Working research document, copied from the author’s notes on 2026-09-07. Rough, long, and unedited apart from removing personal details. The finding page summarizes it.

Does buying organic make a measurable difference to health?

Deep-research pass, 2026-09-07. Companion to the food and ingredient evidence rubric, and written to its conventions: every claim carries a URL, year, n, design, effect size and peer-review status, and multiples are stated with denominators.

Tier labels used below, mapped onto the rubric’s existing evidence language:

Tier	Meaning
RCT	Randomized human trial. In this field, almost all are biomarker trials, never clinical endpoints.
LTSP	Long-term study of people. Prospective cohort, case-control, or a meta-analysis of them. Observational.
MECH	Mechanistic, in vitro, animal, or a composition measurement with no outcome attached.
SPEC	Speculation. Plausible reasoning with no data behind it.

Marks: [V] verified by fetching the source for this document. [A] computed for this document from a raw public dataset, not published anywhere, and therefore usable as a lead but not quotable as a citation. Everything below is [V] unless marked otherwise.

Bottom line, in five bullets

There is no established personal-health case for buying organic. No randomized trial of an organic diet with a clinical endpoint has ever been run. The cohort literature pools to null: overall cancer HR 0.93 (0.78 to 1.12) across three cohorts (Theodoridis 2025), with the largest cohort (Million Women, n=623,080) null and the longest (Danish, median 15 years) null. Confidence: high that no effect is established, low that no effect exists. Those are different statements and this document rests on the first.
The composition differences are real, small, and mixed in direction. Organic crops carry 17% to 69% more polyphenols, 15% less protein and 8% less fibre. Organic milk carries 56% more total n-3 and 69% more ALA, worth roughly 60 to 90 mg/day extra total n-3 per half litre (derived, not a published figure), against 74% less iodine. Several of these estimates, including protein, cadmium, total antioxidant activity (TEAC) and total phenolic acids, are ones Barański’s own authors flag as “less reliable” because the mean percentage difference falls outside the CI of the standardised mean difference. Confidence: high on the composition numbers, moderate on the “too small to matter” reading.
The one defensible precautionary case is pregnancy and toddlers, on cumulative organophosphate exposure. It is precautionary, not evidential: no outcome study links diet-derived residues to a clinical endpoint. But the margins are thinner than the adult numbers suggest. EFSA’s cumulative acetylcholinesterase-inhibition assessment is only 50% to 90% certain that French children stay under the threshold, EPA’s chlorpyrifos food-only estimate for children 1 to 2 is 9.7% of the ssPAD, within about one order of magnitude, and the reference doses themselves are derived from cholinesterase inhibition with uncertainty factors, while the CHAMACOS and Columbia neurodevelopmental effects were observed below them. That puts produce eaten in volume by a pregnant woman or a toddler at the top of the “if you buy organic, buy this” list, with cereals for cadmium second. Confidence: moderate for adults, low to moderate for prenatal and toddler exposure.
The strongest reasons to buy organic are not health reasons. Antimicrobial stewardship (Innes 2021, multidrug-resistant organism aPR 0.43, 0.30 to 0.63, on 39,349 retail meat samples), farmworker and applicator exposure (roughly 500x the general-population urinary glyphosate maximum), and per-hectare biodiversity. All three are collective goods, none is a claim about the buyer’s own body, and none appears in the marketing.
The environmental case is mixed per unit of food. Across 742 systems in 164 LCAs, organic used 25% to 110% more land and had 37% higher eutrophication potential, against 15% less energy and no significant greenhouse-gas difference. What you eat matters more than how it was farmed.

The one-line practical version: buying organic is a defensible ethical and environmental purchase and an unproven nutritional one. If the money is finite, the same dollars spent on more vegetables and more fibre have outcome evidence behind them that organic certification does not.

1. Nutrient content: organic vs conventional crops, milk, meat, eggs

The two large “no difference” reviews

Dangour 2009, Am J Clin Nutr 90:680-5, systematic review, peer-reviewed. 52,471 articles screened, 162 studies retrieved (137 crops, 25 livestock), 55 of satisfactory quality. 1,149 nutrient comparisons from 46 satisfactory crop studies; 11 nutrient categories had 10 or more studies. 3 of 11 differed: nitrogen higher in conventional; phosphorus and titratable acidity higher in organic. The other 8 (vitamin C, phenolic compounds, magnesium, potassium, calcium, zinc, copper, total soluble solids) were null. Conclusion: “there is no evidence of a difference in nutrient quality between organically and conventionally produced foodstuffs.” (https://ajcn.nutrition.org/article/S0002-9165(23)26563-6/fulltext) LTSP/MECH

Smith-Spangler 2012, Ann Intern Med 157(5):348-66, systematic review, peer-reviewed. 17 human studies and 223 food studies. Only phosphorus was significantly higher in organic, and the authors called that difference not clinically significant. Of the 17 human studies, only 3 examined clinical outcomes, all observational, all null. Conclusion verbatim: “The published literature lacks strong evidence that organic foods are significantly more nutritious than conventional foods.” Limitations verbatim: “Studies were heterogeneous and limited in number, and publication bias may be present.” (https://pubmed.ncbi.nlm.nih.gov/22944875/, DARE structured abstract at https://www.ncbi.nlm.nih.gov/books/NBK100554/) LTSP/MECH

The Newcastle meta-analyses that found differences

Barański 2014, Br J Nutr 112:794-811, 343 publications (156 in the weighted meta-analysis), peer-reviewed. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4141693/) MECH

Parameter	Organic vs conventional	95% CI
Total antioxidant activity	+17%	3 to 32
Phenolic acids	+19%	5 to 33
Flavanones	+69%	13 to 125
Flavones	+26%	3 to 48
Flavonols	+50%	28 to 72
Anthocyanins	+51%	17 to 86
Vitamin C	+6%	-3 to 15 (null)
Protein	-15%	-27 to -3
Fibre	-8%	-14 to -2
Total nitrogen	-10%	-15 to -4
Cadmium	-48%	-112 to +16 (MPD CI crosses zero)
Nitrate	-30%	-144 to +84
Nitrite	-87%	-225 to +52
Lead, mercury	no significant difference
Pesticide residue detection frequency	11% (7 to 14) vs 46% (38 to 55)	~4x

Caveats stated by the authors themselves: I-squared above 75% for most composition parameters (76% for ascorbic acid up to 100% for carotenoids and dry matter); “strong or moderate funnel plot asymmetry consistent with publication bias was detected for approximately half of the parameters”; only 11 studies compared pesticide residues, 8 of them on a single crop species, none on cereals, oilseeds or pulses; and for cadmium, nitrate, nitrite, ascorbic acid, protein, total antioxidant activity (TEAC) and total phenolic acids the mean percentage difference fell outside the 95% CI of the standardised mean difference, which the paper flags as making those estimates “less reliable.” The flag therefore applies to the headline antioxidant result as much as to protein and cadmium, and this document applies it in all three places. The consequence is symmetric: the protein and fibre deficits, the cadmium reduction and the antioxidant increases are all estimated on the same flagged basis, so none of the three should be quoted as a precise number. Funding: EU FP6 QUALITYLOWINPUTFOOD plus the Sheepdrove Trust, an organic-farming trust; Charles Benbrook is a co-author. Peer-reviewed criticism exists: Mulet, Br J Nutr 2014;112:1745-7, objects to inclusion of pre-1992 studies and studies from countries with no organic regulation.

The protein and fibre rows are the finding that matters most for the rubric most. The rubric ranks protein second and fibre third among outcome-relevant levers. Barański’s own meta-analysis puts organic crops lower on both. Nobody frames it this way because the paper’s title is about antioxidants, but on the rubric’s own weighting the composition evidence is net-negative for organic crops, not net-positive.

Per-serving arithmetic for protein, derived here, not published. The -15% is measured on crop dry matter, and it is driven by cereals, where grain protein tracks nitrogen fertilisation directly; vegetables and fruit contribute little protein either way. Grains supply roughly 20% of protein intake in a Western diet, so about 16 g/day out of an 80 g/day intake. A 15% cut on that fraction is about -2.4 g/day; extended to all crop-derived protein (roughly 30 g/day) it is about -4.5 g/day. On a 0.8 to 1.6 g/kg/day target that is 2% to 5% of intake, the same order as the n-3 gain in milk and pointing the other way.

Średnicka-Tober 2016 (milk), Br J Nutr 115(6):1043-60, 196 publications (170 on bovine milk), 89 in the weighted meta-analysis, peer-reviewed. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4838834/) MECH

Parameter	Organic vs conventional	95% CI
Total PUFA	+7%	-1 to 15 (null)
n-3 PUFA	+56%	38 to 74
ALA	+69%	53 to 84
VLC n-3 (EPA+DPA+DHA)	+57%	27 to 87
CLA	+41%	14 to 68
Alpha-tocopherol	+13%	1 to 26
Iron	+20%	0 to 41
Iodine	-74%	-115 to -33
Selenium	-21%	-49 to +6 (null)

The authors normalise the very-long-chain fraction themselves. Verbatim: “consumption of half a litre of full-fat milk (or equivalent fat intakes with dairy products) can be estimated to provide 34 and 22% of the actual and 16% (39 mg) and 11% (25 mg) of the recommended daily VLC n-3 PUFA intake with organic and conventional milk consumption, respectively.” The VLC n-3 gain is 14 mg per day, or 5 percentage points of a recommended intake.

But VLC n-3 is not the whole n-3 difference, and quoting only it understates the switch. The headline effects are total n-3 +56% and ALA +69%. Half a litre of full-fat milk carries about 20 g fat, so roughly 19 g of fatty acids. At the ballpark shares reported for bovine milk (total n-3 around 0.5% of fatty acids conventional, around 0.8% organic) that is roughly 90 to 130 mg/day conventional against 140 to 200 mg/day organic, a gain of about 60 to 90 mg/day of total n-3, of which only 14 mg is EPA plus DPA plus DHA. This 60 to 90 mg figure is derived here from the percentage differences and typical milk fat composition; it is not a published number, and the derivation is sensitive to the assumed baseline share. The remainder is ALA, which converts to EPA at single-digit percentages and to DHA at under 1%, so it is not equivalent to the same milligrams of fish oil.

