Abstract
Background
Choline plays key roles in neurotransmitter synthesis and membrane integrity, processes fundamental to brain function. However, evidence from population studies on the association of dietary choline intake with dementia risk is mixed.
Methods
We analyzed data of 5,301 dementia-free participants (mean age = 68.6 years; 59% females) in the Health and Retirement Study and followed them through 2022. Dietary choline intake was assessed with a validated food frequency questionnaire (2013–2014) and energy-adjusted using the residual method. Incident dementia was identified with the Langa-Weir algorithm. Cox models estimated hazard ratios (HRs) across quintiles of total intake of choline and its contributing components. We additionally conducted a dose–response meta-analysis of existing prospective cohorts using a two-stage random-effects model with restricted cubic splines.
Results
During follow-up (median = 8.4 years, interquartile range: 6.4–8.9 years), 506 individuals developed dementia. Total choline intake showed a non-linear association with dementia risk. Compared with the lowest quintile, the multivariable-adjusted HRs (95% CI) for incident dementia across increasing quintiles of total choline were 0.88 (0.66–1.17), 0.87 (0.65–1.17), 0.68 (0.49–0.92), and 0.91 (0.66–1.25) (P-linear-trend: 0.198, P-non-linearity: 0.024). Among choline contributing components, moderate phosphatidylcholine and sphingomyelin intake levels were inversely associated with dementia risk (HR comparing quintile 4 to 1: 0.62, 95%CI: 0.46–0.84 and 0.70, 0.52–0.94, respectively). The meta-analysis of five cohorts (137,607 participants, 2459 dementia cases) showed an L-shaped association of dietary choline intake with dementia risk (P-linear-trend: 0.003, P-non-linearity: 0.036) with a turning point at 465 mg/day.
Conclusions
In this prospective cohort study and dose–response meta-analysis, moderate dietary choline intake was associated with a lower risk of incident dementia in middle-aged and older adults. These findings support further investigation of choline in relation to cognitive aging, particularly the potential non-linear association.
Registry for meta-analyses
CRD420251151642.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12937-026-01354-2.
Introduction
Dementia is a leading cause of disability and dependence with a significant socioeconomic burden as populations age [1, 2]. Due to the limitations in pharmacological treatments [3], lifestyle and dietary interventions have emerged as key strategies for dementia prevention [4, 5]. Among dietary components, choline is involved in pathways relevant to brain health. This essential nutrient serves as a precursor for acetylcholine, a key neurotransmitter involved in memory and learning, and is required for maintaining cell-membrane integrity through its role in phospholipid synthesis, as well as a methyl donor [6–8]. Animal studies have demonstrated the neuroprotective properties of choline, showing that choline supplementation improves synaptic plasticity [9], reduces neuroinflammation, and attenuates amyloid-β pathology [10]. These findings suggest that adequate choline intake may help preserve cognitive function and delay neurodegeneration.
Despite the biological plausibility, the relationship between dietary choline intake and dementia risk in humans remains uncertain [11–13]. Because evidence has been insufficient to establish a Recommended Dietary Allowance, many guidelines rely on Adequate Intake values [14]. Most of the studies conducted to date had inconsistent findings and were limited by cross-sectional or case–control designs. Given the long preclinical phase of dementia that may change dietary behaviors and the performance of recall, such biases could distort the observed associations between nutrient intake and disease risk. Several existing prospective cohort studies on the association between habitual choline intake and incident dementia generated varied findings. For example, in the UK Biobank, dietary choline intake showed a U-shaped association, with both lower and higher intakes (mean intake: 176 and 465 mg/d) linked to increased Alzheimer's disease (AD) risks [15]. In the Rush Memory and Aging Project (MAP), however, higher choline intake (> 350 mg/d) was associated with a lower risk of AD dementia, although a further curve-based analysis suggest a the point of lowest risk for AD to be ∼350 mg/d[16]. While the heterogeneity could emerge from the differences in distributions of choline intake levels, a non-linear dose–response relationship between choline intake and dementia is biologically plausible because while adequate levels of this nutrient are essential for brain functioning, it can also be converted to trimethylamine N-oxide (TMAO), a metabolite related to worse cardiometabolic health [17, 18], which may promote neurodegeneration. Finally, the roles of individual forms of choline on dementia risk are unclear.
Therefore, we hypothesized that dietary choline intake was related to dementia in a non-linear pattern. To test this, we assessed the prospective associations of dietary choline intake and its contributing component with incident dementia using data from the U.S. Health and Retirement Study (HRS) of middle-aged and older adults. We further conducted a dose–response meta-analysis of available prospective cohort studies to provide the most comprehensive and quantitative summary of evidence to date.
Methods
Cohort analysis
Study design and participants
We conducted a prospective cohort study using data from the HRS, a nationally representative survey of U.S. adults aged 50 years and older since 1992. Details of the HRS design, sampling strategy, and data collection procedures have been described previously [19]. Briefly, the HRS collects information on sociodemographic characteristics, health behaviors, medical history, and cognitive function through biennial core interviews.
For the present study, the baseline was specified as the 2013–2014 wave, at which dietary intake was assessed via a validated food frequency questionnaire (FFQ) [20]. Participants were eligible if they were ≥ 55 years of age, had completed the dietary questionnaire, and were free of stroke and dementia at baseline and remained dementia-free during the first two years of follow-up. We excluded individuals with implausible total energy intake (< 500 or > 4,000 kcal/day for females; < 800 or > 4,200 kcal/day for males) [21], or incomplete follow-up for dementia status (Fig. 1A), resulting a final analytic sample of 5,301 participants. The baseline characteristics of the included vs. the excluded participants were shown in Supplementary Table 1.
Fig. 1.
