The STANISLAS cohort is a population-based study of 1,006 families each comprised of at least 2 parents and 2 children (4,295 participants) from the Lorraine region (Eastern France) recruited during 1993–1995 at the Center for Preventive Medicine. The participants were of French origin and free of acute or chronic disease. From 2011 to 2016, 1,705 participants underwent their fourth examination. The STANISLAS study has been described in detail elsewhere (21). The present study focused on the 1,695 participants who underwent their fourth examination and for whom data on food intake were available. After excluding 10 participants without food intake data, 63 with a daily energy intake either below 1,000 or above 5,000 kcal, 27 with cardiovascular (CV) disease (11 with myocardial infarction only, 12 with heart failure only, and 4 with both), 105 with at least one missing subclinical organ damage assessment (19 kidney, 82 CV, and 17 metabolic outcomes), and 198 with missing data regarding baseline characteristics, the present cross-sectional analysis included 1,302 participants (Figure 1).

The research protocols were all approved by the local Ethics Committee (Comité de Protection des Personnes Est III—Nancy—France), and all study participants gave written informed consent. The informed written consent was previously approved by the local Ethics Committee.

Participant characteristics were first described according to total water intake according to HBI quartiles. Differences between the groups were assessed using Fisher's exact test or the chi-square test for categorical variables, and the Kruskal–Wallis non-parametric test for continuous variables.

A graphical representation of HBI component scores was also preformed for the overall population and by generation (Generation 1: ≥ 50 years and Generation 2: <50 years) and sex using a radar chart. The association between HBI or each of its components with each kidney and cardiometabolic outcome was then investigated. Of note, HBI component 4 (diet drinks) alone was not considered since only 22 participants (<2%) had a score different from the majority. Due to the family structure of the STANISLAS cohort, a family random intercept was added in each model to account for the non-independency of members of the same family. For continuous outcomes, linear mixed models with the Kenward-Roger correction for degrees of freedom were used to calculate p-values from conditional F-tests and 95% profile CIs of estimates, using the lmerTest package (35). For binary outcomes, the binomial generalized linear mixed models (with a logit link) were used using the glmmTMB package (36) and p-values were calculated using likelihood-ratio tests. In addition, to ascertain whether the associations between HBI or its components and the studied outcomes were dependent on generation or sex, data were tested for 2-way interactions between HBI or its components and generation or sex (except for binary outcomes, due to the very small number of events). The interaction between HBI and total energy intake was also tested, although the latter was not significant (all p > 0.11). Covariates were selected using a two-step process: (1) univariable multinomial logistic regressions were run with HBI in quartiles as the response variable and variables with p < 0.2 were retained, and (2) a multiple multinomial logistic regression was run with all variables selected during the first step from which variables with p < 0.05 were retained as covariates. The selected covariates were sex, smoking status, and energy intake, the models were further adjusted for age (standardized within sex and generation), generation, income, BMI (body mass index) class, and physical activity. For cardiometabolic outcomes, the models were further adjusted for DBP. To improve model convergence, energy intake and physical activity were standardized. When considering an HBI component as exposure variable, the results were adjusted for the other components, except component 1 (water) due to its collinearity with component 10 (beverage volume). This meant that when considering component 1, the latter was not adjusted for component 10. Model residual checks led to the log-transformation of cIMT and pulse-wave velocity. To facilitate interpretation, beta coefficient and 95% CI were exponentiated for these models.

The median (Q1-Q3) HBI score was 89.7 (78.4–95). Details of all beverage consumptions are presented in Supplementary Table 1. The HBI score was higher (p < 0.001) in women in both generations [91.8 (85.8–95) for Generation 1 (≥ 50 years) and 90.3 (81.8–95) for Generation 2 (<50 years)] compared with men [86.3 (74.8–93.3) for Generation 1, 88.1 (75.0–93.8) for Generation 2]. Differences were observed for sex and generation among the different HBI components. For both generations, women had lower scores than men for “coffee and tea,” indicating that they drank more coffee or tea than recommended. Generation 1 men had the lowest scores for “alcohol,” “total beverage energy,” and “total fluid consumption” (Figure 2).

Participants in the highest HBI quartile (>95) were more likely to be women and non-smokers, have higher incomes, lower energy intake, and blood pressure (Table 2).

There was no strong evidence for an association between any of the studied outcomes and HBI (in quartiles or continuous, Table 3).

