Satarug 2025 - Challenges in toxicological risk assessment of environmental cadmium exposure

Satarug, writing from the Centre for Kidney Disease Research at the Translational Research Institute in Brisbane, reviews the methodology and evidence base behind dietary cadmium exposure guidelines and argues that current threshold-based risk assessments substantially underestimate Cd nephrotoxicity. The review highlights two methodological problems: the JECFA 5.24 microg/g creatinine threshold (derived from a beta2-microglobulinuria endpoint at 300 microg/g creatinine) is too high to be protective for low-iron and low-zinc subpopulations, and the conventional practice of normalizing urinary Cd to creatinine excretion (Ecr) introduces non-differential errors that bias dose-response relationships toward the null. Normalizing to creatinine clearance (Ccr) instead recovers stronger associations between Cd exposure and low eGFR, proteinuria, and CKD. The review then summarizes benchmark dose (BMD) modeling results across six named population datasets and concludes that current dietary Cd guidelines do not adequately protect kidney function.

Key numbers

Dietary Cd exposure guidelines reviewed (Table 1, p.3)

The review tabulates published Cd dietary exposure guidelines based on kidney and bone endpoints.

Dietary Cd exposure guidelines from dosing experiments (Table 2, p.3-4)

Effect of normalization choice on Cd-eGFR association (Tables 3 and 4, p.8-10)

Table 3 (low eGFR per doubling of Cd excretion) is a reanalysis of data from 917 subjects (562 females, 355 males), aged 16-87 years, drawn from Satarug et al. 2024 [76]. Table 4 (low eGFR and proteinuria per 10-fold rise in Cd excretion) is a reanalysis of data from 405 subjects (208 females, 197 males), drawn from Satarug 2024 [80]. Low eGFR is defined as eGFR ⇐60 mL/min/1.73 m^2.

Table 3 — odds of low eGFR per doubling of Cd excretion under two normalization schemes.

Age categories were strong predictors in both models (POR 14.23 for 46-55 y, 141.2 for 66-87 y in Model A; 9.951 for 46-55 y, 198.6 for 66-87 y in Model B; all p⇐0.027). BMI >=24 was associated with POR 2.810 (p=0.028) in Model A and 3.150 (p=0.022) in Model B.

Table 4 — odds of low eGFR and proteinuria per 10-fold rise in Cd excretion.

The Ccr normalization (Model B) produced a 12-fold increase in low-eGFR risk and a 7-fold increase in proteinuria risk per 10-fold rise in Cd excretion, compared with 2.638-fold and 3.685-fold respectively under Ecr normalization (Model A).

PROAST BMD modeling results (Figures 3 and 4, p.11-12)

Source data: Satarug et al. 2024 [80]; total n=405.

Continuous BMD on ECd/Ecr vs total protein excretion (5% increase in protein excretion as BMR):

• Males (n=190): BMDL 0.0212 microg/g creatinine; BMDU 0.757 microg/g creatinine; BMDU/BMDL ratio 35.7
• Females (n=215): BMDL 0.0226 microg/g creatinine; BMDU 0.913 microg/g creatinine; BMDU/BMDL ratio 40.4
• Total (n=405): BMDL 0.0536 microg/g creatinine; BMDU 0.872 microg/g creatinine; BMDU/BMDL ratio 16.3
• Model weights: exponential 0.6840, Hill 0.2794, natural logarithmic 0.0386, inverse exponential 0.0017

Quantal BMD on ECd/Ecr vs 5% prevalence of proteinuria:

• Males (n=190): BMDL5 2.07 microg/g creatinine; BMDU5 5.96 microg/g creatinine; BMDU5/BMDL5 ratio 2.88
• Females (n=215): BMDL5 1.80 microg/g creatinine; BMDU5 5.98 microg/g creatinine; BMDU5/BMDL5 ratio 3.32
• Total (n=405): BMDL5 1.86 microg/g creatinine; BMDU5 5.72 microg/g creatinine; BMDU5/BMDL5 ratio 3.08
• Model weights: logarithmic probability 0.3501, Hill 0.1482, logarithmic logistic 0.1452, Weibull 0.1003, gamma 0.0935, exponential 0.0838, two-stage 0.0789

