In the health sector, there was a significant stir over news that “Artificial sweeteners are associated with accelerated cognitive decline”. Incredibly, this didn’t come from a tabloid newspaper seeking engagement, but from the Instagram account of Ciência USP, one of the most respected higher education institutions – if not the most – in Latin America.
The post, which already has over 17,000 likes, highlights a study conducted in Brazil that suggests regular consumption of low- or no-calorie artificial sweeteners can accelerate cognitive decline, affecting memory and verbal fluency. The results were obtained from monitoring more than 12,000 people over eight years.
Of course, the post also presented information about the study’s limitations. However, what caught my attention most – and what I consider noteworthy – was the engagement of several internet personalities linked to evidence-based medicine, such as Cadu Viterbo and Laura Marise, among others, who attempted to clarify why the post was problematic. This is because, at many points, even using the word “association,” the post gave the impression that it was a study capable of establishing causality.
Despite attempts to stem the tide of misinformation and alarmism, the effort was insufficient. Major newspapers, such as O Globo, began to repost the material in a clickbait format, as in this case: “Most sweeteners are associated with up to 62% faster cognitive decline, shows an unprecedented study from USP.”
In fact, the impact was so great that even Agência SP – the official news production and distribution platform of the São Paulo Government – reproduced the finding, surprisingly also without highlighting that it was an observational study and, therefore, could only verify associations.
Unfortunately, the headline of an article is the first glimpse readers get of its content. If we headline an article something like “Dihydrogen Monoxide: Why Do We Continue to Sell a Substance Found in 100% of Recent Deaths?”, readers are likely to think they’re dealing with an extremely dangerous substance, when in fact it’s just pure water (H₂O).
Of course, in my ideal world, all readers would get past the article title and understand the context. However, as we know, most people don’t read the entire content and – even more tragically, this is true even whlen they share it. This creates almost a “snowball effect”: you see a friend or family member sharing a piece of news, believe they’ve read it, evaluated it, and found it relevant, and then you share it without reading it, expanding the reach of the information.
Given this, I believe we should address the article in its entirety and explain its actual findings and its limitations. However, before unpacking the USP findings, it’s vital to understand the study’s methodological basis: the Longitudinal Study of Adult Health – ELSA – which tracks health determinants and cognitive tests over time.
The ELSA
According to the article “Brazilian Longitudinal Study of Adult Health (ELSA-Brasil): Objectives and Design“, ELSA-Brasil is a cohort study involving 15,105 civil servants from six institutions in different regions of Brazil. In this type of study, a group of people are followed over time by researchers who seek to observe how certain “exposures” – to foods, pollutants, activities – relate to certain “outcomes” – such as health problems. Active or retired employees aged 35 to 74 were considered, excluding pregnant or recently pregnant women, individuals with severe cognitive or communication impairments, and retirees residing outside the metropolitan area of the study centre.
The study aims to investigate the incidence and progression of diabetes and cardiovascular disease, considering biological, behavioural, environmental, psychological, social, and occupational factors. The first examination took place between 2008 and 2010, and the first follow-up between 2012 and 2013.
To establish a health baseline, participants underwent an initial interview about family history, dietary intake, social class, and social capital, as well as an electrocardiogram, anthropometric assessment, fasting blood collection, blood pressure measurement, and cognitive testing.
Three cognitive tests were applied: one on memory – word learning and retention – and two on executive functions, related to attention, concentration and psychomotor speed.
In the field of nutrition, the study “Reproducibility and relative validity of the ELSA-Brasil Food Frequency Questionnaire” evaluated the reliability and relative validity of the Food Frequency Questionnaire (FFQ). From a sample of 15,105 participants, 300 individuals were randomly selected (150 men and 150 women), distributed by age groups (35–54 and 55–74 years) and employee categories. For these participants, the FFQ was administered at the first contact (FFQ1, October 2009) and again in October 2010 (FFQ2). Three 24-hour dietary recalls were conducted, four months apart and between the two FFQs, for comparison of results.
