Metabolic dysfunction–associated steatotic liver disease as a risk factor for cognitive decline: Findings from the ELSA‐Brasil cohort

Abstract

INTRODUCTION

Metabolic dysfunction–associated steatotic liver disease (MASLD) has been linked to adverse brain outcomes. We aim to evaluate whether MASLD is associated with cognitive impairment/decline.

METHODS

We analyzed data from the Estudo Longitudinal da Saúde do Adulto (ELSA‐Brasil study). Global cognitive performance was assessed at baseline and follow‐up. Associations between MASLD and cognition were tested using Poisson regression and linear mixed‐effects models (LMMs).

RESULTS

MASLD was present in 13.6% (n = 918/6754) of participants at baseline. MASLD was not associated with incident cognitive impairment (Relative Risk: 1.01; 95% confidence interval [CI]: 0.83–1.23; p = 0.91). In LMMs, in a fully adjusted model, MASLD was not associated with cognitive decline (estimate: 0.039; 95% CI: –0.109 to 0.031; p = 0.2721) but was associated with the memory domain at the baseline.

DISCUSSION

MASLD is not independently associated with cognitive impairment or cognitive decline in middle‐aged adults.

Highlights

• Metabolic dysfunction–associated steatotic liver disease (MASLD) is not independently associated with cognition in the longitudinal analysis.

• Hepatic steatosis is not independently associated with cognition.

• The effect on cognition is dependent on vascular and metabolic risk factors.

1. BACKGROUND

Metabolic dysfunction–associated steatotic liver disease (MASLD)—a recently redefined term encompassing what was formerly known as nonalcoholic fatty liver disease (NAFLD)—has become the most prevalent chronic liver condition worldwide, affecting an estimated 25%–30% of the adult population. ¹ MASLD is characterized by hepatic steatosis in the presence of cardiometabolic dysfunction, such as obesity, insulin resistance, dyslipidemia, and hypertension. ² These metabolic disturbances, common in low‐ and middle‐income countries experiencing rapid epidemiological transitions, have systemic consequences extending beyond the liver. ³ Emerging literature suggests that MASLD may not only be a hepatic manifestation of systemic metabolic dysfunction but could also serve as a marker—or even a mediator—of extrahepatic conditions, including cardiovascular disease, chronic kidney disease, and, more recently, cognitive decline. ⁴

Cognitive decline and dementia represent a growing global health challenge, particularly in aging populations. Established risk factors include advanced age, lower educational attainment, and cardiovascular and metabolic conditions such as diabetes, hypertension, and dyslipidemia. ⁵ Inflammation, insulin resistance, endothelial dysfunction, and altered lipid metabolism—all commonly observed in MASLD—have also been implicated in the pathogenesis of cognitive decline. ⁴ Based on this shared pathophysiology, several recent studies have explored whether MASLD may be independently associated with cognitive impairment or dementia. ⁶ Although some reports suggest a positive association, these findings remain inconsistent and are frequently limited by cross‐sectional designs, heterogeneous cognitive assessments, small sample sizes, or inadequate adjustment for confounding variables. ⁷ Consequently, it remains unclear whether MASLD plays an independent role in cognitive decline or merely reflects an underlying cardiometabolic burden.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed literature on metabolic dysfunction–associated steatotic liver disease (MASLD) as a risk factor for cognitive decline using traditional sources (e.g., PubMed). MASLD has not been consistently associated with cognitive decline. Relevant citations are appropriately cited.

2. Interpretation: Our findings reinforce the hypothesis that MASLD is not independently associated with cognitive decline. Traditional vascular and metabolic risk factors seem to contribute to this association.

3. Future directions: Future research could consider more refined markers of hepatic severity, such as inflammatory activity, which may have a stronger effect than steatosis. Longitudinal studies with the incorporation of neuroimaging and fluid biomarkers could help clarify whether MASLD contributes indirectly to cognitive aging through vascular and metabolic pathways.

