The decision. Should GI/GL be a marker you steer diet by, alongside (or instead of) fibre and
whole-grain content? GI = how much 50 g of carbohydrate in a food raises postprandial blood glucose
vs a glucose/white-bread standard; GL = GI x the carbohydrate amount. Single primary source
(Jenkins 2024, gold MA), confidence: low — and read the whole page through the source’s provenance:
its lead author invented the GI and the paper is an explicit rebuttal to the WHO/Reynolds
conclusion that GI has «little relevance» (see Guidance divergence below).
The associations are real and consistent, mostly GRADE-low
Jenkins pooled 10 mega-cohorts (>=100 000 each, Richard Doll Consortium), highest-vs-lowest quantile, most-adjusted, fixed-effects. High GI tracks higher risk across all four main outcomes; GL matches GI for T2D and CVD but not cancer/mortality.
| Exposure -> outcome | RR (95% CI) | I2 | GRADE / NutriGrade |
|---|---|---|---|
| GI -> type 2 diabetes | 1.27 (1.21-1.34) | 71% | low / moderate |
| GI -> total CVD | 1.15 (1.11-1.19) | 35% | low / moderate |
| GI -> all-cause mortality | 1.08 (1.05-1.12) | 90% | low / moderate |
| GI -> diabetes-related cancer | 1.05 (1.02-1.08) | 23% | low / moderate |
| GL -> type 2 diabetes | 1.15 (1.09-1.21) | 60% | — |
| GL -> total CVD | 1.15 (1.10-1.20) | 38% | — |
| GL -> cancer / mortality | NS or inverse | — | — |
- The one GRADE-moderate cell is GI -> CVD mortality 1.14 (1.08-1.21); GI -> CHD mortality 1.33 (1.14-1.55) and stroke mortality 1.29 (1.10-1.50) are the largest components. NutriGrade (used deliberately because «GRADE might not adequately assess … prospective cohort dietary assessments») rates the mains moderate — so the certainty verdict itself depends on the instrument chosen.
- These are relative risks on a modest absolute base. For most outcomes the extreme-quantile RR is 1.05-1.15; the decision-weight is small except for the T2D 1.27 arm. E-values exceeded 1.2 (robust to a moderate unmeasured confounder) for the main GI/GL outcomes except GI/GL -> cancer and GL -> mortality. (Jenkins et al., 2024)
Independent lever, or a proxy for the carb-quality pattern? (the crux)
Jenkins’ headline is that low-GI associations are «similar to those for low GI diets» for fibre and whole grains in the same cohorts, and reads this as validating GI as a carbohydrate-quality predictor. The wiki’s read is more cautious — the same data are as consistent with GI being a proxy:
- The markers are collinear. «foods high in carbohydrates also tend to have a high GI» (Jenkins et al., 2024) — GI, GL, fibre and whole-grain content ride together in whole-food diets, so near-identical associations are exactly what proxies of one underlying pattern produce.
- The mutual adjustment is too sparse to separate them. «only 12 (17%) of the 69 GI main outcome assessments controlled for fibre, controlling for fibre or not made little difference»; no fibre or whole-grain assessment controlled for GI. (Jenkins et al., 2024) little difference on 17% is weak evidence of independence, not the mutually-adjusted test that would settle it.
- This is the Is the Food Category Doing Any Work question applied to a construct: similar magnitude + shared cohorts + shared confounding + minimal cross-adjustment cannot attribute the effect to GI rather than the fibre/whole-grain/low-refined-carb pattern that carries it. (inferred from Jenkins et al., 2024)
Direction has human-corroborated mechanistic support (so this is not a GI does nothing verdict): postprandial glucose excursions -> oxidative stress/free radicals -> insulin resistance and vascular dysfunction (Ceriello, Monnier), and acarbose RCTs (STOP-NIDDM, ACE) — which pharmacologically create a low-GI diet «without other dietary changes» — cut incident T2D. (Jenkins et al., 2024) That is a directional mechanism, not outcome-proof that steering food GI moves events independently of fibre.
