Prognostic value of desmoplastic growth patterns in resected hepatocellular carcinoma

Abstract

The prognostic value of histopathological growth patterns is well established in liver metastases, but their significance in hepatocellular carcinoma (HCC) remains uncertain. This study examined the prognostic relevance of the desmoplastic HGP (dHGP) in HCC and explored the impact of the degree of desmoplasia using three assessment methods. We retrospectively analysed 99 patients resected for HCC at two Finnish centres; 74 had adequate tissue for evaluation. HGP was classified as uniform dHGP (44/74) or predominant dHGP (66/74). The degree of desmoplasia was visually estimated as the minimum, maximum, and average distance between tumour cells and adjacent liver parenchyma. Uniform dHGP was not associated with survival in multivariable analysis. Predominant dHGP independently predicted improved overall survival (HR 0.15, 95% CI 0.04–0.51, P = 0.003) and disease-specific survival (HR 0.16, 95% CI 0.04–0.73, P = 0.018). Stronger desmoplasia was associated with improved disease-specific survival across all evaluation methods in multivariable analyses. Uniform and predominant dHGP may be associated with more favourable long-term outcomes in resected HCC, although further investigation is needed, particularly given the small sample size in the predominant non-dHGP group. We present three methods to quantify desmoplasia, providing a basis for future clinical studies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-42320-6.

Introduction

Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related mortality worldwide, with cirrhosis as its primary risk factor^1,2. Globally, hepatitis B is the most common cause of cirrhosis, whereas hepatitis C and alcohol consumption are prominent in Western populations². Although 20–25% of HCC patients are eligible for resection, recurrence rates can reach up to 50% within two years³. Currently, no adjuvant therapies exist post-resection, and validated biomarkers to guide treatment selection are lacking^4,5.

Tumour invasion and stromal remodelling—driven by both tumour and host—play critical roles in disease progression^6,7. The stromal compartment influences tumour growth and behaviour through mechanisms such as desmoplasia (fibrotic response with fibroblasts and extracellular matrix), inflammation, immune response, and angiogenesis^7,8. As tumours progress, distinct histological growth patterns (HGP) emerge at the invasive margin, including desmoplastic HGP (dHGP), replacement HGP (rHGP) and pushing HGP (pHGP)^9,10. Clinically, the dichotomy between dHGP and non-dHGP has emerged as particularly relevant¹⁰. Non-dHGP tumours often exhibit aggressive features, while dHGP is associated with a favourable prognosis in various types of liver metastases^9,11–21. However, the prognostic significance of HGPs in HCC remains less well established.

Prior studies have suggested that tumours with a surrounding desmoplastic rim have a better prognosis in HCC^22–26. However, only one prior study has demonstrated an association between uniform dHGP (dHGP vs any amount of non-dHGP) and favourable survival outcomes in non-cirrhotic HCC, applying the HGP methodology internationally validated for liver metastases²⁷. To date, no studies have validated the prognostic role of predominant dHGP (≥ 50% dHGP vs < 50% dHGP) for HCC. One previous study has analysed the prognostic value of a stronger desmoplasia for resected HCC, finding no association with outcomes²⁵.

This study aimed to assess the association of uniform and predominant dHGP with survival in resected HCC. Additionally, we investigated the prognostic role of the degree of desmoplasia using three novel assessment methods.

Methods

Patient cohort

This retrospective study analysed 99 resection samples of histologically confirmed HCC cases diagnosed in Oulu University Hospital from 1986 to 2022 and Central Finland Central Hospital from 1997 to 2022. Approval of the study was obtained from the Finnish Medicines Agency Fimea (Dnro FIMEA/2021/004928) and the Ethical Committee of the North Ostrobothnia Hospital District (EETTMK 81/2008). The study adhered to the Declaration of Helsinki.

Data collection

The patients were identified from the pathology archives using the ICD-10 code C22.0. Relevant patient information was retrieved from the patient chart and pathology reports. Tumours were staged according to the 8th edition of the TNM classification system. Comorbidities were evaluated using the Charlson Comorbidity Index (CCI), and complications were classified using the Clavien–Dindo classification system. Date and cause of death were collected from Statistics Finland, with data collection extending through the end of 2022.