In this study, we aim to investigate the relationship between MASLD and cognitive decline in participants of the Estudo Longitudinal da Saúde do Adulto (ELSA)‐Brasil cohort, a large, multicenter, and ethnically diverse Brazilian study. Leveraging longitudinal data with standardized and validated measures of liver steatosis, metabolic risk factors, and cognitive performance across multiple domains, we seek to determine whether MASLD is independently associated with cognitive decline over time. Clarifying this association may have important implications for early risk stratification, prevention, and management strategies targeting cognitive decline in individuals with metabolic dysfunction.

2. METHODS

2.1. Participants

The ELSA‐Brasil study is a large, multicenter cohort designed to investigate risk factors for chronic diseases in the Brazilian population. Between 2008 and 2010, it enrolled 15,105 active and retired employees, ages 35 to 74 years, from public universities and research institutions located in six state capitals—Belo Horizonte, Porto Alegre, Rio de Janeiro, Salvador, São Paulo, and Vitória—spanning three regions of the country. ⁸ , ⁹ Participants were followed up with subsequent clinical assessments during 2012–2014 (Wave 2) and 2017–2019 (Wave 3). The study was approved by the respective institutions' research ethics committees and was conducted per the ethical standards of the 1964 Declaration of Helsinki. All participants provided written informed consent for all waves of the study.

2.2. Outcome variables

Cognitive functions were assessed at baseline and again during wave 3 using six standardized tests developed for Brazilian Portuguese speakers. Trained and certified research assistants collected data through face‐to‐face interviews and clinical evaluations. The tests targeted three cognitive domains: memory, executive function, and language. All assessments were administered under standardized conditions in a quiet setting by trained examiners.

For memory and language assessments, the study used a validated and culturally adapted Brazilian Portuguese version of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Memory was evaluated using tests for learning, recall, and word recognition based on a list of 10 words. The memory score reflected the total number of correctly recalled words across these tests. Language function was assessed through semantic verbal fluency (naming as many animals as possible in 1 min) and phonemic fluency (generating words beginning with a specific letter “F” within 1 min). Higher word counts indicated better performance in both tests.

Executive function—encompassing attention, concentration, psychomotor speed, and mental flexibility—was measured using the Trail Making Test Part B. Participants were instructed to connect 24 circles, each labeled with a number or a letter, in an alternating numeric and alphabetic sequence, as quickly as possible without lifting the pen. Because a shorter completion time reflects better performance, the time (in seconds) was multiplied by –1 to align the scoring direction with other tests.

A global cognitive score (GCS) was derived from the individual test results following a standardized procedure. For each test, a z‐score was calculated for each participant by subtracting the mean score and dividing by the standard deviation (SD) of the score for that test. Separate domain mean z‐scores were then generated: the executive function score corresponded directly to the Trail Making Test z‐score; the language score was the average of the semantic and phonemic fluency z‐scores; and the memory score was the average of the z‐scores for learning, recall, and word recognition. This process was applied to baseline and wave 3 data. The GCS was then calculated by averaging the z‐scores of the three domains at each time point. Finally, the GCS at wave 3 was standardized using the baseline mean and SD. Cognitive impairment was defined as a global cognitive z‐score at wave 3 below –1.5, a threshold commonly used in epidemiological cognitive studies to indicate impairment.

2.3. MASLD definition

Liver steatosis was evaluated using ultrasonography at baseline. All research centers used the same equipment model: a high‐resolution B‐mode scanner (SSA‐790A, Aplio XG, Toshiba Medical Systems; https://www.global.toshiba), equipped with a convex array transducer (model PVT‐375BT), configured with a central frequency of 3.5 MHz and a fundamental frequency range of 1.9–5.0 MHz. ¹⁰ , ¹¹ Examinations were performed by board‐certified radiologists or trained technicians, adhering to a standardized protocol and including hepatic beam attenuation. All imaging was interpreted at a centralized reading center in São Paulo, where a senior radiologist also reviewed the exam quality.

