Association between high carbohydrate to dietary fiber ratio and risk of dementia in older adults: analysis from the UK biobank Authors: Eunji Lee, Jiyun Hwang, Dong Woo Kim, Byung Yoon Choi, Hee-Joon Bae, Seung Hyun Won, Young Ho Park, SangYun Kim Published: 2025 ABSTRACT AbstractBackgroundThe dietary carbohydrate-to-fiber ratio (CFR) provides a comprehensive measure of carbohydrate quality; however, its association with cognitive outcomes remains unexplored. Given the distinct health effects of different types of carbohydrates, understanding the impact of CFR on dementia risk and brain structure is valuable.MethodsThis prospective cohort study analyzed 107,785 UK Biobank participants aged ≥ 55 years who completed at least one 24-hour dietary recall between 2011 and 2012, with follow-up until April 2024. CFR was calculated as the ratio of daily carbohydrate intake to fiber intake. Total carbohydrate and dietary fiber intake were included as secondary exposures. CFR outliers were excluded using Tukey’s method (3.0 × interquartile range). The primary outcome was all-cause dementia incidence identified through linked hospital and death registry data. Secondary outcomes were hippocampal volume, entorhinal cortical thickness and frontal cortical thickness. Cox proportional hazards models estimated dementia risk, and multiple linear regression analyzed MRI markers. Models were adjusted for age, sex, education, BMI, physical activity, smoking, alcohol intake, hypertension, diabetes, and total energy intake. Interactions between dietary factors and polygenic risk score (PRS) were examined.ResultsDuring a median follow-up of 11.8 years, 2,000 dementia cases occurred, corresponding to incidence rates of 1.44 and 1.82 per 1,000 person-years in the lowest and highest CFR quartiles, respectively. A higher CFR was significantly associated with increased dementia risk (HR per 1 SD = 1.07, 95% CI: 1.02–1.12, p = 0.003). The association was slightly attenuated in individuals with increased genetic susceptibility (p for interaction = 0.02). Total carbohydrate intake was associated with dementia risk (HR per 1 SD = 1.15, 95% CI: 1.05–1.25, p = 0.001), whereas fiber intake was not (p = 0.99). An elevated CFR was also associated with reduced entorhinal cortical thickness (FDR p = 0.03) and with thinner frontal pole, medial orbitofrontal cortex, and rostral middle frontal cortex (all FDR p = 0.01) among 14,388 participants with MRI data.ConclusionsA higher CFR was associated with increased dementia risk, reduced entorhinal thickness, and thinner frontal cortical regions, suggesting that CFR may serve as a practical indicator of carbohydrate quality with potential implications for dietary and public health approaches to cognitive aging.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12883-025-04583-w.Keywords: Dementia, Older adults, Carbohydrate-to-fiber ratio, Brain MRI, UK biobank Background Dementia is one of the leading causes of health and social burden in aging societies, affecting an estimated 57 million people worldwide [1]. Its incidence rises steeply with advancing age [2], and recent longitudinal analyses suggest that roughly 40% of adults may develop dementia over the course of their lifetime [3]. Although genetic factors play a substantial role in determining individual susceptibility, the onset and progression of dementia are also shaped by long-term lifestyle exposures [4]. Among these, dietary habits—behaviors repeated daily across the lifespan—may exert cumulative effects on brain function through metabolic, inflammatory, and vascular pathways. These pathways may involve alterations in insulin sensitivity, and glycemic variability, which have been linked to neurodegenerative processes. Carbohydrates are a major component of the modern diet [5] and can be classified into digestible and non-digestible forms [6, 7]. Dietary fiber, a classic example of non-digestible carbohydrates, resists enzymatic digestion in the human gastrointestinal tract and benefits metabolic health [8, 9]. In contrast, digestible carbohydrates include simple sugars and complex carbohydrates, with refined carbohydrates—predominant in Western diets—are associated with increased risks of numerous diseases [10, 11]. In real-world dietary patterns, these carbohydrate types do not exist in isolation but are frequently consumed together in various food sources [12, 13]. This dietary complexity presents a challenge when evaluating the health effects of each. Therefore, simply analyzing fiber intake alone, without considering the overall carbohydrate composition, may lead to misleading interpretations. To assess the nutritional profile of carbohydrate intake, the carbohydrate-to-fiber ratio (CFR) has been introduced as a more comprehensive measure [11, 14, 15]. CFR allows for the differentiation between fiber intake that is driven by high overall carbohydrate consumption, and an independent increase in fiber intake. This study aimed to investigate the impact of CFR on cognitive outcomes, focusing on dementia incidence and brain magnetic resonance imaging (MRI) findings, in line with prior neuroimaging research [16] showing that dietary patterns are associated with structural brain differences. We also explored whether the impact of CFR on cognitive outcomes differs according to genetic susceptibility to dementia by analyzing its interaction with the polygenic risk score (PRS) for Alzheimer’s disease (AD). The novelty of this study lies in applying the CFR to dementia incidence in a large, prospective population cohort and in integrating structural MRI outcomes to assess potential neurobiological correlates. Methods