When looking at each HBI component individually, the odds of HTG waist were higher among participants who did not meet the “sugar-sweetened beverages” (0–8% of fluid requirements) criteria (p = 0.009; Table 4).

The “beverage volume” score was associated with LV mass, with lower LV mass for participants who did not meet the criteria (−1.5 [−2.94; −0.07], Table 4). Moreover, participants who did not meet the “full-fat milk” criteria (0% of fluid requirements) had lower cfPWV and cIMT (0.96 [0.92; 0.99] and 0.94 [0.90; 0.97], respectively, Table 4). Sex-specific effects were observed for the association of “low fat milk” / “alcohol” with cfPWV. Men who did not meet the criteria for “low-fat milk” and “alcohol” (0–2 drinks) had higher cfPWV, whereas the reverse was true for women (0–1 drinks) (Table 4).

The association of “low-fat milk” with eGFR was driven by generation. Generation 2 participants who did not meet the criteria for “low-fat milk” (0–16% of fluid requirement) had a lower eGFR while the reverse was true for Generation 1 participants, although to a lesser extent (Generation 1 (≥ 50 years): 3.76 [−0.87; 8.39]; Generation 2 (<50 years): −6.47 [−11.31; −1.63], Table 4). Participants who did not meet the criteria for “coffee and tea” (0–40% of fluid requirements) had a trend for higher eGFR (1.76 [−0.08; 3.6], Table 4). No associations were observed between HBI components and albuminuria.

In the NHANES study, higher HBI was associated with more favorable lipid profiles, lower hypertension risk in men, and lower CRP, but not with MetS (20). Similarly, in the present study, no associations were found between HBI score and either MetS or subclinical organ damage. The HBI score was relatively high [89.7 (78.6–95)] in our cohort compared with that of the NHANES study (63 ± 16) (20). The differences in HBI scores between the two cohorts could be due to two factors. First, our cohort was comprised of initially healthy individuals, along with a relatively homogenous global consumption, thereby limiting the observation of an association with the studied outcomes. Second, culture can influence eating behaviors including beverage intake (37, 38). Although water is the major contributor of beverage consumption in the United States and in France (39–41), the United States is nonetheless characterized by higher consumption of SSBs and milk (37, 42). The total fluid intake was seemingly higher in the United States compared with the present study and other studies in France. It is noteworthy to emphasize that both water and fluid intake were higher in the present study compared with other French studies (39, 43–45). Having different drinking patterns across countries can lead to different scores, thus, the associated risks may not be the same between studies. Of note, the HBI components herein were associated with different outcomes. One may speculate whether compensation between the consumed beverages may counterbalance the overall score and result in the absence of association.

Sugar-sweetened beverages are the largest source of added sugar in the diet and have consequently drawn much attention related to their etiological role in relation to obesity, type 2 diabetes, and CV risk (11, 46, 47). These associations may be partly mediated by adiposity (48). In the European PREDIMED study, higher SSB consumption (> 5 servings/week) was positively associated with MetS incidence (49). Nonetheless, the association between SSB and the presence of MetS remains less consistent (11), whereby associations are seemingly stronger when MetS components are considered individually. This discrepancy has been largely discussed due to the still much-debated clinical utility, consensus assessment, and heterogeneity of MetS.

In the present study, no association was found with MetS. Interestingly, there was a significant association between HTG waist and SSB consumption, although this must be considered cautiously in view of the low number of cases. This observation is in line with previous systematic review data demonstrating positive associations for blood pressure, triglycerides, low-density lipoprotein (LDL) cholesterol, and blood glucose, with an inverse association observed for high-density lipoprotein (HDL) cholesterol (50), and abdominal adiposity (11, 48).

With regard to CV outcomes, high SSB intake has been associated with a higher risk of stroke and myocardial infarction (50, 51). We did not find an association with subclinical vascular damage, i.e., with either IMT, in accordance with a previous study in middle-aged US women (52), or cfPWV, whereas another study conversely reported an association between increased intake in fructose derived from industrial sweetened beverages (> 1 drink/day) and higher cfPWV (13).