Inclusion flowchart for cohort analysis (A) and meta-analysis (B). In panel A, n indicates number of participants; in panel B, k indicates number of publications
Dietary assessment
Dietary intake was assessed using a validated 137-item semi-quantitative FFQ administered in 2013–2014. Participants reported their usual frequency of consumption of each food item over the previous year, with predefined portion sizes. Average daily intakes of nutrients and energy were calculated by multiplying the frequency of each food item by its nutrient content (from the U.S. Department of Agriculture food composition database) and summing across all foods. The FFQ has been extensively validated against multiple assessment methods in assessing dietary intake of foods and nutrients [20, 22]. Total dietary choline intake (mg/day, not including that from supplements) was computed as the sum of free choline, phosphocholine, glycerophosphocholine, phosphatidylcholine, and sphingomyelin. In a previous validation study, choline intake measured by FFQ showed a moderate-to-high correlation with the 7-day dietary record (Spearman correlation: 0.62) [23]. To reduce extraneous variation in total energy intake and energy misreporting, we adjusted the intake levels of choline for total energy intake using the residual methods [24].
Ascertainment of incident dementia
In the HRS, we identified incident dementia using the validated Langa–Weir classification, which incorporates information from cognitive performance measures for self-respondents and from proxy interviews in biennial core interviews [25]. For self-respondents, a 27-point cognitive score was derived from immediate and delayed word recall, serial 7 subtraction, and backward counting. A score of 0–6 indicated dementia. For proxy respondents, an 11-point scale incorporated instrumental activities of daily living (IADLs), proxy ratings of memory, and interviewer assessments. A proxy score of 6–11 denoted dementia. Follow-up time accrued from baseline until dementia diagnosis, death, last contact, or the end of the follow-up, whichever came first.
Covariates
We collected information on multiple covariates for confounding adjustment based on the previous literature, including sociodemographic, lifestyle, clinical factors, and other dietary confounders [16, 21, 26]. Sociodemographic confounders included age, sex defined by self-identity, race (White/Caucasian, Black/African American, or Others), household income (in quartiles, dollar), and duration of formal education. We also included a set of lifestyle variables including smoking status (never, former or current), alcohol drinking status (never, former or current), frequency of vigorous physical activity (< 1 time/week, 1- < 3 times/week, ≥ 3 times/week), body mass index (BMI) categories (< 25.0 kg/m2, 25.0- < 30.0 kg/m2, or ≥ 30.0 kg/m2; calculated as weight in kilograms divided by height in meters squared). Center for Epidemiologic Studies Depression Scale (CES-D) score was included to assess depressive symptoms. Clinical factors as binary categorical variables included hypertension, heart disease, diabetes, and cancer. To further account for confounding from other dietary nutrients, we adjusted the models for total energy intake, intake of vitamin B12, folate, vitamin C, vitamin E, vitamin B6, and lutein, according to a previous study [16]. Missing covariate data (< 5% for all variables) were imputed using multiple imputation by chained equations with five iterations.
Statistical analysis
We described baseline characteristics of participants according to quintiles of total choline intake, expressed as means (standard deviations) or medians (interquartile ranges) for continuous variables and as counts (percentages) for categorical variables.
For the primary analysis, we used Cox proportional hazard models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for incident dementia by total and contributing components of choline intake. The intake levels were modelled as quintile-defined categorical variables. Models were sequentially adjusted for demographic, socioeconomic, lifestyle, and clinical and other dietary covariates. Proportional hazard assumption was verified using the Schoenfeld residual methods. To assess whether the choline-dementia associations differed by study subgroups, stratified analyses were conducted by age group (< 65 and ≥ 65 years), sex, smoking status and alcohol drinking status. Effect modification was formally tested using interaction terms between total choline intake and subgroup variable in fully adjusted models.
To further assess the robustness of our findings, we conducted several sensitivity analyses: (1) to account for potential genetic susceptibility to dementia, we included APOE ε4 carrier status (carrier, non-carrier, or missing) in the fully adjusted model; (2) to reduce residual confounding from overall dietary quality, we further adjusted for adherence to the Alternative Healthy Eating Index–2010 (AHEI-2010);[27] (3) we additionally adjusted the models for dietary fatty acid composition and cholesterol intake; (4) we excluded participants with a baseline history of heart disease to further reduce confounding; (5) we excluded incident dementia cases occurring within the first 4 years of follow-up to further reduce reverse causality; (6) we additionally adjusted the models for pre-baseline cognitive status (in 2012) to account for potential influence of preclinical cognitive impairment; and (7) we used Fine-Gray competing risk analysis and reported sub-distribution hazard ratios to account for the competing risk of non-dementia mortality.
Meta-analysis
Literature search and eligibility
We conducted a systematic review and meta-analysis in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines and registered the protocol with PROSPERO (CRD420251151642) [28]. A comprehensive search of PubMed/MEDLINE, Embase (Ovid), and Web of Science Core Collection was performed from database inception through 18 September 2025 (full strategy provided in Supplementary Table 2). No language or publication date restrictions were applied. Reference lists of all eligible reports and relevant reviews were hand-searched to identify additional studies.
We included prospective cohort of adults (≥ 18 years) that quantitatively assessed dietary choline intake from foods (total choline or specific subtypes such as phosphatidylcholine) using validated dietary assessment methods and reported incident all-cause dementia or subtypes (e.g., AD dementia and vascular dementia). Eligible studies were required to report multivariable-adjusted relative risk (RR), hazard ratio (HR), or odds ratio (OR) estimates, or sufficient data for their calculation. We excluded studies on choline supplementation only; cross-sectional or conventional case–control designs; analyses restricted to populations with a specific pre-existing condition (e.g., Parkinson’s disease or stroke); mechanistic or animal experiments; case reports; narrative reviews; editorials; and conference abstracts lacking full data (Fig. 1B).
Study selection, data extraction and quality assessment
All search results were de-duplicated, and 2 reviewers independently screened titles and abstracts against eligibility criteria, classifying records as “include,” “exclude,” or “uncertain.” Full texts of potentially eligible articles were then assessed independently by the same reviewers. Discrepancies were resolved by discussion or by consultation with a third reviewer.