Quantal BMD on ECd/Ecr vs 5% prevalence of low eGFR:

• Males (n=190): BMDL5 1.26 microg/g creatinine; BMDU5 2.37 microg/g creatinine; BMDU5/BMDL5 ratio 1.88
• Females (n=215): BMDL5 1.26 microg/g creatinine; BMDU5 2.12 microg/g creatinine; BMDU5/BMDL5 ratio 1.68
• Total (n=405): BMDL5 1.19 microg/g creatinine; BMDU5 1.92 microg/g creatinine; BMDU5/BMDL5 ratio 1.61
• Model weights: exponential 0.8740, Hill 0.1231, others ⇐0.0017

BMD modeling across nephrotoxicity endpoints (Table 5, p.13)

BMD modeling on albuminuria and CKD prevalence (Table 6, p.14-15)

Data from 603 subjects (203 males, 400 females) [92].

Albuminuria defined as urinary albumin-to-creatinine ratio >=20 mg/g (men) or >=30 mg/g (women) for Ecr-normalized data; (EAlb/Ccr) x 100 >=20 mg/L filtrate (men) or >=30 mg/L filtrate (women) for Ccr-normalized data. CKD defined as eGFR ⇐60 mL/min/1.73 m^2.

The BMDU/BMDL ratios >=200 for ECd/Ecr in albuminuria mean those BMDL5 and BMDL10 values cannot be reliably defined; the Ccr-normalized BMDLs were tractable for both albuminuria and CKD. The author argues this strengthens the case for Ccr normalization.

Cd absorption and supporting epidemiology

• Cd absorption rate in young Japanese women with low iron intake and low iron status was 24-45 percent, compared with the JECFA-assumed 3-7 percent absorption in iron-replete subjects [45,46].
• Larsson 2015 [58]: risk of breast cancer increased 66 percent for each 0.5 microg/g creatinine increase in Cd excretion.
• Lin 2016 [59]: breast cancer risk elevated 2.24-fold in women with Cd excretion in the top quartile vs the lowest quartile.
• Schwartz 2003 [61]: risk of prediabetes and diabetes rose 48 percent and 24 percent respectively at Cd excretion rates of 1-2 microg/g creatinine after smoking and other confounders were adjusted.
• Wallia 2010 [62]: significant increase in prediabetes risk at Cd excretion rates >=0.7 microg/g creatinine.
• Jiang 2018 [63]: prediabetes risk increased 3.4-fold in obese US men with Cd excretion in the top quartile vs the bottom quartile.
• Kunioka 2022 [95]: meta-analysis of osteoporosis risk in Cd-exposed populations. Low-exposure group: 1.95-fold increased risk comparing Cd excretion >=0.5 vs <0.5 microg/g creatinine. High-exposure group: 1.99-fold comparing >=5 vs <5 microg/g creatinine.
• Xie 2024 [91]: prospective cohort from Switzerland (n=4704) linked Cd exposure to a rapid fall in eGFR.

The review’s principal numerical conclusion is that BMDL values for NAG and total protein excretion (0.060 and 0.054 microg/g creatinine respectively) are roughly 10-fold below the 0.5-0.6 microg/g creatinine mean Cd excretion observed in general populations of many countries, implying that nephrotoxicity already affects a significant fraction of those populations.

Methods (brief)

This is a narrative review by a single author. No new subjects were recruited and no new measurements were performed; the work summarizes published epidemiological datasets and BMD modeling results. BMD modeling exemplified in Section 4.3 used the PROAST software (https://proastweb.rivm.nl, accessed 22 March 2025) developed by the Dutch RIVM. The reanalyses tabulated in Tables 3 and 4 fit logistic regression and odds-ratio models to Cd excretion vs low-eGFR and proteinuria outcomes; mathematical dose-response models compared were the exponential, Hill, natural logarithmic, and inverse exponential models for continuous data and two-stage, logistic logarithmic, Weibull, logarithmic probability, gamma, exponential, and Hill models for quantal (prevalence) data. Model averaging used the Akaike information criterion (AIC) with 200 bootstrap repeats. The U.S. EPA Benchmark Dose Software (BMDS, https://www.epa.gov/bmds, accessed 22 March 2025) is also referenced as an alternative BMD modeling environment. BMR values of 5 percent and 10 percent for prevalence outcomes defined BMDL5/BMDU5 and BMDL10/BMDU10 respectively. The author acknowledges Aleksandra Buha Djordjevic for BMD modeling and Aleksandar Cirovic for figure design.