Hematoxylin and eosin (H&E)-stained diagnostic samples were digitised using Aperio AT2 (Leica Biosystems, Wetzlar, Germany) and viewed using Aperio ImageScope. The same slides used to diagnose the disease were analysed. Tumour grade was reassessed by a pathologist (J.P.V.) if it was not recorded in prior reports. HGP analysis was performed by a single investigator (N.S.), following training and calibration with an experienced pathologist (J.P.V.). Uncertain cases were reviewed jointly with J.P.V. to support consistent application of the criteria.

Evaluation of HGP

We defined dHGP as a rim of fibrous tissue separating tumour cells from the surrounding liver (Fig. 1A). rHGP was characterised by cancer cell plates in direct contact with liver cell plates, utilising sinusoidal vessels to replace hepatocytes (Fig. 1B). In the pHGP, liver cell plates surrounding the tumour cells are pushed away and compressed. pHGP also lacks the desmoplastic rim. HGP classification was conducted on all available H&E-stained samples, blinded to patient outcomes. All sections with tumour–liver interface, regardless of length, contributed to the estimation. The tumour–liver interface was defined as the boundary zone where malignant hepatocytes are adjacent to the surrounding non-tumorous liver parenchyma. Areas near portal tracts or in contact with the desmoplastic rim were excluded. Cases with < 20% of evaluable tumour–liver interface were excluded (n = 25), resulting in a final cohort of 74. In cases showing multiple HGPs, the average proportion of each HGP was estimated. For downstream analyses, patients were categorised as having a uniform dHGP or any presence of non-dHGP (rHGP and pHGP), in line with updated international consensus guidelines for liver metastases²⁷. HGP was also categorised by the predominant HGP, defined as ≥ 50% dHGP vs < 50% dHGP, as recommended in previous guidelines.

Fig. 1.

Examples of histopathological growth patterns (HGPs) and desmoplastic rim measurements in resected HCC. Representative hematoxylin and eosin–stained sections from whole-slide images scanned at 20 × objective magnification. Images shown are cropped regions of interest. (A) Uniform desmoplastic HGP with the assessable invasive front indicated by a black line. (B) Non-desmoplastic HGP. (C–E) Illustration of the minimum (C), maximum (D), and average (E) rim-thickness measurements in uniformly desmoplastic tumours. The boundary of normal liver parenchyma is shown with a dotted line. T, tumour; N, normal liver.

Evaluation of the degree of desmoplasia

Among cases with a uniform dHGP (n = 44), the thickness of the desmoplastic rim was visually estimated by measuring the perpendicular distance from tumour cells to surrounding liver parenchyma (Fig. 1). Measurements included the minimum, maximum, and average thickness. The minimum and maximum represented the smallest and largest values observed across all assessable samples, while the average was calculated as the mean value across all samples. The median number of slides assessed per case was 2 (IQR, 1–3). The uniform dHGP samples were divided into low and high groups, based on median values for each assessment method. The cut-off points for minimum thickness, maximum thickness, and average thickness were 94 μm, 1302 μm, and 516 μm, respectively. The low and high desmoplasia groups were compared with non-dHGP samples, creating three groups for each assessment method.

Outcomes

The primary outcomes were 5-year overall survival and disease-specific survival. The survival outcome was truncated at 5 years. Overall survival was defined as the time from surgical resection to death from any cause or the last follow-up. Disease-specific survival was defined as the time from surgical resection to death specifically due to HCC, with other causes considered censored events.

Statistical analysis

Baseline characteristics were presented as proportions, means with standard deviations (SD), or medians with interquartile ranges (IQR), as appropriate. Categorical variables were compared using the chi-square test. Continuous variables with a normal distribution were analysed using Student’s t-test, whereas the Mann–Whitney U-test was applied to non-normally distributed continuous variables. For comparisons involving three or more groups, one-way ANOVA was used for normally distributed variables, and the Kruskal–Wallis test was applied for non-normally distributed variables. Overall and disease-specific survival were evaluated using the Kaplan–Meier method and compared with the log-rank test. Statistical significance was defined as P < 0.05.

