Browse the corpus

Stroke remains a leading cause of death and long-term disability worldwide. In 2021, approximately 93.8 million people were living with stroke, and more than 11 million new cases occurred globally (). Conventional vascular risk factors account for much of the burden of stroke, but not all of it. Metabolic dysfunction, systemic inflammation, and disturbances in coagulation also contribute to stroke occurrence and recovery. Biomarkers reflecting these processes may offer additional insight into why individuals with otherwise similar clinical profiles experience different outcomes. Liver fibrosis has been linked to cardiovascular and cerebrovascular disease (, ). FIB-4, calculated from age, aspartate aminotransferase, alanine aminotransferase, and platelet count, is widely used as a noninvasive marker in the initial assessment of liver fibrosis (, ). Because it relies on routinely collected clinical data, the index can be obtained without additional testing. It may also reflect broader age-related, metabolic, inflammatory, and hematological changes relevant to stroke, although this makes it less specific as a marker of liver fibrosis alone.

Evidence on FIB-4 and stroke has grown, but remains fragmented. Individual studies have examined incident stroke, ischemic stroke, hemorrhagic events, post-treatment sICH, functional outcome, or mortality, often in markedly different populations (, ). FIB-4 has also been handled differently across studies: some investigators analyzed it continuously, whereas others applied thresholds of 1.30, 1.45, 2.67, or 3.25. These cutoffs were developed for liver-disease assessment rather than for stroke and may not represent comparable exposure across age groups or clinical settings. Previous systematic reviews have mainly focused on broad cardiovascular outcomes, with stroke often embedded within composite endpoints (, ). Evidence specific to stroke and its related outcomes therefore remains uncertain. Findings for ischemic stroke are particularly inconsistent (), and comparisons across studies are complicated by differences in design, population, outcome definition, FIB-4 coding, follow-up duration, and covariate adjustment (). ORs and HRs have also sometimes been discussed together, although they represent different measures of association. The same concern applies to categorical and continuous FIB-4 estimates, which address related but distinct questions.

We therefore conducted a systematic review and meta-analysis to examine the associations of FIB-4 with overall stroke, ischemic stroke, hemorrhagic outcomes, poor functional outcome, and all-cause mortality. Categorical and continuous FIB-4 estimates were evaluated separately, as were ORs and HRs. We also examined cutoff-related differences, between-study heterogeneity, and the robustness of the main findings.

Two investigators independently searched PubMed, Embase, and Cochrane Library (inception–November 2025) for eligible cohort and cross-sectional studies. No language restrictions were applied. The search strategy combined Medical Subject Headings (MeSH) terms and free-text keywords, including “Liver Cirrhosis,” “Liver Fibrosis,” “FIB-4,” “Fibrosis-4,” “Stroke,” “cerebrovascular accident,” and “Risk,” and was expanded to cover stroke subtypes (e.g., “ischemic stroke,” “hemorrhagic stroke”) and their related variants. Search strategies were adapted to the specific syntax and indexing systems of each database. The reference lists of included studies and relevant systematic reviews or meta-analyses were manually screened to identify additional eligible reports. Conference abstracts, trial registrations, and other records without an accessible full original report were not included in the quantitative synthesis. The detailed search strategies are provided in .

Two investigators independently extracted data using a predefined standardized form and cross-checked all entries. Extracted variables included the first author, publication year, country or cohort, study design, sample size, participant characteristics, follow-up duration, FIB-4 modeling approach and cutoff values, outcome definition and assessment time, effect measure, point estimate and 95% CI, numbers of participants and events in each FIB-4 category when available, and covariates included in adjusted models. When several multivariable models were reported for the same exposure-outcome association, the most fully adjusted compatible estimate was selected. When a study reported multiple outcomes, each eligible outcome was extracted separately. Only one estimate from each independent cohort was included in a given meta-analytic model. If a compatible adjusted binary estimate was unavailable but group-specific sample sizes and event counts were reported, an unadjusted OR and 95% CI were reconstructed. Estimates that could not be harmonized because of incompatible outcome definitions, exposure contrasts, effect measures, or overlapping cohorts were retained for narrative synthesis and are listed in . All extracted and reconstructed values were independently verified by a second investigator.

