Diet exposures and the pathograph: disposition of 42 candidates, 2026-09-01
The diet representation audit — diet-representation-audit-2026-09-01.md, added by
PR #10358 — found 42 environmental[] diet entries that carry a supporting,
snippet-backed citation but no influences_mechanisms link, so they never appear
in the mechanism graph. This records what was decided for each.
27 got an edge. 2 remain unsupported. 13 were left alone.
Two of those were reached independently by main while this work was in review,
and main's version won on the merits in both cases — see Superseded by main
below. The change here is 25 new links across 14 entries, plus evidence blocks on
two edges main created.
An earlier revision of this file recorded 20 added, 8 proposed-not-added and 14 left alone. Six of the eight proposed were then resolved by finding a citation that measures the target node, and one entry from the left-alone list the same way — see Resolved by finding node-level evidence below. Nothing was moved by relaxing the rule.
The rule applied
An influences_mechanisms edge asserts that the exposure acts on a named
mechanism node. So the test is not "is this a real risk factor" — it is:
Does the cited snippet name something that maps to a node or phenotype in this file, or does it only say the exposure raises the risk of the disease?
A snippet that says only "X is associated with an increased risk of [the disease]" identifies no node. Those were left alone. This is the distinction the audit flagged and could not make itself: an association is not a mechanism.
Two schema slots were needed, and they are easy to conflate:
causal_link_typeis graph topology — does the edge jump over mechanism steps that are themselves nodes in this file?DIRECT, orINDIRECT_KNOWN_INTERMEDIATESwhen it does.directnesson the evidence item is about the quote — does it assert the claim, or does the claim follow from it by an inference step?
They vary independently. Wilson Disease's dietary copper edge is topologically
DIRECT (intake feeds the hepatic copper pool with nothing in between) but its
evidence is INDIRECT, because the source says lowering intake prevents
re-accumulation and the forward claim is inferred from that removal design.
Added (25 links across 14 entries, plus evidence on two of main's)
| Entry | Exposure | Target node | Effect | Why |
|---|---|---|---|---|
| Celiac Disease | Gluten Exposure | Gluten-Triggered Immune Response | TRIGGERS | Snippet names gluten as required to trigger the enteropathy |
| Celiac Disease | Wheat / Barley / Rye | Gluten-Triggered Immune Response | TRIGGERS | One snippet names all three grains as triggering and maintaining it (3 links) |
| Gastroesophageal Reflux | Alcohol Consumption | Lower Esophageal Sphincter Dysfunction | EXACERBATES | Snippet states the sphincter-pressure reduction |
| Gastroesophageal Reflux | Dietary Factors | Lower Esophageal Sphincter Dysfunction | EXACERBATES | Physiologic evidence for chocolate and high-fat meals on the same node |
| Heart Failure | Alcohol Abuse | Myocardial Contractile Dysfunction | TRIGGERS | Snippet names chronic cardiac dysfunction and alcoholic dilated cardiomyopathy. Edge created by main; this adds its evidence block |
| Hyperlipidemia | High Saturated Fat Diet | Increased LDL Cholesterol | EXACERBATES | Snippet attributes the LDL-C rise to saturated fat substitution |
| Obesity | High-Calorie Diet | Energy Imbalance | TRIGGERS | Inpatient randomised crossover trial; excess intake attributable to processing itself |
| Obesity | Obesogenic Environment | Increased Body Mass Index | PREDISPOSES | Causal-inference systematic review linking outlet density to BMI |
| Osteoporosis | Vitamin D Deficiency | Bone Remodeling Imbalance | PREDISPOSES | Snippet gives the deficiency → hyperparathyroidism → bone loss route |
| Phenylketonuria | Dairy Intake | Hyperphenylalaninemia | EXACERBATES | Randomised crossover trial measuring phenylalanine tolerance from milk protein |
