Assessment of arsenic, cadmium, lead, mercury, and per- and polyfluoroalkyl substances concentrations in human milk and infant formula in the United States: a systematic review

Overview

Thoerig, O’Connor, Spill, and colleagues at the Agriculture, Food, and Nutrition Evidence Center (Fort Worth, TX) conducted a PROSPERO-registered systematic review (CRD42024528756) of published peer-reviewed evidence on arsenic, cadmium, lead, mercury, and per- and polyfluoroalkyl substances (PFAS) in human milk (HM) and infant formula (IF) in the United States. The review was commissioned by the FDA’s Human Foods Program. Searches covered five databases (CAB Abstracts, CENTRAL, CINAHL, Embase, MEDLINE) with no date restrictions through 2 April 2025. The analysis was restricted to US data for primary conclusions; non-US data from countries rated “high” or “very high” on the Human Development Index (HDI) were compiled in heat maps to document evidence available for future systematic reviews. GRADE was used to assess certainty of evidence. For arsenic, cadmium, lead, and mercury in HM and IF, the evidence base consisted primarily of small, geographically unrepresentative studies published before the year 2000, precluding firm conclusions. For PFAS, contemporary evidence was available: PFOA and PFOS were detected in US HM samples (GRADE: moderate), and PFAS were largely undetected in IF (GRADE: moderate).

Study design

PROSPERO-registered (CRD42024528756), PRISMA-compliant systematic review. The review was part of a series of six systematic reviews commissioned by the FDA’s Human Foods Program. Five databases were searched for peer-reviewed articles published in English with no date restrictions: Medline (Ovid), Embase (Ovid), CAB Abstracts (Ovid), CINAHL (EBSCO), and CENTRAL (Cochrane Library). Backward citation searches were conducted, and the Journal of Environmental Exposure Assessment was hand-searched. Inclusion criteria required studies to report measured concentrations of arsenic (total and inorganic), cadmium, lead, mercury (total and methylmercury), or select PFAS (PFOA, PFOS, PFNA, PFHxS) in HM or IF from countries rated “high” or “very high” on the HDI.

Screening and data extraction were performed by two independent reviewers using DistillerSR (Evidence Partners), with disagreements resolved by a third reviewer or discussion. Quantitative concentration data were extracted independently by two reviewers and cross-checked by a third. Medians were prioritized for PFAS (biological data are often skewed); means were extracted for arsenic, cadmium, lead, and mercury because those contaminants were most commonly reported as means. For IF, conclusions were drawn only from studies published after 2000, given changes in ingredient sourcing, manufacturing, and regulations over time. Meta-analysis was planned a priori but was not conducted because of small numbers of contemporary studies and inadequate detail for pooling; narrative synthesis was performed instead.

Risk of bias (ROB) for HM studies used ROBINS-E, ROBINS-I, or RoB 2 depending on study design. ROB for IF studies used a custom instrument developed to address IF-specific sampling concerns because existing tools were designed for human studies. GRADE was applied to each conclusion. Protocol deviations from the a priori registration included restricting primary conclusions to US data only (decided post hoc), adding CAB Abstracts to the search, adding a second data extractor for quantitative data, and creating the custom IF ROB instrument.

Results: human milk studies

Study identification and characteristics

The database search identified 6,255 unique records screened at title/abstract level. Of 686 full-text articles screened, 348 were excluded because they were not relevant to HM. Of the remaining 338 articles, six were excluded at confirmatory full-text screening (no outcome or duplicative data), leaving 332 articles from all countries. Of these, 15 articles from 14 US studies were included for US-specific synthesis: 2 for arsenic, 2 for cadmium, 8 for lead, 2 for mercury, and 4 for PFAS (Table 1 of the paper). Eleven studies were cross-sectional; three were prospective cohort studies with repeated measures. Data were collected between 1968 and 2019 from 14 states plus Washington, DC, primarily on the East and West Coasts. HM samples were collected from a few days to 2 years postpartum.

The technical expert panel (TEP) critically appraised sample collection and analytic methods as “acceptable” for 11 studies and “some concerns” for 4 studies (Klein 2017, Walker 1980, Pitkin 1976, Galster 1976) due to insufficient reporting of methods or quality controls.