Different quality assessment tools were applied according to study design. Cohort studies were evaluated using the Newcastle–Ottawa Scale (NOS), which covers Selection (maximum 4 stars), Comparability (maximum 2 stars), and Outcome (maximum 3 stars). Total NOS scores of 7–9, 5–6, and 0–4 were classified as high, moderate, and low quality, respectively. Cross-sectional studies were evaluated using the 11-item Agency for Healthcare Research and Quality (AHRQ) checklist for cross-sectional/prevalence studies. Each AHRQ item was rated as “Yes,” “No,” or “Unclear”; no numerical total score was calculated. Two investigators performed the assessments independently, with disagreements resolved by discussion or adjudication by a third investigator.

We systematically searched for studies published before November 4, 2025, and identified 3,074 records (PubMed: 262; Embase: 2,757; Cochrane: 55). After removing 228 duplicates, 2,524 records were excluded based on title/abstract screening for ineligible population, exposure, or outcomes. During the initial eligibility assessment of the remaining 322 articles, 189 non-original studies and 13 studies without accessible original data were excluded. After full-text review, 98 studies were further excluded because they did not report an independent association between FIB-4 and the target outcomes. Ultimately, 22 studies were included in the meta-analysis. The study selection process and results are shown in .

This meta-analysis included 20 cohort studies and 2 cross-sectional studies published between 2019 and 2025 (, , ). Individual study sample sizes ranged from 255 to 452,994 participants. The included studies were conducted across 11 countries and regions, including eight studies based on data from China and four studies based on data from the United States. The mean or median age of participants across studies ranged from 53.2 to 79.5 years. Study populations included community-based adults, cardiovascular or metabolic cohorts, and patients with acute or atrial-fibrillation-related stroke. FIB-4 was modeled categorically, continuously, or both; commonly used thresholds were 1.30, 1.45, 2.67, and 3.25, with age-adjusted thresholds in some studies. Outcomes included overall stroke, ischemic stroke, hemorrhagic stroke and its subtypes, symptomatic intracranial hemorrhage (sICH), poor functional outcome, and all-cause mortality. The main characteristics of the included studies are summarized in .

The mean NOS score across the 20 cohort reports was 8.1 of 9; 19 were rated high quality and 1 moderate quality (). The two cross-sectional studies were assessed using the Cross-Sectional/Prevalence Study Quality Assessment tool. Although some methodological details were incompletely reported, both studies applied multivariable adjustment to control for potential confounding factors, and the overall credibility of their results was considered acceptable (, ) ().

Three studies contributed to the exploratory categorical analysis (, , ). Elevated categorical FIB-4 was associated with greater odds of overall stroke (OR = 1.86, 95% CI: 1.63–2.14, p < 0.001; I 2 = 0.0%, P for heterogeneity = 0.580) (). The prospective cohort estimate was OR = 1.93 (95% CI: 1.65–2.24), and the pooled estimate from the two cross-sectional studies was OR = 1.64 (95% CI: 1.21–2.22; I 2 = 0.0%, P for heterogeneity = 0.633); the between-design difference was not significant ( p = 0.353). In two cohort studies, each 1-unit increase in FIB-4 was associated with a higher hazard of overall stroke (HR = 1.08, 95% CI: 1.02–1.14, p = 0.008; I 2 = 0.0%, P for heterogeneity = 0.940) (, ) ().

Two studies contributed categorical ORs (, ). Elevated categorical FIB-4 was associated with greater odds of ischemic stroke (OR = 2.03, 95% CI: 1.72–2.40, p < 0.001; I 2 = 0.0%, P for heterogeneity = 0.786) (). Three studies contributed categorical HRs (, , ). Elevated categorical FIB-4 was associated with a higher hazard of ischemic stroke (HR = 1.55, 95% CI: 1.01–2.35, p = 0.043; I 2 = 68.3%, P for heterogeneity = 0.043; 95% prediction interval, 0.02–154.21) (). One additional study reported no association between continuous FIB-4 and ischemic stroke (HR = 1.00, 95% CI: 0.96–1.05) () ().

