Nucleus of the autoimmune-modifiable-risk cluster — the canonical home for what modifiable
exposures do to the risk of developing an autoimmune inflammatory disease (incidence and risk factors,
NOT treatment of established disease — the same risk-vs-management line the wiki draws for other
conditions). It now spans two distinct autoimmune diseases: rheumatoid arthritis (RA), where the
dose-response evidence is gold-tier, and inflammatory bowel disease (IBD). It holds three gold-tier
meta-analyses across two diseases: smoking->RA and adiposity->RA (each RAISING risk), and dietary fibre /
fruit / vegetables->IBD (mostly LOWERING risk — but fibre is NULL for the UC subtype). The exposures and
diseases DIFFER across facets, so what the cluster licenses is configuration — a map of which lever
moves which disease, plus cross-facet STRUCTURAL observations — NOT an effect-magnitude ranking across
incommensurable exposures and NOT a shared-mechanism claim. The load-bearing emergent move (the recurrence
of ONE structure across two independently-studied diseases — in no single source) is the decomposition
parallel: each autoimmune disease partitions into subtypes on which the same exposure diverges (RA on
serotype; IBD on the CD/UC subtype split), so a subtype-pooled estimate misleads — it can mask a real
effect (a pooled IBD fibre estimate would hide the CD benefit) or blur a concentrated one (a pooled RA
smoking estimate dilutes the seropositive signal) — see the closing synthesis.
Smoking and rheumatoid arthritis — the effect estimate
-
effect_measure / dose-response (relative, vs never-smokers). Risk rises monotonically then flattens:
Pack-years RR (95% CI) Never smokers 1.00 (referent) 1 to 10 1.26 (1.14 to 1.39) 11 to 20 1.70 (1.44 to 2.01) 21 to 30 1.94 (1.65 to 2.27) 31 to 40 2.02 (1.44 to 2.82) >40 2.07 (1.15 to 3.73) -
population_and_comparator. Adults; smokers by lifelong pack-year exposure vs never-smokers. Pooled from 3 prospective cohorts + 7 case-control studies (181,100 subjects, 4,552 RA cases). Cohorts were women-only; case-control mixed-sex. All observational (no RA-prevention RCT is feasible).
-
outcome. Incident RA — a patient-important outcome (a serious chronic autoimmune inflammatory disease), not a surrogate.
-
dose_response_shape — a monotone rise to ~20 pack-years, then a PLATEAU at ~2x, held with uncertainty. «increased risk of developing RA with increasing number of pack-years smoked up to 20 pack-years, and then the relative risk stabilized approximately at the value of 2» (Di Giuseppe et al., 2014). Three facts must travel with the plateau:
- Confidence interval: the top category is imprecise — RR for >40 pack-years = 2.07 (1.15 to 3.73), a wide interval overlapping the 21-30 estimate (1.94, 1.65-2.27). The flattening is therefore partly wide-CI-at-the-edge, not a sharply located knee.
- Studied range (extrapolation boundary): «The median of the highest category of number of pack-years analyzed in each study ranged between 15 and more than 55 pack-years, while the reference group for all studies was never smokers» (Di Giuseppe et al., 2014) — so the >40 arm sits near the sampling edge where data thin.
- Shape statistics: «we observed a limited evi- dence of a non-linear relationship between pack-years of smoking and RA (P = 0.078). However, the restricted cubic spline model resulted in the best fitting model in terms of the Akaike Information Criterion (AIC) … (AICsplines = -70.0; AIClinear = -57.4)» (Di Giuseppe et al., 2014) — the plateau is the best-fitting model but only weakly evidenced as non-linear.
-
uncertainty. No between-study heterogeneity (Pheterogeneity = 0.32; I2 = 12.7%), leave-one-out overall RR 1.92-2.10, no publication bias (Egger P = 0.10); but all 10 studies are NOQAS-moderate, case-control studies carry recall bias, and residual confounding is plausible (case-control pooled RR 2.19 vs cohort 1.83, the authors attribute to fewer confounders adjusted).
-
effect_modifiers. RF serotype — a route-(b) effect modification (below).
-
confidence: low — single gold MA, observational, opens the domain.