| Phenylketonuria | Mammalian Meat / Nut Intake | Hyperphenylalaninemia | EXACERBATES | Named among high-protein foods the diet avoids (2 links) |
| Wilson Disease | Dietary Copper | Hepatic Copper Accumulation | EXACERBATES | Snippet ties intake to copper re-accumulation |
| Irritable Bowel Syndrome | Dietary Triggers | Abdominal Pain, Bloating | EXACERBATES | Low-FODMAP meta-analysis reports both (2 links) |
| Migraine | Dietary Triggers | Headache | TRIGGERS | Provocation studies for caffeine withdrawal and MSG |
| Carotid Stenosis | Smoking and diet | Atherosclerotic carotid plaque formation | PREDISPOSES | Snippet names plaque development specifically |
| Lathyrism | Grass pea overconsumption | Beta-ODAP Receptor Agonism | TRIGGERS | Study attributes the disease to prolonged overconsumption |
| Gout | Red Meat and Organ Meat Intake | Hyperuricemia | EXACERBATES | NHANES III measures serum uric acid rising with meat intake |
| Gout | Shellfish Intake | Hyperuricemia | EXACERBATES | Same source, seafood arm |
| Gout | Beer Intake | Hyperuricemia | EXACERBATES | 78,153 participants; beer raises serum urate where sake at matched ethanol does not |
| Gout | Fructose-Sweetened Soft Drink Intake | Hyperuricemia | EXACERBATES | Meta-analysis measuring serum uric acid, not incident gout |
| Liver Cirrhosis | Alcohol Consumption | Hepatocyte Injury and Death | TRIGGERS | Hepatocyte apoptosis markers ~5x normal in 114 heavy drinkers |
| Polycystic Kidney Disease | Caffeine Consumption | Vasopressin/cAMP-Driven Cyst Expansion | EXACERBATES | Caffeine potentiates desmopressin-driven cAMP in ADPKD cyst cells |
| Coronary Artery Disease | High-Fat Diet | Coronary Endothelial Injury and Subendothelial LDL Retention | PREDISPOSES | Controlled-feeding meta-analysis quantifies the LDL step. Edge created by main; this adds its evidence block |
Every edge carries its own evidence, copied byte-exact from the entry it sits on
so the snippet stays verified and evidence_source cannot drift.
Several edges are deliberately narrower than the entry they hang on. The GERD
dietary edge is scoped to the sphincter effect, because the same review found no
evidence that dietary avoidance improves outcomes. The IBS edge is scoped to
FODMAPs, because the gluten half of that annotation is uncited. The migraine edge
is scoped to caffeine withdrawal and MSG, because the entry's own REFUTE item
withdraws chocolate and aged cheese.
Superseded by main
Two of the 20 were curated independently on main while this sat in review. In
both cases main's version is kept, on the merits rather than by merge order.
Marchiafava-Bignami — dropped entirely. Main linked the same exposure to the
same node using a better sentence from the same paper: "The main
pathophysiological mechanisms involve alcohol consumption, which leads to thiamine
depletion and disrupts various metabolic pathways." That names thiamine depletion,
so main's TRIGGERS / directness: DIRECT is justified where this pass had
PREDISPOSES / INDIRECT — graded down precisely because the quote it used
("predominantly afflicts individuals with chronic alcohol abuse and malnutrition")
did not name the node. Better evidence beats an earlier claim.
Heart Failure — edge kept, evidence added. Main created the same edge with a fuller description and no evidence; this attaches the citation to it.
The history record written for the Marchiafava-Bignami session is left untouched, since history is append-only and it accurately describes what that session did. A later record notes the supersession.
Originally withheld (8), and why
Kept for the record: this is the state each of these was in before the citation search below. Each had good evidence for a risk claim and a plausible target node, but the snippet did not reach the node — so adding the edge would have meant accepting a disease-level association as node-level support. Six were subsequently resolved by finding a citation that does reach the node; two were not.
