Multicenter nomogram integrating clinicopathological features for prognosis in stage II-III deficient/microsatellite instability-high rectal cancer

Abstract

BACKGROUND

Lymphovascular invasion (LVI) and perineural invasion (PNI) are commonly used prognostic markers in colorectal cancer. Their clinical relevance in mismatch repair-deficient/microsatellite instability-high (dMMR/MSI-H) rectal adenocarcinoma, however, has not been well defined. This study aimed to evaluate the prognostic impact of LVI and PNI in stage II-III dMMR/MSI-H rectal cancer patients.

AIM

To assess the prognostic value of LVI and PNI in stage II-III dMMR/MSI-H rectal adenocarcinoma and to develop a preliminary nomogram to predict recurrence risk.

METHODS

We retrospectively analyzed 107 treatment-naive patients with stage II-III dMMR/MSI-H rectal adenocarcinoma who underwent radical resection. LVI and PNI were assessed histopathologically, and their associations with clinicopathological features and recurrence-free survival (RFS) were analyzed using Cox regression and random forest modeling. A prognostic nomogram integrating tumor-node-metastasis stage, LVI, PNI, and body mass index was developed and validated.

RESULTS

LVI and PNI were identified in 21.5% (23/107) and 13.1% (14/107) of cases, respectively. In multivariable analysis, LVI remained independently associated with worse RFS (hazard ratio = 3.31, 95% confidence interval: 1.23-8.91, P = 0.018), while PNI showed a non-significant trend (hazard ratio = 1.75, P = 0.305). The nomogram showed favorable apparent discrimination for 1-, 3-, and 5-year RFS (area under the curves: 0.814, 0.871, and 0.862, respectively).

CONCLUSION

LVI is an independent prognostic factor in dMMR/MSI-H rectal cancer, while PNI is not. The proposed nomogram requires validation in larger prospective cohorts before clinical application.

Key Words: Mismatch repair; Microsatellite instability; Rectal cancer; Lymphovascular invasion; Perineural invasion; Nomogram; Recurrence-free survival

Core Tip: Deficient/microsatellite instability-high rectal cancer has distinct biological features, but postoperative risk assessment is still largely based on tumor-node-metastasis stage. This multicenter exploratory cohort highlights lymphovascular invasion as an independent predictor of recurrence-free survival in stage II-III deficient/microsatellite instability-high rectal adenocarcinoma, while perineural invasion showed no independent prognostic value. A preliminary nomogram integrating tumor stage, lymphovascular invasion, perineural invasion, and body mass index offers a practical framework for recurrence risk stratification, but requires external validation before clinical use.

INTRODUCTION

Colorectal cancer is the third most common malignant tumor in the world and the second most common cause of cancer-related death[1]. As a distinct anatomical subset of colorectal cancer, rectal cancer has exhibited a persistent upward trend in global incidence rates, with a particularly pronounced increase observed in younger populations (under 50 years of age) over the past decade[2]. Deficient mismatch repair or microsatellite instability-high tumors, characterized by unique molecular signatures and immunogenic phenotypes, constitute a clinically relevant subgroup accounting for 10%-15% of rectal adenocarcinoma cases[3].

Immune checkpoint inhibitors have changed the treatment landscape for deficient/microsatellite instability-high (dMMR/MSI-H) rectal cancer, with landmark clinical trials demonstrating exceptional efficacy[4]. Notably, the NICHE-2 study revealed that neoadjuvant dual immune checkpoint blockade (nivolumab plus ipilimumab) has produced high clinical complete response rates in early studies[5]. However, clinical implementation of immunotherapy faces non-negligible challenges, as approximately 10%-15% ultimately discontinue treatment due to immune-related adverse events, with grade 3-4 toxicities occurring in 5%-7% of cases[6]. Some patients discontinue treatment because of immune-related adverse events, severe toxicity, or comorbid conditions that limit its use. For this subset of patients, radical surgical resection remains a critical salvage therapeutic strategy to achieve disease control.