Two overlapping UK Biobank analyses reported positive associations between FIB-4 > 2.67 and incident hemorrhagic outcomes (, ). The reported HRs were 2.00 (95% CI: 1.60–2.60) for hemorrhagic stroke, 2.00 (95% CI: 1.50–2.70) and 2.14 (95% CI: 1.63–2.81) for intracerebral hemorrhage, and 2.20 (95% CI: 1.50–3.50) and 1.90 (95% CI: 1.27–2.84) for subarachnoid hemorrhage; these estimates were not pooled (). Four studies contributed categorical ORs for sICH (, ). Elevated categorical FIB-4 was associated with greater odds of sICH (OR = 2.52, 95% CI: 1.79–3.53, p < 0.001; I 2 = 0.0%, P for heterogeneity = 0.486) (). Two studies contributed continuous ORs (, ). Each 1-unit increase in FIB-4 was associated with greater odds of sICH (OR = 1.33, 95% CI: 1.20–1.48, p < 0.001; I 2 = 0.0%, P for heterogeneity = 0.707) ().

Seven studies evaluated categorical FIB-4 in relation to poor functional outcome, generally defined as a modified Rankin Scale score of at least 3 (, , ). Elevated categorical FIB-4 was associated with greater odds of poor functional outcome (OR = 2.88, 95% CI: 2.47–3.35, p < 0.001; I 2 = 0.0%, P for heterogeneity = 0.763) (). Five studies evaluated continuous FIB-4 (, , ). Each 1-unit increase in FIB-4 was associated with greater odds of poor functional outcome (OR = 1.26, 95% CI: 1.11–1.43, p < 0.001; I 2 = 60.1%, P for heterogeneity = 0.040; 95% prediction interval, 0.86–1.86) ().

Seven independent cohort estimates contributed to the categorical OR analysis (, , , , , ). Elevated categorical FIB-4 was associated with greater odds of all-cause mortality (OR = 2.98, 95% CI: 2.51–3.53, p < 0.001; I 2 = 51.5%, P for heterogeneity = 0.054; 95% prediction interval, 1.97–4.49) (). Two studies contributed categorical HRs (, ). Elevated categorical FIB-4 was associated with a higher hazard of all-cause mortality (HR = 1.93, 95% CI: 1.07–3.48, p = 0.029; I 2 = 97.3%, P for heterogeneity < 0.001). Four studies evaluated continuous FIB-4 (, , , ). Each 1-unit increase in FIB-4 was associated with a higher hazard of all-cause mortality (HR = 1.08, 95% CI: 1.03–1.12, p = 0.001; I 2 = 69.6%, P for heterogeneity = 0.020; 95% prediction interval, 0.92–1.26) (). One additional acute coronary syndrome cohort reported associations of continuous FIB-4 (adjusted OR = 1.24, 95% CI: 1.04–1.47) and FIB-4 > 2.67 versus <1.30 (adjusted OR = 2.35, 95% CI: 1.05–5.28) with in-hospital all-cause mortality (); these estimates were not pooled ().

Cutoff-specific analyses showed positive associations at both lower and upper thresholds, although statistical significance differed for overall stroke. For overall stroke, elevated FIB-4 was significantly associated with greater odds of stroke at the upper cutoff (OR = 1.92, 95% CI: 1.66–2.23; I 2 = 0.0%), whereas the lower-cutoff analysis was not statistically significant (OR = 1.27, 95% CI: 0.92–1.74; I 2 = 71.0%; p = 0.144). For sICH, the pooled ORs were 2.11 (95% CI: 1.40–3.19; I 2 = 0.0%) and 2.52 (95% CI: 1.79–3.53; I 2 = 0.0%) at the lower and upper cutoffs, respectively. The corresponding ORs were 2.48 (95% CI: 2.09–2.95; I 2 = 0.0%) and 2.88 (95% CI: 2.47–3.35; I 2 = 0.0%) for poor functional outcome, and 1.86 (95% CI: 1.39–2.47; I 2 = 94.7%) and 2.98 (95% CI: 2.51–3.53; I 2 = 51.5%) for all-cause mortality (). In leave-one-out analyses, the associations of categorical FIB-4 with sICH, poor functional outcome, and all-cause mortality remained statistically significant after sequential exclusion of individual estimates. The association between continuous FIB-4 and poor functional outcome was also unchanged ().

