.
Univariate and multivariate analysis
Variables that were identified to be significantly associated with serum PSA levels in univariate analysis included lead, cadmium, HDL, age, hypertension history, body mass index, DMDEDUC2, DMDMARTL3, enlarged prostate, coronary heart disease, stroke, PAQ180, smoked at least 100 cigarettes during lifetime, and zinc intake on the first day. The test results for these significant variables were 1.14 ng/mL (1.15, 95% CI, P < 0.0001), 1.05 ng/mL (1.05, 95% CI, P < 0.0001), 0.01 (0.01, 95% CI, P < 0.0001), 0.04 ng/mL (95% CI, P < 0.0001), 0.23 (95% CI, P < 0.0001), 0.03 (95% CI, P < 0.0001), 1.03 (95% CI, P < 0.0001), 0.88 (95% CI, P < 0.0001), 0.84, (95% CI, P < 0.0001), 0.8 (95% CI, P < 0.0001), 0.59 (95% CI, P < 0.0001), 0.90 (95% CI, P < 0.0001), 0.28 (95% CI, P < 0.0001), 0.47 ( 95% CI, P < 0.0001), 0.23 (95% CI, P = 0.0012), 0.44 (95% CI, P < 0.0001), 0.10 (95% CI, P = 0.0047), and 0.14 (95% CI, P < 0.0001), respectively.
Multivariate analysis indicated that body mass index (0.02, 95% CI, P < 0.0001), DMDMARTL3 (0.33, 95% CI, P < 0.0001), enlarged prostate (0.59, 95% CI, P < 0.0001), and PAQ180 (1.54, 95% CI, P < 0.0001) were negatively associated with serum PSA levels, whereas lead exposure (0.08, 95% CI, P = 0.0054) and age (0.03, 95% CI, P < 0.0001) were positively associated with serum PSA levels.
The analyses indicated that lead exposure, body mass index, DMDMART (single), enlarged prostate, and age were significantly associated with elevated serum PSA levels, with only lead exposure having a 10% or more effect size. The results also suggested that cadmium, mercury exposure, or zinc intake was irrelevant to changes in serum PSA levels.
Table 2
Univariate and multivariate analysis.
Exposure | Univariate | Multivariate |
Mercury, total (umol/L) log2 transform | 0.01 (-0.02, 0.03) 0.5889 | -0.00 (-0.03, 0.03) 0.8655 |
Lead (umol/L) log2 transform | 0.20 (0.16, 0.23) < 0.0001 | 0.08 (0.03, 0.14) 0.0054 |
Cadmium (nmol/L) log2 transform | 0.07 (0.04, 0.10) < 0.0001 | 0.02 (-0.02, 0.06) 0.4161 |
Diabetes history | | |
1 | 0 | 0 |
2 | 0.02 (-0.06, 0.11) 0.5932 | 0.08 (-0.04, 0.21) 0.2082 |
Hypertension history | | |
1 | 0 | 0 |
2 | -0.23 (-0.30, -0.17) < 0.0001 | -0.03 (-0.12, 0.07) 0.5847 |
Body mass index,Kg/m2 | -0.03 (-0.03, -0.02) < 0.0001 | -0.02 (-0.03, -0.01) < 0.0001 |
DMDEDUC2.NEW NEW | | |
1 | 0 | 0 |
2 | -0.19 (-0.29, -0.10) < 0.0001 | -0.05 (-0.17, 0.08) 0.4867 |
3 | -0.25 (-0.34, -0.16) < 0.0001 | -0.15 (-0.28, -0.01) 0.0310 |
DMDMARTL.NEW NEW | | |
1 | 0 | 0 |
2 | 0.07 (-0.00, 0.15) 0.0601 | -0.04 (-0.14, 0.06) 0.4439 |
3 | -0.33 (-0.48, -0.18) < 0.0001 | -0.34 (-0.54, -0.13) 0.0011 |
Alcohol (gm) first day | -0.00 (-0.00, -0.00) 0.0176 | -0.00 (-0.00, 0.00) 0.9762 |
HDL | 0.01 (0.01, 0.01) < 0.0001 | 0.00 (-0.00, 0.01) 0.1418 |
Poverty income ratio | -0.02 (-0.04, -0.00) 0.0341 | 0.02 (-0.01, 0.05) 0.2171 |
Enlarged prostate | | |
1 | 0 | 0 |
2 | -0.59 (-0.68, -0.50) < 0.0001 | -0.37 (-0.49, -0.25) < 0.0001 |
7 | 2.97 (0.34, 5.60) 0.0267 | 0 |
9 | -0.90 (-1.32, -0.48) < 0.0001 | -0.65 (-1.27, -0.03) 0.0398 |
LBDLDL | -0.00 (-0.00, -0.00) < 0.0001 | 0.00 (-0.00, 0.00) 0.4904 |
C-reactive protein(mg/dL) | 0.03 (0.00, 0.07) 0.0465 | 0.02 (-0.02, 0.06) 0.4088 |
Glycohemoglobin (%) | -0.02 (-0.05, 0.00) 0.1050 | -0.00 (-0.04, 0.03) 0.8403 |
Triglycerides (mg/dL) | -0.00 (-0.00, -0.00) < 0.0001 | -0.00 (-0.00, -0.00) 0.0032 |
coronary heart disease | | |
1 | 0 | 0 |
2 | -0.28 (-0.40, -0.16) < 0.0001 | 0.01 (-0.15, 0.16) 0.9480 |
stroke | | |
1 | 0 | 0 |
2 | -0.47 (-0.59, -0.35) < 0.0001 | -0.15 (-0.30, -0.00) 0.0452 |
PAQ180 | | |
1 | 0 | 0 |
