The association between liver fibrosis score and incident dementia: A nationwide retrospective cohort study

Abstract

BACKGROUND

We assessed the relationship of liver fibrosis score with incident dementia in a large, national sample.

METHODS

For this retrospective cohort study, data of dementia‐free individuals aged 40‐69 years were derived from electronic records of the largest healthcare provider in Israel. The association between liver fibrosis score (FIB‐4), assessed from routine laboratory measurements, and incident dementia was explored through multivariate cox regression models.

RESULTS

Of the total sample (N = 826,578, mean age 55 ± 8 years at baseline), 636,967 (77%) had no fibrosis, 180,114 (21.8%) had inconclusive fibrosis status and 9497 (1.2%) had high risk for advanced fibrosis. Over a median follow‐up of 17 years, 41,089 dementia cases were recorded. Inconclusive liver fibrosis and advanced fibrosis were associated with increased dementia risk (HR = 1.09, 95%CI: 1.07–1.11 and HR = 1.18, 95%CI: 1.10–1.27, respectively). This association remained robust through seven sensitivity analyses.

CONCLUSIONS

Liver fibrosis assessed through a serum‐based algorithm may serve as a risk factor for dementia in the general population.

Highlights

• Liver fibrosis may predict dementia diagnosis in the general population.

• Inconclusive liver fibrosis was associated with 9% increased dementia risk.

• Advanced liver fibrosis was associated with 18% increased dementia risk.

• Findings remained robust in sensitivity analyses and after adjustments.

1. INTRODUCTION

Dementia is an increasing public health concern. ¹ Because treatment options for dementia are limited, there is a growing need to establish modifiable risk factors for this condition. ²

Increasing evidence points to the role of chronic liver diseases (CLD) in dementia pathogenesis, through multiple mechanisms beyond that of encephalopathy. ³ CLD represents a common pathological outcome of several etiologies, including chronic infection by hepatotropic viruses (mainly hepatitis B and C) and chronic toxic injury, mainly due to excessive alcohol consumption. Yet the most prevalent etiology of CLD is metabolic dysfunction‐associated steatotic liver disease (MASLD), with an estimated global prevalence of 38% among adults. ⁴ MASLD, previously known as non‐alcoholic fatty liver disease (NAFLD), has been associated with multiple health outcomes, including cardiometabolic diseases ⁵ and neurodegeneration. ⁶ However, within the spectrum of MASLD, the presence of liver fibrosis has been shown to constitute the most important factor in determining liver‐related complications ⁷ as well as multiple extrahepatic conditions. ⁸

The advanced stages of liver fibrosis are most often undiagnosed because liver disease in these stages is usually clinically silent. However, accumulating evidence suggests that liver fibrosis, mostly associated with MASLD, is prevalent among individuals in the general population without known liver disease, with estimates ranging between 5.6% and 7.5%. ⁹ , ¹⁰ Yet, fibrosis is reversible through weight reduction and pharmacological interventions. ¹¹ Hence, screening for liver fibrosis in the general population has been considered. ¹⁰ , ¹²

Assessment of liver fibrosis has been conducted using liver biopsy, which is invasive and costly and may be inaccurate due to sampling error. ¹³ Over the last few years, noninvasive methods of assessing liver fibrosis have been developed, including imaging techniques such as transient elastography and serum‐based scores. ¹⁴ Specifically, serum‐based fibrosis scores show promise for screening of liver fibrosis in the general population because they require serum measures that are usually already available or rely on a simple blood draw and have low costs compared to other assessment methods. ¹² FIB‐4 is the serum‐based score primarily utilized in clinical practice for risk stratification and exclusion of advanced fibrosis among individuals with MASLD and is endorsed as a first‐line screening tool in clinical practice. ¹⁵ , ¹⁶ Compared to liver biopsy, FIB‐4 shows a summary area under the receiver operating characteristic (AUROC) of 84% for diagnosing advanced fibrosis. ¹⁷

RESEARCH IN CONTEXT

1. Systematic review: A literature review was conducted to identify studies assessing the implications of chronic liver disease for brain health. Recent evidence suggests that chronic liver disease, and liver fibrosis in particular, may be related to cognitive function, imaging markers of neurodegeneration, and Alzheimer's disease pathology. Few studies have explored the association of liver fibrosis with dementia incidence, and their findings are conflicting.

2. Interpretation: Our findings from a national, population‐based cohort suggest that liver fibrosis, assessed through a validated serum‐based algorithm (FIB‐4), may serve as a predictor of incident dementia in the general population.

3. Future directions: Future research is required to validate the predictive value of FIB‐4 in diverse populations and to clarify the biological mechanisms that underlie the association between liver fibrosis and dementia. Future exploration is also warranted to examine whether improvement in liver fibrosis through lifestyle and pharmacological interventions may ameliorate dementia risk.

