The design and methods of the PREDIMED trial have been described previously [33, 34]. The PREDIMED trial was a randomized, controlled, cardiovascular disease prevention trial based in 11 centers throughout Spain . Institutional Review Boards at all participating centers approved the study protocol.
Eligible participants included men (55–80 year) and women (60–80 year) at high risk for developing cardiovascular disease at enrollment, but had never been diagnosed with cardiovascular disease. High risk was defined as having type 2 diabetes mellitus or at least three of the following major risk factors: current smoking, hypertension, elevated low-density lipoprotein cholesterol levels, overweight or obesity or a familial history of premature coronary heart disease. 7447 participants were recruited between 2003 and 2009. After providing written informed consent, they were randomized to either a traditional Mediterranean diet supplemented with either extra virgin olive oil or tree nuts, or a control (low-fat) diet. The primary cardiovascular disease endpoint was defined as non-fatal acute myocardial infarction, non-fatal stroke or cardiovascular death. Hojiblanca and Patrimonio Communal Olivarero donated extra-virgin olive oil; the California Walnut Commission donated walnuts; Borges donated almonds; La Morella Nuts donated hazelnuts. Sponsors played no role in the design, analysis, interpretation or manuscript writing.
To evaluate the association between fish consumption, mercury exposure and cardiovascular disease, we designed a nested case-control study within the framework of the PREDIMED trial.
Cases and controls
7,232 (97.1%) of PREDIMED participants provided toenail clippings at baseline, within 6 months of randomization. Among the PREDIMED participants who provided toenail samples, we randomly selected 147 of the 288 cases of incident cardiovascular disease (according to the definition of the primary end-point of the trial) during subsequent follow-up. The median follow-up period from time of toenail sampling to time of incident cardiovascular disease was 4.8 years (interquartile range 3.0–5.8 years). Outcomes were identified through repeated contacts with participants, contacts with family physicians, annual medical record review and consultation of the National Death Index. The End-Point Adjudication Committee was blinded to the randomized intervention group and every exposure, and ascertained outcomes. Only confirmed outcomes (non-fatal acute myocardial infarction, non-fatal stroke and cardiovascular death) were included as a case. Cases were randomly matched on age (within 2 years) and sex to 267 controls who were free from cardiovascular disease before December 2010 and had provided toenail samples at baseline. Most cases were matched to two controls, but 23 cases were matched to only one control.
Measurement of exposure
The concentration of total methylmercury in the stored toenails was assessed using instrumental neutron activation analysis (INAA) at the Interfaculty Reactor Institute at Delft University of Technology in Delft, Netherlands. Supplemental materials provide details of analytic methods and validation of these methods .
Specifics of the dietary intervention are provided in detail elsewhere . Registered dietitians conducted one-on-one in-person interviews to collect participants’ dietary information at baseline. We administered a validated 137-item food frequency questionnaire (FFQ) at baseline and yearly thereafter during the follow-up . Participants reported how often, on average over the last year, they consumed each food item (never or almost never, 1–3 times per month, 1 time per week, 2–4 times per week, 5–6 times per week, 1 time per day, 2–3 times per day, 4–6 times per day and >6 times per day). Fish items include white fish (grouper, flounder, sea bream, hake, whiting; 1 serving), dark-meat fish (sardines, tuna, bonito, mackerel, salmon; 1 serving or 130 g); salted fish (cod or salted fish product; 1 serving or 60 g); oysters, clams, mussels (6 units); squid, octopus, cuttlefish (1 serving or 200 g); shellfish (shrimp, prawns, crayfish, etc.; 4–5 pieces or 200 g); natural canned seafood (sardines, anchovies, bonito, tuna; 1 small can or half normal can or 50 g); seafood in oil (sardines, anchovies, bonito, tuna; 1 small can or half normal can or 50 g). We used this information to calculate average fish consumption (in grams per day).
Covariates were measured at baseline and yearly over follow-up. We reviewed medical records and used standardized validated protocols  to collect information on sociodemographic, lifestyle, health, family history and medical diagnoses. A validated Spanish version of the Minnesota Leisure Time Physical Activity Questionnaire  was used to evaluate physical activity. Trained nurses measured weight and height using standardized procedures, and blood pressure using a validated semiautomatic oscillometer in triplicate (Omron HEM_705CP). Yearly glucose tests identified new cases of diabetes. We developed and validated a 14-item Mediterranean diet adherence tool  to repeatedly assess adherence at baseline and during the intervention period. We used the validated FFQ described above to calculate total alcohol and energy intake. Primary care doctors assessed participants for hypercholestaerolemia, hypertension and type 2 diabetes diagnoses.
To examine the association of toenail mercury concentration with cardiovascular disease in the nested case-control study’s total population, we used multivariate-adjusted conditional logistic regression, matching on age and sex. For separate analyses of the intervention groups or control group, we used multivariate-adjusted unconditional logistic regression, adjusting for matching factors. Mercury concentrations (μg/g) were categorized into quartiles based on the distribution among controls. To test for trend, we used the median value of the appropriate quartile of mercury and treated it as a continuous variable.
We again used multivariate-adjusted conditional logistic regression to evaluate mercury exposure in conjunction with fish consumption in the total nested case-control study. Baseline fish consumption was dichotomized into low and high groups according to the PREDIMED 14-item Mediterranean diet adherence tool (high: ≥3 servings/week) . Mercury exposure was dichotomized as above or below the median value. We then created four exposure categories (low mercury/low fish, low mercury/high fish, high mercury/low fish, high mercury/high fish), which we included as indicator variables. The same conditional logistic regression methods were used to evaluate mercury exposure and increase in fish consumption in the nested case-control study. We cross-classified increase in fish consumption from baseline to three years of follow-up (increased or not increased) with baseline toenail mercury concentration (above or below median).
We also used multivariate-adjusted unconditional logistic regression to perform the following sensitivity analyses: 1) excluding diagnosed with cardiovascular disease within one year of baseline; 2) excluding those diagnosed over five years after toenail collection (baseline); 3) excluding participants above the 90th percentile of toenail selenium concentration; 4) stratifying on baseline fish intake (high/low), 5) restricting to participants at very high risk of cardiovascular disease (>70 years old with diabetes and hypertension at baseline); and 6) stratifying on baseline aspirin use (yes/no).
To control for potential confounders, we included in all models those variables based on clinical relevance and previous causal knowledge, as listed in the table footnotes. To evaluate the associations of fish and intake and mercury levels across the distribution of mercury exposure, we constructed restricted cubic splines.
All p-values are two-tailed. Values less than 0.05 are considered statistically significant. We performed all statistical analyses using Stata 12.0.