From each included study, 2 reviewers independently extracted data on cohort characteristics, participant demographics, dietary assessment methods, choline intake categories, follow-up duration, outcome definitions, risk estimates (HR, RR, or OR with 95% CIs), and covariates used for adjustment. Risk of bias was evaluated using the Newcastle–Ottawa Scale (range: 0–9) for cohort studies prior to data analysis.
Data analysis
To fit the dose–response association between choline intake and dementia risk, we fitted restricted cubic spline models (knots at the 10th, 50th, and 90th percentiles of choline intake) using the Greenland-Longnecker methods with the method of moments for CI estimation [29]. HRs or ORs were treated as approximations of RRs given the low absolute incidence of dementia. Between-study heterogeneity was assessed using the I2 statistic. To further provide quantitative results, we calculated pooled RRs for highest versus lowest and medium versus lowest categories of dietary choline intake using a random-effects model [30]. To avoid overlap of reference groups when pooling medium- versus lowest-intake estimates, we extracted only one non-overlapping estimate per study. For studies with > 3 categories, the second-highest category was treated as “medium.” We also conducted a sensitivity meta-analysis excluding the Rush MAP study because its baseline age was substantially older.
All analyses were conducted using R version 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria). Two-sided P-values < 0.05 were considered statistically significant.
Standard protocol approvals, registrations and patient consents
HRS received approval from the University of Michigan institutional review board. All participants in the study signed informed consent. The protocol for the meta-analysis was registered with PROSPERO (CRD420251151642) [28].
Results
Cohort analysis
Baseline characteristics of study participants
The mean age of the 5,301 Health and Retirement Study (HRS) participants was 68.6 (SD: 9.4) years, and 59.2% were females (Table 1). About half (52.1%) had attained a college degree. Compared with those in the lowest quintile of choline intake (median 233.9 mg/d), those in the highest quintile (median: 395.0 mg/d) were more likely to be White, and less likely to be current smokers. In the study population, the primary food contributors of dietary choline included eggs, red and processed meats, poultry, and nuts (Supplementary Fig. 1), and participants with higher choline intake primarily had higher proportional contributions from eggs (Supplementary Fig. 2). The distributions of the total and individual contributing components of choline were shown in Supplementary Table 3.
Table 1.
Baseline characteristics of the study participants in the health and retirement study
| Variable | Overall | Choline Intake | ||||
|---|---|---|---|---|---|---|
| Quintile 1 | Quintile 2 | Quintile 3 | Quintile 4 | Quintile 5 | ||
| N | 5301 | 1061 | 1060 | 1060 | 1060 | 1060 |
| Choline intake, mg/d, (median [IQR]) | 303.5 [268.1, 343.4] | 233.9 [208.8, 247.5] | 275.8 [268.1, 283.1] | 303.5 [297.4, 310.3] | 333.3 [324.8, 343.4] | 395.0 [371.5, 430.7] |
| Age, years, mean (SD) | 68.6 (9.4) | 68.2 (9.5) | 68.8 (9.5) | 68.5 (9.0) | 68.7 (9.2) | 68.7 (9.6) |
| Female, n (%) | 3139 (59.2) | 575 (54.2) | 662 (62.5) | 649 (61.2) | 667 (62.9) | 586 (55.3) |
| Race, n (%) | ||||||
| White/Caucasian | 4124 (78.0) | 797 (75.3) | 812 (77.0) | 815 (77.0) | 861 (81.4) | 839 (79.2) |
| Black/African American | 807 (15.3) | 169 (16.0) | 182 (17.3) | 166 (15.7) | 140 (13.2) | 150 (14.2) |
| Others | 359 (6.8) | 92 (8.7) | 61 (5.8) | 78 (7.4) | 57 (5.4) | 71 (6.7) |
| College degree attainment, n (%) | 2762 (52.1) | 514 (48.4) | 532 (50.2) | 578 (54.5) | 580 (54.7) | 558 (52.6) |
| Smoking status, n (%) | ||||||
| Never | 2456 (46.3) | 491 (46.3) | 526 (49.6) | 482 (45.5) | 483 (45.6) | 474 (44.7) |
| Former | 2307 (43.5) | 439 (41.4) | 442 (41.7) | 480 (45.3) | 463 (43.7) | 483 (45.6) |
| Current | 538 (10.1) | 131 (12.3) | 92 (8.7) | 98 (9.2) | 114 (10.8) | 103 (9.7) |
| Alcohol drinking status, n (%) | ||||||
| Never | 2272 (42.9) | 510 (48.1) | 476 (44.9) | 413 (39.0) | 416 (39.2) | 457 (43.1) |
| Former | 834 (15.7) | 160 (15.1) | 161 (15.2) | 173 (16.3) | 182 (17.2) | 158 (14.9) |
| Current | 2195 (41.4) | 391 (36.9) | 423 (39.9) | 474 (44.7) | 462 (43.6) | 445 (42.0) |
| Vigorous physical activity, times/week, n (%) | ||||||
| < 1 | 1322 (24.9) | 315 (29.7) | 298 (28.1) | 214 (20.2) | 251 (23.7) | 244 (23.0) |
| 1- < 3 | 2125 (40.1) | 393 (37.0) | 413 (39.0) | 450 (42.5) | 424 (40.0) | 445 (42.0) |
| ≥ 3 | 1854 (35.0) | 353 (33.3) | 349 (32.9) | 396 (37.4) | 385 (36.3) | 371 (35.0) |
| Body mass index, kg/m2, n (%) | ||||||
| < 25.0 | 1417 (26.7) | 330 (31.1) | 286 (27.0) | 266 (25.1) | 288 (27.2) | 247 (23.3) |
| 25.0- < 30 | 1973 (37.2) | 389 (36.7) | 413 (39.0) | 421 (39.7) | 377 (35.6) | 373 (35.2) |
| ≥ 30 | 1911 (36.0) | 342 (32.2) | 361 (34.1) | 373 (35.2) | 395 (37.3) | 440 (41.5) |
| CES-D score, mean (SD) | 1.3 (1.9) | 1.4 (2.0) | 1.3 (2.0) | 1.2 (1.9) | 1.1 (1.7) | 1.4 (1.9) |
| Diabetes, n (%) | 1192 (22.5) | 186 (17.5) | 218 (20.6) | 260 (24.5) | 243 (22.9) | 285 (26.9) |
| Hypertension, n (%) | 3138 (59.2) | 632 (59.6) | 617 (58.2) | 617 (58.2) | 623 (58.8) | 649 (61.2) |
| Heart diseases, n (%) | 1166 (22.0) | 236 (22.3) | 233 (22.0) | 234 (22.1) | 234 (22.1) | 229 (21.6) |
Dietary choline intake and incident dementia
During follow-up (median = 8.4 years, interquartile range: 6.4–8.9 years), 506 participants (9.5%) developed incident dementia. In general, intake of dietary choline showed a non-linear association with dementia risk (Table 2). In the fully adjusted model, compared with the lowest quintile, participants in the fourth quintiles (range: 318–357 mg/d) had a 32% (HR: 0.68; 95%CI: 0.49–0.92) lower risk of incident dementia, respectively, compared with those in the lowest quintile. The highest quintile did not show significant association (HR: 0.91; 95%CI: 0.66–1.25), and the overall trend across quintiles was not statistically significant (P-trend = 0.198). The restricted cubic spline analysis revealed a significant non-linearity in the association, with a P-value for non-linearity = 0.024 and a turning point at ~ 344 mg/day (Supplementary Fig. 3).
