Different types of sweetened coffee consumption, genetic predictor of gut microbe, and the risk of metabolic dysfunction-associated steatotic liver disease

Abstract

Background & Aims

Associations between different types of coffee consumption and metabolic dysfunction-associated steatotic liver disease (MASLD) remained inconsistent. We aimed to assess the longitudinal associations of coffee consumption (including unsweetened, sugar-sweetened, artificially sweetened, caffeinated and decaffeinated coffee) with the risk of MASLD, while also exploring the potential coffee type-gut microbial abundance interactions.

Methods

The present cohort study included 185,437 participants free of MASLD at baseline in the UK Biobank. Dietary consumption of different types of coffee was collected through 24-hour dietary recall questionnaires. Incident cases of MASLD were ascertained through linked hospital records and death registries. Genetic risk scores (GRSs) of the relative abundance of intestinal microbiota and the risk of MASLD were calculated using 19 and 5 single nucleotide polymorphisms, respectively. Cox proportional hazards regression model was employed to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations.

Results

During a median follow-up of 10.49 years, we documented 1,536 MASLD cases. Compared with non-consumers, individuals consuming more than 2.5 servings per day of unsweetened coffee, caffeinated coffee, or a combination of both had a lower risk of MASLD, with adjusted HRs (CIs) of 0.70 (0.60–0.82), 0.78 (0.67–0.91), and 0.69 (0.58–0.82), respectively. No significant association was found between sugar-sweetened and artificially sweetened coffee consumption and the risk of MASLD. These associations were consistent across genetic risk levels of abundance of intestinal microbiota and MASLD itself, with no significant interaction observed (all P values for interactions ≥ 0.05).

Conclusion

Higher intake of unsweetened coffee, particularly the caffeinated variety, was significantly associated with a reduced risk of MASLD, irrespective of genetic predisposition related to the abundance of intestinal microbiota or MASLD itself.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12937-026-01289-8.

Introduction

Metabolic dysfunction-associated steatotic liver disease (MASLD), recently renamed from non-alcoholic fatty liver disease, has emerged as the predominant chronic hepatic disorder worldwide 1. Metabolic comorbidities, such as obesity, type 2 diabetes, hyperlipidemia, and hypertension are linked to MASLD 2–4. MASLD has also been found to be associated with more fatal outcomes, including cardiovascular disease (CVD), cancer, and mortality 5–7. With shifts in dietary patterns over recent years, the global prevalence of MASLD has risen rapidly, now estimated at 32.4% [8]. Although the drug resmetirom is approved for the treatment of non-cirrhotic MASLD with moderate to advanced liver fibrosis, it should be used in combination with diet and exercise 9.

Coffee ranks as one of the most widely consumed drinks globally 10. Its beneficial effects have been demonstrated across various health conditions, including hypertension, type 2 diabetes, coronary heart disease, heart failure, arrhythmia, stroke, cardiovascular disease, Parkinson’s disease, cancer, and mortality 11–14. Nonetheless, a limited number of epidemiological studies have concentrated on the links between coffee consumption and the risk of incident MASLD, reporting inconsistent findings. Some studies indicate a positive association 15 while others show no association 16, 17. Additionally, previous studies have highlighted potential differences in the health effects of various coffee subtypes. As an example, both unsweetened and sugar-sweetened varieties of coffee are linked to lower all-cause mortality risks, whereas artificially sweetened coffee does not demonstrate this benefit 11. Similarly, caffeine has been shown to reduce the risk of Parkinson’s disease and type 2 diabetes 18. These variations in health effects may stem from differences in coffee additives, such as sugar, artificial sweeteners, and caffeine content. It remains unclear whether the association between coffee consumption and MASLD varies with coffee additives. Additionally, previous studies have found that the biological effects of added sugar and caffeine are regulated by intestinal flora 19–21. While both dietary and genetic elements influence the onset of MASLD, it remains unknown whether genetically predicted intestinal microbiota abundances or genetic susceptibility to MASLD may modify these associations.

In this study, our objective was to assess the longitudinal associations between the consumption of unsweetened, sugar-sweetened, and artificially sweetened coffee and the risk of MASLD among adults from the UK Biobank study. Then it further probes into the influence of caffeine within each sweetener classification, exploring whether caffeine modulates the beneficial effects of coffee in combination with sugar or sweeteners. Additionally, we investigated whether these associations differ among participants with varying levels of genetically predicted abundances of intestinal microbiota or MASLD risk itself. This study aims to provide a broader insight into the impact of various coffee consumption on MASLD risk, ultimately informing dietary recommendations.

Materials and methods

Study design

Our study utilized data from the UK Biobank, a large prospective cohort study involving over 500,000 individuals aged 37 to 73, collected from 22 research facilities throughout the United Kingdom (England, Wales, and Scotland) during 2006–2010. All participants completed baseline questionnaires that included anthropometric evaluations and reports on medical conditions. In brief, information on lifestyle factors, physical measurements, medical records, and biological samples was collected. Detailed descriptions of the UK Biobank are available elsewhere 22. All participants provided written informed consent prior to enrollment.

In this study, individuals who had filled out the 24-hour dietary recall online questionnaire at least once qualified for participation. Overall, 210,957 participants were initially included. We excluded those with incomplete genetic data (n = 4,499) and discordance between genetic sex and self-reported gender (n = 154), those with an implausible energy intake (defined as 0 or > 20 MJ/day for males, 0 or > 18 MJ/day for females) 23, 24 (n = 508), those not of European descent (n = 8,525) because the genetic instrument used was constructed in White participants. We further excluded participants with prevalent MASLD or other liver diseases (e.g., cirrhosis, liver failure, hepatocellular carcinoma, liver transplant status, alcoholic liver disease, viral hepatitis, etc.), or alcohol/drug abuse at or before baseline (n = 4,212) (details of these diseases were provided in the Supplementary Table 1). In addition, 513 participants were excluded due to loss to follow-up. Finally, we excluded participants who consumed more than one type of coffee (unsweetened, sugar-sweetened, and artificially sweetened), resulting in 7,109 exclusions. This left 185,437 participants eligible for inclusion in the study, comprising 43,093 non-coffee consumers, 104,494 unsweetened coffee consumers, 26,141 sugar-sweetened coffee consumers, and 11,709 artificially sweetened coffee consumers (Supplementary Fig. 1).