Meanwhile the same half litre delivers 53% of daily iodine from organic milk versus 88% from conventional, so for anyone whose iodine comes mainly from dairy, the switch is a net loss on a nutrient with an actual deficiency syndrome behind it. That iodine framing is UK-dairy-model specific. The UK does not iodise table salt and dairy supplies roughly a third of UK iodine intake, which is what makes the -74% load-bearing there. In countries with widespread iodised salt, or for anyone eating fish, eggs or iodised salt regularly, the caveat is much weaker.

The authors also concede “there are virtually no studies in which impacts of organic food consumption on animal or human health or health-related biomarkers were assessed.”

Średnicka-Tober 2016 (meat), Br J Nutr 115(6):994-1011, 67 publications (16 beef, 16 lamb/goat, 14 pork, 17 chicken, 3 rabbit, 1 unspecified), peer-reviewed. Total PUFA +23% (11 to 35); n-3 PUFA +47% (10 to 84). Saturated fat similar, monounsaturated similar or slightly lower. For minerals, antioxidants and most individual fatty acids “the evidence base was too weak for meaningful meta-analyses,” and the authors rate evidence strength “very low or low” for most parameters. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4838835/) MECH

Eggs: no meta-analysis of organic vs conventional egg composition surfaced in this pass. Smith-Spangler’s 223 food studies include eggs but report no significant nutrient difference. Treat organic eggs as unstudied on composition, not as demonstrated-equal.

Verdict on topic 1

Composition differences are real; nutritional relevance is unproven in both directions. The antioxidant increases (17% to 69%) sit on a nutrient class with no established dose-response to a clinical outcome, and carry the same “less reliable” flag as the deficits; polyphenol intake is not a rubric lever at all. The n-3 increases are large in percent and small in milligrams (about 60 to 90 mg/day total n-3 per half litre, derived). The differences running against organic are of the same order: protein about -2.4 to -4.5 g/day (derived, cereals-driven), fibre -8% in crops, iodine -74% in milk in a market without iodised salt.

What would change this: a composition meta-analysis restricted to post-2000 studies from regulated organic systems, category-matched, reporting protein and fibre as primary endpoints rather than as an afterthought. Or an absolute-intake modelling paper showing that the polyphenol delta moves a measured biomarker (endothelial function, LDL oxidation) at realistic servings.

2. Pesticide residues and exposure

A. How much is actually on the food

USDA Pesticide Data Program, calendar year 2024 (published Dec 2025), n=9,872 samples, 597 analytes, 21 commodities. Samples washed 15 to 20 seconds under cold running water before testing. (https://www.ams.usda.gov/sites/default/files/media/PDPAnnualSummary.pdf) MECH

No detectable residue: 42.3%
Below EPA tolerance: more than 99%
Exceeded EPA tolerance: 0.77% (76 samples); 82.9% of those imported; tomatillos alone were 48.7% of all exceedances
Residue of a pesticide with no established tolerance: 3.7% (361 samples)

EFSA, 2024 monitoring data, EFSA Journal 2026;24:10054, more than 125,000 samples. EU-coordinated programme (n=9,842, market-representative): 98.8% compliant, 43.1% no measurable residues, 2.4% above MRL, 1.2% non-compliant after measurement uncertainty. National programmes (n=86,449): 98.2% compliant. EFSA’s conclusion verbatim: “The assessment confirms a low risk to consumer health from the estimated exposure to pesticide residues in the foods tested in 2024.” (https://www.efsa.europa.eu/en/news/pesticide-residues-food-latest-data-released) MECH

Organic food also carries residues, and by more than the marketing implies.

Source	Organic detection rate	Conventional	Ratio
Barański 2014 meta-analysis, 343 pubs	11% (7 to 14)	46% (38 to 55)	~4x
Baker 2002, Food Addit Contam 19(5):427-46, PDP 1994-99	23% (13% excluding organochlorines)	73%	3.2x
USDA NOP Pilot Study 2010-11, 571 USDA-seal samples	42.7% (244/571)	n/a	n/a
EFSA 2022 report, n=6,717 organic	21.0% quantifiable, 2.4% above MRL	3.7% above MRL	~1.5x on exceedance
PDP 2024 raw database, computed for this document [A]	27.4% (20.6% excluding legacy organochlorines)	60.9%	2.2x to 2.9x

The [A] row is an agent’s own join of the USDA PDP 2024 sample and result tables and is not published; treat it as a lead. It does surface a mechanistically important pattern that Baker 2002 independently found: a substantial share of organic detections are legacy persistent organochlorines taken up from soil (dieldrin, heptachlor epoxide, cis-chlordane), banned decades ago, not contemporary sprays. Removing them dropped the organic rate from 27.4% to 20.6% and the conventional rate from 60.9% to only 60.2%.

The USDA NOP pilot figure (42.7% of certified-organic samples with detectable residues) is the number that most undercuts “organic means no residues,” and it comes from USDA’s own testing programme. (https://www.ams.usda.gov/sites/default/files/media/Pesticide%20Residue%20Testing_Org%20Produce_2010-11PilotStudy.pdf)

B. Biomarker intervention trials

Study	n / design	Duration	Result
Lu 2006, EHP 114:260-3	23 children 3-11, single-arm, no control, conventional-first	3d conv, 5d org, 7d conv	MDA median 1.5 to non-detect (p<0.01); TCPy 6.0 to 0.9 ug/L (p<0.001). 5 OP metabolites measured, only 2 moved
Oates 2014, Environ Res 132:105-11	13 Australian adults, randomized single-blind crossover, the only RCT design here	7d per phase	Total DAPs -89% (p=0.013); dimethyl -96%; diethyl -49%, not significant (p=0.170)
Bradman 2015, EHP 123:1086-93	40 Mexican-American children 3-6, 594 first-morning voids, fixed sequence, not randomized	4d conv, 7d org, 5d conv	23 analytes, only 6 analysable. Total DMs -48.7% (-65.7 to -23.2); 2,4-D -25.2%. Null: total DEs -1.2% (p=0.957); 3-PBA -13.3% (p=0.159)
Hyland 2019, Environ Res 171:568-75	16 people, 4 families, 158 samples, conventional-first, no washout	6d + 6d	14 analytes, 13 fell. Clothianidin -82.7%; MDA -95.0%; TCPy -60.7%; mean ~-60.5%
Fagan 2020, Environ Res 189:109898	same 16 people	5d + 5d	Glyphosate -70.9% (-78.0 to -61.7); AMPA -76.7%
Rempelos 2022, Am J Clin Nutr 115:364	27, parallel-group RCT (13 organic, 14 conventional)	2 weeks	Total pesticide excretion 17 vs 180 ug/day (-91%, p<0.0001)
Curl 2015, EHP 123:475-83	cross-sectional MESA, 4,466 FFQs, DAPs in 480	n/a	Produce-matched median DAPs 163 to 121 to 106 nmol/g creatinine across rarely / sometimes / often organic (p<0.02)

One trial went past exposure to a downstream biomarker. ORGANIKO (Makris 2019), Environ Int 126:591-600, PMID 31483785, n=149 Cypriot children, 40-day cluster crossover, measured 8-hydroxy-2-deoxyguanosine (8-OHdG), a urinary marker of oxidative DNA damage, alongside the pesticide metabolites: geometric mean ratio 0.888 (0.808 to 0.976) on the organic phase, roughly an 11% reduction. It is still a biomarker, 8-OHdG has no established mapping to any clinical endpoint, and the design is a school-level crossover rather than individual randomization. But it is the only result in this literature that moves something other than the exposure measure itself,

All RCT or quasi-RCT, all biomarker endpoints, none clinical.

Two structural caveats. First, design: only Oates 2014 was randomized; Lu, Bradman, Hyland and Fagan all ran conventional-first sequences with no washout and no counterbalancing. Second, and more important, Zhang 2008, J Agric Food Chem 56:10638, showed that preformed dialkyl phosphates and TCPy are already present in produce, so part of the measured “decrease” is reduced ingestion of the metabolite itself rather than of the parent pesticide. Bradman 2015 concedes this in the paper.

C. Where those exposures sit against reference doses

None of the seven intervention trials back-calculates absorbed dose against an RfD, PAD or ADI. That is the single largest gap in the pro-organic case: the relative reduction is measured precisely and the absolute dose is never anchored. The anchoring has to come from elsewhere:

Curl 2015: MESA 95th-percentile dietary OP exposure 11 ng/kg-day methamidophos equivalents, against EPA’s own 2006 OP cumulative risk assessment 95th-percentile single-day estimate of 92 ng/kg-day for adults over 50. Roughly 8x below EPA’s high-end figure.
EFSA chronic cumulative acetylcholinesterase-inhibition risk assessment, EFSA Journal 2021;19(2):6392, 47 active substances, 10 EU populations, 2016-2018 monitoring, threshold MOET 100 at the 99.9th percentile: exposure does not exceed the threshold for any of the 10 populations, but the certainty attached to that conclusion is not uniform: above 90% for the adult groups, and as low as 50% to 90% for French children, the weakest band in the assessment. A 50% to 90% certainty statement is close to a coin flip at its lower edge, and it is the single number in this document that most supports a precautionary position. (https://pmc.ncbi.nlm.nih.gov/articles/PMC7873834/)
Glyphosate: biomonitoring-derived intakes correspond to 0.1% to 0.66% of the EFSA ADI (0.5 mg/kg/day); EPA chronic RfD is 1.75 mg/kg/day.
EPA chlorpyrifos, 89 FR 99184 (2024-12-10): food residues alone were 3.2% of the aPAD (females 13 to 49) and 9.7% of the ssPAD (children 1 to 2) at the 99.9th percentile. The 2021 tolerance revocation was driven by drinking water, not food. Note the size of that second number: 9.7% of the population-adjusted dose is within about one order of magnitude of the threshold, not two to five.

Regulatory thresholds are not effect thresholds, and this document should not treat them as if they were. Three things follow.