Table 2.
Hazard ratios of incident dementia across increasing quintiles of total and contributing components of choline in the health and retirement study
| Hazard Ratios (95% Confidence Interval) | P-trend | |||||
|---|---|---|---|---|---|---|
| Quintile 1 | Quintile 2 | Quintile 3 | Quintile 4 | Quintile 5 | ||
| Total choline | ||||||
| Median intake (mg/day) | 234 | 276 | 304 | 333 | 395 | |
| Cases/Person-years | 118/7616 | 101/7721 | 101/7999 | 76/7853 | 110/7784 | |
| Model 1 | 1.00 (Reference) | 0.79 (0.60—1.04) | 0.78 (0.60—1.03) | 0.59 (0.44—0.79) | 0.85 (0.66—1.11) | 0.067 |
| Model 2 | 1.00 (Reference) | 0.87 (0.66—1.15) | 0.91 (0.69—1.20) | 0.68 (0.51—0.91) | 0.94 (0.72—1.22) | 0.271 |
| Model 3 | 1.00 (Reference) | 0.88 (0.66—1.17) | 0.87 (0.65—1.17) | 0.68 (0.49—0.92) | 0.91 (0.66—1.25) | 0.198 |
| Contributing components of choline | ||||||
| Free Choline | 1.00 (Reference) | 1.04 (0.78—1.38) | 0.99 (0.73—1.34) | 0.88 (0.63—1.22) | 0.98 (0.67—1.42) | 0.778 |
| Glycerophosphocholine | 1.00 (Reference) | 0.97 (0.72—1.32) | 1.08 (0.80—1.45) | 1.17 (0.87—1.58) | 1.35 (1.01—1.82) | 0.864 |
| Phosphocholine | 1.00 (Reference) | 1.07 (0.81—1.42) | 0.84 (0.61—1.15) | 1.10 (0.81—1.50) | 1.45 (1.07—1.96) | 0.645 |
| Phosphatidylcholine | 1.00 (Reference) | 0.96 (0.74—1.26) | 0.75 (0.56—1.00) | 0.62 (0.46—0.84) | 0.76 (0.57—1.01) | 0.787 |
| Sphingomyelin | 1.00 (Reference) | 0.78 (0.59—1.04) | 0.80 (0.60—1.08) | 0.70 (0.52—0.94) | 0.81 (0.60—1.09) | 0.094 |
The hazard ratios and 95% confidence intervals for total choline were calculated from Cox proportional hazard models. Model 1 was adjusted for age, sex, and total energy intake. Model 2 was further adjusted for race, education, household income, smoking status, alcohol consumption, physical activity level, and body mass index category. Model 3 (the primary model) included all variables from Model 2 and was additionally adjusted for depressive symptoms, diabetes, hypertension, heart disease, cancer, and intake of vitamin B12, folate, vitamin C, vitamin E, vitamin B6, and lutein. For individual contributing components of choline, estimates from Model 3 were presented
Boldface indicates statistically significant estimate
In the analyses of individual choline-containing components, we observed no significant associations for free choline, while highest glycerophosphocholine and phosphocholine quintiles were associated with a higher dementia risk, with HRs being 1.35 (1.01–1.82) and 1.45 (1.07–1.96), respectively. By contrast, phosphatidylcholine and sphingomyelin showed J-shaped inverse associations, with HRs of 0.62 (0.46–0.84) and 0.70 (0.52–0.94) for the fourth quintiles, respectively, compared with the lowest quintile.
Subgroup and sensitivity analyses
The inverse association between moderate choline intake and incident dementia was generally consistent across study subgroups (Supplementary Table 4). No significant interaction was observed with age, sex, smoking status, or alcohol drinking (all p-interaction > 0.10). Sensitivity analyses further supported the robustness of these findings (Supplementary Table 5). Additional adjustment for APOE genotype, overall dietary quality, or dietary fatty acid compositions and cholesterol intake had minor influences on the associations. Similarly, excluding participants with a baseline history of heart disease did not materially alter the associations. Excluding incident dementia cases occurring within the first 4 years of follow-up slightly strengthened the association (HR 0.58, 95%CI: 0.39–0.86, for the fourth vs. first quintiles). Additionally adjusting for pre-baseline cognitive status and Fine-Gray competing risk analysis yielded similar findings.