Having an adequate fluid intake is essential to avoid dehydration and the subsequent development of health issues or chronic diseases (53). In addition, hydration can have an influence on heart function and hemodynamic response. Indeed, blood volume is regulated by matching total water input and output (54). In the present study, drinking fewer beverages than the recommended fluid requirement was associated with lower LV mass. Nonetheless, this criterion was calculated based on consuming 1 ml liquid for each 1 kcal food consumed, which may be approximative, since not taking into account other lifestyle and personal determinants.

Regarding alcohol, men who did not meet this criterion (0–2 drinks/day) had higher cfPWV. This result is consistent with other studies showing that alcohol consumption is associated with higher cfPWV (55). Interestingly, in a study involving the same cohort, our group previously identified an alcohol pattern in Generation 1 men, and a fast-food and alcohol pattern in Generation 2 men whereby the former was positively associated with cfPWV, but not the latter (14). The present study further revealed that when focusing on beverages only, regardless of the generation, men who did not meet the criteria of alcohol intake had higher cfPWV. The fact that no association was observed for women could be the result of men adhering less to the criteria than women, especially Generation 1 men. Furthermore, the impact of alcohol on CV damage may differ between women and men, in relation to their specific cardiometabolic profiles (56). However, only a whole alcohol category was used in the HBI, i.e., the type of alcohol was not specified, whereas the type of alcoholic beverages may differentially affect the vascular system (57).

The effect of milk on health has been controversial in recent years due to its nature and content in saturated fat, although the presence of some peptides from whey proteins and caseins may have several biological activities (hypocholesterolemic, antihypertensive, antithrombotic, or antioxidant) that have potential beneficial effects on cardiometabolic risks (58). In a recent meta-analysis, milk intake was found to be associated with a reduced risk of hypertension, but not CVD (16). Moreover, the PURE study showed that the consumption of dairy products was associated with a lower risk of mortality and major events related to CV disease (59). Even if low-fat dairy products have become popular in response to the needs of consumers for reduced fat in food products (60), both low-fat dairy products and whole milk have been associated with lower risks of hypertension (60, 61) and better CV health (62). Nevertheless, the HBI score set at 0% of fluid requirements is based on 2% or full-fat milk. In the present study, we observed that drinking full-fat milk was negatively associated with cfPWV and cIMT. In contrast, drinking low-fat milk above 0–16% of fluid requirements was positively, although marginally, associated with PWV in men only.

Drinking low-fat milk above the 0–16% of fluid requirements was associated herein with a lower eGFR in Generation 2 only. Another study found no association between daily consumption of milk, milk products, or low-fat dairy and annual eGFR decline whereas, in a subgroup of participants with mildly decreased eGFR, daily consumption of ≥2 servings of milk and milk products or low-fat dairy was associated with less annual decline in eGFR (63). Their findings are thus in contradiction with the current study. Further studies are therefore needed to better understand this particular aspect. An important feature to take into consideration is the normal kidney function observed in the present cohort.

Coffee and tea contain a mixture of components, notably polyphenols, featuring antioxidant and anti-inflammatory properties. Tea and coffee intake have been associated with multiple health-related effects, notably a decreased prevalence of certain diseases, such as cancer or metabolic diseases (6). However, few studies have investigated the association between coffee intake and CKD prevalence, even less for tea, the results of which are controversial. Of two recent meta-analyses, one found an association between coffee intake and a lower risk of incident CKD (9) while the other did not find any significant association (6, 64). The present study may suggest that drinking coffee and tea beyond 0–40% fluid requirements is associated with higher eGFR. While this criterion is binary, the protective effects of coffee nonetheless appear to be dose-dependent (65). Given that the STANISLAS population displayed a normal eGFR on average, we were unable to focus on CKD per se, although it can be inferred from our results that participants who met the coffee and tea criteria may exhibit a better preserved kidney function.

Overall, the HBI could help in implementing the specific guidelines on all beverage types and amounts for the general population, completing the existing guideline on water and alcohol (2, 17) intake and the extended guideline on SSBs (66).

The strengths of this study are multiple. First, the analyses were based on a large general and initially healthy population-based cohort. Second, our cohort allowed investigating the effects of generation and sex. Third, this study is based on quality data owing to the availability of comprehensive and detailed information relative to diet, kidney function, and extensive CV phenotyping.

However, certain limitations of this study should be acknowledged. First, the results are based on cross-sectional data and thus causality cannot be implied due to the study's observational nature. Second, the HBI was designed based on an American population and recommendations. Nonetheless, the guidelines in terms of beverages are partial.