The organophosphate RfDs and PADs are derived from cholinesterase inhibition, with 10x interspecies and 10x intraspecies uncertainty factors applied to that endpoint. They are not derived from neurodevelopment.
The CHAMACOS and Columbia neurodevelopmental associations were observed at exposures below those thresholds. That is the whole reason EPA retained the additional 10x FQPA children’s safety factor for chlorpyrifos rather than removing it: the agency concluded the cholinesterase endpoint may not be protective for the developing brain.
Mie 2017, the STOA review commissioned by the European Parliament, makes the same point directly: cognitive effects reported at current population exposure levels are not accounted for in the formal risk assessments, because those assessments key on a different endpoint. (https://www.europarl.europa.eu/RegData/etudes/STUD/2016/581922/EPRS_STU(2016)581922_EN.pdf)

Scope note on the “orders of magnitude” framing. “2 to 5 orders of magnitude below the regulatory benchmark” holds for single-compound chronic mean intakes in adults (glyphosate at 0.13% to 4.2% of the ADI, most Winter and Katz combinations below 0.01% of the RfD). It does not hold for the cumulative 99.9th-percentile toddler case, where the chlorpyrifos food-only figure is 9.7% of the ssPAD and EFSA cumulative certainty falls to 50% to 90%.

D. The EWG Dirty Dozen critique

Winter and Katz 2011, J Toxicol 2011:589674, peer-reviewed (academic editor Ian Munro). LifeLine v5.0 probabilistic model, PDP 2004-2008, 2,000 simulated individuals, USDA 1994-96/1998 consumption surveys, 12 commodities by the top 10 detected pesticides on each = 120 combinations. (https://pmc.ncbi.nlm.nih.gov/articles/PMC3135239/) MECH

0 of 120 exceeded the reference dose.
1 of 120 exceeded 1% of the RfD: methamidophos on bell peppers, at 2% of RfD, i.e. the RfD was 49.5x above the exposure.
7 of 120 (5.8%) exceeded 0.1%; 75% were below 0.01%; 40.8% below 0.001%.
The RfD exceeded exposure by more than 1,000x in over 90% of comparisons, and by more than 30,000x for every pesticide on blueberries, cherries and kale.

Conclusions verbatim: “substitution of organic forms of the twelve commodities for conventional forms does not result in any appreciable reduction of consumer risks” and the EWG “methodology used by the environmental advocacy group to rank commodities with respect to pesticide risks lacks scientific credibility.”

The methodological core of the critique: four of EWG’s six ranking indicators are variants of “count of different pesticides detected” (percent of samples with detections, percent with two or more, average number per sample, maximum number on one sample, total number found on the commodity), only one crudely reflects amount, and none reflects either consumption volume or toxicity. That mechanically ranks multi-residue commodities worst regardless of dose.

Be fair to the other side. Winter and Katz has real weaknesses: it reports mean exposures only, with no 95th or 99.9th percentile estimates anywhere; non-detects were assigned zero rather than LOD/2; only the top 10 pesticides per commodity were modelled; there is no cumulative or aggregate assessment across pesticides or exposure routes; and the PMC full text contains no limitations section and no funding or conflict-of-interest statement of any kind. The industry-funded Alliance for Food and Farming has promoted it heavily, though no evidence surfaced that the paper itself was industry-funded, and the funding could not be verified either way. No peer-reviewed EWG rebuttal exists; EWG’s responses appear only on its own site.

Independent replication: Jacobs, Kougias, Louie and Roberts 2024, Crit Rev Toxicol 54(4):215-34, peer-reviewed, applied the same approach to the 2022 Dirty Dozen list against EPA dietary health-based guidance values and found “the estimated daily exposure for each pesticide-produce combination was below the corresponding HBGV for all exposure scenarios.” Conflict flag: all four authors work at commercial risk-consulting firms, and the funding statement could not be retrieved. The authors themselves note acute and cumulative scenarios were not assessed.

E. Does dietary residue exposure at typical levels cause anything?

Strongest evidence for: the Harvard EARTH studies. Critically, the exposure metric is the Pesticide Residue Burden Score, an FFQ-derived proxy that classifies fruits and vegetables as high or low residue using PDP surveillance data. No residues were measured in any participant in any outcome paper.

Validation: Chiu 2018, J Expo Sci Environ Epidemiol 28:31-9, n=90 men: urinary biomarker sum +21% (2 to 44) per serving/day of high-residue produce. Real but weak.
Chiu 2015, Hum Reprod 30:1342-51, 155 subfertile men: Q4 vs Q1 total sperm count -49% (31 to 63).
Chiu 2018, JAMA Intern Med 178:17-26, 325 women, 541 ART cycles: live birth -26% (13 to 37).

The same instrument at much larger sample size, read as the authors read it.

Chiu 2016, J Nutr 146:1084-92, n=189 healthy young men: total sperm count was positively associated with low-to-moderate-residue fruit and vegetable intake, while high-residue intake was unrelated to semen quality. That is the same directional pattern as the 2015 subfertile-men paper, not a refutation of it: the produce benefit shows up where residues are low and is absent where they are high.
Sandoval-Insausti 2021, Environ Int 156:106744, n=180,316, 23,678 cancers, 2.86 million person-years: high-residue produce HR 0.99 (0.97 to 1.01) per serving/day.
Sandoval-Insausti 2022, Environ Int 159:107024, n=160,880, 27,026 deaths: at matched intake, low-residue produce ran 0.64 (0.59 to 0.68) for all-cause mortality while high-residue produce ran 0.93 (0.81 to 1.07). The authors’ own interpretation is not that residues are harmless but that residues may offset the benefit of the produce itself. A 0.93 next to a 0.64 at the same number of servings is a 29-point gap in the same study, not a null.
Chiu 2019, Environ Int 132:105113, n=170,142, 3,707 CHD events: high-residue HR 0.97 (0.72 to 1.30), with the same low-versus-high contrast pointing the same way and wide intervals on the high-residue arm.

The honest read of the EARTH and NHS/HPFS series: the exposure metric is an FFQ proxy with one small validation study behind it (+21% urinary biomarker sum per serving/day), so it cannot separate “residues do something” from “the high-residue basket is a different set of foods”. The results are consistent with a small residue effect that partly cancels the produce benefit, and they are equally consistent with residual confounding by food type. Neither reading is established.

Measured biomarkers linked to hard outcomes in general populations: these studies exist.

Bouchard 2010, Pediatrics 125:e1270-7, PMID 20478945, NHANES n=1,139 children 8 to 15, cross-sectional: urinary dialkyl phosphates and ADHD, OR 1.55 per 10-fold increase in dimethyl alkylphosphates (95% CI 1.14 to 2.10).
Bao 2020, JAMA Intern Med 180:367-74, PMID 31886824, NHANES n=2,116 adults, prospective mortality linkage: urinary 3-PBA (a pyrethroid metabolite) highest versus lowest tertile, all-cause mortality HR 1.56 (1.08 to 2.26) and cardiovascular mortality HR 3.00 (1.02 to 8.32).

The narrow claim that survives is: no organic-versus-conventional study links diet-derived residue biomarkers to a clinical outcome. General-population biomarker cohorts with real outcomes do exist; they are cross-sectional (Bouchard) or modest-n with wide intervals (Bao), and critically neither attributes exposure to a route, so neither can say whether the residues came from food, home pest control, or ambient sources.

Neurodevelopment: the exposure route is the whole story.

Cohort	Effect	Route
CHAMACOS, Bouchard 2011, EHP 119:1189-95, n=329	Q5 vs Q1 prenatal DAP: -7.0 FSIQ points; per 10-fold, -5.6 (-9.0 to -2.2). Postnatal child DAPs not associated	Agricultural. 44% of mothers did farm work while pregnant; maternal DAPs above general US levels
Columbia CCCEH, Rauh 2011, EHP 119:1196, n=265	Per SD cord chlorpyrifos, FSIQ -1.4%, working memory -2.8%	Residential indoor pest control, the exact use EPA cancelled in 2000-2001. Not a dietary cohort
Mount Sinai, Engel 2011, EHP 119:1182-8, n=169	Sum-DEP to FSIQ -2.89 (-6.15 to +0.36), not significant	Urban residential plus dietary
Generation R, van den Dries 2019, EHP 127:017007, n=708	“Small and imprecise”; only the >25-week window significant, -3.9 nonverbal IQ (-7.5 to -0.3)	General Dutch population

Two more measured-exposure results worth having on file.

Eskenazi 2023, Environ Res 217:114712, PMID 36856429, CHAMACOS: glyphosate and AMPA measured in urine in a farmworker-community cohort, extending the CHAMACOS exposure series beyond organophosphates. Same route caveat as the rest of CHAMACOS: agricultural community, not a dietary cohort.
Craddock 2019, Environ Health 18:7, PMID 30634980: a systematic review of neonicotinoid residues, reporting detections across USDA PDP produce sampling and noting the near-total absence of human outcome data for this class. It is the pesticide class where the residue-to-outcome literature is thinnest, and it is also the class with the largest measured drop in the organic feeding trials: clothianidin -82.7% in Hyland 2019.

Chlorpyrifos regulatory history, since it is often cited as proof the system failed: tolerances revoked 86 FR 48315 (2021-08-30); revocation vacated as arbitrary and capricious in Red River Valley Sugarbeet Growers Ass’n v. Regan, 85 F.4th 881 (8th Cir., 2023-11-02); tolerances reinstated 89 FR 7625 (2024-02-05); partial revocation proposed 89 FR 99184 (2024-12-10) sparing 11 crops; no final rule as of 2026-09.

Verdict on topic 2

The exposure reduction is genuine and no clinical outcome has been attached to it. The relative reduction is large (40% to 96% on biomarkers, 4x on detection frequency). The absolute dose is 2 to 5 orders of magnitude below the regulatory anchors for single compounds, at chronic mean intakes, in adults, and far closer than that in the cumulative 99.9th-percentile toddler case (9.7% of the chlorpyrifos ssPAD from food alone; EFSA cumulative certainty down to 50% to 90% for French children). The outcome literature that appears to support a dietary effect rests on an FFQ proxy, and its low-versus-high-residue contrast survives in the large cohorts even though the high-residue arm alone is imprecise.

What would change this: a prospective cohort with measured urinary pesticide biomarkers (not an FFQ proxy) in a general non-agricultural population, linked to a hard endpoint, with enough follow-up for cancer latency. Or an EPA or EFSA cumulative assessment that finds a population subgroup exceeding a threshold from food alone. The EFSA French-children certainty band (50% to 90%) is the place to watch. Also decisive would be a route-attribution study: Bouchard 2010 and Bao 2020 link measured biomarkers to outcomes but cannot say how much of the biomarker came from food.

