Circulating Soluble EGFR Correlates with Tumor Burden and Hepatic Reserve but Lacks Independent Prognostic Value in Hepatocellular Carcinoma: A Prospective Turkish Cohort

Abstract

Purpose

Serum soluble epidermal growth factor receptor (sEGFR) is a circulating extracellular-domain EGFR fragment that is biologically distinct from membrane-bound tissue EGFR, and its clinical meaning in hepatocellular carcinoma (HCC) has not been defined. We evaluated whether baseline serum sEGFR is related to tumor burden and hepatic reserve, and whether it carries prognostic information independent of established clinical variables.

Patients and Methods

This prospective single-center cohort included 84 patients with HCC evaluated between November 2014 and May 2017. Baseline serum sEGFR was measured by ELISA. Overall survival (OS) was defined from baseline serum sample collection to death from any cause or last follow-up and was analyzed using Kaplan-Meier estimates, lg-rank testing, and Cox proportional hazards models with bootstrap internal validation. The incremental value of sEGFR over alpha-fetoprotein (AFP) and MELD-Na was tested in a combined model.

Results

Lower serum sEGFR was associated with larger tumor size, higher AFP, higher aspartate aminotransferase and bilirubin levels, worse MELD-Na and Child-Turcotte-Pugh scores, lower insulin-like growth factor-1, and distant metastasis. Patients with sEGFR ≤13 ng/mL had shorter median OS than those with sEGFR >13 ng/mL (5.4 vs 12.1 months; log-rank p=0.0027; univariable HR 2.06, 95% CI 1.27–3.34). After adjustment for tumor size, log-transformed AFP, MELD-Na, portal vein thrombosis, and age, sEGFR was not associated with OS (HR 0.99, 95% CI 0.95–1.03, p=0.69), and the estimate remained close to the null across 1000 bootstrap resamples. Adding continuous sEGFR to a model containing AFP and MELD-Na did not improve discrimination (C-index 0.657 vs 0.656).

Conclusion

Low serum sEGFR identifies an adverse tumor-liver phenotype and shorter unadjusted survival, but it is not an independent prognostic factor and adds no discriminative information beyond AFP and MELD-Na. Serum sEGFR is best regarded as a marker of disease phenotype; it does not perform as a stand-alone prognostic test.

Introduction

Hepatocellular carcinoma (HCC) remains a leading cause of cancer mortality worldwide. In GLOBOCAN 2022, liver cancer ranked sixth in global cancer incidence and third in cancer-related mortality, accounting for more than 865,000 new cases and more than 757,000 deaths.¹ Prognosis in HCC is difficult to estimate because survival is shaped by two overlapping processes: progression of the malignant tumor and functional decline of the underlying liver. This dual biology separates HCC from many other solid tumors, in which anatomic stage alone often carries greater prognostic weight. In clinical practice, patients with similar tumor burden may have markedly different outcomes depending on hepatic reserve, portal hypertension, performance status, and treatment tolerance.^2,3

Current prognostic systems reflect this clinical reality. The Barcelona Clinic Liver Cancer (BCLC) framework links tumor burden, vascular invasion, extrahepatic spread, performance status, and liver function to stage-specific treatment recommendations.^2–4 Hepatic reserve is most commonly assessed using the Child-Turcotte-Pugh (CTP) classification, while the Model for End-Stage Liver Disease (MELD) and MELD-Na scores provide more objective laboratory-based estimates of liver disease severity.^5–7 The albumin-bilirubin (ALBI) grade was later developed to refine liver-function assessment in HCC using only serum albumin and bilirubin.⁸ These systems remain central to clinical decision-making, but they do not fully capture circulating tumor-host biology. Serum biomarkers may therefore provide complementary information, particularly when they reflect both malignant progression and the condition of the non-tumoral liver.

Most circulating protein markers studied in HCC are secreted or shed glycoproteins whose serum concentrations rise with tumor mass. Alpha-fetoprotein (AFP) remains the reference example, and glypican-3, Dickkopf-1, Golgi protein 73, and osteopontin have each been proposed as tumor-associated serum markers that are elevated in HCC relative to non-malignant liver disease.^9–12 Secreted lectin-like glycoproteins such as zymogen granule protein 16B have more recently been described as regulators of tumor progression and immune microenvironment remodeling across several cancers.¹³ In each case the working assumption is that a secreted tumor product rises as the tumor grows. A circulating receptor ectodomain need not behave this way, because its serum level depends on where the receptor is expressed and how it is processed, and on the synthetic capacity of the liver, as well as on tumor secretion. Whether a soluble receptor fragment behaves like a tumor-secreted marker in HCC has not been examined, and that question is the starting point of the present study.

The epidermal growth factor receptor (EGFR/ErbB1) pathway has biologic relevance in chronic liver injury and HCC. EGFR signaling participates in hepatocyte proliferation, liver regeneration, fibrogenesis, inflammatory signaling, and neoplastic transformation.¹⁴ Its role in HCC, however, is not linear. Tissue EGFR expression has been reported in a substantial proportion of conventional HCCs, but its relationship with gene copy number, clinicopathologic features, and survival has been inconsistent.¹⁵ Experimental and human tissue data also indicate that EGFR effects are compartment-specific: macrophage EGFR may promote inflammation-driven hepatocarcinogenesis, whereas hepatocyte EGFR can retain injury-protective functions in some settings.¹⁶ This biology argues against interpreting all EGFR-related measurements as equivalent. In particular, circulating EGFR should not be treated as a simple surrogate for tumor-cell membrane EGFR expression.