Study design and participants This study was a prospective cohort analysis using data from the UK Biobank (UKB), a large-scale cohort study that recruited approximately 500,000 participants aged 37–73 years between 2006 and 2010. The baseline assessments included dietary, demographic, and health-related information. A detailed protocol is described elsewhere [17, 18]. Participants who completed at least one 24-hour dietary recall between 2011 and 2012 were included if they were White, aged ≥ 55 years, and had no prior diagnosis of dementia at the time of assessment, based on linked hospital inpatient, death registry, and primary care records. The age criterion of 55 years or older, as in previous studies [19, 20], was selected to include an age group with a higher risk of dementia, as its incidence is relatively low in younger populations. We restricted the analytic sample to White participants, who comprise the vast majority of the UKB cohort [21], to ensure adequate statistical power and to reduce the risk of unstable estimates stemming from small subgroup sizes. This approach is consistent with prior UKB studies [22, 23] analyzing dietary and lifestyle factors. Individuals with extreme values (as defined as < 800 or > 4200 kcal/day for males and < 600 or > 3500 kcal/day for females) [24] in total energy intake or self-reported atypical diets on the survey day, in which respondents indicated that their eating habits differed from usual, were excluded to reduce bias from extreme dietary behaviors. Participants with missing data on exposure, outcome, or key covariates were excluded from the analyses (listwise deletion), as illustrated in the flow diagram (Fig. 1). Selection of Study Participants in the UK Biobank. Selection of study participants in the UK Biobank. Among 283,118 participants with at least one dietary assessment. Additional exclusions were made for age under 55 years, non-White ethnicity, extreme energy intake, atypical diet, withdrawal, missing covariate data, and outliers in the carbohydrate-to-fiber ratio, resulting in 107,785 participants included in the final analysis Participants were followed from the date of dietary assessment until the first diagnosis of dementia, death, or the censoring date of April 30, 2024, whichever occurred first. The median follow-up duration was 11.8 years (1.27 million person-years in total). Exposure assessment Dietary intake was assessed using the Oxford WebQ, a tool designed for large population studies [25], which provided comprehensive data on habitual dietary patterns [26]. By collecting participant-reported 24–hour dietary data [24], this tool automatically estimates the intake of macro- and micronutrients. All 24-hour dietary recalls were completed between 2011 and 2012, preceding both the onset of dementia follow-up and MRI acquisition (which began in 2014). Participants with a recorded diagnosis of dementia at or before the dietary assessment were excluded. For participants who completed multiple recalls, the average intake across assessments was used such as previous studies [27, 28], whereas values from a single recall were retained to maximize sample size and representativeness. The Oxford WebQ has demonstrated good validity even when administered once, with improved precision and reproducibility when repeated [25, 26]. The primary exposure variable was CFR, which was calculated by dividing the daily carbohydrate intake (grams per day) by the daily fiber intake (grams per day). CFR was modeled on the raw scale. Dietary fiber was quantified using the Englyst method [29], which measures non-starch polysaccharides and excludes resistant starch. To identify and exclude outliers in the CFR, we applied Tukey’s method with a threshold of 3.0 times the interquartile range (IQR) in the primary analysis [30]. Total carbohydrate and fiber intake, components of the CFR, were considered secondary exposure variables. Outcome measures The primary outcome was the incidence of all-cause dementia, as defined by the UKB’s algorithmically determined outcomes [31]. This definition includes cases in which clinical records (any primary care data, hospital inpatient data and death register records) and ICD-10 codes (Additional File 1; Tabel S1) are collected. Participants with self-reported dementia diagnoses that lacked formal clinical confirmation were excluded from the analysis to improve diagnostic accuracy. For this cohort analysis, the cutoff date for capturing dementia incidence was April 2024. Participants were censored at the time of death or the end of follow-up, and death was not treated as a competing risk. MRI data was acquired as part of the UKB imaging extension using identical Siemens Skyra 3 T scanners equipped with 32-channel head coils across imaging centers. Harmonized acquisition protocols and standardized preprocessing and quality-control procedures were implemented by the UKB imaging team, as previously described [32, 33]. The key MRI markers included hippocampal volume (measured in mm³), entorhinal cortical thickness (measured in mm) and frontal cortical regions (measured in mm) —including the thickness of the superior frontal, rostral middle frontal, lateral orbitofrontal, medial orbitofrontal and frontal pole. These markers were included to evaluate the potential impact of dietary factors on brain structure, as intermediate indicators of dementia risk, as suggested by previous studies [16]. Covariates Covariates were selected on the basis of their potential to confound the association between dietary factors and dementia risk, and they were informed by prior literature and clinical relevance. Total energy intake, as