Current clinical guidelines predominantly utilize postoperative tumor-node-metastasis (TNM) staging to direct adjuvant therapy, yet this framework fails to incorporate critical prognostic indicators like lymphovascular invasion (LVI) and perineural invasion (PNI), causing notable outcome variability within matched-stage groups[7-9]. Most evidence derives from microsatellite stable (MSS) or molecularly unstratified populations. In rectal cancer patients undergoing neoadjuvant chemoradiotherapy followed by total mesorectal excision, the presence of LVI correlated with an increased incidence of distant metastasis, subsequently contributing to compromised overall survival outcomes[10]. Per The National Comprehensive Cancer Network guidelines, LVI/PNI positivity in stage II rectal cancer constitutes a category 1 recommendation for adjuvant chemotherapy, demonstrating a 5%-8% improvement in 5-year disease-free survival[11]. In stage III patients, the co-occurrence of LVI and PNI is significantly associated with elevated regional lymph node metastasis burden and occult micrometastatic foci, resulting in a marked deterioration of 5-year disease-free survival[12,13]. However, these observations are largely derived from unselected or MSS-dominant populations. In locally advanced (II-III stages) dMMR/MSI-H rectal adenocarcinoma, a systematic prognostic evaluation system has not been fully established, and there is currently no clear pathological stratification indicator that can further guide individualized treatment strategies. dMMR/MSI-H tumors exhibit distinct biological features, including high tumor mutational burden, abundant immune cell infiltration, and unique immune microenvironment characteristics. These features may influence the prognostic significance of traditional pathological markers like LVI and PNI, as the immune-mediated control of tumor dissemination could alter the clinical implications of vascular and PNI. Body mass index (BMI) was included as a potential prognostic factor based on emerging evidence suggesting that obesity and nutritional status may influence immune function and treatment outcomes in cancer patients, though its specific role in dMMR/MSI-H rectal cancer remains to be established. In addition, carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) are tumor markers commonly used in colorectal cancer management. Elevated levels of these markers have been associated with advanced disease stage and worse prognosis, making them important prognostic indicators in clinical practice. Therefore, we aimed to systematically evaluate the impact of LVI, PNI, BMI, CEA, CA19-9 on recurrence-free survival (RFS) and develop a prognostic stratification model in patients with stage II-III dMMR/MSI-H rectal cancer.

MATERIALS AND METHODS

Patients

This multicenter retrospective study included 107 treatment-naive patients with stage II-III rectal adenocarcinoma who underwent radical resection from 2013 to 2023 across five tertiary institutions: Sun Yat-sen University Cancer Center, The Sixth Affiliated Hospital of Sun Yat-sen University, Guangdong Provincial People’s Hospital, Nanfang Hospital, and The Fifth Affiliated Hospital of Sun Yat-sen University.

The inclusion criteria were as follows: (1) Underwent radical resection for rectal adenocarcinoma; (2) Postoperative pathological staging confirmed as stage II or III; (3) Definitive dMMR/MSI-H status confirmed by immunohistochemistry or polymerase chain reaction analysis; (4) Complete pathological reports documenting LVI and PNI; and (5) Minimum follow-up duration of 12 months. The exclusion criteria included: (1) History of current malignancy; (2) Lynch syndrome; (3) Receipt of preoperative neoadjuvant therapy; (4) Posttreatment survival time of less than 1 month; and (5) Incomplete follow-up data.

Histopathology

Pathological assessment was performed independently at each participating institution rather than through centralized review. To minimize diagnostic variability, all tumor specimens were reviewed in a blinded manner by two board-certified pathologists at each participating institution, based on the 8th edition of the American Joint Committee on Cancer staging system[14]. Discordant cases were resolved by joint review and consensus discussion when necessary. LVI and PNI were evaluated via hematoxylin and eosin staining using morphological criteria under conventional light microscopy. LVI was defined as the unequivocal presence of tumor cell clusters within endothelial-lined lymphatic or vascular channels. Of note, the diagnosis was based on hematoxylin and eosin morphology and did not require ancillary immunohistochemical confirmation[15]. PNI was defined as histologically confirmed tumor cell infiltration into the perineural space, circumferential encasement of nerves, or penetration through the nerve sheath[16]. The loss of expression in any of the four key mismatch repair proteins (MutL homolog 1, MutS homolog 2, MutS homolog 6, and PMS1 homolog 2) is primarily assessed via immunohistochemistry. Tumors lacking expression in ≥ 1 of these proteins are classified as deficient mismatch repair, whereas intact expression across all four proteins defines proficient mismatch repair[17].