Serum EGFR measured by ELISA is best interpreted as serum soluble epidermal growth factor receptor (sEGFR), an immunoreactive circulating extracellular-domain EGFR signal. Soluble EGFR species may arise from alternative EGFR transcripts encoding truncated extracellular-domain isoforms and from proteolytic ectodomain shedding of the full-length receptor.^17,18 These circulating species lack the transmembrane and intracellular kinase domains that define membrane-bound EGFR signaling. As a result, serum sEGFR may behave differently from tissue EGFR and may reflect receptor processing, ligand-binding capacity, host epithelial or hepatic biology, systemic disease state, or a combination of these factors. Soluble ErbB-family proteins have also been proposed to modulate ligand availability and receptor-family signaling, supporting the concept that sEGFR is a distinct biological analyte rather than a diluted form of membrane EGFR.^19,20

Clinical data on serum sEGFR in HCC remain limited. The most directly relevant HCC-specific evidence comes from an exploratory serum N-glycoproteomic study in which lower circulating sEGFR was associated with metastatic HCC.²¹ That finding provides a biologically plausible precedent for examining whether low baseline sEGFR identifies patients with adverse tumor features. No study, to our knowledge, has tested serum sEGFR against standard liver-reserve indices, examined its survival relevance in a prospectively assembled clinical cohort, or asked whether it adds prognostic information to established variables. We therefore evaluated baseline serum sEGFR in a prospective Turkish cohort of patients with HCC, focusing on its associations with tumor burden, hepatic reserve, other circulating biomarkers, and overall survival, and whether any survival association persisted after adjustment for established prognostic variables.

Materials and Methods

Patients and Study Design

This was a single-center observational study of a prospective cohort of patients with HCC evaluated at Hacettepe University Cancer Institute between November 2014 and May 2017. A total of 84 adult patients aged 18 years or older were enrolled. Baseline serum samples were obtained at study enrollment or first admission before biomarker analysis. Patients were followed prospectively for survival. Serum sEGFR was measured later on the stored baseline samples, after ethics approval had been obtained in March 2015; the study is therefore a retrospective analysis of prospectively collected samples and data. The cohort is the same prospectively assembled series in which serum insulin-like growth factor-1 was previously evaluated;²² the present analysis addresses a different analyte and a different scientific question, and no analytic parameter from that report was carried over.

HCC was diagnosed histopathologically or radiologically according to criteria compatible with the American Association for the Study of Liver Diseases and European Association for the Study of the Liver diagnostic recommendations. For cirrhotic patients without histologic confirmation, HCC was diagnosed from characteristic arterial-phase hyperenhancement with washout on multiphasic computed tomography or dynamic contrast-enhanced magnetic resonance imaging. A diagnosis of cirrhosis was based on histology where available; otherwise it was inferred from concordant radiologic, laboratory, and clinical evidence of portal hypertension and hepatic decompensation. Patients with an additional malignancy other than HCC were excluded.

No formal a priori sample size calculation was performed, because this was an exploratory biomarker analysis of an existing prospective cohort with a fixed number of participants. The precision that the cohort affords is described under Statistical Analysis.

Clinical and Laboratory Variables

Baseline demographic, clinical, laboratory, and tumor-related variables were prospectively recorded. These included age, sex, hepatitis B virus and hepatitis C virus status, cirrhosis status, CTP class and score, MELD, MELD-Na, BCLC stage, AFP, liver-function tests, serum bilirubin, albumin, international normalized ratio, aspartate aminotransferase, alanine aminotransferase, tumor size, portal vein thrombosis, macrovascular invasion, lymph node involvement, and distant metastasis. Baseline insulin-like growth factor-1 (IGF-1) was also evaluated as a comparator serum marker reflecting hepatic reserve.

Serum sEGFR Measurement

Peripheral blood samples were collected at baseline and stored at −80°C until analysis. Before assay, samples were thawed and brought to room temperature as specified in the manufacturer’s protocol. Serum EGFR concentrations were measured in duplicate using a commercial sandwich enzyme-linked immunosorbent assay (Human EGFR ELISA Kit, Elabscience, catalog no. E-EL-H0060) according to the manufacturer’s protocol. In this assay, samples and calibrators are added to microplate wells pre-coated with an EGFR-specific capture antibody and incubated; a biotinylated detection antibody and an avidin-horseradish peroxidase conjugate are then added sequentially with intervening wash steps; a tetramethylbenzidine substrate is added, the reaction is stopped, and optical density is read at 450 nm. Concentrations were interpolated from the calibrator curve, and the mean of duplicate wells was used for analysis. The manufacturer reports an assay sensitivity of 0.19 ng/mL. Intra- and inter-assay coefficients of variation were not recorded during the original study period, were not reported in the previously published analyses of this cohort, and cannot be reconstructed retrospectively, so no assay precision data are available; this is acknowledged as a limitation. Because the assay detects circulating EGFR protein in serum, the measured analyte was interpreted as serum sEGFR, representing an immunoreactive extracellular-domain EGFR signal rather than membrane-bound EGFR.

Study Outcomes

The primary outcome was overall survival (OS), defined as the time from baseline serum sample collection to death from any cause or last follow-up. Patients who were alive at last follow-up were censored on that date. Landmark survival probabilities at 12, 24, and 36 months were estimated from the Kaplan-Meier curves. Five-year survival could not be estimated because the maximum follow-up from serum sampling was 45.9 months.

Statistical Analysis

Descriptive statistics were used to summarize baseline clinical and laboratory characteristics. Continuous variables were reported as mean with standard deviation or median with interquartile range, as appropriate. Categorical variables were summarized as frequencies and percentages. Normality was assessed using the Shapiro–Wilk test.

Associations between serum sEGFR and continuous clinical or laboratory variables were assessed using correlation analyses. Spearman correlation coefficients were emphasized for non-normally distributed variables, and Pearson correlation was used where appropriate. Groupwise comparisons of serum sEGFR were performed using the Mann–Whitney U-test for two-group comparisons and the Kruskal–Wallis test for comparisons involving more than two groups. As an exploratory sensitivity analysis, correlations between sEGFR and tumor size, AFP, and MELD-Na were repeated within BCLC strata (very early or early, intermediate, and advanced or terminal); these subgroup estimates are reported for transparency and are not powered for inference.