Elevated categorical FIB-4 was associated with greater odds of all-cause mortality in all prespecified subgroups. The pooled associations were similar between retrospective and prospective cohorts (OR = 3.24, 95% CI: 2.37–4.42 vs. OR = 2.86, 95% CI: 2.34–3.51; P for subgroup difference = 0.185), between middle-aged and older populations (OR = 3.00, 95% CI: 2.44–3.68 vs. OR = 2.93, 95% CI: 1.99–4.33; P for subgroup difference = 0.562), and between acute-stroke/short-term and cardiovascular-metabolic/long-term studies (OR = 2.97, 95% CI: 2.17–4.06 vs. OR = 3.04, 95% CI: 2.44–3.79; P for subgroup difference = 0.659). The pooled OR was higher in Asian than in Western studies (OR = 3.80, 95% CI: 2.78–5.19 vs. OR = 2.79, 95% CI: 2.36–3.29; P for subgroup difference = 0.042) ().

Univariable meta-regression did not identify significant moderation of the association between categorical FIB-4 and all-cause mortality by study design (ratio of ORs = 0.87, 95% CI: 0.62–1.23; p = 0.443), geographic region (ratio of ORs = 0.74, 95% CI: 0.51–1.06; p = 0.103), age category (ratio of ORs = 0.99, 95% CI: 0.65–1.51; p = 0.960), or clinical setting/follow-up duration (ratio of ORs = 1.06, 95% CI: 0.70–1.61; p = 0.789) ().

This meta-analysis of 22 observational studies found that higher FIB-4 was associated with overall stroke, ischemic stroke, sICH, poor functional outcome, and all-cause mortality. The clearest and most consistent associations were seen for sICH, poor functional outcome, and mortality in the categorical analyses. Evidence for stroke occurrence was less straightforward. The overall-stroke OR combined one prospective estimate of incident stroke with two cross-sectional estimates of prevalent stroke, while the ischemic-stroke HR analysis showed substantial heterogeneity and an exceptionally wide prediction interval. FIB-4 may therefore capture clinically relevant risk across several stroke-related settings, although the magnitude of the association is unlikely to be uniform and the available data do not support a causal interpretation.

Evidence on stroke occurrence came mainly from prospective cohorts in which higher FIB-4 was associated with incident stroke (, ). The two cross-sectional studies added information on prevalent stroke, but could not establish whether elevated FIB-4 preceded the event. The pooled categorical estimate for overall stroke is therefore best viewed as exploratory. In the continuous analysis, stroke hazard increased modestly with each 1-unit increase in FIB-4. A similar association has been reported in patients with acute coronary syndrome (), although that population differs from the stroke cohorts included here. Associations with post-stroke outcomes were more consistent in the categorical OR analyses. Higher FIB-4 was associated with sICH, poor functional outcome, and all-cause mortality, in keeping with findings from acute ischemic stroke and large-vessel-occlusion cohorts (). These estimates should not be compared directly with those for stroke occurrence because the populations, treatments, outcome definitions, and follow-up periods differed. Continuous models produced smaller relative effects, and the prediction intervals for poor functional outcome and mortality included the null. FIB-4 may carry prognostic information across its range, but the effect expected in a new clinical setting remains uncertain.

FIB-4 is better understood as a composite clinical marker than as a direct measure of hepatic fibrosis. Age, aspartate aminotransferase, alanine aminotransferase, and platelet count each reflect more than liver status; together, they may also capture ageing, acute illness, metabolic burden, and hematological change. Two patients with the same score may therefore have quite different underlying conditions. This is particularly relevant in stroke research because age is built into the index and is itself a major determinant of stroke risk and prognosis. Conventional FIB-4 thresholds become less specific with advancing age (), and statistical adjustment cannot completely remove this structural overlap. Timing also matters. In population-based cohorts, FIB-4 probably reflects pre-existing hepatic and cardiometabolic risk. During acute stroke, however, aminotransferase levels and platelet counts may change in response to physiological stress, infection, treatment, nutritional status, or stroke severity. A value obtained on admission may therefore reflect both chronic vulnerability and the acute event itself. This makes reverse causation a genuine concern in prognostic studies. Future studies should distinguish pre-stroke, admission, and serial FIB-4 measurements rather than treating them as equivalent.

