Abstract

Background and aim:

Liver enzyme elevations between one and two times the upper limit of normal (ULN) are common in outpatient practice and often dismissed as clinically insignificant, yet whether this range harbours meaningful subphenotypes, and whether the hepatocellular–cholestatic distinction applies at borderline levels, has not been systematically examined. We investigated whether, in adult outpatients with persistent borderline elevation confirmed using sex-specific ULN, the hepatocellular and cholestatic patterns carry distinct metabolic and inflammatory signatures.

Materials and methods:

This single-centre observational outpatient study was conducted at Elazığ Fethi Sekin City Hospital between October 2024 and January 2026. Eight hundred adults with standardised fasting biochemistry and haematology panels at baseline and third-month follow-up were analysed. Liver enzymes were classified as normal, borderline, or overtly elevated by the maximum fold-increase above the sex-specific ULN; borderline cases were subcategorised as hepatocellular, cholestatic, or overlap. Clinical, biochemical, inflammatory, and iron parameters were compared, and independent associations were examined using multivariable logistic regression.

Results:

Enzyme levels were normal in 548 patients (68.5%), borderline in 211 (26.4%), and overtly elevated in 41 (5.1%). Among the 211 borderline patients, 105 (49.8%) exhibited a hepatocellular pattern, 77 (36.5%) a cholestatic pattern, and 29 (13.7%) an overlap pattern. The hepatocellular pattern was associated with higher ferritin, serum iron, and transferrin saturation (TSAT), and a lower AST/ALT ratio; in multivariable analysis, ferritin was the only independent correlate (OR = 1.523; 95% CI: 1.199–1.935; p = 0.001). The cholestatic pattern showed higher HbA1c, higher systemic immune-inflammation index (SII), and lower TSAT; the independent predictors were HbA1c (OR = 1.591; 95% CI: 1.168–2.168; p = 0.003), SII (OR = 2.205; 95% CI: 1.228–3.960; p = 0.008), and TSAT (OR = 0.679; 95% CI: 0.489–0.943; p = 0.021), preserved in the non-diabetic subgroup. The overlap pattern exhibited features of both pure patterns.

Conclusion:

Persistent borderline liver enzyme elevation can be stratified into biochemically distinct subphenotypes when sex-specific reference ranges are applied. The cholestatic pattern is predominantly associated with glycaemic burden, systemic inflammation, and reduced iron transport, whereas the hepatocellular pattern shows an iron-predominant signature. A pattern-based approach may carry clinical value in outpatient management; findings from full models with limited events-per-variable, particularly for SII, are hypothesis-generating and require longitudinal validation.

1 Introduction

Elevated liver enzymes are among the laboratory findings most frequently encountered in outpatient practice (1). While marked elevations typically prompt a systematic evaluation, values between one and two times the upper limit of normal (ULN) are commonly dismissed as clinically insignificant and attributed to transient fluctuations, measurement error, or incidental findings of uncertain relevance (2). By contrast, the 2017 clinical guideline of the American College of Gastroenterology (ACG) defines this range as a “borderline” (<2 × ULN) category, highlights that it may be clinically relevant, and recommends reassessment within a few months (3).

The biological significance of this range has not been systematically addressed. The distinction between hepatocellular and cholestatic patterns has to date been examined primarily in the context of overt liver injury — particularly in drug-induced liver injury — and the two phenotypes have been shown to carry distinct clinical courses and outcomes (4). By contrast, large-scale clinical studies linking liver enzyme levels or their variability over time to cardiovascular events and mortality have treated enzyme elevation as a single category and have not incorporated the hepatocellular–cholestatic distinction into clinical risk stratification (5). The same approach persists in the setting of metabolic disease: enzyme elevation in MASLD patients has been shown to be strongly associated with long-term cardiometabolic mortality, yet this association has been evaluated independently of the underlying biochemical profile. Consequently, although the clinical relevance of this distinction is well defined in overt disease, it has not been specifically examined in the borderline range (6).

The biochemical basis for this distinction is important. The hepatocellular pattern is defined by AST and ALT; whereas ALT is relatively liver-specific, AST is also present in skeletal muscle, the heart, and erythrocytes, so its elevation is not always of hepatic origin (7). The cholestatic pattern is defined by ALP and GGT; ALP can also rise from extrahepatic sources such as bone, placenta, and intestine, while GGT, although more sensitive to the liver, is readily influenced by alcohol and enzyme-inducing medications (8). Despite these limitations, AST, ALT, ALP, and GGT remain the core markers used in the clinical evaluation of hepatocellular and cholestatic phenotypes, owing to their standardised reference ranges and inclusion in routine panels. Whether these two phenotypes carry distinct metabolic and inflammatory signatures at borderline levels is unknown.