| Entry | Exposure | Proposed target | The problem |
|---|---|---|---|
| Gout | Red Meat and Organ Meat | Hyperuricemia | Cohort measures incident gout, not urate. Purine → urate is textbook, but this snippet does not say it |
| Gout | Shellfish Intake | Hyperuricemia | Same; the entry's own explanation already notes it measures incident gout rather than flares |
| Gout | Beer Intake | Hyperuricemia | UK Biobank per-drink association with incident gout |
| Gout | Fructose-Sweetened Soft Drink | Hyperuricemia | Meta-analysis of incident gout; note claims urate production |
| Liver Cirrhosis | Alcohol Consumption | Hepatocyte Injury and Death | Steep dose-response over 3M participants, but for cirrhosis risk, not hepatocyte injury |
| Coronary Artery Disease | High-Fat Diet | Coronary Endothelial Injury and Subendothelial LDL Retention | Cochrane RCT meta-analysis measures CVD events; the entry's own note says the LDL effect is small |
| Generalized Anxiety Disorder | Caffeine | (a phenotype) | Controlled trial shows GAD patients are abnormally caffeine-sensitive, but names no node or specific symptom |
| Tyrosinemia Type I | Catabolic and dietary stress | Toxic FAA/MAA accumulation | GeneReviews avoid-list; naming circumstances to avoid is not a mechanism measurement |
The four Gout entries are the cleanest illustration of the whole problem, and the
best fix is probably not a judgement call at all — it is a citation that measures
serum urate after a purine load, which would turn all four into ordinary
EXACERBATES edges on Hyperuricemia.
That is what happened. Six of these eight, and one entry from the left-alone list, were resolved by citation rather than by ruling. See the next section; the two that remain are the two where no such citation exists.
Resolved by finding node-level evidence (7)
The rule was never "these exposures are not real". It was that the cited quote stopped at disease risk. That is a citation problem, so the fix is a citation — and for seven of these it exists. Each one below is now an ordinary edge with evidence that measures the target node itself. No edge here was added by relaxing the rule.
| Entry | Exposure | Node | New citation | What it measures |
|---|---|---|---|---|
| Gout | Red Meat and Organ Meat | Hyperuricemia | PMID:15641075 | Serum uric acid across meat intake quintiles, NHANES III, n=14,809 |
| Gout | Shellfish | Hyperuricemia | PMID:15641075 | Same study's seafood arm |
| Gout | Beer | Hyperuricemia | PMID:36930152 | Serum urate by beverage at 20 g ethanol per standard drink, n=78,153 |
| Gout | Fructose-Sweetened Soft Drink | Hyperuricemia | PMID:32683776 | Serum uric acid, +0.18 mg/dL highest vs lowest intake |
| Liver Cirrhosis | Alcohol Consumption | Hepatocyte Injury and Death | PMID:34305638 | Caspase-cleaved CK-18 (M30) and M65 in 114 heavy drinkers |
| Polycystic Kidney Disease | Caffeine Consumption | Vasopressin/cAMP-Driven Cyst Expansion | PMID:12397042 | Proliferation and transepithelial fluid secretion in ADPKD cyst epithelium under caffeine |
| Coronary Artery Disease | High-Fat Diet | Coronary Endothelial Injury and Subendothelial LDL Retention | PMID:40885400 | LDL cholesterol in 24 randomised complete-feeding trials |
Three of these are worth reading closely, because they change what the entry says rather than just adding to it:
- The four Gout edges turn on the difference between incident gout and serum
urate. Every citation the entries already carried measured who develops
gout, which is a disease-risk claim. Choi's NHANES analysis measures serum uric
acid itself, so it reaches the
Hyperuricemianode directly, and the beer study is stronger still: sake at the same standardised ethanol content shows no urate association, so the effect is attributable to beer rather than to ethanol. - The Liver Cirrhosis edge is graded
DIRECT, notINDIRECT, because M30 is caspase-cleaved cytokeratin-18 released by dying hepatocytes. The quote measures the node, not a proxy for it. - The Coronary Artery Disease evidence contradicts the edge's own
description. That description said the trial effect on LDL is small, which came
from free-living trials. Controlled feeding gives 15.72 mg/dL when the diets
differ in saturated fat, so the qualifier was removed from the edge and the
entry-level
notesrewritten to name both designs rather than leave a reader hitting the older claim first. The link is still gradeddirectness: INDIRECT, because the trials lower saturated fat and the forward claim is inferred from that removal. The snippet runs through the study's own null arm — no significant LDL difference when the diets are saturated-fat-matched — so the quote carries its own control instead of the caveat living only in prose.