To increase the accuracy of records, participants detailed all food and beverages consumed, using an album with life-size photos of utensils to estimate portions, as well as instructions on the days of the week and month to record.
The FFQ, consisting of 114 items, was semi-quantitative, assessing the frequency and average quantity of foods consumed in the last 12 months. Questions included sweetened beverages (“with sugar,” “with sweetener,” and “without sugar”) and eating habits, such as consumption of fast food and sweets. Each region included up to two local items, such as cuscuz paulista and Japanese food in São Paulo.
The final sample consisted of 281 participants (136 men and 145 women), with 54.8% between 35 and 54 years old and 45.2% between 55 and 74 years old. High variability in energy and nutrient intake was observed between FFQ1 and FFQ2, with lower averages in the second questionnaire, although carbohydrates, fibre, iron, potassium and zinc presented higher average values in FFQ2.
The intraclass correlation coefficient – ICC – is an essential statistical measure in this context, as it assesses both consistency and absolute agreement between instruments, that is, whether the methods generated results very similar to each other. The ICC showed agreement of 0.55 (proteins) to 0.83 (vitamin E) between nutrients, being generally higher for micronutrients.
Average energy and macronutrient values also varied: energy (ICC 0.51), proteins (0.38) and lipids (0.44). The percentage of agreement between the FFQ2 and food records ranged from 36.6% (fibre) to 48% (calcium), with an average disagreement of 13.5%.
In quantitative terms, the FFQ2 overestimated intake: on average, 783 kcal more for energy, 85 g for carbohydrates and 9.35 mg for iron, attributable to characteristics of the instrument, such as an extensive list of foods and the non-exclusion of extreme values.
Considering the ICCs found, it is concluded that the ELSA-Brasil FFQ presents reasonable relative validity to classify participants according to intake levels, allowing multiple dietary analyses.
One of the outcomes of ELSA-Brasil was the article “Examining the Usage Patterns of Non-Nutritive Sweeteners among Non-Diabetic Individuals: Insights from the Longitudinal Study of Adult Health (ELSA-Brasil)“, which estimated the prevalence of regular consumption of non-nutritive sweeteners – NNS – and their associated factors among non-diabetic individuals at the study’s baseline.
Exclusion criteria were applied: participants who did not respond to the food consumption section; underwent bariatric surgery; who reported dietary changes in the previous six months; presented implausible energy intake; or were diagnosed with diabetes mellitus or taking medication for the condition.
The final sample included 9,226 individuals, and regular NNS consumption was defined as ingestion of at least one NNS-sweetened product at least once per day.
The prevalence of regular NNS consumption was 27.5%, with coffee being the main source (20.9%). Regarding the sample characteristics, 39.2% were overweight, 64.5% had no family history of diabetes, 68.3% did not report prior hypertension, and 60.7% had higher education. Women had a higher prevalence of consumption (30.5%).
Regarding lifestyle, 80.3% reported alcohol consumption, 42.9% were smokers, and 78.3% had low levels of leisure-time physical activity. A progressive increase in the prevalence of NNS was observed with advancing age and higher BMI categories, in addition to a direct association with higher education level and per capita income. Other factors associated with consumption were: not being married, family history of diabetes, hypertension, and alcohol consumption.
In the adjusted exploratory analysis, women were almost twice as likely to consume alcohol as men, and white individuals were 50% more likely than black individuals. Participants aged 65 to 74 were 1.4 times more likely to consume alcohol than those aged 35 to 44, while obese individuals were 6.1 times more likely.
Having a family history of diabetes, systemic arterial hypertension, and alcohol consumption increased the likelihood of regular NNS consumption by about 20%, while moderate or intense levels of leisure-time physical activity increased the odds by 30%. Having higher education and an income above two minimum wages increased the likelihood of regular consumption by 80%, compared to the lowest income brackets.
Broadly speaking, this was the “theoretical basis” behind the study that generated so much controversy on social media. From the outset, some limitations can be identified.
