The one move this page makes (type-A, present in neither source alone): periodontitis is a
single modifiable exposure whose two most-discussed systemic outcome arms sit at different
evidence grades, so they license different decisions. Treating the gums to improve glycaemic
control rests on interventional RCT evidence (moderate certainty); the dementia link is
observational association only, confounded by a shared risk-factor cluster and with
reverse-causation live. So treat periodontitis to lower HbA1c is decision-supported for the
right stratum, while treat gums to prevent dementia is not supported at the same grade —
same exposure, opposite warrant. Neither Simpson (glycaemic, no dementia) nor Dibello (dementia,
no treatment trial) makes this decomposition; it is induced by placing them side by side.
(Dibello et al., 2024; inferred from Simpson et al., 2022)
Arm 1 — glycaemic control (interventional, causal-grade)
Exposure: subgingival instrumentation (scaling and root planing / non-surgical periodontal treatment) vs no active treatment / usual care, in people with both periodontitis and diabetes (almost all T2D).
- Effect (primary, 3-4 months): absolute HbA1c reduction -0.43% (4.7 mmol/mol), 95% CI -0.59% to -0.28%, 30 studies, n=2443. «Moderate-certainty evidence from 30 studies (2443 analysed participants) showed an absolute reduction in HbA1c of 0.43% (4.7 mmol/mol) 3 to 4 months afer treatment of periodontitis (95% confidence interval (CI) -0.59% to -0.28%; -6.4 mmol/mol to -3.0 mmol/mol).» (Simpson et al., 2022) Against a control median HbA1c of 7.74%, this moves a patient from ~7.7% toward ~7.3%.
- Certainty: ⊕⊕⊕⊝ MODERATE (GRADE), downgraded one level «for high risk of bias, largely due to lack of blinding» (Simpson et al., 2022); NOT downgraded for I2=70% (effects consistent across time points), and a low-RoB sensitivity analysis supported the finding. The authors judge it a «clinically significant amount» and consider «Further trials … unlikely to change the overall conclusion» (Simpson et al., 2022).
- Durability — the horizon caveat (decision-relevant). The effect at 6 months is smaller, -0.30% (95% CI -0.52% to -0.08%, 12 studies, n=1457); the 12-month -0.50% (95% CI -0.55% to -0.45%) rests on a single study (n=264). (Simpson et al., 2022) So the benefit is well-evidenced over 3-6 months; whether it is maintained long-term without re-treatment is essentially untested (one small study is not a durability claim). Read this as a repeated-intervention lever, not a one-off cure.
- Harms: where measured, none or mild; serious adverse events similar across arms — but most studies did not evaluate harms (insufficient evidence).
- Outcome is a surrogate. HbA1c is a glycaemic surrogate, not a patient-important endpoint; the trials did not measure microvascular/macrovascular complications or mortality -> Surrogate Outcomes. HbA1c has good (but not perfect) causal transmission to microvascular outcomes from the wider diabetes-trial literature, so it is a reasonable target — but the 0.43% is demonstrated on the marker, not on the complications.
Layer-1 sizing [inferred from @simpson2022periodontitis]. A ~0.4 percentage-point HbA1c drop is a modest adjunct, smaller than a glucose-lowering drug started for the same purpose — it does not replace pharmacotherapy or diet/weight levers. Its value is that it is (i) a different channel (treating an inflammatory/infectious source, not glucose directly), so plausibly additive to drug and diet effects, and (ii) low-harm with an independent oral-health rationale. For a person with untreated periodontitis and T2D it is a real, cheap, non-substitutable increment; it is not a big rock.
Arm 2 — dementia and cognition (observational, association-only)
Exposure here is different: having periodontal disease (present/absent), not treating it. Every constituent study is observational — there is no trial that treating periodontitis prevents dementia.
- PD -> incident dementia: RR 1.22 (95% CI 1.10-1.36), 8 studies, n=3,076,684 dementia-free at baseline, mean follow-up 11 y. «PD was associated to higher risk of incident dementia (RR 1.22, 95% CI 1.10 to 1.36) with sig- nificant heterogeneity across the studies (I2= 95%, p< 0.001)» (Dibello et al., 2024). Heterogeneity is very high (I2=95%); no small-study effect (Egger p=0.473); trim-and-fill raised it to 1.29 (1.16-1.43). Paper’s GRADE: moderate (driven by the vast pooled n).
- PD -> cognitive impairment (cross-sectional): RR 1.25 (1.11-1.40), I2=20%, but Egger p=0.01 small-study effect; trim-and-fill 1.19 (1.06-1.34). PD -> cognitive decline (prospective): RR 3.01 (1.52-5.95), I2=84%, and the effect shrinks with age (z=-2.41, p=0.016) — a pattern more consistent with selection/reverse-causation than a clean dose. (Dibello et al., 2024)
- PD -> depression: a NULL. RR 1.07 (95% CI 0.95-1.21), CI crosses 1 — no association (Dibello et al., 2024).
Why the dementia arm cannot carry a causal recommendation (three source-stated reasons).
