The subjects in NHANES III were civilian, non-institutionalized persons in the United States 2 months of age or older. The survey was conducted from 1988 to 1994. Approximately 40,000 persons were selected to participate in the survey. The subjects for the present study were 16,573 men and women, 20 years old or older, who participated in an examination at a mobile examination center.

Subjects attended one of three daily sessions at a mobile examination center: morning, afternoon, or evening. The sessions could last up to three and one-half hours. Subjects were asked to fast over-night for the morning session and for six hours before an afternoon or evening session.

The sample design was a stratified, multistage probability design. In the first stage of sampling, 81 primary sampling units (PSUs) were selected. The PSUs were individual counties or adjacent counties. Thirteen of the large PSUs were divided into 21 survey locations and the remaining 68 PSUs had one survey location. The 89 survey locations or ‘stands’ were randomly divided into two phases. Phase I consisted of 44 locations visited from 1988 to 1991. Phase II consisted of 45 locations visited from 1991 to 1994. Later stages of sampling included area segments, households, and sample persons. Details about the survey design can be found in Ezzati, Massey, Waksberg, Chu, and Maurer [6].

Venous blood samples were taken at mobile examination centers or during home examinations given to persons who could not go to a mobile examination center. Blood lead was measured by atomic absorption spectrometry in persons one year and older. The limit of detection for the blood lead measurements was 1 μg/dL. Values below the limit of detection were assigned a value of 0.7 μg/dL. The percent of measured values below the limit of detection was 7.93%.

Details about the measurement of blood lead and the other blood and urine measurements can be found in the NHANES III laboratory manual [7].

Thyroid stimulating hormone was measured in blood serum samples of persons 12 years and older by a chemiluminescence immunometric assay. The limit of detection was 0.01 μU/mL. Values below the limit of detection were assigned a value of 0.00 μU/mL. The percent of measured values below the limit of detection was 0.41%.

Total circulating thyroxine was measured in blood serum samples of persons 13 years and older by an enzyme-based homogenous immunoassay. The limit of detection was 0.5 μg/dL. Values below the limit of detection were assigned a value of 0.4 μg/dL. The percent of measured values below the limit of detection was 0.21%.

Serum antimicrosomal antibody was measured by radio-immunoassay in persons 12 years and older. The limit of detection was 0.5 U/mL. Values below the limit of detection were assigned a value of 0.3 U/mL. The percent of measured values below the limit of detection was 86.75%.

Serum anti-thyroglobulin antibody was measured by radio-immunoassay in persons 12 years and older. The limit of detection was 1.0 U/mL. Values below the limit of detection were assigned a value of 0.7 U/mL. The percent of measured values below the limit of detection was 89.22%.

Spot urine specimens were collected at mobile examination centers by the clean-catch technique into sterile 250 mL polyethylene containers.

A Jaffé rate reaction, where creatinine reacts with picrate in an alkaline solution to form a red creatinine-picrate complex, was used to measure urinary creatinine in persons six years or older. The limit of detection was 10 mg/dL. Values below the limit of detection were assigned a value of 7.9 mg/dL. The percent of measured values below the limit of detection was 0.37%.

Urinary iodine is measured by using the reduction-oxidation reaction between ceric and arsenite catalyzed by iodide in persons six years or older. The limit of detection was 0.2 μg/dL. Values below the limit of detection were assigned a value of 0.1 μg/dL. The percent of measured values below the limit of detection was 0.03%.

Cotinine was measured in blood serum samples of persons four years and older by an enzyme immunoassay. The limit of detection was 0.05 ng/mL. Values below the limit of detection were assigned a value of 0.035 ng/mL. The percent of measured values below the limit of detection was 12.15%.

The bone densities of five areas of the proximal femur were measured using dual energy x-ray absorptiometry. Measurements were made on women who were not pregnant and men, 20 years or older. The femoral neck, trochanter, intertrochanter, Ward’s Triangle, and the total region were measured. Bone mineral density of the total region (g/cm²) was used in the present analysis. The missing values for pregnant women were set to zero so that they would not be excluded from the analysis.

The demographic variables that were used included race-ethnicity (DMARETHN), sex (HSSEX), and age at interview (HSAGEIR). Ages of 90 years old or above were coded as 90 years old.

A variable (MXPSESSR) indicating the period of day that a person was examined at a mobile examination center was used. Examinations were scheduled for morning, afternoon, and evening sessions.

Body mass index (BMPBMI) was calculated from the measurements of the weight in kilograms (BMPWT) and the standing height in centimeters (BMPHT).

The variable used to indicate pregnancy (MAPF12R) was from the Mobile Examination Center Adult Questionnaire that was administered to all persons, 17 years or older, that were examined at a mobile examination center. Positive responses were coded as ‘1’. ‘Blank but applicable’ and ‘Don’t know’ were coded as missing. All other values were coded as ‘0’.