Serum sEGFR was analyzed both as a continuous variable and as a dichotomized variable. A cutoff of 13 ng/mL was selected based on the cohort median. Receiver operating characteristic analysis with Youden index identified a similar cutoff of 12.95 ng/mL; this value was considered supportive of the median-based threshold and was not interpreted as an externally validated cutoff. The threshold applies to this cohort only and has not been verified in an external population.

OS from baseline serum sample collection was estimated using the Kaplan-Meier method, and survival distributions were compared using the log-rank test. Univariable Cox proportional hazards regression was used to evaluate associations between candidate variables and OS. A multivariable Cox proportional hazards model was then fitted including serum sEGFR as a continuous variable, tumor size, log-transformed AFP, MELD-Na, portal vein thrombosis, and age. The proportional hazards assumption was assessed using scaled Schoenfeld residuals. Model discrimination was assessed using Harrell’s C-index. Internal validation of the multivariable model was performed by bootstrap resampling with 1000 replicates: the model was refitted in each resample, apparent and test discrimination were compared, and the mean optimism was subtracted from the apparent C-index to yield an optimism-corrected estimate. The stability of the sEGFR coefficient was summarized as the median bootstrap hazard ratio with its 2.5^th to 97.5^th percentile interval and as the proportion of resamples in which sEGFR reached nominal significance.

To quantify incremental prognostic value, a parsimonious base model containing log-transformed AFP and MELD-Na was compared with extended models that added serum sEGFR, first as a continuous variable and then as the dichotomized variable. Improvement was assessed by the likelihood ratio test and by the change in apparent and optimism-corrected C-index (500 bootstrap resamples per model). As a further exploratory analysis, OS was compared by sEGFR group within patients who received sorafenib, and a sEGFR-by-sorafenib interaction term was tested in the whole cohort to assess whether the association between sEGFR and survival differed according to treatment. The minimum detectable effect was calculated from the observed number of events using the Schoenfeld formula: with 69 deaths, a two-sided alpha of 0.05, 80% power, and an exposure prevalence of 0.51, the cohort could detect a hazard ratio of 1.96 or larger for a binary predictor and 1.40 or larger per standard deviation for a continuous predictor. All statistical tests were two-sided, and p<0.05 was considered statistically significant. Survival analyses, bootstrap validation, and combined-model comparisons were performed in Python 3.12 using the lifelines and SciPy libraries.

Ethical Aspects

This study was planned and conducted in accordance with the Declaration of Helsinki. The study protocol was reviewed and approved by the Ethics Committee of Hacettepe University Faculty of Medicine (approval date: March 4, 2015; approval number: 15/153). Written informed consent was obtained from all participants before enrollment and blood sample collection.

Results

Baseline Patient Characteristics

A total of 84 patients with HCC were included in the analysis. The median age was 64 years (range, 19–90 years). Thirteen patients were female and 71 were male. Hepatitis B virus infection was present in 36 patients, and hepatitis C virus infection was present in 8 patients. According to CTP class, 58 patients were CTP class A, 22 were class B, and 3 were class C; CTP class was unavailable for one patient. BCLC stage distribution was as follows: very early stage in 1 patient, early stage in 18 patients, intermediate stage in 13 patients, advanced stage in 50 patients, and terminal stage in 2 patients. At the time of analysis, 69 patients had died. Baseline demographic, clinical, laboratory, and tumor-related characteristics are summarized in Table 1.

Table 1.

Baseline Demographic and Clinical Characteristics of the Study Cohort

Characteristic	Overall Cohort (N=84)
Age, years, median (range)	64 (19–90)
Sex, n (%)
Female	13 (15.5)
Male	71 (84.5)
Hepatitis B virus-positive, n (%)	36 (42.9)
Hepatitis C virus-positive, n (%)	8 (9.5)
Cirrhosis status, n (%)
No	35 (41.7)
Yes	48 (57.1)
Not reported	1 (1.2)
CTP class, n (%)
A	58 (69.0)
B	22 (26.2)
C	3 (3.6)
Not reported	1 (1.2)
BCLC stage, n (%)
Very early	1 (1.2)
Early	18 (21.4)
Intermediate	13 (15.5)
Advanced	50 (59.5)
Terminal	2 (2.4)
AFP, ng/mL, median (IQR)	128.50 (6.04–1369.90)
AST, U/L, median (IQR)	50.00 (36.25–73.50)
Bilirubin, mg/dL, median (IQR)	1.08 (0.76–1.77)
MELD-Na score, median (IQR)	9.00 (6.00–13.00)
IGF-1, ng/mL, median (IQR)	18.31 (8.02–47.58)
Tumor size, cm, median (IQR)	7.10 (4.00–11.93)
Tumor size >5 cm, n/N (%)	52/82 (63.4)
Portal vein thrombosis, n/N (%)	36/83 (43.4)
Macrovascular invasion, n/N (%)	39/83 (47.0)
Lymph node involvement, n (%)	23 (27.4)
Distant metastasis, n (%)	17 (20.2)
Deaths during follow-up, n (%)	69 (82.1)
Median OS from baseline serum sample collection, months	7.38

Notes: Percentages are based on the overall cohort unless a nonmissing denominator (n/N) is specified.

Abbreviations: AFP, alpha-fetoprotein; AST, aspartate aminotransferase; BCLC, Barcelona Clinic Liver Cancer; CTP, Child-Turcotte-Pugh; IGF-1, insulin-like growth factor-1; IQR, interquartile range; MELD-Na, Model for End-Stage Liver Disease-sodium; OS, overall survival.

Serum sEGFR Distribution

Serum sEGFR was measured in all 84 patients. The mean serum sEGFR level was 13.03 ng/mL with a standard deviation of 7.39 ng/mL. The median serum sEGFR level was 12.94 ng/mL, with an interquartile range of 7.55–19.79 ng/mL and an observed range of 0.324–26.30 ng/mL. Using a median-based cutoff of 13 ng/mL, 43 patients were classified as having serum sEGFR ≤13 ng/mL and 41 patients as having serum sEGFR >13 ng/mL. Receiver operating characteristic analysis yielded an AUC of 0.591, with a Youden-index cutoff of 12.95 ng/mL, closely corresponding to the median-based threshold.