Univariable analysis was conducted to assess long-term survival. Multivariable Cox regression analysis was performed, adjusting for the following covariates: age (continuous), gender (male/female), TNM stage (1 vs ≥ 2), and tumour grade (1–2 vs 3). Proportional hazards assumptions were evaluated prior to model interpretation. For uniform dHGP, assumptions were supported by parallel log–minus–log survival curves. For predominant dHGP, visual assessment was limited by the small size of the non-dHGP reference group; therefore, assumptions were formally assessed using time-dependent covariates, with no evidence of violation. To assess the robustness of the hazard ratio given the small size of the non-predominant dHGP reference group, we performed a leave-one-out influence analysis restricted to this group. The Cox proportional hazards model was refitted repeatedly, each time excluding one non-predominant dHGP patient (n = 8), using the same model specification, covariate adjustment, and censoring as in the primary analysis. Robustness was evaluated by summarising the range (minimum–maximum) of hazard ratios across the refitted models. In addition, a subgroup analysis was performed among patients with predominant dHGP, excluding cases with R1 resection to account for its potential influence on survival differences between groups. Statistical analyses were conducted using IBM SPSS Statistics version 29.0 (IBM Corp., Armonk, NY).

Results

A total of 99 resection samples were analysed. The patient selection process and derivation of the final study cohort are summarised in the flow chart (Fig. S1). In 25 cases, there was insufficient TLI to assess the HGP, resulting in 74 samples in the final dataset. Baseline characteristics of included and excluded patients are presented in the supplementary material, demonstrating a significantly higher prevalence of multifocal tumours and more advanced TNM and BCLC stages among excluded patients (Table S1). The median follow-up time was 4.2 years (IQR, 1.5–5.2). The median age was 67.5 (IQR, 64.0–74.1), and the median CCI was 1.3. Most patients were male (62.2%) with well- or moderately-differentiated tumours (87.8%). Six patients (8.7%) had an R1 resection. No patients in this cohort received neoadjuvant therapy, whereas two patients had received locoregional therapy prior to surgery. The 5-year overall survival rate for the entire cohort was 54.1%, while the disease-specific survival rate was 69.5% (Fig. 2). The multivariable model included four covariates, yielding an events‑per‑variable ratio of approximately 6.25 based on the 25 deaths observed in the cohort.

Fig. 2.

Survival outcomes according to desmoplastic growth patterns. (A–B) Overall and disease-specific survival in patients with uniform desmoplastic HGP. (C–D) Overall and disease-specific survival according to predominant desmoplastic HGP.

Uniform dHGP

Uniform dHGP was observed in 44 patients (59.5%). The non-dHGP mainly represented rHGP, and a predominant pHGP was not observed in any case of this cohort. The presence of any amount of non-dHGP was associated with having a positive resection margin (Table S2). Among patients with uniform dHGP, 5-year overall survival and disease-specific survival rates were 60.5% and 77.8%, respectively (Fig. 2). In patients with any amount of non-dHGP, the 5-year overall survival and disease-specific survival rates were 43.3% and 56.3%, respectively (Fig. 2). Having a uniform dHGP was not significantly linked with improved 5-year overall survival (HR 0.51. 95% CI 0.23–1.13) in univariable analysis, while it was a predictor of improved disease-specific survival (HR 0.35. 95% CI 0.13–0.91, P = 0.032) (Table 2). However, after adjusting for age, sex, TNM stage and grade, a uniform dHGP was significantly associated with better 5-year overall (HR 0.42. 95% CI 0.19–0.97, P = 0.042) and disease-specific survival (HR 0.31. 95% CI 0.11–0.86, P = 0.025) (Table 2).

Table 2.

Hazard ratios (HR) with 95% confidence intervals (CI) of mortality comparing resected HCC patients according to the uniform and predominant histopathological growth pattern (HGP).

