Baseline characteristics of the study population (N=5224) are summarized in Table 1, stratified by sex. Characteristics for all three NHANES cycles utilized, including missing units, are summarized in Supplementary Table 1, and the workflow for data cleaning and sample assembly is shown in Figure 1. From the initial sample, we first omit children under the age of 10 (n = 7142), leaving a sample size of 22258. We then omit participants lacking urinary glyphosate (74% missing), creatinine (6.7% missing), BMI (5.5% missing) or poverty-income ratio information (10.7% missing). The degree of missing units in the exposure makes imputation of the exposure for inference purposes impossible. However, a comparison of summary information from the complete case sample (Table 1) and the original dataset (Supplementary Table 1) suggests the complete case sample is representative. The sample is roughly balanced by sex, with 2605 males and 2619 females. The largest single ethnic group is Non-Hispanic White (about 38%), followed by Non-Hispanic Black (about 20%). Hispanics as a group (Mexican American and ‘Other Hispanic’) comprised about 25%. Non-Hispanic Asians comprised about 22% and multiracial participants less than 10%. The cohort was older, with 52% of participants age 40 or over, and 26% age 60 or over. Covariates used in EFA (N=22,258) to create the MetS score were imputed using a conservative donor-based single-imputation method before proceeding with EFA (see Figure 1 for details). Features used in the EFA to create the MetS score are summarized in Table 2.
Table 2. Covariates used to create Metabolic Syndrome (MetS) score (N=22,258).
Feature1
|
Min
|
25%
|
50%
|
75%
|
Max
|
Waist circumference
|
49.5
|
81.8
|
94.5
|
107
|
177.9
|
Systolic blood pressure
|
64.67
|
108
|
117.33
|
130.67
|
231.33
|
Diastolic blood pressure
|
0
|
60
|
68
|
76
|
135.33
|
Glycoheme (HbA1c)
|
3.5
|
5.2
|
5.5
|
5.8
|
17.5
|
Fasting Glucose
|
21
|
93
|
100
|
109
|
479
|
HDL
|
6
|
42
|
51
|
62
|
226
|
Triglycerides
|
10
|
57
|
85
|
129
|
4233
|
Triglyceride : HDL Ratio
|
0.13
|
1.03
|
1.65
|
2.69
|
103.24
|
Mean Arterial Pressure (MAP)
|
71.56
|
122.89
|
134.44
|
150.22
|
279.56
|
1Variables and composite variables (the latter shown below double line at bottom of table) used in score are in boldface. MAP = SBP + 1/3 (SBP - DBP). MetS score is estimated on as many samples as possible to create a score broadly applicable to a general population. Missingness is summarized in Supplementary Table 2.
We explored associations between the outcome and exposure, plus patterns of associations between both the outcome and the exposure with covariates identified as confounders or potential effect modifiers (age, sex, race). Exposure and outcome are summarized by Quartile in Table 3A-C. Unadjusted associations between the standardized MetS outcome score and glyphosate, categorized by quartile, are significant for the reference quartile, and quartiles 3 (Q3) and 4 (Q4), with a suggestion of an inverted U-shaped dose-response (maximum at Q3; Table 3A).
Figure 1. Work flow and sample sizes showing sample size evolution from the initial raw sample (three NHANES cycles, 2013-2018) to final complete case sample with the created MetS score.
Table 3. Bivariate analysis. Note that every level value is relative to reference.
a. Bivariate association between MetS scale outcome and glyphosate by exposure quartile.
Glyphosate
|
Estimate
|
95% CI
|
p-value
|
Quartile 1 (ref)
|
-0.072
|
-0.126 ,-0.018
|
-
|
Quartile 2
|
0.071
|
-0.005, 0.148
|
0.069
|
Quartile 3
|
0.127
|
0.051, 0.204
|
0.001
|
Quartile 4
|
0.090
|
0.014, 0.167
|
0.021
|
b. Exposure and outcome by Race/ethnicity.