Assessment of coffee consumption

Dietary information was gathered through an online 24-hour dietary recall questionnaire known as the Oxford WebQ, which evaluates the variety and amounts of food intake, and has been validated in detail elsewhere 25. Over a year, participants were asked to fill out the survey five times to reflect the seasonal changes in their diet from April 2009 to June 2012.

During each 24-hour dietary recall, participants indicated the quantity of coffee drinks they had ingested in the preceding 24 h. Participants reported the precise number of cups of coffee consumed per day according to the instructions. Participants were also asked to specify if they added sugar or artificial sweeteners (any brand) and the quantity of teaspoons included. In every dietary recall, there were four distinct categories of coffee consumers: non-consumers, unsweetened coffee consumers, sugar-sweetened coffee consumers, and artificially sweetened coffee consumers. Given that participants could complete the 24-hour dietary recall up to five times, individuals consuming coffee at any given dietary recall were categorized as coffee consumers, while the rest were deemed non-consumers. Participants who consumed identical coffee types (unsweetened, sugar-sweetened, or artificially sweetened) were categorized in dietary records as a sole consumer, while others were grouped as overlapping consumers and were excluded. Coffee consumer categories are shown in Supplementary Table 2. Finally, we excluded 7,109 overlapped consumers (as detailed previously). To further explore whether the beneficial effects of coffee are due to caffeine, we classified the participants’ coffee consumption into two categories: caffeinated and decaffeinated, within each type of sweetened coffee. To reduce confusion about size, the online dietary questionnaire offered detailed guidelines for standard serving sizes, such as mugs or cups. Participants who indicated consuming over 10 drinks from all types of coffee were requested to confirm their answers. A single beverage had an approximate volume of 250 mL.

Assessment of outcomes

MASLD, identified as either hospitalization or mortality from MASLD or MASH (metabolic dysfunction-associated steatohepatitis), was the main result. This result was pinpointed using hospital inpatient records, categorized as primary or secondary diagnoses (UKB data-field 41270), along with death records, encompassing both underlying or contributory causes of death (UKB data-fields 40001 and 40002). Using ICD-10 and the Expert Panel Consensus Statement 26, MASLD was defined as ICD-10 code K76.0 (fatty [change of] liver, not elsewhere classified) and K75.8 (other specified inflammatory liver diseases) (Supplementary Table 1). The tracking of participants spanned from their visit to the evaluation facility to their demise, diagnosis, or final follow-up date (30th September 2021 for England, 28th February 2018 for Wales, and 31st July 2021 for Scotland), whichever came first. During the follow-up, each participant was censored at the first event of MASLD or death, whichever occurred first. In the absence of incidents or fatalities, participants were monitored until the end of follow-up.

Assessment of covariates

We used the baseline questionnaire to assess the following potential confounders: age, sex, body mass index (BMI), Townsend deprivation index (TDI), education level (college/university degree, other degrees or unknow), smoking status (current, former, never or unknow), alcohol status (current, former, never or unknow), physical activity (low, moderate, high or unknow), total energy intake (< median or ≥ median), total sugar intake (grams per day), healthy diet score (0–3, 4, 5, 6, or > 7), tea intake (drinks per day), cancer (yes or no), CVD (yes or no), diabetes (yes or no), hypertension (yes or no), and hyperlipidaemia (yes or no). BMI was calculated as weight (kg)/height squared (m²). The TDI served as a measure of socioeconomic standing, with negative TDI values signifying comparative wealth. Education level was based on self-report of highest qualification achieved and classified as college/university degree (College or University degree) and other degrees (A levels/AS levels or equivalent, O levels/GCSEs or equivalent, CSEs or equivalent, NVQ or HND or HNC or equivalent, Other professional qualifications or none of the above). Based on the guidelines of the International Physical Activity Questionnaire (IPAQ) 27, participants were classified into groups of low, moderate, and high activity levels according to categorical standards, which are shown in Supplementary Methods. We used multiple imputation to impute missing values for education, smoking, alcohol consumption, and physical activity for subsequent analyses to reduce the bias that might be introduced. According to a previous study 28, healthy diet score was calculated based on the adequate consumption of fruits (≥ 3 servings/ day), vegetables (≥ 3 servings/day), whole grains (≥ 3 servings/day), fish (shell) (≥ 2 servings/week), dairy products (≥ 2 servings/day), and vegetable oils (≥ 2 servings/day); and reduced or no consumption of refined grains (≤ 2 servings/day), processed meats (≤ 1 serving/week), unprocessed red meats (≤ 2 servings/week), industrial trans-fat (≤ 2 servings/week), sugar-sweetened beverages and sodium. The details are provided in Supplementary Methods and Supplementary Table 4. Participants will receive one point for each of the 12 dietary goals they achieve, with a maximum possible score of 12 points. A higher score indicates a healthier diet. Hypertension was defined as a systolic blood pressure ≥ 140 mmHg and/or a diastolic blood pressure ≥ 90 mmHg, self-reported hypertension, or antihypertensive medication use 29, 30. Diabetes was defined as fasting glucose ≥ 7.0 mmol/L, glycated hemoglobin (HbA1c) ≥ 48.0 mmol/L, self-reported diabetes, or a history of diabetes medication use 29. Hyperlipidemia was defined as total cholesterol ≥ 5.17 mmol/L, triglycerides ≥ 1.70 mmol/L, low-density lipoprotein cholesterol ≥ 3.37mmol/L, self-reported hyperlipidemia, or a history of medication use for hyperlipidemia 31.