Meta-analysis
Characteristics of included publications
In the systematic review, 4 prospective cohort studies met the inclusion criteria and together included 132,306 participants (Table 3). These cohorts spanned diverse age ranges but were all conducted in Western countries, with follow-up periods ranging from 7.7 to 21.9 years. Ylilauri et al. followed 2,497 middle-aged males (mean baseline age 53 years) from the Kuopio Ischaemic Heart Disease Risk Factor Study in Finland for an average of 21.9 years [31]. Yuan et al. analyzed 3,224 Framingham Heart Study Offspring Cohort participants in the U.S. (mean baseline age 54.5 years; 53.8% females) over 16.1 years of follow-up [13]. Karosas et al. conducted a prospective analysis of 991 older adults from the Rush MAP in Chicago, U.S., with a mean follow-up of 7.7 years [16]. Niu et al. evaluated 125,594 UK Biobank participants (mean baseline age 56.1 years; 55.8% females) over a mean follow-up of 11.8 years [15]. Across all studies, the Newcastle–Ottawa Scale (NOS) scores were 7 or 8 (Supplementary Table 6), indicating overall good methodological quality.
Table 3.
Characteristics of published reports on the association between dietary choline intake and incident dementia
| First Author, Year | Country | Data Source/Cohort | Sample size | Mean age at baseline | %Female | Dietary assessment method | Outcome(s) | Mean/median follow-up duration |
|---|---|---|---|---|---|---|---|---|
| Ylilauri, M.P.T., 2019 [31] | Finland | Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD) | 2497 | 53.0 years | 0% | 4-day food records | Incident dementia | 21.9 years |
| Yuan, J., 2022 [13] | United States | Framingham Heart Study Offspring Cohort | 3224 | 54.5 years | 53.80% | Food Frequency Questionnaire (FFQ) | Incident dementia and AD | 16.1 years |
| Karosas, T., 2025 [16] | United States | Rush Memory and Aging Project (MAP) | 991 | 81.4 years | 74.50% | FFQ | Clinical diagnosis of AD | 7.7 years |
| Niu, Y.Y., 2025 [15] | United Kingdom | UK Biobank | 125,594 | 56.1 years | 55.80% | Web-based 24-h dietary recalls (Oxford WebQ) | Incident dementia, AD, and MCI | 11.8 years |
Pooled estimates for association between dietary choline intake and incident dementia
Across five prospective cohorts (four previous studies plus the present study, total N = 137,607 participants; 2,459 incident dementia cases), the two-stage random-effects dose–response meta-analysis revealed a non-linear association between dietary choline intake and risk of incident dementia (P-nonlinearity = 0.036). Dementia risk declined with increasing choline intake from the lowest observed levels up to 465 mg/d and then leveled off with no further reduction (Fig. 2), which is consistent with the association observed in the cohort analysis. Between-study heterogeneity was moderate and statistically non-significant (I2 = 37%, P-heterogeneity = 0.12). In sensitivity analyses excluding the Rush MAP study, the nadir remained at approximately 407 mg/d (Supplementary Fig. 4).
Fig. 2.
Dose–response meta-analysis on the association between dietary choline intake and incident dementia. The solid blue line represents the estimated restricted cubic spline (RCS) model for the pooled hazard ratio, with the shaded band indicating the 95% confidence interval (P-value for linear trend: 0.003; for non-linearity: 0.036; turning point: 465 mg/d). Each point and error bar indicate a point estimate of hazard ratio (95% confidence interval) reported by a study. The spline was fitted with knots located at the 10th, 50th, and 90th percentiles of the choline intake distribution. The model was derived from a two-stage random-effects meta-analysis of 5 studies, with the method of moments estimator used for the between-study variance and the Greenland & Longnecker covariance approximation. Between-study heterogeneity was moderate and statistically non-significant (I2 = 37%, P-heterogeneity = 0.12)
Quantitatively, the pooled HR for medium-to-high versus lowest intake levels was 0.71 (95% CI: 0.53–0.94; I2 = 45%, P-heterogeneity = 0.12), while that for highest versus lowest intake levels was 0.73 (0.46–1.17; I2 = 59%, P-heterogeneity = 0.045) (Supplementary Fig. 5), which confirmed the dose–response associations.
Discussion
In a prospective cohort study of U.S. adults and an accompanying dose–response meta-analysis of five prospective cohorts (N = 137,607; 2,459 incident dementia cases), we observed that higher dietary choline intake was associated with a lower risk of incident dementia in a non-linear manner. In the HRS cohort, participants with moderate choline intake experienced the lowest dementia risk, and further analyses suggested that the inverse association was strongest for phosphatidylcholine and sphingomyelin. When we pooled the HRS data with four previous cohort studies, the risk reached its lowest level at ~ 465 mg/day, beyond which additional intake may confer limited or no benefit.
Our findings extend and reconcile a heterogeneous body of observational research on choline and cognitive outcomes. Earlier prospective investigations generally reported inverse associations between dietary choline or phosphatidylcholine intake and risks of cognitive decline or AD. However, the magnitude and shape of these associations have varied: some studies suggested a linear inverse relation, whereas others non-linear associations. For example, in a 22-year longitudinal study in the China Health and Nutrition Survey (CHNS), higher intake of choline was related to better cognitive function in a linear manner (P-trend < 0.0001) [12]. On the contrary, in the UK Biobank, moderate dietary choline intake (332.89–353.93 mg/d), rather than higher intake, was associated with lower risk of dementia, AD, and mild cognitive impairment [15]. Similarly, in the Rush MAP, total choline intake was related to a lower risk of dementia, with a point of lowest risk for AD to be ∼350 mg/d [16]. The pattern observed in the CHNS may reflect differences in choline food sources, with a greater contribution from plant-based foods and lower contributions from dairy and other animal-source foods compared with Western cohorts. These differences in background diet and lifestyle suggest that the dose–response relationship observed in one population thus may not be directly generalizable to others. In the present study, the analyses among HRS participants generally showed a non-linear association of choline intake with dementia risk, with the third and fourth (adjusted intake to 1647 kcal/day: 291– < 357 mg/day) but not the highest quintile (357 mg/d or above) related to a lower dementia risk. By synthesizing all available longitudinal evidence, our meta-analysis clarifies this inconsistency and indicates that moderate, but not necessarily high, choline intake was associated with the lowest risk. Importantly, we observed statistically significant lower risk of dementia only for the moderate intake category rather than the highest intake category, which could be attributable both to insufficient accuracy in estimation and lack of true effect in the highest category. Nonetheless, because the upper portion of the RCS in HRS, particularly beyond approximately 395 mg/day, was informed by only about 10% of the sample, uncertainty is greater in this range and the apparent plateau at higher intakes should be interpreted cautiously.
