Artificial sweeteners have been linked to an increased risk of type 2 diabetes: a meta-analysis of more than 721,000 people showed a 31% increase in relative risk.

High or regular consumption of artificial sweeteners has been associated with a higher risk of developing type 2 diabetes in the long term. This is the conclusion reached by the authors of a new systematic review and meta-analysis published on August 11, 2026, in the journal Frontiers in Nutrition. The pooled analysis included 721,003 adults followed in six large prospective cohort studies in the United States, Europe, France, and Japan.

During a follow-up period of 5 to 20 years, 38,445 new cases of type 2 diabetes were recorded, with a total of 7,280,376 person-years of observation. In the pooled sample, the crude incidence rate was approximately 5.28 cases per 1,000 person-years. The primary comparison was between individuals with the highest consumption of artificial sweeteners or artificially sweetened beverages and those who rarely consumed them or did not consume them at all.

The pooled odds ratio was 1.31, with a 95% confidence interval of 1.16 to 1.49. In other words, high exposure was statistically associated with an approximately 31% higher relative incidence of type 2 diabetes. However, this is crucial: the meta-analysis combines observational studies and does not prove that sweeteners directly cause diabetes. The authors themselves rated the overall certainty of the evidence, according to the GRADE system, as low.

Furthermore, there was significant variability between studies: the statistical heterogeneity index I² reached 86.02%. Therefore, the figure of "31%" cannot be taken as a universal individual risk for any individual and any sweetener. Different studies examined different products—from total artificial sweeteners to diet soda—and used different definitions of "high" consumption.

Key results	Data
Type of work	Systematic review and meta-analysis
Number of studies	6 prospective cohorts
Total participants	721 003
New cases of type 2 diabetes	38,445
Summary observation	7,280,376 person-years
Duration of individual studies	5-20 years
General HR	1.31
95% confidence interval	1.16-1.49
Relative difference	≈ +31%
I²	86.02%
95% prediction interval	0.97-1.78
GRADE level of evidence	Short
Causality has been proven	No

[1]

What exactly did the researchers analyze?

The authors searched for prospective studies in all available publications up to March 31, 2025. The search included PubMed, Embase, Web of Science, and the China National Knowledge Infrastructure. Only studies of adults aged 18 years and older who were free of type 2 diabetes at baseline and who then followed the participants over time to record new cases of the disease were included. The analysis was performed according to the PRISMA guidelines.

The initial search yielded 108 publications. After removing duplicates, screening by title and abstract, and evaluating the full texts, only six studies met the criteria. This is an important feature of the study: the enormous 721,000 participants creates the impression of an extremely large evidence base, but the statistical results are actually based on just six independent cohort assessments.

All included studies assessed sweetener consumption using validated dietary questionnaires. Some studies directly analyzed total artificial sweetener consumption, while most others used artificially sweetened beverages or diet soda as the primary marker. Therefore, the results of the meta-analysis cannot be automatically generalized to each individual sweetener, such as sucralose, saccharin, aspartame, or acesulfame potassium.

New cases of diabetes were identified by self-reporting and subsequently confirmed by medical records, standardized diagnostic criteria, or national registries. For the meta-analysis, the authors used maximally adjusted models from the original studies, accounting for age, body mass index, and various lifestyle factors, including diet, smoking, alcohol, and physical activity.

Methodological characteristics	How it's done
Sources	PubMed, Embase, Web of Science, CNKI
Last search date	March 31, 2025
Age	≥18 years old
Design	Prospective observational studies only
Diabetes at the start	Was excluded
Exposition	Artificial sweeteners/artificially sweetened beverages
Nutrition assessment	Validated food questionnaires
Quality of research	Newcastle-Ottawa Scale
Quality scores	7-8 of 9
Statistical model	Random effects
The main measure	Hazard ratio

[2]

What six separate studies showed

The highest risk estimate was obtained in the French NutriNet-Santé cohort, published by Debras et al. Among 105,588 individuals followed for an average of 9.1 years, the highest versus lowest intake of artificial sweeteners was associated with a HR of 1.69—that is, an approximately 69 percent increase in the relative rate of diabetes incidence. In this cohort, 972 new cases of the disease were recorded.

