Association between alcohol consumption and kidney stones in American adults：2007-2016 NHANES

Zhen Zhou, Zhicong Huang, Guoyao Ai, Xin Guo, Guohua Zeng et al. (+1)•View original
HighlightsBalancedalloy voice
Let’s zoom in on what this paper actually did statistically to answer a very practical question: is alcohol consumption associated with having kidney stones? First, the main finding in one sentence: using nationally representative NHANES data from 2007 to 2016, the authors found no statistically significant association between alcohol consumption—either lifetime or in the past 12 months—and the prevalence of kidney stones after full adjustment for confounders. Why that matters: prior studies have been mixed, and clinicians and patients often wonder whether drinking raises or lowers stone risk; this analysis suggests any apparent signal is likely explained by other factors rather than alcohol itself. Now, how did they analyze it? The study is cross-sectional and uses adult participants in NHANES with both kidney stone history and alcohol data. Exposure came from 10 Alcohol Use Questionnaire items, labeled Q1–Q10, spanning lifetime and recent drinking. Because NHANES is a complex, multistage survey, the authors followed NHANES guidance and applied the recommended sampling weights, merging weights across five continuous cycles in a way appropriate to the variable collected on the smallest number of respondents. In other words, the statistics were designed to produce population-representative estimates. They began by describing baseline characteristics by kidney stone history using weighted proportions. Differences between groups were tested with a survey-weighted chi-square test for categorical variables. Notably, even continuous variables were treated as categorical in the baseline table, again summarized as weighted proportions. To assess the alcohol–stone relationship, they ran three logistic regression models with kidney stone history as the outcome and each alcohol variable as the exposure: 1) a crude model with no covariates; 2) Model I adjusted for age, gender, and race; and 3) Model II fully adjusted for a wide range of potential confounders: age, gender, race, marital status, education, recreational physical activity, smoking, asthma, overweight status, gout, congestive heart failure, coronary heart disease, angina, stroke, cancer, diabetes, hypertension, and BMI. The type of alcohol was not specified in NHANES, so it wasn’t modeled. Importantly, they analyzed the ten alcohol variables separately, often categorizing continuous exposure measures and estimating trends across categories. They also performed univariate analyses of each covariate with kidney stone history to show crude associations. Statistical significance was assessed with two-tailed P < 0.05, and analyses were conducted in R and EmpowerStats. According to Table 2 (summarizing odds ratios across the three models), initial crude or minimally adjusted associations (for example, Q1 looked protective, Q4 looked risky) disappeared after full adjustment. Table 3 lists the univariate covariate associations, and the flow chart on page 8 (Figure 1) shows the inclusion and exclusion process across NHANES cycles and the creation of the Q1–Q10 samples. The key takeaway: a survey-weighted, multivariable logistic regression framework—moving from crude to fully adjusted models—demonstrated that any apparent link between alcohol and kidney stones is not robust once confounding factors are accounted for.

Let’s zoom in on what this paper actually did statistically to answer a very practical question: is alcohol consumption associated with having kidney stones?

First, the main finding in one sentence: using nationally representative NHANES data from 2007 to 2016, the authors found no statistically significant association between alcohol consumption—either lifetime or in the past 12 months—and the prevalence of kidney stones after full adjustment for confounders. Why that matters: prior studies have been mixed, and clinicians and patients often wonder whether drinking raises or lowers stone risk; this analysis suggests any apparent signal is likely explained by other factors rather than alcohol itself.

Now, how did they analyze it? The study is cross-sectional and uses adult participants in NHANES with both kidney stone history and alcohol data. Exposure came from 10 Alcohol Use Questionnaire items, labeled Q1–Q10, spanning lifetime and recent drinking. Because NHANES is a complex, multistage survey, the authors followed NHANES guidance and applied the recommended sampling weights, merging weights across five continuous cycles in a way appropriate to the variable collected on the smallest number of respondents. In other words, the statistics were designed to produce population-representative estimates.

They began by describing baseline characteristics by kidney stone history using weighted proportions. Differences between groups were tested with a survey-weighted chi-square test for categorical variables. Notably, even continuous variables were treated as categorical in the baseline table, again summarized as weighted proportions.

To assess the alcohol–stone relationship, they ran three logistic regression models with kidney stone history as the outcome and each alcohol variable as the exposure: 1) a crude model with no covariates; 2) Model I adjusted for age, gender, and race; and 3) Model II fully adjusted for a wide range of potential confounders: age, gender, race, marital status, education, recreational physical activity, smoking, asthma, overweight status, gout, congestive heart failure, coronary heart disease, angina, stroke, cancer, diabetes, hypertension, and BMI. The type of alcohol was not specified in NHANES, so it wasn’t modeled.

Importantly, they analyzed the ten alcohol variables separately, often categorizing continuous exposure measures and estimating trends across categories. They also performed univariate analyses of each covariate with kidney stone history to show crude associations. Statistical significance was assessed with two-tailed P < 0.05, and analyses were conducted in R and EmpowerStats. According to Table 2 (summarizing odds ratios across the three models), initial crude or minimally adjusted associations (for example, Q1 looked protective, Q4 looked risky) disappeared after full adjustment. Table 3 lists the univariate covariate associations, and the flow chart on page 8 (Figure 1) shows the inclusion and exclusion process across NHANES cycles and the creation of the Q1–Q10 samples.

The key takeaway: a survey-weighted, multivariable logistic regression framework—moving from crude to fully adjusted models—demonstrated that any apparent link between alcohol and kidney stones is not robust once confounding factors are accounted for.