New prognostic and predictive scoring systems for colorectal cancer liver metastases

Abstract

Colorectal cancer liver metastases (CRLM) remain a major therapeutic challenge, shaped by marked biological heterogeneity and unpredictable clinical outcomes. While classical Clinical Risk Scores (CRS) continue to inform surgical decision making, their static nature limits their relevance in an era defined by molecular profiling, dynamic biomarkers, and personalized therapy. Advances in genomics and liquid biopsy technologies have expanded the understanding of CRLM beyond traditional clinicopathological parameters, enabling real-time disease monitoring and more refined risk stratification. In parallel, dynamic scoring systems that incorporate evolving clinical and molecular data are redefining how risk is assessed and managed throughout the course of treatment. This review revisits classical CRS, explores emerging molecular and dynamic tools, and discusses their potential integration into clinical practice to improve patient selection, guide treatment timing, and refine the overall management of CRLM.

Highlights

• •CRS remain essential but fail to capture the biological complexity of CRLM.
• •Genomics, histological growth patterns, and ctDNA uncover liver-specific tumor heterogeneity beyond anatomical staging.
• •Integrating dynamic biomarkers enables adaptive risk stratification and precision management of liver metastases.

Introduction

Colorectal cancer (CRC) is a major global health problem, with 1.9 million new cases reported in 2020, and it is projected to reach 2.5 million by 2035.¹ Liver metastases develop in more than half of patients, which has a substantial impact on prognosis and therapeutic decision making.^2,3

A wide range of therapeutic strategies are available for patients with colorectal cancer liver metastases (CRLM). Hepatic resection continues to be the cornerstone of curative treatment, achieving 5- and 10-year overall survival (OS) rates of 42% and 25%, respectively, in comparison to the 5-year OS of 15% achieved with systemic therapy alone.⁴ Despite these outcomes, more than half of the patients undergoing surgery experience recurrence, most commonly within the liver.⁵ This underscores the necessity for refined stratification tools capable of more accurately predicting outcomes.

Clinical Risk Scores (CRS), widely employed in clinical practice, were initially developed to estimate the recurrence risk following hepatic resection. While helpful in guiding surgical decision making, they do not fully address other critical aspects, such as long-term prognosis or predicting response to systemic therapy. In recent years, several novel prognostic and predictive tools have emerged, providing opportunities for refined patient stratification. It is imperative that clinical, biological, pathological, and radiological parameters be integrated.6, 7, 8, 9, 10, 11

This review offers an updated and integrative perspective on prognostic assessment in CRLM, integrating classical CRS with emerging molecular, histopathological, and liquid biopsy-based tools. This approach aims to reflect the ongoing transition from static, CRS-driven models toward biologically informed and dynamically adaptable prognostic frameworks.

Current clinical risk scores for colorectal cancer liver metastases

Preoperative risk scores assessing suitability for surgical resection

During the 1980s and 1990s, CRS were formulated as prognostic tools to stratify patients with CRLM based on post-resection relapse probability. The initial scoring system emerged in 1996,¹² but the most widely recognized is the Fong score, introduced in 1999. This score represents the classical anatomical–clinical approach, relying exclusively on tumor stage, disease-free interval (<12 months), number of metastases (>1), diameter of the largest metastasis (>5 cm), and carcinoembryonic antigen (CEA) level >200 ng/ml. Patients exhibiting a score of 0 showed a 60% 5-year recurrence-free survival (RFS) rate, whereas those with a score of 5 had only a 14% survival rate.¹³ Subsequent models and meta-analyses identified similar prognostic factors reinforcing their prognostic relevance.^4,14, 15, 16, 17, 18, 19, 20, 21

Despite the widespread use of the Fong score, its ability to estimate long-term outcomes has been challenged by newer models that integrate tumor morphology and molecular biomarkers. The Tumor Burden Score (TBS) marked a shift toward a morphology-based paradigm by improving risk stratification through a quantitative representation of tumor morphology. This framework was further advanced by the Genetic and Morphology Evaluation (GAME) score, which marked a critical step toward biologically informed risk stratification, including KRAS status, alongside established clinical variables.^6,8,22

In line with this evolving paradigm, the score proposed by Villard et al.¹⁰ introduced a refined set of selection criteria that demonstrated clear survival discrimination between risk groups in both internal and external validation cohorts, despite still largely lacking molecular integration.¹⁰ Table 1 summarizes the most relevant prognostic scores.