Associations Between Serum sEGFR and Continuous Variables

Serum sEGFR showed significant inverse correlations with several tumor- and liver-related variables. Lower serum sEGFR was associated with larger tumor size (Spearman ρ=−0.374, p=0.0005) and higher AFP level (ρ=−0.385, p=0.0003). Lower serum sEGFR was also associated with higher aspartate aminotransferase level (ρ=−0.261, p=0.018), higher bilirubin level (ρ=−0.242, p=0.027), and higher MELD-Na score (ρ=−0.240, p=0.029). Serum sEGFR was inversely associated with CTP score and positively correlated with IGF-1 level (ρ=+0.264, p=0.015). Serum sEGFR was not significantly correlated with VEGF-A (p=0.135) or glypican-3 (p=0.675). Correlation analyses are shown in Table 2.

Table 2.

Correlations Between Serum sEGFR and Continuous Clinicobiochemical Variables

Variable	Direction	Spearman rho	p-value
Tumor size	Inverse	−0.374	0.0005
AFP	Inverse	−0.385	0.0003
AST	Inverse	−0.261	0.0177
Bilirubin	Inverse	−0.242	0.0273
MELD-Na score	Inverse	−0.240	0.0289
CTP score	Inverse	−0.247	0.0244
IGF-1	Positive	+0.264	0.0152
VEGF-A	Not significant	−0.164	0.1352
Glypican-3	Not significant	+0.046	0.6749

Notes: Spearman correlations are shown because most variables were not normally distributed.

Abbreviations: AFP, alpha-fetoprotein; AST, aspartate aminotransferase; CTP, Child-Turcotte-Pugh; IGF-1, insulin-like growth factor-1; MELD-Na, Model for End-Stage Liver Disease-sodium; sEGFR, serum soluble epidermal growth factor receptor; VEGF-A, vascular endothelial growth factor-A.

Groupwise Comparisons of Serum sEGFR

Serum sEGFR differed significantly according to tumor size category, AFP category, aspartate aminotransferase category, distant metastasis status, and CTP class. Lower sEGFR levels were observed among patients with tumors >5 cm compared with those with tumors ≤5 cm, among patients with AFP >400 ng/mL compared with those with AFP ≤400 ng/mL, and among patients with aspartate aminotransferase >50 U/L compared with those with ≤50 U/L. Serum sEGFR levels were also lower in patients with distant metastasis and differed across CTP classes. A sex-based difference was observed in groupwise analysis, although this comparison should be interpreted in the context of the small number of female patients.

Serum sEGFR did not differ significantly according to portal vein thrombosis, macrovascular invasion, lymph node involvement, cirrhosis status, alanine aminotransferase category, or BCLC class. Groupwise comparisons are presented in Table 3.

Table 3.

Groupwise Comparisons of Serum sEGFR According to Clinical and Tumor-Related Variables

Variable/Group	n	Serum sEGFR, ng/mL, Median (IQR)	p-value
Tumor size ≤5 cm	30	17.25 (10.03–22.28)	0.0100
Tumor size >5 cm	52	10.94 (5.71–17.94)
AFP ≤400 ng/mL	53	16.70 (8.53–21.30)	0.0008
AFP >400 ng/mL	30	7.76 (2.54–14.41)
AST ≤50 U/L	42	16.71 (10.87–19.88)	0.0123
AST >50 U/L	40	8.92 (5.18–17.94)
Distant metastasis absent	67	15.27 (8.23–19.90)	0.0075
Distant metastasis present	17	7.03 (2.45–11.03)
CTP class A	58	14.98 (8.14–19.90)	0.0397
CTP class B	22	11.73 (7.78–17.77)
CTP class C	3	2.45 (1.70–2.82)
Female	13	19.90 (12.54–22.70)	0.0091
Male	71	12.49 (6.88–18.03)
Portal vein thrombosis absent	47	15.62 (8.23–20.10)	0.1681
Portal vein thrombosis present	36	10.90 (7.01–18.73)
Macrovascular invasion absent	44	15.48 (8.28–20.00)	0.2392
Macrovascular invasion present	39	10.96 (6.88–18.84)
Lymph node involvement absent	61	15.27 (8.04–19.90)	0.1602
Lymph node involvement present	23	10.84 (5.88–15.75)
Cirrhosis absent	35	13.59 (5.58–20.25)	0.9743
Cirrhosis present	48	12.56 (7.81–18.31)
ALT ≤50 U/L	58	14.13 (8.17–19.88)	0.1869
ALT >50 U/L	24	10.92 (5.74–18.19)
BCLC very early/early	19	16.69 (7.23–22.25)	0.5407
BCLC intermediate	13	13.29 (5.84–19.77)
BCLC advanced/terminal	52	12.52 (7.74–18.19)

Notes: P-values were derived from Mann–Whitney U-tests for two-group comparisons and Kruskal–Wallis tests for comparisons involving more than two groups.

Abbreviations: AFP, alpha-fetoprotein; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BCLC, Barcelona Clinic Liver Cancer; CTP, Child-Turcotte-Pugh; IQR, interquartile range; sEGFR, serum soluble epidermal growth factor receptor.

Overall Survival According to Serum sEGFR

The median OS from baseline serum sample collection was 7.38 months. Patients with serum sEGFR ≤13 ng/mL had shorter median OS than patients with serum sEGFR >13 ng/mL. Median OS was 5.42 months in the sEGFR ≤13 ng/mL group and 12.06 months in the sEGFR >13 ng/mL group. The difference between survival curves was statistically significant by log-rank testing (χ²=9.012, p=0.0027). Landmark survival probabilities in the whole cohort were 41.7% (95% CI 31.1–51.9) at 12 months, 23.5% (95% CI 15.0–33.0) at 24 months, and 16.8% (95% CI 9.2–26.2) at 36 months. In the sEGFR ≤13 ng/mL group these were 32.6% (95% CI 19.3–46.5), 13.6% (95% CI 5.3–25.6), and 5.4% (95% CI 1.0–15.7), compared with 51.2% (95% CI 35.1–65.2), 33.6% (95% CI 19.7–48.1), and 28.8% (95% CI 14.9–44.4) in the sEGFR >13 ng/mL group. Because the longest follow-up from serum sampling was 45.9 months, 5-year survival could not be estimated. Kaplan-Meier survival curves for the whole cohort and according to serum sEGFR group are shown in Figures 1 and 2, respectively.