The plateau is mechanistically motivated — distinct from a measurement artifact
The wiki’s [PRIOR — CONTESTED] on dose-response knees/plateaus holds that a measured-monotone-then-flat
curve is weak evidence of a true plateau, because measurement error can hide a knee but not manufacture
one, so the operative default stays «every reduction pays». This case sits slightly differently from the
vault’s other plateaus, and the difference is worth recording:
- Unlike the ESC fruit/veg plateau (an observed flattening with no offered mechanism) or the coffee plateau that dissolved into a smoking artifact (Grosso, on The U-Shaped Association Artifact), here the authors name a saturation mechanism for the flattening: «The ob- served non-linear shape of the dose-response associ- ation in this meta-analysis is compatible with this triggering mechanism» (Di Giuseppe et al., 2014) — a triggering (threshold-then-saturate) immune process would plateau once the trigger is fully engaged.
- But this is compatibility, not proof, and it does not overturn the prior: the top-arm CI is wide, the non-linearity is only P=0.078, and mechanism licenses direction not curve shape (Net Effect vs Intended Effect). Held as a second explicit plateau with a mechanistic rationale, not as a demonstrated knee. The decision default is unchanged and conservative — see below.
Effect modification by RF serotype — a route-(b) positive whose modifier is an OUTCOME subphenotype
Smoking’s relative effect is larger for seropositive RA. Highest-vs-lowest pack-year category: RF-positive RR «2.47 (95% CI 2.02 to 3.02; Pheterogeneity = 0.88), while it was 1.58 (95% CI 1.15 to 2.18, Pheterogeneity = 0.39) among RF-negative cases. These estimates were statistically significantly different (P-value 0.022)» (Di Giuseppe et al., 2014).
This is a genuine route-(b) finding, not the route-(a) arithmetic mirage: the relative effect (the ratio itself) differs across strata with a significant heterogeneity test — the thing route (b) requires and route (a) cannot produce -> Baseline Risk and the Relative-Absolute Split. Its mechanistic basis is a gene-environment interaction: «smok- ing interacts with HLA-DR SE genes in triggering … immunity against citrullinated proteins» and «RF-positive RA is more likely to be associated with the HLA-DR4 shared epitope … than RF-negative RA» (Di Giuseppe et al., 2014).
The load-bearing caveat: the modifier is a subphenotype of the OUTCOME, not a pre-exposure stratum of the person. RF status is measured at RA diagnosis, so this cannot stratify a never-diseased person for targeting — nobody knows which serotype of RA they would develop. It is etiologic heterogeneity (smoking specifically drives the seropositive, HLA-linked path), which strengthens the causal interpretation of the smoking->RA association but supplies no actionable person-level stratifier. RF-negative also rests on only 2 studies, so the contrast is imprecise.
Adiposity (BMI) and rheumatoid arthritis — the effect estimate
The second RA exposure. A gold-tier SR + dose-response MA of 11 studies (7 case-control + 4 cohort; Europe n=7, North America n=4; published 1994-2014), motivated because prior evidence was «contradictory».
-
effect_measure (relative, random-effects). Higher adiposity raises RA risk, modestly:
Contrast RR (95% CI) I2 Obese (BMI >30) vs non-obese (<30) 1.25 (1.07 to 1.45) 63% Obese vs normal weight (18.5-24.99) 1.31 (1.12 to 1.53) 60.1% Overweight (25-29.99) vs normal weight 1.15 (1.03 to 1.29) 46.3% Dose-response, per 5 kg/m2 increment 1.03 (1.01 to 1.05) 70.0% -
population_and_comparator. Adults classified by BMI category; obese/overweight vs normal weight (or vs non-obese). BMI self-reported in 7 of 11 studies — a crude, error-prone adiposity proxy (the authors note BMI «cannot measure the percentage of body fat»), so measurement error attenuates toward the crude exposure -> Ectopic Fat and Depot-Specific Risk, Measurement Error in Dietary Assessment.
-
outcome. Incident RA — the same patient-important outcome as the smoking arm.