In the largest European cohort, EPIC-InterAct, the analysis included 340,234 participants and 12,403 cases of diabetes. Those who consumed at least one serving of an artificially sweetened beverage daily had a HR of 1.18 compared with those who drank such beverages less than once a month. In the US MESA study, daily diet soda consumption was associated with a HR of 1.36 compared with no consumption.

In a Japanese cohort of 36,173 participants, 2,037 cases of diabetes were identified, and the highest intake of artificially sweetened beverages was associated with a HR of 1.43. In the US Health Professionals Follow-up Study, which was followed for 20 years, the HR was 1.31.

The largest modern American study, by Pacheco et al., combined data from the Nurses' Health Study, Nurses' Health Study II, and the Health Professionals Follow-up Study. It involved 191,805 people, and the number of new cases of diabetes reached 19,940. Among those consuming at least two servings of artificially sweetened beverages per day, the HR was 1.11 compared with those consuming less than one serving per month—the smallest effect size among the six included studies, but it was also statistically significant.

Study	Participants	Cases of diabetes	Observation	HR
Debras, NutriNet-Santé	105,588	972	9.1 years	1.69
Romaguera, EPIC-InterAct	340 234	12,403	11.3 years	1.18
Nettleton, MESA	6,814	413	7 years	1.36
Sakurai, Japan	36,173	2,037	5 years	1.43
de Koning, HPFS	40,389	2,680	20 years	1.31
Pacheco, NHS/NHS II/HPFS	191,805	19,940	7.5 years	1.11
United	721 003	38,445	-	1.31

[3]

What does "31% higher risk" actually mean?

A hazard ratio of 1.31 is the ratio of the rates of disease occurrence in two groups over time. This doesn't mean that a diet soda drinker has a "31% chance of developing diabetes," nor does it represent a 31 percentage point increase in individual risk. If the baseline absolute risk is small, the corresponding absolute increase may be significantly smaller; if the baseline risk is high, the absolute difference may be larger.

The meta-analysis also did not yield a single, universal absolute risk estimate for high and low exposure, as baseline diabetes rates, follow-up duration, age, and intake definitions varied greatly between cohorts. The rate of 5.28 cases per 1,000 person-years applies to the entire pooled population and is not a specific incidence among sweetener consumers.

Another important figure is the 95% prediction interval, which ranges from 0.97 to 1.78. Unlike the confidence interval of the pooled estimate, this range attempts to indicate the potential effect size in a new study comparable to the included studies. Since the lower bound of 0.97 is below 1, the result means that in a future study with a similar population, a virtually complete absence of a statistical association cannot be ruled out.

This is why the GRADE assessment was downgraded to low certainty. The pooled effect is statistically convincing, all six studies point in the same direction, and the leave-one-out analysis did not disrupt it. However, high heterogeneity, the observational design, and the small number of individual cohorts prevent the HR of 1.31 from being converted into evidence of causal harm.

Indicator	Correct interpretation
HR 1.31	The relative rate of diabetes onset is approximately 31% higher
This does not mean	"31% of people will develop diabetes"
95% CI 1.16-1.49	The pooled estimate is statistically different from 1
Prediction interval 0.97–1.78	In a new population, the effect could potentially be much weaker.
GRADE: low	Confidence in causal inference is limited
Absolute risk	Depends on the individual's baseline risk

[4]

All studies showed a link, but they were very different from each other.

The direction of the effect was remarkably consistent: all six studies had point estimates of HRs greater than one. When excluding each individual study, the pooled HR remained elevated, ranging from 1.23 to 1.37. This suggests that the overall result was not driven by a single, particularly large or unusual cohort.

At the same time, the I² value was 86.02%, which corresponds to very high statistical heterogeneity. Simply put, the studies varied more than would be expected from random error alone. One study found a result of 1.11, while another found a result of 1.69, and these estimates cannot be considered as measuring exactly the same effect.