Hepatic steatosis was defined primarily based on hepatic beam attenuation, which, in this cohort, has demonstrated the highest accuracy for detecting steatosis when compared to high‐resolution computed tomography (CT) in individuals with nonalcoholic fatty liver disease (or NAFLD). ¹¹ Hepatic beam attenuation was assessed by the visibility of intrahepatic vessels and the diaphragm posterior to the right hepatic lobe. Steatosis was considered absent when hepatic attenuation appeared normal with full visualization of the diaphragm. Abnormal attenuation was categorized as mild (diaphragm visibility >50%), moderate (diaphragm visibility <50%), or severe (diaphragm not visible). ¹¹ Ultrasound image quality was evaluated based on the visualization of four anatomic landmarks within the axial plane of the right hepatic lobe: the anterior hepatic surface, posterior hepatic surface, gallbladder, and portal vein. Exams in which fewer than two of these structures were identifiable were classified as having “poor” image quality. Participants without ultrasound or with low/undetermined quality images were excluded from the analysis.

MASLD was defined, according to a recent definition, as ultrasound‐based steatosis plus the presence of at least one of the following cardiometabolic risk factors: overweight/obesity defined by elevated body mass index (BMI) or elevated waist circumference (WC), prediabetes and diabetes, high blood pressure, hypertriglyceridemia, and low high‐density lipoprotein (HDL) cholesterol. ² The definitions of cardiometabolic risk factors used in this study, as defined in Rinella et al. 2023, ² are as follows:

• −Height was measured using a standardized stadiometer, and weight was obtained with a calibrated electronic scale. WC was measured with a non‐elastic tape positioned at the midpoint between the lower margin of the last palpable rib and the top of the iliac crest along the mid‐axillary line. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²). The presence of overweight or obesity was defined as a BMI ≥23 kg/m² for participants who self‐identified as Asian, and a BMI ≥25 kg/m² for all other self‐reported ethnic groups. Alternatively, WC thresholds of >94 cm for men and >80 cm for women were also considered indicative of excess adiposity.
• −Prediabetes and diabetes were defined by blood tests performed in each visit: fasting plasma glucose ≥100 mg/dL (5.6 mmol/L), 2 h 75 g oral glucose tolerance test ≥140 mg/dL (7.8 mmol/L), or glycated hemoglobin (HbA1c) ≥5.7% (39 mmol/mol).
• −Blood pressure was measured three times during the clinic visits, and the average of the final two readings was used for analysis. The use of antihypertensive medication was defined by the simultaneous presence of: (1) self‐reported use of medication for arterial hypertension within the preceding 2 weeks, and (2) verification of at least one antihypertensive agent through inspection of prescriptions or medication packaging provided by the participant. Elevated blood pressure was classified as a mean systolic blood pressure ≥130 mmHg and/or a mean diastolic blood pressure ≥85 mmHg, or the confirmed use of antihypertensive medication.
• −Hypertriglyceridemia was identified by fasting plasma triglyceride levels ≥150 mg/dL (1.7 mmol/L) or the confirmed use of triglyceride‐lowering medications, such as fibrates or nicotinic acid. Medication use was verified through prescriptions and medication packaging presented by the participant during the clinical visit. ¹²
• −Low HDL cholesterol was classified as fasting HDL cholesterol concentrations ≤40 mg/dL (1.0 mmol/L) in males or ≤50 mg/dL (1.3 mmol/L) in females.

In addition, we excluded from the analysis participants with excessive alcohol consumption, defined as the equivalent of alcohol intake of 210 g/week for men and 140 g/week for women. According to current guidelines, ² participants with a self‐reported history of liver disease (cirrhosis or hepatitis) were excluded, as this was the only available information on pre‐existing hepatic conditions at baseline.

