Artificially Sweetened Beverages and Weight Change: Findings from Three Prospective Cohort Studies of U.S. Adults

Abstract

Background:

Whether habitually consuming artificially sweetened beverages (ASBs) or water in place of sugar-sweetened beverages (SSBs) is associated with weight changes is unclear.

Objective:

To evaluate the association of increasing ASBs or water as replacements for SSBs with changes in weight/body mass index (BMI).

Methods:

We analyzed data from the Nurses’ Health Study (NHS), NHS II, and Health Professionals Follow-up Study (HPFS), prospective U.S. cohorts. Females and males aged 26–65 years were followed for 24–32 years with biennial questionnaires updating medical, lifestyle, and anthropometric data. Multivariable-adjusted linear models estimated associations between changes in ASB intake, substitution of SSBs with ASBs or water, and four-year weight and BMI changes. Latent class growth models and multinomial logistic models estimated the odds of weight-trajectory membership based on beverage intake.

Results:

Among 143,409 participants (median follow-up 28 years), the average weight gain per four-year interval was 1.3 kg (5^th–95^th percentile: −6.3 to 9.1 kg). Pooled estimates indicated that each three-serving/week increase in ASB was associated with −0.18 kg weight change (95%CI: −0.26, −0.11) and −0.06 kg/m² BMI change (95%CI: −0.09, −0.03) over four-year intervals, with stronger inverse associations among adults with overweight/obesity and higher SSB intake. Replacing three SSB serving/week with ASB was associated with lower weight (−1.39 kg, 95%CI: −1.50, −1.28) and BMI gains (−0.50 kg/m² 95%CI: −0.54, −0.45). Similar estimates were observed when replacing SSB with water (−1.39 kg, 95%CI: −1.48, −1.30 and −0.49 kg/m², 95%CI: −0.53, −0.46). Replacing ASB with water was associated with modest reductions in weight (−0.11 kg, 95%CI: −0.19, −0.03) and BMI gains (−0.04 kg/m², 95%CI: −0.08, −0.01).

Conclusion:

Increases in ASB intake were inversely associated with weight/BMI gains over time, particularly among adults with overweight/obesity and higher SSB intake. Replacing SSBs with ASBs -or ideally water- was associated with decreases in weight/BMI changes in a dose-dependent manner.

INTRODUCTION

The role of artificially sweetened beverages (ASBs) as a replacement for sugar-sweetened beverages (SSBs) on long-term weight change remains controversial (1). Artificial sweeteners, also known as non-nutritive or non-caloric/low-calorie sweeteners, are used as food and beverage additives to provide sweetness without contributing calories or eliciting a glycemic response (2,3). In the United States (U.S.), the Food and Drug Administration has approved the use of saccharin, aspartame, acesulfame potassium, sucralose, neotame, advantame, steviol glycosides, and mogrosides (4). U.S. survey data indicate that SSB intake has declined modestly since 2000 but still contributes substantially to total energy and added sugars (5,6). In contrast, ASB intake has increased over the same period (7), suggesting a population-level shift from SSBs to ASBs (8).

Extensive research has examined whether replacing SSBs with ASBs might promote weight maintenance and downstream cardiometabolic improvements (9). Estimates from randomized clinical trials (RCTs) indicate that replacing SSBs [1000 (250–1750) mL/day] with ASBs [1000 (250–2000) mL/day] reduces weight (−1.06 kg; 95% CI, −1.71 to −0.41 kg), and body mass index (BMI) (−0.32 kg/m²; 95% CI, −0.58 to −0.07 kg/m²) in adults (10). In contrast, estimates from some observational studies have associated ASBs (in varying dosages) with increased risks of the conditions they aim to prevent, such as weight (0.06 kg; 95% CI, 0.05 to 0.07 kg) and BMI gain (0.05 kg/m²; 95% CI, 0.03 to 0.07 kg/m²) (11).

Debate persists over the long-term effects of ASBs on body weight, largely due to differences in study design and analytic methods. In RCTs, weight change is usually compared between ASB consumers and those consuming alternative beverages or sweetener formulations, which often entitles caloric displacement in the ASB group. Conversely, most observational studies assess weight changes relative to ASB intake without specifying a comparison beverage or accounting for caloric displacement. Observational studies also often rely on baseline or prevalent ASB intake, implicitly assuming that intake remains stable during follow-up. This approach, often referred to as naïve (1), is susceptible to measurement error in relation to future changes in weight (12), prone to reverse causation, and does not allow for analyses of within-person changes in ASB intake over time. Additionally, individuals consuming ASBs at baseline may have already reached a new steady-state in weight after previous changes in SSB and ASB intake not accounted for in the analysis. As in RCTs, prospective cohort studies should pair changes in ASB intake with concurrent changes in weight and BMI using repeated measures to maximize consistent, robust, and biologically plausible associations (12–14).

Due to the conflicting evidence base, the World Health Organization (WHO) has issued a conditional recommendation against using artificial sweeteners for weight control (15). The recommendation also highlighted the importance of future studies that include multiple, sequential exposure assessments, account for varying consumption patterns, and reduce susceptibility to reverse causation.

Thus, our study aimed to evaluate the association between long-term changes in ASB intake and subsequent changes in weight/BMI, to estimate how replacing SSBs with ASBs or water is associated with changes in weight/BMI, and to assess whether these estimated changes follow a dose-dependent pattern among U.S. females and males. We hypothesized that there is no association between habitual ASB intake and long-term weight or BMI change. We further expected that substituting SSBs with ASBs or water would be associated with lower weight and BMI gains, with water showing stronger inverse associations.

METHODS

Study design and population

We used data from three large ongoing prospective cohort studies: Nurses’ Health Study (NHS), Nurses’ Health Study II (NHS II), and Health Professional Follow-up Study (HPFS). The NHS initially enrolled 121,701 females (aged 30–55 years) in 1976. The NHS II enrolled 116,429 females (aged 25–42 years) in 1989, and the HPFS enrolled 51,529 males (aged 40–75 years) in 1986. Detailed cohort descriptions have been published previously (16).

At baseline, participants completed self-administered questionnaires covering personal characteristics, diet, medical history, lifestyle, and other health-related indicators. These assessments are updated every 2 to 4 years (17,18). For the current analysis, the baseline was defined as the year when detailed dietary assessments were first administered: 1986 for the NHS, 1991 for the NHS II, and 1986 for the HPFS; the most recent dietary follow-up years were 2010 for the NHS, 2019 for the NHS II, and 2018 for the HPFS.