Although many studies have focused on the implications of liver fibrosis among individuals with confirmed MASLD, ¹⁸ accumulating data stress the utility of liver fibrosis as a predictor of multiple health outcomes in the general population. This includes cardiometabolic risk factors, ¹⁹ cardiovascular and cerebrovascular morbidity and mortality, ²⁰ , ²¹ , ²² and postoperative mortality. ²³ Evidence additionally suggests a relationship between liver fibrosis and reduced cognitive performance, neurodegeneration, ²⁴ , ²⁵ , ²⁶ and amyloid beta (Aβ), and tau brain pathology, ²⁷ thus stressing the possible contribution of liver fibrosis in the general population to dementia pathogenesis. Moreover, liver fibrosis, particularly due to MASLD, shares common risk factors with brain aging, including obesity, hypertension, and systemic atherosclerosis. ²⁸ Some recent studies additionally assessed the link between liver fibrosis and dementia risk; however, findings are contradictory. ²⁴ , ²⁹ , ³⁰

We aimed to examine the associations between liver fibrosis and incident dementia in a population‐based national cohort study using a digital dataset from Clalit Health Services (CHS), the largest health maintenance organization (HMO) in Israel. The National Health Insurance law in Israel mandates all residents in the country to join one of the four official HMOs and prohibits discrimination based on any background or circumstances in the provision of healthcare services. This ensures equitable access to healthcare and minimizes selection bias in research based on CHS data. Moreover, loss to follow‐up from CHS due to transfer to a different HMO is negligible. In addition, the use of health records from an HMO enabled the longest follow‐up for the association between liver fibrosis and incident dementia to date. We hypothesize that liver fibrosis will be related to increased dementia risk in this “real‐world” setting of Israeli adults.

2. METHODS

2.1. Study sample

The CHS is the largest HMO in Israel, providing a broad array of private and public healthcare services for nearly 5 million enrollees, representing ∼52% of the general Israeli population. The CHS electronic health registry continuously harmonizes individual‐level information from clinical encounters, laboratory tests, and medication claims.

The initial study population included all 1,098,582 CHS members aged 40 to 69 years who were enrolled for at least 1 year on January 1, 2006 (index date; follow‐up initiation). Of those, 854,096 had information on all the laboratory measures needed to calculate the FIB‐4 score. We further excluded 3341 who had already been diagnosed with dementia at cohort entry. Additionally, we excluded 9069 with recorded excessive alcohol consumption and those with liver and gastrointestinal diseases (eg, hepatitis B and C, ulcerative colitis) to focus on the MASLD etiology and minimize an effect of these conditions on the serum measures included in the FIB‐4 algorithm. Hence, our final sample included 826,578 participants (Figure 1). The follow‐up period ended on December 31, 2022.

FIGURE 1.

Flow diagram of study sample.

The study protocols were approved by the CHS ethical review board.

2.2. Assessment of liver fibrosis

To calculate FIB‐4, we used the most recent laboratory information available 5 years prior to the index date (ie, between January 1, 2001 and December 31, 2005). The FIB‐4 index was used to estimate participants' risk for advanced liver fibrosis using the following formula: Age(years) × AST[U/L]/(platelets [10⁹/L] × (ALT[U/L])^1/2). ³¹ The FIB‐4 score has been widely validated ³² and recommended by the European Association for the Study of the Liver (EASL). ¹⁴ All the FIB‐4 components were measured during routine blood tests conducted periodically in the CHS. Extreme values were treated as valid, and there were no erroneous values (ie, ≤0). Participants were classified into three levels of fibrosis using established cut‐off values: FIB‐4 < 1.3 indicated no fibrosis, FIB‐4 1.3 to 2.66 was inconclusive with regard to the presence of advanced fibrosis, and FIB‐4 ≥ 2.67 indicated high risk of advanced fibrosis. ³³

2.3. All‐cause dementia ascertainment

Dementia diagnosis was based on the internal dementia registry of CHS. In this registry, dementia is recorded if at least one of the following criteria is met: (1) chronic diagnosis of dementia according to ICD‐9 codes; (2) a diagnosis of dementia in a hospital discharge letter; (3) purchase of at least four monthly prescriptions of an anti‐dementia medication within 12 months (based on ACT codes N06DA02 and No6DA03). A list of International classifiaction of disease, 9th revision (ICD‐9) codes included for the dementia diagnoses is presented in Supplementary Table 1.

2.4. Covariate ascertainments

Potential confounders were assessed at baseline and included sociodemographic, lifestyle, and health factors. Age was categorized into 5‐year bands. For ethnicity, participants were grouped as either Arab, Orthodox Jew, or Jew. Since no direct measure of ethnicity exists in CHS, a proxy measure was ascertained according to the majority ethnic group residing in the catchment area of each primary care clinic based on periodic Israeli Central Bureau of Statistics reports. ³⁴ The peripherality index was constructed based on the Israeli Bureau of Statistics' “Peripherality Index of Localities and Local Authorities,” which takes into account population size, proximity to other local authorities in the country, and the proximity to the Tel‐Aviv District. ³⁵ Participants were grouped into the following peripherality categories: 1 to 4, 5 to 7, and 8 to 10, indicating distant peripherality, proximate peripherality, and centrality, respectively.

Socioeconomic status designation, grouped into four categories, was based on members' listed home address. The POINTS Location Intelligence Company uses commercial propriety information to continuously augment neighborhood‐level socioeconomic information gathered during a 2008 Israel Central Bureau of Statistics census. Participants were grouped into the following categories: 1 to 3, 4 to 5, 6 to 7, and 8 to 10, indicating lowest, medium‐low, medium‐high, and highest status, respectively.

Baseline morbidity was based on pre‐existing diagnoses specified in the Clalit chronic disease register. The CHS epidemiology unit applies disease‐specific algorithms to compile and maintain electronic health registers for over 100 chronic conditions, combining community and hospital‐discharge diagnoses, laboratory data, medication claims, and linkage with Ministry of Health national registers. ³⁶

2.5. Statistical analysis

Sociodemographic, clinical, and laboratory characteristics were evaluated in the total sample and by FIB‐4 categories. Normally distributed values were presented as mean ± SD, skewed variables were presented as median (Q1, Q3), and categorical variables were presented as N (%). We used Cox regression models to assess the relationship between FIB‐4 categories and incident dementia. Participants who did not develop dementia were censored at the date of death, upon leaving the HMO, or at the end of follow‐up.

