Lot-to-lot variation in lead in US infant food: 13,076 disclosed production lots and FDA Total Diet Study data

Preprint. This manuscript has not been peer reviewed. It is under review at Public Health and Toxicology. Introduction. California Assembly Bill 899 requires manufacturers of baby food sold in California to test each finished product for four toxic elements at least monthly and to publish the results. They do so across dozens of incompatible per-brand portals. Existing surveys sample each product once, so they cannot see how much one product varies from lot to lot. Methods. We harvested and harmonized these disclosures into one open dataset of 13,076 lots from 32 brands, holding 47,782 element determinations, and compared it against FDA's baby and toddler food sampling programs and the FDA Total Diet Study. We partitioned lead variance within and between products, tested lot sequences for serial dependence, and estimated the lead removed by rejecting over-level lots. Results. The top 5 percent of lots carry 20.5 percent of the measured lead. Against the applicable FDA action levels, 138 lots in covered categories exceed their level (4.46 percent of covered lots with a measured value), and a further 163 lots in uncovered snack categories sit above the same values. Within a brand, a median 57.8 percent of lead variance sits between lots of one product rather than between products, and a typical product spans a 4-fold range across its lots. High-lead lots arrive in streaks in production order, in 20 of 56 product series against 2.8 expected by chance, of which 17 survive correction for multiple testing, though laboratory batch effects cannot be excluded for 9 of the 17. In the Total Diet Study the top 5 percent of samples carry 60.5 percent of measured lead. Conclusions. The variation that matters lies between lots of the same product, so a limit written for the finished product reaches the problem only after the contaminated ingredient has already been used. Testing sited at the ingredient lot would reach it earlier. The complete data and code archive supporting this manuscript is deposited separately; its DOI is activated when that deposit is published.