cup of coffee on white ceramic saucer

Do Sugar Substitutes Speed Brain Aging?

A Brazilian cohort tied higher sweetener intake to slightly faster cognitive decline, but not at 60 and over and not in complete-case analysis.

A shaky study inspired “Sugar substitutes linked to faster brain aging” headlines was corrected twice, the link was absent in people 60 and over, and it disappeared when the analysis was limited to participants measured at all three waves.

A Brazilian cohort study is circulating again under a headline about six sweeteners and faster brain aging. The paper has been corrected twice, and its association was absent in people 60 and over and absent when the analysis was limited to participants tested at all three waves.

What this means

In 12,772 Brazilian civil servants followed a median of eight years, those reporting the most low- and no-calorie sweeteners at the start declined slightly faster on tests of global cognition and verbal fluency than those reporting the least. The difference was absent at 60 and over, and it disappeared in a complete-case analysis. Nothing here shows that sweeteners caused cognitive change, and one corrected cohort cannot settle the question in either direction.

What the study measured

ELSA-Brasil enrolled 15,105 civil servants in six Brazilian cities; 12,772 had usable dietary and cognitive data at the first wave, mean age 51.9. Diet came from a 114-item food frequency questionnaire, given once at baseline. Cognition was tested three times, about four years apart, from 2008-10 to 2017-19; median follow-up was eight years.

Most of the exposure was not the pink packet: mean intake was 92.1 mg a day, and 63.8 mg of it was sorbitol, a sugar alcohol in sugar-free candy, gum and coffee sweeteners; erythritol was near zero at 0.1 mg a day. The top third of intake began at 102.4 mg a day. That is a population sweetening coffee and buying diet versions of everyday products, not one working through tabletop sachets.

The estimate the headlines ran with

On global cognition, the top tertile’s decline was steeper than the bottom’s by a modeled difference of 0.024 z-score units (95% CI -0.040 to -0.016); the middle tertile fell between at -0.016 (-0.032 to -0.001). The paper’s own figures place these as slopes on an age timescale; the ‘extra years’ figure the authors attach to them carries that model’s assumptions. Verbal fluency moved further: -0.040 (-0.064 to -0.024) for the top third against the bottom, trend test p < 0.001. The authors describe those slopes as 35% and 62% faster decline, or about 1.3 and 1.6 extra years of cognitive aging — derived figures, published without intervals, the second assuming the lowest-intake group aged normally.

Memory is the domain the press framing leans on. The top tertile’s estimate was -0.024 (-0.040 to -0.004), an interval excluding zero, but the trend test across tertiles came in at p = 0.058, just above 0.05. On Trail-Making Test B, a timed measure of processing speed, there was no association for combined intake or any single sweetener.

Two nulls the headline left out

The first is age: the interaction was strong (p < 0.001). Below 60 the pattern held — verbal fluency -0.040 (-0.064 to -0.016), global cognition -0.024 (-0.040 to -0.008). At 60 and over, no tertile was associated with any cognitive outcome — the age range where dementia is diagnosed and an effect would be easiest to detect.

The second is the sensitivity analysis: restricted to people tested at all three waves, no association between tertiles and decline. The published estimate therefore depends on inverse probability weighting, which adjusts for those who died or stopped attending. That is legitimate, but it makes the headline number a modeled quantity; the unweighted comparison in the same paper is flat. The analyses also draw on different people: 5,869 participants under 55 got no cognitive battery at wave two, so the complete-case sample skews older. To read an interval like that, see How to Read Three Numbers in Nutrition News.

What the two corrections changed

Neurology has corrected this paper twice. The first erratum ran Dec. 9, 2025 (105(11):e214393, doi:10.1212/WNL.0000000000214393, PMID 41197106). The second ran online Jul. 30, 2026 and in the Aug. 25, 2026 issue (107(4):e218380, doi:10.1212/WNL.0000000000218380, PMID 42531540). Both notices are recorded on the article page and in the journal’s correction history. The corrected article differs from the original version in several estimates — the daily-use results and some individual-sweetener intervals among them — while the tertile estimates quoted below were unchanged by the re-analysis.

One point to read carefully: the abstract reports its tertile estimates only for participants younger than 60 — top-tertile verbal fluency -0.040 (-0.064 to -0.016) and global cognition -0.024 (-0.040 to -0.008) — and those are the paper’s own age-stratified results, not a contradiction of anything. The whole-sample estimates quoted above (-0.040 to -0.016 and -0.064 to -0.024) are a different analysis in the same paper.