Liver enzyme levels differ markedly by sex. The ACG 2017 clinical guideline defines the “healthy” ALT level as 29–33 IU/L in men and 19–25 IU/L in women, thereby formally endorsing the clinical use of sex-specific thresholds (3). This difference rests on a multifactorial biological foundation: the liver is one of the organs exhibiting the most pronounced sexual dimorphism, with sex hormones and growth hormone shaping liver-specific gene expression and metabolic activity, giving rise to distinct baseline enzyme profiles (9). This biological basis is also consistent with the sex-specific differences in prevalence and progression observed in metabolic liver diseases, most notably MASLD (formerly NAFLD) (10). Consequently, the use of a single threshold — particularly in women — carries the risk of systematically misclassifying enzyme elevations.

In light of these considerations, the present study aimed to examine, in adult outpatients with persistent borderline enzyme elevation confirmed by standardised repeat measurements and classified using sex-specific upper limits of normal, the metabolic and inflammatory signatures of the hepatocellular and cholestatic patterns in the outpatient setting.

2 Materials and methods2.1 Study design and setting

This is a single-centre observational outpatient study based on standardised repeat measurements, conducted at the Department of Internal Medicine, Elazığ Fethi Sekin City Hospital, between October 2024 and January 2026. The study was approved by the Clinical Research Ethics Committee and carried out in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants. The study has been reported in accordance with the STROBE guideline for the reporting of observational studies.

2.1.1 Inclusion criteria

Patients meeting all of the following criteria were included in the study:

Age ≥18 years

Adult outpatients presenting to the internal medicine clinic

Availability of a complete, standardised fasting biochemistry and haematology panel at both the baseline and the three-month follow-up visit

Provision of written informed consent

2.1.2 Exclusion criteria

Patients with any of the following were excluded from the study:

Missing key laboratory data

Initiation or modification of any hepatotoxic medication within the 4 weeks preceding enrolment (including paracetamol, NSAIDs, antibiotics, antifungals, or statins)

Oral contraceptive use (newly initiated or ongoing; to minimise hormonal confounding of enzyme reference ranges)

Pre-existing chronic liver disease — viral hepatitis B or C, autoimmune hepatitis, primary biliary cholangitis, primary sclerosing cholangitis, alcohol-related liver disease, or cirrhosis of any aetiology — verified through ICD-10 diagnostic codes and the national patient tracking system

Presence of symptoms suggestive of hepatobiliary disease at enrolment — including jaundice, right upper quadrant pain, pruritus, or unexplained fatigue — assessed by structured clinical history

The analysed cohort therefore comprised adult outpatients without clinical manifestations of hepatobiliary disease.

An a priori sample-size calculation was performed using G*Power 3.1.9.7 software (F tests—linear multiple regression: fixed model, R2 deviation from zero). Assuming a small-to-medium effect size (Cohen’s f2 = 0.10), a two-sided α of 0.05, and 90% power, the required minimum total sample sizes were 215 for the primary 10-predictor multivariable model, 199 for the pre-specified 8-predictor reduced pattern-specific model, and 267 for the pre-specified 18-predictor full pattern-specific model. These calculations were used to define the design requirements for the overall analytical framework, including the planned pattern-specific modelling strategy. Over the course of the study, 1,125 consecutive outpatients were screened. After applying the pre-specified inclusion and exclusion criteria, 325 patients were excluded, and the remaining 800 patients were included in the final analyses. A detailed breakdown of the screening process and reasons for exclusion is provided in Supplementary Figure S1 (STROBE-style flow diagram).

2.2 Classification of liver enzyme elevation

Sex-specific upper limits of normal (ULN) were defined for all four enzymes according to the standard reference ranges of our hospital laboratory: ALT (women >25 U/L, men >35 U/L), AST (women >25 U/L, men >35 U/L), GGT (women >35 U/L, men >55 U/L), and ALP (women >100 U/L, men >120 U/L). These thresholds are consistent with the ACG 2017 clinical guideline’s recommendation for the use of sex-specific ULN (3). Each patient was assigned to one of three groups based on the maximum fold-increase above the applicable sex-specific ULN across the four enzymes: normal (≤1 × ULN), borderline elevation (>1–<2 × ULN), and overt elevation (≥2 × ULN). Borderline cases were further divided into three groups according to pattern: hepatocellular (AST and/or ALT elevated, ALP and GGT normal), cholestatic (ALP and/or GGT elevated, AST and ALT normal), and overlap (both patterns present simultaneously). As an additional descriptor, the AST/ALT ratio was calculated.