Log(Glyphosate) exposure by Race
|
|
Race/Ethnicity
|
Estimate
|
95% CI
|
p-value
|
Pr(>F)
|
White
|
0.530
|
0.504, 0.555
|
-
|
F = 8.35
p < 0.0001
|
Mex - American
|
-0.076
|
-0.123, -0.029
|
0.002
|
Other Hispanic
|
-0.030
|
-0.086, 0.026
|
0.291
|
Black
|
0.038
|
-0.005, 0.081
|
0.083
|
Asian
|
-0.110
|
-0.164, -0.055
|
0.0001
|
Multiracial
|
0.037
|
-0.039, 0.114
|
0.33
|
Standardized MetS scale (outcome) by Race
|
|
White
|
-0.040
|
-0.084 , 0.004
|
-
|
F = 19.0
p < 0.0001
|
Mex American
|
0.089
|
0.008, 0.170
|
0.030
|
Other Hispanic
|
0.083
|
-0.013, 0.180
|
0.090
|
Black
|
0.112
|
0.038, 0.187
|
0.003
|
Asian
|
-0.063
|
-0.156, 0.031
|
0.188
|
Multiracial
|
0.018
|
-0.114, 0.149
|
0.792
|
c. Exposure and Outcome by Age Category
Log(Glyphosate) exposure by Age Grouping
|
|
Age Category
|
Estimate
|
95% CI
|
p-value
|
Pr(>F)
|
10-19yrs
|
0.586
|
0.553, 0.62
|
-
|
F = 22.6
p <0.0001
|
20-39yrs
|
-0.136
|
-0.182, -0.09
|
<0.0001
|
40-59yrs
|
-0.125
|
-0.170, -0.08
|
<0.0001
|
>60yrs
|
-0.025
|
-0.070, 0.02
|
0.281
|
Standardized MetS scale (outcome) by Age Grouping
|
|
10-19yrs
|
-0.695
|
-0.745, -0.645
|
-
|
F=4113
p < 0.0001
|
20-39yrs
|
0.385
|
0.317, 0.454
|
<0.0001
|
40-59yrs
|
0.966
|
0.898, 1.034
|
<0.0001
|
>60yrs
|
1.308
|
1.240, 1.376
|
<0.0001
|
d. Exposure and outcome by Sex
Glyphosate exposure by Sex
|
Sex
|
Estimate
|
95% CI
|
p-value
|
Male (ref)
|
0.546
|
0.524, 0.568
|
-
|
Female
|
-0.067
|
-0.099, -0.036
|
<0.0001
|
Standardized MetS Scale (outcome) by Sex
|
Male (ref)
|
0.070
|
0.031, 0.108
|
-
|
Female
|
-0.139
|
-0.193, -0.085
|
<0.0001
|
We find two race-ethnicity levels (Mexican-American and NH Asian) with significantly differe t mean estimates from the reference (White) for log-transformed glyphosate from regression (Table 3A; p < 0.0001). For the outcome MetS score, Mexican-American and Black participants have significantly higher scores than reference, and the result from ANOVA is also highly significant (p < 0.0001) (Table 3B). A visualization of differences in exposure by race is shown in Supplementary Figure 2A.
Unadjusted regressions of exposure and outcome with age-category are consistent, showing significant differences in exposure between the reference level (ages 10-19) and ages 20-39 and 40-59 years, which are slightly reduced relative to the reference (p = 0.004). Estimates for the age 60 and above level are not significantly different than the reference level, interestingly (Supplementary Figure 2B). The MetS score is significantly different across all age categories, which is unsurprising (Table 3C). Finally, there are significantly different exposure and outcome levels by sex (reference Male); women have on average slightly but significantly lower exposure levels (p <0.0001) and slightly lower MetS scores than men (p < 0.0001) in unadjusted regressions (Table 3D). Bivariate regressions show a surprising result for age category, in that the highest exposures are among the ages 10 to 19 group and ages above 65, consistent with exposure results from the 2013-14 cycle only by Ospina et al. (2022). Results for MetS score are as expected, not much difference across race/ethnicity, but the score rises with age, and is slightly higher among men than women.
Exploratory Factor Analysis resulted in all positive weightings and yielded a single index score. The weightings are shown in a factor analysis diagram (Figure 2): glyphosate and glucose dominate, but just slightly. Biometric measurements waist circumference and MAP are similarly weighted and triglyceride : HDL ratio the least, though well within the range of inclusion.
Summary statistics for the index are listed by quartile in Table 1 and for the component covariates, Table 2. Missing units for the component covariates are summarized in Supplemental Table 2.
As demonstrated in Cavero-Redondo et al. (2019), we validate our score by using it to predict vascular damage (associated with MetS) according to albumin-to-creatinine ratio (e.g., ACR
); a typical, though not alone sufficient, metric of clinical performance for a MetS score (Shi et al 2020, Saadi et al. 2020). Results from the validation procedure suggest the sensitivity, specificity and accuracy increase with the ACS cutoff (Supplemental Table 3), with an error rate of 12% for an ACR cut-off of 30 and above (microalbuminaria), and 2% for an ACR cut-off of 300 and above (albuminuria). Again, because microalbuminaria is associated with MetS but not completely predictive of it, this is not an ideal test.
Figure 2. Factor Analysis Diagram. Variable loadings for the MetS index created via factor analysis on hot-deck imputed risk factor metrics (Table 2).
Adjusted associations between standardized metabolic score and glyphosate categorized by quartile, from regressions adjusted for scaled BMI, transformed and standardized creatinine, sex (reference Male), age category (reference 10-19 years), race-ethnicity (reference White) and standardized income to poverty ratio are shown in Figure 3 and Table 4. Results for adjusted quartile models are significant and stronger than unadjusted results (Table 3A) and show the same slight inverted U-shaped dose-response, with peak estimates at the third quartile. In the adjusted model, score estimates increase with age and related estimates are slightly elevated over reference and over other races for Asian and multiracial participants. This reverses in stratified models. Female sex is protective.