We initially adjusted for sociodemographic characteristics, namely, age, sex, ethnicity, and peripherality, and additionally adjusted for lifestyle and health characteristics, namely, smoking, obesity, and prevalence of the following health conditions: hyperlipidemia, ischemic heart disease (IHD), coronary heart failure, hypertension, stroke, carotid artery stenosis, diabetes, depression, anxiety, and Parkinson's disease.

Further, we tested whether age, sex, socioeconomic status, ethnicity, obesity, smoking, prevalent diabetes, IHD, and history of stroke modified the link between estimated liver fibrosis and incident dementia by including interaction terms in the models. When interaction terms showed a p for interaction < 0.05 in our fully adjusted models, we further conducted subgroup analyses stratifying the analyses by the modifier. Seven sensitivity analyses were performed: (1) using a cut‐off of 2.0 instead of 1.3 for differentiation between no fibrosis and inconclusive fibrosis among individuals aged ≥65 years; these age‐specific thresholds have been proposed to increase specificity for advanced fibrosis in this age group ³⁷ ; (2) including only individuals with non‐alcoholic fatty liver disease (NAFLD) diagnosis; (3) excluding participants who were diagnosed with dementia during the first 5 years of follow‐up (2005 to 2009); (4) excluding individuals who purchased glucocorticoid in 2005 and between 2000 and 2005; (5) excluding individuals with a breast cancer diagnosis at baseline or tamoxifen claims between 2000 and 2005; (6) excluding prevalent stroke or carotid stenosis diagnosis at baseline; and (7) only those whose FIB‐4 scores were based on laboratory data from 2005 (rather than 2000 to 2005). Supplementary Table 2 outlines the rationale behind the selection of potential confounding variables and sensitivity analyses.

All p values were based on likelihood‐ratio tests. All statistical analyses were performed using Stata version 16.0 (StataCorp).

3. RESULTS

Baseline characteristics of the study sample overall and by fibrosis level are presented in Table 1. The mean age of the study participants at cohort entry was 55 ± 8 years, and 468,737 (56.7%) were women. Among the total sample, 86.7% were Jews and 11.8% were Arabs, and the majority of participants were at medium‐high socioeconomic status (30.9%). Furthermore, NAFLD was present in 1.8% of the sample, and the FIB‐4 score indicating advanced fibrosis was present in 9497 subjects (1.1%), while the mean FIB‐4 score was 1.07 ± 1.05. The baseline characteristics of the subsample of participants who were not included in our analyses are presented in Supplementary Table 3.

TABLE 1.

Baseline sociodemographic and clinical characteristics of study participants by fibrosis status (N = 826,578).

	FIB‐4 score categories
	No fibrosis (<1.3)	Inconclusive (1.3 to 2.66)	Fibrosis (≥2.67)	Total population
Characteristics	N = 636,967	N = 180,114	N = 9,497	N = 826,578
Sex, female	373,419 (58.62%)	90,849 (50.44%)	4,469 (47.06%)	468,737 (56.71%)
FIB‐4 score	0.86 [0.69 to 1.05]	1.55 [1.41 to 1.80]	3.21 [2.88 to 4.00]	0.97 [0.74 to 1.27]
Age, years	52.95 ± 7.76	59.99 ± 7.13	60.91 ± 7.33	54.58 ± 8.18
Age, years, categories
35 to 44	118,205 (18.56%)	6,402 (3.55%)	364 (3.83%)	124,971 (15.12%)
45 to 54	272,339 (42.76%)	37,359 (20.74%)	1,712 (18.03%)	311,410 (37.67%)
55 to 64	191,919 (30.13%)	81,393 (45.19%)	3,849 (40.53%)	277,161 (33.53%)
65 to 74	54,504 (8.56%)	54,960 (30.51%)	3,572 (37.61%)	113,036 (13.68%)
Ethnicity, Arab *	80,043 (12.58%)	16,192 (9.00%)	1,009 (10.63%)	97,244 (11.78%)
Peripherality index
Distant periphery	101,790 (15.98%)	28,166 (15.64%)	1,627 (17.13%)	131,583 (15.92%)
Proximate periphery	282,873 (44.41%)	78,735 (43.71%)	4,290 (45.17%)	365,898 (44.27%)
Centrality	249,916 (39.24%)	72,673 (40.35%)	3,554 (37.42%)	326,143 (39.46%)
Socioeconomic status
Lowest	94,334 (14.81%)	22,081 (12.26%)	1,481 (15.59%)	117,896 (14.26%)
Medium‐low	171,487 (26.92%)	48,787 (27.09%)	2,844 (29.95%)	223,118 (26.99%)
Medium‐high	195,574 (30.70%)	57,123 (31.71%)	2,747 (28.92%)	255,444 (30.90%)
Highest	138,122 (21.68%)	42,585 (23.64%)	1,833 (19.30%)	182,540 (22.08%)
AST, IU/L	19.40 ± 6.34	25.08 ± 11.00	46.38 ± 56.75	20.95 ± 10.36
ALT, IU/L	22.07 ± 12.88	22.83 ± 17.81	34.33 ± 63.12	22.38 ± 15.64
GGT, IU/L	27.56 ± 31.05	30.33 ± 42.85	62.37 ± 119.25	28.65 ± 36.79
AFP, ng/mL	2.50 (1.80 to 3.70)	2.60 (1.90 to 3.91)	3.30 (2.30 to 5.00)	2.51 (1.90 to 3.88)
Ferritin, ng/mL	76.07 ± 143.35	99.67 ± 152.81	155.64 ± 355.37	82.75 ± 151.83
Platelets, 109/L	271.10 ± 63.46	200.53 ± 41.73	141.53 ± 57.77	254.24 ± 67.19
Non‐alcoholic fatty liver disease	10,773 (1.69%)	4,024 (2.23%)	408 (4.30%)	15,205 (1.84%)
Prevalent malignancy	37,798 (5.93%)	17,996 (9.99%)	1,574 (16.57%)	57,368 (6.94%)
Prevalent breast cancer	10,390 (1.63%)	5,280 (2.93%)	362 (3.81%)	16,032 (1.94%)
Tamoxifen use	4,823 (%)	2,911 (%)	241 (%)	7,975 (%)
Glucocorticoid use	175,754 (27.59%)	56,085 (31.14%)	3,426 (36.07%)	235,265 (28.46%)
Prevalent diabetes	87,819 (13.79%)	28,828 (16.01%)	2,151 (22.65%)	118,798 (14.37%)
Obesity	82,670 (12.98%)	21,236 (11.79%)	1,341 (14.12%)	105,247 (12.73%)
Hyperlipidemia	214,069 (33.61%)	75,701 (42.03%)	3,818 (40.20%)	293,588 (35.52%)
History of depression	35,329 (5.55%)	11,051 (6.14%)	624 (6.57%)	47,004 (5.69%)
History of anxiety	27,775 (4.36%)	9,249 (5.14%)	476 (5.01%)	37,500 (4.54%)
History of alcohol abuse	0 (0.00%)	0 (0.00%)	0 (0.00%)	0 (0.00%)
Ever smoking	186,879 (29.34%)	49,011 (27.21%)	2,657 (27.98%)	238,547 (28.86%)
History of drug abuse	2,604 (0.41%)	562 (0.31%)	60 (0.63%)	3,226 (0.39%)
Parkinson's disease	3,786 (0.59%)	1,427 (0.79%)	154 (1.62%)	5,367 (0.65%)
Ischemic heart disease	52,428 (8.23%)	30,331 (16.84%)	2,223 (23.41%)	84,982 (10.28%)
Congestive heart failure	5,353 (0.84%)	3,725 (2.07%)	481 (5.06%)	9,559 (1.16%)
Hypertension	177,265 (27.83%)	70,976 (39.41%)	4,178 (43.99%)	252,419 (30.54%)
History of stroke	16,073 (2.52%)	7,624 (4.23%)	575 (6.05%)	24,272 (2.94%)