3. Health outcomes in organic consumers

Cancer

Baudry 2018, JAMA Intern Med 178:1597-1606, NutriNet-Sante, n=68,946 (78% female), mean follow-up 4.56 years (SD 2.08), 1,340 incident cancers, peer-reviewed. Organic score 0 to 32 across 16 product types. (https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2707948) LTSP

Model 1 (age, sex): HR 0.70 (0.60 to 0.83)
Model 2 (+ occupation, education, marital status, income, physical activity, smoking, alcohol, family cancer history, BMI, height, energy, mPNNS-GS diet quality score, fibre, processed meat, red meat, parity, menopause, HRT, OCP): 0.75 (0.63 to 0.88), p-trend .001
Model 3 (+ ultra-processed food, fruit and vegetable intake, PCA dietary patterns): 0.76 (0.64 to 0.90), p-trend .003
Absolute risk reduction 0.6%.

Site-specific, Q4 vs Q1: postmenopausal breast 0.66 (0.45 to 0.96); non-Hodgkin lymphoma 0.14 (0.03 to 0.66); prostate 1.00 (0.63 to 1.60); colorectal 0.87 (0.48 to 1.57).

The NHL result rests on 2 cases in Q4 out of 47 NHL cases total, with a non-monotone gradient across quartiles (1.00, 0.98, 1.19, then 0.14). It is not an effect estimate in any useful sense.

Buried in the discussion and rarely quoted: the association was not significant among participants with high overall dietary quality, nor in men, nor in never smokers, current smokers, younger adults, or those without family cancer history. Under a pesticide-toxicity hypothesis the effect should persist among good eaters; under a diet-quality-confounding hypothesis it should vanish there, and it vanished. Conflict disclosure: Lairon serves as scientific expert to two French foundations funded by organic-food interests.

The accompanying editorial (Hemler, Chavarro, Hu, JAMA Intern Med 2018;178:1606) notes organic intake “is notoriously difficult to assess, and its self-report is highly susceptible to confounding by positive health behaviors and socioeconomic factors.”

Bradbury 2014, Br J Cancer 110:2321-6, Million Women Study, n=623,080, mean follow-up 9.3 years, 53,769 cancers, peer-reviewed. Usually/always vs never organic: all cancer RR 1.03 (0.99 to 1.07); breast 1.09 (1.02 to 1.15), i.e. the wrong direction; NHL 0.79 (0.65 to 0.96); soft tissue sarcoma 1.37 (0.82 to 2.27). Only 2 of 16 sites were significant, in opposite directions; the authors explicitly invoke multiple testing. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4007233/) LTSP

Andersen 2023, Eur J Epidemiol 38:315, Danish Diet Cancer and Health, n=41,928, median 15 years, 9,675 cancers. Overall cancer null. Two site-specific results reached significance and they point in opposite directions: NHL 1.97 (1.28 to 3.04), p-trend .05 toward higher risk, and stomach cancer 0.50 (0.32 to 0.78) toward lower risk. (https://link.springer.com/article/10.1007/s10654-022-00951-9) LTSP

Three cohorts, three different NHL answers: 0.14, 0.79, 1.97.

Multiple testing is the explanation for all of them, applied symmetrically. Each cohort tested roughly a dozen to sixteen cancer sites. At that many tests, one or two significant results per cohort is the expectation under the null. That reading has to cover the Danish NHL 1.97 and stomach 0.50, the French NHL 0.14 (2 cases in Q4), and the Million Women breast 1.09 equally. None of the four is an effect estimate that should be quoted on its own, and this document should not lean on the ones that happen to support its conclusion while dismissing the ones that do not.

Meta-analyses. Theodoridis 2025, Life (Basel) 15:160, PMID 40003569, 3 cohorts (277,410 in the quantitative analysis): overall cancer HR 0.93 (0.78 to 1.12), I-squared 84%; breast 1.01 (0.81 to 1.26); colorectal 1.01 (0.93 to 1.10); NHL 0.70 (0.17 to 2.94), I-squared 90%. All null. (https://pmc.ncbi.nlm.nih.gov/articles/PMC11856173/)

Kasper 2025, PMID 40630609: the organic relative-effect estimate is 1.03 (1.00 to 1.05), pooled from 2 studies, with GRADE certainty “low”. The E-values of 1.06 (incidence) and 1.04 (mortality) that circulate with this paper belong to its pooled sustainable-diet estimates, not to the organic estimate. No published E-value exists for the organic estimate specifically. (https://pmc.ncbi.nlm.nih.gov/articles/PMC12235399/) LTSP

Metabolic syndrome, obesity, type 2 diabetes

Baudry 2018, Eur J Nutr 57:2477, cross-sectional, n=8,174: MetS prevalence ratio T3 vs T1 0.69 (0.61 to 0.78).
Kesse-Guyot 2017, Br J Nutr 117:325, prospective, n=62,224, mean 3.1 years: overweight OR 0.77 (0.68 to 0.86), obesity OR 0.69 (0.58 to 0.82).
Kesse-Guyot 2020, IJBNPA 17:136, n=33,256, 293 T2D cases, 4.05 years: Q5 vs Q1 0.65 (0.43 to 0.97). But women 0.35 (0.19 to 0.63) versus men 1.61 (0.89 to 2.91), an implausible sex reversal for a pesticide mechanism, and sensitivity analyses excluding early cases were “strongly attenuated.” (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7653706/)
Andersen 2023, Diabetes Res Clin Pract 205:110972, n=41,286, 4,843 T2D cases: highest vs never, women 0.88 (0.74 to 1.05), men 0.89 (0.75 to 1.05). Null.
Aljahdali 2022, Eur J Nutr 61:1255, US HRS/HCNS: crude models inverse, “No significant associations were detected in the fully adjusted models.”

Pattern: the two analyses with the tightest confounder control are the two that go null. All LTSP.

Children: allergy and eczema

Kummeling 2008, Br J Nutr 99:598, KOALA birth cohort, n=2,764, outcomes to age 2. Result verbatim: “Consumption of organic dairy products was associated with lower eczema risk (OR 0.64 (95% CI 0.44, 0.93)), but there was no association of organic meat, fruit, vegetables or eggs, or the proportion of organic products within the total diet with the development of eczema, wheeze or atopic sensitisation.” (https://pubmed.ncbi.nlm.nih.gov/17761012/) LTSP

The aggregate exposure was null. One of six food categories was significant, and it is the one category least relevant to pesticide residues.

PARSIFAL. Alfven 2006, Allergy 61:414, cross-sectional, n=14,893, ages 5 to 13, five countries: Steiner/anthroposophic children rhinoconjunctivitis aOR 0.69 (0.56 to 0.86), sensitisation 0.73 (0.58 to 0.92), described by the authors as “less pronounced and not as consistent between countries” than the farm-child effect (0.50 and 0.53).

Flöistrup 2006, JACI 117:59, n=6,630, decomposes the lifestyle and answers the attribution question directly. The significant components are antibiotics in year 1 (asthma 2.79, 2.03 to 3.83; rhinoconjunctivitis 1.97, 1.26 to 3.08; eczema 1.63, 1.22 to 2.17), early antipyretics, and MMR vaccination. Conclusion verbatim: “Certain features of the anthroposophic lifestyle, such as restrictive use of antibiotics and antipyretics, are associated with a reduced risk of allergic disease.” Biodynamic and organic food is not among the identified protective factors. (https://pubmed.ncbi.nlm.nih.gov/16387585/) LTSP

Waser 2007, Clin Exp Allergy 37:661, same cohort: the dietary signal is farm milk (asthma aOR 0.74, 0.61 to 0.88), and “other farm-produced products were not independently related to any allergy-related health outcome.” Raw farm milk is a distinct exposure from organic certification.

Later: Pernin-Schneider 2025, Allergy, PARIS cohort n=1,258: sensitisation to any allergen aOR 0.60 (0.40 to 0.91) but no association with asthma, eczema, rhinitis or food sensitisation. No RCT with an allergy endpoint exists.

Pregnancy

Torjusen 2014, BMJ Open 4:e006143, MoBa, n=28,192 nulliparous, 1,491 pre-eclampsia cases (5.3%). High organic vegetable intake (n=1,951): crude 0.76 (0.61 to 0.96); adjusted 0.79 (0.62 to 0.99), p=.043. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4160835/) LTSP

The other six organic exposures were all null: fruit 0.92, cereals 1.02, milk/dairy 1.03, eggs 0.93, meat 1.09, and the combined organic index 0.92 (0.76 to 1.12). One of seven, no multiplicity correction, unvalidated organic questions, and severe pre-eclampsia/HELLP null at 0.97 (0.65 to 1.43). In the same table the healthy-food-pattern tertile 3 vs 1 was 0.74 (0.64 to 0.85), a larger and far tighter effect than the organic variable.

Twelve years, zero replications, verified against Liu 2023, Adv Nutr 14:12, a systematic table of all six observational organic-in-pregnancy studies.

Brantsaeter 2016, EHP 124:357, MoBa, n=35,107 male singletons, 74 hypospadias cases: any organic adjusted 0.42 (0.25 to 0.70) on 21 exposed cases; fruit and eggs flip sign on full adjustment (1.15 and 1.40); cryptorchidism null (0.91, 0.66 to 1.26). Authors call for replication. (https://pmc.ncbi.nlm.nih.gov/articles/PMC4786987)

Trials

Zero randomized trials of organic versus conventional diet with a clinical endpoint have ever been run. Confirmed against PubMed, ClinicalTrials.gov API (4 registered organic-diet RCTs, all biomarker-primary), and four systematic reviews. No Cochrane review exists.