Table 1.

Summary of principal clinical risk scores for colorectal liver metastases

Score	External cohort	Presurgery			Postsurgery	Other
		Primary tumor	Liver metastasis	Molecular profile
Nordlinger et al. (1996)12	—	Serosal invasion lymph Node + DFS < 24 months	Size ≥ 5 cm n ≥ 4	—	Margin involved <1 cm	Age > 60 years
Fong et al. (1999)13	—	Lymph node + CEA > 200 ng/ml DFS < 12 months	Size ≥ 5 cm n > 1	—	—	—
TBS (2018)6	Yes	—	TBS 3-8 (1 point) ≥9 (2 points)	—	—	—
GAME (2018)8	Yes	Lymph node + CEA ≥ 20 ng/ml	TBS 3-8 (1 point) ≥9 (2 points) Extrahepatic disease	KRAS	—	—
Villard et al. (2022)10	Yes	Lymph node + right-sided	Size ≥ 5 cm 1-2 (0 points) 3-9 (1 point) 10-40 (2 points) Non-resectability Extrahepatic disease	—	—	Age > 60 years

CEA, carcinoembryonic antigen; DFS, disease-free survival; GAME, Genetic and Morphology Evaluation; TBS, Tumor Burden Score.

While incremental refinements have been made, no existing preoperative score has yet achieved consistent reproducibility across independent datasets. These CRS remain fundamentally limited by their exclusive reliance on static anatomical and morphological variables and their purely prognostic nature, unable to capture tumor biology. Biological and dynamic biomarkers are needed to more accurately guide upfront surgical and systemic treatment decisions.

Post-operative risk scores evaluating recurrence risk and guiding complementary systemic therapies

Beyond preoperative assessment, post-operative pathology provides complementary biological insight into treatment response and recurrence risk. Parameters such as margin status, tumor regression grade (TRG), and features of the immune microenvironment refine prognostication and inform decisions regarding adjuvant therapy.

Historically, margin status dominated post-operative risk assessment, reflecting a predominantly technical paradigm. While negative margins (>1 mm, R0) remain associated with improved oncological outcomes,23, 24, 25, 26, 27 emerging data suggest that their prognostic relevance may vary according to tumor biology, including factors such as primary tumor characteristics and molecular profiling.28, 29, 30 In selected cases, vascular R1 resections may still achieve acceptable outcomes.³¹

With the introduction of neoadjuvant chemotherapy, histopathological response gained relevance as a surrogate marker of biological aggressiveness.³² The TRG system captures treatment-induced tumor regression and has consistently demonstrated strong prognostic value, with improved long-term outcomes observed in patients achieving major histological response.33, 34, 35, 36

More recently, attention has shifted toward the immune microenvironment. High infiltration of activated immune cells within liver metastases has been associated with improved tumor control and survival, whereas immunosuppressive components correlate with poorer prognosis.37, 38, 39, 40 Although immune-based scoring systems are well established in localized CRC, their clinical implementation in CRLM remains exploratory.41, 42, 43

Importantly, these post-operative features may help inform adjuvant treatment decisions in a setting where randomized trials have improved recurrence-related endpoints without consistently translating into an OS benefit after CRLM resection. In this context, integrating pathology-based risk factors with post-operative circulating tumor DNA (ctDNA) assessment may support a more biologically grounded selection of patients for adjuvant treatment, while sparing low-risk individuals from unnecessary toxicity.

Emerging biomarkers to refine prognostic stratification in colorectal cancer liver metastases

To address the limitations of traditional CRS, novel biomarkers such as molecular profiling, liquid biopsy, and histological growth patterns (HGPs) offer valuable tools for refining prognostic assessment. These markers provide complementary information on tumor biology, therapeutic sensitivity, and residual disease, enabling more accurate selection of candidates and ultimately contributing to improved oncological outcomes.