Notes: Values are presented as mean ± SD, median [Q1, Q3] or N (%).

Abbreviations: AFP, alpha–fetoprotein; ALT, alanine transaminase; AST, aspartate transaminase; GGT, gamma‐glutamyltransferase.

^*Otherwise Jews.

During a mean follow‐up time of 15.2 ± 4.0 years and a median of 17 years, 41,089 (5%) were diagnosed with dementia. Statistically significant differences in dementia incidence were observed between the three liver fibrosis categories (Figure 2). After adjustment for age, sex, ethnicity, and peripherality, inconclusive fibrosis status was related to a 5% increase in dementia risk, and high risk for advanced fibrosis was associated with an 18% greater risk for dementia compared to the risk of those with no fibrosis (hazard ratio [HR] = 1.05, 95% confidence interval [CI]: 1.03 to 1.07 and HR = 1.18, 95% CI: 1.10 to 1.26, respectively). Further adjustment for sociodemographic, lifestyle, and health factors resulted in similar associations (HR = 1.09, 95% CI: 1.07 to 1.11 and HR = 1.18, 95% CI: 1.10 to 1.27, for inconclusive and high risk of fibrosis vs no fibrosis, respectively) (Table 2).

FIGURE 2.

Kaplan–Meier curves for cumulative risk of dementia by liver fibrosis status.

TABLE 2.

Association between liver fibrosis and dementia risk in total sample.

	Model 1			Model 2
Fibrosis category *	N	Events/person years	HR (95% CI)	N	Events/person years	HR (95% CI)
No fibrosis	598,216	22,991/9,837,282	1.00 (ref.)	598,216	22,991/9,837,282	1.00 (ref.)
Inconclusive	170,287	15,245/2,641,322	1.05 (1.03 to 1.07)	170,287	15,245/2,641,322	1.09 (1.07 to 1.11)
Fibrosis	8,897	826/118,811	1.18 (1.10 to 1.26)	8,897	826/118,811	1.18 (1.10 to 1.27)

Notes: Model 1: Adjusted for age, sex, ethnicity, and peripherality.

Model 2: Additionally adjusted for smoking, obesity, hyperlipidemia, ischemic heart disease, hypertension, history of stroke, carotid artery stenosis, congestive heart failure, diabetes, depression, anxiety, and Parkinson's disease.

Bold font indicates p value ≤ 0.05.

^*FIB‐4 < 1.3, 1.3 to 2.66, and ≥2.67 indicating no fibrosis, inconclusive, and fibrosis, respectively.