The authors specifically examined possible causes of this variation and found that the definition of exposure was particularly important. When the analysis was restricted to studies of artificially sweetened beverages or diet soda, the heterogeneity decreased from approximately 86% to 68.8%, although it remained significant. Differences in populations and duration of follow-up may also have played a role, but their individual contribution was not statistically significant.

This is one of the main arguments against the conclusion that "all sweeteners increase the risk of diabetes by 31%." Most of the included data pertains specifically to beverages, whereas the French study assessed the total intake of artificial sweeteners. It is impossible to reliably determine from this study whether aspartame, sucralose, saccharin, acesulfame potassium, or their combinations behave similarly.

The main problem: people at high risk of diabetes are more likely to switch to “diet” foods themselves

This is the most compelling alternative explanation for the results, and the authors discuss it directly. People with obesity, prediabetes, a family history of diabetes, or already deteriorating metabolic indicators are significantly more likely to decide to give up regular soda and switch to sugar-free drinks. After a few years, some of these people develop diabetes—but the initial high risk may not have been due to the sweetener.

This mechanism is called reverse causality. A prospective design mitigates this problem: all individuals who already had diabetes at the start of the study were excluded, and only new cases were considered. However, it cannot completely rule out situations where someone already has obesity, insulin resistance, or prediabetes but does not yet meet formal criteria for diabetes, and therefore begins using sweeteners.

All six studies controlled for body mass index, and the models additionally adjusted for various dietary and lifestyle factors. However, statistical adjustment is not the same as randomization. For example, the degree of visceral adiposity, weight history, dietary characteristics, family risk factors, motivation for weight loss, or metabolic status at baseline may have been insufficiently measured.

Therefore, it's entirely possible that two components exist: part of the observed association may reflect the actual adverse effects of certain sweeteners or patterns of their consumption, while part may reflect the fact that these products are particularly frequently chosen by people with an already elevated risk of diabetes. This is precisely what cannot be reliably distinguished from existing cohort data.

Possible explanation for HR 1.31	How much was he excluded?
The real metabolic effect of sweeteners	Possible
Initial obesity	Statistically taken into account
Prediabetes before official diagnosis	It cannot be completely ruled out
Switching to diet foods due to high risk	Possible
Unhealthy overall diet	Partially taken into account
Smoking/alcohol/activity	Considered in a number of models
Residual mixing	It is impossible to completely exclude

[5]

How sweeteners could theoretically affect glucose metabolism

The first proposed mechanism involves the gut microbiome. Several experimental and clinical studies show that certain non-caloric sweeteners can alter the composition and functional activity of gut microorganisms, and through them, the individual's response to glucose. The authors of the meta-analysis consider this mechanism as one possible explanation for the long-term association.

Of particular interest is a 2022 study in Cell cited by the authors. In it, the effects of several non-caloric sweeteners on glucose tolerance were found to be variable and dependent on the individual microbiome. This further highlights the problem of generalization: the biological effects of the "sweetener" category can vary greatly depending on the specific molecule and the individual.

The second potential mechanism involves sweet taste receptors in the intestine. These are involved not only in taste sensation but also in endocrine signaling associated with the secretion of insulin and incretin hormones, including glucagon-like peptide-1. Theoretically, chronic exposure to a strong sweet taste without a corresponding energy intake could alter the normal regulation of these systems, although the clinical significance of this effect remains controversial.

The third mechanism concerns behavior and the reward system. The authors discuss the possibility that prolonged exposure to a very sweet taste may influence food preference, sweetness perception, appetite, and energy intake patterns. However, the new meta-analysis directly measured none of these mechanisms: the study demonstrates an epidemiological association, while the mechanistic explanations are based on other studies.

Possible mechanism	Proposed action
Gut microbiome	Altered glucose response and insulin sensitivity
Sweet taste receptors in the intestines	Altered endocrine signaling
Incretin system	Possible alteration of GLP-1 and other signals
Reward centers	Changes in food preferences
Maintaining a high sweetness diet	Possible effects on appetite and eating behavior
This meta-analysis has proven this.	No, these are supposed mechanisms.