Calculation of the genetic risk score for fatty liver and gut microbial abundance

According to previous studies, the genetic risk score of gut microbial abundance (GMA-GRS) 32, 33 and fatty liver-related GRS (FLD-GRS) 34–36 were constructed based on distinct single-nucleotide polymorphisms (SNPs) identified from GWAS, respectively. For GMA-GRS, we pruned SNPs based on R² < 0.2 and a 500 kb clustering window to find the SNP with the lowest P-value for each cluster, and 1 SNP(rs35866622) was excluded due to a high R². Finally, a total of 19 and 5 SNPs were used to construct the GMA- and FLD-GRS. Related information for each SNP was shown in Supplementary Table 3. Based on the selected SNPs, the GRSs were calculated separately as follows: GRS = (β1 × SNP1 + β2 × SNP2 + ... + βn × SNPn) * (n/sum of the β coefficients), where SNPn is the risk allele number of each SNP. A higher GMA-GRS indicates a higher genetically determined gut microbial abundance. A higher FLD-GRS score indicated a higher genetic predisposition to MASLD. For further analysis, we divided participants into two groups (low or high GRS) according to the median of GRS. Detailed information can be found in the Supplementary Methods. Detailed information regarding genotyping, quality control, and imputation methodologies is available on the UK Biobank website (http://biobank.ctsu.ox.ac.uk/).

Statistical analysis

Baseline characteristics are presented as mean (standard deviation, SD) for continuous variables and number (percentage) for categorical variables according to the type of coffee consumer (non-coffee consumers, unsweetened coffee consumers, sugar-sweetened coffee consumers, and artificially sweetened coffee consumers). Coffee consumers were categorized into three groups based on daily intake: > 0 to 1.5 drinks/day, > 1.5 to 2.5 drinks/day, and > 2.5 drinks/day. Taking non-coffee consumers as reference, Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% confidence intervals (95% CIs) for the prospective associations of different types of coffee consumption with MASLD. In addition, we examined the dose-response associations between coffee consumption and the risks of MASLD using the restricted cubic spline analysis. The Schoenfeld residual method was used to check the proportional hazard assumption, and no violation was found. Four models were developed for analysis. The crude model was the model without any adjustment. Model 1 was adjusted for age (continuous), sex (male or female), and BMI (continuous, kg/m²). Model 2 was further adjusted for Townsend deprivation index (continuous, years), education level (college/university degree, other degrees or unknow), smoking status (current, former, never or unknow), alcohol status (current, former, never or unknow), physical activity (low, moderate, high or unknow), healthy diet score (0–3, 4, 5, 6, or > 7), total energy (< median or ≥ median), total sugar (continuous, g/d), and tea (continuous, drinks/d) based on Model 1 .Model 3 was further adjusted for hypertension (yes or no), diabetes (yes or no), cancer (yes or no), CVD (yes or no), and hyperlipidaemia (yes or no) based on Model 2. Moreover, in the analyses related to GRSs (e.g., subgroup analyses according to GRS categories), the first 10 principal components of ancestry and the genotype measurement batch were further adjusted.

To explore the potential modifying effect of GMA-GRS and FLD-GRS on the associations between coffee consumption and incident MASLD, analyses were stratified by GRS categories (< median and ≥ median). The interactions between different types of coffee consumption and GRSs were tested by adding multiplicative interaction terms into the fully adjusted model. Among participants with any coffee consumption, joint associations of the different types of coffee consumption and GRSs with MASLD were assessed using combined variables with 2 × 4 categories. The combination of high GRS and the lowest coffee consumption was the reference group.

We conducted several secondary analyses to test the robustness of our results. To assess whether the associations of different types of coffee consumption with the risks of MASLD differed by population subgroups, we stratified the participants by potential effect modifiers including sex (male or female), age (< 60 or ≥ 60 years), BMI (≤ 24.9 or > 24.9), education level (college/university degree, other degrees), Townsend deprivation index (< average or ≥ average), smoking status (never or ever), and alcohol intake (< average or ≥ average). In addition, we did the following sensitivity analyses. First, we excluded participants who developed MASLD during the first 2 years of follow-up. Second, we used multiple imputation by chained equations to impute other missing data with 5 imputations and repeated the main analysis using the complete data set to test the influence of missing variables. Third, we excluded participants with excessive alcohol (> 20 g/d in females or > 30 g/d in males) to further control for the health effects associated with drinking behaviors. Fourth, we excluded participants who had chronic diseases (including CVD, cancer, and diabetes) at baseline to reduce confounding bias and to clarify independent associations between exposure and outcome. Fifth, we evaluated the competing risk of MASLD deaths using Fine and Gray’s competing-risk analysis. Sixth, MASLD is defined by the ICD-10 code K76.0 to capture early stages of the disease. Seventh, participants who changed their coffee type during the follow-up period were included to reduce selection bias. Last, to further verify the causal effect of caffeine on MASLD, we performed MR analysis of caffeine and MASLD (details are provided in Supplementary Methods).

Analyses were performed using R, version 4.3.3, with the package “rcssci”, SAS software, version 9.4 for Windows (SAS Institute), and GraphPad Prism 9.0.0. Statistical tests were 2-sided, and P values less than 0.05 were considered statistically significant.

Results

Baseline characteristics

A total of 185,437 participants (mean [SD] age: 56.22 [7.90] years; 82,779 males [44.64%]) were enrolled. Overall, 170,498 (76.76%) were coffee consumers. Unsweetened coffee [(n = 104,494 (56.35%)] was the most commonly consumed, followed by sugar-sweetened coffee [n = 26,141 (14.10%)] and artificially sweetened coffee [n = 11,709 (6.31%)]. During a median of 10.49 years of follow-up, we documented 1,536 cases of MASLD. Non-consumers tended to be younger and consumed more tea. Unsweetened coffee consumers have a lower Townsend deprivation index, were less educated, more likely to be never-smokers and had healthier diets. Sugar-sweetened coffee consumers were predominantly male, with a higher Townsend deprivation index, consumed more energy and sugar, and generally had less healthy diets. In addition, artificially sweetened coffee consumers tended to be older, had a higher BMI, and a higher prevalence of hypertension, diabetes, and hyperlipidaemia (Table 1).

Table 1.