measured in kcal, was included to account for overall dietary intake. Most covariates, including demographic, lifestyle, and clinical factors were assessed at baseline (2006–2010), preceding the dietary assessments. Demographic variables such as age, sex, and lifestyle factors, including smoking status (never, previous, or current), frequency of alcohol intake (daily or almost daily, 3–4 times per week, 1–2 times per week, 1–3 times per month, or never), total physical activity (categorized as low, medium, or high based on the International Physical Activity Questionnaire group [34]), body mass index (underweight (< 18.5), normal weight (18.5–24.9), overweight (25–29.9), or obese (≥ 30)), and typical sleep duration (< 7 h, 7–8 h, or > 8 h), were considered. Socioeconomic status was assessed using the Townsend deprivation index (lower, moderate, higher), and education level (university degree, vocational training, upper secondary, lower secondary or none of the qualifications) was considered. Medical history variables, such as hypertension and type 2 diabetes diagnosed by a doctor, were also included. Furthermore, genetic predisposition to dementia was addressed using the PRS for AD, which was precalculated and uploaded by the UKB [35]. The PRS was standardized to a z-score (mean = 0, SD = 1) before inclusion in the models. In the MRI analyses, total intracranial volume and mean cortical thickness were additionally used as covariates. Statistical analysis Baseline characteristics were summarized by CFR quartiles, with continuous variables presented as medians (IQRs) and categorical variables presented as counts and percentages. Group comparisons across quartiles were conducted using the Kruskal-Wallis test for continuous variables and the Chi-square test for categorical variables. The time-to-event was defined as the duration from the date of the first completed questionnaire to the date of dementia diagnosis, death, or last follow-up, whichever occurred first. Participants without an event by the end of follow-up were censored. In the primary time-to-event analyses, incident dementia was modeled using a Cox proportional hazards approach to estimate cause-specific hazards, in which dementia was the sole event of interest and deaths were treated as censoring events. Cox proportional hazard models were used to examine associations between dietary factors and incident dementia, adjusting for covariates including age, sex, hypertension status, type 2 diabetes status, education level, Townsend deprivation index, smoking status, drinking status, physical activity, body mass index, sleep duration, total energy intake, and PRS. The proportional hazards assumption was verified using Schoenfeld residuals and no major violations were identified. We evaluated potential non-linearity using restricted cubic splines, and no evidence of non-linearity was observed; therefore, no transformation (e.g., log-scale) was applied. The primary effect estimates were reported as hazard ratios per one–standard deviation (1 SD) increase in each dietary variable. For each Cox model, the SD of the exposure was calculated from the analytic dataset, and per-SD hazard ratios were obtained by scaling the model coefficient and its standard error by this SD prior to exponentiation. Because this transformation is not directly applicable when interaction terms are included, models with interaction terms were reported only on the original exposure scale. This procedure yields scale-independent, directly comparable effect estimates across dietary exposures. Survival distributions across CFR quartiles were also compared using the Kaplan-Meier method, and differences were evaluated using the log-rank test. Pairwise comparisons, with CFR quartile 1 as the reference group, were conducted without p value adjustments. Effect modification was assessed by using interaction terms between dietary factors and PRS in Cox models. Quartile-based analyses were performed only for descriptive comparison and interpretative purposes. Associations between dietary factors and MRI markers were evaluated using multiple regression analyses, adjusting for the same covariates as those used in the survival analyses. Additionally, total intracranial volume and mean cortical thickness were included as covariates when analyzing hippocampal volume and regional cortical thickness, respectively. Outliers were excluded using Tukey’s method (values greater than 3 × IQRs) in all analysis. This approach [36] is commonly used in nutritional epidemiology to accommodate the right-skewed distribution of dietary variables and to limit the influence of extreme observations. Statistical significance was defined as a two-sided p value < 0.05, and all the analyses were conducted using R version 4.4.1. The main analyses used the survival, survminer packages. Sensitivity analyses Several sensitivity analyses were performed to evaluate the robustness of the findings. First, to examine whether metabolic health conditions influenced the observed associations, we repeated the analyses after excluding participants with metabolic comorbidities. Individuals with hypertension, diabetes mellitus, or obesity (BMI ≥ 30 kg/m²) were classified as having metabolic comorbidities. Third, we assessed the influence of extreme dietary values by applying multiple outlier exclusion methods. Additional analyses were performed using the MedCouple (MC) boxplot method, a 1% trimming method, and models without outlier exclusion. Ethics approval pprovalThis study was conducted using de-identified data from the UKB, which was approved by the North West Multi-Centre Research Ethics Committee. The