Arsenic in human milk (Table 2)

Two cross-sectional studies assessed and detected arsenic in US HM:

• Klein 2017 (Boston, MA; collected 2013): mean total arsenic 3.47 +/- 0.84 ug/L from 20 participants; collected mid-feed at 2 weeks to 2 years postpartum; arsenic speciation not reported; TEP appraisal “some concerns” due to lack of method performance detail; overall ROB: high (missing data).
• Carignan 2015 (New Hampshire Birth Cohort; collected 2012-2013): mean arsenic 0.33 +/- 0.17 ug/L from 9 participants (median 0.31, Q1-Q3 0.25-0.44 ug/L; mean and SD calculated using Meta-Analysis Accelerator); collected at 2-7 weeks postpartum; home water source from a private unregulated well; speciation not investigated due to low concentrations; TEP appraisal “acceptable”; overall ROB: high (selection bias); 6 of 9 (table) or 5 of 9 (text) samples detected.

The evidence did not support a conclusion about total or inorganic arsenic in US HM: sample sizes were small (n = 9 and n = 20), limited to one geographic region, and demographic characteristics were insufficient to assess generalizability.

Cadmium in human milk (Table 2)

Two cross-sectional studies assessed cadmium in US HM:

• Klein 2017 (Boston, MA; 2013): cadmium not detected in 20 samples; TEP appraisal “some concerns”; overall ROB: high (missing data).
• Murthy 1971 (Cincinnati, OH; 1968): mean cadmium 19 +/- 27 ug/L detected in 22 samples from 13 participants; TEP appraisal “acceptable”; overall ROB: high (missing data).

The evidence did not support a conclusion: small sample sizes from one geographic region, little demographic detail, and one article was published over 50 years ago.

Lead in human milk (Table 2)

Eight studies (2 prospective cohort, 6 cross-sectional) assessed and detected lead in US HM. Data collection spanned 1968-2013 across 9 states and Washington, DC. Mean lead concentrations ranged from 0.77 +/- 0.45 ug/L (Klein 2017; Boston, 2013) to 26 +/- 11 ug/L (Dillon 1974; NY/KY/CA/TN, 1970-1973). The 2013 Klein study was the only one to collect data within the last 25 years and had the lowest mean concentration. In the two prospective cohort studies (Sowers 2002; Rockway 1984), repeated measures showed no statistically significant changes in lead across lactation time points.

Concentration values by study (from Table 2):

The evidence did not support a conclusion about lead in US HM: only 2 of 8 studies collected data within the past 25 years, most studies were at high ROB primarily due to missing data, and geographic representation was limited.

Mercury in human milk (Table 2)

Two cross-sectional studies, both published before 1976, assessed and detected mercury in US HM. Neither study reported methylmercury speciation.

• Pitkin 1976 (Iowa; date not reported): mean total mercury 0.93 +/- 0.23 (SE) ug/L from 32 participants at 56+ days postpartum; TEP appraisal “some concerns”; ROB: some concerns.
• Galster 1976 (Anchorage/Bethel, Alaska; date not reported): total mercury concentrations in HM collected a few days postpartum from 21 participants; coastal area mean 7.6 +/- 2.7 (SE) ug/L (n = 11); interior area mean 3.2 +/- 0.8 (SE) ug/L (n = 5); coastal participants had approximately 2-fold higher mercury than those in the interior, attributed to greater seafood consumption; TEP appraisal “some concerns”; ROB: high.

The evidence did not support a conclusion: concerns about sample collection and analytic methods, and data were collected approximately 50 years ago.

PFAS in human milk (Table 4)

Four studies (3 cross-sectional, 1 prospective cohort) assessed PFAS in US HM, with samples collected from 2004 to 2019 from approximately 540 participants across different parts of the United States. All four studies were critically appraised as “acceptable” by the TEP.

PFOA and PFOS were detected in 3 of 4 studies (Criswell 2023, Zheng 2021, Tao 2008; not detected in von Ehrenstein 2009):

Values are median (Q1-Q3) except Tao 2008 which reports median (range). Criswell 2023 values were converted from ng/mL (multiply by 1,000). Zheng 2021 also reported a median total PFAS of 121 (103-190) pg/mL. PFNA and PFHxS were detected in only 2 of 4 studies in 79/559 and 68/559 samples, respectively.