Participants were excluded at baseline if they returned only the baseline questionnaire, or reported a history of any major disease, including cardiovascular disease (heart disease, angina, coronary artery bypass graft, stroke, pulmonary embolism), diabetes, cancer, respiratory diseases (emphysema, active tuberculosis, chronic bronchitis), chronic kidney disease, neurodegenerative disorders (Amyotrophic Lateral Sclerosis, Parkinson’s disease, multiple sclerosis, Alzheimer’s disease), gastric conditions (ulcerative colitis, gastric or duodenal ulcer, gastric surgery or intestinal bypass), or systemic lupus erythematosus. Additional baseline exclusions included having more than 70 missing items on the questionnaire, implausible energy intake (<600 or >3500 kcal/day for women; <800 or >4200 kcal/day for men), missing information for body weight or beverage intake (ASBs, SSBs, or water), and being age 65 years or older to prevent imprecision and reverse causation bias due to age-related lean mass loss.

During follow-up, participants were excluded at the time of death, loss to follow-up, or upon reaching age 65 years. We additionally excluded participants starting six years before any diagnosis of diabetes, cancer, respiratory diseases, chronic kidney disease, neurodegenerative disorders, gastric conditions, or systemic lupus erythematosus. Participants were also excluded for missing outcome data (weight and BMI change in subsequent cycles). Pregnant or lactating females were excluded from the analyses for that cycle and at least one year postpartum due to acute changes in body weight. In sensitivity analyses, we excluded participants using GLP-1 receptor agonists, including, exenatide, liraglutide, or dulaglutide (1.4% of the sample).

The final analyses included 51,805 females from the NHS, 65,425 females from the NHS II, and 26,179 males from the HPFS, resulting in a total sample size of 143,409 participants. The flowchart outlining participant inclusion and exclusions at baseline and during follow-up is presented in the Supplementary material (Figure S1). Participants signed informed consent and the boards of the Brigham and Women’s Hospital, and the Harvard T.H. Chan School of Public Health approved the study.

Dietary assessments

Dietary intake was assessed using validated and reproducible self-administered semiquantitative food-frequency questionnaires (SFFQ) to determine the usual intake frequency of specified portion sizes over the previous year (>130 items). Information on each item and its validation has been published previously (17,18). Validation studies revealed that beverages had the highest reproducibility and corrected validity, displaying a corrected correlation coefficient for ASB intake reported between SFFQ and two 1-week diet records of 0.86 (95%CI: 0.83, 0.88) in females and 0.83 (95%CI: 0.80, 0.86) in males (19). The estimated intakes of food groups (fruits, vegetables, whole grains, refined grains, potatoes [including boiled or mashed potatoes and French fries], potato chips, whole-fat dairy products, low-fat dairy products, sweets, desserts, processed meats, unprocessed red meats, and fried foods), beverages (e.g., coffee, tea, alcohol, and commercial processed drinks) were evaluated at baseline and every four years thereafter. Nutrients from food and beverage intake were subsequently derived for each assessment using nutrient composition databases, drawing from the U.S. Department of Agriculture Nutrient Database for Standard Reference, the Food and Nutrient Database for Dietary Studies, manufacturers’ data, and updated Harvard biochemical information (16,20).

For beverages on the SFFQ, participants reported their usual intake of a glass, bottle, or can serving, based on nine frequency categories (ranging from “never” to “≥ six times/day”). ASB intake was estimated by aggregating the consumption frequencies of diet sodas (with caffeine, without caffeine). SSB intake was determined by combining the frequencies of regular soda (with and without caffeine) and other sugar-sweetened soft drinks. Water intake was specified as the frequency of consuming plain water. Changes in each beverage intake were calculated as the differences between sequential evaluations for diet soda, regular soda, and water over four-year intervals during follow-up (NHS: 1986–2010; NHS II: 1991–2019; HPFS: 1986–2018).

Body weight and BMI changes

Height and weight were self-reported at enrollment. Weight was subsequently updated with each biennial follow-up questionnaire. Four-year changes in weight were calculated as the difference between sequential assessments during follow-up, spanning 24 years in the NHS (1986–2010), 28 years in the NHS II (1991–2019), and 32 years in the HPFS (1986–2018). BMI was calculated for each assessment interval using baseline height and the corresponding weight from each follow-up [BMI = weight (kg) / height (m²)]. Previous validation studies found a Spearman correlation coefficient of 0.96 when comparing self-reported weight measurements to those assessed in-person by staff in these cohorts (21).

Covariates

Questionnaires for all cohorts collected information on lifestyle factors, including race, age, physical activity, alcohol use, sleep duration, coffee and tea consumption, cigarette smoking, hours spent sitting and watching television, medication use, dieting behaviors, socioeconomic status at the neighborhood level, and other important covariates. These factors were assessed at baseline and updated every 2–4 y. Further details on the data collection methods for these lifestyle factors can be found elsewhere (16).

Missing covariate data was handled with a missing indicator for categorical variables [missing for categories for sleep duration (16.3%), time spent watching television (8.5%), and smoking status (0.9%)] and last value carried forward for continuous variables [missing for physical activity (2.4%), dietary information (9.5%), and socioeconomic status at the neighborhood level (0.2%)].

Statistical Analysis

We calculated changes in ASBs and changes in body weight and BMI as the difference between baseline and follow-up for each four-year interval.

We fitted multivariable generalized linear models with an unstructured correlation matrix and robust variance estimation to account for the use of within-person repeated measures. The primary models evaluated the association between changes in ASB intake, expressed per three-serving-per-week increase, and subsequent changes in weight (kg) and BMI (kg/m²) over four-year intervals to reflect intake patterns of potential clinical relevance. To complement these estimates and in alignment with prior literature and dietary-substitution frameworks (22), we additionally modeled changes in beverage intake expressed per one-serving-per-day increase.

Both ASB change and the changes in weight and BMI were modeled as continuous variables in the primary analysis. To mitigate the influence of extreme values, exposure and outcome outliers were truncated at the 0.5^th and 99.5^th centiles.

We used theoretical substitution models to evaluate how replacing SSBs with ASBs relates to weight and BMI changes. To estimate these substitutions, we simultaneously modeled continuous four-year changes (per one serving/day and per three servings/week) in the intake of ASB and SSB as independent variables in the multivariable linear model. The difference in the β coefficients for ASBs and SSBs was used to calculate the point estimates for the substitution estimate. The 95 % CI was estimated using the variance and covariance matrices (23). This approach acknowledges that ASB and SSB intake are not mutually exclusive. We then conducted similar models for the concurrent changes in SSB and water with weight and BMI changes.

We also evaluated changes in weight and BMI associated with categorical changes in beverage intake. ASB intakes were classified into stable intake of 0.05 to 0.5 servings/day, four-year increments/decrements of 0.5 to 1 serving/day, and four-year increments/decrements of more than 1 serving/day. Theoretical categorical substitution models were used to assess weight/BMI changes associated with replacing SSBs with ASBs or water and replacing ASBs with water using the same change in intake intervals.