Concurrently, microsatellite instability status was evaluated by analyzing five microsatellite loci (BAT-25, BAT-26, D2S123, D5S346, and D17S250) recommended by the National Cancer Institute using polymerase chain reaction-based methods. Instability in 2 loci confirms a diagnosis of microsatellite instability-high, while < 2 unstable loci are categorized as microsatellite instability-low or MSS, depending on the number of affected sites[18].

Follow-up and endpoints

Patients received structured clinical surveillance through outpatient visits every 3 months to 6 months or telemedicine consultations. Systematic monitoring included digital rectal examination, chest and abdominal computed tomography, serum tumor marker assessments (CEA, CA19-9), colonoscopy screening and documentation of recurrence patterns, metastatic progression, and mortality events. The primary endpoints were RFS (time from surgery to recurrence/metastasis), with events adjudicated by an independent oncology review panel. The last follow-up was conducted in November 2024.

Statistical analysis

Statistical analyses were conducted using R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria. http://www.r-project.org) with the survival and survminer packages. Categorical variables were compared using Pearson’s χ² test or Fisher’s exact test for low-frequency categories. Survival outcomes were analyzed via the Kaplan-Meier method, complemented by log-rank tests to statistically compare survival distributions between groups. Univariable Cox regression analyses were performed to identify prognostic indicators of RFS and calculate hazard ratios (HRs) with 95% confidence intervals (CIs). Variables with P < 0.05 in univariable analysis were included in the multivariable Cox regression model using backward stepwise selection with a significance level of P < 0.05 for retention. The random forest method was also used to select variables with clinical significance. Statistical significance was set at a two-tailed P < 0.05. A nomogram integrating clinicopathological covariates was developed to predict RFS probabilities at 1-, 3-, and 5-year intervals. Given the limited sample size, the cohort was not split into training and validation sets. Instead, bootstrap internal validation with 500 resamples was employed, as this method provides a more stable and reliable estimate of model performance and corrects for optimism (overfitting) in situations where data is limited compared to a single random split.

Time-dependent receiver operating characteristic curves were used to assess the model’s ability to distinguish outcomes at different time points. Decision curve analysis was used to evaluate clinical benefit. The area under the curve was used to measure the model’s discriminative performance. A two-sided P value < 0.05 was considered statistically significant.

RESULTS

Clinicopathological characteristics

A total of 107 patients with pathologically confirmed mismatch repair-dMMR/MSI-H rectal adenocarcinoma were included. The cohort was predominantly male (65.4%, 70/107), and the median age was 53.82 years (range: 20.25-90.25). Postoperative histopathological staging classified the cohort as stage II (68.2%, 73/107) and stage III (31.8%, 34/107). Pathological assessments identified LVI in 21.5% (23/107) and PNI in 13.4% (14/107) of cases. Preoperative serum tumor marker analysis revealed elevated CEA (> 5 ng/mL) in 16.8% (18/107) and CA19-9 (> 37 U/mL) in 26.2% (28/107). During a median follow-up of 54 months (range: 4-146), disease recurrence/metastasis events were identified in 21 patients (19.6%) (Table 1).

Table 1 Baseline clinicopathological characteristics of the patients.

Patient characteristics		Number	Percentage
Age (years)	mean ± SD	53.82 ± 12.42
Age (years)	≥ 53	42	39.3%
	< 53	65	60.7%
Sex	Male	70	65.4%
	Female	37	34.6%
BMI (kg/m2)	mean ± SD	22.79 ± 3.38
MutL homolog 1	+	45	42.1%
	-	62	57.9%
PMS1 homolog 2	+	53	49.5%
	-	54	50.5%
MutS homolog 2	+	40	37.4%
	-	67	52.6%
MutS homolog 6	+	49	45.8%
	-	58	54.2%
Deficient mismatch repair protein	1	36	33.6%
	2	62	57.9%
	3	9	8.4%
T stage	1	1	0.9%
	2	3	2.8%
	3	87	81.3%
	4	16	15.0%
N stage	0	73	68.2%
	1	22	20.6%
	2	12	11.2%
AJCC TNM staging system	II	73	68.2%
	III	34	31.8%
LVI	Positive	23	21.5%
	Negative	84	78.5%
PNI	Positive	14	13.1%
	Negative	93	86.9%
CA19-9 (35 U/mL)	mean ± SD	47.2 ± 135.5
CEA (5 ng/mL)	mean ± SD	8.2 ± 26.0
CA19-9 (35 U/mL)	Elevated	18	16.8%
	Normal	87	81.3%
	Missing data	2	1.9%
CEA (5 ng/mL)	Elevated	28	26.2%
	Normal	77	72.0%
	Missing data	2	1.9%
Recurrence	Yes	21	19.6%
	No	86	80.4%