Two questions were used to determine menopausal status: ‘Have you had a period in the past 12 months?’ (MAPF21) and ‘Were both ovaries removed or only one?’ (MAPF28). From these variables, a variable called ‘menopause’ was created with three levels: no, yes (women who had not had a period in the last 12 months), and surgical (women with both ovaries removed). The levels of menopause are mutually exclusive.

One question was used to determine hormone pill use: ‘How many months ago did you stop taking estrogen or female hormone pills or are you still taking them?’ (MAPF39S), one question was used to determine hormone use via vaginal creams, suppositories, or injections: ‘How many months ago did you stop using the vaginal cream, suppository, or injection or are you still taking them?’ (MAPF43S), and one question was used to determine hormone patch use: ‘How many months ago did you stop using the hormone patches or are you still using them?’ (MAPF47S).

A variable (PHPFAST) indicating the number of hours since a person last ate or drank was used. This variable was from the Phlebotomy Screening Questionnaire.

Two questions from the Household Adult Questionnaire were used to indicate thyroid disease: ‘Has a doctor ever told you that you had goiter?’ (HAC1J) and ‘Has a doctor ever told you that you had other thyroid disease?’ (HAC1K).

Variables (HAX9) which give the drug class codes of medications a person reports currently taking were used to determine thyroid or antithyroid drug use. From these variables, a variable called ‘thyroid drug’ was created with two levels: no and yes (a person reports currently taking one or more thyroid or antithyroid drugs).

The computer program SAS^® (Release 9.4, SAS Institute, Inc., Cary, North Carolina) was used to analyze the survey data. Survey procedures were used that took into account the complex survey design. Two design variables were used, pseudo-PSU (SDPPSU6) and pseudo-stratum (SDPSTRA6). The total MEC-examined sample final weight (WTPFEX6) was used. Analytical guidelines for NHANES III are available [8,9].

Linear regression analysis was performed between the log₁₀ serum thyroid stimulating hormone concentration and the log₁₀ blood lead concentration. This model also included race-ethnicity, sex, age, session, body mass index, pregnant, menopause, hormone pill use, vaginal cream use, hormone patch use, urinary creatinine, log₁₀ urinary iodine, log₁₀ serum cotinine, serum thyroxine, log₁₀ serum antimicrosomal antibody, log₁₀ serum anti-thyroglobulin antibody, total bone mineral density, length of fast, goiter, other thyroid disease, and thyroid drug as covariates, as well as the sex × pregnant × log₁₀ blood lead concentration interaction.

Linear regression analysis was also performed between serum thyroxine and the log₁₀ blood lead concentration. This model also included race-ethnicity, sex, age, session, body mass index, pregnant, menopause, hormone pill use, vaginal cream use, hormone patch use, urinary creatinine, log₁₀ urinary iodine, log₁₀ serum cotinine, log₁₀ serum antimicrosomal antibody, log₁₀ serum anti-thyroglobulin antibody, total bone mineral density, length of fast, goiter, other thyroid disease, and thyroid drug as covariates, as well as the sex × pregnant × log₁₀ blood lead concentration interaction.

In order to determine the lowest concentrations of blood lead at which a relationship could be detected, two variables for blood lead were created. If the lead value was less than or equal to a cutoff value, the first lead variable was assigned the value and the second lead variable was set to zero. If the lead value was greater than the cutoff value, the first lead variable was set to zero and the second lead variable was assigned the value. The cutoff value varied from the next to lowest value of blood lead to the highest value and was incremented in steps of 0.1. The slopes of the two lead variables were calculated at each step. The lowest blood lead value at which a relationship could be detected was defined as the cutoff value for which the slope of the first lead variable was statistically significant (p < 0.05) at the cutoff value and all subsequent greater values. Covariates were included in the models, and separate estimates were made for each level of sex by including interaction terms with sex.

SigmaPlot^® (Version 11.0, Systat Software, Inc., San Jose, California) was used to make the graph. Simple linear least squares regression was used to fit the data in the graphs.

No relationship was found between the blood lead level and the concentration of serum thyroid stimulating hormone in men or women.

In previous environmental studies, Abdelouahab et al. [4] found an inverse relationship between the concentrations of serum thyroid stimulating hormone and blood lead in women, but not in men, and Meeker et al. [5] found an inverse relationship in men. The median blood lead levels in these studies (1.74, 3.10, and 1.5 μg/dL) are similar to the mean blood lead level in the present study (3.55 μg/dL).

Abdelouahab et al. [4] included age, smoking status, selenium, total plasma lipid concentration, and oestro-progestative hormone intake as covariates for women. The men and women were analyzed in separate models. The age of the 87 women ranged from 18 to 70 years old. Concentrations of lead were log transformed. The concentrations of thyroid hormones were not.