2.4. Covariates

At baseline, age (in years), sex, and self‐reported skin color/race (Brazilian census categories of White, Black, Brown [mixed‐race], Asian, and Indigenous) were collected. We also included as covariates dementia risk factors, as follows:

• −Low education: we used the categories “incomplete primary education” and “complete primary education” to define low educational achievement, which currently lasts 8 years in Brazil.
• −Depression: assessed using the Brazilian version of Clinical Interview Schedule‐Revised (CIS‐R), a structured interview for measuring and diagnosing non‐psychotic psychiatric morbidity present over the past 7 days. ¹³ The diagnosis of a current depressive episode was determined based on several CIS‐R questions, employing an algorithm derived from the International Classification of Diseases, Tenth Revision (ICD‐10) diagnostic criteria. ¹⁴ For this analysis, we included mild, moderate, and severe depression, with and without somatic symptoms.
• −Smoking: active smoking was evaluated based on self‐reporting. This is a categorical variable with three categories: “never smoked,” “former smoker,” and “smoker.” We define active smoking for those who answer “smoker” to this question.
• −Physical inactivity: assessed through the International Physical Activity Questionnaire–Long Form. ¹⁵ Participants self‐reported the time spent engaging in moderate and vigorous activity in the last week during leisure time. Physical inactivity was defined as <150 min of moderate or <75 min of leisure‐time vigorous physical activity per week. ¹⁶

2.5. Statistical methods

Continuous variables were described using means and SDs or medians and interquartile ranges (IQRs), as appropriate, whereas categorical variables were summarized as frequencies and percentages. Group comparisons were performed using Student's t‐test, Mann–Whitney U test, or chi‐square test, depending on variable type and distribution. A two‐sided p‐value < 0.05 was considered statistically significant. We applied three main analytical approaches to evaluate the relationship between MASLD and cognitive outcomes: (1) Poisson regression with robust variance to estimate the relative risk of cognitive impairment at wave 3, (2) mixed linear regression to assess changes in cognitive performance over time, and (3) sensitivity analysis using hepatic steatosis as the primary exposure. All models were progressively adjusted for relevant confounders, including sociodemographic and dementia risk factors.

2.5.1. Poisson regression analysis

We used Poisson regression models with robust variance to estimate the relative risk (RR) of incident cognitive impairment at wave 3, defined as a global cognitive z‐score below –1.5 SD. The primary exposure was MASLD status at baseline. Initial models were adjusted for age, sex, and race. Subsequent models included additional adjustments for educational attainment, smoking status, physical activity, and depression. Finally, we perform a model including the risk factors defining MASLD, namely diabetes, hypertension, obesity, hypertriglyceridemia, and low HDL cholesterol. Results are presented as adjusted relative risks with 95% confidence intervals (CIs).

2.5.2. Mixed linear regression analysis

To investigate longitudinal changes in cognitive performance, we fitted a linear mixed‐effects model with random intercepts for individuals to account for within‐subject correlation. We used the “lmer” function from the “lme4” package in R. The primary outcome was the global cognitive z‐score, calculated at baseline and wave 3. The main predictor was MASLD status at baseline. Time was modeled as a categorical variable, and an interaction term between MASLD and time was included to assess differential cognitive decline across groups. Models were progressively adjusted for the same covariates described in the Poisson regression analysis. Beta coefficients and 95% CIs are reported, reflecting the average change in cognitive performance over time and the effect of MASLD.

2.5.3. Sensitivity analysis

We conducted a sensitivity analysis using hepatic steatosis as the primary exposure to explore the robustness of the association between MASLD and cognitive performance. We then repeated the main linear mixed‐effects model with steatosis as the predictor, adjusting for the same covariates as in the primary analysis. This analysis aimed to evaluate whether the observed association with cognition was attributable to hepatic fat accumulation alone or to the broader metabolic dysfunction captured by the MASLD definition.

We also performed the same analysis (using MASLD as the independent factor) with each of the three domains (memory, executive function, and language) as the dependent variable.

All statistical analyses were conducted using R (R Foundation for Statistical Computing, Vienna, Austria) version 4.4.3. ¹⁷

3. RESULTS

Of the 15105 participants initially assessed, 6754 individuals were included in the analysis (Figure 1), with a mean (SD) follow‐up time of 8.1 (0.6) years. The median age was 50 years (IQR: 44–56), with a predominance of women (n = 3990; 59.1%) and individuals self‐identifying as White (n = 3862; 57.6%). MASLD was present in 13.6% (n = 918) of participants, and hepatic steatosis in 13.7% (n = 928). Compared to those without MASLD, affected individuals were older, more frequently male, exhibited a less favorable profile of dementia risk factors, and had lower global cognitive performance (Table 1). In the fully adjusted Poisson regression model (6256 included in the analysis, see Figure 1), MASLD was not significantly associated with cognitive impairment (RR [95% CI]: 1.01 [0.83 to 1.23], p = 0.915) (Table 2).