BMI: Body mass index; AJCC: American Joint Committee on Cancer; TNM: Tumor-node-metastasis; LVI: Lymphovascular invasion; PNI: Perineural invasion; CEA: Carcinoembryonic antigen; CA19-9: Cancer antigen 19-9.

Associations between LVI, PNI and clinicopathological characteristics

We next examined the relationships between LVI, PNI, and clinicopathological features using χ² tests or Fisher’s exact tests for categorical variables. Patients with LVI positivity demonstrated significant associations with advanced lymph node stage (P < 0.001), higher clinical stage distribution (stage III: 86.9% vs 43.5%, P < 0.001), and significantly elevated recurrence rates (56.5% vs 9.5%, P < 0.001). Parallel patterns were observed in PNI-positive subgroups (stage III: 57.1% vs 17.2%, P = 0.003) (Tables 2 and 3). Patients with LVI positivity exhibited significantly inferior 5-year RFS (P < 0.001) compared to LVI-negative counterparts (Figure 1A). Similarly, PNI status was associated with reduced 5-year RFS (P < 0.001) (Figure 1B), highlighting the independent prognostic value of these histopathological markers.

Figure 1 Random forest analysis and survival curves. A: Lymphovascular invasion status; B: Perineural invasion status; C: Kaplan-Meier curves illustrate recurrence-free survival. pTNM: Pathological tumor-node-metastasis; LVI: Lymphovascular invasion; PNI: Perineural invasion; BMI: Body mass index; CEA: Carcinoembryonic antigen; CA19-9: Cancer antigen 19-9.

Table 2 Relationship between lymphovascular invasion and clinical characteristics, n (%).

Variables		LVI negative (n = 84)	LVI positive (n = 23)	P value
Age (≥ 53 years/< 53 years)		34 (40.5)/50 (59.5)	8 (34.8)/15 (65.2)	0.787
BMI (kg/m2)	mean ± SD	23.1 ± 3.3	55.8 ± 3.3	0.798
Sex (female/male)		29 (34.5)/55 (65.5)	8 (34.8)/15 (65.2)	1.000
CA19-9 (< 35 U/mL/≥ 35 U/mL)	Missing data	70 (83.3)/13 (15.5)/1 (1.2)	17 (73.9)/5 (21.7)/1 (4.3)	0.643
CEA (< 5 ng/mL/≥ 5 ng/mL)	Missing data	61 (72.6)/22 (26.2)/1 (1.2)	16 (69.6)/6 (26.1)/1 (4.3)	1.000
PNI (negative/positive)		77 (91.7)/7 (8.3)	16 (69.6)/7 (30.4)	0.015
T stage (1/2/3/4)		1 (1.2)/1 (1.2)/69 (82.1)/13 (15.5)	0 (0)/2 (8.7)/18 (78.3)/3 (13.0)	0.261
N stage (0/1/2)		64 (76.2)/16 (19.0)/4 (4.8)	9 (39.1)/6 (26.1)/8 (34.8)	< 0.001
AJCC TNM staging system (II/III)		73 (86.9)/11 (13.1)	10 (43.5)/13 (56.5)	< 0.001
Recurrence (no/yes)		76 (90.5)/8 (9.5)	10 (43.5)/13 (56.5)	< 0.001

LVI: Lymphovascular invasion; BMI: Body mass index; CA19-9: Cancer antigen 19-9; CEA: Carcinoembryonic antigen; PNI: Perineural invasion; AJCC: American Joint Committee on Cancer; TNM: Tumor-node-metastasis.

Table 3 Relationship between perineural invasion and clinical characteristics, n (%).