-
dose_response_shape — positive and statistically NONLINEAR over the studied range, but with NO located knee and NO U/J arm. «there was evidence of a nonlinear association between BMI and RA risk (Pnon-linearity = 0.005; Figure 4b)» (Qin et al., 2015). Two facts must travel with this:
- The nonlinearity is stronger than smoking’s (P=0.005 here vs P=0.078 for smoking) — yet the paper does not locate a threshold/knee/plateau and the text does not state the curve’s concavity (only Figure 4b shows it). So the shape is positive, nonlinear, knee-unlocated; a bare threshold must not be read off it, and the per-5-kg/m2 coefficient is a single-regression summary that cannot itself show a knee -> Energy Adjustment and What a Diet Coefficient Means.
- No U/J to adjudicate. The referent is normal weight; underweight (<18.5) was not separately pooled, so there is no protective/elevated low-BMI arm reported — the low end is understudied (a G-gap), NOT a checked-and-cleared protective arm. The The U-Shaped Association Artifact discipline has no arm to test here; the honest statement is absence. Studied range: the obese category is open-ended (>30), so the high-BMI arm is imprecise and near the sampling edge.
-
uncertainty — the defining weakness is HETEROGENEITY, not imprecision. I2 = 63-70% throughout, and meta-regression did NOT attribute it to region/gender/design/sample-size/year — the inconsistency is unexplained. No publication bias (dose-response Egger P=0.915, Begg P=0.532); robust to one-study-removal (obese vs non-obese 1.24-1.35). But the authors close on residual confounding: «the possibility of residual confounding cannot be ruled out in these studies» (Qin et al., 2015) — this is the residue AFTER most studies adjusted for age/gender/smoking/alcohol/parity, not an unadjusted gap; and any residual smoking confounding would, if anything, bias the positive BMI->RA association TOWARD the null (smokers tend to be leaner while carrying higher RA risk), so it does not manufacture the finding.
-
effect_modifiers. Higher in females (female obese vs non-obese 1.27, 1.04-1.54; proposed sex-hormone modifier); higher in cohort than case-control designs; ACPA serotype — but the direction is CONTESTED (below), unlike smoking’s settled RF-positive signal.
-
mechanism (directional, marked — authors say it is unproven). «Although the mechanism by which obesity or higher BMI could lead to RA remains unclear», candidate routes are adipose-tissue inflammation («Obesity is often considered a systemic inflammatory condition with increased levels of inflammatory cytokines, including tumor necrosis factor-alpha and interleukin-6»), leptin as a «pro-inflammatory adipokine» sustaining «autoreactive cell proliferation», altered sex-hormone metabolism, and shared genetic predisposition (Qin et al., 2015). This is a DIFFERENT pathway from smoking’s HLA-citrullination immune triggering — the adiposity->cytokine/leptin route touches Inflammation as a Modifiable Lever on an autoimmune (not CVD-event) outcome.
-
confidence: low — gold MA but observational, high unexplained heterogeneity, self-reported BMI, residual confounding acknowledged, effect modest.
Smoking vs adiposity — a CERTAINTY comparison, not a magnitude ranking
The two RA exposures invite a which is the bigger lever question. The op-weave parameter table shows why a magnitude ranking is not licensed — but a certainty comparison is, and it runs one way:
| Parameter | Smoking (Di Giuseppe 2014) | Adiposity (Qin 2015) | Same quantity? |
|---|---|---|---|
| Exposure metric | cumulative pack-years | BMI (kg/m2) | NO — incommensurable dose scales |
| Referent | never-smokers | normal weight / non-obese | NO — different referents |
| Top-category RR | 2.07 (1.15 to 3.73) at >40 py | 1.31 (1.12 to 1.53) obese vs normal | NO — different exposure ceilings + units |
| Per-unit dose RR | spline, no single per-unit reported | 1.03 (1.01 to 1.05) per 5 kg/m2 | NO |
| Heterogeneity I2 | 12.7% (none) | 63-70% (high, unexplained) | YES — smoking far more consistent |
| Publication bias | none (Egger 0.10) | none (Egger 0.915) | YES — both clean |
| Serotype modifier | RF-positive, robust (P-diff=0.022) | ACPA-positive pooled but CONTESTED | ~partial — same axis, different robustness |
- Magnitude ranking — NOT licensed. The four exposure-metric rows are all NO: a pack-year and a kg/m2 are not comparable doses, and the referents differ, so smoking is the stronger RA lever cannot be read off the RR magnitudes. (That the top smoking RR ~2.0 exceeds the obese RR ~1.3 is suggestive, not established, because the exposure ceilings are set by different metrics.)