[6]

Why short randomized trials often fail to find such harm

At first glance, this seems contradictory. Prospective studies following people for years have consistently found a link between artificial sweeteners and diabetes, while many randomized trials show no significant reduction in glucose or insulin levels. The authors of the new meta-analysis believe these results cannot be directly compared with each other.

Randomized trials are best at establishing causality because baseline differences between groups are randomly distributed. However, most trials of non-caloric sweeteners have been relatively short and have examined interim outcomes—body weight, glucose, or insulin levels—rather than the occurrence of diabetes 10-20 years later. Such experiments simply don't have the time horizon to detect a rare long-term outcome.

Observational studies have both advantages and disadvantages. They allow us to follow tens or hundreds of thousands of real-life people over many years and record thousands of cases of diabetes. However, the researcher doesn't control who drinks diet soda and why, meaning reverse causality and residual confounding remain.

Therefore, the current state of evidence cannot be summed up as either "sweeteners have been proven to cause diabetes" or "randomized trials have proven their complete long-term metabolic neutrality." A definitive answer requires significantly longer-term experimental approaches or carefully conducted studies simulating targeted clinical trials with large observational data—an approach proposed by the authors.

The results are consistent with the World Health Organization's position, but this does not mean a ban on all sweeteners.

Back in 2023, the World Health Organization issued a conditional recommendation against the use of non-sugar sweeteners as a long-term strategy for weight control or to reduce the risk of chronic non-communicable diseases. This was based on data showing no convincing long-term benefit for fat loss, and observational studies suggesting a possible association with type 2 diabetes, cardiovascular disease, and mortality.

A new meta-analysis moves in the same direction and updates precisely one of these questions—the incidence of diabetes. Its pooled HR of 1.31 is slightly higher than the estimate for artificially sweetened beverages reported in a previous systematic review by the World Health Organization, but directly comparing these figures is difficult due to the different sets of studies and inclusion criteria.

Another important point: the World Health Organization's recommendation is not a toxicological conclusion that approved sweeteners are dangerous at permitted doses. The organization specifically emphasizes that its document does not revise the acceptable daily levels of individual substances established by relevant expert committees. The issue at hand is whether using sugar-free sweetness as a long-term strategy for preventing obesity and chronic diseases.

This certainly doesn't mean it's better to return to a sugar-heavy diet. The World Health Organization has specifically stated that excess free sugars also pose a significant risk, and recommending non-sugar sweeteners doesn't mean sugar is becoming a healthier alternative. The overall goal is to gradually reduce the craving for excessively sweet tastes and choose predominantly unsweetened beverages and foods.

Question	What follows from the data
Are sweeteners better than sugar in calories?	Usually contain little or no energy
Has long-term diabetes prevention been proven?	No
Is there an observational link with diabetes?	Yes
Does WHO prove the toxicity of permitted doses?	No
Should sweeteners be replaced with sugar?	No
WHO's overall strategy	Reduce the overall sweetness of the diet

[7]

Aspartame, sucralose, saccharin and stevia cannot be combined into one substance.

The meta-analysis' title refers to "artificial sweeteners," but the actual data structure is much more complex. Five of the six included studies primarily classified people by their consumption of artificially sweetened beverages. Only the French NutriNet-Santé study provided a more detailed assessment of total sweetener intake. Therefore, the pooled figure largely reflects diet drink consumption patterns rather than a pharmacological assessment of a specific molecule.

Different sweeteners have completely different chemical structures, absorption, and metabolism. Some are almost entirely absorbed in the small intestine, others interact more with the gut microbiome, and some are plant-derived glycosides. The new meta-analysis did not have enough studies to reliably compare individual compounds.

Therefore, from an HR of 1.31, it is impossible to conclude that "sucralose increases the risk by 31%" or "aspartame increases the risk of diabetes by a third." Such statements would require separate meta-analyses with sufficiently precise accounting of the doses of the specific substance and the duration of consumption. The authors explicitly mention research on specific types of sweeteners as one of the next necessary avenues.