Several biological mechanisms support a protective role for moderate choline intake. Choline is the direct precursor of acetylcholine and crucial for memory and learning [6]. Adequate choline may help preserve cholinergic function and delay the synaptic losses that characterize AD [32]. Choline is required for phosphatidylcholine and sphingomyelin synthesis that maintain neuronal membrane architecture and facilitate signal transduction [33]. Through its conversion to betaine, choline serves as a methyl donor, lowering circulating homocysteine, which is linked to cognitive decline and cerebrovascular pathology [34, 35]. Choline’s involvement in one-carbon metabolism also influences DNA methylation and epigenetic regulation of neurodegenerative pathways [7]. Adequate phosphatidylcholine supports lipoprotein assembly and may attenuate impaired mitochondrial function, increased oxidative stress and disrupted phosphatidylcholine metabolism by certain genetic factors [36]. However, choline also serves as a key substrate for gut microbial formation of trimethylamine (TMA), which is converted in the liver to TMAO that could harm vascular and metabolic health [17]. Although the role of TMAO in neurodegeneration is still elusive, this double-edged role may explain the plateau in the dose–response association curve [37]. The unfavorable associations observed for glycerophosphocholine and phosphocholine should be interpreted cautiously. Given the distinct biochemical properties and metabolic pathways of different choline-containing compounds, these associations may reflect complex interactions involving intestinal absorption, hepatic and gut-microbial metabolism, and endogenous tissue turnover, rather than a direct adverse effect.
The present study has several strengths. First, we leveraged a large, well-characterized U.S. cohort with repeated cognitive assessments and long follow-up, allowing for robust control of demographic, lifestyle, and dietary covariates. Second, we complemented these data with a comprehensive meta-analysis, yielding the most detailed dose–response evaluation of choline and dementia risk to date. Third, by distinguishing choline-contributing component intake, we provide novel insight into the specific dietary sources most relevant to neurodegeneration. From a public-health perspective, these findings highlight moderate choline intake as a potentially modifiable dietary factor for dementia prevention. While current dietary guidelines in the U.S. and China, and other countries often recommend an Adequate Intake of 400–500 mg/day [38, 39], our pooled results suggest that moderate intake levels were associated with the lowest observed dementia risk. This may inform dietary guidance that balances the benefits of choline-rich foods (e.g., eggs, fish, lean meats, legumes) with the potential vascular concerns related to excessive intake and TMAO generation. The pooled pattern was materially unchanged after exclusion of the Rush MAP study, suggesting that the overall non-linear association was not driven solely by this older cohort.
Several limitations should be considered while interpreting our findings. First, choline intake (dietary only) was estimated from food-frequency questionnaires or other self-reported tools administered only at baseline, which is prone to recall bias and measurement error and may attenuate the associations. As dietary intake can change over time, but repeated dietary assessments were unavailable in most cohorts (including the HRS), this may dilute the associations. Because the FFQ captured choline from foods but not supplements, total choline exposure may have been underestimated in some participants. Such misclassification is likely to be non-differential with respect to subsequent dementia status and would attenuate the observed associations. Although we adjusted for multiple dietary and lifestyle covariates, residual confounding by other micronutrients or overall dietary quality remains possible. Second, the potential non-differential misclassification of dementia could also attenuate the observed non-linear associations. Third, as subtle cognitive decline could alter eating habits, reverse causality could not be eliminated, although we have excluded early dementia cases in the primary analysis. Given the long prodromal phase of dementia, early disease-related changes in appetite or food choice might have led to lower reported choline intake, potentially exaggerating an inverse association. Fourth, heterogeneity across studies was moderate, which may reflect the differences in dietary assessment tools, follow-up duration, and dietary backgrounds and augment the uncertainty of evidence, especially in the risk estimation for the highest choline intake level. As we were only able to include five cohort studies in the meta-analysis due to the limitation of previous publications, future investigations, especially from non-Western populations, are needed for populations with other genetic and cultural backgrounds. Finally, we were unable to evaluate gene-diet or microbiota-diet interactions, both of which could modify choline metabolism and its cognitive effects.
In conclusion, this prospective cohort study and accompanying dose–response meta-analysis provide evidence that dietary choline intake is associated with a lower risk of dementia in a non-linear manner, with moderate intake associated with the lowest risk. Future mechanistic studies and randomized trials are needed to confirm causality and elucidate the interplay of choline metabolism with genetics and the gut microbiome to refine dietary recommendations aimed at the primary prevention of dementia.
Supplementary Information
Acknowledgements
The University of Michigan Health and Retirement Study (HRS) was supported by the National Institute on Aging (NIA U01AG009740) and the Social Security Administration. The authors extends sincere gratitude to all participants and staff of HRS for their dedicated efforts in data collection and processing.