We found statistically significant interactions of fibrosis status with sex, age, socioeconomic status, ethnicity, prevalent diabetes, IHD, and stroke with regard to their relationship with dementia risk (Table 3). Stratified models revealed that the association between estimated liver fibrosis status and dementia risk was somewhat higher in women compared to men; however, these differences were small and may not have clinical significance (HR = 1.11, 95% CI: 1.08 to 1.14 and HR = 1.22, 95% CI: 1.10 to 1.35 for inconclusive and high risk of advanced fibrosis, respectively, compared to no fibrosis in women compared to HR = 1.08, 95% CI: 1.04 to 1.11 and HR = 1.17, 95% CI: 1.06 to 1.29 in men). The strongest associations between estimated liver fibrosis and dementia incidence were in the 45‐ to 54‐year age group, where compared to individuals with no fibrosis, those with inconclusive fibrosis had 18% (HR = 1.18, 95% CI: 1.08 to 1.30) increased dementia risk, and those at high risk of advanced fibrosis had more than two‐fold risk for dementia (HR = 2.17; 95% CI: 1.61‐2.93). Lastly, we found significant interactions between prevalent diabetes, IHD and stroke history and estimated liver fibrosis status. Compared to individuals with no fibrosis, those with inconclusive score or at high risk of advanced fibrosis had a higher risk of dementia in those without prevalent diabetes (HR = 1.11, 95% CI: 1.09 to 1.14 and HR = 1.24, 95% CI: 1.14 to 1.34, respectively) but not in those with prevalent diabetes (HR = 1.02, 95% CI: 0.98 to 1.07 and HR = 1.04, 95% CI: 0.91 to 1.19, respectively). Similarly, only individuals with no history of stroke showed significant associations between estimated liver fibrosis status and dementia risk (HR = 1.10, 95% CI: 1.08 to 1.13 and HR = 1.18, 95% CI: 1.10 to 1.27 in participants with no stroke history vs HR = 0.96, 95% CI: 0.89 to 1.03 and HR = 1.16, 95% CI: 0.94 to 1.43 in those with stroke history, comparing inconclusive and advanced fibrosis to no fibrosis, respectively). Lastly, stronger associations were observed in individuals without IHD (HR = 1.10, 95% CI: 1.08 to 1.13 and HR = 1.20, 95% CI: 1.10 to 1.30) than in those with IHD (HR = 1.05, 95% CI: 1.00 to 1.10 and HR = 1.15, 95% CI: 1.01 to 1.31 for inconclusive and advanced fibrosis compared to no fibrosis, respectively) (Table 3).

TABLE 3.

Association between liver fibrosis and dementia risk stratified by participants’ characteristics.

	Model 1				Model 2
Fibrosis category1	N	Events/person years	HR (95% CI)	P for interaction	N	Events/person years	HR (95% CI)	P for interaction
Sex
Men
No fibrosis	247,307	9,067/4,004,074	1.00 (ref.)	<0.0001	247,307	9,067/4,004,074	1.00 (ref.)	<0.0001
Inconclusive	84,244	7,077/1,287,877	1.04 (1.01 to 1.08)		84,244	7,077/1,287,877	1.08 (1.04 to 1.11)
Fibrosis	4,690	417/62,207	1.16 (1.05 to 1.28)		4,690	417/62,207	1.17 (1.06 to 1.29)
Women
No fibrosis	350,909	13,924/5,833,208	1.00 (ref.)		350,909	13,924/5,833,208	1.00 (ref.)
Inconclusive	86,043	8,168/1,353,445	1.07 (1.04 to 1.10)		86,043	8,168/1,353,445	1.11 (1.08 to 1.14)
Fibrosis	4,207	409/56,604	1.22 (1.11 to 1.35)		4,207	409/56,604	1.22 (1.10 to 1.35)
Age, years
35 to 44
No fibrosis	109,899	406/1,889,307	1.00 (ref.)	<0.0001	109,899	406/1,889,307	1.00 (ref.)	0.0002
Inconclusive	5,892	31/100,921	1.41 (0.97 to 2.03)		5,892	31/100,921	1.42 (0.99 to 2.05)
Fibrosis	332	3/5,305	2.56 (0.82 to 7.97)		332	3/5,305	2.20 (0.70 to 6.86)
45 to 54
No fibrosis	255,256	3,142/4,325,314	1.00 (ref.)		255,256	3,142/4,325,314	1.00 (ref.)
Inconclusive	35,083	562/585,562	1.16 (1.06 to 1.27)		35,083	562/585,562	1.18 (1.08 to 1.30)
Fibrosis	1,590	44/23,561	2.35 (1.74 to 3.16)		1,590	44/23,561	2.17 (1.61 to 2.93)
55 to 64
No fibrosis	181,330	10,446/2,900,400	1.00 (ref.)		181,330	10,446/2,900,400	1.00 (ref.)
Inconclusive	77,046	5,260/1,216,806	1.06 (1.02 to 1.09)		77,046	5,260/1,216,806	1.09 (1.05 to 1.13)
Fibrosis	3,616	249/48,812	1.23 (1.08 to 1.39)		3,616	249/48,812	1.21 (1.06 to 1.37)
65 to 74
No fibrosis	51,731	8,997/722,262	1.00 (ref.)		51,731	8,997/722,262	1.00 (ref.)
Inconclusive	52,266	9,392/738,033	1.04 (1.01 to 1.07)		52,266	9,392/738,033	1.08 (1.04 to 1.11)
Fibrosis	3,359	530/41,133	1.12 (1.02 to 1.22)		3,359	530/41,133	1.12 (1.03 to 1.23)
Diabetes
Yes
No fibrosis	81,650	7,100/1,217,716	1.00 (ref.)	<0.0001	81,650	7,100/1,217,716	1.00 (ref.)	<0.0001
Inconclusive	27,016	3,701/370,094	1.03 (0.99 to 1.07)		27,016	3,701/370,094	1.02 (0.98 to 1.07)
Fibrosis	1,992	224/22,544	1.08 (0.94 to 1.23)		1,992	224/22,544	1.04 (0.91 to 1.19)
No
No fibrosis	516,566	15,891/8,619,566	1.00 (ref.)		516,566	15,891/8,619,566	1.00 (ref.)
Inconclusive	143,271	11,544/2,271,229	1.11 (1.08 to 1.14)		143,271	11,544/2,271,229	1.11 (1.09 to 1.14)
Fibrosis	6,905	602/96,267	1.25 (1.15 to 1.36)		6,905	602/96,267	1.24 (1.14 to 1.34)
Ischemic heart disease
Yes
No fibrosis	49,129	3,863/720,337	1.00 (ref.)	<0.0001	49,129	3,863/720,337	1.00 (ref.)	0.0001
Inconclusive	28,611	3,377/399,094	1.00 (0.96 to 1.05)		28,611	3,377/399,094	1.05 (1.00 to 1.10)
Fibrosis	2,077	238/24,573	1.13 (0.99 to 1.29)		2,077	238/24,573	1.15 (1.01 to 1.31)
No
No fibrosis	549,087	19,128/9,116,945	1.00 (ref.)		549,087	19,128/9,116,945	1.00 (ref.)
Inconclusive	141,676	11,868/2,242,228	1.06 (1.03 to 1.08)		141,676	11,868/2,242,228	1.10 (1.08 to 1.13)
Fibrosis	6,820	588/94,238	1.17 (1.08 to 1.27)		6,820	588/94,238	1.20 (1.10 to 1.30)
Stroke history
Yes
No fibrosis	15,101	2,068/199,232	1.00 (ref.)	<0.0001	15,101	2,068/199,232	1.00 (ref.)	<0.0001
Inconclusive	7,206	1,273/90,503	0.92 (0.85 to 0.99)		7,206	1,273/90,503	0.96 (0.89 to 1.03)
Fibrosis	528	91/5,173	1.21 (0.98 to 1.49)		528	91/5,173	1.16 (0.94 to 1.43)
No
No fibrosis	583,115	20,923/9,638,049	1.00 (ref.)		583,115	20,923/9,638,049	1.00 (ref.)
Inconclusive	163,081	13,972/2,550,819	1.07 (1.04 to 1.09)		163,081	13,972/2,550,819	1.10 (1.08 to 1.13)
Fibrosis	8,369	735/113,639	1.17 (1.09 to 1.26)		8,369	735/113,639	1.18 (1.10 to 1.27)