FIGURE 1.

Flowchart of participants’ selection.

TABLE 1.

Characteristics of the individuals included, divided by the presence of MASLD.

	All individuals included (n = 6754)	MASLD– (n = 5836, 86.4%)	MASLD+ (n = ​​918, 13.6%)	p‐value
Age, years, median (IQR)	50 (44, 56)	49 (44, 56)	52 (46, 58)	< 0.001
Female, n (%)	3990 (59.1)	3592 (61.5)	398 (43.4)	< 0.001
Skin color/race, n (%) White Brown (mixed‐race) Black Asian Indigenous	3862 (57.6) 1659 (24.8) 928 (13.8) 203 (3.0) 49 (0.7)	3314 (57.2) 1449 (25.0) 804 (13.9) 178 (3.1) 45 (0.8)	548 (60.2) 210 (23.1) 124 (13.6) 25 (2.7) 4 (0.4)	0.412
Low education, n (%)	501 (7.4)	415 (7.1)	86 (9.4)	0.018
Obesity, n (%)	1436 (21.3)	938 (16.1)	498 (54.3)	< 0.001
Smoking, n (%)	752 (11.1)	653 (11.2)	99 (10.8)	0.760
Depression, n (%)	263 (3.9)	232 (4.0)	31 (3.4)	0.435
Physical inactivity, n (%)	4905 (74.0)	4169 (72.8)	736 (81.1)	< 0.001
Diabetes, n (%)	814 (12.1)	551 (9.4)	263 (28.6)	< 0.001
Hypertension, n (%)	2010 (29.8)	1555 (26.7)	455 (49.6)	< 0.001
Global Cognitive Z Score Baseline, median (IQR)	0.15 (−0.52, 0.67)	0.18 (−0.48, 0.69)	−0.02 (−0.75, 0.53)	< 0.001
Global Cognitive Z Score Wave 3, median (IQR)	0.08 [−0.64, 0.63]	0.10 [−0.61, 0.64]	−0.08 [−0.86, 0.55]	< 0.001
Global Cognitive Z Score <1.5 SD at Wave 3, n (%)	667 (9.9)	555 (9.5)	112 (12.2)	0.013

Abbreviations: IQR, Interquartile range; MASLD, Metabolic dysfunction‐associated steatotic liver disease; SD, Standard deviation.

TABLE 2.

Poisson with robust variance analysis of the effect of MASLD on cognitive impairment at Wave 3.

	RR	95% CI	p‐value
Model 1 a	1.28	1.06–1.55	0.0105
Model 2 b	1.25	1.03–1.51	0.0256
Model 3 c	1.07	0.89–1.28	0.4808
Model 4 d	1.01	0.83–1.23	0.9151

Abbreviation: MASLD, Metabolic dysfunction‐associated steatotic liver disease.

^a (without adjustment)

^b (adjusted by age, sex, and race)

^c (adjusted by age, sex, race, education, smoking, depression, physical inactivity)

^d (adjusted by age, sex, race, education, smoking, depression, physical inactivity, diabetes, hypertension, obesity, hypertriglyceridemia, and low HDL cholesterol)

We performed a series of linear mixed‐effects models to examine the association between MASLD and global cognitive performance over time, progressively adjusting for potential confounders (Table 3). In the unadjusted model (Model 1), MASLD was significantly associated with lower cognitive performance at baseline (β = −0.193, 95% CI −0.267 to −0.110, p < 0.001), whereas the interaction term with time was not significant. After adjusting for age, sex, and race (Model 2), the association between MASLD and cognition at baseline remained statistically significant (β = −0.105, 95% CI −0.175 to 0.035, p = 0.003), and the time interaction was non‐significant. In Model 3, additional adjustment for socioeconomic and lifestyle variables (education, smoking, depression, and physical inactivity) slightly attenuated the association (β = −0.076, 95% CI −0.142 to −0.010, p = 0.023), without evidence of an interaction with time. However, in the fully adjusted model (Model 4), which further included metabolic and cardiovascular risk factors (obesity, diabetes, hypertension, hypertriglyceridemia, and low HDL cholesterol), the association between MASLD and cognition at baseline was no longer significant (β = −0.039, 95% CI −0.109 to 0.031, p = 0.272). Across all models, time remained consistently associated with cognitive decline, whereas the MASLD*time interaction was not significant (Figure 2).