	Predominantly non-dHGP (N = 8) HR (95% CI)	Predominantly dHGP (N = 66) HR (95% CI)	P value	Any amount of non dHGP (N = 30)	Uniform dHGP (N = 44)	P value
5-year overall mortality
Crude	1 (reference)	0.24 (0.08–0.67)	0.007	1 (reference)	0.51 (0.23–1.13)	0.098
Adjusteda	1 (reference)	0.22 (0.07–0.67)	0.007	1 (reference)	0.42 (0.19–0.97)	0.042
5-year disease-specific mortality
Crude	1 (reference)	0.24 (0.07–0.75)	0.015	1 (reference)	0.35 (0.13–0.91)	0.032
Adjusteda	1 (reference)	0.25 (0.07–0.91)	0.036	1 (reference)	0.31 (0.11–0.86)	0.025

^aAdjusted for age (continuous), sex (female, male), TNM stage (1, 2 or higher) and tumour grade (1–2, 3).

Significant values are bold.

Predominant dHGP.

A total of 66 patients (89.2%) were classified as having a predominant dHGP, while 8 (10.8%) had a predominant non-dHGP. In the predominant dHGP group, 44 patients (66.6%) were categorised as having a uniform dHGP. A predominant non-dHGP was associated with female sex (P = 0.022) and R1-resection (P = 0.002) (Table 1). Among patients with a predominant dHGP, the 5-year overall survival and disease-specific survival rates were 58.2% and 73.3%, respectively (Fig. 2). Patients with a predominant non-dHGP had generally worse survival, but follow-up data were insufficient to estimate 5-year survival rates (Fig. 2). The predominant dHGP was associated with better 5-year overall (HR 0.24. 95% CI 0.08–0.67, P = 0.007) and disease-specific survival (HR 0.24. 95% CI 0.07–0.75, P = 0.015) in univariable analysis (Table 2). The predominant dHGP remained independently associated with overall survival (HR 0.22. 95% CI 0.07–0.67, P = 0.007) and disease-specific survival (HR 0.25. 95% CI 0.07–0.91, P = 0.036) in multivariable analysis (Table 2). Across the leave-one-out refits of the non-predominant dHGP group, the estimated 5-year overall survival hazard ratios ranged from 0.18 to 0.29, while disease-specific hazard ratios ranged from 0.19 to 0.30. This limited variation indicates low sensitivity of the results to any individual reference patient. After exclusion of R1 resections, neither univariable nor multivariable analyses demonstrated a statistically significant survival difference according to predominant HGP (Table S3). A complete univariable and multivariable Cox regression results for all covariates included in the survival models are displayed in the supplementary material (Table S4-S5).

Table 1.

Baseline characteristics for resected HCC stratified by the predominant histopathological growth pattern (HGP).

		Predominant HGP
		dHGP, N = 66	Non-dHGP, N = 8	P value
Age, years. Median (IQR)		69.7 (65.0–74.0)	66.4 (52.0–75.3)	0.389
Male. N (%)		44 (66.7%)	2 (25.0%)	0.022
Charlson Comorbidity Index. Mean (SD)		1.3 (0.8)	1.3 (1.0)	0.896
Cirrhosis. N (%)		31 (47.0%)	2 (25.0%)	0.241
Child- Pugh Score. N (%)	A	51 (82.3%)	7 (87.5%)	0.713
	B	11 (17.7%)	1 (12.5%)
	C	0 (0%)	0 (0%)
Alpha-fetoprotein, μg/L. Median (IQR)		5 (3–19)	1442 (3–3209)	0.467
Tumour size, mm. Median (IQR)		45 (30–60)	68 (38–100)	0.136
Unifocal tumour. N (%)		53 (80.3%)	7 (87.5%)	0.626
BCLC Stage. N (%)	Very early (0)	2 (3.0%)	0 (0%)	0.924
	Early (A)	55 (83.3%)	7 (87.5%)
	Intermediate (B)	8 (12.1%)	1 (12.5%)
	Advanced (C)	1 (1.5%)	0 (0%)
	Terminal (D)	0 (0%)	0 (0%)
TNM stage. N (%)	Stage 1	45 (68.2%)	6 (75.0%)	0.636
	Stage 2	17 (25.8%)	2 (25.0%)
	Stage 3	3 (4.5%)	0 (0%)
	Stage 4	1 (1.5%)	0 (0%)
Tumour grade. N (%)	Grade 1	23 (34.8%)	3 (37.5%)	0.915
	Grade 2	35 (53.0%)	4 (50.0%)
	Grade 3	8 (12.1%)	1 (12.5%)
Positive resection margin. N (%)		3 (4.9%)	3 (37.5%)	0.002