The evidence supported two conclusions at GRADE moderate certainty (downgraded once for ROB due to selection bias and missing data in all studies): (1) median PFOA in US HM ranged from undetected to 36.1 pg/mL; (2) median PFOS in US HM ranged from undetected to 106 pg/mL. PFNA and PFHxS had inconsistent detection frequencies precluding a conclusion.

Results: infant formula studies

Study identification and characteristics

Of 6,255 unique records screened, 128 articles underwent full-text screening after the first round. Five were excluded at confirmatory full-text screening (outcome and study design), and one article from the Journal of Environmental Exposure Assessment was added, leaving 16 US IF studies: arsenic (n = 5), cadmium (n = 4), lead (n = 12), mercury (n = 1), and PFAS (n = 2). All study designs were cross-sectional except for two prospective cohort studies. The seven studies published between 2002 and 2024 are summarized in Table 6 of the paper; the nine published before 2000 are in supplemental materials. All studies were critically appraised as “acceptable” except one published in 1980 that lacked quality-control information for lead.

Arsenic in infant formula (Table 8)

Five studies assessed arsenic in IF. All products reported (n = 27) were powdered and not diluted before sample processing. Arsenic was detected in 3 of 5 studies and not detected in 2:

• TatahMentan 2024 (US; n = 2 milk-based products with rice starch): arsenic detected, range 11-20 ng/g; cadmium 2-3 ng/g in same products; lead not detected; analytic method ICP-MS (Agilent 7900).
• Jackson 2012 (Hanover, NH; n = 15 products from 5 main brands purchased locally): arsenic detected in 11 of 14 dairy-based products (range 2.6 +/- 0.44 to 9.38 +/- 0.31 ng/g for 7 dairy products without rice; range 6.02 +/- 0.26 to 8.19 +/- 0.63 ng/g for 2 dairy products with rice; 11.89 +/- 0.64 ng/g for 1 nondairy product with rice); speciation analysis (anion-exchange chromatography coupled to ICP-MS) found 100% inorganic arsenic in the 9 of 15 products with total arsenic above 6 ng/g; arsenic not detected in any of the 3 soy-based products; analytic method ICP-MS (Agilent 7700x).
• Vela 2004 (Cincinnati, OH; n = 3 samples, iron-fortified and low-iron): arsenic detected, range 12 +/- 2 to 17 +/- 3 ng/g (iron-fortified), 16 +/- 1 ng/g (low iron); analytic method ICP-MS model PQ3 (Thermo Elemental).
• Olu-Owolabi 2007 (multiple US cities; n = 1 soy with iron in glass bottle): arsenic not detected; analytic method wet oxidation plus radial ICP emission spectrometry.
• Ikem 2002 (Auburn, AL; n = 2 products, one milk-based iron-fortified in metal container, one milk-based iron-fortified hypoallergenic in paper container; and 2 soy-based iron-fortified products in metal or paper container): arsenic not detected in any; analytic method ICP-OES (Perkin-Elmer DV 3300).

Arsenic concentrations across the detecting studies ranged from 2.6 to 20 ng/g. The EU maximum guideline for powdered infant formula is 20 ng/g (0.02 mg/kg); all reported concentrations were at or below this value. The evidence did not support a conclusion: arsenic was inconsistently detected, sample sizes were small, sampling schemes were inadequate, and product descriptions were insufficient to confirm whether the 27 products assessed were unique.

Cadmium in infant formula (Table 8)

Three contemporary studies (TatahMentan 2024, Olu-Owolabi 2007, Ikem 2002) assessed cadmium in 8 IF products (all powdered, undiluted). Cadmium was detected only in TatahMentan 2024 (2-3 ng/g in 2 milk-based products with rice starch) and not detected in Olu-Owolabi 2007 or Ikem 2002.

The 1971 Murthy study detected cadmium in all 4 IF products (diluted 1:1; original form, water source, and dilution timing not reported), with values ranging from 20 +/- 4 to 39 +/- 7 ng/g; soy-based products had higher concentrations than dairy-based products. The evidence did not support a conclusion: results were inconsistent, sample preparation varied, sample size was small, sampling schemes were inadequate, and brand or manufacturer information was lacking.