Institutional Review Board of Seoul National University Bundang Hospital reviewed the study and exempted it from further ethical approval (IRB No. X-2408-918-901). This research was conducted using the UKB Resource under Application Number 104,323. Results Characteristics of the study participants A total of 107,785 participants were included in the study based on eligibility criteria and after the exclusion of CFR outliers (Fig. 1). The demographics of the participants excluded due to outliers are shown in Supplementary Additional File 1: Table S2. As shown in Table S2, participants excluded by Tukey’s method tended to have lower educational attainment and a higher prevalence of smoking, heavier drinking patterns, and obesity compared with those included in the analysis. They also exhibited markedly lower fiber intake, resulting in unusually high CFR values. The participants included in the analysis were categorized into quartiles (Q1–Q4) according to their CFR levels (Table 1). Women were more prevalent in Q1, whereas men were slightly more represented in Q4. There were slight differences across CFR quartiles, with higher education and physical activity being more common in lower quartiles and smoking and hypertension showing a marginally higher prevalence in higher quartiles. Table 1Baseline characteristics by quartiles of Carbohydrate-to-Fiber ratioCharacteristicsOverall(n = 107785)Carbohydrate-to-fiber intake ratioP valueQ1 (9.22–10.95)(n = 26947)Q2 (12.10-13.15)(n = 26946)Q3 (14.24–15.52)(n = 26946)Q4 (17.33–21.07)(n = 26946)Age62 (59–65)62 (59–65)62 (59–65)62 (59–65)61 (58–65)< 0.01Female57,046 (52.9)17,483 (64.9)15,195 (56.4)13,412 (49.8)10,956 (40.7)< 0.01Hypertension history38,255 (35.5)9075 (33.7)9317 (34.6)9566 (35.5)10,297 (38.2)< 0.01Type 2 diabetes history6179 (5.7)1749 (6.5)1542 (5.7)1427 (5.3)1461 (5.4)< 0.01Education level< 0.01University degree60,532 (56.2)16,095 (59.7)16,187 (60.1)15,155 (56.2)13,095 (48.6)Vocational training11,086 (10.3)2405 (8.9)2504 (9.3)2786 (10.3)3391 (12.6)Upper secondary6195 (5.7)1548 (5.7)1570 (5.8)1588 (5.9)1489 (5.5)Lower secondary or none of the qualifications29,972 (27.8)6899 (25.6)6685 (24.8)7417 (27.5)8971 (33.3)Townsend deprivation index< 0.01Lower21,179 (19.6)5100 (18.9)5415 (20.1)5562 (20.6)5102 (18.9)Moderate64,937 (60.2)16,102 (59.8)16,399 (60.9)16,362 (60.7)16,074 (59.7)Higher21,669 (20.1)5745 (21.3)5132 (19.0)5022 (18.6)5770 (21.4)Smoking status< 0.01Never57,035 (52.9)14,334 (53.2)14,800 (54.9)14,624 (54.3)13,277 (49.3)Previous44,145 (41.0)11,313 (42.0)10,865 (40.3)10,766 (40.0)11,201 (41.6)Current6605 (6.1)1300 (4.8)1281 (4.8)1556 (5.8)2468 (9.2)Drinking status< 0.01Daily or almost daily28,872 (26.8)6885 (25.6)7326 (27.2)7264 (27.0)7397 (27.5)3 or 4 times per week27,059 (25.1)6772 (25.1)7027 (26.1)6870 (25.5)6390 (23.7)1 or 2 per week24,765 (23.0)6176 (22.9)6251 (23.2)6139 (22.8)6199 (23.0)1 to 3 times per month20,622 (19.1)5411 (20.1)4905 (18.2)5080 (18.9)5226 (19.4)Never6467 (6.0)1703 (6.3)1437 (5.3)1593 (5.9)1734 (6.4)Total physical activity< 0.01Low28,402 (26.4)6186 (23.0)6746 (25.0)7306 (27.1)8164 (30.3)Moderate41,099 (38.1)10,145 (37.6)10,459 (38.8)10,410 (38.6)10,085 (37.4)High38,284 (35.5)10,616 (39.4)9741 (36.2)9230 (34.3)8697 (32.3)Body mass index< 0.01Underweight (< 18.5)517 (0.5)191 (0.7)140 (0.5)92 (0.3)94 (0.3)Normal weight (18.5–24.9)37,616 (34.9)10,391 (38.6)10,147 (37.7)9351 (34.7)7727 (28.7)Overweight(25-29.9)46,841 (43.5)10,878 (40.4)11,473 (42.6)11,948 (44.3)12,542 (46.5)Obese (≥ 30)22,811 (21.2)5487 (20.4)5186 (19.2)5555 (20.6)6583 (24.4)Typical sleeptime< 0.01< 7 h23,636 (21.9)6055 (22.5)5632 (20.9)5681 (21.1)6268 (23.3)7–8 h76,125 (70.6)19,018 (70.6)19,456 (72.2)19,248 (71.4)18,403 (68.3)> 8 h8024 (7.4)1874 (7.0)1858 (6.9)2017 (7.5)2275 (8.4)Carbohydrate-to-fiber intake ratio13.69 (11.56–16.31)10.23 (9.22–10.95)12.63 (12.10-13.15)14.85 (14.24–15.52)18.72 (17.33–21.07)Total carbohydrate (g/day)246.46 (202.55-294.89)221.24 (179.87-266.67)245.64 (205.88-289.73)256.28 (213.76-303.13)263.13 (216.15-314.81)< 0.01Fiber intake (g/day)17.92 (14.22–22.11)22.09 (18.17–26.73)19.47 (16.25–22.98)17.24 (14.34–20.38)13.67 (10.89–16.65)< 0.01Total energy intake (kcal/day)1981.17 (1665.58-2339.30)1828.03 (1538.35-2174.06)1974.39 (1680.27-2307.73)2037.82 (1737.72-2388.32)2084.02 (1739.70–2469)< 0.01Polygenetic risk scores-0.11 (-0.66-0.61)-0.10 (-0.66-0.62)-0.11 (-0.65-0.62)-0.11 (-0.66-0.60)-0.11 (-0.66-0.61)0.59Dementia cases2000460482483575Incidence per 1000 person-years1.581.441.531.531.82< 0.01The values for categorical variables are given as counts (percentages); values for continuous variables are given as medians (interquartile ranges). Continuous variables were analyzed using the Kruskal-Wallis test, and categorical variables were analyzed using the Chi-square test Baseline characteristics by quartiles of Carbohydrate-to-Fiber ratio Overweight (25-29.9) Typical sleep time The values for categorical variables are given as counts (percentages); values for continuous variables are given as medians (interquartile ranges). Continuous variables were analyzed using the Kruskal-Wallis test, and categorical variables were analyzed using the Chi-square test Associations between dietary factors and dementia risk Carbohydrate-to-fiber ratio The survival analysis revealed a significant association between CFR and dementia risk. The Kaplan-Meier method revealed significant differences across CFR quartiles (log-rank test, p = 0.001; Fig. 2). Post hoc comparisons revealed a significant difference between Q1 and Q4, with individuals in Q4 having a greater risk of dementia (HR = 1.22, 95% CI: 1.07–1.38, p = 0.002). These