shows the characteristics of the 42 male members of the Faroese whaling society examined in this study. Among the 41 men with detailed whale consumption data, 26 (63%) consumed three or more whale meals per month. We included the binary (yes/no) weekly alcohol consumption variable in the models for model robustness because the three-categories alcohol variable had no significant associations with the outcomes and the effects of mercury changed only marginally after its inclusion as a confounder in the models. Serum concentrations of cholesterol and triglycerides were considered normal for the age groups examined and were only weakly correlated with indicators of mercury exposure (correlations ranged from 0.07 to 0.16, *p* > 0.3).

| **Table 1**Characteristics of the 42 male members of the Faroese whaling society. |

shows the geometric mean and interquartile range of the mercury concentrations. All exposure biomarkers showed relatively wide ranges, where the highest mercury concentration was 50-fold higher than the lowest in toenails and hair. Mercury levels in hair samples taken 7 years earlier were higher than the levels of the current hair samples. Among the four exposure biomarkers, mercury in toenails showed the closest correlation with hair mercury levels. The average ratio of mercury concentrations in hair to those in toenails was approximately 3. The PCB concentrations were comparatively high and were positively associated with the mercury concentrations (correlation coefficients between 0.47 and 0.54, *p* < 0.002).

| **Table 2**Distribution of mercury and PCB concentrations in specimens used as exposure biomarkers among the 42 Faroese whaling men. |

The mean IMT was highly correlated with the maximal and minimal measures (). We anticipated the results because all three outcomes measured the distance from the right and left CCAs but with different summary estimates. Likewise, correlations between CVRR%, C-CVLF, and C-CVHF were high as components of HRV (). The systolic and diastolic BPs correlated well (

*r* = 0.79) and were weakly and negatively correlated with CVRR. The latencies of peaks III and V at 20 Hz and 40 Hz were highly correlated (), and the mean latencies were similar to those of a study of Faroese children at 14 years of age (

Murata et al. 2004).

| **Table 3**Distribution of outcome variables among the 42 Faroese whaling men. |

Mercury concentrations in current hair and whole-blood samples were highly correlated (*r* = 0.94, *p* < 0.001). To obtain an SEM consisting of exposure variables with independent error terms, we included the results only for toenails, blood, and hair from 7 years previously. This model showed that toenail measurement had the smallest error component. Thus, the toenail mercury biomarker had the highest correlation with the estimated latent exposure (0.98, compared with 0.83 for blood and 0.46 for hair from 7 years before), thereby suggesting that it is the best indicator of long-term MeHg exposure in this population. When combined with the outcome variables, the results again showed that, among the three mercury biomarkers, toenail mercury measurement was the best indicator, with the smallest imprecision and the highest correlation to the latent MeHg exposure ().

| **Table 4**Factor loadings and estimated correlation to each of the latent variables for mercury exposure and outcome groups in an SEM.^{a} |

An SEM was constructed to determine the overall effect of mercury exposure on the groups of outcome variables with adjustment for confounders (). PCB exposure showed no statistically significant associations with the outcomes, and inclusion of PCB as a confounder in the models changed the effects of mercury only marginally. The results are therefore reported without PCB adjustment. High correlations among related measures within this small sample size limited the number of outcome variables that could be included in latent variables. We included all variables shown in in the final model (). Among the HRV indicators, CVRR% was the best indicator of the combined outcomes, with the smallest imprecision and the highest correlation to the latent HRV variable. Likewise, peak V at 20 Hz has slightly better correlation with the latent BAEP latent variable. For CVF, the diastolic BP and heart rate were the better indicators.

For comparison, results are reported both for major individual effect variables and for the latent variables (). We also performed separate models for the latent exposure variable with each of the outcome groupings and obtained similar results. Among the three mercury measures, the coefficients from the regression models revealed some differences. For the CVRR% outcome, for example, the coefficient for the blood concentration was in the direction opposite of expectation, but was close to zero for other exposure biomarkers and the latent exposure variable. Similar results were obtained with the other two HRV measures. In contrast, for both CVF and IMT, tendencies are similar for the three individual exposure parameters and the latent exposure variable, thus supporting a causal linkage. The confidence intervals for the latent exposure interval also take into account the temporal variability of the exposures, and the wider confidence intervals are reflected by increased upper confidence limits. Because standard regression analysis fails to take into account measurement error in exposure variables, the confidence intervals may be too narrow.

| **Table 5**Change in outcome, expressed in percentage of outcome SD (and corresponding 95% confidence intervals) associated with 1 SD increase in exposure.^{a} |