2.3 Laboratory parameters and inflammatory indices

The routine baseline panel included the following parameters: complete blood count, liver enzymes (AST, ALT, ALP, GGT), creatine kinase, lactate dehydrogenase, amylase, albumin, renal function tests (urea, creatinine), glucose, HbA1c, lipid profile (total cholesterol, HDL, LDL, triglycerides), iron parameters (serum iron, ferritin, TSAT, UIBC), thyroid function tests (TSH, free T3, free T4), vitamin B12, 25-OH vitamin D, and CRP; all participants underwent abdominal ultrasonography. MASLD was defined according to the current criteria (11) as the coexistence of hepatic steatosis on imaging and at least one cardiometabolic risk factor. Within this framework, body mass index (≥25 kg/m2) was used in the anthropometric assessment, and a diagnosis of hypertension or use of antihypertensive medication was used in the blood pressure assessment. Biochemical analyses were performed using a Beckman Coulter AU5800 automated chemistry analyser, and complete blood count analyses were performed using a Beckman Coulter DxH 800 analyser. Intra-assay coefficients of variation were <2.5% for AST, ALT, and GGT, and <2.0% for ALP.

The following composite inflammatory indices were derived from routine haematological parameters: neutrophil-to-lymphocyte ratio (NLR), neutrophil-to-monocyte ratio (NMR), systemic immune-inflammation index (SII = platelet × neutrophil / lymphocyte), and aggregate index of systemic inflammation (AISI = neutrophil × monocyte × eosinophil / lymphocyte). FIB-4 and APRI were calculated as non-invasive fibrosis scores. eGFR was estimated using the CKD-EPI 2021 creatinine equation (12). Hepatic steatosis was graded by abdominal ultrasonography (absent, Grade 1, 2, or 3). Smoking was quantified in pack-years. Alcohol consumption was assessed by self-report and classified as harmful or non-harmful according to the EASL–EASD–EASO 2024 MASLD guideline (13). Comorbidity diagnoses and medication-use patterns were verified through patient self-report, ICD-10 diagnostic codes, the Central Physician Information System, and e-Prescription records. BMI was recorded as the mean of height/weight measurements obtained at the two visits.

2.4 Statistical analysis

Continuous variables are expressed as mean ± SD or median (IQR) according to distribution, which was assessed by the Shapiro–Wilk test. Two-group comparisons were performed using the Mann–Whitney U test or the independent-samples t-test, and multi-group comparisons were performed using the Kruskal–Wallis test. Categorical variables were compared using Pearson’s chi-square test or Fisher’s exact test. Associations between continuous variables were assessed by Spearman rank correlation. Binary logistic regression was applied to compare pattern subgroups with the normal group and with each other. A full multivariable model and a pre-specified reduced model were fitted in parallel. Continuous variables were standardised, and odds ratios (OR) are reported per 1-SD increment. Multicollinearity was evaluated using the variance inflation factor (VIF). Sex-by-covariate interaction terms were pre-specified and included in the models. As a pre-specified sensitivity analysis addressing potential collinearity between HbA1c and age, an age-residualisation procedure was additionally performed for the cholestatic model: HbA1c was first regressed on age using ordinary least squares (HbA1c = β₀ + β₁ × Age + ε), and the resulting residuals, representing the component of HbA1c variance orthogonal to age, were entered in place of raw HbA1c into the multivariable logistic regression model. Per the Frisch–Waugh–Lovell theorem, the partial coefficient of a covariate in a multivariable linear regression is mathematically equivalent to the coefficient obtained by first residualising that covariate against all other predictors; this analysis therefore provides an explicit test of whether the HbA1c–pattern association is attributable to shared variance with age. As a sensitivity analysis, the models were repeated in the subcohort without hepatic steatosis. Internal validity was assessed using non-parametric bootstrap with 1,000 iterations. For the cholestatic and hepatocellular subgroups, pattern-specific post-hoc statistical power for the independently significant associations was additionally estimated using Hsieh’s (1998) approximation for logistic regression, assuming standardised predictors and the observed event proportions. Given the limited events-per-variable in the overlap subgroup (n = 29; EPV < 2 with 18 covariates), an adjusted multivariable model was not fitted for this subgroup, which is accordingly reported descriptively. All tests were two-sided, and p < 0.05 was considered statistically significant. Analyses were performed using IBM SPSS Statistics version 26.0 (IBM Corp., Armonk, NY, United States).