The same applies to stevia. Current World Health Organization guidelines include steviol glycosides in the category of non-sugar sweeteners when discussing long-term strategies for reducing dietary sweetness, but a current meta-analysis does not attribute the observed association with diabetes specifically to stevia.

How convincing is the result given the statistics?

A strength of the study is its very large total number of participants—over 700,000—and long-term follow-up, which averaged over ten years. Unlike cross-sectional studies, where beverage consumption and diabetes are measured simultaneously, all included cohorts first measured dietary intake and then recorded new cases of the disease, enhancing the temporal consistency of the association.

The quality of individual cohorts on the Newcastle-Ottawa scale ranged from 7 to 8 points out of 9, which the authors classified as high. All studies used adjusted estimates, and the results were retained after sequentially removing each individual study. Egger's test revealed no statistically significant evidence of small studies or obvious publication bias.

However, testing for publication bias with only six studies has limited statistical power: the absence of a significant test does not guarantee the absence of a problem. Even more serious are the high heterogeneity of I² and the wide prediction interval. Therefore, the data set simultaneously appears consistent in direction and uncertain in the precise magnitude of the effect.

It is this combination that explains the final GRADE rating of "low certainty." A large sample size alone does not transform an observational association into causal evidence. 721,000 people allow for highly accurate detection of a statistical association, but does not eliminate systematic differences between those who choose artificially sweetened foods and those who do not.

Key limitations of the new meta-analysis

The first limitation is the inclusion of only six studies. Although the overall population is large, the effects of each product type, gender, age, and specific sweetener cannot be studied reliably. Furthermore, the studies varied significantly in their definitions of exposure, which was a major source of high heterogeneity.

Second, diet was assessed by self-reporting. Even validated dietary questionnaires do not allow perfect recall of all beverage and food intakes over a long period. People may also change their habits after developing obesity, hypertension, or prediabetes. Therefore, a single baseline or periodically updated assessment does not always reflect true long-term exposure.

Third, residual confounding and possible reverse causality. The authors emphasize that people with metabolic problems are more likely to choose artificial sweeteners. Controlling for pre-diabetes and adjusting for body weight reduces this problem, but does not eliminate it entirely.

Finally, the meta-analysis protocol was not pre-registered in PROSPERO. Pre-registration typically helps demonstrate that the inclusion criteria and statistical plan were defined before the final results were obtained. The authors themselves cite the lack of registration as one of the limitations of their study.

Limitation	Why is it important?
Only 6 studies	Limits subgroup analyses
I² = 86.02%	Studies vary widely
Observational design	No proof of causality
Self-report of nutrition	Exposure errors may occur.
Reverse causation	People at high risk are more likely to choose diet foods
Residual confounding	It is impossible to statistically eliminate all differences.
Different definitions of "high consumption"	There is no single dose
No PROSPERO registration	Methodological limitation
GRADE	Low confidence

[8]

What does the result mean in practice?

The study does not provide grounds for claiming that someone who occasionally uses sweetener in their coffee should expect a 31% increase in their personal risk of diabetes. The exposures in the original studies varied, and most of the data pertained to regular consumption of artificially sweetened beverages. Therefore, applying a single pooled figure to a specific dose of a specific substance would be statistically inappropriate.

On the other hand, the results undermine the common notion that switching from regular sugary soda to unlimited diet drinks automatically translates into proven long-term diabetes prevention. A new meta-analysis shows that diabetes is more common among regular consumers of such products in prospective cohorts, even after numerous statistical adjustments.

However, returning to sugar-sweetened beverages is not recommended. The World Health Organization recommends reducing free sugar intake and simultaneously suggests not relying on non-sugar sweeteners as a long-term weight-control strategy. Water remains the most neutral beverage in terms of sweetness, and a long-term strategy may involve gradually reducing your overall addiction to very sweet tastes.