Meeker et al. [5] included age, body mass index, and current smoking. The age of the 219 men ranged from 18 to 55 years old. Concentrations of lead were log transformed and grouped into quartiles. The concentrations of thyroid stimulating hormone were log transformed.

In the present analysis, removing all the covariates from the model for thyroid stimulating hormone did not result in statistically significant regression coefficients for blood lead for men (b = 0.00730, p = 0.7132), non-pregnant women (b = 0.01899, p = 0.3186), or pregnant women (b = −0.02726, p = 0.8159), so adjustment by the covariates was not responsible for the lack of statistical significance.

The concentration of serum total thyroxine decreased as the blood lead level increased in non-pregnant and pregnant women, but not in men. The decrease was greatest in the pregnant women.

In a previous environmental study, Abdelouahab et al. [4] found no relationship between the concentrations of serum total thyroxine and blood lead in women or men. The regression coefficients for men (b = −1.93, p > 0.10) and women (b = −0.36, p > 0.10) were negative but not statistically significant. It is possible the sample size was not large enough to detect a relationship.

Two sex specific hypotheses can be made about the decreases in total thyroxine seen in the present study. First, lead could inhibit the sialylation of thyroid binding globulin by estrogens and reduce the concentration of thyroid binding globulin and thus total thyroxine in the blood. Thyroid binding globulin is a carrier protein that binds to thyroxine [10]. Estrogens increase the concentration of thyroid binding globulin and thyroxine [11]. Estrogens increase the amount of thyroid binding globulin that is sialylated [12]. Sialylated thyroid biding globulin has a lower clearance rate in the liver than unsialylated thyroid binding globulin [13].

Secondly, during pregnancy lead could also inhibit the sialylation of human chorionic gonadotropin which would lead to a decrease in the production of thyroxine by the thyroid. Desialylated forms of human chorionic gonadotropin have a higher affinity for the thyroid stimulating hormone receptors in thyroid membranes [14]. A desialylated form of human chorionic gonadotropin has been shown to be an antagonist of adenylate cyclase activity induced by thyroid stimulating hormone [15].

Covariates were chosen based on previous analyses of thyroid hormone data, especially, Hollowell et al. [16] and Blount et al. [17]. Urinary creatinine was included in the models to adjust for urine volume for urinary iodine. This method of adjustment is recommended by Barr et al. [18]. Serum total thyroxine was included as a covariate for serum thyroid stimulating hormone to control for an indirect effect of lead via the negative feedback of thyroxine [19].

Total bone mineral density was included as a covariate to control for the relationships between thyroid hormones and blood lead level due to effect of thyroid hormones on bone metabolisim. Thyroid hormones are necessary for bone growth and maintenance [20]. Thyroid stimulating hormone is positively related to bone mineral density in men [21] and women [22], and hyperthyroidism is associated with lower bone mineral density and a greater risk of fractures [23].

Lead is stored in bone [24] and can alter bone cell function [25], and blood lead levels are related to bone metabolism. In women 40 to 59 years old from NHANES III, there was an inverse relationship between blood lead level and bone mineral density [26]. Biological markers of bone turnover were positively related to blood lead levels in a study of Japanese female farmers [27].

The method for determining the lowest concentration was a modification of the method used by Lanphear, Dietrich, Auinger, and Cox [28]. They calculated regression coefficients between blood lead concentrations and cognitive performance in children with blood lead concentrations at < 10, < 7.5, < 5.0, and < 2.5 μg/dL, and determined the lowest subset in which the relationships were still statistically significant (< 5 μg/dL). Children were successively dropped out of the model at each step.

The modified method used here incremented the blood lead concentration by steps of 0.1 μg/dL to get a more precise estimate of the concentration at which the relationship becomes consistently statistically significant. Instead of dropping subjects out of the model, their blood lead levels are placed in a second blood lead variable so that the sample size remains constant.

For serum total thyroxine, the lowest concentration of blood lead at which a relationship could be detected was 2.1 μg/dL for the non-pregnant women and 3.9 μg/dL for the pregnant women. The current elevated blood lead level for adults as defined by the Centers for Disease Control and Prevention is ≥ 5 μg/dL (https://www.cdc.gov/niosh/topics/ables/default.html).

In terms of clinical significance, from Table 4, for the non-pregnant women a 10 times increase in blood lead, e.g., from 1 to 10 μg/dL, would result in a 0.5 μg/dL decrease in serum total thyroxine. For the pregnant women, a 10 times increase would result in a 2 μg/dL decrease in serum total thyroxine. The normal range for total thyroxine is 4.5 to 12.5 μg/dL (http://globalrph.com/labs/t/), so a 10 times increase would probably not be clinically relevant for a non-pregnant women. For a pregnant women, a 10 times increase could potentially take a woman below the normal range or into the normal range depending or what her initial level was.