Abbreviations
- AD
Alzheimer’s disease
- AHEI-2010
Alternative healthy eating index–2010
- APOE
Apolipoprotein E
- BMI
Body mass index
- CES-D
Center for epidemiologic studies depression scale
- CI
Confidence interval
- CIs
Confidence intervals
- FFQ
Food frequency questionnaire
- HR
Hazard ratio
- HRs
Hazard ratios
- HRS
Health and retirement study
- IADLs
Instrumental activities of daily living
- MAP
Memory and aging project
- MCI
Mild cognitive impairment
- NIA
National institute on aging
- NOS
Newcastle–Ottawa scale
- OR
Odds ratio
- PRISMA
Preferred reporting items for systematic reviews and meta-analyses
- RDA
Recommended dietary allowance
- RR
Relative risk
- TMA
Trimethylamine
- TMAO
Trimethylamine N-oxide
- UKB
UK biobank
- USDA
United States Department of Agriculture
Authors’ contributions
HC, YL, and CY conceived and designed the study. HC conducted statistical analysis and wrote the first draft of the paper. TL and MW collected data and provided statistical expertise. HC, TL, MH, JZ, ZG, JY, BL, DL, YL, and CY contributed to the interpretation of the results and critical revision of the manuscript for important intellectual content and approved the final version of the manuscript. YL and CY are the guarantors. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding
The current study was supported by the Fundamental Research Funds for the Central Universities (226–2025-00178) and the "Leading Goose" R&D Program of Zhejiang (2026C02A1149). The funding sources did not participate in the design or conduct of the study; collection, management, analysis or interpretation of the data; or preparation, review, or approval of the manuscript.
Data availability
Data used for cohort analysis could be obtained from the website of HRS (hrs.isr.umich.edu/dataproducts/access-to-public-data). Data used for the meta-analysis were obtained from the individual publications, which are available from the respective publishers.
Declarations
Ethics approval and consent to participate
The Health and Retirement Study (HRS) was approved by the Institutional Review Board of the University of Michigan. All participants provided written informed consent prior to participation. The current analyses were conducted using de-identified, publicly available data and were therefore exempt from additional institutional review. The meta-analysis component of this study used data extracted from published studies and did not require separate ethical approval.
Consent for publication
Not applicable. This study did not involve the publication of any individual-level or identifiable personal data.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Yanhui Lu, Email: luyanhui@bjmu.edu.cn.
Changzheng Yuan, Email: chy478@zju.edu.cn.
References
- 1.Livingston G, Huntley J, Liu KY, Costafreda SG, Selbæk G, Alladi S, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. Lancet. 2024;404:572–628. [DOI] [PubMed] [Google Scholar]
- 2.Nichols E, Steinmetz JD, Vollset SE, Fukutaki K, Chalek J, Abd-Allah F, et al. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. Lancet Public Health. 2022;7:e105–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Fox NC, Belder C, Ballard C, Kales HC, Mummery C, Caramelli P, et al. Treatment for Alzheimer’s disease. Lancet. 2025;406:1408–23. [DOI] [PubMed] [Google Scholar]
- 4.Reuben DB, Kremen S, Maust DT. Dementia Prevention and Treatment: A Narrative Review. JAMA Intern Med. 2024;184:563–72. [DOI] [PubMed] [Google Scholar]
- 5.Chen H, Cortese M, Flores-Torres MH, Tessier AJ, Wang DD, Kang JH, et al. Dietary Patterns and Indicators of Cognitive Function. JAMA Neurol. 2026;83(4):382–91. [DOI] [PMC free article] [PubMed]
- 6.Klein J. Membrane breakdown in acute and chronic neurodegeneration: focus on choline-containing phospholipids. J Neural Transm. 2000;107:1027–63. [DOI] [PubMed] [Google Scholar]
- 7.Bekdash RA. Neuroprotective effects of choline and other methyl donors. Nutrients. 2019;11:2995. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Kenny TC, Scharenberg S, Abu-Remaileh M, Birsoy K. Cellular and organismal function of choline metabolism. Nat Metab. 2025;7:35–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Shahraki S, Esmaeilpour K, Shabani M, Sepehri G, Rajizadeh MA, Maneshian M, et al. Choline chloride modulates learning, memory, and synaptic plasticity impairments in maternally separated adolescent male rats. Int J Dev Neurosci. 2022;82:19–38. [DOI] [PubMed] [Google Scholar]
- 10.Judd JM, Jasbi P, Winslow W, Serrano GE, Beach TG, Klein-Seetharaman J, et al. Inflammation and the pathological progression of Alzheimer’s disease are associated with low circulating choline levels. Acta Neuropathol. 2023;146:565–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Liu L, Qiao S, Zhuang L, Xu S, Chen L, Lai Q, et al. Choline Intake Correlates with Cognitive Performance among Elder Adults in the United States. Behav Neurol. 2021;2021:2962245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Huang F, Guan F, Jia X, Zhang J, Su C, Du W, et al. Dietary choline intake is beneficial for cognitive function and delays cognitive decline: a 22-year large-scale prospective cohort study from China Health and Nutrition Survey. Nutrients. 2024;16:2845. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Yuan J, Liu X, Liu C, Ang AF, Massaro J, Devine SA, et al. Is dietary choline intake related to dementia and Alzheimer’s disease risks? Results from the Framingham Heart Study. Am J Clin Nutr. 2022;116:1201–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Office of Dietary Supplements - Choline. [cited 2025 Sept 26]. Available from: https://ods.od.nih.gov/factsheets/Choline-HealthProfessional/?uid=fe9d8f204d12ds16.