Endpoints used as kidney injury indicators include urinary beta2-microglobulin (beta2M, molecular weight 11,800 Daltons), N-acetyl-beta-D-glucosaminidase (NAG), retinol binding protein (RBP), albumin, total protein, and estimated glomerular filtration rate (eGFR). Cd excretion was expressed either as ECd/Ecr (urinary Cd normalized to creatinine excretion, microg/g creatinine) or as ECd/Ccr (urinary Cd normalized to creatinine clearance, microg/L glomerular filtrate). The author’s central methodological argument is that Ecr normalization introduces non-differential error that biases dose-response toward the null, while Ccr normalization recovers unbiased dose-response relationships.

The study is total Cd throughout; no Cd speciation is discussed. The review is single-authored and was not externally funded.

Implications

Certification: The review provides multiple lines of evidence that current JECFA (0.83 microg/kg bw/day, 5.24 microg/g creatinine threshold) and even EFSA (0.36 microg/kg bw/day, 1 microg/g creatinine threshold) dietary Cd guidelines do not protect against early nephrotoxicity, with BMDL values for total protein excretion as low as 0.054 microg/g creatinine when Ccr normalization is used. It contributes occurrence-of-effect data on tubular and glomerular endpoints to the Cd-exposure threshold case for HMTc product categories that drive dietary Cd burden (rice, leafy and root vegetables, cocoa products, organ meats). The 24-45 percent Cd absorption rate observed in iron-deficient women is a quantitative anchor for vulnerable-population Cd burden modeling. The author’s recommendation that Ccr should replace Ecr as the urine-normalization denominator is a methodological point with downstream implications for how future Cd cohorts in the HMI evidence base should be evaluated.

Courses: Useful as a recent (2025) reference for the BMD methodology section of any Cd dose-response course module, including the distinction between continuous and quantal BMD modeling, the role of the BMDU/BMDL ratio in characterizing statistical uncertainty, and the comparison of PROAST and BMDS software. It is also a teaching reference for the Ecr-vs-Ccr normalization debate. The breast cancer (Larsson 2015, Lin 2016), prediabetes/diabetes (Schwartz 2003, Wallia 2010, Jiang 2018), and osteoporosis (Kunioka 2022, Wallin 2013) epidemiology summarized in Sections 3.2 and 5 supplies course-grade examples of low-dose Cd effects below the 5.24 microg/g creatinine threshold.

App: This review does not measure foods, so it does not contribute occurrence values to the contamination_profile of any ingredient. It contributes vulnerable-population evidence (children and adolescent females [49,50]; women of reproductive age [51,52]) that low body iron stores are associated with elevated blood and urinary Cd, and it contributes occurrence-of-effect data on tubular and glomerular endpoints relevant to infant-and-child Cd threshold work.

Wiki pages this source may touch

• cadmium
• jecfa-cadmium-ptmi
• efsa-cadmium-twi
• epa-iris-cadmium-rfd
• atsdr-cadmium-mrls

Verification notes

DOI 10.3390/toxics13050404 matches the Toxics 2025, 13, 404 article received 9 April 2025, revised 6 May 2025, accepted 15 May 2025, published 16 May 2025. License is CC BY 4.0 per the publisher copyright statement on page 1. This is a narrative review by a single author (Soisungwan Satarug, Centre for Kidney Disease Research, Translational Research Institute, Brisbane, Australia); the author declared no conflicts of interest and the work received no external funding. The acknowledgments name Aleksandra Buha Djordjevic for BMD modeling assistance and Aleksandar Cirovic for figure design but neither is listed as an author.