Baseline characteristics of participants ^a

Characteristic	Overall	Non-coffee consumers	Coffee consumers			P value b
			Unsweetened	Sugar-sweetened	Artificially sweetened
Participants, n (%)	185,437 (100.00%)	43,093 (23.24%)	104,494 (56.35%)	26,141 (14.10%)	11,709 (6.31%)
Age (mean, SD), y	56.22 ± 7.90 c	54.95 ± 8.04	56.47 ± 7.71	56.66 ± 8.22	57.71 ± 7.70	< 0.001
Sex (Male, %)	82,779 (44.64%)	17,980 (41.72%)	43,721 (41.84%)	15,844 (60.61%)	5,234 (44.70%)	< 0.001
BMI (mean, SD) (kg/m2)	26.89 ± 4.62	27.00 ± 4.87	26.74 ± 4.53	26.46 ± 4.10	28.77 ± 5.01	< 0.001
Townsend deprivation index (mean, SD)	-1.70 ± 2.79	-1.49 ± 2.88	-1.84 ± 2.72	-1.54 ± 2.90	-1.63 ± 2.85	< 0.001
Education, n (%)						< 0.001
College/university degree	26,005 (14.02%)	5,034 (11.68%)	16,606 (15.89%)	3,117 (11.92%)	1,248 (10.66%)
Other degrees	158,607 (85.53%)	37,803 (87.73%)	87,550 (68.26%)	22,859 (87.45%)	10,395 (88.78%)
Unknow	825 (0.44%)	256 (0.59%)	338 (0.32%)	165 (0.63%)	66 (0.56%)
Smoking status, n (%)						< 0.001
Current	13,827 (7.46%)	2,860 (6.64%)	6,173 (5.91%)	3,553 (13.59%)	1,241 (10.60%)
Former	66,326 (35.77%)	14,323 (33.24%)	36,992 (35.40%)	9,529 (36.45%)	5,482 (46.82%)
Never	104,831 (56.53%)	25,796 (59.86%)	61,108 (58.48%)	12,973 (49.63%)	4,954 (42.31%)
Unknow	453 (0.24%)	114 (0.26%)	221 (0.21%)	86 (0.33%)	32 (0.27%)
Drinking status, n (%)						< 0.001
Current	175,316 (94.54%)	39,185 (90.93%)	100,373 (96.06%)	24,816 (94.93%)	10,942 (93.45%)
Former	5,144 (2.77%)	1,903 (4.42%)	2,218 (2.12%)	622 (2.38%)	401 (3.42%)
Never	4,850 (2.62%)	1,961 (4.55%)	1,847 (1.77%)	679 (2.60%)	363 (3.10%)
Unknow	127 (0.07%)	44 (0.10%)	56 (0.05%)	24 (0.09%)	3 (0.03%)
Physical activity, n (%)						0.002
Low	28,630 (15.44%)	6,968 (16.17%)	15,522 (14.85%)	4,042 (15.46%)	2,098 (17.92%)
Moderate	66,645 (35.94%)	14,909 (34.60%)	38,748 (37.08%)	9,010 (34.47%)	3,978 (33.97%)
High	61,940 (33.40%)	14,177 (32.90%)	35,246 (33.73%)	8,855 (33.87%)	3,662 (31.28%)
Unknow	28,222 (15.22%)	7,039 (16.33%)	14,978 (14.33%)	4,234 (16.20%)	1,971 (16.83%)
healthy diet score (mean, SD)	4.63 ± 1.66	4.53 ± 1.69	4.86 ± 1.60	3.96 ± 1.62	4.46 ± 1.63	< 0.001
Tea intake, drinks/d*	2.17 ± 1.87	2.90 ± 2.14	2.01 ± 1.72	1.86 ± 1.74	1.63 ± 1.69	< 0.001
Total energy, kJ/d	8604.35 ± 2373.73	8453.78 ± 2496.69	8531.02 ± 2262.40	9229.38 ± 2483.37	8417.41 ± 2410.51	< 0.001
Total sugar, g/d	124.11 ± 47.47	122.93 ± 50.41	120.49 ± 44.12	142.46 ± 50.63	119.81 ± 48.44	< 0.001
Free sugar, g/d	59.91 ± 35.07	60.87 ± 38.17	54.11 ± 29.74	83.83 ± 39.49	54.75 ± 33.71	0.039
Hypertension, n (%)	97,528 (52.59%)	21,903 (50.83%)	54,250 (51.92%)	14,287 (54.65%)	7,088 (60.53%)	< 0.001
Diabetes, n (%)	9,693 (5.23%)	2,280 (5.29%)	5,306 (5.08%)	653 (2.50%)	1,454 (12.42%)	< 0.001
Hyperlipidemia, n (%)	20,094 (10.84%)	4,333 (10.05%)	10,809 (10.34%)	3,001 (11.48%)	1,951 (16.66%)	< 0.001
Completed 24-h dietary recalls (mean, SD), n	2.16 ± 1.18	1.92 ± 1.11	2.29 ± 1.19	2.10 ± 1.17	2.04 ± 1.13	< 0.001

^a BMI Body mass index, SD Standard deviation

^bAnalysis of covariance or Chi-square test

^cContinuous variables were presented as mean (standard deviation), and categorical variables were shown as percentages

*1 drink is equal to approximately 250 mL

Coffee consumption and the risk of MASLD

As shown in Table 2, coffee consumption was negatively associated with the risk of MASLD after adjustments (P for trend < 0.001). Compared with non-consumers, the multivariable-adjusted HRs (95% CIs) of incident MASLD for those with 0-1.5, 1.5–2.5, and > 2.5 drinks/day were 0.71 (0.62, 0.81), 0.73 (0.63, 0.85), and 0.78 (0.68, 0.89), respectively. Table 3 shows that unsweetened coffee and caffeinated coffee consumption were negatively associated with the risk of MASLD. As shown, compared with non-consumers, the adjusted HRs (95% CIs) of incident MASLD for those with more than 2.5 drinks/d of unsweetened, sugar-sweetened, artificially sweetened, caffeinated, and decaffeinated coffee were 0.70 (0.60, 0.82), 0.95 (0.73, 1.23), 1.00 (0.78, 1.28), 0.79 (0.68, 0.91) and 0.90 (0.68, 1.17), respectively. Moreover, the study found that the nonlinear associations between total coffee, unsweetened coffee, and caffeinated coffee consumption and the risk of MASLD were statistically significant (all P values for non-linear < 0.001) (Fig. 1).

Table 2.