findings highlight significant disparities in survival distributions across CFR quartiles, with the most pronounced differences between the lowest and highest quartiles. Kaplan-Meier Survival Curve by Carbohydrate-to-Fiber Ratio Quartile. The Kaplan-Meier survival curves illustrate the survival probabilities over time for the four strata based on carbohydrate-to-fiber ratio (CFR) quartile. Colored lines represent each CFR quartile. A log-rank test revealed a statistically significant difference in survival distributions among the quartiles (p = 0.0005) A significant association was observed between CFR and dementia risk. A higher CFR, indicative of lower dietary fiber relative to carbohydrate intake, was associated with an increased risk of dementia (HR per 1 SD = 1.07, 95% CI per 1 SD: 1.02–1.12, p = 0.003; Table 2; Fig. 3) A model including the interaction term between CFR and PRS on dementia risk was statistically significant (interaction term; HR = 0.99, 95% CI: 0.98–1.00; p = 0.02; Table 2). To further illustrate this interaction, stratified analyses across PRS tertiles showed a modest decreasing pattern in the CFR coefficients (Table S3), indicating that the CFR–dementia association may vary by genetic susceptibility. Table 2Hazard ratios (HRs) and 95% confidence intervals (CI) for dementia risk according to dietary factorsModelHR95% CIP valueHR per 1SD95% CI (per SD)P value[Carbohydrate-to-fiber intake ratio (CFR)] [Model 1; Without interaction] CFR1.021.01–1.030.0031.071.02–1.120.003 Polygenetic risk scores (PRSs)1.841.77–1.91< 0.011.821.75–1.90< 0.01 [Model 2; With interaction] CFR1.031.01–1.04< 0.011.101.05–1.160.0001 PRS2.161.88–2.49< 0.012.141.86–2.46< 0.01 CFR*PRS0.990.98-1.000.02[Fiber (g/day)] [Model 1; Without interaction] Fiber1.000.99–1.010.991.00030.95–1.050.99 PRS1.841.77–1.91< 0.011.821.75–1.89< 0.01 [Model 2; With interaction] Fiber1.000.99–1.011.001.00010.94–1.061.00 PRS1.831.64–2.04< 0.011.811.63–2.02< 0.01 Fiber*PRS1.001.00-1.010.95[Carbohydrates (g/day)] [Model 1; Without interaction] Carbohydrates1.0021.001–1.0030.0011.151.05–1.250.001 PRS1.831.76–1.90< 0.011.821.75–1.89< 0.01 [Model 2; With interaction] Carbohydrates1.0021.001–1.0030.0011.171.07–1.230.0005 PRS2.001.76–2.29< 0.012.001.74–2.27< 0.01 Carbohydrates*PRS1.0000.999-1.0000.17The values are hazard ratios (HRs) with 95% confidence intervals (CIs) and P values, estimated using Cox proportional hazards models. Models were adjusted for covariates, including age, sex, hypertension history, type 2 diabetes history, education level, Townsend deprivation index, smoking status, drinking status, total physical activity, body mass index, typical sleep time and total energy intake Hazard ratios (HRs) and 95% confidence intervals (CI) for dementia risk according to dietary factors The values are hazard ratios (HRs) with 95% confidence intervals (CIs) and P values, estimated using Cox proportional hazards models. Models were adjusted for covariates, including age, sex, hypertension history, type 2 diabetes history, education level, Townsend deprivation index, smoking status, drinking status, total physical activity, body mass index, typical sleep time and total energy intake Relationship Between Carbohydrate-to-Fiber Ratio and the Dementia Risk. The relationship between carbohydrate-to-fiber ratio (CFR) and the dementia risk. The solid line represents the estimated relative hazard, whereas the shaded gray area indicates the 95% confidence interval. The x-axis represents CFR, and the y-axis represents the relative hazard Fiber In the analysis of the relationship between dietary fiber intake and the risk of dementia, no remarkable association was observed (HR per 1 SD = 1.00, 95% CI: 0.95–1.05; p = 0.99; Table 2). The interaction effect between fiber intake and PRS was also not statistically significant (interaction term; HR = 1.00, 95% CI: 1.00–1.01; p = 0.95). Carbohydrate intake Total carbohydrate intake was associated with an increased risk of dementia (HR per 1 SD = 1.15, 95% CI: 1.05–1.25; p = 0.001; Table 2). The interaction effect between total carbohydrate intake and PRS was not significant (interaction term; HR = 1.000, 95% CI: 0.999–1.000; p = 0.17) for dementia risk. Associations between dietary factors and MRI markers The results of multiple regression analyses examining the associations between dietary factors and MRI markers—including hippocampal volume, entorhinal cortical thickness, and frontal cortical regions—are summarized in Supplementary Additional File 1: Table S4. These analyses were conducted among the 14,388 participants with available MRI data, and the demographic characteristics of this MRI subsample are presented separately in Supplementary Additional File 1: Table S5. A significant negative correlation was found between CFR and entorhinal cortical thickness (coefficient = -0.0015, 95% CI: -0.0027 to -0.0023; FDR p = 0.03), suggesting that a higher CFR is linked to reduced entorhinal cortical thickness. Higher CFR was also associated with thinner frontal cortical regions, including the frontal pole (coefficient = − 0.02, 95% CI: − 0.03 to − 0.01; FDR p = 0.01), medial orbitofrontal cortex (coefficient = − 0.02, 95% CI: − 0.02 to − 0.01; FDR p = 0.01), and rostral middle frontal cortex (coefficient = − 0.01, 95% CI: − 0.02 to − 0.004; FDR p = 0.01). In contrast, dietary fiber intake showed significant positive associations with multiple frontal cortical regions, whereas no significant associations were observed for hippocampal volume or entorhinal cortical thickness. Higher fiber intake was associated with greater cortical thickness in the frontal pole (coefficient = 0.01, 