Age-stratified models also placed the estimated margins of exposure for infants and young children far below the benchmark of 10,000 used to identify a public-health concern for a genotoxic carcinogen. These results come from a screening-level assessment that combined separate literatures rather than measuring feed and milk from the same Bangladeshi herds. The review assembled evidence from PubMed, Scopus, Web of Science and ScienceDirect for studies published from January 2000 through January 2026.

Of 318 records identified, 96 were removed as duplicates and 188 were excluded during title and abstract screening; 32 full texts were assessed, and 12 reports met the eligibility criteria. The 20 analysed datasets represented division-, site- or milk-type-specific strata within 10 distinguishable sources, including raw, pasteurised and ultra-high-temperature milk. Nine datasets were classified as being at low risk of bias, five at moderate risk and six at high risk.

All measurements used competitive enzyme-linked immunosorbent assay, or ELISA, and none was confirmed chromatographically. The investigators also prespecified subgroup analyses by region, milk type and risk of bias, along with sensitivity analyses intended to test whether clustered datasets, methodological choices or unusually high measurements altered the results. Aflatoxin B1 is the most prevalent aflatoxin congener in food and feed and the most potent naturally occurring hepatocarcinogen, according to the source article.

When lactating animals consume aflatoxin B1-contaminated feed, hepatic cytochrome-P450 hydroxylation converts part of the toxin to aflatoxin M1, which is secreted into milk. Aflatoxin M1 retains about one-tenth of the carcinogenic potency of its parent compound but resists pasteurisation and ultra-high-temperature processing, allowing it to remain in milk after those treatments. Bangladesh’s tropical monsoon climate supports fungal growth during crop filling, harvest and on-farm storage, while maize, a cereal particularly susceptible to aflatoxigenic fungi, is now a principal energy ingredient in compound feed.

The country’s expanding dairy sector remains dominated by smallholder and informal supply chains with limited cold-chain infrastructure, fragmented quality control and weak enforcement. Although aflatoxin also contaminates rice and other foods, the assessment focused on the milk pathway and did not estimate dietary exposure from those additional sources. The pooled concentration distribution was strongly right-skewed, making the choice of central measure important.

The sample-weighted arithmetic mean across datasets was 143 nanograms per litre, the sample-weighted geometric mean was 92 nanograms per litre, the median of dataset means was 88 nanograms per litre and the unweighted mean was 118 nanograms per litre. Typical regional raw-milk means ranged from 35 to 120 nanograms per litre, but one Dhaka dataset reported a mean of 699.1 nanograms per litre. Removing that single dataset, which represented 8% of all samples, reduced the weighted mean to 93 nanograms per litre, a 35% decrease.

The interquartile range of dataset means was 57 to 111 nanograms per litre, whereas the complete range extended from 6.6 to 699.1 nanograms per litre. Prevalence and mean concentration were largely decoupled: two datasets with 100% positivity had among the lowest means, and the highest concentration occurred at a prevalence close to the pooled value. The assessment therefore relied on the concentration distribution rather than treating the proportion of positive samples as a sufficient measure of exposure.

Milk type did not significantly explain differences in prevalence, whereas region did. Raw milk had a pooled prevalence of 60.4%, compared with 59.0% in pasteurised milk and 69.5% in ultra-high-temperature milk, but the between-subgroup test found no significant difference among the three categories. The ultra-high-temperature estimate was not considered interpretable because it was based on only two datasets, one reporting 20% positivity and the other 100%, and its prediction interval spanned the complete possible range.

Region showed a more consistent pattern: prevalence was 43.7% in Chattogram, 55.3% in Dhaka and 66.8% across other divisions and national samples, with a Q-between value of 13.5, two degrees of freedom and a nominal p value of 0.001. This was the only pre-specified subgroup comparison to remain significant after Holm adjustment, with an adjusted p value of 0.004, although the authors classified it as hypothesis-generating rather than an established geographic gradient. Residual heterogeneity was also 21% to 23% in exploratory one-covariate meta-regressions, leaving variation potentially associated with unmeasured differences in season, analytical kit, detection threshold, sampling frame and position within the supply chain.

The feed evidence showed that the upstream contamination burden was compatible with the milk observations, but it did not establish a causal relationship within Bangladesh. Published surveys reported aflatoxin B1 at 0 to 160 micrograms per kilogram in poultry feed and 1 to 103 micrograms per kilogram in maize, with both matrices frequently exceeding the 20 micrograms per kilogram general feed maximum. Compound beef-cattle feed ranged from 7.78 to 14.66 micrograms per kilogram, while oil cakes, bran, rice straw and soybean meal had means of approximately 9.8 to 12 micrograms per kilogram.

Although many cattle-feed means remained below 20 micrograms per kilogram, they exceeded the European Union limit of 5 micrograms per kilogram for dairy-cattle feed. Maize and other feed ingredients also contained fumonisins, zearalenone and sterigmatocystin, revealing a multi-mycotoxin burden not captured by the aflatoxin-specific risk metrics. Feed and milk measurements both showed a June-to-August maximum, consistent with a humidity-driven mechanism, but the series came from independent studies, sites and years and therefore did not demonstrate coupling within a shared production system.