Every existing RCT primary endpoint is a biomarker: ORGANIKO (Makris 2019, PLoS ONE, n=149 children, 40-day crossover, urinary 3-PBA GMR 0.297); Curl 2019, Environ Int, n=20 pregnant, 24 weeks, the longest ever run, whose paper states “we are not evaluating health effects”; Rempelos 2022 AJCN 115:364 (pesticide excretion -91%) and its companion AJCN 116:1278 on micronutrients, which was null; Grinder-Pedersen 2003, J Agric Food Chem 51:5671, n=16, 22-day crossover, quercetin +60%, most antioxidant markers null, and protein oxidation higher on organic; OrgTrace (NCT00738166), n=33, null for zinc, copper and carotenoids. RCT

Confounding, which is the decisive section

Baudry Q1 to Q4 baseline gradient (all p<.001 unless noted):

	Q1 (least organic)	Q4 (most organic)
BMI	24.46	22.92
Postsecondary education	56.9%	71.1%
Income above 2,700 EUR	18.8%	28.2%
Current smoker	16.4%	13.8%
Fibre, g/day	17.88	22.60
Red meat, g/day	48.72	31.44
Processed meat, g/day	23.67	15.12
mPNNS-GS diet score	7.41	8.19
High physical activity	24.0%	17.2% (p=.03)

Note the physical activity gradient runs backwards: Q4 organic consumers were less active. Secondary sources claiming otherwise are wrong; this is from the published Table 1.

Million Women: usually/always vs never organic: BMI 25.4 vs 26.3; current smoker 11% vs 16%; strenuous activity more than weekly 55% vs 35%; red and processed meat 43% vs 58%; fibre 15.0 vs 12.8 g/day.

E-values. No published E-value exists for Baudry, nor for the organic estimate in Kasper 2025. Computed for this document (VanderWeele-Ding, from the published point estimates): overall Q4 HR 0.76 gives E = 1.96, and the CI bound 0.90 gives E = 1.46.

The bias-factor arithmetic. An unmeasured confounder at RR 1.5 on both the exposure and the outcome has a bias factor of 1.125 (1.5 x 1.5 / (1.5 + 1.5 - 1)). Applied to Baudry’s CI bound of 0.90 that gives 1.01, so it moves the interval to null. Applied to the point estimate of 0.76 it gives 0.86, so it does not erase the point estimate; that needs roughly RR 1.96 on both, which is a large confounder.

The measured gradients are not the candidate confounder. Red meat, income, education, BMI, smoking, physical activity, fibre and diet quality are all measured and adjusted for in Baudry’s Model 2 and Model 3. Their baseline gradients show that the organic-consumer population differs, which is why residual confounding is plausible, but they are not themselves candidates for the unmeasured confounder an E-value describes. The E-value question is what remains after that adjustment, and the honest answer is that nobody knows.

Healthy-user-bias benchmark. Feng/Wang 2022, Front Nutr 9:831470, UK Biobank n approximately 400,000: adjustment for confounders reduced the likelihood-ratio statistics for raw vegetable intake and CVD by 82% (incidence) and 87% (mortality); the authors conclude “residual confounding is likely to account for much, if not all, of the observed associations.” (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8901125/)

Negative-control-style checks. No negative-control-outcome analysis has been published for any organic cohort. The closest thing available is a within-family design: Instanes 2024, BMC Medicine 22, MoBa n=40,707, outcome ADHD and autism-spectrum symptom scores at age 8 (maternal-report scales, not diagnoses). (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11492991/) LTSP

What it actually shows, stated carefully:

Between-family, organic food score was associated with slightly worse ADHD (beta +0.03) and ASD (beta +0.07) symptom scores.
The sibling comparison rests on 5,534 discordant siblings, 13.6% of the sample, and the intervals widen accordingly. Sibling ADHD beta -0.07 (-0.15 to 0.01), ASD -0.001 (-0.12 to 0.12).
In the low range of organic intake the ASD sibling estimate is beta -0.09, pointing protective.

The right reading: the only within-family design in this literature is uninformative rather than confirmatory. It measures neurodevelopmental symptom scores, not cancer or any other endpoint elsewhere in this document; the sibling sample is a small discordant subset with intervals wide enough to contain moderate effects in both directions; and one of its own sub-estimates points protective. It does not transfer to other outcomes.

One honest point for the other side: MoBa does not fit the healthy-user template cleanly. Torjusen 2010, BMC Public Health 10:775, found frequent organic consumers in MoBa were more likely to smoke, drink alcohol and be in the lowest income group, while also being more likely to be highly educated and to eat a diet scoring better on the cohort’s own healthy-food pattern. The net confounding direction in MoBa is mixed and undetermined, not reversed, which weakens the standard healthy-user rebuttal without turning it around. So the MoBa pre-eclampsia and hypospadias findings are not straightforwardly explained by classical healthy-user bias. That is the best structural argument in the pro-organic literature.

Verdict on topic 3

The outcome literature does not support a health benefit, and its internal contradictions are the signature of multiple testing across roughly 16 cancer sites in three cohorts. Pooled results are null; the largest cohort (623,080) is null; the longest (15 years) is null; site-specific findings reverse sign between cohorts; aggregate organic exposures are consistently null while individual subcategories pop; and the only within-family analysis is uninformative rather than confirmatory. The published E-values of 1.04 and 1.06 belong to Kasper’s pooled sustainable-diet estimates, not to the organic estimate, and no E-value has been published for organic specifically.

What would change this: a replication of the MoBa pre-eclampsia finding in a cohort where organic consumers are not the healthy-user group, or a within-family/sibling analysis with enough discordant pairs to have real precision, in either direction. A negative-control-outcome analysis on Baudry (pick an outcome organic food cannot plausibly cause and show it is null) would be cheap and decisive; nobody has run one.

4. Contaminants

Cadmium

Barański 2014: MPD -48% (-112 to +16), CI crosses zero; significance came only from the standardised mean difference, SMD -1.45 (-2.52 to -0.39). The paper itself flags cadmium as one of the parameters where “MPD were outside the 95% CI of SMD, and therefore these should be seen as less reliable.” When crop types were analysed separately, the cadmium difference was significant for cereals only, not vegetables or fruits. GRADE reliability: moderate. MECH

Exposure context: EFSA CONTAM TWI 2.5 ug/kg bw/week. EFSA 2012, EFSA J 10(1):2551: middle-bound European weekly average 2.04 ug/kg bw, P95 3.66, toddlers 4.85 (nearly 2x the TWI). Dietary contributions: grains 26.9%, vegetables 16.0%, starchy roots and tubers 13.2%. Ferrari 2013, Food Addit Contam A 30(4):687-97, probabilistic across 14 national surveys: 14.8% to 31.2% of the population exceeds the TWI. (https://www.efsa.europa.eu/en/efsajournal/pub/2551)

Arithmetic, mine, not a measured finding: crop groups are roughly 56% of European cadmium intake, so a full 48% cut across all of them would move the mean from 2.04 to about 1.5 ug/kg bw/week; restricted to cereals, where the finding is actually significant (26.9% of intake), it moves 2.04 to about 1.78. The mean is already below the TWI either way. The effect lands on the 15% to 31% upper tail and on toddlers, not on a typical adult. This is the strongest single personal-health case for organic in the entire document, and it is still a “less reliable” effect estimate applied to a tail.

Nitrate

Barański 2014: nitrate -30% (-144 to +84), nitrite -87% (-225 to +52), both detected only by the unweighted meta-analysis and both flagged “less reliable.”

But lower nitrate is not a benefit. EFSA 2008, EFSA J 689:1-79: ADI 3.7 mg/kg bw/day = 222 mg/day for a 60 kg adult; 400 g of mixed vegetables at median nitrate gives 157 mg/day, within the ADI; exceedance is confined to about 2.5% of the population in some Member States eating only leafy vegetables in high amounts. Conclusion verbatim: “Epidemiological studies do not suggest that nitrate intake from diet or drinking water is associated with increased cancer risk” and “the estimated exposures to nitrate from vegetables are unlikely to result in appreciable health risks, therefore the recognised beneficial effects of consumption of vegetables prevail.”

And the RCT evidence runs the other way. Siervo 2013, J Nutr 143:818-26, 16 crossover RCTs, n=254: systolic BP -4.4 mmHg (-5.9 to -2.8), p<0.001. Bahadoran 2017, Adv Nutr 8:830-8, 22 RCTs: SBP -3.55 (-4.55 to -2.54), DBP -1.32 (-1.97 to -0.68). RCT

Net read: the lower nitrate in organic vegetables is a small negative, not a positive. The nitrosamine concern is a processed-meat issue (IARC Group 1), which the rubric already handles in the processing block. This is the same sign-inversion trap the rubric documents for “uncured” meat, running in the opposite direction.

Mycotoxins

Wang 2024, Compr Rev Food Sci Food Saf 23(3):e13363, systematic review and meta-analysis, peer-reviewed. The standard weighted meta-analysis of concentrations found a significant production-system effect only for deoxynivalenol, with concentrations about 50% HIGHER in conventional cereals (p<0.0001). Conclusion: “contamination levels are similar in organic and conventional cereals used for human consumption,” and keeping ochratoxin A under the EU limit of 3.0 ug/kg is a challenge for both systems. Per-toxin directions for OTA, aflatoxin and T-2/HT-2 could not be verified beyond the abstract. (https://pubmed.ncbi.nlm.nih.gov/38720588/) MECH

The “no fungicides means more mycotoxins” hypothesis is not supported. It was a reasonable prior; the data do not bear it out.

Pathogens

Smith-Spangler 2012, verbatim: “Escherichia coli contamination risk did not differ between organic and conventional produce” and “Bacterial contamination of retail chicken and pork was common but unrelated to farming method.” No difference in symptomatic Campylobacter infection.

Outbreaks: Harvey, Zakhour and Gould, J Food Prot 2016;79(11):1953-8, found 18 outbreaks, 779 illnesses, 258 hospitalisations, 3 deaths, 1992-2014 linked to organic foods, but state “our study cannot ascribe risk of foodborne outbreaks based on production method.” There is no denominator, so no rate comparison is possible. Manure timing is regulated (USDA NOP: 90 or 120 days between raw manure application and harvest). (https://pmc.ncbi.nlm.nih.gov/articles/PMC7881495/) LTSP/MECH

Antibiotic-resistant bacteria on meat

This is the one contaminant axis where organic wins cleanly on measurement.