TABLE 3.

Association of MASLD with cognitive decline.

	β*	95% CI	p‐value
Model 1 (without adjustment)
MASLD	−0.193	−0.267 to −0.110	< 0.001
MASLD*time interaction	0.010	−0.041 to 0.063	0.6768
Time, years	−0.110	−0.128 to −0.090	< 0.001
Model 2 (adjusted by age, sex and race)
MASLD	−0.105	−0.175 to 0.035	0.0032
MASLD*time interaction	0.011	−0.041 to 0.063	0.6766
Time, years	−0.108	−0.127 to −0.089	< 0.001
Model 3 (adjusted by age, sex and race, education, smoking, depression, physical inactivity)
MASLD	−0.076	−0.142 to −0.010	0.0235
MASLD*time interaction	0.007	−0.045 to 0.056	0.7821
Time, years	−0.106	−0.125 to −0.087	< 0.001
Model 4 (adjusted by age, sex and race, education, smoking, depression, physical inactivity, diabetes, hypertension, obesity, hypertriglyceridemia and low HDL cholesterol)
MASLD	−0.039	−0.109 to 0.031	0.2721
MASLD*time interaction	0.005	−0.047 to 0.058	0.8374
Time, years	−0.105	−0.124 to −0.085	< 0.001

Abbreviations: CI, Confidence Interval; HDL, High‐Density Lipoprotein; MASLD, Metabolic dysfunction‐associated steatotic liver disease.

^*β: decline of z‐scores

FIGURE 2.

Mean cognitive function at baseline and cognitive decline between the two waves among those with and without Metabolic dysfunction‐associated steatotic liver disease (MASLD) at baseline. *Adjusted by age, sex, education, smoking, depression, physical inactivity, hypertension, diabetes, obesity, hypertriglyceridemia, and low HDL cholesterol. MASLD, Metabolic dysfunction‐associated steatotic liver disease.

In the sensitivity analysis, we found the same results as in the MASLD analysis. When controlled by all the variables, the association with hepatic steatosis was lost (β = −0.046, 95% CI −0.111 to 0.023, p = 0.19) (Table S1). In addition, the interaction term between steatosis and time (β = 0.001, 95% CI −0.042 to 0.062, p = 0.71) was not significant, indicating that steatosis did not significantly modify the rate of cognitive decline over the follow‐up period. When we analyzed the cognitive domains as the dependent variable, we found that executive function (β = 0.0168, 95% CI −0.0397 to −0.0733, p = 0.5605) and language (β = 0.0135, 95% CI −0.0569 to 0.0840, p = 0.7065) were not associated with MASLD at baseline and were not associated with cognitive decline over time. However, memory domain was significantly associated with MASLD at baseline, even after controlling for sociodemographic and dementia risk factors (β = −0.0690, 95% CI −0.1341 to −0.0040, p = 0.0376), despite not being associated with cognitive decline in the follow‐up (see Tables S2 to S4).

4. DISCUSSION

Here we examined the association between MASLD and cognitive outcomes in a large, well‐characterized Brazilian cohort, called the ELSA‐Brasil study. Our findings show no evidence that MASLD is independently associated with either incident cognitive impairment or cognitive decline over time. These results remained robust in models adjusted for key sociodemographic factors and known non‐metabolic dementia risk factors.