Notes: Model 1: Adjusted for age, sex, ethnicity, and peripherality.

Model 2: Additionally adjusted for smoking, obesity, hyperlipidemia, ischemic heart disease, hypertension, history of stroke, carotid artery stenosis, congestive heart failure, diabetes, depression, anxiety, and Parkinson's disease.

Bold font indicates p value ≤ .05.

The results of our sensitivity analyses are presented in Table 4. Similar findings were obtained in analyses using age‐specific cut‐offs and when excluding subjects with dementia diagnosis in the first 5 years of follow‐up, persons receiving glucocorticoids and those with breast cancer or receiving tamoxifen, persons with a history of stroke or carotid stenosis, and when restricting to those whose serum measures for the FIB‐4 calculation were obtained from laboratory tests done in 2005 only (Table 4). Despite similar or even larger estimates of the associations, the analysis among individuals with NAFLD was nonsignificant, probably due to the small subsample of participants with both NAFLD and high risk of advanced fibrosis (n = 40).

TABLE 4.

Association between liver fibrosis and dementia risk, sensitivity analyses.

	Model 1			Model 2
	N	Events/person years	HR (95% CI)	N	Events/person years	HR (95% CI)
1. Using age‐specific cutoffs
No fibrosis	640,989	30,629/10,445,061	1.00 (base)	640,989	30,629/10,445,061	1.00 (base)
Inconclusive	127,514	7,607/2,033,543	1.08 (1.05 to 1.11)	127,514	7,607/2,033,543	1.11 (1.08 to 1.14)
Fibrosis	8,897	826/118,811	1.17 (1.09 to 1.25)	8,897	826/118,811	1.16 (1.08 to 1.24)
2. Include only those with NAFLD at baseline
No fibrosis	10,142	578/166,041	1.00 (ref.)	10,142	578/166,041	1.00 (ref)
Inconclusive	3,796	391/59,095	1.01 (0.88 to 1.15)	3,796	391/59,095	1.02 (0.89 to 1.16)
Fibrosis	381	40/4,803	1.37 (0.99 to 1.89)	381	40/4,803	1.32 (0.95 to 1.83)
3. Exclude incident dementia cases occurring in the first 5 years of follow‐up
No fibrosis	595,264	20,039/9,828,082	1.00 (ref.)	595,264	20,039/9,828,082	1.00 (ref.)
Inconclusive	168,330	13,288/2,635,176	1.05 (1.03 to 1.08)	168,330	13,288/2,635,176	1.09 (1.06 to 1.11)
Fibrosis	8,734	663/118,332	1.12 (1.03 to 1.21)	8,734	663/118,332	1.13 (1.04 to 1.22)
4a. Exclude those with glucocorticoid claims 2005
No fibrosis	542,038	20,142/8,932,669	1.00 (ref.)	542,038	20,142/8,932,669	1.00 (ref.)
Inconclusive	152,668	13,446/2,377,859	1.05 (1.03 to 1.08)	152,668	13,446/2,377,859	1.09 (1.07 to 1.12)
Fibrosis	7,654	708/104,422	1.16 (1.08 to 1.25)	7,654	708/104,422	1.16 (1.08 to 1.25)
4b. Exclude those with glucocorticoid claims 2000 to 2005
No fibrosis	434,201	15,046/7,155,105	1.00 (ref.)	434,201	15,046/7,155,105	1.00 (ref.)
Inconclusive	117,456	9,816/1,833,944	1.04 (1.02 to 1.07)	117,456	9,816/1,833,944	1.09 (1.06 to 1.12)
Fibrosis	5,721	509/78,419	1.15 (1.05 to 1.25)	5,721	509/78,419	1.17 (1.07 to 1.27)
5. Exclude those with breast cancer at baseline or tamoxifen claims 2000 to 2005
No fibrosis	587,943	22,430/9,682,849	1.00 (ref.)	587,943	22,430/9,682,849	1.00 (ref.)
Inconclusive	165,040	14,795/2,567,637	1.05 (1.03 to 1.08)	165,040	14,795/2,567,637	1.09 (1.07 to 1.12)
Fibrosis	8,538	801/115,016	1.19 (1.10 to 1.27)	8,538	801/115,016	1.19 (1.11 to 1.28)
6. Exclude those with history of stroke or carotid stenosis
No fibrosis	578,439	20,506/9,566,410	1.00 (ref.)	578,439	20,506/9,566,410	1.00 (ref.)
Inconclusive	160,718	13,644/2,516,046	1.07 (1.04 to 1.09)	160,718	13,644/2,516,046	1.10 (1.08 to 1.13)
Fibrosis	8,219	716/111,797	1.17 (1.09 to 1.26)	8,219	716/111,797	1.18 (1.10 to 1.27)
7. Include only those with FIB‐4 scores bases on labs in 2005
No fibrosis	361,712	15,892/5,921,081	1.00 (ref.)	361,712	15,892/5,921,081	1.00 (ref.)
Inconclusive	117,709	11,327/1,811,781	1.04 (1.01 to 1.07)	117,709	11,327/1,811,781	1.08 (1.05 to 1.11)
Fibrosis	6,510	646/84,710	1.20 (1.11 to 1.30)	6,510	646/84,710	1.20 (1.11 to 1.30)

