants. Median pure alcohol consumption per day stratified by sex and context of drinking can be seen in Fig. S2 and Fig. S3 (Online Resource).

Figure 1 shows the average pure alcohol intake per person stratified by sex, weekday, and context of drinking. Regardless of weekday and context, men drank more than women. Both sexes consumed more pure alcohol during mealtime than outside. The amount of pure alcohol intake was lower at the beginning of the week, increased from Thursday onwards and reached its peak on Saturday. On Saturday, men consumed on average 23.7 g pure alcohol during and 8.5 g outside mealtime and women on average 11.1 g pure alcohol during and 3.4 g outside mealtime. On Saturday only, both sexes consumed on average more than the maximum daily recommendation given by the EKAL [19].

Between the years 2015 and 2018, the following number of deaths were documented in Switzerland: 84,959 all-cause deaths, 16,082 CVD deaths, 37,202 all-cancer deaths, 3327 colorectal cancer deaths, 1821 liver cancer deaths, 2360 UADT cancer deaths, 3064 breast cancer deaths among women, 1256 diabetes deaths, and 17,362 deaths attributable to the eight specific cancer sites group.

The SMR at the district level are shown in Fig. 2. All-cause and all-cancer maps revealed a similar pattern with high SMR mainly in the western region and low SMR mainly in the central region. For CVD mortality high SMR were detected in the northwestern and eastern region. In contrast, many districts with low SMR were observed in the southwestern region. No clear pattern was detected for any of the specific cancer type mortalities, except for liver cancer: CH-French and CH-Italian districts tended to have higher SMR than CH-German districts. High diabetes SMR were more prevalent in central and northwestern region, whereas lower diabetes SMR were observed in the northeastern and southwestern region.

Table 2 shows the results of the generalized linear regression models, with occasional drinkers used as reference group. In general, the consumption of alcoholic beverages tended to increase mortality rates, especially for all-cancer and UADT cancer mortality. Heavy, during mealtime drinkers had a higher risk of all-cancer (RR = 1.05, 95% CI 1.00, 1.10), breast cancer (RR = 1.10, 95% CI 1.00, 1.21), and UADT cancer (RR = 1.19, 95% CI 1.09, 1.31) mortality. Furthermore, heavy, outside mealtime drinkers had an increased UADT cancer (RR = 1.20, 95% CI 1.00, 1.43) and diabetes mortality (RR = 1.35, 95% CI 1.07, 1.71). Even for safe drinkers there was evidence for an increased risk of all-cancer (safe_during: RR = 1.07, 95% CI 1.02, 1.13), liver cancer (safe_outside: RR = 1.27, 95% CI 1.07, 1.51), and UADT cancer (safe_during: RR = 1.15, 95% CI 1.04, 1.26) mortality. Interestingly, the abstainers showed evidence for an increased all-cancer (RR = 1.07, 95% CI 1.00, 1.14) and UADT cancer (RR = 1.21, 95% CI 1.07, 1.37) mortality.

The residuals of the generalized regression models were investigated for spatial autocorrelation at the district level using global Moran’s I statistic. The results of the spatial autocorrelation analysis are shown in Table 3. Only for all-cancer mortality, there was evidence for spatial autocorrelation (observed global Moran’s I: 0.144; expected global Moran’s I: − 0.014). The significant local Moran’s I values are visualized in a LISA cluster map (Online Resource Fig. S4). In total, 5 districts showed evidence for spatial clusters or spatial outliers (Fig. 3).

Based on results of the INLA model (Table 4), there was evidence for an increased all-cancer mortality among abstainers (RR = 1.07, 95% CI 1.00, 1.14); safe, during mealtime drinkers (RR = 1.08, 95% CI 1.03, 1.13); and heavy, during mealtime drinkers (RR = 1.04, 95% CI 1.00, 1.09) compared to occasional drinkers.

Districts with a statistically significant structured spatial component are shown in Fig. 3. The districts Uster (RR = 0.91, 95% CI 0.82, 0.99), Hinwil (RR = 0.91, 95% CI 0.81, 1.00), and Meilen (RR = 0.89, 95% CI 0.78, 0.99) revealed evidence for a decreased all-cancer mortality rate, whereas the districts canton Neuchâtel (RR = 1.12, 95% CI 1.01, 1.27), Jura Bernois (RR = 1.13, 95% CI 1.01, 1.31), Seeland (RR = 1.14, 95% CI 1.01, 1.33), and Biel (RR = 1.18, 95% CI 1.01, 1.40) revealed evidence for an increased all-cancer mortality rate.

The adjustments for diet quality via the AHEI did not meaningfully change the results of our study. The SMR for the cancer sites with evidence for a carcinogenic effect of alcoholic beverages [17, 18], which are not presented in Fig. 2, are shown in Online Resource Fig. S5. The results of the negative binomial regression model for mortality from the eight specific cancer sites were similar to those for all-cancer mortality (Online Resource Table S2). The global Moran’s I statistic revealed no evidence for spatial autocorrelation (observed Moran’s I: − 0.060, expected Moran’s I: − 0.014) and therefore, no INLA model was fitted.