These include: the fact that this is an observational study, which, due to its methodological nature, does not allow for the inference of causality, only correlation; the use of dietary assessment tools that, although fundamental in epidemiological investigations, are subject to significant biases, such as memory bias (given the difficulty of accurately recalling food consumption from previous days or months) and social desirability bias (responding in a socially acceptable way, such as underestimating the consumption of ultra-processed foods or overestimating the consumption of foods considered healthy); and the possibility of residual confounding variables, even after statistical adjustments, which may influence the findings.
With these considerations in mind, we can now move on to discussing the USP study.
Do sweeteners accelerate decline?
Published in the journal Neurology, the article “Association Between Consumption of Low and No-Calorie Artificial Sweeteners and Cognitive Decline” investigated the relationship between the consumption of seven low or no-calorie sweeteners – LNCS – and cognitive decline over eight years, based on the hypothesis that greater consumption would be associated with faster cognitive decline.
The ELSA-Brasil sample was evaluated at three time points (2008–2010, 2012–2014, and 2017–2019). At baseline, participants with incomplete data, extreme caloric intake, or a history of Parkinson’s disease were excluded, resulting in 12,772 individuals.
In subsequent measurements, losses occurred due to deaths, non-attendance, or missing data, totaling 5,784 and 10,707 participants evaluated, respectively. It is worth noting that, unlike the first two assessments, in which cognitive tests were administered only to participants over 55 years of age, the third assessment included all individuals.
The seven LNCS were identified from tabletop sweeteners and the composition of light and diet beverages. Based on this, combined consumption was calculated, and participants were categorised into tertiles, divided into three concentration groups: first tertile (0.02–37.2 mg), second tertile (37.3–102.3 mg), and third tertile (102.4–856.5 mg).
Furthermore, participants were classified according to the frequency of LNCS use into two categories: “no consumption/sporadic consumption (less than once a day)” and “daily consumption”.
Cognitive performance was assessed every four years. Episodic memory – responsible for receiving and storing information about specific episodes or temporal events – was measured using the Consortium to Establish a Registry for Alzheimer’s Disease word list, a battery of tests capable of assessing cognitive status and its changes, especially in older populations.
Semantic and phonemic verbal fluency tests were used to assess language and executive function – in which participants typically have one minute to produce as many words as possible from a given category or that begin with a specific letter.
Processing speed and executive function were assessed with version B of the Trail-Making Test.
The lifestyle and sociodemographic variables assessed were similar to those of other ELSA-Brasil studies, including BMI, diabetes, self-reported alcohol and tobacco consumption, and leisure-time physical activity. One difference was the classification of the diet as healthy or unhealthy, based on the MIND standard, which encourages green leafy vegetables, berries, and olive oil, and discourages fried foods, sweets, and red meat.
After eight years, participants in the highest consumption tertile were found to have a 32% greater rate of memory decline compared to those in the first tertile. Furthermore, compared to the first tertile, participants in the two highest tertiles had 110% and 173% greater rates of verbal fluency decline, respectively, and 35% and 62% greater declines in global cognition, corresponding to an excess cognitive aging of 1.3 to 1.6 years. Aspartame was associated with faster decline in memory, verbal fluency, and global cognition; acesulfame-K and erythritol affected memory and global cognition; tagatose had no significant effect.
When considering age, it was found that this factor modified the associations. Consumption in the highest tertile was associated with a faster decline in verbal fluency and global cognition in participants under 60 years of age. Interestingly, among participants aged 60 and over, no significant association was observed between tertiles of LNCS consumption and cognitive decline.
Stratifying by obesity, diabetes, and diet, only the presence of diabetes modified the relationship between total LNCS consumption and cognitive decline. In participants without diabetes, combined LNCS consumption in the highest tertiles was associated with a faster decline in verbal fluency and global cognition. In participants with diabetes, combined consumption in the highest tertiles was associated with accelerated decline in memory and global cognition.