- 15.Niu Y-Y, Yan H-Y, Zhong J-F, Diao Z-Q, Li J, Li C-P, et al. Association of dietary choline intake with incidence of dementia, Alzheimer disease, and mild cognitive impairment: a large population-based prospective cohort study. Am J Clin Nutr. 2025;121:5–13. [DOI] [PubMed] [Google Scholar]
- 16.Karosas T, Wallace TC, Li M, Pan Y, Agarwal P, Bennett DA, et al. Dietary choline intake and risk of Alzheimer’s dementia in older adults. J Nutr. 2025;155:2322–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Organ CL, Otsuka H, Bhushan S, Wang Z, Bradley J, Trivedi R, et al. Choline diet and its gut microbe–derived metabolite, trimethylamine N-oxide, exacerbate pressure overload–induced heart failure. Circ Heart Fail. 2016;9:e002314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Romano KA, Vivas EI, Amador-Noguez D, Rey FE. Intestinal Microbiota Composition Modulates Choline Bioavailability from Diet and Accumulation of the Proatherogenic Metabolite Trimethylamine-N-Oxide. mBio. 2015;6: 10.1128/mbio.02481-14. [DOI] [PMC free article] [PubMed]
- 19.Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JWR, Weir DR. Cohort profile: the Health and Retirement Study (HRS). Int J Epidemiol. 2014;43:576–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Yuan C, Spiegelman D, Rimm EB, Rosner BA, Stampfer MJ, Barnett JB, et al. Relative validity of nutrient intakes assessed by questionnaire, 24-hour recalls, and diet records as compared with urinary recovery and plasma concentration biomarkers: findings for women. Am J Epidemiol. 2018;187:1051–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chen H, Ding Y, Dhana K, Agarwal P, Beck T, Rajan KB, et al. Sweetened beverages and incident all-cause dementia among older adults. JAMA Psychiat. 2025;82:801–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Al-Shaar L, Yuan C, Rosner B, Dean SB, Ivey KL, Clowry CM, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire in men assessed by multiple methods. Am J Epidemiol. 2021;190:1122–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Yuan C, Spiegelman D, Rimm EB, Rosner BA, Stampfer MJ, Barnett JB, et al. Validity of a dietary questionnaire assessed by comparison with multiple weighed dietary records or 24-hour recalls. Am J Epidemiol. 2017;185:570–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Willett W. Nutritional epidemiology. OUP USA; 2013. p. 547. [Google Scholar]
- 25.Crimmins EM, Kim JK, Langa KM, Weir DR. Assessment of cognition using surveys and neuropsychological assessment: the Health and Retirement Study and the Aging, Demographics, and Memory Study. J Gerontol B Psychol Sci Soc Sci. 2011;66(Suppl 1):i162-171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Liu Y-H, Gao X, Na M, Kris-Etherton PM, Mitchell DC, Jensen GL. Dietary Pattern, Diet Quality, and Dementia: A Systematic Review and Meta-Analysis of Prospective Cohort Studies. J Alzheimers Dis. 2020;78:151–68. [DOI] [PubMed] [Google Scholar]
- 27.Chiuve SE, Fung TT, Rimm EB, Hu FB, McCullough ML, Wang M, et al. Alternative Dietary Indices Both Strongly Predict Risk of Chronic Disease. J Nutr. 2012;142:1009–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Chen H, Wu M, Li T, Yuan C. Dietary Choline Intake and Incident Dementia: A Dose–Response Meta-Analysis of Prospective Cohort Studies. PROSPERO 2025 CRD420251151642. Available from: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251151642.
- 29.Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol. 1992;135:1301–9. [DOI] [PubMed] [Google Scholar]
- 30.DerSimonian R, Kacker R. Random-effects model for meta-analysis of clinical trials: an update. Contemp Clin Trials. 2007;28:105–14. [DOI] [PubMed] [Google Scholar]
- 31.Ylilauri MP, Voutilainen S, Lönnroos E, Virtanen HE, Tuomainen T-P, Salonen JT, et al. Associations of dietary choline intake with risk of incident dementia and with cognitive performance: the Kuopio Ischaemic Heart Disease Risk Factor Study. Am J Clin Nutr. 2019;110:1416–23. [DOI] [PubMed] [Google Scholar]
- 32.Chartampila E, Elayouby KS, Leary P, LaFrancois JJ, Alcantara-Gonzalez D, Jain S, Gerencer K, Botterill JJ, Ginsberg SD, Scharfman HE. Choline supplementation in early life improves and low levels of choline can impair outcomes in a mouse model of Alzheimer’s disease. Slutsky I, Huguenard JR, editors. eLife; 2024;12:RP89889. [DOI] [PMC free article] [PubMed]
- 33.Spence MW. Sphingomyelin biosynthesis and catabolism. In: Phosphatidylcholine metabolism. CRC Press; 1989. [Google Scholar]
- 34.da Costa K-A, Gaffney CE, Fischer LM, Zeisel SH. Choline deficiency in mice and humans is associated with increased plasma homocysteine concentration after a methionine load. Am J Clin Nutr. 2005;81(2):440–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Chiuve SE, Giovannucci EL, Hankinson SE, Zeisel SH, Dougherty LW, Willett WC, et al. The association between betaine and choline intakes and the plasma concentrations of homocysteine in women1. Am J Clin Nutr. 2007;86:1073–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.von Maydell D, Wright SE, Pao P-C, Staab C, King O, Spitaleri A, Bonner JM, Liu L, Yu CJ, Chiu C-C, et al. ABCA7 variants impact phosphatidylcholine and mitochondria in neurons. Nature. 2025;647:462–71. [DOI] [PMC free article] [PubMed]
- 37.Yaqub A, Vojinovic D, Vernooij MW, Slagboom PE, Ghanbari M, Beekman M, et al. Plasma trimethylamine N-oxide (TMAO): associations with cognition, neuroimaging, and dementia. Alzheimers Res Ther. 2024;16:113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Institute of Medicine (US) Standing Committee on the Scientific Evaluation of Dietary Reference Intakes and its Panel on Folate, Other B Vitamins, and Choline. Dietary reference intakes for thiamin, riboflavin, niacin, vitamin B6, folate, vitamin B12, pantothenic acid, biotin, and choline [Internet]. Washington (DC): National Academies Press (US); 1998. [PubMed] [Google Scholar]
- 39.Chinese Nutrition Society. Dietary reference intakes for China (2023). 2023rd ed. People’s Medical Publishing House.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data used for cohort analysis could be obtained from the website of HRS (hrs.isr.umich.edu/dataproducts/access-to-public-data). Data used for the meta-analysis were obtained from the individual publications, which are available from the respective publishers.