For people with diabetes or other medical conditions, the situation also requires an individualized approach. The new meta-analysis examined the occurrence of diabetes in people without the disease at baseline, not the effectiveness of a specific sweetener for glycemic control in an already diagnosed patient. Therefore, it does not provide grounds for independently changing a therapeutic diet without taking into account total calorie intake, body weight, carbohydrates, and the recommendations of a healthcare professional.

What the study showed—and what it didn't prove

Correct conclusion	Incorrect conclusion
High intake was associated with an HR of 1.31	"Sweeteners cause diabetes in 31% of people"
All six cohorts showed HR >1	Causality has been proven
The association remained after adjusting for body mass index.	Reverse causality is completely excluded
Most of the data relates to diet drinks	Every sweetener has the same effect.
The French cohort yielded an HR of 1.69	Each person increases the risk by 69%
Pacheco et al. obtained an HR of 1.11	The effect is the same in all populations
Mechanisms through the microbiome are possible	Microbiome has been shown to be the cause of the observed risk
WHO does not recommend NSS for long-term weight control	WHO has declared all sweeteners toxic.
The evidence is rated as low	The research is useless
More research is needed	The existing data can be completely ignored

Funding and conflict of interest

The study was conducted by researchers from Xuzhou Medical University and the Geriatric Hospital of Nanjing Medical University in China. The authors are Siyuan Lu, Yi Lu, Rui Xu, Wenfeng Hou, Ting Xu, and Wei Tang. The publication was peer-reviewed and accepted on July 20, 2026, after an initial submission on May 20 and a revision on July 8.

Funding for this study was provided by Jiangsu Province Key R&D Plan Social Development Project, grant BE2023774, with Wei Tang listed as the recipient.

The authors declared that the work was conducted without any commercial or financial relationships that could be construed as a potential conflict of interest. They also stated that generative artificial intelligence was not used in the preparation of the manuscript.

Thus, the article does not declare funding from sugar or artificial sweetener manufacturers. However, the absence of a commercial conflict does not change the key methodological limitations: all data remain observational, and the overall certainty of evidence, according to GRADE, is low.

Results

A new meta-analysis in Frontiers in Nutrition includes one of the largest prospective data sets on this topic to date—721,003 individuals, 38,445 cases of type 2 diabetes, and over 7.28 million person-years of follow-up. High, compared with low or no, artificial sweetener consumption was associated with a HR of 1.31, or approximately a 31% increase in the relative rate of diabetes incidence.

The result was robust: removing any of the six studies did not change the direction of the association, and the pooled HR remained between 1.23 and 1.37. All six cohorts also had point estimates greater than one. This makes the observed association noteworthy, especially since follow-up in individual studies extended for up to 20 years.

However, the effect size cannot be taken literally as a proven causal effect. The heterogeneity reached 86%, the prediction interval included one, the overall level of evidence was low, and people with obesity and prediabetes are more likely to switch to diet products. Therefore, the most accurate formulation is: long-term high consumption of artificially sweetened foods is a marker of increased diabetes risk, but it has not yet been established what portion of this association is actually due to the sweeteners themselves.

In practice, the study supports a more cautious strategy: not considering either regular sugar or large amounts of artificial sweeteners as an optimal long-term dietary staple. The World Health Organization also recommends reducing overall dietary dependence on intense sweetness, rather than simply constantly replacing sugar with non-sugar sweeteners.

News source

Lu S., Lu Y., Xu R., Hou W., Xu T., Tang W. The longitudinal association between artificial sweetener intake and the risk of type 2 diabetes among adults aged 18 years and older: a systematic review and meta-analysis. Frontiers in Nutrition. 2026;13:1886458. Published August 11, 2026.

This is a systematic review and meta-analysis of six prospective cohort studies with a total sample of 721,003 adults. The main pooled outcome was HR 1.31; 95% CI 1.16-1.49, with high heterogeneity I² 86.02% and low overall certainty of evidence according to GRADE.

DOI: 10.3389/fnut.2026.1886458.