- Reverse causation — «studies with a follow-up period of at least 10 years have suggested that this association may be explained by reverse causality»; «poor peri- odontal health may not necessarily cause dementia, but rather results from it» (Dibello et al., 2024). A long prodrome degrades self-care (including oral hygiene) years before diagnosis -> The U-Shaped Association Artifact.
- Shared confounders — «gender and soci- oeconomic status can be common risk factors asso- ciated with both dementia and PD» (Dibello et al., 2024); the discussion also names cardiovascular disease, diabetes, low education, smoking and drinking. These are largely the same cardiometabolic / socioeconomic / behavioural cluster the Lancet Commission already counts as dementia risk factors -> Dementia Prevention and Modifiable Risk Factors, so a raw PD-dementia RR is heavily confounded by already-counted levers.
- Confounding under-addressed — «confounding factors were not adequately considered in over 50% of the articles» (Dibello et al., 2024), on top of I2=95% heterogeneity.
Mechanism is directional only. Bacteremia/red complex pathogens seeding the brain,
neuroinflammation, amyloid-β and tau (from preclinical work), tooth-loss/mastication routes — all
mechanistic/animal, no human interventional confirmation
(Dibello et al., 2024).
Mechanistic plausibility is not outcome evidence (the mechanism-directional rule): it justifies
holding the lever as a candidate, not recommending it for dementia.
Synthesis — the cross-arm decomposition (distinction, NOT tension)
The two arms answer different questions with different quantities, so this is a distinction,
not a joined-issue tension. The parameter table’s Same quantity? column is NO on every row
(this cross-arm decomposition is the wiki’s own synthesis move
(Dibello et al., 2024; inferred from Simpson et al., 2022)):
| Parameter | Arm 1 — glycaemic (Simpson) | Arm 2 — dementia (Dibello) | Same quantity? |
|---|---|---|---|
| Exposure | periodontal treatment (SRP) vs usual care | having periodontal disease (present/absent) | NO |
| Design | 35 parallel RCTs (interventional) | 8 observational cohorts (dementia sub-arm) | NO |
| Effect measure | HbA1c mean difference -0.43% | dementia RR 1.22 | NO |
| Outcome | HbA1c (glycaemic surrogate), 3-4 mo | incident dementia, ~11 y | NO |
| Certainty | GRADE moderate (RCT, RoB-downgraded) | observational; confounded + reverse-causal | NO |
| Causal claim licensed | YES — treating lowers HbA1c | NO — association only, no treatment trial | NO |
The decision payoff [inferred from @simpson2022periodontitis; @dibello2024periodontal]. Rank one exposure’s arms by evidence strength, not by which is more talked-about:
- Act on Arm 1 for a person with untreated periodontitis and T2D — a genuine, low-harm, probably-additive glycaemic increment (modest, surrogate-based, durability-limited to ~6 mo).
- Hold Arm 2 as a candidate lever, not a recommendation — the dementia association is real but confounded and possibly reverse-causal, and the interventional test (does treating PD lower dementia incidence?) has not been done. an RCT/quasi-experimental periodontal-treatment -> cognitive-outcome study before any dementia-prevention claim.
- A tempting indirect route, explicitly discounted. Because T2D is itself a Commission dementia risk factor, one could argue periodontal treatment -> better glycaemic control -> lower dementia risk. This is speculative and double-counts: the HbA1c effect is small and short-horizon, the diabetes->dementia contribution is modest, and it routes through an already-counted lever — not an independent dementia benefit (Dibello et al., 2024; inferred from Simpson et al., 2022).
Confidence
medium — carried by Arm 1’s moderate-certainty RCT evidence, which is the page’s strongest and
most decision-relevant claim. Arm 2 is explicitly the weaker, low-grade, association-only
leg. A high would be unearned: Arm 1’s outcome is a surrogate over a short horizon, and Arm 2
cannot support a causal recommendation at all. Coherence, not validity (R1): no operation
here grades either arm against a realized complication or dementia outcome; the loop is open.
Self-critique [run 2026-08-23, before commit]
- Not laundered. The two arms are kept individually attributed (Simpson = glycaemic RCT; Dibello = observational dementia) — no number is blurred across arms. The cross-arm claim is tagged as this page’s own synthesis move.
- Distinction, not tension (not-joined guard applied). The parameter table is NO on every row
— different exposure (treat vs have), design, measure, outcome, horizon — so it is filed as a
decomposition/distinction, never a
[[tension]]. Nocontradicts:edge. - No overclaim on dementia. The dementia arm is stated as association with all three
source-given caveats (reverse causation, shared confounders, confounding under-addressed); the
treat gums to prevent dementiaclaim is explicitly withheld, and the indirect glycaemic route is discounted rather than asserted. No independence (type-E) claimed between the two sources — they share no data, but they also do not corroborate one claim; they decompose one exposure. - Durability honesty. The 12-month HbA1c figure is flagged as single-study rather than read as sustained benefit — the horizon caveat the task and source both bear.