Regarding individual sweeteners, in non-diabetics, erythritol and xylitol accelerated the decline in memory and global cognition, while sorbitol affected memory, verbal fluency, and global cognition. In diabetics, aspartame and acesulfame-K accelerated the decline in memory, verbal fluency, and global cognition; saccharin and xylitol affected memory and global cognition; and tagatose accelerated the decline in executive function.
Given these findings, the authors conclude that LNCS consumption was associated with an accelerated rate of cognitive decline during eight years of follow-up in this large cohort of middle-aged and older adults, suggesting the possibility of long-term detrimental effects of LNCS consumption, particularly artificial sweeteners and sugar alcohols, on cognitive function. Possible mechanisms that could explain these results include neurotoxicity, neuroinflammation induced by toxic metabolites of artificial sweeteners, and possible alterations in the gut microbiota.
Therefore, opting for more natural alternatives, such as tagatose, or other sugar alternatives may help mitigate the potentially harmful association observed. Future research is needed to confirm these findings and investigate whether other alternatives to refined sugar, such as honey, may be effective in this regard.
Among the limitations highlighted by the researchers, it is worth noting that long-term studies are subject to participant loss – as seen, since the first measurement, of the 15,105 individuals, only 3,857 completed all the cognition tests.
Furthermore, although a validated food frequency questionnaire was used, the possibility of reporting errors cannot be ruled out. Many covariates were self-reported, which may introduce social desirability bias.
Similarly, although regression models were adjusted for clinical and lifestyle variables, residual confounding cannot be excluded, particularly because health behaviours tend to co-occur, and some groups may consume more LNCS due to lifestyle and clinical history.
Personally, I find this quite plausible, and Figure 4 of the article supports this interpretation: the highest tertiles of total sweetener consumption – although they had minimal impacts in patients without diabetes – showed more significant associations with cognitive decline in patients with diabetes. This raises the possibility that the findings are not necessarily caused by the sweeteners, but rather by the disease itself when left untreated – something much more likely.
Another noteworthy aspect is the relationship between tagatose and cognitive decline in diabetic patients. Although this sweetener did not demonstrate statistically significant negative effects in most tests and even produced potentially neutral or beneficial effects on global cognition, the subsample of diabetic patients was associated with a more rapid decline in executive function. In theory, this contradicts the researchers’ conclusion that more natural options may be good alternatives to LNCS.
Furthermore, diet was assessed only at baseline, which may not reflect dietary changes over time and lead to an underestimation – or overestimation – of the associations between LNCS and cognition.
Finally, due to the unavailability of neuroimaging, it was not possible to investigate structural brain changes or potential mechanisms that explain the associations between LNCS consumption and cognitive decline in specific groups.
Ultimately, given the clear limitations, I can say that sweeteners – at least based on the evidence presented in this study and many others – remain safe. What’s concerning isn’t the metallic or bitter aftertaste some leave, but rather the fact that USP’s Instagram account didn’t take a strong stance and remove the alarmist language from the post.
References
- SUNDAR, S. et al. Sharing without clicking on news in social media. Nature Human Behaviour volume 9, pages156–168 (2025).
- AQUINO, E. et al. Brazilian Longitudinal Study of Adult Health (ELSA-Brasil): Objectives and Design. American Journal of Epidemiology 175(4):315-24. February 2022.
- MOLINA, M. et al. Reproducibility and relative validity of the ELSA-Brasil Food Frequency Questionnaire. Cad. Public Health 29 (2). Feb 2013.
- MOREIRA, T. Examining the Usage Patterns of Non-Nutritive Sweeteners among Non-Diabetic Individuals: Insights from the Longitudinal Study of Adult Health (ELSA-Brasil). Nutrients. 2023 Nov 15;15(22):4785.
- GONÇALVES, N. et al. Association Between Consumption of Low- and No-Calorie Artificial Sweeteners and Cognitive Decline: An 8-Year Prospective Study. Neurology. 2025 Oct 7;105(7):e214023.
This story was originally published by Revista Questão de Ciência in Brazil. It is translated and reprinted here with permission.