Figure 1.

Kaplan-Meier curve showing overall survival (OS) in the entire hepatocellular carcinoma cohort (n=84; 69 deaths). OS was defined as the time from baseline serum sample collection to death from any cause; patients alive at last follow-up were censored, and censoring is shown by vertical ticks. Median OS was 7.38 months. Numbers at risk are shown below the curve.

Figure 2.

Kaplan-Meier curves showing OS stratified by baseline serum soluble epidermal growth factor receptor (sEGFR) level. OS was defined from baseline serum sample collection; censoring is shown by vertical ticks. The dark red solid line represents sEGFR ≤13 ng/mL (n=43; 40 deaths; median OS 5.42 months) and the dark blue dashed line represents sEGFR >13 ng/mL (n=41; 29 deaths; median OS 12.06 months); log-rank p=0.0027. Numbers at risk are shown below the curves.

Cox Regression Analyses

In univariable Cox regression analysis, dichotomized serum sEGFR was associated with OS. Patients with serum sEGFR ≤13 ng/mL had increased mortality risk compared with those with serum sEGFR >13 ng/mL (HR 2.060, 95% CI 1.272–3.338, p=0.0033). Serum sEGFR analyzed as a continuous variable was not significantly associated with OS in univariable analysis (HR 0.974, 95% CI 0.943–1.007, p=0.1166). Sensitivity parameterizations using serum sEGFR per interquartile-range increase and log-transformed serum sEGFR were also not significant.

A multivariable Cox proportional hazards model including serum sEGFR as a continuous variable, tumor size, log-transformed AFP, MELD-Na, portal vein thrombosis, and age included 80 patients and 66 events (11 events per covariate), with a Harrell C-index of 0.663. In this model, serum sEGFR was not associated with OS after adjustment (HR 0.992, 95% CI 0.952–1.033, p=0.6874). MELD-Na and age were associated with OS in the adjusted model, whereas tumor size showed borderline association and log-transformed AFP and portal vein thrombosis were not significant. Bootstrap internal validation with 1000 resamples gave a mean optimism of 0.03 and an optimism-corrected C-index of 0.63. The sEGFR hazard ratio had a bootstrap median of 0.99 (2.5^th to 97.5^th percentile 0.95–1.04) and reached nominal significance in only 7.1% of resamples, so the estimate was consistently close to the null across resamples and did not depend on a single model fit. Cox regression analyses are summarized in Table 4.

Table 4.

Univariable and Multivariable Cox Regression Analyses for Overall Survival from Baseline Serum Sample Collection

Variable	Univariable HR (95% CI)	p-value	Multivariable HR (95% CI)	p-value
sEGFR ≤13 vs >13 ng/mL	2.060 (1.272–3.338)	0.0033	Not included	-
sEGFR, continuous	0.974 (0.943–1.007)	0.1166	0.992 (0.952–1.033)	0.6874
sEGFR, per IQR increase	0.725 (0.485–1.083)	0.1166	Not included	-
log(sEGFR)	0.884 (0.713–1.098)	0.2651	Not included	-
Tumor size	1.082 (1.037–1.130)	0.0003	1.061 (0.998–1.127)	0.0583
log(AFP)	1.108 (1.039–1.182)	0.0018	1.060 (0.986–1.141)	0.1166
MELD-Na	1.050 (1.018–1.083)	0.0020	1.056 (1.018–1.096)	0.0035
Portal vein thrombosis	2.071 (1.280–3.349)	0.0030	1.169 (0.613–2.229)	0.6357
Age	1.019 (0.998–1.040)	0.0844	1.025 (1.004–1.046)	0.0204

Notes: Overall survival was calculated from baseline serum sample collection to death or last follow-up. The multivariable model included 80 patients and 66 events. Apparent Harrell C-index=0.663; bootstrap internal validation (1000 resamples) gave a mean optimism of 0.03 and an optimism-corrected C-index of 0.63. The proportional hazards assumption was satisfied on scaled Schoenfeld residuals.

Abbreviations: AFP, alpha-fetoprotein; CI, confidence interval; HR, hazard ratio; IQR, interquartile range; MELD-Na, Model for End-Stage Liver Disease-sodium; sEGFR, serum soluble epidermal growth factor receptor.

Incremental Value of Serum sEGFR in a Combined Model

A base model containing log-transformed AFP and MELD-Na (82 patients, 67 events) had an apparent C-index of 0.657 and an optimism-corrected C-index of 0.650 (Table 5). Adding serum sEGFR as a continuous variable did not improve the model: the apparent C-index was 0.656, the optimism-corrected C-index 0.641, the likelihood ratio test p=0.56, and the adjusted hazard ratio 0.99 (95% CI 0.96–1.03). When sEGFR was instead entered as the dichotomized variable, the low-sEGFR group retained an association with mortality (HR 1.75, 95% CI 1.07–2.88, p=0.027; likelihood ratio p=0.026), but the gain in discrimination was negligible (apparent C-index 0.666; optimism-corrected 0.653, a change of 0.003). When the dichotomized variable was entered into the full prespecified model in place of continuous sEGFR, the association was attenuated and no longer significant (HR 1.59, 95% CI 0.95–2.65, p=0.079), consistent with sEGFR carrying information that overlaps with tumor burden and liver reserve.

Table 5.