3 Results3.1 Cohort composition and pattern distribution

The cohort had a mean age of 50.4 ± 16.3 years, and 57.4% were female. Enzyme levels were normal in 548 patients (68.5%), borderline elevated (>1–<2 × ULN) in 211 (26.4%), and overtly elevated (≥2 × ULN) in 41 (5.1%) (baseline characteristics are presented in Supplementary Table S1). Within the borderline elevation group, 105 patients (49.8%) exhibited a hepatocellular pattern, 77 (36.5%) a cholestatic pattern, and 29 (13.7%) an overlap pattern. The baseline characteristics of the normal and borderline elevation groups are presented in Table 1.

VariableNormal (n = 548)Borderline (n = 211)pAge (years)50.0 (38–63)52.0 (41–63)0.421Female sex, n (%)310 (56.6%)130 (61.6%)0.239BMI (kg/m2)28.4 (25.5–31.0)29.0 (25.4–31.6)0.183HbA1c (%)5.7 (5.3–6.3)5.9 (5.5–7.2)<0.001TG/HDL ratio2.6 (1.7–4.5)3.1 (1.8–4.7)0.091Triglycerides (mg/dL)131 (91–204)163 (99–227)0.004Total cholesterol (mg/dL)187 (159–214)202 (170–230)<0.001Fasting glucose (mg/dL)96 (86–115)101 (89–149)0.002Ferritin (μg/L)34.0 (15–64)45.0 (20–90)0.002TSAT (%)21.8 (15.3–29.8)20.7 (13.5–29.1)0.350AISI0.13 (0.07–0.26)0.13 (0.07–0.26)0.875NLR1.85 (1.48–2.38)1.82 (1.48–2.42)0.698NMR7.70 (6.24–9.28)7.83 (6.31–9.51)0.490CRP (mg/L)3.2 (1.8–5.4)2.6 (1.6–5.8)0.589Albumin (g/L)43.0 (41–45)43.0 (41–46)0.091Haemoglobin (g/dL)14.4 (13.4–15.5)14.6 (13.5–15.8)0.160HCT (%)42.1 (39.6–45.1)42.8 (39.8–46.0)0.095eGFR (mL/min/1.73m2)100.0 (89–113)102.0 (90–115)0.616Hepatic steatosis, n (%)321 (58.6%)122 (57.8%)0.915Smoking, n (%)188 (34.3%)70 (33.2%)0.834Alcohol use, n (%)63 (11.5%)21 (10.0%)0.632Diabetes mellitus, n (%)132 (24.1%)72 (34.1%)0.007

Baseline characteristics: normal versus borderline elevation groups.

Mann–Whitney U test for continuous variables; Pearson chi-square or Fisher exact test for categorical variables. Continuous variables: median (IQR); categorical variables: n (%). The overt elevation group (n = 41, ≥2 × ULN) is excluded from this table and its baseline characteristics are provided in Supplementary Table S1. BMI, Body mass index; HbA1c, Glycated haemoglobin; TG/HDL, Triglyceride-to-HDL cholesterol ratio; HCT, Haematocrit; TSAT, Transferrin saturation; AISI, Aggregate index of systemic inflammation; NLR, Neutrophil-to-lymphocyte ratio; NMR, Neutrophil-to-monocyte ratio; CRP, C-reactive protein; eGFR, Estimated glomerular filtration rate.

3.2 Raw comparisons: pattern biochemical profiles

The proportion of participants meeting MASLD criteria was similar across the normal, hepatocellular, and cholestatic groups (55.8, 49.5, and 58.4%, respectively; 72.4% in the overlap group); no significant between-group difference was observed (chi-square p = 0.16).

Compared with the normal group, the hepatocellular pattern was characterised by higher levels of ferritin, serum iron, TSAT, creatine kinase, and LDH, together with a characteristically low AST/ALT ratio (all p < 0.05). Inflammatory indices (NMR, SII, NLR) were lower in the hepatocellular group than in normal controls (all p < 0.01).

In the cholestatic group, HbA1c, inflammatory indices (NMR, SII, NLR), white blood cell and neutrophil counts, total cholesterol, and fasting glucose were markedly higher, whereas serum iron and TSAT were lower (all p ≤ 0.018). The cholestatic pattern showed a female predominance (70.1% vs. 56.6% in the normal group, p = 0.033); no significant sex difference was observed in the hepatocellular pattern.