Effect modification by BMI category (kg/m² <25, 25–30, and ≥30) at the start of each four-year interval was evaluated. It should be noted that the association between changes in ASB intake and changes in weight/BMI might reflect the average estimate from participants who either (i) actively reduced SSB intake or (ii) simply added ASBs to their habitual diet without reducing SSBs. To address these contrasting dietary scenarios, we also evaluated effect modification by concurrent four-year changes in SSB intake.

To account for heterogeneity in long-term weight change, we identified weight-change trajectories across the three combined cohorts using latent class growth models. Age served as the underlying time scale. This specific analysis was restricted to 24 years to harmonize the follow-up period across all cohorts. Using a two-step, Bayesian Information Criterion strategy, we first selected the optimal number of trajectory groups and then the functional form (polynomial order) for time within each group. We evaluated models with up to five trajectory groups and up to cubic time terms. The four-group model with linear time terms provided the best balance of fit and interpretability. Participants were assigned to the trajectory for which they had the highest posterior probability. We then examined how 4-year changes in beverage intake (per 3-servings-per-week increment) related to weight-change trajectory membership using multinomial logistic regression models with a generalized logit link, considering trajectory 1 (weight maintenance) as the reference.

All models were adjusted for age (continuous; years), questionnaire cycle, waist circumference in 1986 (NHS and HPFS) and 1991 (NHS II), and BMI at the beginning of each four-year period (continuous; kg/m²). Models also adjusted for baseline intake and four-year changes in individual dietary components (continuous; servings/day), including items typically classified as ultra-processed under the NOVA system (24), to reduce residual confounding related to the high consumption of ultra-processed foods in the U.S. diet (25). A full list of dietary items is provided in the Supplementary material, which includes fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets and desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Models further adjusted for the following lifestyle variables at the beginning of each four-year interval as well as for their changes over the subsequent 4 years: physical activity (continuous; Met-hours/week), sleep duration (categorical; (≤6, 7, 8, >8 hours/day), time spent sitting and watching television [baseline only in NHS and NHS II (0–1, 2–5, 6–20, 21–40, >40 hours/week) and baseline and four-year change in HPFS (continuous)], smoking status (categorical; stayed never, stayed former, stayed current, former to current, never to current, current to former), alcohol use (continuous; grams/day). Models were also adjusted for dieting and weight-loss behaviors at the beginning of each four-year interval (categorical variables [yes/no]; following a low-calorie, low-fat, or low-carbohydrate diet; skipping meals or fasting; increasing exercise for weight loss purposes; experiencing unintentional weight loss; consuming diet pills; following a commercial weight loss program; resuming or increasing smoking or alcohol intake for weight loss purposes; consuming commercial dietary supplements; prescription of weight loss drugs; consuming fenfluramine, phentermine, or dexfenfluramine; pursuing intentional weight loss in the past two years; and limiting dietary portion sizes). Models further included adjustments for socioeconomic status (continuous, composed score) at the neighborhood level at the beginning of each four-year interval. Models assessing only changes in ASB intake additionally adjusted for baseline ASB and for baseline and changes in SSB and water intake. Substitution models were adjusted for the baseline intake of both beverages involved in the substitution and for baseline and changes in the third beverage during follow-up.

The primary analysis was conducted without model adjustment for total energy intake because it may lie on the causal pathway between ASB intake and weight changes (12). ASBs may lower energy intake by displacing SSBs (10) or, conversely, may contribute to a compensatory increase in energy intake (26). In either case, energy intake functions as a mediator of the association between changes in ASB intake and weight changes. To quantify this, we estimated the proportion of the ASB-weight association mediated by energy intake from SSBs using mediation analysis for generalized linear models, with confidence intervals obtained via direct counterfactual imputation and 1,000 bootstrap iterations (27). Nonetheless, we also fit supplemental models adjusted for total energy as a mechanistic sensitivity analysis to capture potential compensatory changes in energy intake.

Analyses of each cohort were conducted separately using SAS version 9.4 (SAS Institute, Cary, NC) (28). Subsequently, cohorts were pooled through fixed-effect inverse-variance meta-analysis (29). Pooled analysis combined estimates from the females’ cohorts (NHS and NHS II) and estimates from the combined population (NHS, NHS II, and HPFS). The findings were visually represented in figures generated using STATA version 18 (StataCorp LLC, TX) (30), and further refined with BioRender (31). A two-sided alpha level of <0.05 was used to determine statistical significance.

RESULTS

Descriptive characteristics

Females enrolled in the NHS had a four-year baseline average age of 55.3 ± 6.5 years and presented a mean weight of 69.5 ± 13.8 kg and a BMI of 25.8 ± 4.8 kg/m² (Table 1). Females from the NHS II cohort had an average baseline age of 43.9 ± 7.0 years and a mean weight of 70.1 ± 16.1 kg and BMI of 25.8 ± 5.6 kg/m². Males from the HPFS had an average age of 54.4 ± 7.1 years at baseline, with an average weight of 83.1 ± 12.3 kg and BMI of 25.9 ± 3.4 kg/m².

Table 1.

Baseline characteristics and four-year changes among 143,409 females and males enrolled in the Nurses’ Health Study I & II and Health Professionals Follow-up Study, followed for 24 years (NHS; n=51,805), 28 years (NHS II; n=65,425), and 32 years (HPFS; n=26,179).