Associations of coffee consumption with incident MASLD ^a

Outcomes	Non-coffee consumers	Coffee intake, drinks per day b			P for trend c
		> 0-1.5 drinks/d	> 1.5–2.5 drinks/d	> 2.5 drinks/d
Incident MASLD/person-years	472/428,649	357/525,451	284/401,358	423/484,973
Crude model	1.00 (Reference)	0.62 (0.54, 0.71) g	0.65 (0.56, 0.75)	0.80 (0.70, 0.91)	0.004
Model 1d	1.00 (Reference)	0.68 (0.59, 0.78)	0.69 (0.60, 0.80)	0.76 (0.67, 0.87)	< 0.001
Model 2e	1.00 (Reference)	0.70 (0.61, 0.81)	0.73 (0.63, 0.84)	0.77 (0.68, 0.88)	< 0.001
Model 3f	1.00 (Reference)	0.71 (0.62, 0.81)	0.73 (0.63, 0.85)	0.78 (0.68, 0.89)	< 0.001

^aBMI Body mass index, CVD Cardiovascular disease, MASLD Metabolic dysfunction-associated steatotic liver disease

^b1 drink is equal to approximately 250 mL

^cTrend tests were conducted by including coffee intakes as ordinal variables

^dModel 1: Adjusted for age, BMI, and sex

^eModel 2: Additionally adjusted for Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, and tea based on model 1

^fModel 3: Additionally adjusted for cancer, CVD, diabetes, hypertension, and hyperlipidemia based on model 2

^gHazard ratio (95% confidence interval) (all such values)

Table 3.

Associations of coffee consumption with MASLD by coffee type and caffeination status separately ^a

	Non-coffee consumers	Coffee intake, drinks per day b			P for trend c
		> 0-1.5 drinks/d	> 1.5–2.5 drinks/d	> 2.5 drinks/d
Unsweetened coffee
Incident MASLD/person-years	472/428,649	214/371,656	178/301,207	278/363,471
Crude model	1.00 (Reference)	0.52 (0.45, 0.62) g	0.54 (0.46, 0.64)	0.70 (0.60, 0.81)	< 0.001
Model 1 d	1.00 (Reference)	0.59 (0.50, 0.69)	0.59 (0.49, 0.70)	0.68 (0.58, 0.78)	< 0.001
Model 2 e	1.00 (Reference)	0.62 (0.52, 0.73)	0.62 (0.52, 0.74)	0.70 (0.60, 0.81)	< 0.001
Model 3 f	1.00 (Reference)	0.62 (0.53, 0.73)	0.62 (0.52, 0.74)	0.70 (0.60, 0.82)	< 0.001
Sugar-sweetened coffee
Incident MASLD/person-years	472/428,649	97/115,349	60/70,286	71/73,908
Crude model	1.00 (Reference)	0.76 (0.61, 0.95)	0.78 (0.59, 1.02)	0.87 (0.68, 1.12)	0.042
Model 1 d	1.00 (Reference)	0.87 (0.70, 1.08)	0.89 (0.68, 1.17)	0.96 (0.75, 1.24)	0.455
Model 2 e	1.00 (Reference)	0.87 (0.69, 1.08)	0.89 (0.67, 1.16)	0.91 (0.70, 1.17)	0.260
Model 3 f	1.00 (Reference)	0.89 (0.71, 1.11)	0.93 (0.70, 1.22)	0.95 (0.73, 1.23)	0.489
Artificially sweetened coffee
Incident MASLD/person-years	472/428,649	46/38,447	46/29,866	74/47,594
Crude model	1.00 (Reference)	1.09 (0.80, 1.47)	1.41 (1.04, 1.91)	1.42 (1.11, 1.81)	< 0.001
Model 1 d	1.00 (Reference)	0.88 (0.65, 1.19)	1.15 (0.85, 1.56)	1.10 (0.86, 1.41)	0.377
Model 2 e	1.00 (Reference)	0.86 (0.64, 1.17)	1.14 (0.84, 1.55)	1.03 (0.80, 1.32)	0.686
Model 3 f	1.00 (Reference)	0.82 (0.60, 1.11)	1.08 (0.80, 1.47)	1.00 (0.78, 1.28)	0.964
Caffeinated coffee
Incident MASLD/person-years	472/428,649	292/419,073	206/302,040	312/345,718
Crude model	1.00 (Reference)	0.64 (0.55, 0.74)	0.63 (0.53, 0.74)	0.83 (0.72, 0.96)	0.003
Model 1 d	1.00 (Reference)	0.68 (0.59, 0.79)	0.66 (0.56, 0.77)	0.78 (0.68, 0.90)	< 0.001
Model 2 e	1.00 (Reference)	0.71 (0.61, 0.82)	0.69 (0.58, 0.81)	0.78 (0.68, 0.90)	< 0.001
Model 3 f	1.00 (Reference)	0.72 (0.62, 0.83)	0.70 (0.59, 0.82)	0.79 (0.68, 0.91)	< 0.001
Decaffeinated coffee
Incident MASLD/person-years	472/428,649	47/67,709	45/43,820	60/54,607
Crude model	1.00 (Reference)	0.63 (0.47, 0.85)	0.93 (0.69, 1.27)	0.99 (0.76, 1.30)	0.445
Model 1 d	1.00 (Reference)	0.67 (0.50, 0.91)	0.94 (0.69, 1.28)	0.89 (0.68, 1.16)	0.204
Model 2 e	1.00 (Reference)	0.71 (0.52, 0.96)	0.99 (0.73, 1.35)	0.91 (0.69, 1.19)	0.343
Model 3 f	1.00 (Reference)	0.70 (0.52, 0.95)	0.99 (0.73, 1.35)	0.90 (0.68, 1.17)	0.298

^aBMI body mass index, CVD cardiovascular disease, MASLD metabolic dysfunction-associated steatotic liver disease

^b1 drink is equal to approximately 250 mL

^cTrend tests were conducted by including coffee intakes as ordinal variables

^dModel 1: Adjusted for age, BMI, and sex

^eModel 2: Additionally adjusted for Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, and tea based on model 1

^fModel 3: Additionally adjusted for cancer, CVD, diabetes, hypertension, and hyperlipidemia based on model 2

^gHazard ratio (95% confidence interval) (all such values)

Fig. 1.