The paper is total Cd throughout; no inorganic vs total speciation is discussed, so metals: [Cd] is correct (the cadmium vocabulary does not split iCd vs tCd as arsenic and mercury do). Frontmatter ingredients: [] and products: [] because the review is methodological and does not measure foods; the dietary-intake matrices entry captures the exposure-route context. The matrices vocabulary entries risk-assessment-methodology, urine, blood, kidney, bone, and dietary-intake describe the review’s subject matter and are precedented in similar review-type sources (e.g., chandravanshi-shiv-kumar-2021-cadmium-developmental-toxicity-infants-children-review uses infant-cohort, developmental-toxicity, etc.).

All instrument and software vendor names retained in the Methods section fall under the Part 12 scientific-method exception (PROAST software from RIVM, U.S. EPA Benchmark Dose Software). No brand-name products were sampled. No HMTc threshold values are proposed in the page body; the Cd dietary guideline values reported (JECFA 0.83 microg/kg bw/day, EFSA 0.36 microg/kg bw/day, etc.) are quotations from the published guidelines being reviewed, not wiki-side threshold proposals. No “consistent with prior literature” or other Part 2 cross-source synthesis is asserted; the review’s claims about Ecr vs Ccr non-differential error and the inadequacy of the JECFA threshold are reported as the author’s argument, not endorsed.

Multiple cohort datasets are named (Sweden Suwazono 2006 n=790; Jiangshan China Wang 2016 n=934; Japan Hayashi 2024 n=112; Thailand Satarug 2022 n=734; Thailand Satarug 2024 n=405; Switzerland Xie 2024 n=4704; combined n=917 and n=603 reanalyses) but these are review references, not the present paper’s primary subjects; sample_n: null and the sample_population field documents that this is a review.

The age-category POR values in Table 3 (e.g., 141.2 for the 66-87 y group, 95% CI 17.87-1116) are taken verbatim from the PDF and are extremely wide because the youngest age group (16-45 y) has very few low-eGFR cases. The 95% CIs are wide but the values are as printed.

Fresh-context audit subagent (2026-06-02) returned REVISE verdict; Checks 1 (numerical fidelity, all tabulated values cross-checked against pp.3-15), 3 (speciation/methods), and 4 (Part 12 brand firewall) clean. Findings applied: (a) Table 3/4 framing — verified against PDF p.9 footnote (Table 3 n=917 from [76]) and p.10 footnote (Table 4 n=405 from [80]); the prior overarching sentence “Both tables are reanalyses of data from 917 subjects” was wrong and was rewritten to distinguish the two tables. (b) [[health/cadmium-developmental-toxicity]] removed from “Wiki pages this source may touch” — verified that wiki/health/cadmium-developmental-toxicity.md does not exist (the sibling chandravanshi2021 review wikilinks the same slug, suggesting the target page should exist but is missing; flagged as a separate issue, not for this page to create). (c) Implications App paragraph — softened “supports the case for tighter Cd thresholds in the I and C product rows” to “contributes occurrence-of-effect data on tubular and glomerular endpoints relevant to infant-and-child Cd threshold work” to stay on the contributes-data side of the Part 2 boundary; replaced “premenopausal women, adolescent girls, pregnant women” with the source-faithful “children and adolescent females [49,50]; women of reproductive age [51,52]” per PDF p.5. Findings retained as flagged (not rejected, not modified): Check 2 ⚠️ on matrices vocabulary (urine, blood, kidney, bone, dietary-intake, risk-assessment-methodology) — the standard matrices vocabulary is scoped to consumer-food matrices; review-type sources extend it to subject-matter matrices, precedented by sibling reviews. Flagged for Karen’s confirmation if the matrices vocabulary should be formally extended. Check 5 ⚠️ on Implications Certification paragraph noting HMTc product categories that drive dietary Cd burden — retained because the page stays on “contributes occurrence-of-effect data for category X” wording, the allowed side of the Part 2 boundary.

Page history

The five most recent substantive edits to this page. The full version history lives in git; when DOI minting comes online (see schema docs), each entry below will also link to a version-pinned DataCite DOI.