Significant values are in bold.

Degree of desmoplasia

Evaluating the degree of desmoplasia according to minimum, maximum and average desmoplastic rim thickness showed a strong correlation (r_s = 0.52–0.81, P < 0.001). Stronger desmoplasia was associated with improved 5-year disease-specific survival in univariable analyses across all three metrics (Table S6) and remained statistically significant in multivariable analysis (Table S6). Baseline characteristics and longitudinal survival rates are provided in the supplementary material (Tables S7–S9 and Fig. S2).

Discussion

Our findings suggest that both predominant and uniform dHGP at the tumour–liver interface may be associated with improved prognosis in resected HCC. Predominant dHGP showed the strongest association, but the predominantly non-dHGP reference group comprised only 11% of the cohort, increasing susceptibility to random variation and instability of effect estimates. In addition, the substantial overlap between the uniform and predominant definitions limits inference about their distinct prognostic significance. These findings should therefore be viewed as hypothesis-generating and require validation in larger, contemporary cohorts.

One prior study has evaluated dHGP in non-cirrhotic HCC¹⁰, reporting an association between uniform dHGP and better long-term outcomes¹⁰. Our findings extend these observations to a cohort that also included cirrhotic HCCs, with a particularly consistent association observed for disease-specific survival. The previous study observed a higher proportion of non-dHGP (68%) cases, compared to our cohort (41%)¹⁰. The role of HGP has been studied more extensively in colorectal cancer liver metastases, demonstrating a link between uniform dHGP and favourable long-term outcomes^11,16,17. Interestingly, the proportion of dHGP cases has been smaller (23–25%) in these studies, potentially reflecting the aggressive nature of metastatic tumours^11,16,17. Colorectal cancer often exhibits more infiltrative tumour growth and distinct pathophysiological features. Furthermore, uniform dHGP has been associated with lower recurrence rates following local treatment for colorectal cancer liver metastases, and it has predicted favourable outcomes for chemo-naïve colorectal cancer metastases, showing potential for guiding clinical decision-making^16,17.

To our knowledge, this is the first study to suggest that predominant dHGP is a favourable predictor of survival in resected HCC. However, only 11% of the cohort were categorised as predominantly non-dHGP, making the results more susceptible to random variation. However, leave-one-out influence analyses demonstrated that the magnitude and direction of the survival association remained stable across refitted models, indicating low sensitivity to individual reference patients. The observed survival differences may partly reflect the higher proportion of R1 resections in the predominant non-dHGP group, as subgroup analysis excluding these patients did not show a statistically significant difference in survival outcomes. The association between a predominant dHGP and improved prognosis also seems consistent for colorectal cancer liver metastases^9,18–21. In HCC, macroscopic encapsulation has previously been linked with better survival, both when identified on gross pathology inspection and as complete encapsulation on MRI^22,28. A visible desmoplastic rim has also been associated with fewer microsatellite nodules and reduced microvascular invasion²⁵. In parallel, the Liver Cancer Study Group of Japan has developed guidelines to standardise histological criteria based on tumour border and architecture²⁶. While these approaches may capture overlapping aspects of tumour biology, the systematic assessment of HGP appears to offer a reproducible framework for prognostic stratification in HCC, in addition to liver metastases¹⁰.