Notes: Model 1: Adjusted for age, sex, ethnicity and peripherality.

Model 2: additionally adjusted for smoking, obesity, hyperlipidemia, Ischemic heart disease, hypertension, history of stroke, carotid artery stenosis, congestive heart failure, diabetes, depression, anxiety, and Parkinson's disease.

Abbreviations: HR, hazard ratio; NAFLD, non‐alcoholic fatty liver disease.

Bold font indicates p value ≤ 0.05.

4. DISCUSSION

In this retrospective cohort study, we examined the associations between estimated liver fibrosis and dementia risk in a nationally representative sample of middle‐age and older adults in Israel. Our findings demonstrated that individuals with advanced liver fibrosis according to FIB‐4 had an 18% higher risk of dementia compared to those without liver fibrosis, and those with an inconclusive result of liver fibrosis had a 9% higher risk. The association between estimated liver fibrosis and dementia risk was significantly modified by sex, age, and the presence of comorbidities, with slightly stronger associations among women, middle‐aged individuals, and those free of diabetes, cardiovascular disease, and stroke. The observed associations remained robust in seven sensitivity analyses, including after restriction to participants with NAFLD and exclusion of dementia cases occurring in the first years of follow‐up.

Previous findings regarding the link between liver fibrosis and dementia risk are sparse and conflicting. Like our exploration, a prior study among the UK Biobank participants that was comparable to our study with regard to the ascertainment of liver fibrosis and incident dementia, as well as the mean age of the participants, found an excessive dementia risk of 52% in those with liver fibrosis compared to those without, an estimate larger compared to the 18% found in our study. ³⁸ In contrast, a Rotterdam study found no associations between liver fibrosis assessed through elastography and dementia risk. ²⁹ Lastly, in an Italian longitudinal study on aging, liver fibrosis according to NAFLD fibrosis score (NFS), another serum‐based algorithm, was not associated with increased risk of all‐cause dementia overall; however, a four‐fold increased risk was observed among those who were physically frail. ³⁰ Of note, our estimate of liver fibrosis in the study sample was 1.1%, much lower compared to the estimates of 2.2%, 5.8%, and 15.7% in the UK Biobank, Rotterdam, and Italian studies, respectively. This discrepancy in fibrosis rates could reflect real differences in estimated liver fibrosis prevalence between the distinct populations, or it may be attributed to methodological differences and sample selection, particularly the fact that our study participants represented the total Israeli population.

In the context of prior literature, our research adds to the existing knowledge base by highlighting the importance of liver fibrosis as a possible risk factor for dementia in a “real‐world” setting of middle‐aged and older adults in Israel. Compared to the general Israeli population, the representation of men and Arabs was somewhat lower in our sample (ie, 56.7% vs 50.3% and 11.8% vs 15%, respectively). ³⁵ Because there was no selection for our study, we postulate that this discrepancy may reflect genuine sex and ethnic differences in utilization of healthcare services and routine check‐ups. Moreover, to our knowledge, our study is the largest and has the longest follow‐up duration. This is of particular importance when assessing potential risk factors for dementia, which is characterized by a long prodromal phase. In addition, when addressing the link between liver fibrosis and dementia incidence, reverse causality may occur particularly during the first years of follow‐up and may dilute or even reverse the associations, as was the case in the Rotterdam study exploration. In that prior study, the researchers hypothesized that the reversed association could be explained by the possible weight loss that often accompanies the preclinical phase of dementia ³⁹ and in turn can reduce liver fibrosis. ⁴⁰ In that context, the exclusion of dementia cases occurring in the first 5 years of follow‐up yielded similar findings in our exploration, thereby strengthening our hypothesis of an association between estimated liver fibrosis and dementia risk.