The PKD edge is evidence_source: IN_VITRO — cultured patient-derived cyst
epithelium, not patients. That is a weaker footing than the other six and is
recorded as such rather than smoothed over. Its snippet is the abstract's
concluding sentence, which names proliferation and transepithelial fluid
secretion; the same abstract identifies those two as the mechanisms underlying
cyst expansion, so the quote reaches the node instead of stopping at cAMP and
inferring the rest.
Still unsupported (2)
Searched and not found. These are not withheld on judgement any more; there is no quote to add.
| Entry | Exposure | Why no edge |
|---|---|---|
| Generalized Anxiety Disorder | Caffeine | The one human imaging study (PMID:21972425, double-blind crossover fMRI) is 14 healthy males, not GAD patients, and its amygdala result is a tolerance finding — the caffeine effect correlated negatively with habitual intake. It does not support Amygdala Hyperactivity, and no other node in the entry matches caffeine's adenosine-antagonist action |
| Tyrosinemia Type I | Catabolic and dietary stress | The node is Toxic fumarylacetoacetate and maleylacetoacetate accumulation. FAA and MAA are not measured in patients — succinylacetone is the clinical surrogate — and no study links catabolic stress to either. The GeneReviews avoid-list remains the only source |
The GAD case is the more interesting of the two: the gap is not missing literature but a missing node. Caffeine acts on adenosine signalling, and the entry models GABA, serotonin, noradrenergic, HPA and amygdala mechanisms with no adenosine node to attach to. Adding one is a curation decision beyond this PR's scope.
Left alone (13)
All have solid evidence for a disease-level risk association and no node the snippet reaches. This is the correct resting state, not a backlog.
Breast Carcinoma (alcohol), Colon Adenocarcinoma (alcohol), Essential Hypertension (alcohol), Familial Hypercholesterolemia (alcohol), HPV-Negative Head and Neck Cancer (ethanol), Laryngeal SCC (alcohol), Oral Cavity SCC (ethanol), Postcricoid Region Cancer (tobacco and alcohol), Obstructive Sleep Apnea (alcohol), Scurvy (vitamin C deficiency), Thyroid Follicular Carcinoma (iodine deficiency), Type 2 Diabetes Mellitus (high-calorie diet), Congestive Splenomegaly (alcohol).
Four are worth singling out:
- Laryngeal SCC and Obstructive Sleep Apnea already say so themselves.
The laryngeal entry's
notesrecord that it "makes a risk-association claim and describes no ethanol or acetaldehyde mechanism"; the OSA explanation says the meta-analysis "measures the association, not the airway-relaxation mechanism". Those entries were curated correctly and need nothing. - Congestive Splenomegaly has a different problem: its description argues alcohol → cirrhosis → portal hypertension → splenomegaly, but the cited snippet is about anemia from direct bone-marrow toxicity. The evidence does not support the claim the entry makes. That is an evidence defect, not a missing edge, and it is not fixed here.
- Polycystic Kidney Disease has moved. It was in this list because its only citation was a GeneReviews avoid-list; it now carries an edge on the strength of PMID:12397042. See the resolved section above.
- Scurvy stays. Vitamin C → collagen prolyl hydroxylation is textbook, but
searches for a human or fibroblast study stating it returned guinea-pig and
mouse work and general biochemistry reviews, nothing quotable that names the
Collagen Synthesis Impairmentnode in humans. The entry's case report is still the only source.
One thing fixed in passing
Wilson Disease graded PMID:36010023 as OTHER on a treatment and left
evidence_source absent on the environmental entry, which counts as
HUMAN_CLINICAL. Copying the quote onto the new edge made
check-snippet-grading fail. The paper is a narrative review of dietary
recommendations with no primary data, so OTHER is right; the environmental item
was corrected to match rather than the new edge being graded to fit.