Dose-response associations of total and different types of coffee consumption and MASLD. Results were adjusted for age, BMI, sex, Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, tea, cancer, CVD, diabetes, hypertension, and hyperlipidemia

After categorizing coffee consumers based on both sugar type and caffeine content, within the unsweetened coffee group, the HRs (95% CIs) of MASLD for those with > 0-1.5, > 1.5–2.5, and > 2.5 drinks of caffeinated coffee were 0.64 (0.54, 0.77), 0.57 (0.46, 0.69), and 0.69 (0.58, 0.82), respectively. However, no significant associations were noted between other types of coffee intake and the risk of MASLD (P > 0.05) (Table 4).

Table 4.

Associations of coffee consumption with MASLD by coffee type and caffeination status jointly ^a

	Non-coffee consumers	Coffee intake, drinks per day b			P for trend c
		> 0-1.5 drinks/d	> 1.5–2.5 drinks/d	> 2.5 drinks/d
Unsweetened coffee
Caffeinated coffee (n = 968/772,014)	1.00 (Reference)	0.64 (0.54, 0.77) de	0.57 (0.46, 0.69)	0.69 (0.58, 0.82)	< 0.001
Decaffeinated coffee (n = 576/120,728)	1.00 (Reference)	0.60 (0.41, 0.89)	1.02 (0.71, 1.45)	0.92 (0.68, 1.26)	0.403
Sugar-sweetened coffee
Caffeinated coffee (n = 661/212,575)	1.00 (Reference)	0.88 (0.69, 1.12)	0.88 (0.65, 1.19)	0.99 (0.65, 1.32)	0.587
Decaffeinated coffee (n = 493/26,101)	1.00 (Reference)	1.03 (0.58, 1.84)	0.77 (0.32, 1.86)	0.59 (0.22, 1.59)	0.281
Artificially sweetened coffee
Caffeinated coffee (n = 594/82,291)	1.00 (Reference)	0.75 (0.52, 1.07)	1.19 (0.84, 1.67)	1.08 (0.81, 1.44)	0.548
Decaffeinated coffee (n = 499/19,447)	1.00 (Reference)	0.76 (0.38, 1.52)	1.05 (0.50, 2.21)	0.91 (0.51, 1.63)	0.688

^aBMI Body mass index, CVD Cardiovascular disease, MASLD Metabolic dysfunction-associated steatotic liver disease

^b1 drink is equal to approximately 250 mL

^cTrend tests were conducted by including coffee intakes as ordinal variables

^dEstimates were hazard ratios (95% CIs) from multivariable Cox regression models adjusted for variables in age, BMI, sex, Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, tea, cancer, CVD, diabetes, hypertension, and hyperlipidemia

^eHazard ratios (95% confidence interval) (all such values)

We further examined the joint associations of different types of coffee (based on sugar type and caffeine content) consumption with the risk of MASLD (Fig. 2). As shown, different combinations of unsweetened and caffeinated coffee consumption were negatively associated with the risk of MASLD. In addition, there was a significant interaction between unsweetened coffee and caffeine content with respect to MASLD (P value for interaction < 0.001). However, no significant association was observed between the combinations of sugar-sweetened or artificially sweetened coffee with caffeine content and the risk of MASLD, and no significant interaction between them was found (all P values for interactions > 0.05).

Fig. 2.

The associations of joint classifications of coffee consumption, types of sweetened coffee, and caffeine content on incident MASLD in the UK Biobank. Multivariable Cox proportional regression was adjusted for variables in age, BMI, sex, Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, tea, cancer, CVD, diabetes, hypertension, and hyperlipidemia

Associations of coffee types, FLD-GRS, and GMA-GRS with MASLD risk

As shown in Fig. 3, no significant interaction between different types of coffee consumption and GRSs for GMA or MASLD itself with respect to MASLD risk was found (all P values for interactions > 0.05). Unsweetened and caffeinated coffee consumption was negatively associated with the risk of MASLD across different levels of FLD-GRS (all P for trend < 0.01). Compared with non-consumers, the HRs (95% CIs) of MASLD for those with the highest frequency of unsweetened coffee consumption were 0.71 (0.56, 0.90) and 0.70 (0.57, 0.84) in the low and high FLD-GRS groups, respectively. Similarly, for the highest levels of caffeinated coffee consumption, the HRs (95% CIs) were 0.73 (0.57, 0.92) and 0.83 (0.69, 1.00) in the low and high FLD-GRS groups, respectively. Similar to the grouping by FLD-GRS, subgroup analyses according to GMA-GRS showed that the association between unsweetened coffee consumption and the risk of MASLD, as well as the association between caffeinated coffee consumption and the risk of MASLD, were significant in both low and high GMA-GRS groups (both P for trend < 0.01). Compared with non-consumers, the HRs (95% CIs) of MASLD for those with the highest frequency of unsweetened coffee consumption were 0.71 (0.57, 0.89) and 0.70 (0.56, 0.86) in the low and high GMA-GRS groups, respectively. For those with the highest frequency of caffeinated coffee consumption, the HRs (95% CIs) of MASLD were 0.79 (0.64, 0.97) in the low GMA-GRS group and 0.78 (0.64, 0.96) in the high GMA-GRS group.

Fig. 3.

Stratified analysis of associations between different types of coffee consumption and risk of MASLD according to GRS. Multivariable Cox proportional regression was adjusted for variables in age, BMI, sex, Townsend deprivation index, education level, smoking status, alcohol status, physical activity, healthy diet score, total energy, total sugar, tea, cancer, CVD, diabetes, hypertension, and hyperlipidemia

Supplementary Figs. 2–5 showed the joint associations of coffee types, FLD-GRS, and GMA-GRS with MASLD risk. As shown in Supplementary Fig. 2, compared with participants high FLD-GRS and lowest unsweetened coffee consumption, those who consumed unsweetened coffee had a lower risk of MASLD, regardless of GRS. Similarly, as shown in Supplementary Fig. 3, compared with participants with a FLD-GRS and the lowest consumption of caffeinated coffee, those who consumed caffeinated coffee had a lower risk of MASLD, regardless of GRS.

Furthermore, the combined analysis of the GMA-GRS and coffee consumption and the risk of MASLD is presented in the Supplementary Fig. 4. We observed that compared with low GMA-GRS and lowest unsweetened coffee consumption, participants who drank unsweetened coffee had a lower risk of MASLD, regardless of GRS. However, there were no significant results for sugar-sweetened and artificially sweetened coffee. As shown in Supplementary Fig. 5, compared with low GMA-GRS and the lowest unsweetened coffee consumption, participants who drank caffeinated coffee had a lower risk of MASLD, regardless of GRS. However, there were no significant results for decaffeinated coffee.