- Certainty comparison — licensed, and it favors smoking. The rows that ARE the same quantity (heterogeneity, publication bias, robustness of the serotype modifier) are comparable, and they run uniformly: smoking’s estimate is homogeneous (I2 12.7%) with a robust route-(b) modifier; adiposity’s is high-heterogeneity (I2 63-70%, unexplained) with a contested modifier and self-reported exposure. So on the Layer-1 axis (effect size × certainty), smoking is the better-ESTABLISHED RA lever; adiposity is the more provisional one.
RA is not one disease — serotype etiologic heterogeneity across both exposures
A structural observation present in neither source alone: RA partitions on the ACPA/citrullination (seropositive vs seronegative) axis — «distinct genetic etiologies of those two RA subsets» — and BOTH modifiable exposures show serotype-specific signals (a random confounder would not respect the serotype boundary). For SMOKING this strengthens the causal reading (a consistent RF-positive signal); for ADIPOSITY it does NOT — the signal’s direction is inconsistent (below), and a serotype signal that cannot agree on which serotype is weak causal evidence, not strong. Note also that the two arms measure DIFFERENT antibody markers — smoking’s is RF-based (Di Giuseppe), adiposity’s ACPA-based (Qin); related but non-identical, so the shared serotype axis is an approximation, not one measured quantity. The two also differ in how settled the direction is:
- Smoking -> seropositive, settled. RF-positive RR 2.47 vs RF-negative 1.58, significantly different (P=0.022), with an HLA-DR shared-epitope / citrullination mechanism.
- Adiposity -> serotype-specific but direction CONTESTED. The pooled subgroup reads «the association of BMI or obesity with RA risk in ACPA- positive RA rather than ACPA-negative RA» (Qin et al., 2015), yet its own constituent studies point the OTHER way (Wesley: obesity->ACPA-negative; Pedersen: BMI selectively ACPA-negative; Lu: both), and the authors flag «the great heterogeneity may reflect modification» of the ACPA relationship. So adiposity’s serotype direction is UNSETTLED.
The move is emergent-but-measured (type A held weak): the cluster of RA risk factors organizes along the same outcome-serotype axis, but only smoking’s arm is directionally settled — do NOT launder this into a clean both act on seropositive RA parallel. Confidence stays low.
Dietary fibre, fruit, and vegetables and IBD — the effect estimates (the second autoimmune disease)
The cluster’s first non-RA facet: a gold-tier SR + dose-response MA (Milajerdi 2021) of 11 studies (12 effect sizes; published 1992-2018; 478,604 participants; Sweden/US/Australia/Denmark/multi-country) on dietary fibre / fruit / vegetable intake -> incident IBD. Design honesty: the abstract frames it as «prospective cohort studies», but inclusion was «All prospective cohort or nested case-control studies» (6 cohorts + 5 nested case-control) (Milajerdi et al., 2021) — nested case-control from a cohort keeps exposure prospective, but the cohorts-only framing overstates design purity.
-
effect_measure (relative, highest-vs-lowest intake category; random-effects). All protective except fibre->UC:
Exposure -> outcome RR (95% CI) I2 Verdict Fibre -> UC 1.09 (0.88, 1.34) 0.0% NULL Fibre -> CD 0.59 (0.46, 0.74) 0.0% protective Fruit -> UC 0.69 (0.55, 0.86) 87.0% protective (FRAGILE) Fruit -> CD 0.47 (0.38, 0.58) 32.1% protective Vegetable -> UC 0.56 (0.48, 0.66) 72.0% protective Vegetable -> CD 0.52 (0.46, 0.59) 78.9% protective -
population_and_comparator. Adults + some adolescents (10-80 y); highest vs lowest intake category (a contrast, NOT a dose target). Exposure by FFQ (most) or bespoke questionnaire.
-
outcome. Incident IBD and its two subtypes (UC, CD) — a patient-important autoimmune inflammatory disease, «a chronic, relapsing intestinal inflammatory disorder», not a surrogate.