95% CI: 0.003 to 0.02; FDR p = 0.03), lateral orbitofrontal cortex (coefficient = 0.007, 95% CI: 0.001 to 0.01; FDR p = 0.03), medial orbitofrontal cortex (coefficient = 0.01, 95% CI: 0.01 to 0.02; FDR p = 0.003), rostral middle frontal cortex (coefficient = 0.01, 95% CI: 0.002 to 0.01; FDR p = 0.03), and superior frontal cortex (coefficient = 0.01, 95% CI: 0.001 to 0.01; FDR p = 0.05). No significant associations were detected between total carbohydrate intake and any MRI markers. Sensitivity analysis First, after excluding participants with metabolic comorbidities—defined as hypertension, diabetes mellitus, or obesity—the principal findings remained broadly consistent with the main results (Supplementary File 1: Table S6). The direction and magnitude of the hazard ratio estimate for CFR, total carbohydrate intake, and dietary fiber intake did not materially change, although statistical significance was attenuated for several variables. Interaction patterns between dietary exposures and PRS similarly paralleled those observed in the main analysis, and the fluctuations in significance were not substantial enough to alter the overall conclusions. These findings suggest that metabolic health status does not account for the observed associations between dietary factors and dementia risk. Second, sensitivity analyses applying multiple outlier-handling approaches were done, including the MC method, 1% trimming (Supplementary File 1: Figures S3–S4). The associations between CFR and dementia risk remained statistically significant under the MC and 1% trimming methods, whereas statistical significance diminished in the untrimmed analyses, although the direction of association was preserved. The CFR–PRS interaction largely showed similar patterns across trimming strategies, with reduced significance in the untrimmed dataset. In the MRI analyses (Supplementary File 1: Tables S7–S9), the association between CFR and entorhinal cortical thickness remained consistently significant across all models. The associations of dietary fiber intake with hippocampal volume and entorhinal cortical thickness showed modest variation in statistical significance depending on the outlier-handling method, but none of these differences were substantial enough to alter the interpretation of the results. Likewise, the associations between CFR or dietary fiber intake and frontal cortical regions showed some fluctuation in significance across methods, but the overall interpretation remained unchanged. In contrast, total carbohydrate intake showed no measurable association with any of the MRI markers across all sensitivity models. Discussion In this large cohort study, we found that a lower CFR was associated with a reduced risk of dementia. While higher carbohydrate intake was linked to increased risk, dietary fiber intake was not significantly related to dementia risk. The observed association between CFR and dementia risk was comparable to other lifestyle risk factors previously reported in UKB, such as obesity (HR 1.14), current smoking (HR 1.23) [37]. Dietary fiber supports gut microbiota and short-chain fatty acid production, enhancing gut-brain communication by modulating immune responses, reducing inflammation, and supporting neurotransmitter production [38, 39]. It also increases brain-derived neurotrophic factor levels, improving neuroplasticity and cognitive performance [39, 40]. Refined carbohydrates and free sugars are rapidly digested and absorbed, leading to sharp increases in blood glucose levels insulin secretion [41, 42]. Chronic hyperinsulinemia and the subsequent development of insulin resistance in the brain may impair the clearance of beta-amyloid, a key pathological feature of AD [43, 44]. Excessive intake of these carbohydrates can alter the gut microbiota composition, promoting systemic inflammation, which in turn may compromise the integrity of the blood-brain barrier and contribute to neuro-inflammation [45, 46]. The long-term consumption of such unhealthy carbohydrates has been associated with neurocognitive deficits [47, 48]. Collectively, these findings support a potential pathway linking high CFR to dementia via insulin resistance and neuroinflammation, possibly mediated through gut–brain axis dysregulation. Unlike previous studies [49, 50] suggesting a protective effect of fiber, our findings revealed no significant association between the amount of fiber intake alone and dementia risk. This discrepancy may reflect the complexity of modern dietary patterns, as previously discussed, underscoring the importance of considering overall carbohydrate composition rather than fiber intake alone in relation to cognitive outcomes. Previous studies have shown that the physiological effects of dietary fiber differ in terms of solubility and viscosity [51–54]. However, because the UKB dietary data do not distinguish between fiber types or functional properties, we were unable to assess these differences; total fiber was therefore treated as a composite variable, which may partly explain the null association. A related limitation is that the UKB provides fiber values only under the Englyst [29] non-starch polysaccharide method, precluding reconstruction of AOAC [55] total dietary fiber and limiting the ability to characterize the full spectrum of fiber components. Even so, analyzing CFR allowed us to evaluate the balance between carbohydrate and fiber intake—a perspective not previously examined in relation to cognitive or neuroimaging outcomes—and future work incorporating both Englyst and AOAC metrics will be essential to more precisely delineate these relationships. In this context, CFR serves as