Secondary analyses

Subgroup analyses revealed consistent results with the main analyses, indicating that unsweetened/caffeinated coffee consumption was associated with reduced risk of MASLD across different subgroups (Supplementary Tables 5–6). The inverse association with MASLD was stronger for unsweetened coffee in participants with a BMI > 24.9 kg/m² or higher education level, but stronger for caffeinated coffee in those aged > 60 years or with a BMI ≤ 24.9 kg/m² (all P values for interactions < 0.05).

As shown in Supplementary Table 7, sensitivity analyses demonstrated consistent results with the main analysis under various conditions: (1) excluding participants in the first two years of follow-up; (2) multiple imputation by chained equations to impute other missing data.; (3) excluding participants with excessive alcohol; (4) excluding participants with major chronic diseases (including CVD, cancer, and diabetes) at baseline; (5) conducting a competing risk analysis; (6) MASLD is defined separately using K76.0; (7) participants who changed coffee type during follow-up were included.

MR analysis results of caffeine and MASLD showed that the per 80-mg increase in caffeine consumption was associated with a decreased risk of MASLD (combined OR = 0.85, 95% CI: 0.73–0.99, P = 0.037) (Supplementary Fig. 6).

Discussion

In this large population-based prospective cohort study, we found that unsweetened caffeinated coffee consumption was significantly associated with a lower risk of MASLD. However, no significant association was found between sugar-sweetened or artificially sweetened coffee consumption and the risk of MASLD. Furthermore, when unsweetened coffee was further subdivided based on caffeine content, we found that only unsweetened caffeinated coffee demonstrated a protective effect against MASLD, while unsweetened decaffeinated coffee was not associated with MASLD. We also used the Mendelian randomization method to further validate the protective effect of caffeine intake on MASLD. These associations remained consistent across different genetic risk categories for fatty liver or gut microbial abundance, as well as subgroup and sensitivity analyses.

Previous studies have explored the association between coffee consumption and MASLD, yet the findings have varied. Several previous meta-analyses 37–39 supported the protective role of coffee consumption on liver fibrosis in patients with MASLD. An observational study conducted in middle-aged and older South Korean adults also revealed that a higher habitual coffee consumption (> 3 cups/day) was associated with a lower risk of MASLD 15. Yet, a certain study revealed no association between coffee consumption and MASLD. For instance, a cross-sectional study involving almost 3,000 South Italian participants found that there was no association between coffee consumption and the prevalence of hepatic steatosis 16. The varied results observed might stem from the differing design of the study, ethnic background, sample size, and measurement of MASLD. Most importantly, consumption of different kinds of coffee, including unsweetened, sugar-sweetened, artificially sweetened, caffeinated, and decaffeinated coffee, could demonstrate different associations with healthy outcomes. As a previous study showed, unsweetened coffee is significantly associated with a reduced risk of depression and anxiety, but the associations between sugar-sweetened or artificially sweetened coffee and mental illness are inconsistent 40. However, no previous study has explored the associations of different types of sweetened coffee consumption with the risk of MASLD. We first found a significant inverse association between unsweetened coffee consumption and the risk of MASLD, while sugar-sweetened or artificially sweetened coffee intake was not. Presently, it is generally accepted that diet is crucial in the development of chronic liver conditions 41, 42. Beverage intake, such as coffee, is considered one of the protective factors for chronic liver disease. Over 1000 different compounds are found in coffee, encompassing caffeine, diterpenoid alcohols, niacin, and antioxidants like chlorogenic acid (CGA) and tocopherols 43. Some of the beneficial components have been elucidated. Previous study indicates caffeine’s ability to lower lipid concentrations and prevent fat accumulation in mice fed a high-fat diet 44. The mechanism may be that caffeine boosts the expression of low-density lipoprotein receptors expression by directly binding to epidermal growth factor receptor (EGFR) and activating the EGFR-ERK1/2 signalling pathway 45. This aligns with our finding that caffeinated coffee is capable of diminishing the risk of MASLD. The results of the MR analysis also confirmed this point. Perhaps because there are only two SNPs of caffeine, further research will be needed in the future to verify our hypothesis. Furthermore, coffee with high CGA (a natural polyphenol compound) concentrations can modulate glucose intolerance and improve/decrease MASLD development in obese rats 46; coffee polyphenols could reduce the build-up of fat in the liver by downregulating the expression of lipogenic genes driven by SREBP-1c 47. In our study, we did not find an association between coffee with sugar or artificially sweeteners and MASLD. Suggesting that adding sugar or artificial sweeteners to coffee may mask or offset the benefits of drinking coffee. It has been demonstrated that a diet rich in fructose, sucrose, or HFCS tends to cause fatty liver in experimental animals 48–54. In addition, artificial sweeteners 55 are proven risk factors for MASLD. Excessive intake of artificial sweeteners may lead to imbalances in the gut microbiome and trigger the release of proinflammatory agents, potentially elevating the risk of MASLD 56.In the present study, we did not find evidence of interactions between coffee consumption and the FLD-GRS, suggesting that the beneficial association between a higher level of unsweetened caffeinated coffee consumption and the risk of MASLD was not modified by genetic risk profiles. However, the absence of significant interactions might be attributed, in part, to the limited proportion of genetic risk explained by the included SNPs. Thus, the interpretation of these results requires caution. Moreover, previous studies have found that the gut microbiota modulates postprandial nutrient metabolism and generates metabolites that interact with the host through multiple molecular mechanisms, thereby influencing the incidence of disease 57, 58. Previous studies have also found that coffee can affect liver disease through intestinal flora in animal studies 59 or clinical trials 60, but there were no results from large-scale population cohort. Therefore, we for the first time investigated the interactions of coffee consumption and GMA-GRS on the risk of MASLD. However, we did not find evidence of interactions between coffee consumption and the GMA-GRS. There seems to be an obvious reason behind this result. Previous studies have shown that genetic factors explain less than 10% of the variation in gut microbiota abundance 33. Therefore, GMA-GRS can only represent genetic predisposition and only reflect partial abundance. However, gut microbiota is influenced not only by genetics but also by immune responses and dietary factors. These influences may obscure the beneficial association between high coffee intake and the risk of MASLD 57. Future research is needed to directly use the abundance of intestinal flora to verify our findings. Additionally, we found that BMI had significant interactions with unsweetened and caffeinated coffee on the outcome. Specifically, we found that the beneficial association between unsweetened coffee and MASLD was more pronounced in overweight populations. A previous study found that moderate caffeine consumption was associated with lower all-cause mortality in adults, especially those who were overweight 61. This means that components in coffee, such as caffeine, may have more significant beneficial effects on metabolism in overweight or obese individuals. In addition, education level interacted significantly with unsweetened coffee and age with caffeinated coffee on outcomes. Earlier research has indicated that education has a beneficial impact on health-related lifestyles and behaviours, such as smoking 62–64, drinking 64, or care seeking 65, 66. In individuals aged > 60 years, lower to moderate caffeine coffee consumption (< 2.5 drinks/day) is associated with a reduced risk of MASLD. This phenomenon may be attributed to the fact that, compared with high-dose intake, lower to moderate doses of caffeine are more likely to match the intrinsic metabolic capacity of older adults. Such a dosage range can more effectively alleviate the age-related propensity for hepatic lipid accumulation. Concurrently, lower to moderate caffeine intake exerts less impact on the cardiovascular system, thereby minimizing the risk of precipitating cardiovascular complications. While choosing unsweetened and caffeinated coffee may be part of their overall healthy lifestyle, other healthy behaviours may enhance the beneficial association between unsweetened caffeinated coffee intake and the risk of MASLD.