Incremental Prognostic Value of Serum sEGFR Added to a Base Model Containing AFP and MELD-Na

Model	Apparent C-Index	Optimism-Corrected C-Index	sEGFR HR (95% CI)	sEGFR p-value	Likelihood Ratio p
Base: log(AFP) + MELD-Na	0.657	0.650	-	-	-
Base + sEGFR (continuous)	0.656	0.641	0.990 (0.956–1.025)	0.559	0.560
Base + sEGFR ≤13 vs >13 ng/mL	0.666	0.653	1.753 (1.067–2.880)	0.027	0.026

Notes: 82 patients and 67 events. Optimism-corrected C-index from bootstrap resampling (500 resamples per model). The likelihood ratio test compares each extended model with the base model. When the dichotomized variable replaced continuous sEGFR in the full prespecified model (tumor size, log(AFP), MELD-Na, portal vein thrombosis, age; 80 patients, 66 events), the association was attenuated and not significant (HR 1.586, 95% CI 0.948–2.653, p=0.079).

Abbreviations: AFP, alpha-fetoprotein; CI, confidence interval; HR, hazard ratio; MELD-Na, Model for End-Stage Liver Disease-sodium; sEGFR, serum soluble epidermal growth factor receptor.

Exploratory Analysis within Sorafenib-Treated Patients

Thirty-seven patients received sorafenib; the baseline serum sample was obtained before sorafenib was started in 19, on the same day in 5, and after treatment had begun in 13. Within the sorafenib-treated group, median OS was 7.82 months in patients with sEGFR ≤13 ng/mL (n=18, 16 events) and 18.40 months in those with sEGFR >13 ng/mL (n=19, 13 events). The difference did not reach significance (log-rank p=0.071; HR 1.95, 95% CI 0.93–4.08, p=0.076) and was of the same magnitude as the unadjusted association in the whole cohort. In the whole cohort, the interaction between sEGFR group and sorafenib treatment was not significant (HR 0.86, 95% CI 0.33–2.28, p=0.77), so there was no evidence that the association of sEGFR with survival differed according to whether sorafenib was given. Among the 19 patients sampled before sorafenib initiation, survival did not differ by sEGFR group (log-rank p=0.51). These analyses were exploratory, were limited by small numbers and non-randomized treatment allocation, and provide no evidence that sEGFR predicts benefit from sorafenib.

Stage-Stratified Correlations

When correlations were repeated within BCLC strata (Table 6), the inverse associations of sEGFR with tumor size (ρ=−0.489, p<0.001) and MELD-Na (ρ=−0.564, p<0.001) were concentrated in the advanced or terminal stratum (n=52), with a borderline inverse association with AFP (ρ=−0.271, p=0.055). In the intermediate stratum (n=13) only the AFP correlation reached nominal significance (ρ=−0.555, p=0.049), and in the very early or early stratum (n=19) no association with tumor size was present while the MELD-Na correlation was positive (ρ=+0.620, p=0.006). These small-stratum estimates are unstable and are reported for transparency; the cohort-level associations are driven mainly by patients with advanced disease, who formed 62% of the sample.

Table 6.

Exploratory Stage-Stratified Spearman Correlations Between Serum sEGFR and Tumor Size, AFP, and MELD-Na

BCLC Stratum	Variable	n	Spearman rho	p-value
Very early/early	Tumor size	18	−0.010	0.968
Very early/early	AFP	19	−0.330	0.168
Very early/early	MELD-Na	18	+0.620	0.006
Intermediate	Tumor size	13	−0.237	0.436
Intermediate	AFP	13	−0.555	0.049
Intermediate	MELD-Na	13	−0.039	0.900
Advanced/terminal	Tumor size	51	−0.489	<0.001
Advanced/terminal	AFP	51	−0.271	0.055
Advanced/terminal	MELD-Na	52	−0.564	<0.001

Notes: Subgroup sizes are small and these estimates are not powered for inference; they are reported to show that the cohort-level associations are concentrated in advanced-stage disease.

Abbreviations: AFP, alpha-fetoprotein; BCLC, Barcelona Clinic Liver Cancer; MELD-Na, Model for End-Stage Liver Disease-sodium; sEGFR, serum soluble epidermal growth factor receptor.

Discussion

In this prospective cohort of patients with HCC, lower serum sEGFR was associated with a clinically adverse phenotype characterized by larger tumor size, higher AFP, higher aspartate aminotransferase and bilirubin levels, worse MELD-Na and CTP scores, lower IGF-1, distant metastasis, and shorter unadjusted OS from baseline serum sample collection. Patients with serum sEGFR ≤13 ng/mL had substantially shorter median OS than those with higher levels. However, serum sEGFR was not associated with OS after adjustment for tumor size, AFP, MELD-Na, portal vein thrombosis, and age, the estimate remained close to the null under bootstrap resampling, and continuous sEGFR added no discriminative information to a model containing AFP and MELD-Na. These findings support a cautious interpretation: low circulating sEGFR marks an adverse tumor-liver phenotype rather than serving as a stand-alone prognostic test.

Serum sEGFR behaved differently from the secreted glycoprotein markers that make up most of the HCC biomarker literature. AFP, glypican-3, Dickkopf-1, Golgi protein 73, and osteopontin are tumor-associated products whose serum concentrations are elevated in HCC and, in the case of AFP, rise with tumor mass,^9–12 and secreted lectin-like glycoproteins such as zymogen granule protein 16B have been shown to promote tumor progression directly.¹³ Serum sEGFR moved in the opposite direction, falling as tumor burden increased and as hepatic reserve deteriorated. A soluble receptor ectodomain is not a tumor secretion product in the same sense. Its concentration in serum depends on receptor expression and proteolytic processing in host tissues, including the non-tumoral liver, and may fall as functional hepatic mass is lost. The present data fit this interpretation and suggest that sEGFR is better understood as a host-derived indicator of combined tumor and liver status than as a conventional tumor marker.