Tumor molecular profiling

Molecular biology plays a pivotal role in determining prognosis, predicting therapeutic response, and elucidating mechanisms of resistance in CRC. Advances in molecular profiling have highlighted the clinical impact of genetic alterations, guiding personalized treatment strategies. Consequently, the European Society for Medical Oncology (ESMO) and American Society of Clinical Oncology (ASCO) guidelines recommend assessing BRAF, RAS (KRAS, NRAS), and mismatch repair status before initiating first-line therapy.^3,44

Among these alterations, mutations in RAS are present in ∼50% of CRC cases,⁴⁵ with KRAS being the most studied. Since the GAME score first incorporated KRAS as a prognostic factor, multiple models have confirmed its negative impact on OS. It reduces median survival from 70 months in wild-type patients to 19-50.9 months in mutated cases.⁸ A recent meta-analysis reported an odds ratio of 1.49 for worse survival outcomes.⁴⁶

On the other hand, the BRAF V600E mutation is found in ∼10% of metastatic CRC cases but is significantly less frequent in resectable CRLM (1%-6.1%).⁴⁵ The underlying explanation is that the BRAF V600E mutation is strongly associated with aggressive tumor behavior, high recurrence rates, and significantly reduced survival.^47,48 Even in these cases, the benefit of surgery remains controversial; however, hepatic resection appears to improve OS compared with systemic therapy alone, albeit with limited impact.^49,50 A recent meta-analysis reported a 1.98-fold higher risk of death compared with wild-type tumor, with a 3-year OS of 54% versus 82.9% and higher recurrence rates, both hepatic and extrahepatic.⁵¹

Another key molecular subset of CRC includes tumors with mismatch repair-deficient or microsatellite instability (dMMR/MSI), which comprise 4%-5% of all CRC cases and correlate with better prognosis but poor response to chemotherapy.⁵² However, these patients show exceptional responses to immune checkpoint inhibitors (ICIs), with a 50% pathological complete response rate compared with 13% with chemotherapy.⁵³ Immunotherapy, either as monotherapy or in combination, has significantly improved OS compared with chemotherapy alone and is now integrated into clinical practice as first- or subsequent-line therapy.54, 55, 56 The clinical interpretation of dMMR/MSI status in the management of CRLM remains complex. Retrospective analyses have reported worse OS following hepatic resection in dMMR/MSI patients, raising concerns about the benefit of surgery in this subgroup.^57,58 This may be partially explained by the limited responsiveness of dMMR/MSI tumors to conventional chemotherapy. The potential role of integrating immunotherapy with curative-intent local treatments, including liver resection, remains to be fully investigated.

These biomarkers should be urgently considered in prognostic scoring systems to support clinical decision making, not only for refined prognosis but also to influence treatment selection, particularly in the decision between upfront resection and neoadjuvant chemotherapy. In fact, an interesting strategy to explore would be the development of molecularly tailored prognostic models, specifically designed for each distinct molecular profiling.

Histological growth patterns

HGPs represent an emerging prognostic factor in the field of prognostic factors for CRLM. Several distinct types have been described, although their characterization has yet to be widely adopted in routine clinical practice.

Among the identified HGPs, the desmoplastic or encapsulated growth pattern (DGP) and the replacement growth pattern (RGP) are particularly notable. The DGP is often associated with a better prognosis, characterized by a well-defined stromal response that may inhibit tumor spread. In contrast, the pure RGP, where tumor cells replace hepatocytes without any clear barrier, observed in 20%-25% of patients, is linked to a poorer prognosis.^59,60

Recent research suggests that DGP is associated with a pro-inflammatory immune phenotype, indicating a more immunologically active tumor microenvironment. This association has raised interest in exploring its potential as a predictive biomarker for immunotherapy response.⁶¹ The RGP is further connected with the overactivation of the WNT/β-catenin pathway and increased glycolytic activity, contributing to immune evasion and more aggressive tumor behavior.⁶² HGPs also reflect distinct vascularization mechanisms; DGP is angiogenesis-dependent and may benefit from anti-vascular endothelial growth factor therapies, while RGP relies on vessel co-option, infiltrating the liver parenchyma and utilizing pre-existing sinusoidal vasculature, features that are associated with resistance to antiangiogenic treatments.⁶³