a more comprehensive metric for evaluating the balance between beneficial and harmful carbohydrates in the diet [56]. Previous studies have employed CFR in the context of various health outcomes [57–59], with an optimal threshold suggested to be below 10:1. However, this cutoff has primarily been derived either from studies on metabolic and cardiovascular conditions [60] or from research that examined CFR values for individual food items rather than considering overall dietary intake [61]. Interestingly, the lowest CFR quartile, which presented the lowest dementia risk, had a median CFR of 10.23, closely aligning with the previously suggested optimal threshold of 10:1. These findings offer new and valuable insights into the potential relevance of CFR for cognitive health, supporting the idea that the recommended balance of carbohydrate intake warrants further exploration. In addition to CFR, lifestyle factors tend to interact consistently with inherent characteristics such as genetics. In this context, PRS serves as a powerful tool, as it quantifies genetic susceptibility to specific diseases [62] and offers valuable insights into the risk of complex disorders such as dementia. Recent studies suggest that various lifestyle factors, including smoking, physical activity, alcohol consumption, and diet, can interact with PRS to modulate dementia risk [63–65]. The relationship between PRS and diet yielded inconsistent findings [64, 66–68], indicating that dietary effects have not yet been firmly established. In our study, individuals with a lower PRS exhibited a stronger effect on CFR, suggesting that those with a higher genetic predisposition may be slightly less sensitive to the effects of a higher CFR. While the interaction was statistically significant, its effect size was modest. This underscores the importance of considering genetic susceptibility when evaluating the role of CFR in dementia risk. In addition, unlike previous studies [66, 69] that have examined overall dietary patterns, our study focused on a specific dietary factor. Carbohydrate quality or balance can also be assessed using established metrics such as the glycemic index (GI) and glycemic load (GL) [70]. These metrics directly capture postprandial glycemic and insulinemic responses, and recent studies [71, 72] have further shown that high-GI/GL dietary patterns are associated with greater risks of cognitive decline and incident dementia. Implementing GI/GL [73] in large-scale datasets, however, requires matching each food item to external GI tables, a process that introduces substantial assumptions and variability due to differences in food composition and preparation methods. In contrast, CFR used in the present study can be derived directly from existing nutrient data, offering greater reproducibility and practical feasibility. CFR also provides an intuitively interpretable summary of dietary patterns in terms of whether carbohydrate intake is accompanied by sufficient dietary fiber. Nevertheless, CFR does not capture postprandial glycemic responses per se and should be viewed as a complementary measure rather than a substitute for GI/GL. Future work comparing CFR and GI/GL within the same cohort would help clarify the extent to which these metrics convey overlapping or distinct mechanistic and predictive information. In our MRI analyses, higher CFR was associated with thinner entorhinal cortical thickness, whereas no significant association was observed with hippocampal volume. Because both regions are known to be vulnerable in the earlier phases of AD [74], this discrepancy may appear unexpected. However, rather than reflecting differences in underlying pathology, this pattern is more plausibly explained by differences in measurement sensitivity and variability between the two markers. Entorhinal cortical thickness represents a small, focal cortical structure and is therefore sensitive to subtle thinning [22], whereas hippocampal volume is a more variable metric that is strongly influenced by age, intracranial volume, vascular factors, lifestyle factors, and other nonspecific sources of variability [75, 76]. Given these characteristics, especially in a large population-based cohort such as the UKB, it is likely that relatively small CFR-related effects were more readily detectable in entorhinal cortical thickness, whereas the higher variability of hippocampal volume may have reduced statistical power to identify comparable associations. Beyond medial temporal structures, higher CFR was also associated with thinner cortex across several frontal regions, including the frontal pole, orbitofrontal cortex, and rostral middle frontal cortex. Prior neuroimaging studies [77, 78] of diet quality have also reported that associations in frontal and other association cortices are sometimes detected more readily than those in medial temporal regions, likely because they capture cumulative cardiometabolic influences rather than AD-specific neurodegeneration. Thus, the regional pattern observed in our study—entorhinal susceptibility with additional frontal involvement—is consistent with known pathways through which carbohydrate quality may influence structural brain integrity. The effect sizes observed in our analyses were modest. This is expected and consistent with previous large-scale neuroimaging–nutrition literature [16]. Given that diet exerts its effects through multiple long-term metabolic and inflammatory pathways—and that structural MRI measures are shaped by diverse genetic, vascular, and lifestyle factors—large effect sizes are not typically observed in population-based cohorts. Therefore, the magnitude of the