This interpretation is important because EGFR biology in HCC is not equivalent across compartments. EGFR signaling has well-established relevance in liver injury, regeneration, fibrogenesis, inflammation, and hepatocarcinogenesis.¹⁴ Tissue EGFR expression has been reported in a substantial proportion of conventional HCCs, but associations with clinicopathologic variables, gene copy number, and survival have not been uniform.¹⁵ Experimental data also indicate that EGFR signaling can have cell-specific effects in the liver, with macrophage EGFR contributing to inflammation-driven hepatocarcinogenesis and hepatocyte EGFR retaining injury-protective functions in selected contexts.¹⁶ Serum sEGFR should therefore not be interpreted as a direct proxy for tumor-cell membrane EGFR. Circulating sEGFR represents extracellular-domain EGFR immunoreactivity and may arise from alternative EGFR transcripts encoding soluble isoforms, proteolytic ectodomain shedding, or both.¹⁷ These soluble species lack the transmembrane and intracellular kinase domains and may have biological behavior distinct from membrane-bound EGFR signaling.^18–20

The inverse association between serum sEGFR and tumor burden was one of the clearest findings in this cohort. Lower sEGFR correlated with larger tumor size and higher AFP, and patients with distant metastasis had lower sEGFR levels than those without metastasis. These findings are consistent with the survival separation observed using the 13 ng/mL threshold, because tumor size, AFP, and metastatic spread each reflect clinically meaningful disease burden. At the same time, the tumor-related pattern was not uniform across all anatomic markers of disease extension. Serum sEGFR did not differ significantly according to portal vein thrombosis, macrovascular invasion, or lymph node involvement. Serum sEGFR therefore appears to track overall disease burden more than any single marker of vascular or nodal spread. The stage-stratified analysis supports this reading. The inverse associations with tumor size and MELD-Na were concentrated in patients with advanced disease, whereas the small early-stage stratum showed no tumor-size association and an unstable positive correlation with MELD-Na. The cohort-level findings therefore describe sEGFR behavior mainly in advanced HCC and should not be extrapolated to early-stage or surveillance populations without direct study.

The association between sEGFR and hepatic reserve was also clinically relevant. Lower sEGFR correlated with higher bilirubin, higher MELD-Na, higher CTP score, and lower IGF-1. These findings place sEGFR within a tumor-host context rather than a purely tumor-centered framework. IGF-1 is largely liver-derived and decreases with impaired hepatic synthetic function; it has also been incorporated into objective modifications of CTP-based assessment in HCC.^22,23 The positive correlation between sEGFR and IGF-1 therefore supports the possibility that circulating sEGFR is partly linked to hepatic reserve or systemic host condition. This may also explain why sEGFR did not retain a survival association after adjustment. Once tumor size and MELD-Na were included in the multivariable model, much of the prognostic information carried by sEGFR may have been accounted for by these dominant clinical determinants.

The HCC-specific literature on circulating sEGFR remains limited, but the available data are directionally compatible with the present findings. In an exploratory serum N-glycoproteomic study, lower circulating sEGFR was associated with metastatic HCC, and experimental validation showed lower sEGFR expression and secretion in metastatic-potential HCC cell lines compared with non-metastatic lines.²¹ The present study extends that clinical direction by showing that low sEGFR is also associated with larger tumor size, higher AFP, impaired liver-reserve indices, and shorter unadjusted OS from baseline serum sample collection. However, the two studies should not be treated as directly interchangeable. The prior work focused on metastatic status and biomarker discovery rather than survival prognostication, and differences in assay platform, cohort composition, disease stage, and cutoff selection limit numerical comparisons. The common finding is limited but clinically relevant: lower serum sEGFR is seen in patients with more adverse HCC features.

Serum correlations cannot establish why circulating sEGFR falls in advanced disease. Three hypotheses, which are not mutually exclusive, are consistent with the data. First, loss of functional hepatic mass may reduce the pool of hepatocyte membrane EGFR available for ectodomain shedding, which would explain the parallel decline of sEGFR and IGF-1 with worsening MELD-Na. Second, tumor cells with metastatic potential may express and secrete less soluble EGFR, as reported in HCC cell lines,²¹ so that a larger or more aggressive tumor contributes proportionally less circulating ectodomain. Third, the ligand-binding capacity of soluble ectodomains may be consumed in states of high EGF-family ligand production associated with inflammation and regeneration. Distinguishing these possibilities will require paired serum and tissue sampling with quantification of membrane EGFR and metalloprotease activity, in vitro measurement of ectodomain shedding in hepatocyte and HCC cell lines under conditions that model hepatic injury, and in vivo models in which tumor burden and liver function can be varied independently. Such experiments were beyond the scope of a clinical cohort study and would need to follow it.

The adjusted analysis helps define the role of sEGFR rather than weakening the unadjusted survival result. The survival difference between the ≤13 ng/mL and >13 ng/mL groups was evident when OS was calculated from baseline serum sample collection, with median OS of 5.42 and 12.06 months, respectively. The univariable Cox model also showed increased mortality risk in the low-sEGFR group. After adjustment, however, continuous sEGFR was not associated with OS. The combined-model analysis shows what this means in practice. Continuous sEGFR added nothing to the discrimination of a two-variable model containing AFP and MELD-Na. The dichotomized variable retained a nominal association in that parsimonious model, but the change in optimism-corrected C-index was 0.003, which has no clinical meaning, and in the full prespecified model the association was attenuated to a non-significant trend (HR 1.59, p=0.079). The prognostic information carried by sEGFR therefore overlaps substantially with that carried by tumor burden and liver reserve. In practical terms, sEGFR is more informative as a phenotype marker than as an isolated prognostic test, and any future multimarker application would need to demonstrate incremental value that this cohort did not show.