-
The subtype split — IBD is not one disease (a B/F decomposition, the load-bearing finding). Fibre is NULL for UC (1.09) but protective for CD (0.59), while fruit and vegetables protect BOTH subtypes. The same exposure (fibre) diverging by IBD subtype is a genuine decomposition — and it is not a heterogeneity artifact: both fibre pooled estimates are homogeneous (I2 = 0.0%). «no significant association was found between dietary fiber intake and risk of UC (RR: 1.09; 95% CI: 0.88, 1.34; I2 = 0.0%) … However, pooling 6 effect sizes from 5 studies revealed a significant inverse association between dietary fiber intake and risk of CD (RR: 0.59; 95% CI: 0.46, 0.74; I2 = 0.0%)» (Milajerdi et al., 2021). Author’s rationale: «UC is limited to the colon, whereas CD can occur anywhere throughout the GI tract … some dietary fibers are fermented in the distal colon» (Milajerdi et al., 2021).
-
dose_response_shape. Fibre->CD nonlinear (Pnonlinearity < 0.001), «the highest risk reduction was seen for fiber intake >22 g/d», linear «additional 10 g/d … associated with a 14% reduction in CD risk» (Milajerdi et al., 2021). Carry the caveats: the >22 g/d figure is the studied high category, not a validated knee (no CI given on a knee location, FFQ exposure); fibre->UC had NO gradient (Plinearity = 0.09; Pnonlinearity = 0.52). Vegetable nonlinear analysis was «not able» to be run (too few studies).
-
uncertainty — the protective fruit/veg arms are FRAGILE. Two facts:
- Fruit->UC dissolves under subgrouping: «the negative association disappeared in studies conducted exclusively in females and in those with a large sample size (for both, RR: 0.94; 95% CI: 0.69, 1.28)», and the raw estimate carries I2 = 87.0%. An «inverted U-shaped nonlinear association» appeared (RRs > 1.00 at 1-3 servings/d) but was nonsignificant (Pnonlinearity = 0.68) (Milajerdi et al., 2021) — an unadjudicated U-arm, held as absence-of-verdict, not a protective plateau -> The U-Shaped Association Artifact.
- Fruit->CD rests on 2 same-team, women-only cohorts: «these 2 studies were done by the same team and that they recruited only women» (Milajerdi et al., 2021) — NOT independent backing (shared team + single sex).
-
confounding / causal discipline (BLOCKING — these are the streetlight/confounding danger zone).
- Reverse causation acknowledged but unaddressed: preclinical/undiagnosed IBD causes GI symptoms that cut fibre/fruit/veg intake, manufacturing a spurious protective association. The authors flag the channel — «dietary intake by patients with IBD can change by the disease stage. Included studies did not provide sufficient data about disease severity» (Milajerdi et al., 2021) — but report NO lag analysis and NO early-case exclusion, and some studies used baseline (not repeated) intake. Prospective/nested design mitigates but does not clear the long preclinical phase.
- Observed-healthy-population trap: high fruit/veg intake co-travels with overall healthy lifestyle, and only 4 of 11 studies adjusted for dietary intake at all; fibre «is usually derived from legumes, whole grains, fruit, or vegetables», so the food matrix is not isolated — a pattern-derived estimate, not a causal effect of the isolated nutrient -> Is the Food Category Doing Any Work.
- Measurement error: «Given the use of FFQs for dietary assessment in most studies, misclassification of participants in terms of dietary intake should also be taken into account» (Milajerdi et al., 2021) -> Measurement Error in Dietary Assessment.
- Publication bias NOT tested in the main text (no Egger/Begg) — weaker on this axis than the RA arms, which reported clean tests.
-
mechanism (directional, marked — a DIFFERENT pathway from the RA exposures). Proposed route: fibre/fruit/veg «influence the composition and function of the gut microbiota to affect immune responses and immunological homeostasis», via SCFA/butyrate production, reduced colonic permeability, colonocyte energy (Milajerdi et al., 2021). This is a gut-microbiota/SCFA pathway — NOT the smoking->HLA-citrullination immune-triggering route nor the adiposity->adipokine/cytokine route. Human evidence is the observational associations; the SCFA chain is in vitro/animal, admitted directionally only.
-
confidence: low — gold MAs but observational; the protective diet arms carry unaddressed reverse causation, food-matrix confounding, FFQ measurement error, untested publication bias, and (for fruit/veg) high heterogeneity and fragile subgroups.
Closing cross-disease synthesis — are the autoimmune levers SHARED or DISEASE-SPECIFIC?