Despite these insights, the molecular mechanisms underlying HGPs remain largely unknown. Approximately 50%-60% of liver metastases present a mixed HGP pattern, complicating their clinical interpretation. Since the latest consensus on HGP in liver metastases, significant efforts have been made to validate their prognostic importance in CRC management.^64,65 To integrate this complexity into clinical decision making, the Histopathological, Clinical, and Molecular (HICAM) score was developed. This model incorporates HGPs along with age, TNM (tumor–node–metastasis) stage, TBS, CEA levels ≥20 ng/ml, primary tumor resection, molecular status, and inflammatory parameters. The HICAM score demonstrated significant OS differences between risk groups, with a median OS of 105.0 months in low-risk patients versus 30.4 months in high-risk patients (P < 0.001). However, its clinical application is limited by small sample sizes and the need for external validation.¹¹

The clinical utility of HGPs is limited by their post-operative assessment, restricting their role in guiding initial treatment decisions. In this context, radiomics has emerged as a non-invasive alternative, applying quantitative analysis to medical imaging to capture features reflective of tumor biology. Despite being derived mainly from small cohorts, radiomic texture and homogeneity features have shown promising sensitivity for predicting HGPs in CRLM,^66,67 potentially enhancing preoperative risk stratification and supporting more personalized treatment strategies.

Liquid biopsy

Despite the value of tumor tissue-based molecular profiling, its static nature fails to capture real-time tumor evolution. This limitation has driven the rise of liquid biopsy, allowing dynamic monitoring of minimal residual disease (MRD) and treatment response through a minimally invasive technique.⁶⁸

Preoperative detection of ctDNA in patients with CRLM has been associated with an increased risk of recurrence and reduced survival outcomes.^69,70 A meta-analysis encompassing multiple studies reported a pooled hazard ratio (HR) of 1.98 [95% confidence interval (CI) 1.04-4.36] for RFS in patients with detectable ctDNA before surgery compared with those without detectable ctDNA.⁷¹ In a multicenter observational cohort study involving 212 patients, Kobayashi et al. found that among 40 patients with available preoperative blood samples, 32 (80%) had detectable ctDNA. Of these, 20 patients (63%) developed recurrence, whereas only 1 patient (13%) without detectable preoperative ctDNA experienced recurrence during the follow-up period.⁷² This suggests that preoperative ctDNA positivity is a significant predictor of recurrence risk. However, it is important to note that the prognostic value of preoperative ctDNA may be influenced by various factors, including tumor burden, treatment history, and the sensitivity of the ctDNA assay used.^73,74 Therefore, while preoperative ctDNA detection holds promise as a prognostic marker, further studies are needed to standardize its use and determine its utility in guiding neoadjuvant therapy decisions.

Post-operative ctDNA status serves as a robust indicator of MRD and is strongly predictive of recurrence. Patients with detectable ctDNA following curative-intent liver resection exhibit significantly higher recurrence rates and shorter OS.^71,74,75 For instance, a pooled analysis revealed that detectable post-operative ctDNA was associated with an HR of 3.12 (95% CI 2.27-4.28) for recurrence and an HR of 5.04 (95% CI 2.53-10.04) for death.⁷¹ In a prospective cohort study by Reinert et al., assessment of ctDNA status immediately after surgery of liver metastases stratified patients into high- and low-risk groups for recurrence. The study reported an HR of 7.6 (95% CI 3.0-19.7, P < 0.0001) for recurrence in patients with detectable post-operative ctDNA. Notably, the positive predictive value of ctDNA for early relapses was 100%, further underscoring its clinical utility.⁷⁵ These findings were consistent with the results of a study conducted by Tie et al., which evaluated 54 patients undergoing curative-intent resection for CRLM. Detectable ctDNA after surgery was associated with significantly worse RFS (HR 6.3, 95% CI 2.58-15.2, P < 0.001) and OS (HR 4.2, 95% CI 1.5-11.8). Importantly, detection of ctDNA at the end of treatment (surgery ± adjuvant chemotherapy) predicted a 5-year RFS of 0%, compared with 75.6% in patients with undetectable ctDNA (HR 14.9, 95% CI 4.94-44.7, P < 0.001).⁷⁶ Moreover, patients who cleared ctDNA during adjuvant chemotherapy had better outcomes, suggesting a potential role for ctDNA as an indicator of treatment efficacy, which could guide real-time clinical decisions and personalize post-operative management.⁷⁷

Taken together, these studies highlight the role of post-operative ctDNA as a powerful biomarker for early relapse and treatment guidance. Its incorporation into clinical practice could enable risk-adapted surveillance strategies and the selection of candidates for intensified perioperative treatment.