	NHS (n = 51,805)		NHS II (n = 65,425)		HPFS (n = 26,179)
	Four-year mean intake (mean ± SD)	Four-year mean change in intake (5th — 95th percentile)	Four-year mean intake (mean ± SD)	Four-year mean change in intake (5th — 95th percentile)	Four-year mean intake (mean ± SD)	Four-year mean change in intake (5th — 95th percentile)
General characteristics
Age (years)	55.3 ± 6.5	—	43.9 ± 7.0	—	54.4 ± 7.1	—
Anthropometric assessment
BMI (kg/m2)	25.8 ± 4.8	0.4 (−2.4 — 3.4)	25.8 ± 5.6	0.7 (−2.6 — 4.3)	25.9 ± 3.4	0.3 (−1.7 — 2.2)
Weight (kg)	69.5 ± 13.8	1.2 (−6.8 — 9.1)	70.1 ± 16.1	2.0 (−6.8 — 11.3)	83.1 ± 12.3	0.8 (−5.4 — 6.8)
Waist Circumference (in)	21.0 ± 14.3	NA†	NA†	NA†	25.7 ± 17.2	NA†
Dietary factors (servings/day)
Energy intake (kcal)	1769 ± 521	−6 (−710 — 694)	1810 ± 555	−9 (−772 — 754)	1995 ± 617	−3 (−779 — 776)
SSBs	0.3 ± 0.5	0.0 (−0.6 — 0.6)	0.4 ± 0.8	0.0 (−0.9 — 0.7)	0.4 ± 0.6	0.0 (−0.7 — 0.6)
ASBs	0.6 ± 0.9	0.0 (−1.0 — 1.0)	0.9 ± 1.3	−0.1 (−1.6 — 1.4)	0.6 ± 1.0	0.0 (−0.8 — 0.7)
Water	2.9 ± 1.9	0.0 (−5.2 — 3.9)	2.8 ± 2.0	0.0 (−3.5 — 3.5)	2.4 ± 1.8	0.0 (−2.0 — 2.0)
Coffee	2.2 ± 1.7	0.1 (−5.4 — 5.8)	1.5 ± 1.6	0.0 (−2.0 — 1.8)	1.9 ± 1.8	0.0 (−2.0 — 1.6)
Fruits	1.6 ± 1.2	0.0 (−1.6 — 1.5)	1.3 ± 1.0	0.1 (−1.3 — 1.5)	1.6 ± 1.2	0.1 (−1.4 — 1.6)
Vegetables	3.3 ± 1.9	0.0 (−2.4 — 2.6)	3.9 ± 2.2	0.1 (−2.6 — 2.9)	4.0 ± 2.0	0.1 (−2.5 — 2.7)
Nuts	0.2 ± 0.4	0.0 (−0.4 — 0.6)	0.3 ± 0.4	0.1 (−0.4 — 0.8)	0.4 ± 0.6	0.0 (−0.6 — 0.8)
Whole grains	1.4 ± 1.1	0.1 (−1.7 — 1.9)	1.3 ± 1.0	0.0 (−1.8 — 1.6)	1.6 ± 1.4	0.0 (−1.9 — 2.0)
Refined grains	1.6 ± 1.2	0.0 (−1.9 — 1.8)	1.6 ± 1.1	−0.1 (−1.9 — 1.6)	1.5 ± 1.2	0.0 (−1.9 — 1.7)
Potatoes	0.3 ± 0.3	0.0 (−0.4 — 0.4)	0.3 ± 0.2	0.0 (−0.4 — 0.3)	0.3 ± 0.3	0.0 (−0.4 — 0.4)
Potato chips	0.1 ± 0.2	0.0 (−0.3 — 0.3)	0.1 ± 0.2	0.0 (−0.4 — 0.3)	0.2 ± 0.2	0.0 (−0.4 — 0.3)
Dairy	2.1 ± 1.4	0.0 (−2.1 — 1.9)	2.2 ± 1.4	0.0 (−2.2 — 2.2)	1.9 ± 1.4	0.0 (−1.9 — 1.8)
Other sweets and desserts	1.1 ± 1.2	0.0 (−1.5 — 1.6)	1.1 ± 1.0	−0.1 (−1.4 — 1.2)	1.4 ± 1.3	0.0 (−1.6 — 1.6)
Processed meats	0.2 ± 0.3	0.0 (−0.4 — 0.4)	0.2 ± 0.3	0.0 (−0.4 — 0.4)	0.3 ± 0.4	0.0 (−0.5 — 0.4)
Unprocessed red meats	0.7 ± 0.5	0.0 (−0.8 — 0.6)	0.7 ± 0.5	0.0 (−0.7 — 0.7)	0.8 ± 0.6	0.0 (−0.8 — 0.7)
Fried foods	0.1 ± 0.1	0.0 (−0.1 — 0.1)	0.1 ± 0.1	0.0 (−0.1 — 0.1)	0.1 ± 0.1	0.0 (−0.3 — 0.1)
Lifestyle indicators
Alcohol (g/day)	5.8 ± 9.9	−0.1 (−8.0 — 7.7)	4.4 ± 8.0	0.7 (−4.9 — 8.5)	11.5 ± 15.0	0.3 (−11.9 — 13.7)
Physical activity (MET-hr/wk)	18.0 ± 22.9	0.6 (−27.5 — 30.0)	20.4 ± 25.6	1.0 (−30.1 — 34.2)	30.3 ± 30.3	4.0 (−39.4 — 52.2)
Sleep duration (hr/day)	7.0 ± 1.0	NA†	7.0 ± 0.9	NA†	7.0 ± 0.8	NA†
Smoking status (cigarettes/day)	1.7 ± 0.7	3.1 (1.0 — 6.0)	1.4 ± 0.6	2.3 (1.0 — 6.0)	1.8 ± 1.1	2.7 (1.0 — 5.0)
Time spent watching television (hr/wk)	11.7 ± 10.6	2.8 (1.0 — 4.0)	3.8 ± 1.2	2.5 (1.0 — 4.0)	4.1 ± 2.4	−0.1 (−10.5 — 10.5)
Socioeconomic status (composed score)	0.2 ± 3.8	0.0 (−2.4 — 1.7)	0.0 ± 3.8	−0.01 (−2.8—2.9)	0.0 ± 3.6	0.0 (−2.0 — 2.4)

Abbreviations: NHS = Nurses’ Health Study; BMI = body mass index; hr = hour; wk = week.

^†Insufficient repeated assessments for obtaining a change estimate.

The average weight gain over the four-year periods was 1.3 kg (5th to 95th percentile: −6.3 to 9.1 kg), translating to an estimated gain of 7.8 kg over 24 years. The cohort-specific average weight gain over four years was 1.2 kg (−6.8 to 9.1 kg) in the NHS (7.2 kg in 24 years), 2.0 kg (−6.8 to 11.3 kg) in the NHS II (14.0 kg in 28 years), and 0.8 kg (−5.4 to 6.8 kg) in the HPFS (6.4 kg in 32 years).

Among NHS participants, the four-year average ASB intake was 0.6 ± 0.9 servings/day, with an average four-year change of 0.0 (5^th to 95^th percentile: −1.0 to 1.0) servings/day. For NHS II participants, the four-year average intake was 0.9 ± 1.3 servings/day, with a four-year average change of −0.1 (5^th to 95^th percentile: −1.6 to 1.4) servings/day. In the HPFS cohort, the four-year average intake was 0.6 ± 1.0 servings/day, with a corresponding four-year change of 0.0 (5^th to 95^th percentile: −0.8 to 0.7) servings/day.

Compared to ASBs, SSB intake was less frequent. Among NHS female participants, the four-year mean intake of SSBs was 0.3 ± 0.5 servings/day, with a four-year average change of 0.0 (5^th to 95^th percentile: −0.6 to 0.6) servings/day. NHS II participants had a four-year average intake of 0.4 ± 0.8 servings/day, with a four-year average change of 0.0 (5^th to 95^th percentile: −0.9 to 0.7) servings/day. Male participants showed a four-year average intake of 0.4 ± 0.6 servings/day, with a corresponding four-year average change of 0.0 (5^th to 95^th percentile: −0.7 to 0.6) servings/day.

Table 1 provides an overview of four-year mean dietary intakes and corresponding four-year changes within each cohort, coupled with anthropometric and lifestyle data.