Prior studies have reported associations between hepatic steatosis, predominantly in the form of NAFLD, and poorer cognitive performance. ¹⁸ , ¹⁹ , ²⁰ However, many of these analyses did not fully adjust for the constellation of metabolic abnormalities frequently co‐occurring with NAFLD/MASLD or were cross‐sectional. ⁶ , ¹⁹ , ²¹ Our study adds to more recent work, suggesting that when these covariates are properly controlled for, the independent contribution of MASLD to cognitive outcomes diminishes significantly. ²² , ²³ Our results reinforce the hypothesis that the burden of metabolic dysregulation, rather than hepatic involvement per se, is more pertinent to cognitive aging.

Studies specifically evaluating the effect of MASLD on cognition are scarcer, as this concept is more recent. Recently, four MASLD cohorts showed conflicting results, similar to previous studies on NAFLD. Gao et al. (2024) ²⁴ evaluated the data from the UK Biobank and showed an association between MASLD and incident dementia. However, the definition of steatosis was not performed with imaging or biopsy evaluation, so their results are difficult to interpret or compare with ours. Vataja et al. (2025) evaluated a nationally representative Finnish cohort and showed that in an 11‐year follow‐up, MASLD (steatosis defined by fatty liver index [FLI]) was associated with poorer word‐list learning. ²⁵ Alternatively, Bao et al. (2024) evaluated the UK Biobank dataset and showed that MASLD (with steatosis measured by FLI) was associated with incident vascular dementia but a lower risk of Alzheimer's disease. ²⁶ In the same direction, a post hoc analysis of the Australian study ASPREE and ASPREE‐XT cohort study showed that, in a follow‐up of ≈6 years, MASLD was associated with reduced rates of incidence dementia of controlling for sociodemographic and cardiometabolic variables (Clayton‐Chubb et al., 2024). ²⁷ These results suggest that the relationship between MASLD and cognitive outcomes is complex and may depend on the specific population studied, the definition and ascertainment of MASLD, the methods used to assess cognitive function, and the type of dementia considered. Discrepancies among studies could also be influenced by the degree of adjustment for confounding cardiometabolic conditions, which are both components of MASLD and independent risk factors for cognitive impairment.

Our longitudinal analyses demonstrated that MASLD was initially associated with cognitive decline, but this association disappeared once metabolic components were included in the model. This attenuation highlights the potential for overattributing causality to MASLD in studies not adequately accounting for these covariates. Given the increasing prevalence of MASLD worldwide, especially in middle‐income countries like Brazil, these findings are clinically relevant. They suggest that public health efforts to mitigate cognitive decline should continue to focus on early identification and management of metabolic risk factors, rather than targeting MASLD in isolation.

The lower prevalence of MASLD observed in our sample (19%) compared with estimates from the general population (25%–30%) ¹ likely reflects characteristics of the ELSA‐Brasil cohort. Participants are civil servants from universities and research institutions, who tend to have higher educational attainment and better access to health care, leading to healthier lifestyle profiles and earlier management of metabolic risk factors. In addition, our definition of MASLD was based on standardized biochemical and anthropometric criteria, which may have led to a more conservative classification compared with studies using imaging or liver stiffness measures. Together, these aspects likely contributed to a lower apparent prevalence in our study.

To better understand this association, we performed a sensitivity analysis to evaluate the effect of isolated hepatic steatosis on cognitive performance, controlling for the same sociodemographic and dementia risk factors as the main analysis. Of interest, we found the same results, that is, when controlled for cardiometabolic risk factors, the association with cognition was lost, suggesting that hepatic steatosis may act more as a marker of underlying metabolic dysfunction than as an independent etiological factor in cognitive decline. This is in line with existing literature showing that hepatic steatosis is not associated independently with an increased risk of cognitive impairment and dementia. For example, Gerber et al. (2021), evaluating the CARDIA study, showed that NAFLD was associated inversely with cognitive scores but lost its association with cognition in a 5‐year follow‐up when controlling for cardiovascular diseases, such as hypertension, diabetes, and hypertriglyceridemia. ²³ Therefore, it is plausible that MASLD/NAFLD may act more as a syndromic marker of metabolic burden than as a causative agent in cognitive deterioration.