Toward a dynamic and biology-driven informed approach

Prognostic and predictive models play distinct yet complementary roles in the management of CRLM. Prognostic tools describe the expected natural history of the disease independently of treatment, whereas predictive models estimate the likelihood of benefit from specific therapeutic interventions. Integrating these dimensions with emerging dynamic biomarkers is essential to move beyond static risk assessment and toward more individualized clinical decision making (Figure 1).

Figure 1.

Proposedalgorithm for integration of prognostic and predictive tools along the therapeutic timeline in CRLM. ctDNA, circulating tumor DNA; HGP, histological growth pattern; TIL, tumor-infiltrating lymphocyte; TRG, tumor regression grade.

Despite the widespread use of perioperative chemotherapy in resectable CRLM, uniform systemic strategies have not resulted in consistent OS benefits, as shown in the European Organisation for Research and Treatment of Cancer (EORTC) 40983 trial and subsequent meta-analyses.^78,79 Classical CRS, developed as prognostic rather than predictive tools, were often extrapolated to guide treatment intensity, implicitly assuming that poorer prognosis would translate into greater benefit from systemic therapy. This assumption failed to account for biological heterogeneity and treatment sensitivity, likely contributing to the modest and inconsistent outcomes observed with standardized perioperative approaches.

A more biologically informed framework is therefore required. Tumor biology not only refines baseline risk but can also identify patients in whom effective systemic therapies may meaningfully alter disease behavior, thereby influencing both treatment strategy and surgical opportunity. This paradigm shift is exemplified by BRAF V600E-mutated CRC, historically associated with poor outcomes but now therapeutically actionable through biologically matched combinations of targeted therapy and chemotherapy, as demonstrated in the BREAKWATER trial.⁸⁰ Similarly, dMMR/MSI tumors show profound and durable sensitivity to ICIs, challenging conventional decisions regarding whether and when surgery should be pursued following sustained systemic disease control.^81,82

Beyond baseline molecular classification, dynamic biomarkers provide treatment-sensitive insight into tumor behavior. Longitudinal changes in CEA, ctDNA kinetics, and radiomic features can capture depth of response and residual disease ahead of conventional imaging. In particular, ctDNA clearance correlates with chemosensitivity, whereas persistent positivity identifies ongoing molecular disease and frequently anticipates clinical relapse. In this context, ctDNA serves a dual role across the disease course: as a dynamic marker of treatment response and as a post-operative indicator of molecular residual disease. Post-operative ctDNA assessment therefore represents a promising approach to refine adjuvant treatment decisions in a setting where no validated biomarkers currently guide treatment escalation, continuation, or omission after CRLM resection. However, clinical translation remains constrained by limited prospective predictive validation and the lack of standardized, decision-grade thresholds. The next step should therefore be the deliberate integration of these biomarkers into prospective clinical trials and multidisciplinary decision making to redefine current therapeutic algorithms. Such an approach may help improve patient selection for systemic therapy and surgery, ensuring appropriate timing aligned with tumor biology and treatment responsiveness, and gradually advancing precision oncology in CRLM.

Conclusion

Overall, the evidence reviewed supports a transition from static, anatomy-driven risk models toward a more integrated, biology-informed approach to CRLM. While clinical and pathological variables remain foundational, they are insufficient to capture disease dynamics or to guide treatment decisions across the entire disease course.

The incorporation of molecular and dynamic readouts offers an opportunity to move from retrospective risk estimation toward prospective, biology-aligned decision making, enabling more rational selection of patients for systemic therapy, local treatment, and surgery, as well as improved timing of these interventions. Importantly, the clinical impact of this shift will depend not on isolated biomarkers, but on their rigorous validation, standardization, and thoughtful integration into prospective trials and multidisciplinary practice. How this transition is achieved will ultimately define the next stage of precision oncology in CRLM.