The prognostic role of stronger desmoplasia has been investigated in one previous HCC study, which found no association with survival outcomes²⁵. In colorectal liver metastasis, one study has evaluated the prognostic significance of the degree of desmoplasia and reported improved outcomes with stronger desmoplasia²⁹. In clear cell renal cell carcinoma, Xi et al. demonstrated a linear relationship between the thickness of the desmoplastic rim and prognosis³⁰. However, methodological differences between these studies are substantial^25,29,30. Our study introduces clearly defined measurement criteria for assessing the degree of desmoplasia, which are readily applicable to future research investigating its prognostic relevance.

Tumours with an rHGP, which constitute the vast majority of the non-dHGP group, may utilise vascularisation from the adjacent liver through co-option, without relying on vascular endothelial growth factor (VEGF)³¹. This mechanism enables efficient vascularisation with minimal hypoxia¹¹. In colorectal cancer liver metastases, rHGP has also been linked with adverse histopathological features of the primary tumour, including tumour budding, attributed to upregulated signalling pathways driving cell motility^32,33. In contrast, dHGP tumours have been linked with lymphocyte aggregates associated with anti-tumour immunity^34,35. Female sex has also been linked with well-encapsulated tumours and overall better prognosis, possibly related to oestrogen receptor–positive HCCs more frequently demonstrating complete tumour encapsulation and improved survival outcomes³⁶. However, the opposite association was found in our cohort, as a predominant HGP was linked with male sex.

The extended accrual period represents an important limitation, as clinical and pathological documentation was incomplete in earlier years, restricting the availability of several covariates. In particular, vascular invasion was not routinely reported and could not be reliably assessed, precluding adjustment for this key prognostic factor and raising the possibility of residual confounding. The aetiology of underlying liver disease was also incompletely documented. In addition, evolving clinical practice over time, including higher resection rates in recent years and more invasive treatments in patients with greater morbidity, may have influenced the results, while sensitivity analyses restricted to more recently operated patients were not statistically feasible due to the limited sample size. The retrospective design and relatively small cohort further constrain statistical power, reflecting the low incidence of hepatocellular carcinoma in Finland and its frequent diagnosis at advanced stages that preclude curative resection. Moreover, histopathological growth pattern assessment was performed by a single observer without formal interobserver agreement testing, and separate cohort analyses for Central Finland Central Hospital were not feasible because of the small number of cases. Finally, the small predominant non-dHGP reference group and overlap between uniform and predominant HGP definitions limit the stability and interpretability of the survival estimates.

In conclusion, this study suggests that the tumour–liver interface growth pattern evaluation methods established for liver metastases may also be applied to HCC. Further research is needed to clarify the clinical relevance of these results. Additionally, we introduced three approaches for measuring the degree of desmoplasia, which may serve as a foundation for future studies exploring its clinical relevance.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the study’s conception and design. Material preparation and data collection from patient charts were performed by Niklas Sarelin, Valtteri Kairaluoma and Olli Helminen. The methodology was designed by Niklas Sarelin, Olli Helminen and Juha P. Väyrynen. Juha P. Väyrynen and Jan Böhm assisted in picking out, retrieving and digitising the histological samples. Niklas Sarelin performed the histological analysis. Valtteri Kairaluoma assisted Niklas Sarelin in statistical analysis. All authors reviewed, validated and approved the final manuscript. The work was supervised by Juha P. Väyrynen and Olli Helminen, and administered by Niklas Sarelin.

Funding

Financial support was provided to the corresponding author by the Mary and Georg C. Ehrnrooth Foundation, Vaasan Lääkäriyhdistys rf, and The Swedish Cultural Foundation in Finland.

Data availability

The research data analysed during the current study are available from the corresponding author upon reasonable request. Sharing the data will require additional ethical approval.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

The Finnish Medicines Agency Fimea (Dnro FIMEA/2021/004928) and the Ethical Committee of the North Ostrobothnia Hospital District (EETTMK 81/2008) approved the study. The study was performed in accordance with the Declaration of Helsinki. The need to obtain informed consent from the study participants was waived by the Finnish Medicines Agency (FIMEA/2021/004928).