*2.2. Ethics*

The NCHS Research Ethics Review Board (ERB) approved the 2013–2014 study protocol (protocol 2011-17). NHANES and all participants provided written informed consent.

#### *2.3. Interactions under Study*

Table 1 shows the drug–dietary supplement interactions studied. We considered medication groups if they represented a prevalence of use higher than 2% of the population included in the database.



Tetracycline interacts with divalent ions such as calcium, magnesium, and zinc, forming a relatively stable and poorly absorbed chelate, preventing absorption of the antibiotic due to a lower amount of calcium in the gu<sup>t</sup> available to be absorbed. This interaction may reduce or even abolish the therapeutic effect of the antibiotic, thereby diminishing anti-infectious e fficiency. For this reason, tetracycline should be taken one hour before or two hours after meals [3,5].

Thiazide diuretics can cause increased calcium reabsorption in distal tubules of the kidneys, which contributes to hypercalcemia. Another cause of hypercalcemia is the excess of vitamin D, for example, through high doses of oral supplements, which increases the absorption of calcium in the gut. Due to the retention of calcium in the body, metabolic alkalosis may be developed [5].

Finally, angiotensin II receptor blockers are potassium-sparing and can, therefore, have additional hyperkalaemic e ffects if combined with potassium supplements or salt substitutes containing potassium. The use of potassium supplements is the main risk factor for developing hyperkalaemia, as this causes a rapid rate of increase in serum potassium levels. Other contributory risk factors such as poor renal function, heart failure, and diabetes should also be considered, as they are associated with a faster rate of hyperkalaemia progress [5,8].

#### **3. Population and Prescription Medication Information**

The sample was composed of tetracyclines, thiazides, and/or angiotensin II receptor blocker users. These drugs were chosen due to the potential severity of their interactions and their high prevalence of use in the American population. Prescription medication information was obtained through the Prescription Medication subsection included in The Dietary Supplement and Prescription Medication section of the Sample Person Questionnaire. This section provides personal information on the use of prescription medication in the month prior to the participant's interview. The name of the medication was provided by the participant to the interviewer, who entered it into the computer where it was automatically matched to a generic drug name and code. Medication is presented following the WHO Drug Statistics Methodology of ATC index [9].

*Tetracyclines*: We considered a patient to be a tetracycline consumer if he/she reported having taken a medication with a generic drug name included in the ATC group J01AA.

*Thiazides*: We considered a patient to be a thiazide consumer if he/she reported having taken a medication with a generic drug name included in the ATC group C03AA.

*Angiotensin II receptor blockers*: We considered a patient to be an angiotensin II receptor blocker consumer if he/she reported having taken a medication with a generic drug name included in the ATC group C09CA.

#### **4. Dietary Supplement Information**

Dietary supplement information was obtained through the dietary supplement subsection also included in The Dietary Supplement and Prescription Medication section of the Sample Person Questionnaire. This subsection allows for collection of personal data on the use of dietary supplements

in the month prior to the participant's interview. Interviewers reported the supplement product name, which was automatically disaggregated to up to 34 nutrients.

We considered a participant to be a nutrient supplement consumer if he/she had taken any supplement containing at least one of the nutrients under study (Table 1).

## **5. Other Covariates**

Demographic covariates were sex, age, race/ethnicity (Hispanic origin: Mexican-American, other Hispanic, non-Hispanic white, non-Hispanic black, other race), educational level (primary, secondary, university), civil status (married/with partner, widow/er, divorced, single), citizenship (American/non-American), and country of birth (U.S./Other country).

We also considered body mass index (BMI) and physical activity. BMI was determined from height and weight measured by health technicians previously trained by an expert anthropometrist. We categorised this variable as follows: underweight (<18.5), normal weight (18.5–24.9), overweight (25.0–29.9), and obesity type I (30.0–34.9) and obesity type II and type III (≥35.0).

Physical activity was self-reported and measured through a question asking if participants did any moderate-intensity sports in a typical week. The answer was dichotomic (yes/no).

## **6. Statistical Analysis**

Prevalence rates were calculated for each potential interaction described in Table 1. The reference population consisted of those who reported having taken one of the medications under study.

A multivariate logistic regression analysis was conducted to determine the factors associated with a higher probability of having a potential interaction using the presence/absence of the interaction as the dependent variable and demographic and clinical variables as independent variables. The regression model provided odds ratios (OR) and 95% confidence intervals for the associations between the dependent variable and each of the independent variables. Crude associations obtained from bivariate logistic regressions were also presented.

To test potential interactions between demographic and clinical variables, such as age and physical exercise or gender and BMI, we tested the association between the probability of having a potential interaction and the interacting term of the dependent variables. The interacting terms were not statistically significant and, therefore, were not included in the final model.

All analyses were performed taking into account the appropriate weights. This procedure was developed with the goal of obtaining nationally representative estimates and accounting for unequal probability of selection derived from study design and non-response.

STATA 13.1MP was used to perform the statistical analyses.