Lead in infant formula (Table 8)

Three contemporary studies (TatahMentan 2024, Olu-Owolabi 2007, Ikem 2002) assessed lead in 8 IF products and did not detect it. Nine older studies (1971-1997) detected lead in IF regardless of variations in protein base, preparation method, or water source. The lowest mean lead concentration in an older study was 0.24 +/- 0.06 ng/g (Thompson 1993; milk-based concentrate, dilution information not reported). The highest mean was 540 +/- 110 ng/g (Murthy 1971; soy-based, diluted 1:1 with unknown water source). The evidence did not support a conclusion: most data were from before 1997, and contemporary studies had small sample sizes and lacked brand and manufacturing information.

Mercury in infant formula (Table 8)

One study (Ikem 2002) assessed mercury in IF. Five powdered and undiluted milk- or soy-based IF products were tested; mercury was not detected in any. Results from a single small study were not representative of the US market and did not support a conclusion.

PFAS in infant formula (Table 8)

Two studies assessed PFAS in IF: Tao 2008 and Bogdan 2024. These collectively assessed 38 IF products:

• Tao 2008 (Washington, DC/Boston, MA; 2007): 21 products from 5 manufacturers representing over 99% of the US market; largely milk- or soy-based, including organic and nonorganic; powdered and concentrated liquid diluted with Milli-Q water; PFOA not detected; PFOS detected in 1 of approximately 21 dairy-based products at a mean of 0.054 +/- 0.003 ng/g (detected as a powder, not in diluted form); PFNA not detected; PFHxS detected in range 1.35-3.59 pg/mL in 2 of 21 samples; analytic method: HPLC (Agilent 1100 Series) coupled with API 2000 electrospray triple-quadrupole mass spectrometer.
• Bogdan 2024 (Minnesota; December 2021 to January 2023): 17 different powdered products including dairy, soy, nondairy amino acid hypoallergenic blends, medical formulas, organic, non-GMO, and with and without rice starch; sampling was informed by WIC purchasing statistics; diluted with ultrapure water at 20% weight/volume; PFOA not detected in any of the 17 products (12 dairy, 4 soy, 1 amino acid); PFOS not detected in 16/17 products; PFHxS not detected; PFNA not detected; analytic method: liquid chromatography tandem-mass spectrometry (Shimadzu Prominence, AB Sciex 5500Q).

The evidence supported a conclusion (GRADE: moderate) that PFAS were largely not detected in IF assessed in US studies. The certainty was downgraded for publication bias because industry and regulatory data are routinely collected but not published in peer-reviewed literature.

Analytical methods described

Methods across included studies are detailed in Tables 1 and 6 of the paper. For HM studies, instruments ranged from early atomic absorption spectrophotometry (Dillon 1974, Lamm 1973, Murthy 1971, Rabinowitz 1985) to ICP-MS (Klein 2017, Carignan 2015) and liquid chromatography-mass spectrometry for PFAS (Criswell 2023, Zheng 2021, Tao 2008, von Ehrenstein 2009). For IF studies, instruments included ICP-MS (TatahMentan 2024, Jackson 2012), ICP-OES (Ikem 2002), and wet oxidation with ICP emission spectrometry (Olu-Owolabi 2007). The TEP (Galusha, Huset, and an additional expert from NIST) appraised each study as “acceptable” or “some concerns.”

LOD/LOQ: Not reported by this systematic review. Individual primary studies reported their own detection limits; imputation methods for below-LOD values varied across PFAS studies (LOQ/sqrt(2) in Criswell 2023; LOD/sqrt(2) in von Ehrenstein 2009).

CRM recovery: Not reported by this systematic review.