The cluster now spans two autoimmune diseases and three exposures, so the banked question is whether the modifiable levers generalize across autoimmune disease or are disease-specific. The parameter table shows a magnitude/lever comparison is not licensed (incommensurable exposures and diseases), and forbids the tempting fake move of a shared inflammatory mechanism:
| Parameter | RA (Di Giuseppe / Qin) | IBD (Milajerdi) | Same quantity? |
|---|---|---|---|
| Exposures studied | smoking (pack-years), adiposity (BMI) — both RAISE risk | fibre (CD only), fruit, vegetables — LOWER risk (fibre NULL for UC) | NO — different exposures, opposite direction |
| Disease | rheumatoid arthritis (joints) | inflammatory bowel disease (gut) | NO — different organ system |
| Proposed mechanism | HLA-citrullination immune triggering (smoking); adipokine/cytokine (adiposity) | gut-microbiota / SCFA / colonic permeability | NO — distinct pathways |
| Subtype partition | serotype (seropositive / seronegative) | subtype (CD / UC) | ~parallel STRUCTURE, different axis |
| Exposure diverges across the partition? | YES (smoking->seropositive; adiposity contested) | YES (fibre->CD only, null for UC) | YES — same structural fact |
- Levers are DISEASE-SPECIFIC, not shared. No exposure in the cluster has been shown to move both RA and IBD: the RA risk factors (smoking, adiposity) and the IBD-associated exposures (fibre, fruit, vegetables) are disjoint, run in opposite directions, and act through distinct mechanisms. The rows that ARE the same quantity are the last two — the STRUCTURE of the evidence, not any shared cause. Do NOT read a shared inflammation lever off the two facets: inflammation is a label spanning distinct pathways (citrullination vs adipokine vs SCFA/microbiota), and treating it as one lever is the trumps/magic-bullet appetite the telos warns against -> Inflammation as a Modifiable Lever. This is the OPPOSITE of the cardiometabolic-vascular case, where a shared upstream substrate DOES license one lever across diseases -> Shared Modifiable Levers Across Age-Related Diseases; the autoimmune axis does not (yet) support it.
- The emergent cross-source observation (type A, held weak — a methodological reading, not a novel mechanism): the decomposition parallel. Each source performs its OWN subtype split (Di Giuseppe on serotype, Milajerdi on CD/UC); what no single source shows is that the SAME structure recurs across two independently-studied autoimmune diseases — in BOTH, the same exposure diverges across the partition (smoking concentrates on seropositive RA; fibre protects CD but not UC). So the cross-disease lesson is methodological, not causal: an autoimmune-risk estimate pooled across subtypes can mislead — masking a real effect (a pooled IBD fibre estimate would hide the CD benefit) or blurring a concentrated one (a pooled RA smoking estimate dilutes the seropositive signal). This is a type-A/C decomposition move — a claim about how to READ autoimmune-risk evidence — held weakly (it is a recurring structure, not a mechanism), and honest precisely because it claims no shared cause.
- A named gap, NOT a smoothed parallel — smoking and IBD. Smoking is known to affect IBD in a subtype-split, direction-OPPOSITE way (harmful for CD, apparently protective for UC) — itself the reverse of its RA-raising effect. That would make smoking the one exposure touching both diseases, and a striking third instance of the subtype-divergence structure. But this MA is about diet, not smoking — Milajerdi supplies no smoking-IBD estimate — so the observation is, explicitly NOT sourced here, and left as a G-gap: it is owed a dedicated smoking-IBD source before any cross-disease smoking claim is banked. Named, not asserted.
Decision relevance
-
There is no safe low level for RA risk. Risk is already elevated at 1-10 pack-years (RR 1.26, 1.14-1.39) — «statistically significant even among those smoking less than 10 pack- years» (Di Giuseppe et al., 2014). The decision-change is that even light smoking raises RA risk; the exposure has no threshold below which it is inert here.
-
The plateau does NOT license fatalism. «It stabilized» above ~20 pack-years describes incident risk by cumulative exposure; it is not a claim that further smoking is harmless (it plateaus at double the never-smoker risk) nor that cessation stops paying. Whether stopping reduces RA risk is a separate question this MA does not answer (a G-gap; the authors’ companion cessation cohort is not ingested here).