Association of a three-serving-per-week increase in ASBs with weight and BMI changes.

Over four-year intervals, a three-serving-per-week increase in ASB intake was associated with a net weight change of −0.17 kg (95% CI: −0.25, −0.09 kg) among females and −0.26 kg (95% CI: −0.45, −0.06 kg) in males (Figure 1). The combined estimate for both females and males was −0.18 kg (95% CI: −0.26, −0.11 kg) for each three-serving increase of ASBs. Results were consistent for BMI. Females exhibited a −0.06 change in kg/m² (95% CI: −0.09, −0.02 kg/m²), males a −0.09 change in kg/m² (−0.16, −0.02 kg/m²), and the estimate for the combined population resulted in −0.06 kg/m² (−0.09, −0.03 kg/m²). In mediation analysis we found that this association was partially mediated by reducing energy intake from SSBs (pooled estimate: 20.6% [95% CI: 15.0, 26.3%] for weight change and 19.3% [95% CI: 13.1, 25.4%] for BMI change).

Figure 1.

Association between a three-serving-per-week increase in artificially sweetened beverages and three-serving-per-week beverage replacements with changes in weight (kg) and BMI (kg/m²) over four-year intervals in three U.S. cohorts.

Models were multivariable linear regressions with robust variance and were adjusted for age, questionnaire cycle, waist circumference (1986 for NHS and HPFS; 1991 for NHS II), and BMI at the start of each 4-year interval. Dietary adjustments included baseline intake and 4-year changes in fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets/desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Lifestyle adjustments included baseline and 4-year changes in physical activity, sleep duration, smoking status, and alcohol intake, as well as sitting/television time (baseline only for NHS and NHS II; baseline and changes for HPFS). Models also included weight-control behaviors at each interval: low-calorie, low-fat, or low-carbohydrate diet; meal skipping/fasting; exercising for weight loss; unintentional weight loss; diet pill use; commercial weight-loss programs; smoking or alcohol changes for weight loss; dietary supplements; prescription weight-loss drugs; fenfluramine, phentermine, or dexfenfluramine use; intentional weight loss in the prior 2 years; and portion-size restriction. Models also included neighborhood socioeconomic status at interval start. Models assessing changes in ASB intake further adjusted for baseline ASB and for baseline and changes in SSB and water intake. Substitution models were adjusted for the baseline intake of both beverages involved in the substitution, as well as for baseline and changes in the third beverage during follow-up.

Cohort-specific estimates from the NHS (n = 51,805; 1986–2010) NHS II (n = 65,425; 1991–2019), and the HPFS (n = 26,179; 1986–2018) were meta-analyzed using fixed-effects, inverse-variance models. Female-only analyses included NHS and NHS II, and male-only analyses included HPFS.

Abbreviations: ASB = artificially sweetened beverages; SSB = sugar-sweetened beverages; s/d = serving/day; NHS = Nurses’ Health Study; NHS II = Nurses’ Health Study II; HPFS = Health Professionals Follow-up Study.

Substituting three-serving-per-week of SSBs with ASBs and subsequent weight and BMI changes

Replacing three servings per week of SSBs with the equivalent amount of ASBs was associated with a −1.55 kg (95% CI: −1.68, −1.43 kg) change in weight in females and −0.82 kg (95% CI: −1.06, −0.59 kg) in males. The combined estimate for this substitution was −1.39 kg (95% CI: −1.50, −1.28 kg). The BMI change observed was −0.59 kg/m² (95% CI: −0.64, −0.53 kg/m²) for females and −0.26 kg/m² (95% CI: −0.35, −0.18 kg/m²) for males, with a corresponding change for the combined population of −0.50 kg/m² (95% CI: −0.54, −0.45 kg/m²).

Substituting three-serving-per-week of SSBs with water and subsequent weight and BMI changes

Changes in weight and BMI were observed when replacing three servings per week of SSBs with the same amount of water. Estimates for this substitution were −1.70 kg (95% CI: −1.81, −1.59 kg) and −0.65 kg/m² (95% CI: −0.70, −0.61 kg/m²) changes for females, and −0.71 kg (95% CI: −0.87, −0.55 kg) and −0.22 (95% CI: −0.28, −0.16 kg/m²) changes in males. Combined estimates for both females and males resulted in −1.39 kg (95% CI: −1.48, −1.30 kg) and −0.49 kg/m² (95% CI: −0.53, −0.46 kg/m²).

Substituting three-serving-per-week of ASBs with water and subsequent weight and BMI changes

Replacing three servings per week of ASB with an equal amount of water was associated with a modest weight change of −0.14 kg (95% CI: −0.23, −0.06 kg) and BMI change of −0.06 kg/m² (95% CI: −0.10, −0.03 kg/m²) in females. No association was displayed in males, with estimates for changes in weight of 0.10 kg (95% CI: −0.11, 0.30 kg) and BMI of 0.04 kg/m² (95% CI: −0.03, 0.11 kg/m²). Combined estimates resulted in a small weight change of −0.11 kg (95% CI: −0.19, −0.03 kg) and BMI change of −0.04 kg/m² (95% CI: −0.08, −0.01 kg/m²), respectively.

Figure 1 summarizes the estimates for the association between four-year changes in ASB intake and weight and BMI changes, along with estimates for substituted beverages (as per three servings/week). Cohort-specific and pooled estimates are presented in Table S1 and Table S2 in the Supplementary material.

Dose-response analyses: Weight and BMI changes relative to categorical changes in beverage intake

Stepwise increases in ASB intake were associated with progressively less weight and BMI gain. This dose-dependent association was particularly pronounced among participants who increased their intake by one or more daily servings. In categorical substitution models, we consistently observed dose-dependent associations, where greater replacements of SSBs with ASBs were associated with progressively less weight and BMI gain over time. Similar dose-dependent associations were observed with the gradual replacement of SSBs with water and ASBs with water. Figure 2 (panel A and panel B) summarize the estimates for dose-response associations. Cohort-specific estimates are available in Table S3, and Table S4 in the Supplementary material.

Figure 2.

Association between categorized changes in beverage intake and changes in weight (kg; Panel A) and BMI (kg/m²; Panel B) over four-year intervals in three U.S. cohorts.

Beverage-intake changes were categorized as small (0.05–0.5 servings/day), moderate (0.5–1 serving/day), or large (>1 serving/day) increases or decreases, with “no change” serving as the reference category.