2 | -0.05 (-0.16, 0.05) 0.3225 | 0.03 (-0.07, 0.14) 0.5191 |
3 | -0.23 (-0.37, -0.09) 0.0012 | -0.09 (-0.22, 0.05) 0.2155 |
4 | -0.44 (-0.62, -0.27) < 0.0001 | -0.06 (-0.24, 0.12) 0.5022 |
7 | -0.64 (-3.32, 2.05) 0.6429 | -0.85 (-3.36, 1.67) 0.5104 |
9 | -1.54 (-2.89, -0.20) 0.0247 | -1.84 (-3.11, -0.58) 0.0043 |
Age,year | 0.04 (0.03, 0.04) < 0.0001 | 0.03 (0.02, 0.03) < 0.0001 |
Smoked at least 100 cigarettes in life | | |
1 | 0 | 0 |
2 | -0.10 (-0.16, -0.03) 0.0047 | 0.03 (-0.06, 0.13) 0.4768 |
Zinc (mg) first day log2 transform | -0.14 (-0.17, -0.10) < 0.0001 | 0.00 (-0.05, 0.05) 0.9102 |
VITD | 0.00 (0.00, 0.00) 0.0180 | -0.00 (-0.00, 0.00) 0.3987 |
Adjusted model
A non-adjusted model and a fully adjusted model were attempted to analyze the associations between serum PSA levels with lead, mercury, cadmium exposure, and zinc intake (Table 3). After adjustment of the variables, including hypertension history, body mass index (kg/m2), diabetes history, DMDEDUC, DMDMARTL, alcohol (gm) first day, HDL, poverty income ratio, enlarged prostate, LBDLDL, C-reactive protein (mg/dL), glycohemoglobin (%), triglycerides (mg/dL), coronary heart disease, stroke, PAQ180, age (year), race/ethnicity, smoked at least 100 cigarettes during lifetime, VITD, and zinc (mg) first day, the effect size for lead exposure was reduced by nearly 60% from 0.196 (multivariate analysis) to 0.079 (P = 0.00752). The other three variables (mercury exposure, cadmium exposure, and zinc intake) were not significantly associated with changes in serum PSA levels.
GAM
In the four-subgroup data (Q1–Q4), we observed a non-equidistant change of regression coefficients when linear regression models were used to analyze the relationship between PSA levels and the four variables of interest, i.e., lead exposure, cadmium exposure, mercury exposure, and zinc intake. Therefore, GAM was used to identify the non-linear relationship between these four variables and PSA levels. The advantage of this model is that it allows other variables to be adjusted using a function, which are then included in a regression model. If a non-linear relationship is identified, an inflection point is calculated using a recursive algorithm, followed by an analysis with a weighted linear regression model to integrate data points on both sides of the inflection point. Eventually, a simple linear regression model or a piecewise linear regression model can be established based on the relationship between the logarithmic likelihood ratio with 0.01 selected as a cutoff. In GAM analyses, the variables that had been adjusted included hypertension history, body mass index (kg/m2) (smooth), diabetes history, DMDEDUC, DMDMARTL, alcohol (gm) first day (smooth), HDL (smooth), poverty income ratio, enlarged prostate, LBDLDL (smooth), C-reactive protein (mg/dL) (smooth), glycohemoglobin (%) (smooth), triglycerides (mg/dL) (smooth), coronary heart disease, stroke, PAQ180, age (year) (smooth), race/ethnicity, smoked at least 100 cigarettes during lifetime, VITD (smooth), and zinc (mg) first day (smooth). The results showed that only lead exposure was significantly associated with changes in serum PSA levels after the adjustment. Univariate analysis showed a significant association between zinc intake and serum PSA levels, whereas multivariate analysis did not confirm this association; however, when zinc was adjusted in the GAM analysis, a significant association between lead exposure and serum PSA levels was observed. The sum of the evidence indicated that the zinc in blood influences serum PSA levels in an indirect manner, which is also supported by the fact that zinc does not directly interfere with PSA pathways.