Placed against the rest of the evidence

Two beverage cohorts frame the question and disagree. In Framingham Offspring — 1,484 people aged 60 and over — daily artificially sweetened soft drink carried a hazard ratio of 2.89 for Alzheimer disease dementia against none (95% CI 1.18 to 7.07); over ten years of surveillance the sample saw 81 incident dementia cases, 63 of them consistent with Alzheimer disease. A 2025 pooled analysis of six US cohorts — 10,974 adults aged 65 and over, 2,445 cases over 116,067 person-years — found for artificially sweetened beverages 1.00 per serving per week (0.99 to 1.01) and 1.00 (0.83 to 1.21) for highest against lowest intake. The null is larger, newer and better powered; neither covers total intake in midlife.

We read the corrected article in full; the beverage studies only as abstracts, the editorial only as a record. The beverage question is its own story: Do Diet Drinks Work as Well as Water for Weight Loss?.

What remains standing

Strip out the amplification and something survives: in middle-aged adults, higher intake went with slightly faster decline on two measures, in rough dose order. The authors also report a wider gap in diabetes: interaction p = 0.016, with -0.063 (-0.105 to -0.021) in the 1,409 participants with diabetes against -0.018 (-0.034 to -0.002) in the 11,363 without. That subgroup carries an obvious confounder — people with diabetes use more sweeteners and start from higher risk — and a signal a trial could test in middle-aged adults.

The paper cannot close the question. It has no neuroimaging, so the mechanisms usually invoked — neuroinflammation, the blood-brain barrier, the microbiome — go untested. Diet was measured once, leaving eight years of reformulation and shifting market share invisible. Cohorts cannot separate sweeteners from the behaviors that travel with them, and the sugar-alcohol results have no independent replication we could find. A small, model-dependent association in one cohort is grounds for a trial, not for a headline about six products — and whether these compounds affect cognition across decades is still open.

The article text carries no funding or disclosure statement; the publisher’s page states under study funding that the authors report no targeted funding, and under disclosure that the authors report no relevant disclosures, with full disclosures at Neurology.org/N. Those are the journal’s own statements, found in an archived capture of the page, and are recorded, not audited.

Terms explained

  • z-score — a score in standard deviations from a group’s average; 0.024 is small against the spread.
  • Tertile — one of three equal groups ranked by intake; comparisons are highest against lowest.
  • Inverse probability weighting — reweighting by how likely each participant was to stay in the study, so those who dropped out are represented.
  • Complete-case analysis — an analysis keeping only participants with data at every time point; here, all three waves.
  • Sugar alcohol (polyol) — a low-calorie sweetener such as sorbitol or xylitol; sorbitol supplied most of the intake here.

Sources

  1. Gonçalves NG, Martinez-Steele E, Lotufo PA, Bensenor I, Goulart AC, Barreto SM, Giatti L, de Faria CP, Molina MDCB, Caramelli P, Marchioni DM, Suemoto CK. Association Between Consumption of Low- and No-Calorie Artificial Sweeteners and Cognitive Decline: An 8-Year Prospective Study. Neurology. 2025;105(7):e214023. doi:10.1212/WNL.0000000000214023 · PMID: 40902134 · https://doi.org/10.1212/WNL.0000000000214023
  2. Erratum for ref. 1. Neurology. 2025;105(11):e214393. doi:10.1212/WNL.0000000000214393 · PMID: 41197106 · https://doi.org/10.1212/WNL.0000000000214393
  3. Correction for ref. 1. Neurology. 2026;107(4):e218380. doi:10.1212/WNL.0000000000218380 · PMID: 42531540 · https://doi.org/10.1212/WNL.0000000000218380
  4. Pase MP, Himali JJ, Beiser AS, et al. Sugar- and Artificially Sweetened Beverages and the Risks of Incident Stroke and Dementia: A Prospective Cohort Study. Stroke. 2017;48(5):1139-1146. doi:10.1161/STROKEAHA.116.016027 · PMID: 28428346
  5. Chen H, Ding Y, Dhana K, et al. Sweetened Beverages and Incident All-Cause Dementia Among Older Adults. JAMA Psychiatry. 2025;82(8):801-809. doi:10.1001/jamapsychiatry.2025.1230 · PMID: 40531522
  6. Holland TM. The Dark Side of Sweet: Neurocognitive Consequences of Artificial Sweeteners. Editorial. Neurology. 2025;105(7):e214129. doi:10.1212/WNL.0000000000214129 · PMID: 40902133 · https://doi.org/10.1212/WNL.0000000000214129
  7. Six popular sugar substitutes linked to faster brain aging. The Times of India. Oct. 30, 2026. https://timesofindia.indiatimes.com/etimes/wellness/six-popular-sugar-substitutes-linked-to-faster-brain-aging/articleshow/135367563.cms

Related Articles