Models were multivariable linear regressions with robust variance and were adjusted for age, questionnaire cycle, waist circumference (1986 for NHS and HPFS; 1991 for NHS II), and BMI at the start of each 4-year interval. Dietary adjustments included baseline intake and 4-year changes in fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets/desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Lifestyle adjustments included baseline and 4-year changes in physical activity, sleep duration, smoking status, and alcohol intake, as well as sitting/television time (baseline only for NHS and NHS II; baseline and changes for HPFS). Models also included weight-control behaviors at each interval: low-calorie, low-fat, or low-carbohydrate diet; meal skipping/fasting; exercising for weight loss; unintentional weight loss; diet pill use; commercial weight-loss programs; smoking or alcohol changes for weight loss; dietary supplements; prescription weight-loss drugs; fenfluramine, phentermine, or dexfenfluramine use; intentional weight loss in the prior 2 years; and portion-size restriction. Models also included neighborhood socioeconomic status at interval start. Models assessing changes in ASB intake further adjusted for baseline ASB and for baseline and changes in SSB and water intake. Substitution models were adjusted for the baseline intake of both beverages involved in the substitution, as well as for baseline and changes in the third beverage during follow-up.

Cohort-specific estimates from the NHS (n = 51,805; 1986–2010) NHS II (n = 65,425; 1991–2019), and the HPFS (n = 26,179; 1986–2018) were meta-analyzed using fixed-effects, inverse-variance models. Female-only analyses included NHS and NHS II, and male-only analyses included HPFS.

Abbreviations: ASB = artificially sweetened beverages; SSB = sugar-sweetened beverages; s/d = serving/day; NHS = Nurses’ Health Study; NHS II = Nurses’ Health Study II; HPFS = Health Professionals Follow-up Study.

Effect modification by BMI category at the start of each four-year interval

In analyses stratified by BMI at the start of each four-year interval, participants with a healthy BMI who increased their ASB intake by three servings per week displayed no change in weight [0.03 kg (95% CI: −0.05 to 0.12 kg)] or BMI [0.02 kg/m² (95% CI: −0.01 to 0.06 kg/m²)]. However, the association became more pronounced and linear among participants with overweight and obesity. For participants with overweight (BMI = 25 to <30 kg/m²), a three-serving-per-week increase in ASB intake was associated with an estimated weight change of −0.20 kg (95% CI: −0.32 to −0.09 kg) and a BMI change of −0.08 kg/m² (95% CI: −0.12 to −0.04 kg/m²). Participants with obesity (BMI ≥ 30 kg/m²) exhibited an estimated weight change of −0.47 kg (95% CI: −0.65 to −0.29 kg) and a BMI change of −0.16 kg/m² (95% CI: −0.24 to −0.09 kg/m²).

Figure 3 summarizes the estimates for the association between changes in ASB (as per three servings per week) and weight/BMI changes, stratified by baseline BMI classification. Cohort-specific and pooled estimates are available in Table S1 and Table S2 in the Supplementary material.

Figure 3.

Association between a three-serving-per-week increase in artificially sweetened beverages and changes in weight (kg) and BMI (kg/m²) over four-year intervals, stratified by baseline BMI classification in three U.S. cohorts.

Models were multivariable linear regressions with robust variance stratified by baseline BMI categories (healthy weight, overweight, obesity). Models were adjusted for age, questionnaire cycle, waist circumference (1986 for NHS and HPFS; 1991 for NHS II), and BMI at the start of each 4-year interval. Dietary adjustments included baseline intake and 4-year changes in fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets/desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Lifestyle adjustments included baseline and 4-year changes in physical activity, sleep duration, smoking status, and alcohol intake, as well as sitting/television time (baseline only for NHS and NHS II; baseline and changes for HPFS). Models also included weight-control behaviors at each interval: low-calorie, low-fat, or low-carbohydrate diet; meal skipping/fasting; exercising for weight loss; unintentional weight loss; diet pill use; commercial weight-loss programs; smoking or alcohol changes for weight loss; dietary supplements; prescription weight-loss drugs; fenfluramine, phentermine, or dexfenfluramine use; intentional weight loss in the prior 2 years; and portion-size restriction. Models also included neighborhood socioeconomic status at interval start. Models assessing changes in ASB intake further adjusted for baseline ASB and for baseline and changes in SSB and water intake. Substitution models were adjusted for the baseline intake of both beverages involved in the substitution, as well as for baseline and changes in the third beverage during follow-up.

Cohort-specific estimates from the NHS (n = 51,805; 1986–2010) NHS II (n = 65,425; 1991–2019), and the HPFS (n = 26,179; 1986–2018) were meta-analyzed using fixed-effects, inverse-variance models. Female-only analyses included NHS and NHS II, and male-only analyses included HPFS.

Abbreviations: ASB = artificially sweetened beverages; SSB = sugar-sweetened beverages; s/d = serving/day; NHS = Nurses’ Health Study; NHS II = Nurses’ Health Study II; HPFS = Health Professionals Follow-up Study.

Effect modification by four-year changes in SSB intake

In analyses stratified by categorical changes of SSB and ASB intake, we observed that participants who reduced their SSB intake experienced less weight and BMI gain over the four-year intervals, irrespective of changes in their ASB intake (whether it increased, decreased, or remained unchanged). Conversely, participants who increased their SSB intake displayed more weight and BMI gain over the same intervals, irrespective of changes in ASB intake.

Detailed estimates for each of these scenarios are illustrated in Figure 4, with cohort-specific estimates provided in Table S5 and Table S6 in the Supplementary material.

Figure 4.

Association between stratified changes in beverage intake and changes in weight (kg) and BMI (kg/m²) over four-year intervals in three U.S. cohorts.

Models were multivariable linear regressions with robust variance, stratified by 4-year categorical changes in SSB and ASB intake (increase, decrease, no change). The ‘no change in beverage intake’ category served as the reference group. Models were adjusted for age, questionnaire cycle, waist circumference (1986 for NHS and HPFS; 1991 for NHS II), and BMI at the start of each 4-year interval. Dietary adjustments included baseline intake and 4-year changes in fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets/desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Lifestyle adjustments included baseline and 4-year changes in physical activity, sleep duration, smoking status, and alcohol intake, as well as sitting/television time (baseline only for NHS and NHS II; baseline and changes for HPFS). Models also included weight-control behaviors at each interval: low-calorie, low-fat, or low-carbohydrate diet; meal skipping/fasting; exercising for weight loss; unintentional weight loss; diet pill use; commercial weight-loss programs; smoking or alcohol changes for weight loss; dietary supplements; prescription weight-loss drugs; fenfluramine, phentermine, or dexfenfluramine use; intentional weight loss in the prior 2 years; and portion-size restriction. Models also included neighborhood socioeconomic status at interval start. Models assessing changes in ASB intake further adjusted for baseline ASB and for baseline and changes in SSB and water intake. Substitution models were adjusted for the baseline intake of both beverages involved in the substitution, as well as for baseline and changes in the third beverage during follow-up.

Cohort-specific estimates from the NHS (n = 51,805; 1986–2010) NHS II (n = 65,425; 1991–2019), and the HPFS (n = 26,179; 1986–2018) were meta-analyzed using fixed-effects, inverse-variance models.