Smith-Spangler 2012: risk of isolating bacteria resistant to 3 or more antibiotics was higher in conventional chicken and pork, risk difference 33% (21% to 45%).
Innes 2021, EHP 129(5), NARMS 2012-2017, 39,349 retail meat samples, 342 facilities: multidrug-resistant organism prevalence 3.9% conventional versus 0.90% organic; adjusted prevalence ratio 0.43 (0.30 to 0.63). (https://pmc.ncbi.nlm.nih.gov/articles/PMC8114881/) MECH

The contrast is not what it looks like. The reported comparison is organic product processed at split (organic-and-conventional) facilities versus conventional product processed at conventional-only facilities, and the dataset contains only 3 organic-only facilities. So the estimate bundles the production standard together with the processing facility, and the facility effect cannot be separated from the farming effect. The direction is almost certainly right; the magnitude is entangled.
The upstream driver, for scale. FDA’s 2023 Summary Report on antimicrobials sold for use in food-producing animals is the quantity side of the same question: US sales of medically important antimicrobials for food animals have sat in the range of roughly 6 million kg per year since the drop that followed the 2017 veterinary-feed-directive rules, with cattle and swine taking the large majority. [R] recalled, not fetched for this document; check the current Summary Report before quoting the tonnage. Organic certification forbids routine antimicrobial use outright, which is what makes the stewardship argument a quantity argument rather than a residue one.

Does it translate to personal infection risk? The best-powered test says largely no. Day 2019, Lancet Infect Dis 19:1325-35: 20,243 human faecal samples, retail chicken 104/159 (65%) ESBL-positive, 0/400 fruit and vegetable samples; but ST131 accounted for 188/293 (64%) of human bloodstream isolates and only 2 of 218 food and veterinary isolates. Conclusion verbatim: “non-human reservoirs made little contribution to invasive human disease.” (https://pubmed.ncbi.nlm.nih.gov/31653524/) LTSP

So the AMR case for organic meat is a public-goods argument about resistance ecology, not a personal-health argument about your next infection. That is still a good reason to buy it. It is just not the reason usually given.

Hormones

rBST milk. FDA: “bST is a large protein. Like most dietary proteins, bST is degraded by digestive enzymes in the gastrointestinal tract and not absorbed intact.” The EU ban (Council Decision 1999/879/EC) is on animal-welfare grounds, not consumer safety. (https://www.fda.gov/animal-veterinary/product-safety-information/bovine-somatotropin-bst)

Measured retail milk. Vicini 2008, J Am Diet Assoc 108:1198-1203, n=334 retail samples across 48 states:

Analyte	Conventional	rbST-free	Organic	p
bST	~0.005 ng/mL	~0.005	~0.005	ns
IGF-1	3.12 ng/mL	3.14	2.73	<0.05
Progesterone	12.0 ng/mL	12.8	13.9	0.019
Estradiol	4.97 pg/mL	6.63	6.40	0.045

On the two steroid hormones actually measured, organic milk was HIGHER, not lower. Milk raises circulating IGF-1 by about 13.8 ng/mL across 8 RCTs (https://pubmed.ncbi.nlm.nih.gov/19746296/), attributed to protein and amino acids driving endogenous IGF-1, not to ingested bovine IGF-1. MECH

Beef implants. Cimino 2024, J Expo Sci Environ Epidemiol, 397 retail beef samples (321 products), LC-MS/MS plus NHANES intake modelling: progesterone detected in 24%, melengestrol acetate 18%, epitestosterone 17%, testosterone 6%; estradiol, estradiol benzoate, trenbolone and zeranol had no or single detections. Maximum hazard quotient under usual intake 0.29 (MGA, 99th percentile, boys 1 to 5); no hazard quotient exceeded 1.0. (https://pmc.ncbi.nlm.nih.gov/articles/PMC12069096/) MECH

No study compares organic and conventional dairy or beef consumers on any hormone-mediated endpoint. Note the jurisdiction quirk: the EU bans growth implants outright (96/22/EC), so EU conventional beef is equivalent to US organic on this axis.

Glyphosate

Residues. FDA Pesticide Residue Monitoring Program: FY2022 animal food, 103 tested, 65 detected (63.1%), 0 violative; FY2021, 38 tested, 18 detected, 0 violative. Zero glyphosate tolerance violations across FY2019, 2021, 2022.

Dose. EPA chronic RfD 1.75 mg/kg bw/day; EFSA ADI 0.5 mg/kg bw/day (2023 renewal, NOAEL 53 mg/kg bw/day from a 90-day dog study). From Vicini 2021, Compr Rev Food Sci Food Saf 20:1243-1300: JMPR 2019 across 17 GEMS cluster diets gives 0.7% to 4.2% of the ADI, median 2.7%; EFSA-guidance probabilistic modelling at the 99.9th percentile gives 0.58% (adults) and 0.90% (children 2 to 6); back-calculation from urinary biomarkers gives about 0.13% of the ADI. MECH

Hazard classification divergence. IARC Monograph Vol 112 (2015): Group 2A, limited human evidence (NHL), sufficient animal, strong mechanistic. EPA and EFSA: not likely carcinogenic at expected exposures. Four concrete reasons for the split: (1) hazard versus risk, IARC does not model dose; (2) evidence pool, IARC used published literature only while EPA and EFSA also used unpublished registrant GLP studies; (3) active ingredient versus formulation, much of the genotoxicity evidence comes from POEA-containing formulations, not glyphosate alone; (4) statistics, trend tests versus pairwise, historical controls, multiplicity correction. EPA’s 2020 interim decision had its human-health portion vacated by the Ninth Circuit (NRDC v. EPA, No. 20-70787, 2022-06-17) and EPA withdrew it in Sept 2022. EFSA 2023 found no critical area of concern.

Epidemiology. Andreotti 2018, JNCI 110:509-16, Agricultural Health Study, prospective, n=54,251 applicators, 440 exposed NHL cases: NHL at the highest quartile of intensity-weighted lifetime days, RR 0.87 (0.64 to 1.20), p-trend 0.95. Zhang 2019, Mutat Res Rev, 6 studies: meta-RR for NHL at highest cumulative exposure 1.41 (1.13 to 1.75); structural criticism is that the a priori rule selected, from each study, the highest cumulative exposure with the longest lag, which biases upward when studies report many metrics. LTSP

Applicator versus dietary, the decisive gap. Acquavella 2004, EHP 112:321-6 (funded by Monsanto, with the authors’ affiliation and funding disclosed in the paper; the exposure measurements themselves are straightforward biomonitoring and no reanalysis has challenged them): on application day 60% of farmers had detectable urinary glyphosate (geometric mean 3 ppb, max 233 ppb); spouses 4% (max 3 ppb). NHANES 2013-2014 general population: adjusted geometric mean 0.37 to 0.51 ug/L. That is roughly 2 to 3x at the geometric mean and about 500x at the applicator maximum. Fagan 2020 shows an organic diet cuts urinary glyphosate by 70.9%, moving an exposure from about 0.13% of the ADI to about 0.04% of it. Precisely measured, toxicologically trivial. The open glyphosate question is occupational, not dietary.

Verdict on topic 4

Organic is better on antibiotic-resistant bacteria (aPR 0.43, facility effect entangled) and probably better on cadmium in cereals; equal on pathogens and mycotoxins; and worse or neutral on nitrate, hormones in milk, and iodine. Glyphosate is a real regulatory controversy about applicators that has almost nothing to do with what is on your plate.

What would change this: a cadmium biomarker study (urinary or blood Cd) in matched organic versus conventional eaters, which nobody has done and which is entirely feasible. On AMR, a study linking retail-meat resistome to human colonisation and then to invasive disease in the same population; Day 2019 suggests the chain breaks at the second link.

5. Ultra-processed organic: is the label a nutrition signal?

By regulation, no, and it is worth stating precisely why

7 CFR 205.301: “organic” means at least 95% organic agricultural ingredients by weight, excluding water and salt; “made with organic” means at least 70%. There is no sugar, sodium, fat or calorie criterion anywhere in the rule, and because salt is excluded from the denominator, sodium is formally unbounded even in a “100 percent organic” product. 7 CFR 205.605 (the National List) explicitly permits carrageenan, xanthan and gellan gum, mono- and di-glycerides, alginates, sodium citrate and silicon dioxide. Organic cane sugar and organic palm oil are fully compliant.

This is a structural point the rubric already knows in another form. Organic certification is an input-method certification. It says nothing about composition, exactly as “uncured” says nothing about nitrite.

Empirically, a weak positive signal that is mostly category composition

Meadows 2021, Nutrients 13(9):3020, Label Insight/NielsenIQ data covering roughly 85% of US packaged sales, n=8,240 organic vs 72,205 conventional, per 100 g:

	Organic	Conventional	Delta
Energy	202 kcal	265 kcal	-63
Total sugar	11.3 g	16.4 g	-5.1
Added sugar	7.2 g	13.5 to 14.2 g	-6.3 to -7.0
Saturated fat	2.7 to 3.0 g	4.1 to 4.6 g	-1.4 to -1.9
Sodium	273 mg	469 mg	-196

Trans fat present in 8% vs 40%. Odds of a product being organic: per ultra-processed ingredient OR 0.68, per cosmetic additive OR 0.63, per 10 g added sugar OR 0.88, all p<0.001. (https://pmc.ncbi.nlm.nih.gov/articles/PMC8469099/) MECH

Two disqualifying caveats. The authors are Environmental Working Group staff (EWG has taken disclosed organic-industry donations), and critically the design does not match on food category, so the deltas partly reflect organic’s skew toward grain- and produce-based categories rather than like-for-like reformulation of the same product.

European product-level data with NOVA and nutrient profiling:

Richonnet 2021, Nutrients 14(1):171, n=1,155 French child-marketed products (organic n=199): NOVA 4 was 65.3% of organic versus 92.6% of non-organic; NOVA 1 was 26.1% versus 2.5%. But 94.9% of the whole sample failed the WHO Europe nutrient profile model. (https://pmc.ncbi.nlm.nih.gov/articles/PMC8747148/)
Pellegrino 2024, Nutrients 16(11):1656, n=735 Swiss child-marketed products: WHO NPM compliance 21.6% organic versus 4.9% non-organic (p<0.001). But 75% of organic products were still ultra-processed and 78% still failed the nutrient profile model. (https://pmc.ncbi.nlm.nih.gov/articles/PMC11175003/)

No published Open Food Facts analysis stratifying Nutri-Score or NOVA by organic label was found, and no USDA Branded Food Products Database or Mintel analysis. This is a real gap.