We perform the same analysis using the cognitive domains available in our sample to verify their association with MASLD, both at baseline and throughout the follow‐up. Although executive function and language were not associated with MALSD, we found that the memory domain (measured through learning, recall, and word recognition) was associated with MASLD at baseline (but not at follow‐up), even after controlling for sociodemographic and dementia risk factors. This is in line with some evidence in the literature that showed that NAFLD ¹⁹ , ²³ is associated with memory function. A recent study by Miao et al. ²⁸ showed, using functional magnetic resonance imaging, that individuals with NAFLD had volume loss in the left hippocampus, in subregions of subiculum and presubiculum, which can partially explain the association with reduced memory function.

The association between MASLD and lower cognitive performance may reflect shared vascular and metabolic mechanisms rather than a direct neurodegenerative effect. MASLD is characterized by systemic inflammation, insulin resistance, oxidative stress, and endothelial dysfunction, all of which have been implicated in cerebral microvascular injury and impaired neurovascular coupling. ²⁹ Furthermore, MASLD encompasses cardiometabolic risk factors—such as hypertension, diabetes, and dyslipidemia—that contribute to both vascular and Alzheimer's‐type pathology. ³⁰ Collectively, these findings support the hypothesis that the MASLD–cognition relationship arises from a complex interplay of metabolic, vascular, and inflammatory processes rather than from a single linear causal pathway.

This study has several important strengths. First, it was conducted in a large, well‐characterized sample, allowing for robust statistical power and the inclusion of multiple relevant confounders in the analytical models. Second, it was performed in a middle‐income country setting, providing valuable data from a population that is underrepresented in the literature on metabolic liver disease and cognition. This enhances the generalizability of the findings to diverse socioeconomic and ethnic contexts, particularly in Latin America. Third, the diagnosis of MASLD was based on liver ultrasonography, an imaging method with higher diagnostic accuracy compared with biochemical or surrogate indices frequently used in epidemiological studies.

Our study has some limitations that must be acknowledged. First, although the cognitive assessment battery employed in ELSA‐Brasil is validated, it may not capture subtle or domain‐specific impairments, particularly in younger individuals. In addition, despite rigorous adjustment, residual confounding from unmeasured variables such as diet, inflammatory markers, or genetic susceptibility cannot be ruled out. The relatively short follow‐up period and midlife age range of the cohort may also have limited our ability to detect the long‐latency effects of metabolic dysfunction on cognition. Furthermore, participants in ELSA‐Brasil are public sector employees, which limits generalizability to the Brazilian population. However, their schooling levels also reduce confounding effects of extreme social deprivation and illiteracy on cognitive testing, conferring internal validity. Although the ELSA‐Brasil participants are predominantly middle‐aged adults, this allows the investigation of early, subclinical cognitive changes associated with MASLD, which are critical for understanding and preventing later‐life cognitive decline.

In conclusion, findings from this large, well‐characterized Brazilian cohort indicate that MASLD is not an independent risk factor for cognitive decline. Rather, the association appears to be mediated by the constellation of cardiometabolic risk factors that define MASLD. These results support the need for continued focus on the prevention and management of metabolic comorbidities as a strategy for mitigating cognitive aging.

Future studies with longer follow‐ups, more sensitive cognitive testing, and more precise liver phenotyping—such as elastography or magnetic resonance imaging—are warranted to further clarify the role of liver health in cognitive trajectories across the life course.

CONFLICT OF INTEREST STATEMENT

R.M. Castilhos served on the scientific advisory board for Eli Lilly. L.G. Sampaio has no conflict to declare. N. Feter has no conflict to declare. B. Duncan has no conflict to declare. M.I. Schmidt has no conflict to declare. Any author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants in this study provided written informed consent.

Supporting information

Supporting Information

Supporting Information

Castilhos RM, Sampaio LG, Feter N, Duncan B, Schmidt MI. Metabolic dysfunction–associated steatotic liver disease as a risk factor for cognitive decline: Findings from the ELSA‐Brasil cohort. Alzheimer's Dement. 2026;22:e71088. 10.1002/alz.71088