Abbreviations: ASB = artificially sweetened beverages; SSB = sugar-sweetened beverages; s/d = serving/day; NHS = Nurses’ Health Study; NHS II = Nurses’ Health Study II; HPFS = Health Professionals Follow-up Study.

Associations between three-serving-per-week increase in beverages with long-term weight trajectory patterns.

Latent class growth models assessing the three cohorts identified four distinct long-term weight-change trajectories among females and males combined (Figure 5, panel A). The “weight maintenance” group, comprising 41.7% of participants, showed minimal change, with an average weight change of −1.4 kg from baseline to year 24. The “slow weight-gain” group (40.2%) experienced an estimated gain of 7.0 kg, while the “moderate weight-gain” group (14.7%) showed a larger increase of 17.2 kg. The “rapid weight gain” (3.3%) group exhibited a substantial increase of 32.1 kg over the 24-year period.

Figure 5.

Latent trajectories of weight change (kg; Panel A) and odds of weight-change trajectories according to beverage intake (per 3 servings/week; Panel B) in three U.S. cohorts.

Panel A shows four latent trajectories of weight change from baseline over 24 years of follow-up (to harmonize the follow-up period across all cohorts): rapid weight gain, moderate weight gain, slow weight gain, and weight maintenance, identified using latent class growth models. Values represent mean weight change (kg) at each follow-up time point. Class sizes (n and %) are shown in the legend.

In Panel B, odds ratios (ORs) and 95% confidence intervals (CIs) for membership in each weight-change trajectory (rapid, moderate, or slow weight gain) compared to the weight-maintenance trajectory (reference) are shown per 3-servings/week higher intake of SSB, ASB, or water. Models were multinomial logistic regressions and were adjusted for age, questionnaire cycle, waist circumference (1986 for NHS and HPFS; 1991 for NHS II), and BMI at baseline. Dietary adjustments included baseline intake and 4-year changes in fruits, vegetables, whole and refined grains, processed and red meats, nuts, sweets/desserts, fried foods, low-fat and whole-fat dairy, coffee, tea, and other industrially processed products. Lifestyle adjustments included baseline and 4-year changes in physical activity, sleep duration, smoking status, and alcohol intake, as well as sitting/television time (baseline only for NHS and NHS II; baseline and changes for HPFS). Models also included weight-control behaviors at each interval: low-calorie, low-fat, or low-carbohydrate diet; meal skipping/fasting; exercising for weight loss; unintentional weight loss; diet pill use; commercial weight-loss programs; smoking or alcohol changes for weight loss; dietary supplements; prescription weight-loss drugs; fenfluramine, phentermine, or dexfenfluramine use; intentional weight loss in the prior 2 years; and portion-size restriction. Models also included neighborhood socioeconomic status at interval start.

Abbreviations: ASB = artificially sweetened beverages; SSB = sugar-sweetened beverages; CI = confidence interval; HPFS = Health Professionals Follow-up Study; NHS = Nurses’ Health Study; NHS II = Nurses’ Health Study II; OR = odds ratio.

In multivariate-adjusted multinomial models, higher 4-year increases in SSB intake were strongly associated with greater odds of belonging to less favorable weight-gain trajectories compared with weight maintenance. For each 3 servings per week increase in SSBs, the odds ratio (OR) for membership in the rapid weight gain trajectory was 1.31 (95% CI, 1.19–1.46), and 1.08 (95% CI, 1.01–1.15) for the moderate weight gain trajectory, with no association observed for the slow weight gain group (Figure 5, panel B). In contrast, increases in ASB intake were associated with lower odds of rapid weight gain (OR, 0.93; 95% CI, 0.87–0.99), and were not related to membership in the other trajectories. Similarly, higher water intake was associated with reduced odds of rapid weight gain (OR, 0.93; 95% CI, 0.90–0.96), but showed no associations with the remaining trajectory groups.

Sensitivity analyses of one-serving-per-day beverage changes

To ensure comparability with prior literature and with common dietary-substitution frameworks based on one-for-one serving replacements, we further modeled beverage changes as one-serving-per-day increments. Pooled estimates combining females and males indicated that each one-serving increase in ASB intake was associated with weight and BMI changes of −0.06 kg (95% CI: −0.09, −0.04) and −0.02 kg/m² (95% CI: −0.03, −0.01), respectively, over four-year intervals.

Replacing one SSB serving with one ASB serving was associated with an estimated change of −0.49 kg (95% CI: −0.53, −0.45) and −0.18 kg/m² (95% CI: −0.19, −0.16), and replacement with water yielded corresponding changes of −0.51 kg (95% CI: −0.54, −0.47) and −0.18 kg/m² (95% CI: −0.20, −0.17). Replacing one ASB serving with water was associated with a weight change of −0.04 kg (95% CI: −0.06, −0.01) and BMI change of −0.01 kg/m² (95% CI: −0.03, 0.00).

Consistent with the primary analysis, the associations between each one-serving increase in ASB intake and lower weight or BMI gain were stronger among participants with overweight or obesity.

Cohort-specific and pooled estimates for each one-serving-per-day increase in beverage intake, as well as the corresponding substitution and BMI-stratified analyses, are presented in Tables S7 and Table S8 of the Supplementary material.

Sensitivity analyses adjusting for energy intake and excluding GLP-1 consumers.

To account for potential compensatory changes in energy intake following ASB consumption, we additionally adjusted the ASB–weight-change models for total energy intake and excluded participants (1.4%) using GLP-1 receptor agonists (exenatide, liraglutide, or dulaglutide) in the 2017 cycle (data available in NHS II only). Results remained consistent and are presented in Tables S9 and Table S10 of the Supplementary material.

DISCUSSION

In this study of three large U.S.-based cohorts, participants followed for between 24 to 32 years who increased their ASB intake experienced less weight and BMI gain over four-year intervals. Modeled substitution of SSBs with ASBs or water resulted in larger, more favorable weight and BMI changes. Replacing ASBs with water was also associated with slight but favorable outcome changes.

We consistently observed dose-dependent associations in categorical substitution models. Stepwise replacements of SSBs with ASBs were associated with progressively less weight and BMI gains, with even stronger associations when replacing SSBs with water. Although ASBs are non-caloric, gradually substituting them with water also resulted in progressively less weight and BMI gain. However, four-year changes in water intake were smaller than those of ASBs and SSBs, which may explain why larger SSB-to-ASB replacements (≥1 serving/day) had a more pronounced change in weight and BMI than SSB-to-water substitutions. The association between increased ASB intake and lower weight and BMI gain was stronger in participants with overweight and obesity.