As shown in Table 3, the logarithmic likelihood ratios for lead, cadmium, and mercury calculated using the GAM model were 0.168, 0.059, and 0.399, respectively, which were all greater than the threshold of 0.01 and suggested a piecewise linear regression model. The analysis with the piecewise linear regression model indicated that when the cadmium concentration was less than 6.06 µmol/L, an increase of cadmium by a unit led to a slight increase in PSA levels by 0.08 units. Such a small effect is equivalent to a 2% change compared with the standard used in clinical practice for diagnosis, i.e., PSA at 4 ng/mL. There was no such association when cadmium concentrations were greater than 6.06 µmol/L. No association was identified between serum PSA levels and lead or mercury.
Table 3
Adjusted model and generalized additive model.
Exposure | Non-adjusted model | Fully-adjusted model | GAM model |
(β, 95%CI, P value) | (β, 95%CI, P value) | (β, 95%CI, P value) |
Lead (umol/L) log2 transform | 0.196 (0.158, 0.234) < 0.00001 | 0.079 (0.021, 0.136) 0.00752 | 0.070 (0.013, 0.128) 0.01718 |
Q1(-6.15 to -3.86) | 0 | 0 | 0 |
Q2 (-3.84 to -3.32) | 0.278 (0.186, 0.371) < 0.00001 | 0.164 (0.035, 0.293) 0.01258 | 0.143 (0.014, 0.272) 0.02969 |
Q3 (-3.31 to -2.75) | 0.321 (0.229, 0.413) < 0.00001 | 0.152 (0.024, 0.281) 0.02014 | 0.132 (0.003, 0.260) 0.04431 |
Q4 (-2.74 to -0.67) | 0.431 (0.339, 0.523) < 0.00001 | 0.142 (0.009, 0.275) 0.03679 | 0.115 (-0.020, 0.249) 0.09426 |
P for trend | 0.133 (0.104, 0.162) < 0.00001 | 0.037 (-0.005, 0.079) 0.08412 | 0.029 (-0.013, 0.072) 0.17701 |
Cadmium (nmol/L) log2 transform | 0.073 (0.045, 0.101) < 0.00001 | 0.035 (-0.005, 0.075) 0.08658 | 0.027 (-0.014, 0.067) 0.19236 |
Q1 (-0.17 to 0.97) | 0 | 0 | 0 |
Q2 (1.03 to 1.72) | 0.234 (0.141, 0.326) < 0.00001 | 0.040 (-0.086, 0.166) 0.53269 | 0.031 (-0.094, 0.156) 0.62969 |
Q3 (1.76 to 2.62) | 0.404 (0.312, 0.496) < 0.00001 | 0.165 (0.039, 0.290) 0.01035 | 0.145 (0.020, 0.270) 0.02351 |
Q4 (2.63 to 6.29) | 0.283 (0.191, 0.376) < 0.00001 | 0.124 (-0.007, 0.255) 0.06269 | 0.092 (-0.040, 0.224) 0.17315 |
P for trend | < 0.00001 | 0.02186 | 0.07494 |
Mercury, total (umol/L) log2 transform | 0.006 (-0.016, 0.028) 0.58892 | -0.010 (-0.040, 0.021) 0.53175 | -0.008 (-0.039, 0.023) 0.60519 |
Q1 (-1.03 to 1.29) | 0 | 0 | 0 |
Q2 (1.32–2.30) | -0.032 (-0.126, 0.062) 0.50588 | -0.000 (-0.124, 0.124) 0.99601 | -0.001 (-0.125, 0.122) 0.98525 |
Q3(2.31 to 3.31) | 0.014 (-0.080, 0.109) 0.76542 | -0.021 (-0.146, 0.105) 0.74448 | -0.019 (-0.144, 0.106) 0.76646 |
Q4 (3.32 to 8.74) | 0.028 (-0.067, 0.122) 0.56739 | -0.038 (-0.169, 0.092) 0.56309 | -0.043 (-0.173, 0.086) 0.51268 |
P for trend | 0.36689 | 0.50872 | 0.47009 |
Zinc (mg) first day log2 transform | -0.136 (-0.174, -0.097) < 0.00001 | -0.016 (-0.090, 0.058) 0.67581 | 0.042 (-0.154, 0.239) 0.67255 |
low | 0 | 0 | 0 |
medium | -0.139 (-0.221, -0.058) 0.00079 | -0.071 (-0.178, 0.035) 0.19049 | -0.091 (-0.205, 0.023) 0.11889 |
high | -0.325 (-0.407, -0.244) < 0.00001 | -0.112 (-0.239, 0.015) 0.08366 | -0.176 (-0.340, -0.013) 0.03465 |
P for trend | <0.00001 | 0.07877 | 0.08324 |
Non-adjusted model adjusts for: None |
Fully-adjusted model adjust for: Hypertension history, body mass index (kg/m2), diabetes history, DMDEDUC, DMDMARTL, alcohol (gm) first day, HDL, poverty income ratio, enlarged prostate, LBDLDL, C-reactive protein (mg/dL), glycohemoglobin (%), triglycerides (mg/dL), coronary heart disease, stroke, PAQ180, age (year), race/ethnicity, smoked at least 100 cigarettes during lifetime, VITD, zinc (mg) first day. |