This study has several strengths, including its prospective design, large sample size, and wide range of coffee consumption information. We also considered the impact of gut microbial abundance on the risk of MASLD.

This study also has several limitations. Firstly, given the inherent limitations of observational design, the potential for lingering confounding or bias arising from factors that are not measured or known remains a consideration. Second, participants in our study were of European heritage, who resided in areas with lower socioeconomic deprivation and led healthier lives compared to the broader population. This suggests a “healthy volunteer” bias 67, which may limit the generalizability of our findings to other populations. However, accurate evaluations of the link between exposure and disease typically don’t necessitate a representative population 67. Third, self-reported exposures assessed by the 24-h diet recall questionnaires were subject to inevitable measurement error and recall bias, which in turn leads to misclassification of intake, although the questionnaire has been validated 25, 68. Fourth, the consumption of coffee relied on the self-reported data of participants, accompanied by the potential for reporting bias. Hence, the potential for inaccurate reporting cannot be completely ruled out. Furthermore, since the evaluation of coffee consumption was limited to the baseline, categorizing participants as one type of coffee consumer appears imprecise, as they might consume various coffee varieties over time. However, the follow-up duration may be insufficient to capture all long-term effects or changes in coffee consumption patterns over time. Considering only a single coffee type and not mixing coffee types over time could lead to bias in exposure classification. Participants who changed coffee types during follow-up were excluded, potentially introducing selection bias. Potential alterations in coffee consumption post-baseline could have impacted our risk estimates. Future studies are required to explore how variations in coffee consumption over time affect the risk of MASLD. A high degree of reproducibility for assessing nutrient intake over time has been demonstrated in a previous study 69. Therefore, we assumed that baseline coffee intake remained constant during follow-up. Fifth, the lack of quantitative data on milk, cream, or other additives that are commonly added to coffee precludes the assessment of their impact on the results. Sixth, associations between unsweetened/caffeinated coffee and lower MASLD risk are broadly supported by the evidence, but causal claims (especially via MR) should be interpreted cautiously due to weak instruments and residual confounding. Seventh, FLD-GRS and GMA-GRS captured only a small fraction of the genetic variation in MASLD and gut microbiome composition, with an explained variance ratio (R²) is less than 10%, the results need to be interpreted carefully. Eighth, since only two SNPs were included in the MR analysis and the sample size was very small, the genetically predicted caffeine exposure may be negatively associated with MASLD risk. However, this analysis fell well short of the conventional requirement for the number of instrumental variables in MR studies ( ≥3 SNPs), leading to extremely low statistical power. Finally, MASLD was identified through the application of hospital inpatient data alongside death register records. This approach potentially introduced bias due to the higher likelihood of hospitalization and subsequent diagnosis of MASLD in individuals with serious comorbidities compared to those without. Focusing on these clinically significant or advanced phenotypes may lead to an underestimation of the overall incidence. Therefore, our findings are primarily applicable to advanced or confirmed MASLD, and future studies need to confirm our results further.

Conclusions

In summary, this prospective cohort investigation revealed a negative association between the consumption of unsweetened caffeinated coffee and the risk of MASLD, regardless of genetic predisposition. The results of our study endorse the idea that drinking unsweetened caffeinated coffee might contribute to a healthy way of living, aiming to prevent and control MASLD among the broader populace.

Authors’ contributions

Wenqi Liu: conceptualization; formal analysis; writing–original draft. Xinrui Xu: conceptualization; formal analysis; writing–original draft. Qing Chang: formal analysis. Honghao Yang: writing–review & editing. Zheng Ma: formal analysis. Tingjing Zhang: formal analysis. Yuhong Zhao: writing–review & editing; supervision. Lu Zhao: conceptualization; writing–review & editing; supervision. Yang Xia: conceptualization; data curation; writing–review & editing; supervision; funding acquisition. All authors mentioned above made substantial contributions to the content of the paper. All authors read and approved the final manuscript.

Funding

This study was supported by the LiaoNing Revitalization Talents Program (grant number XLYC2203168 to Yang Xia) and the General Program of Natural Science Foundation of Liaoning Province (grant numbers: 2025-MS-221 to Yang Xia).The funders had no role in the conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

Data availability

Data are available in a public, open access repository. This research has been conducted using the UK Biobank Resource under Application Number 524293. The UK Biobank data are available on application to the UK Biobank (https://www.ukbiobank.ac.uk/).

Declarations

Ethics approval and consent to participate

No applicable.

Consent for publication

All authors approved the final manuscript and the submission to this journal.

Competing interests

The authors declare no competing interests.