Figure 3. Change in MetS score by Quartile Glyphosate, full model with all covariates. covariates (N = 5224). Model is adjusted as shown by sex (reference Male), age category (reference 10-19 years), race-ethnicity (reference Non-Hispanic White), scaled BMI, standardized creatinine, and standardized income-poverty ratio. See Table 4 for estimates, confidence intervals, and p-values.
Table 4. Associations between standardized Metabolic Score, glyphosate exposure and covariates (N = 5224). Models are adjusted by sex (ref. Male), age category (ref. 10-19 years), race-ethnicity ( ref. Non-Hispanic White), scaled BMI, standardized creatinine, and standardized income-poverty ratio.
Covariate
|
Level/Reference
|
Estimate
|
95% CI
|
p-value
|
(Intercept)
|
|
-0.604
|
-0.681, -0.527
|
<0.0001
|
Glyphosate
Categorized by Quartile
Ref: Quartile 1
|
Quartile 2
|
0.092
|
0.029, 0.155
|
0.0043
|
Quartile 3
|
0.185
|
0.119, 0.250
|
<0.0001
|
Quartile 4
|
0.177
|
0.107, 0.248
|
<0.0001
|
BMI
|
(standardized)
|
0.352
|
0.328, 0.376
|
<0.0001
|
Creatinine
|
(sqrt, standardized)
|
-0.067
|
-0.094, -0.040
|
<0.0001
|
Sex (Female)
|
Reference Male
|
-0.233
|
-0.278, -0.188
|
<0.0001
|
Age Category
Ref: 10-19 years
|
20-39 years
|
0.180
|
0.114, 0.247
|
<0.0001
|
40-59 years
|
0.702
|
0.634, 0.770
|
<0.0001
|
>60 years
|
1.072
|
1.004, 1.139
|
<0.0001
|
Race/Ethnicity
Ref: Non-Hispanic White
|
Mexican American
|
0.175
|
0.108, 0.242
|
<0.0001
|
Other Hispanic
|
0.147
|
0.069, 0.224
|
0.0002
|
Black
|
0.135
|
0.074, 0.197
|
0.0002
|
Asian
|
0.193
|
0.117, 0.269
|
<0.0001
|
Multi
|
0.187
|
0.081, 0.293
|
0.00055
|
Income-Poverty Ratio
|
(standardized)
|
-0.049
|
-0.072, -0.027
|
0.00002
|
We explored the possibility of effect modification by age category, race-ethnicity and sex in a series of stratified, adjusted models. Data were stratified on the key variable and models run separately on each stratum. Results for models stratified on age category are shown in Supplemental Table 4. While there are significant differences in quartile estimates by age category (relative to reference level), confidence intervals overlap. MetS scores increase slightly, but inconsistently, with age (Figure 4). Among the 10 to 19 years cohort, only the fourth quartile is significantly different than reference (0.18, 95% CI: 0.035, 0.316, p = 0.015). For the age 20 to 39 cohort, scores rise significantly on quartile 2, drop and are marginal on Q3, and rise slightly to become significant on Q4. Cohorts age 40 and above show the inverted U-shaped dose-response profile. Results are strongest (p = 0.00005 at Q3) for participants 60 years and above. We do not correct for multiplicity.
While confidence intervals for the models stratified on race-ethnicity do overlap, there are two to three-fold differences in the effect sizes, relative to White participants, of estimates for Mexican-Americans (Q4 is 2.1 x greater than reference/White Q4), Other Hispanic (Q4 is 3.5 x larger) and Black participants (Q4 is a factor of 2.44 greater). Significant estimates for Mexican
Figure 4. Change in MetS score by Quartile Glyphosate Exposure from adjusted regression models stratified on age category. Reference level for age category in full model is ages 10-19 years (<19yrs). See supplemental table 3 for details including sample sizes and p-values. To simplify presentation, only associations for outcome and exposure by quartile are shown for each age grouping.
American and Black participants show the inverted U-shaped dose response profile we observe in the study population as a whole (Figure 5, Supplemental Table 5).
Finally, estimates for models stratified on sex show a significant and inverted U-shaped profile for females and a significant and increasing profile for males. Aside from the patterns of increase across levels, values are consistent between the sexes (Supplemental Table 6), and confidence intervals overlap, suggesting no evidence of effect modification by sex.
Figure 5. Change in MetS score by Quartile Glyphosate Exposure from adjusted regression models stratified on race/ethnicity. Reference level for race in full model is Non-Hispanic White. See supplemental Table 4 for details including sample sizes and p-values. To simplify presentation, only associations for outcome and exposure by quartile are shown for each grouping.