Why the estimate is fragile — measurement error, twice over
GI/GL inherit FFQ error (Measurement Error in Dietary Assessment) plus GI-table assignment error (Atkinson 2008 tables mapped onto FFQ foods; local tables for Asian FFQs; PURE bread-scale converted). And the studied contrast is narrow: lowest-quantile mean GI 58, highest 67 (glucose scale), which the authors attribute to «a small range of low GI and GL values, possibly related to health-conscious individuals». (Jenkins et al., 2024) A narrow, error-laden contrast attenuates a real gradient toward the null and makes any dose-response knee unlocatable — consistent with the wiki’s The Underivable Optimum reading. No plateau/knee was located; monotone «over the studied range» only.
Guidance divergence — a certainty-grading dispute, not a data clash
The 2019 WHO-sponsored Reynolds/Mann series and the 2023 WHO carbohydrate guideline endorsed fibre and whole grains but held «GI … was not considered to be a relevant dietary factor for the prevention of chronic diseases». (Jenkins et al., 2024) Jenkins calls that «counterintuitive» and this MA is the rebuttal. But this is largely NOT a dispute about the associations — Jenkins notes the WHO appendix itself carried significant GI associations. It is a disagreement about whether GRADE-low cohort evidence is strong enough to act on (GRADE downgrades observational data; NutriGrade, which Jenkins co-develops, does not), plus a live conflict of interest (the GI’s inventor, ICQC co-chair, IDF-guideline author, extensive food-industry funding). Not filed as a joined tension (the not-joined check: same evidence base, different certainty threshold) — see the G-gap. (inferred from Jenkins et al., 2024)
Decision relevance
- GI/GL is not a separate big rock. On this evidence, steering by fibre and whole-grain content captures most of what steering by GI would, because the markers are collinear and GI’s independent contribution is unproven. The practical advice — more whole grains, legumes, intact/minimally-refined carbohydrate, fewer refined/sugary carbs — is the same one fibre and whole grains already yield.
- The T2D and CVD-mortality arms are the ones with any decision weight (RR 1.27 and 1.14, moderate where anything is); cancer and GL-mortality are the softest (E-value <1.2, GL inverse/null).
- Where GI plausibly adds is a stratum signal, not a population lever: the impaired-glucose / insulin-resistant, where postprandial excursions and the acarbose RCTs bite -> Carbohydrate Restriction and Type 2 Diabetes Remission, Continuous Glucose Monitoring as a Health Intervention.
Certainty and gaps
confidence: low— single primary source, deeply-conflicted lead author, GRADE-low mains, an unresolved proxy-vs-lever question, and a narrow error-laden exposure contrast. The magnitudes are sound (gold MA); the reframe (GI as an independent carb-quality lever) is not established.- Gaps (G): the mutually-adjusted GI-net-of-fibre-and-wholegrain test the independence claim needs is largely absent (only 17% adjusted for fibre); no long-term GI RCT on hard outcomes (acarbose is the pharmacological proxy); the WHO/Reynolds GI-dismissal is not extracted as its own claim, so a clean joined tension is deferred to a Revisit.
Self-critique [run 2026-08-13, before commit]
- Food-category launder check (the guard for this ingest). The page does not credit GI when
fibre/whole-grain/matrix may be the actor — the crux section leads with the proxy reading and tags the
independence doubt as the wiki’s. No
[E-independent]claimed (there is one primary source). - Over-claim. No superlative scoped to the vault’s holdings; the source’s own reframe is reported as the authors’ interpretation and explicitly not adopted. The RRs are extracted, the proxy reading is .
- Symmetric standards. The magnitudes are held even though they run with a plausible mechanism; the reframe is discounted for COI and the wiki flags the source’s convenient asymmetry (a confounder invoked only for the disconfirming GL-cancer arm). The conflict does not delete the pooled numbers.
- Not-joined / counter-passage. The GI-vs-WHO clash was checked and found to be a certainty-grading disagreement over shared evidence -> recorded as a distinction, no tension filed.