The health halo, which is the actionable finding here

Schuldt and Schwarz 2010, Judgment and Decision Making 5(3):144-50. Study 1, n=114, both groups shown the identical 160 kcal panel: relative-calorie rating 3.94 (organic) versus 5.17, F(1,112)=26.17, p<.001, d=0.97; “how often should this be eaten” 3.68 versus 2.76, d=0.89. Study 2, n=214: leniency about skipping exercise, d=0.27. (https://www.sas.upenn.edu/~baron/journal/10/10509/jdm10509.pdf) RCT (these are randomized label manipulations, so they are experimental evidence)

Durand 2025, Food Qual Prefer, systematic review and meta-analysis, 10 articles / 21 studies: pooled r = 0.40. Besson 2025, J Hum Nutr Diet, n=198: the organic label caused calorie underestimation for high-calorie items, and frequent nutrition-label readers showed a stronger halo.

Do not cite Lee, Shimizu, Kniffin and Wansink 2013, Food Qual Prefer 29(1):33-39. It is not formally retracted (Crossref and the Retraction Watch database were checked directly on 2026-09-07), but two methodologically near-identical Wansink label-bias-taste papers have been retracted.

Verdict on topic 5

“Organic” on a packaged food is a weak positive marker at the population level and no information at all about a specific product. It is a category proxy (organic products skew toward grain and produce categories), not a formulation signal, and 75% of organic child-marketed products are still ultra-processed. Against that weak signal sits an experimentally demonstrated d = 0.97 calorie underestimation from the label itself, which is one of the largest effect sizes in this entire document.

What would change this: a category-matched analysis (same NOVA subgroup, same shelf, organic versus conventional) on added sugar, sodium and energy density. Meadows 2021 has the data to do it and did not.

6. The price premium and what else the money buys

USDA ERS, Carlson and Jaenicke, ERR-209 (May 2016), hedonic model isolating the organic attribute from package size, type, store type, month, region and brand; Nielsen Homescan, roughly 40,000 households (2004-06) rising to 60,000 (2007-10). (https://ers.usda.gov/sites/default/files/_laserfiche/publications/45547/59472_err209.pdf)

2010 premiums, percent above nonorganic:

Product	Premium	Product	Premium
Spinach	7%	Celery	44%
Granola	22%	Coffee	47%
Carrots	27%	Yogurt	52%
Potatoes	28%	Spaghetti sauce	53%
Apples	29%	Canned beans	54%
Baby food	29 to 31%	Salad mix	60%
Bread	30%	Milk	72%
Soup	33%	Eggs	82%

All 17 products cost more organic; 16 of 17 exceeded a 20% premium. Meat and poultry could not be analysed (sold by weight without UPC). USDA ERS Organic Situation Report, 2025 (EIB-281) gives a current range of “less than 10 percent to more than 120 percent.” No post-2010 US retail hedonic replication exists.

What the same money buys instead. USDA ERS: meeting the fruit and vegetable recommendation costs $2.50 to $3.00 per day (2022 prices, 155 items); 77% of vegetables cost under $0.80 per cup-equivalent; dried pinto beans are $0.17 per cup-equivalent. CDC MMWR, 2019 BRFSS, n=294,566: only 12.3% of US adults meet the fruit recommendation and 10.0% meet the vegetable recommendation. NHANES 2007-2010, n=8,957: 39.9% rate cost “very important” in food shopping, rising to 60.2% below 130% of the federal poverty level.

And the outcome evidence for the substitute is in a different league:

Aune 2017, Int J Epidemiol 46(3):1029-56, 95 studies, 142 publications, per 200 g/day of fruit and vegetables:

Outcome	RR (95% CI)	I-squared	studies
Coronary heart disease	0.92 (0.90 to 0.94)	0%	15
Stroke	0.84 (0.76 to 0.92)	73%	10
Cardiovascular disease	0.92 (0.90 to 0.95)	31%	13
Total cancer	0.97 (0.95 to 0.99)	49%	12
All-cause mortality	0.90 (0.87 to 0.93)	83%	15

Risk reductions continue to 800 g/day. (https://pubmed.ncbi.nlm.nih.gov/28338764/) LTSP

Reynolds 2019, Lancet 393:434-45, 185 prospective studies (about 135 million person-years) plus 58 RCTs (n=4,635), per 8 g/day of fibre: all-cause mortality 0.93 (0.90 to 0.95), CHD 0.81 (0.73 to 0.90), type 2 diabetes 0.85 (0.82 to 0.89), colorectal cancer 0.92 (0.89 to 0.95). The RCT arm gives body weight -0.37 kg (-0.63 to -0.11), GRADE High and systolic BP -1.27 mmHg. Already the top-ranked evidence in the project rubric. LTSP + RCT

Honest limit on the substitution argument: Aune has no RCT arm, Reynolds’ RCT arm measures surrogates only, and neither meta-analysis stratified by organic status. But that is the comparison that matters: the fibre and produce literature is 185 prospective studies and 58 RCTs; the organic outcome literature is three cohorts that disagree with each other and zero clinical RCTs.

Verdict on topic 6

At a 20% to 82% premium with a null outcome literature, organic is a poor use of a marginal food dollar relative to buying more produce and more fibre. The useful arithmetic: spinach carried the smallest premium in ERR-209 (7%) and dried beans at $0.17 per cup-equivalent are among the densest fibre sources, so both the 800 g/day produce target and the 25 to 29 g/day fibre floor are reachable inside the $2.50 to $3.00 daily ERS budget with zero organic spend.

What would change this: organic premiums converging toward zero (three of seventeen ERR-209 products did trend down steadily, spinach from 57.5% to 7.2%), at which point the argument becomes moot rather than won.

7. Which specific foods, if any, have a defensible case

Ranked by strength of the case on health grounds only. None reaches “strong.”

Rank	Food	Case	Strength
1	Produce eaten in volume by a pregnant woman or a toddler	Cumulative organophosphates. EFSA’s cumulative acetylcholinesterase certainty falls to 50% to 90% for French children (against above 90% for adults); EPA’s chlorpyrifos food-only estimate for children 1 to 2 is 9.7% of the ssPAD; the RfDs are cholinesterase-derived and the CHAMACOS and Columbia effects sit below them	Precautionary, and the strongest precautionary case here. No outcome study attaches diet-derived residues to a clinical endpoint, so this is a margin-of-safety argument, not an effect estimate
2	Cereals and grain products	Cadmium. Barański’s Cd effect was significant for cereals only; grains are 26.9% of European Cd intake; 14.8% to 31.2% of the population already exceeds the TWI and toddlers average 1.9x it	Weak but real. The effect estimate is one the authors themselves call “less reliable,” and it lands on a tail, not a mean
3	Meat and poultry	Multidrug-resistant organisms, aPR 0.43 (0.30 to 0.63) on 39,349 NARMS samples, with the facility effect entangled	Real measurement, but the personal-health link breaks: Day 2019 shows non-human reservoirs contribute little to invasive human disease. This is a stewardship argument, not a personal-health one
4	High-residue produce eaten daily in quantity (strawberries, spinach, peppers)	4x lower residue detection frequency	Defensible only if residue reduction is valued for itself. Winter and Katz: 0 of 120 combinations exceeded the RfD
5	Milk	Total n-3 +56%, ALA +69%, roughly 60 to 90 mg/day more total n-3 per half litre (derived) of which 14 mg is VLC n-3	Net negative in a market without iodised salt. The n-3 gain is bought with a 74% iodine reduction at a 72% price premium; where salt is iodised the iodine objection is weak and the case is simply small
6	Eggs	none found	No case. No composition meta-analysis exists, and the premium is the highest measured at 82%
7	Packaged organic food	none	No case, plus a measured downside: the label produces a d=0.97 calorie underestimation

The reasons that actually hold up, which are not health reasons

Out of scope as instructed, but the brief asks them to be named, and honesty requires it, because these are the strongest arguments in the entire dossier:

Antimicrobial stewardship. Innes 2021, aPR 0.43 (0.30 to 0.63) on 39,349 retail meat samples, is the largest organic-versus- conventional effect anywhere in this document. It is a collective-action good, not a personal one, which is exactly why it never appears in marketing.
Farmworker and applicator exposure. Acquavella 2004: applicator maximum urinary glyphosate 233 ppb versus a NHANES general-population geometric mean of 0.37 to 0.51 ug/L, roughly 500x. Every neurodevelopmental cohort with a robust effect (CHAMACOS, -7.0 IQ points) measured agricultural-proximity exposure. The pesticide harm in this literature is real, and it lands on the people who grow the food, not on the people who eat it.
Animal welfare, on which no evidence was gathered here and none is needed to see that it is a values question, not an empirical one.
Biodiversity per hectare, which is real, though note the honest accounting below.

The environmental case is genuinely mixed per unit of food, and this should not be overstated either. Clark and Tilman 2017, Environ Res Lett 12:064016, 742 agricultural systems across 164 published LCAs: per unit of food, organic systems used 25% to 110% more land, 15% less energy, had 37% higher eutrophication potential, and showed no significant difference in greenhouse gas emissions (4% lower) or acidification (13% higher). The authors’ own framing is that differences between production systems are smaller than differences between food types, i.e. what you eat matters more than how it was farmed. (https://iopscience.iop.org/article/10.1088/1748-9326/aa6cd5)

Strongest counter-evidence to my own bottom line

Stated as strongly as I can make it, because the bottom line above is a negative claim and negative claims are the easiest to over-sell.