associations identified here, although small, should be interpreted as biologically plausible and aligned with effect-size ranges reported in prior literature. This study examined the relationship between dietary patterns and cognitive outcomes. Diet is not isolated behavior but a composite lifestyle factor that co-occurs with physical activity, smoking, alcohol use, and socioeconomic conditions, forming a clustered behavioral profile that complicates causal interpretation. Reflecting this complexity, the recent expert consensus report [4] did not designate diet as an independent causal risk factor for dementia, noting the difficulty of separating dietary effects from the broader lifestyle and contextual factors with which they co-vary. Nevertheless, because diet represents a pervasive, lifelong exposure, its relevance to cognitive health remains substantial. Prior studies [79–81] generally report modest diet–dementia associations—typically within the range of HR 1.01–1.20 per unit or per standard deviation—largely due to substantial measurement error inherent in self-dietary recalls, and to residual confounding arising from correlated lifestyle behaviors that cannot be fully removed in observational settings. Although such structural limitations persist, the CFR in this study showed consistent associations across MRI markers and dementia incidence, suggesting that the findings are unlikely to be explained solely by confounding artifacts and that CFR may capture aspects of carbohydrate quality not reflected by fiber alone. Limitation There are several limitations to this study. First, although we adjusted for major lifestyle and nutritional factors, residual confounding by unmeasured dietary indices may persist. Second, the MRI analysis was cross-sectional rather than longitudinal, restricting the ability to determine the directionality of the association between dietary intake and brain structural changes. Third, as this study used data from the UKB, several constraints inherent to this resource should be acknowledged. Because UKB is an observational cohort, reverse causation cannot be fully ruled out, especially given the long preclinical phase of dementia. Dietary intake was assessed using one or more 24-hour recalls, which may introduce intra-individual variability. Moreover, the UKB comprises volunteers who are generally healthier and of higher socioeconomic status than the general population, introducing selection bias that may limit generalizability. Finally, dementia subtypes were not differentiated in the present analysis, preventing the assessment of whether the associations between CFR and risk differ across Alzheimer’s disease, vascular dementia, or other subtypes. Future studies should consider interventional designs, such as randomized controlled trials, to establish causality. Conclusion Despite these limitations, this is the first large-scale population-based investigation to examine the relationship between CFR and both dementia incidence and brain structural markers. This investigation demonstrated that maintaining a balanced intake of carbohydrates and fiber may be important for cognitive health, especially among individuals with lower genetic vulnerability. These findings emphasize the role of dietary balance and suggest that integrating CFR with genetic risk profiles may inform future precision nutrition strategies. Supplementary Information Supplementary Material 1. Acknowledgements The authors thank the Division of Statistics in Medical Research Collaborating Center at Seoul National University Bundang Hospital for the statistical analyses.This study was supported by grant No. 18-2023-0004 from the SNUBH Research Fund and by a National Research Foundation of Korea grant funded by the Korean government (Ministry of Science and ICT; No. RS-2024-00461936). Abbreviations Carbohydrate-to-fiber ratio Magnetic resonance imaging Polygenic risk score Alzheimer’s disease UK Biobank Interquartile range International classification of diseases Hazard ratios Confidence intervals MedCouple Authors’ contributions Lee, Park, and SY Kim contributed to the study concept and design. Data acquisition, analysis, and interpretation were performed by Lee, Hwang, Won, and Park. Lee, Won, and Park drafted the manuscript. Critical revision for important intellectual content was conducted by DW Kim, Choi, Bae, Won, Park, and SY Kim. All authors read and approved the final manuscript. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data availability The data used in this study are available from the UK Biobank upon reasonable request and approval through their data access process (https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access). Declarations Ethics approval and consent to participate This study was conducted using data from the UK Biobank. The UK Biobank obtained informed consent from all participants. Use of the data in this study was approved by the Institutional Review Board of Seoul National University Bundang Hospital (IRB No. B-2304-819-104). All procedures involving human participants were conducted in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments (last revised in 2013). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Clinical trial number Not applicable. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Seung Hyun Won, Email: shwon0213@gmail.com. Young Ho Park, Email: kumimesy@snubh.org. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The data used in this study are available from the UK Biobank upon reasonable request and approval through their data access process (https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access).