Since SSB and ASB intake are not mutually exclusive behaviors, we conducted stratified analyses by changes in SSB intake. Weight and BMI changes were largely independent of ASB intake. Compared to participants with no change in either beverage, those who reduced SSB intake over four-year intervals experienced less weight gain, regardless of their ASB intake changes. Conversely, those who increased SSB intake experienced more weight gain, irrespective of their ASB intake. Overall, stratified findings suggest that replacing SSBs with ASBs may be more closely associated with lower weight and BMI gain among individuals with obesity and those with higher habitual SSB intake.

To better capture heterogeneity in long-term weight change and to identify periods of accelerated gain or loss that may be obscured in analyses relying on mean changes across intervals, we fit latent growth trajectory models and identified four mutually exclusive weight-change trajectories. Compared with weight maintenance, increases in ASB and water intake were associated with lower odds of belonging to unhealthy weight-change trajectories, whereas increases in SSB intake were associated with progressively higher odds of belonging to the moderate and rapid weight-gain trajectories.

Discordant findings in previous cohort analyses (2,11,32,33) likely stem from methodological discrepancies. Some studies assessed baseline or prevalent ASB intake and estimated subsequent weight changes, while others used a more robust longitudinal framework that modeled concurrent changes in ASB intake and weight changes over time. Additional variability arises from examining ASBs in isolation or explicitly modeling their substitution for SSBs (34).

In an effort to address these methodological constraints, a WHO committee conducted a meta-analysis of RCTs and observational studies to evaluate the health implications of artificial sweetener consumption (9). For adult weight change, the analysis comprised 29 RCTs, four prospective cohort studies of continuous exposure, and five prospective cohort studies of categorical exposure. For BMI change, the analysis included 23 RCTs and five observational studies of continuous exposure. Across RCTs comparing baseline with final assessments, consumers of artificial sweeteners experienced a mean difference (MD) of −0.71 kg (95% CI: −1.13 to −0.28 kg) in weight and −0.14 kg/m² (95% CI: −0.30 to 0.02 kg/m²) in BMI relative to non-consumers. Although observational in design, our findings are consistent with these RCT-based estimates.

The WHO meta-analysis (9) of observational studies reported no association between continuous artificial sweetener exposure and weight change [MD −0.12 kg (95% CI: −0.40 to 0.15 kg)] or between high and low categorical exposure [MD −0.01 kg (95% CI: −0.67 to 0.64)] but found a modest positive association with BMI gain [MD 0.14 kg/m² (95% CI: 0.03 to 0.25 kg/m²)]. Our findings differ from these pooled estimates. These discrepancies can be attributed to acknowledged limitations in the WHO Nutrition and Food Safety Department’s artificial sweetener guideline (15,34). Notably, the WHO meta-analysis (9) relied heavily on prospective cohort studies that assessed prevalent or baseline artificial sweetener exposure, an approach prone to bias due to behavior clustering, residual confounding, and reverse causality (35). In contrast, our study examined changes in ASB intake alongside subsequent changes in weight and BMI using repeated assessments.

Several biological mechanisms have been proposed to explain how ASBs may influence weight changes. Some RCTs suggest that SSB avoidance through ASBs could support short-term weight control by reducing energy intake (36,37). Observational evidence has described complementary biological pathways. One hypothesis is that sustained ASB intake could activate sweet taste receptors that initiate hyperinsulinemia and disrupt the cephalic phase response, leading to metabolic dysregulation (38). ASBs may also contribute to microbiota dysbiosis, potentially promoting adiposity and insulin resistance (38,39). While these mechanisms are supported largely by in vitro and animal models, their direct translation to humans remains uncertain and has been questioned by complementary research (40,41). In mediation analysis, we found that a proportion of the ASB-weight changes association is mediated by energy intake from SSBs. Notably, despite being non-caloric, our findings also showed that replacing ASBs with water was associated with small but favorable changes in weight and BMI, suggesting the involvement of biological pathways beyond energy displacement from SSB avoidance.

This study has several strengths. We leveraged large prospective data from both females and males, collected repeatedly over 24-, 28-, and 32-year follow-up periods. Our analyses accounted for within-person variation and time-varying confounding. We evaluated beverage substitutions, assessed dose-response associations, and explored effect modification by baseline BMI classification and habitual SSB intake. We also identified subgroups within the population that followed distinct and otherwise undetectable patterns of weight change, providing a more nuanced understanding of long-term weight dynamics. We evaluated changes in ASB intake relative to subsequent changes in weight and BMI using sequential assessments, an approach that improves on prior studies relying on baseline or prevalent ASB exposure by better accounting for physiological adaptation following weight changes and reducing reverse causation. Detailed dietary data allowed adjustment for multiple potential confounders, including baseline and changes in intake of numerous foods and beverages. Although dietary items were not explicitly grouped by NOVA categorization, these covariates captured processed and ultra-processed consumption at the item level.

Limitations of our study include being restricted to U.S. health professionals, who may differ from the general population. Although ASB and SSB intake were relatively low, the distribution of dietary intake overlapped with national estimates (7) and was consistent across all three cohorts, supporting reasonable external validity. Our analyses focused on carbonated and non-carbonated beverages as the primary sources of artificial sweeteners in the U.S. (42). We were unable to disaggregate the contributions of individual sweeteners in these beverages, examine the broader range of ultra-processed products containing artificial sweeteners or evaluate those added by consumers to food and beverages. This design, however, enabled standardized beverage-substitution modeling, which would be more complex with varied products and portion sizes. Although ASB formulations evolved over time, we captured most of the aspartame intake, as ASBs account for the large majority (~86%) of aspartame in the diet (43,44). Similarly, most U.S. SSBs are sweetened with high-fructose corn syrup (HFCS). Although HFCS and sucrose contain similar proportions of glucose and fructose, HFCS provides these sugars in free form, allowing faster absorption and has been associated with weight gain and downstream metabolic disturbances. Therefore, the SSB-to-ASB/water substitution estimates in our study may differ in regions where SSBs are sweetened with cane or beet sugar.

In conclusion, increases in ASB intake were associated with less weight and BMI gains over time, particularly among persons with overweight or obesity and those with higher SSB intake. Replacing SSBs with ASBs was associated with larger reductions in weight and BMI in a dose-dependent manner. Increases in ASB and water intake were inversely associated with unhealthy long-term weight-change patterns, whereas increases in SSBs were consistently associated with trajectories marked by greater weight gain. Overall, however, water emerged as the best substitute for both SSBs and ASBs.

Abbreviations:

ASBs

Artificially sweetened beverages

SSBs

Sugar-sweetened beverages

BMI

Body mass index

NHS

Nurses’ Health Study

NHS II

Nurses’ Health Study II

HPFS

Health Professionals Follow-up Study

FDA

Food and Drug Administration

WHO

World Health Organization

SFFQ

Semiquantitative food-frequency questionnaires

RCTs

Randomized Controlled Trials

Data Availability

Data described in the manuscript, code book, and analytic code will be made available upon reasonable request pending approval from the corresponding author.