The observed associations are biologically plausible, although no single pathway is likely to account for them. Metabolic liver disease clusters with obesity, diabetes, hypertension, and dyslipidemia, all of which increase vascular risk, and hepatic fibrosis has been linked to subsequent cardiovascular events (). Liver disease can also alter both procoagulant and anticoagulant pathways, offering a possible connection to ischemic as well as hemorrhagic complications (). A further link may be atrial fibrillation, which has been associated with higher FIB-4 in patients with acute ischemic stroke (). Even so, the present data cannot separate an effect of liver fibrosis itself from the broader cardiometabolic and systemic burden reflected by FIB-4.

Differences in how FIB-4 was defined are an important source of uncertainty. The commonly used thresholds were not developed in stroke populations. In patients with HIV/HCV coinfection, a value below 1.45 had a negative predictive value of 90% for excluding advanced fibrosis, whereas a value above 3.25 had a specificity of 97% for identifying advanced fibrosis (). In nonalcoholic fatty liver disease, thresholds of 1.30 and 2.67 yielded a negative predictive value of 90% and a positive predictive value of 80%, respectively (). Some included cohorts also used age-adjusted definitions. These thresholds are not interchangeable, and the same numerical value may not represent the same biological state across populations. Results from the cutoff analyses mostly pointed in the same direction, although the lower-cutoff estimate for overall stroke was not statistically significant. The larger effects observed at upper thresholds require caution. Raising the cutoff changes both the exposed and reference groups, so the difference cannot be taken as evidence of a dose–response relationship or more severe histological fibrosis. Continuous modeling avoids some of this arbitrariness, but a 1-unit increase in FIB-4 may not have exactly the same meaning across age groups, populations, and clinical settings.

Heterogeneity was uneven across the analyses. Several categorical models had low I 2 values, but some were based on only two to four estimates, so apparent homogeneity may simply reflect limited power. The categorical HR analysis for ischemic stroke and several continuous or mortality analyses were clearly more variable. Prediction intervals remained above 1 only for categorical FIB-4 and all-cause mortality; they crossed the null for ischemic-stroke HRs, continuous poor functional outcome, and continuous mortality. Neither subgroup analyses nor meta-regression identified a consistent explanation for this variation. The nominal regional difference seen in the subgroup analysis was not reproduced in meta-regression and may have arisen by chance. Leave-one-out analyses showed that the principal estimates were not driven by a single study, but they do not address residual confounding or other systematic bias.

Several features strengthen this analysis. Categorical and continuous FIB-4 were examined separately, ORs and HRs were not combined, prediction intervals were reported where appropriate, and overlapping cohorts were handled to avoid double counting. The limitations are more important to the interpretation. All included studies were observational, two were cross-sectional, and some ORs were reconstructed from event counts without adjustment. Populations, treatments, outcome definitions, follow-up periods, thresholds, and the timing of FIB-4 measurement also varied. Overlap among UK Biobank and NHANES reports further reduced the effective independence of the evidence and made a unique participant total unreliable.

Many syntheses included only a few estimates. Between-study variance and prediction intervals are therefore imprecise, even when the pooled effect is statistically significant. No model contained 10 or more independent estimates, so funnel plots and Egger tests would not have been informative. High NOS scores are reassuring about study design and reporting, but they do not remove selection bias, exposure misclassification, reverse causation, or residual confounding. The included studies also evaluated association rather than predictive performance. Whether FIB-4 adds useful information beyond age, stroke severity, atrial fibrillation, and established vascular risk factors remains unknown. FIB-4 is inexpensive and readily available, which makes it an attractive candidate marker. The present results do not support using it alone to guide treatment or prognosis. Prospective studies should prespecify age-appropriate thresholds, standardize the timing of laboratory measurement, and report categorical and continuous estimates with comparable adjustment. Serial measurements may prove more informative than a single value. Individual-participant-data analyses could address non-linearity and interactions with age or treatment, while prediction studies should test whether adding FIB-4 improves discrimination, calibration, and clinical decision-making beyond existing models.