Regulatory context

The paper notes the following regulatory context (as published):

• No mandatory federal limits or proposed action levels for arsenic, cadmium, lead, mercury, or PFAS in IF exist in the United States as of the publication date. The FDA works with manufacturers to monitor environmental contaminants but published per-product data are limited.
• EU maximum guideline for inorganic arsenic in powdered infant formula: 0.020 mg/kg (20 ng/g). All detected arsenic concentrations in the included IF studies were at or below this value.
• WHO cadmium exposure: levels at or below 20.6 ug/kg body weight per month may be safe for children 6 months to 12 years; whether US HM exposures approach this threshold is unclear from the available evidence.
• CDC recommendation for HM: discard milk containing 40 ug/dL (400 ug/L) or more of lead to avoid toxicity; the highest reported lead value in this review was 26 +/- 11 ug/L (Dillon 1974), orders of magnitude below that threshold.
• WHO “normal” for mercury in HM: 1.4-1.7 ug/L; Iowa HM values in Pitkin 1976 (0.93 ug/L) were below this, but Alaska coastal values in Galster 1976 (3.2-7.6 ug/L) were 2-5 times above it.
• FDA Total Diet Study (TDS): a potential mechanism for more systematic IF surveillance, as described by the paper; currently presents hybrid mean values across IF types rather than product-specific data.

Evidence fitness

Evidence tier A. This is the most current and rigorous US systematic review of toxic elements and PFAS in human milk and infant formula available as of 2025. It is the canonical secondary-synthesis reference for the US infant-feeding contaminants evidence base through 2025. Its primary contribution is documenting the absence of contemporary, nationally representative US data on arsenic, cadmium, lead, and mercury in both HM and IF — a gap that contextualizes why international occurrence data are needed for threshold-setting.

Supports: regulatory context for US absence of mandatory limits on these contaminants in IF; systematic review of PFAS detection in US HM (PFOA and PFOS detected, moderate certainty); systematic review of PFAS largely undetected in US IF (moderate certainty); documentation of data gaps for arsenic, cadmium, lead, and mercury in HM and IF; methodological framework for evaluating primary studies.

Does not support: direct concentration data for contemporary US IF or HM for arsenic, cadmium, lead, or mercury (evidence base is too old, too small, and too geographically narrow). The underlying primary studies are the authoritative sources for quantitative occurrence data.

Limitations

The majority of studies for arsenic, cadmium, lead, and mercury were conducted before the year 2000, limiting relevance to contemporary formulations and exposure levels. Only 2 of 14 HM studies and 7 of 16 IF studies were published since 2000. Most HM studies were from a single geographic region (New England). No studies reported speciated arsenic in IF; all speciation data come from a single study (Jackson 2012) that found 100% of detectable arsenic was inorganic in products where total arsenic exceeded 6 ng/g. No studies reported methylmercury speciation in IF. The custom ROB instrument for IF studies was not validated or tested for inter-rater reliability. The review synthesized only published peer-reviewed literature; FDA’s internal monitoring data for IF are not publicly reported and were therefore not captured.

Implications

Certification: This systematic review is the highest-tier secondary-synthesis source for HMT&C infant formula rows (milk-based and soy-based powder and ready-to-feed liquid). It documents that the US peer-reviewed evidence base for contemporary heavy metal concentrations in IF is too thin to support concentration-based conclusions, reinforcing reliance on international occurrence data for threshold-setting. The absence of mandatory US federal limits for arsenic, cadmium, lead, and mercury in IF is confirmed here and is itself directly relevant to the HMT&C regulatory-gap justification.

Courses: Useful as a teaching example of systematic review methodology applied to infant feeding toxicology (PRISMA, GRADE, ROB assessment); illustrates how evidence gaps in a high-income country monitoring program create knowledge deficits for infant safety.

App: Does not add contemporary product concentration data for US IF. The historical range of lead in pre-1997 IF (0.24 to 540 ng/g) illustrates the magnitude of contamination reduction achieved by manufacturing changes and regulatory action; this historical context can inform the app’s risk-calibration narrative for formula-fed infants.

Microbiome: No direct microbiome endpoint addressed.

Related evidence

• Breastmilk
• Infant Formula, Powder (Non-Soy)
• Soy-based infant formula
• Infant Formula, Powder (Non-Soy)
• Infant Formula, Powder (Soy-Based)
• Infant Formula, RTF Liquid (Non-Soy)
• Infant Formula, RTF Liquid (Soy-Based)
• Lead
• Cadmium
• Arsenic, Inorganic
• Arsenic, Total
• Mercury, Total

Update history

No substantive edit history is available in this build. The full commit record is available in git.