The findings also should not be interpreted as evidence for EGFR-targeted treatment selection. EGFR signaling is biologically relevant in HCC, but clinical trials of EGFR-directed therapy in unselected advanced HCC have been disappointing. Cetuximab monotherapy produced limited clinical activity without objective responses in a Phase II HCC study, and the Phase III SEARCH trial showed that adding erlotinib to sorafenib did not improve OS.^24,25 These therapeutic results do not negate the biological relevance of EGFR signaling, but they argue against using serum sEGFR as a surrogate for actionable EGFR dependence. Whether sEGFR predicts benefit from systemic therapy is a separate question from whether it is prognostic, and the exploratory analysis in sorafenib-treated patients does not support a predictive role. The survival trend by sEGFR within treated patients mirrored the unadjusted prognostic association in the whole cohort, the sEGFR-by-sorafenib interaction was null, and no difference was seen among the patients sampled before treatment began. Two features of the cohort limit any stronger inference. Enrollment took place between November 2014 and May 2017, when sorafenib was the only approved systemic therapy for HCC and before the first regulatory approval of an immune checkpoint inhibitor for HCC in September 2017,²⁶ so no patient received immunotherapy and systemic treatment was essentially limited to sorafenib; and treatment allocation was determined clinically, without randomization, so any comparison within treated subgroups is confounded by the tumor-burden and liver-reserve variables with which sEGFR is associated. In addition, 18 of the 37 sorafenib-treated patients had their serum sample drawn on or after the day treatment started, which precludes a true baseline predictive design. The present study is best framed as a circulating biomarker analysis, not as a predictive biomarker study for anti-EGFR or immune therapy.

Several strengths support the internal validity of the analysis. The cohort was prospectively assembled, and serum samples were collected at baseline. The study included detailed clinical, tumor-related, and liver-reserve variables, allowing sEGFR to be evaluated in relation to both malignancy-associated and hepatic-reserve domains. Serum sEGFR was measured in the full cohort, and the analysis included correlation testing, groupwise comparisons, Kaplan-Meier survival analysis, univariable Cox regression, and multivariable Cox modeling. The survival time origin was aligned with the biomarker sampling time, the multivariable model was prespecified and internally validated by bootstrap, and the incremental-value question was tested directly rather than inferred. The final interpretation was aligned with the multivariable model rather than based solely on dichotomized survival curves.

The study has limitations. It was conducted at a single center and included 84 patients, limiting statistical power and increasing the risk of model instability; with 69 events the cohort could detect a hazard ratio of about 2.0 for a binary predictor and about 1.4 per standard deviation for a continuous predictor, so a smaller independent effect of sEGFR cannot be excluded, and no a priori sample size calculation was performed. The cohort included a large proportion of patients with advanced-stage disease and was predominantly HBV-related, which may limit generalizability to Western, NASH-associated, or early-stage surveillance populations; the stage-stratified analysis showed that the principal associations were carried by the advanced-stage majority. The 13 ng/mL threshold was derived from the cohort median and supported by a similar Youden-index cutoff, but it applies only to this cohort and requires external validation before clinical use. Intra- and inter-assay coefficients of variation for the ELISA were not recorded and are not available, which limits the reproducibility assessment that other laboratories can make. Follow-up from serum sampling did not exceed 46 months, so 5-year survival could not be estimated and long-term prognostic stratification remains untested. Serial sEGFR measurements were not available, so changes during treatment or progression could not be evaluated. Paired tumor tissue was not analyzed, preventing direct comparison between serum sEGFR and tissue EGFR expression. The ELISA-based measurement cannot define the exact molecular source of circulating EGFR immunoreactivity or distinguish among soluble isoforms generated by alternative transcripts, shedding, or other mechanisms. Treatment heterogeneity, including locoregional therapies such as transarterial chemoembolization and thermal ablation whose combinations themselves influence survival,²⁷ may also have influenced outcomes; the exploratory analysis within sorafenib-treated patients was small and non-randomized; and the small number of female patients limits interpretation of sex-based differences.

The study has two implications for current understanding. It adds a soluble receptor ectodomain to the small group of circulating markers whose levels fall, not rise, as HCC burden increases and hepatic function declines, and it shows that this decline is closely coupled to standard liver-reserve indices and to IGF-1. It also provides a direct and internally validated test of incremental value and finds none, which argues against adding sEGFR to prognostic panels on the strength of unadjusted survival curves. Future studies should test serum sEGFR in larger and externally validated HCC cohorts with broader etiologic and stage distributions. Paired serum and tissue analyses would help clarify whether circulating sEGFR relates to tumor EGFR expression, non-tumoral liver biology, systemic receptor processing, or a combination of these sources. Serial measurements could determine whether sEGFR changes with tumor progression, locoregional therapy, systemic treatment, or hepatic decompensation. If sEGFR is to be evaluated within a multimarker model in the future, the relevant question is whether it improves discrimination, calibration, or risk reclassification beyond AFP, tumor size, BCLC stage, MELD-Na, ALBI grade, and IGF-1 in an independent cohort, and that question should be answered before any panel incorporating sEGFR is proposed.

Conclusion

In this prospective cohort of patients with HCC, lower serum sEGFR was associated with adverse tumor-related and liver-related features, including larger tumor size, higher AFP, impaired liver-function parameters, worse MELD-Na and CTP scores, lower IGF-1, and distant metastasis. In unadjusted analysis, patients with serum sEGFR ≤13 ng/mL had shorter OS from baseline serum sample collection than those with higher levels. Continuous sEGFR was not associated with OS before or after adjustment; in multivariable analysis serum sEGFR was not an independent predictor of OS, the estimate remained close to the null under bootstrap resampling, and sEGFR added no discriminative value to a model containing AFP and MELD-Na. The two results are compatible. The unadjusted survival difference arises from the association of low sEGFR with tumor burden and reduced hepatic reserve, and the adjusted analysis shows that sEGFR carries no prognostic information beyond those variables. Serum sEGFR should therefore be regarded as a marker of disease phenotype without independent prognostic capacity. Its potential contribution to a multimarker model remains hypothetical, was not supported by the incremental-value analysis in this cohort, and would require demonstration in independent, larger, and stage-diverse populations with paired serum and tissue data.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data Sharing Statement

The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Ethics Approval and Informed Consent

The study was designed and conducted in accordance with the Declaration of Helsinki. Approval was granted by the Ethics and Institutional Research Committees of Hacettepe University Faculty of Medicine. Written informed consent was obtained from all participants before enrollment and blood sample collection.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.