For the mechanistic check, the researchers transferred a published yield-dependent carry-over function from controlled trials of 12 high-yielding Israeli-Holstein cows to three ration scenarios representing Bangladeshi smallholder practice. The function predicted transfer of approximately 0.67% to 1.13% at milk yields of 5 to 15 kilograms per day, with reconstructed milk concentrations spanning 24 to 282 nanograms per litre and a central prediction of 80 nanograms per litre. That envelope contained 18 of the 20 observed dataset means, or 90%, and the central prediction was close to the observed geometric mean of 92 nanograms per litre.

Reproducing the sample-weighted mean of 143 nanograms per litre required a ration-weighted aflatoxin B1 concentration of about 18 micrograms per kilogram, while reproducing the extreme Dhaka mean would have required about 89 micrograms per kilogram. The two tails, at 6.6 and 699.1 nanograms per litre, lay outside the envelope. This result shows that the independent feed and milk literatures were quantitatively compatible under the transferred model; it does not show that a particular feed sample caused a particular milk residue.

Moreover, the carry-over relationship was extrapolated below the 30-to-46-kilogram daily-yield range represented in its original calibration. Using Bangladesh-specific milk-intake and body-weight parameters, the central scenario estimated aflatoxin M1 exposure at 4.22 nanograms per kilogram of body weight per day for infants aged 0.5 to 1 year, 2.29 for toddlers aged 1 to 3 years, 1.08 for children aged 4 to 9 years, 0.38 for adolescents aged 10 to 17 years and 0.24 for adults aged 18 years or older. Corresponding margins of exposure were 95, 175, 372, 1,055 and 1,649, all below the benchmark of 10,000, while hazard indices were 21.1, 11.4, 5.4, 1.9 and 1.2, all above one.

Under a high-concentration scenario of 250 nanograms per litre, the infant margin fell to 54. The central model estimated incremental hepatocellular carcinoma incidence of 0.0108 cases per 100,000 infants per year, rising to 0.0189 in the high scenario. These cancer estimates were order-of-magnitude indicators rather than case counts because they depended on an aflatoxin M1 relative potency factor of 0.1 and an HBsAg-weighted potency model using an approximately 5.4% hepatitis-B surface antigen carrier prevalence.

A 100,000-iteration Monte Carlo simulation incorporated variability in milk concentration, intake and body weight, treating the three inputs as independent. Median margins of exposure were 157 for infants, 286 for toddlers, 612 for children, 1,720 for adolescents and 2,735 for adults; fifth-percentile values were 30, 55, 119, 340 and 543, respectively. The simulation produced a margin below 10,000 in 100% of infant and toddler iterations, 99.7% of child iterations, 96.2% of adolescent iterations and 90.4% of adult iterations.

For infants, 32.8% of iterations produced a margin below 100, while the median hazard index was 12.8 for infants, 7.0 for toddlers, 3.3 for children, 1.2 for adolescents and 0.7 for adults. These probabilistic medians were higher than the deterministic margins because the fitted concentration distribution was centred on the 92 nanogram per litre geometric mean rather than the 143 nanogram per litre arithmetic mean. The numerical exposure estimates therefore varied with the central tendency used, but the principal conclusion remained stable: no tested scenario or restriction raised the infant margin above 158, at least 60-fold below the 10,000 benchmark.

Several limitations constrain the assessment. All included measurements relied on competitive ELISA without chromatographic confirmation, and inconsistent reporting of detection limits, recovery and within-dataset concentration distributions prevented the investigators from reconstructing individual exposures. Most surveys used convenience samples from markets, collection points or farms rather than probability samples of the national milk supply.

Six datasets had uncertain provenance or a high risk of bias, and four reports contributed multiple strata, so the 20 datasets did not represent 20 fully independent sources. A 50% downward correction for possible ELISA overestimation reduced the weighted concentration to 96 nanograms per litre but left the infant margin at 142. Restricting the evidence to fully referenced peer-reviewed datasets produced a weighted mean of 172 nanograms per litre and an infant margin of 79, while removing the influential Dhaka dataset produced a mean of 93 nanograms per litre and a margin of 146.

Alternative heterogeneity estimators left pooled prevalence between 58.7% and 59.0%, and no plausible single uncertainty or tested combination reversed the finding. The authors identified paired feed-and-milk sampling from the same herds across monsoon and dry seasons, chromatographic confirmation, individual milk-consumption data and replication of the high-concentration Dhaka milk shed as priorities for converting this literature-based screening assessment into a more precise national exposure model.

Subject of Research: Systematic review and meta-analysis of aflatoxin contamination in Bangladeshi milk and animal feed, with modelled dietary risk by age.

Article Title: Aflatoxin M1 in Milk and Aflatoxin B1 in Animal Feed in Bangladesh: A Systematic Review, Meta‐Analysis and Literature‐Based Dietary Risk Assessment

Article References: Fattah, M. A., & Habib, M. A. (2026). Aflatoxin M 1 in Milk and Aflatoxin B 1 in Animal Feed in Bangladesh: A Systematic Review, Meta‐Analysis and Literature‐Based Dietary Risk Assessment. Veterinary Medicine and Science, 12(6), Article e71284. https://doi.org/10.1002/vms3.71284

Image Credits: AI Generated

DOI: 10.1002/vms3.71284

Keywords: aflatoxin M1, aflatoxin B1, Bangladesh, milk contamination, animal feed, food safety, dietary exposure, infant nutrition, hepatocellular carcinoma, meta-analysis, Aflatoxin, Milk