In the direct comparison between the two borderline patterns, the cholestatic pattern was characterised by higher HbA1c and inflammatory indices, whereas the hepatocellular pattern showed higher iron markers and creatine kinase together with a lower AST/ALT ratio (all p < 0.01). Patients in the cholestatic group were significantly older than those in the hepatocellular group (median 56 vs. 46 years, p < 0.001). All univariable comparisons are presented in Table 2.

VariableNormal (n = 548)Hepatocellular (n = 105)p†Cholestatic (n = 77)p†p‡Ferritin (μg/L)34.0 (15–64)48.0 (21–100)0.00536.0 (17–80)0.3360.225Serum iron (μg/dL)76.0 (54–105)85.0 (58–118)0.02567.0 (42–83)0.002<0.001TSAT (%)21.8 (15.3–29.8)24.4 (16.3–33.9)0.03718.4 (10.5–23.3)0.002<0.001CK (U/L)82 (59–111)106 (70–159)<0.00168 (52–87)0.009<0.001LDH (U/L)176 (155–202)191 (171–216)<0.001184 (159–211)0.1110.133AST/ALT ratio1.17 (1.00–1.46)0.77 (0.64–1.00)<0.0011.13 (0.91–1.31)0.094<0.001Free T3 (pmol/L)3.39 (3.1–3.7)3.50 (3.2–3.8)0.0793.37 (3.1–3.7)0.8340.165HbA1c (%)5.70 (5.3–6.3)5.70 (5.3–6.3)0.6726.10 (5.7–8.1)<0.001<0.001Neutrophil (×109/L)4.05 (3.3–4.9)3.80 (3.1–4.7)0.0364.75 (4.1–6.1)<0.001<0.001WBC (×109/L)6.98 (5.9–8.2)6.99 (5.5–8.2)0.3987.61 (6.5–9.3)<0.001<0.001NMR7.70 (6.2–9.3)7.24 (5.4–8.5)<0.0018.92 (7.5–10.0)<0.001<0.001SII477 (350–642)420 (312–539)0.008592 (459–781)<0.001<0.001AISI0.13 (0.07–0.26)0.12 (0.07–0.22)0.6080.14 (0.06–0.30)0.6590.459NLR1.85 (1.5–2.4)1.64 (1.3–2.1)0.0042.06 (1.6–2.8)0.018<0.001CRP (mg/L)3.2 (1.8–5.4)2.6 (1.6–5.1)0.1333.1 (1.7–6.6)0.7320.176Total chol. (mg/dL)186 (159–214)200 (166–230)0.014208 (180–230)0.0020.476Fasting glucose (mg/dL)96 (86–115)96 (88–120)0.719106 (91–167)<0.0010.009Albumin (g/L)43 (41–45)44 (42–46)0.00642 (40–45)0.3610.008HCT (%)42.1 (39.6–45.1)43.1 (40.4–46.5)0.03541.9 (38.9–45.4)0.7900.112Female sex, n (%)310 (56.6%)56 (53.3%)0.61454 (70.1%)0.0330.033Age (years)50.0 (38–63)46 (38–58)0.11356 (47–66)0.007<0.001

Raw comparison of biochemical profiles: hepatocellular and cholestatic patterns versus normal group and versus each other.

†vs normal group; ‡hepatocellular vs cholestatic direct comparison. Mann–Whitney U for continuous variables; Pearson chi-square for categorical variables. Continuous variables: median (IQR). CK, Creatine kinase; LDH, Lactate dehydrogenase; HbA1c, Glycated haemoglobin; WBC, White blood cell count; TSAT, Transferrin saturation; NMR, Neutrophil-to-monocyte ratio; SII, Systemic immune-inflammation index; AISI, Aggregate index of systemic inflammation; NLR, Neutrophil-to-lymphocyte ratio; CRP, C-reactive protein; HCT, haematocrit.

3.3 Multivariable analysis: pattern predictors

In the multivariable logistic regression analysis adjusted for 18 variables, ferritin was identified as the only variable independently associated with the hepatocellular pattern (OR = 1.523 for standardised ferritin; 95% CI: 1.199–1.935; p = 0.001). In this model, no metabolic or inflammatory index reached independent significance (n = 653, EPV = 5.8, all VIF < 3.5).