Additional secondary analysis showed that a high risk of advanced liver fibrosis according to FIB‐4 was associated with dementia risk among individuals with a diagnosis of NAFLD, although the findings were not statistically significant, probably due to limited statistical power. Of note, the predictive utility of the FIB‐4 score has been studied mostly among individuals with NAFLD, not in the general population, and showed high validity compared to liver biopsy, which remains the gold standard. ³² Although some studies among the general population showed only modest performance in predicting liver morbidity and mortality, ⁴¹ accumulating investigations highlight the possible role of FIB‐4 as a predictor of cardiovascular diseases ²¹ and other health outcomes ⁴² , ⁴³ in the general population. In addition to the advantage of being non‐invasive, requiring readily available serum measures from routine blood tests, our results, together with previous findings, underscore the possible utility of FIB‐4 as a useful biomarker for dementia risk in a broad population at high risk.

The link between liver fibrosis and dementia incidence is supported by prior studies demonstrating an association of liver fibrosis with cognitive impairment ²⁴ , ²⁶ and brain atrophy. ⁶ , ²⁴ The effects on brain health could stem from vascular damage as prior evidence also demonstrated a link between liver fibrosis and endothelial dysfunction, ⁴⁴ subclinical atherosclerosis, ⁴⁵ cerebral small vessel disease, ⁴⁶ and clinical stroke. ⁴⁷ Yet recent evidence also shows that hepatic dysfunction may impair clearance of peripheral Aβ ⁴⁸ and therefore enhance Alzheimer's disease pathology. ²⁷ Inflammation, which characterizes liver fibrosis, ⁴⁹ is a possible underlying mechanism that explains our finding as it affects both vascular ⁵⁰ and neurodegenerative pathologies. ⁴⁹ , ⁵¹ , ⁵² The gut microbiome may additionally play a role in the observed link between liver fibrosis and incident dementia. Indeed, significant differences in gut composition were observed comparing individuals with and without liver fibrosis, ⁵³ with recent hypotheses considering intestinal permeability and gut dysbiosis as a main trigger for fibrinogenesis. ⁵⁴ , ⁵⁵ Similarly, recent literature points to a correlation between the gut microbiota composition and Alzheimer's disease pathology and suggests that gut dysbiosis may be present even in individuals with preclinical dementia. ⁵⁶

Our findings imply that the association between estimated liver fibrosis and dementia risk may be strongest when fibrosis was assessed in midlife rather than later life. This is in line with consistent evidence from well‐established risk factors for dementia, such as vascular and behavioral factors, which are also known to have the strongest associations with incident dementia when measured at midlife rather than in older ages. ⁵⁷ , ⁵⁸ In addition, our results demonstrate an association between estimated liver fibrosis and dementia incidence, particularly among individuals who were free of diabetes, IHD, and history of stroke, with the most pronounced difference shown between subjects with and without diabetes. Because these are age‐related health conditions, the latter observation may simply echo our age‐stratified results showing a stronger association when liver fibrosis is measured at midlife, before these conditions were diagnosed. Alternatively, these results may stem from the differential performance of the FIB‐4 algorithm in individuals with and without these health conditions. Indeed, prior literature points to a decreased performance of non‐invasive tests for liver fibrosis, ⁵⁹ particularly FIB‐4, ⁶⁰ in subjects with diabetes.

The main strength of this study is the retrospective cohort design with a population‐based nationally representative sample, which reduces the risk for recall and selection biases. Additional strengths are the large sample with a long follow‐up, the use of detailed information from electronic health records, and the multiple sensitivity analyses. Yet several limitations are worth acknowledging. First, because the dementia definition relied on ICD‐9 codes, an over‐ or underestimation of dementia incidence may have occurred. Yet we extended the dementia definition to further include information from hospital‐discharge letters and purchase of medications. Furthermore, because liver fibrosis is largely underdiagnosed, it may be assumed that misclassification of dementia diagnosis would be non‐differential and, hence, can result only in an underestimation of the true association. Second, dementia severity and subtypes could not be ascertained because this information is not recorded in the health records. Lastly, despite the adjustment for multiple covariates, the presence of residual confounding is plausible. For example, data on moderate alcohol consumption are not being routinely recorded in the CHS database. However, alcohol consumption in Israel is overall low. ⁶¹

In conclusion, we demonstrate a robust association between estimated liver fibrosis and dementia risk in a nationally representative sample of middle‐aged and older adults. These findings contribute further support to the importance of liver fibrosis as a newly recognized risk factor for dementia. Moreover, our findings stress the possible utility of a non‐invasive, simply measured tool in dementia risk stratification of high‐risk individuals from the general population and, hence, may guide timely intervention strategies for dementia prevention. Future observational studies are warranted to validate our findings in additional populations using clinically based assessments of dementia and its subtypes and while considering the presence of cardiometabolic comorbidity. Lastly, future research is needed to deepen our understanding of the underlying mechanisms of the liver–brain axis, as well as to identify effective lifestyle and pharmaceutical interventions to reduce dementia risk through the amelioration of liver fibrosis.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting information.

CONSENT STATEMENT

Due to the retrospective nature of this study, the requirement for consent statement has been waived.

Supporting information

Supporting Information

Supporting Information

Weinstein G, Schonmann Y, Yeshua H, Zelber‐Sagi S. The association between liver fibrosis score and incident dementia: A nationwide retrospective cohort study. Alzheimer's Dement. 2024;20:5385–5397. 10.1002/alz.14033