Baudry 2018 survived an unusually heavy dietary adjustment. The HR moved only 0.70 to 0.76 after adding a validated diet-quality score, PCA dietary patterns, ultra-processed food, fruit and vegetable intake, fibre, red meat and processed meat. Against the Feng/Wang benchmark, where adjustment removed 82% to 87% of the likelihood-ratio statistic for vegetables and CVD, that is remarkably little attenuation. Either the organic signal is unusually robust, or the covariate set is capturing less of the confounding structure than it appears to. I cannot distinguish those from the published data.
MoBa does not fit the healthy-user template. Torjusen 2010 documented that frequent organic consumers in MoBa smoked more, drank more and earned less, while also being better educated and eating better on the cohort’s own diet score. The confounding direction there is mixed rather than the usual healthy-user pattern, and yet pre-eclampsia came out at 0.79 and hypospadias at 0.42. That is a real structural argument and the only clean rebuttal is the multiplicity problem (1 of 7 exposures in Torjusen).
The exposure contrast is not in doubt, only its consequences. Rempelos 2022 is a parallel-group RCT showing 17 versus 180 ug/day of pesticide excretion, a 91% reduction. If any dietary residue effect exists at all, organic is a highly effective intervention against it. My bottom line rests entirely on the dose being irrelevant, which is a regulatory-threshold argument, and regulatory thresholds have moved before.
Winter and Katz, my single most load-bearing citation, has real weaknesses. It reports means only with no upper-percentile estimates, treats non-detects as zero, models only the top 10 pesticides per commodity, performs no cumulative or aggregate assessment, and carries no funding or COI statement. The independent replication (Jacobs 2024) was written entirely by commercial risk consultancies. The “0 of 120” figure is the strongest single number in this document and it comes from a paper I cannot fully audit.
The EFSA cumulative assessment is less certain in children. Certainty that the MOET threshold is not exceeded is above 90% for two adult groups but only 50% to 90% for French children, the weakest band in the assessment. That is not a null result; it is an admission of near-coin-flip uncertainty in exactly the subgroup where a precautionary argument would apply. Add that the thresholds themselves are cholinesterase-derived, that EPA kept the 10x FQPA children’s factor for chlorpyrifos precisely because that endpoint may not protect the developing brain, and that Mie 2017 says the cognitive findings are not in the formal assessments at all.
Cadmium in cereals is a genuine open question. 14.8% to 31.2% of Europeans exceed the TWI and toddlers average nearly 2x it. A 48% reduction in the largest contributing food group is not nothing, and no biomarker study has ever tested whether organic eaters have lower body cadmium. That study is easy to run and has not been run.
General-population biomarker cohorts with hard outcomes exist. Bouchard 2010 (NHANES n=1,139 children, urinary DAP, ADHD OR 1.55 per 10-fold) and Bao 2020 (NHANES n=2,116, urinary 3-PBA, all-cause mortality HR 1.56, cardiovascular 3.00) are real, published and measured rather than FFQ-derived. Neither can attribute the exposure to diet, and one is cross-sectional, so neither is an organic study. But the category is not empty, and if the route question is ever settled these are the studies that would matter.
The low-versus-high-residue contrast is a coherent alternative reading of the large cohorts. At matched servings, Sandoval-Insausti 2022 gives 0.64 for low-residue produce and 0.93 for high-residue produce, and Chiu 2016 finds sperm count positively associated with low-residue produce and flat for high-residue. The authors read that as residues partly offsetting the produce benefit. I read it as an FFQ instrument separating food categories. Both readings fit the data and I cannot rule theirs out.
Absence of evidence is doing a lot of work here. There has never been a randomized trial of organic food with a clinical endpoint, and the longest biomarker trial ever run is 24 weeks. “No demonstrated benefit” and “no benefit” are not the same statement, and I am relying on the first while writing something that reads like the second.

Hardest open questions

The questions this document cannot answer, and that any future pass should attack first. Roughly in order of how much the answer would move the bottom line.

Route attribution. Bouchard 2010 and Bao 2020 link measured urinary pesticide metabolites to ADHD and to mortality in general populations. What fraction of those biomarkers is diet-derived, versus residential pest control, ambient drift, or preformed metabolites already present in produce (Zhang 2008)? Without route attribution neither study says anything about organic food, and with it they would be the most important studies in the field.
Is the cumulative toddler margin actually adequate? The reference doses are derived from cholinesterase inhibition with standard uncertainty factors. The neurodevelopmental associations in CHAMACOS and Columbia appeared at exposures below them, and EPA’s response was to keep the 10x FQPA factor rather than to change the endpoint. At 9.7% of the chlorpyrifos ssPAD from food alone, and 50% to 90% EFSA certainty for French children, what is the real margin for a high-consuming toddler?
Do organic eaters have lower body cadmium? No biomarker study has ever been run. The composition effect is significant for cereals only, the estimate is one the authors flag as “less reliable”, and 14.8% to 31.2% of Europeans already exceed the tolerable weekly intake. A urinary or blood cadmium comparison in matched organic and conventional eaters is cheap and would settle the strongest composition-based case in this document.
Residue offset or category confounding? Sandoval-Insausti 2022 gives 0.64 for low-residue and 0.93 for high-residue produce at matched servings, and Chiu 2016 finds the same shape for sperm count. Is that residues cancelling part of the produce benefit, or is the high-residue basket simply a different set of foods? The Pesticide Residue Burden Score cannot distinguish them.
Would a negative-control-outcome analysis break Baudry? Pick an outcome organic food cannot plausibly cause, run it on NutriNet-Sante, and see whether it comes back protective. Nobody has done this, it is inexpensive, and it would be close to decisive either way.
How much residual confounding survives Baudry’s adjustment? The HR moves only from 0.70 to 0.76 across a very heavy covariate set, which is unusually little attenuation against the Feng/Wang benchmark. Either the signal is robust or the covariates are not capturing the confounding structure. The E-value needed to erase the point estimate is about 1.96 on both exposure and outcome, which is large but not absurd for an unmeasured lifestyle variable.
Is the AMR difference a production standard or a facility effect? Innes 2021 compares organic product from split facilities against conventional product from conventional-only facilities, with only 3 organic-only facilities in the dataset. How much of aPR 0.43 survives a design that can separate farming practice from processing plant?
Does a within-family design with real precision exist anywhere? Instanes 2024 is the only one, its sibling comparison rests on 5,534 discordant siblings, and its intervals are wide enough to contain moderate effects in both directions. A larger discordant-sibling analysis, on any outcome, would be worth more than another between-family cohort.
Does the polyphenol difference move any biomarker at realistic servings? The 17% to 69% increases carry the same “less reliable” flag as the protein and cadmium estimates, and no absolute-intake modelling paper has shown the delta shifting endothelial function, LDL oxidation or anything else measured.
Is the packaged-food advantage real after category matching? Meadows 2021 shows -196 mg sodium and -6.3 g added sugar per 100 g, uncategorised. A same-shelf, same-NOVA-subgroup comparison would show whether that is reformulation or category composition. The data exist and the analysis has not been published.

What to add to the rubric

Recommendation: organic gets zero scored weight and one new flag. Concretely:

No nutrient-block or ingredient-block weight, in either direction. The composition evidence does not support a bonus, and on the rubric’s own top-ranked levers it points the wrong way: Barański found protein -15% (-27 to -3) and fibre -8% (-14 to -2) in organic crops, and organic milk carries 74% less iodine. A rubric that weights protein at 24 and fibre at 18 cannot coherently award points for a certification associated with less of both.
Add one flag, in the same family as the existing “uncured is marketing” rule. Proposed wording for the traps section:

3. Certification is an input standard, not a composition standard. “Organic” under 7 CFR 205.301 means at least 95% organic agricultural ingredients excluding water and salt, with no sugar, sodium, fat or calorie criterion anywhere in the rule; sodium is formally unbounded. 7 CFR 205.605 permits carrageenan, xanthan gum, mono- and di-glycerides and silicon dioxide. Organic-labelled packaged foods do average better (Meadows 2021, n=8,240 vs 72,205: -196 mg sodium and -6.3 g added sugar per 100 g) but that is category composition, not formulation, and 75% of organic child-marketed products are still ultra-processed (Pellegrino 2024, n=735). Never let an organic label offset a processing or additive hit; score the panel and the ingredient list exactly as if the word were not there.
Add the health halo to the headline findings as a scoring-hygiene note. Schuldt and Schwarz 2010 (n=114, identical 160 kcal panel, d=0.97) and Durand 2025 (meta-analysis, 21 studies, pooled r=0.40) make this one of the largest effect sizes in the whole rubric, and it is an effect on the evaluator, not on the food. It belongs next to the disclosure-depth trap for the same reason: both are ways the measurement instrument gets fooled.
Add a sign-inversion entry for nitrate, mirroring the existing “uncured” rule. The rubric already knows vegetable nitrate is the same anion with the opposite sign from cured-meat nitrite. Organic vegetables are lower in nitrate (Barański -30%), and given Siervo 2013 (16 RCTs, SBP -4.4 mmHg) that is a small negative for organic, not a positive. Anyone porting the “organic has less nitrate” talking point into the scorer would get the sign backwards.
One thing NOT to add for the general case: a pesticide or residue term. Zero of 120 commodity-by-pesticide combinations exceeded the RfD (Winter and Katz 2011), the EFSA cumulative MOET is above threshold for all 10 EU populations, and no organic-versus-conventional study has ever linked diet-derived residue biomarkers to a clinical outcome. (General-population biomarker cohorts with outcomes do exist, Bouchard 2010 and Bao 2020, but neither attributes the exposure to food.) Adding a residue term would repeat exactly the failure mode the rubric’s own headline warns about: paying attention to a clean-label axis while protein, fibre, sodium and energy density go unattended.
The one exception to hold open: a precautionary note for pregnancy and toddlers. Not a score, not a weight, and explicitly labelled SPEC. The supporting facts are the EFSA cumulative certainty floor of 50% to 90% for French children, EPA’s chlorpyrifos food-only estimate of 9.7% of the ssPAD for children 1 to 2, the fact that the reference doses are cholinesterase-derived while the neurodevelopmental effects in CHAMACOS and Columbia were seen below them, and EPA’s retention of the 10x FQPA children’s factor. It changes what a parent might buy; it does not change how any food is scored.

Net: organic belongs in the rubric as a Tier 3 entry (drop the penalty AND the bonus; the label is not evidence), plus a transparency flag against the health halo. It is the mirror image of “natural flavors”: a disclosure and marketing issue rather than a health one, with the twist that here the marketing works in the consumer’s disfavour by making them eat more of the thing.

/research/health/rubric/ - the grading conventions this document follows
The scoring implementation this document recommends not changing is not published. [meal-service comparison removed; it is a separate private evaluation]