GAM model was adjusted for: Hypertension history, body mass index (kg/m2) (smooth), diabetes history, DMDEDUC2.NEW NEW, DMDMARTL.NEW NEW, alcohol (gm) first day (smooth), HDL (smooth), poverty income ratio, enlarged prostate, LBDLDL (Smooth), C-reactive protein (mg/dL) (smooth), glycohemoglobin (%) (smooth), triglycerides (mg/dL) (smooth), coronary heart disease, stroke, PAQ180, age (year) (smooth), race/ethnicity, smoked at least 100 cigarettes during lifetime, VITD (smooth), zinc (mg) first day (smooth). Generalized additive models were applied. |
Subgroup analysis
Univariate analyses showed that lead, cadmium, and zinc were all significantly associated with changes in serum PSA levels. Multivariate analyses, using a fully adjusted model, and GAM analysis only detected an association between lead and serum PSA. Nevertheless, when zinc was treated as a variable being adjusted in multivariate analysis, the fully adjusted model, or GAM, associations between serum PSA and lead or cadmium could be detected. Thus, we divided the sample into three strata (T1: 0.19–8.69 (low), T2: 8.75–14.03 (median), and T3: 14.12–315.17 (high)) based on zinc intake, and analyzed each of these three strata separately using various models including a non-adjusted model, an adjusted model I, and an adjusted model II (Table 4). The results showed that only in the T3 (high level of zinc intake) stratum, exposure to cadmium or lead could increase serum PSA levels: an increase of cadmium by 1 unit led to an increase in PSA by 1.06 unit, while an increase in lead by 1 unit yielded an increase in PSA by 1.09 units. Such effects are equivalent to 25% of changes compared with the standard for diagnosis, which is quite significant in clinical practice. No such associations were detected in the T1 (low level of zinc intake) or T2 (median level of zinc intake) strata. These results suggest that increased zinc intake may interfere with serum cadmium to increase serum PSA levels in a synergistic manner.
Table 4
Exposure | Non-adjusted model | Adjust model I | Adjust model II |
Zinc (mg) first day log2 transform | -0.136 (-0.174, -0.097) < 0.00001 | -0.016 (-0.090, 0.058) 0.67581 | 0.042 (-0.154, 0.239) 0.67255 |
Zinc (mg) first day log2 transform | | | |
T1 | 0 | 0 | 0 |
T2 | -0.139 (-0.221, -0.058) 0.00079 | -0.071 (-0.178, 0.035) 0.19049 | -0.091 (-0.205, 0.023) 0.11889 |
T3 | -0.325 (-0.407, -0.244) < 0.00001 | -0.112 (-0.239, 0.015) 0.08366 | -0.176 (-0.340, -0.013) 0.03465 |
P for trend | <0.00001 | 0.07877 | 0.08324 |
Non-adjusted model adjusts for: none. |
Adjusted model I adjusts for: demography univariate.
Adjusted model II adjusts for: all univariate in Table I.
Table 5
Interaction relationships.
Model | DR1TZINC.T3: LOW | DR1TZINC.T3: MEDIUM | DR1TZINC.T3: HIGH | P interaction |
Mercury, total (umol/L) log2 transform | | | | |
Model | -0.01 (-0.06, 0.04) 0.6586 | 0.01 (-0.05, 0.06) 0.7373 | -0.04 (-0.09, 0.02) 0.1873 | 0.4921 |
Lead (umol/L) log2 transform | | | | |
Model II* | 0.06 (-0.04, 0.16) 0.2390 | 0.04 (-0.06, 0.15) 0.4020 | 0.12 (0.03, 0.22) 0.0122 | 0.5056 |
Cadmium (nmol/L) log2 transform | | | | |
Model II* | -0.03 (-0.10, 0.03) 0.3253 | 0.07 (-0.01, 0.14) 0.0748 | 0.08 (0.01, 0.14) 0.0268 | 0.0249 |