For the cholestatic pattern, HbA1c (OR = 1.591; 95% CI: 1.168–2.168; p = 0.003), SII (OR = 2.205; 95% CI: 1.228–3.960; p = 0.008), and TSAT (OR = 0.679; 95% CI: 0.489–0.943; p = 0.021) were identified as independent predictors (n = 625, EPV = 4.3). Owing to the limited EPV, a pre-specified eight-variable reduced model was applied (EPV = 9.6), in which HbA1c (OR = 1.456, p < 0.001), TSAT (OR = 0.614, p = 0.003), female sex (OR = 1.976, p = 0.026), and ferritin (OR = 1.345, p = 0.022) were confirmed. Directional consistency was preserved across both model specifications (Table 3). A forest-plot visualisation of the multivariable ORs is presented in Figure 1. A graphical summary integrating the pattern-specific signatures with their clinical interpretations is provided in Supplementary Figure S2.

VariableHepatocellular OR (95% CI)pCholestatic OR (95%CI)pVIFHbA1c0.970 (0.711–1.323)0.8471.591 (1.168–2.168)0.003**2.2TG/HDL ratio1.180 (0.966–1.441)0.1050.989 (0.752–1.300)0.9361.1NLR0.843 (0.505–1.408)0.5140.789 (0.477–1.305)0.3563.5NMR1.070 (0.880–1.302)0.4961.069 (0.867–1.317)0.5341.0SII0.844 (0.517–1.378)0.4972.205 (1.228–3.960)0.008**2.4AISI1.017 (0.751–1.378)0.9110.962 (0.725–1.277)0.7911.3CRP0.620 (0.252–1.529)0.2990.771 (0.518–1.148)0.2011.5Ferritin1.523 (1.199–1.935)0.001**1.306 (0.995–1.713)0.0541.4TSAT1.086 (0.853–1.384)0.5020.679 (0.489–0.943)0.021*1.3Age1.125 (0.761–1.664)0.5541.381 (0.881–2.163)0.1592.9Female sex1.377 (0.790–2.403)0.2591.812 (0.935–3.511)0.0781.4eGFR1.445 (0.980–2.128)0.0631.116 (0.749–1.661)0.5902.5Albumin1.228 (0.959–1.572)0.1041.225 (0.926–1.620)0.1541.3BMI0.973 (0.750–1.262)0.8351.170 (0.893–1.533)0.2531.4Alcohol0.883 (0.431–1.811)0.7351.103 (0.458–2.653)0.8271.1Steatosis grade0.884 (0.700–1.116)0.3000.965 (0.744–1.251)0.7861.1DM diagnosis1.237 (0.571–2.678)0.5900.688 (0.301–1.573)0.3762.3Hepatotoxic drug0.931 (0.427–2.027)0.8571.159 (0.562–2.390)0.6901.2

Multivariable logistic regression: hepatocellular and cholestatic patterns versus normal group.

Hepatocellular model: n = 653, events = 105, EPV = 5.8. Cholestatic model: n = 625, events = 77, EPV = 4.3. Reference category: normal group. All models adjusted for all listed variables. Continuous variables standardised; ORs expressed per 1 SD increment. Reduced cholestatic model (EPV = 9.6, 8 pre-specified variables): HbA1c OR = 1.456 (p < 0.001), TSAT OR = 0.614 (p = 0.003), female sex OR = 1.976 (p = 0.026), ferritin OR = 1.345 (p = 0.022) — directionally consistent with full model. OR: odds ratio; CI: confidence interval; VIF: variance inflation factor; HbA1c: glycated haemoglobin; TG/HDL: triglyceride-to-HDL cholesterol ratio; NLR: neutrophil-to-lymphocyte ratio; NMR: neutrophil-to-monocyte ratio; SII: systemic immune-inflammation index; AISI: aggregate index of systemic inflammation; CRP: C-reactive protein; TSAT: transferrin saturation; BMI: body mass index; eGFR: estimated glomerular filtration rate; DM: diabetes mellitus. *p < 0.05; *p < 0.01.

Forest plot of multivariable odds ratios for borderline hepatocellular and cholestatic patterns versus the normal group.

Adjusted odds ratios (OR) with 95% confidence intervals from separate binary logistic regression models comparing the borderline hepatocellular pattern (blue; n = 105) and the borderline cholestatic pattern (orange; n = 77) to the normal group (n = 548). Models were adjusted for 18 covariates (HbA1c, TSAT, ferritin, SII, NLR, NMR, AISI, CRP, albumin, age, sex, BMI, eGFR, TG/HDL rati