-
RA joins the big-rock case for not smoking, via a distinct mechanism. Smoking is already the vault’s largest mortality lever (Smoking and Mortality, all-cause HR ~3.0); RA incidence is one more patient-important outcome it moves. But unlike the cardiometabolic-vascular diseases that share an upstream substrate (Shared Modifiable Levers Across Age-Related Diseases), RA runs through an immune-triggering / HLA-citrullination path — so the shared-substrate reading does not extend to it, and smoking’s breadth across disease is mechanistically heterogeneous, not one pathway.
-
Adiposity is a MODEST, uncertain RA add-on — RA is not the reason to manage weight. The RA-specific effect is small (obese vs normal 1.31; per-5-kg/m2 1.03) and high-heterogeneity, so for someone deciding about weight, RA risk is a minor additional consideration dwarfed by adiposity’s cardiometabolic and mortality effects held elsewhere in the vault — exactly as RA is a minor add-on among smoking’s harms. The decision-change is small: adiposity joins the modifiable RA risk factors, but does not by itself move a weight decision that the big-rock outcomes have not already settled. Absolute RA benefit of weight reduction would scale with baseline RA risk (route (a)), but the relative effect is modest to begin with.
-
For IBD, the decision-change is subtype-conditioned and modest in confidence. Higher fibre/fruit/ vegetable intake is associated with lower IBD risk, but: fibre’s benefit is CD-specific (null for UC); the fruit/veg arms are confounder- and reverse-causation-exposed and partly fragile; and none is a causal effect of the isolated nutrient. So the honest advice is that a fibre-and-produce-rich diet is consistent with lower IBD (especially CD) risk and is already recommended for larger outcomes (mortality, CHD, T2D, colorectal cancer) — IBD-risk reduction is a plausible additional reason at low confidence, not an independent big lever, and «22 g/d» is a studied-category label, not a target. The exposure is one someone would want to increase anyway; IBD does not change that decision, it adds a low-confidence supporting reason.
Gaps (G)
What the assembled sources collectively cannot answer (the wiki’s own gap-identification).
- Reversibility — BOTH exposures are priced as cumulative/cross-sectional risk, not as the effect of change: neither the smoking-cessation nor the weight-loss dose-response on RA risk is in hand. Owed.
- Adiposity’s low-BMI arm + serotype direction — underweight (<18.5) was not separately pooled (no U/J adjudicable), and the ACPA-serotype direction for adiposity is contested across the constituent studies; a cleaner-measured cohort or an MR instrument is owed to settle both.
- Causality of the diet->IBD arms — reverse causation (preclinical IBD altering intake) is acknowledged but never addressed (no lag analysis / early-case exclusion); food-matrix confounding is uncontrolled (4/11 studies adjusted for diet); publication bias untested. An MR instrument or a lag-analysed cohort is owed to move any diet->IBD arm from association toward effect.
- Smoking and IBD — the one exposure that may touch BOTH cluster diseases (subtype-split, direction opposite to its RA effect) is a named G-gap: no smoking-IBD source is held, so the cross-disease smoking observation stays, not banked. Closing it would give a third instance of the subtype- divergence structure and the cluster’s first genuinely cross-disease exposure.
- A third autoimmune disease — the shared-vs-disease-specific verdict rests on two diseases (RA, IBD) showing DISJOINT levers; a third (e.g. type 1 diabetes, MS, psoriasis) would test whether disease-specificity is the rule or an artifact of these two happening not to overlap.
- True shape — for smoking, separating biological saturation from thinning data at high exposure; for adiposity, locating the significant nonlinearity’s knee (unlocated here). Both need cleaner exposure measurement or a genetic instrument.
The loop is open
No operation here has graded the don’t smoke / manage adiposity, to lower RA risk or the eat more fibre/produce, to lower IBD risk recommendations against a realized outcome. The synthesis says a well-informed advisor would count RA among the harms of smoking and excess adiposity, and would count lower IBD (especially CD) risk among the plausible benefits of a fibre-and-produce-rich diet; whether acting on any of them lowered an individual’s autoimmune-disease risk is a validity fact the wiki cannot see. All are observational — no autoimmune-prevention RCT is feasible — and the diet->IBD arms carry unaddressed reverse causation and food-matrix confounding on top of that, so they sit at lower confidence than the smoking->RA arm.