Variables		PNI negative (n = 93)	PNI positive (n = 14)	P value
Age (≥ 53 years/< 53 years)		37 (39.8)/56 (60.2)	5 (35.7)/9 (64.3)	1.000
BMI (kg/m2)	mean ± SD	23.0 ± 3.2	23.0 ± 3.2	0.700
Gender (female/male)		37(39.8)/56 (60.2)	0 (0)/14 (100.0)	0.009
CA19-9 (< 35 U/mL/≥ 35 U/mL)	Missing data	77 (82.8)/15 (16.1)/1 (1.1)	10 (71.4)/3 (21.4)/1 (7.1)	0.831
CEA (< 5 ng/mL/≥ 5 ng/mL)	Missing data	68 (73.1)/24 (25.8)/1 (1.1)	9 (64.3)/4 (28.6)/1 (7.1)	0.982
LVI (negative/positive)		77 (82.8)/16 (17.)	7 (50.0)/7 (50.0)	0.015
T stage (1/2/3/4)		1 (1.1)/2(2.2)/78 (83.9)/12 (12.9)	0 (0)/1 (7.1)/9 (64.3)/4 (28.6)	0.283
N stage (0/1/2)		71 (76.3)/14 (15.1)/8 (8.6)	2 (14.3)/8 (57.1)/4 (28.6)	< 0.001
AJCC TNM staging system (II/III)		77 (82.8)/16 (17.2)	6 (42.9)/8 (57.1)	0.003
Recurrence (no/yes)		78 (83.9)/15 (16.1)	8 (57.1)/6 (42.9)	0.047

PNI: Perineural invasion; BMI: Body mass index; CA19-9: Cancer antigen 19-9; CEA: Carcinoembryonic antigen; LVI: Lymphovascular invasion; AJCC: American Joint Committee on Cancer; TNM: Tumor-node-metastasis.

Random forest results

The feature importance analysis of the random forest model demonstrates significant variations in the predictive contributions of clinical variables for patients’ RFS (Figure 1C). The importance scores reveal that pathological TNM staging (0.379) exhibits the highest predictive value, followed by LVI (0.154), PNI (0.108), and BMI (0.012).

In contrast, clinical variables including CEA and CA19-9 show relatively lower contributions. The importance scores for CEA and CA19-9 were substantially lower than that of pathological TNM staging, suggesting that while these serum markers contribute to prognosis, they may have limited independent predictive value in this molecular subtype.

Univariable and multivariable Cox regression results

Univariable Cox proportional hazards regression analysis revealed significantly worse RFS in patients with LVI (HR = 7.34, 95%CI: 3.02-17.86, P < 0.001) and PNI (HR = 5.11, 95%CI: 1.92-13.60, P = 0.001). Stage III patients showed an 8.98-fold higher recurrence risk compared to stage II patients (95%CI: 3.61-22.34, P < 0.001). Using variables that showed significance in univariable analysis, we constructed a multivariable Cox proportional hazards regression model including LVI, PNI, and TNM staging.

The multivariable analysis identified TNM stage (adjusted HR = 5.23, 95%CI: 1.75-15.62, P = 0.003) and LVI (adjusted HR = 3.31, 95%CI: 1.23-8.91, P = 0.018) as independent prognostic factors, while PNI did not demonstrate independent prognostic significance for recurrence risk (adjusted HR = 1.75, 95%CI: 0.60-5.14, P = 0.305) (Table 4).

Table 4 Results of the univariate and multivariate Cox regression analysis for recurrence-free survival among the clinical characteristics.

Variables	Univariate Cox analysis		Multivariate Cox analysis
	HR (95%CI)	P value	HR (95%CI)	P value
Age (years)		0.326
mean ± SD	1.02 (0.98-1.05)
BMI (kg/m2)		0.278
mean ± SD	0.92 (0.80-1.07)
LVI		< 0.001		0.018
Negative	Reference		Reference
Positive	7.34 (3.02-17.86)		3.31 (1.23-8.91)
PNI		0.001		0.305
Negative	Reference		Reference
Positive	5.11 (1.92-13.60)		1.75 (0.60-5.14)
Gender		0.200		0.768
Male	Reference
Female	0.52 (0.19-1.42)		1.19 (0.37-3.87)
AJCC TNM staging system		< 0.001		0.003
II	Reference		Reference
III	8.98 (3.61-22.34)		5.23 (1.75-15.62)
CA19-9 (35 U/mL)		0.582
< 35	Reference
≥ 35	0.71 (0.21-2.42)
CEA (5 ng/mL)		0.765
< 5	Reference
≥ 5	1.16 (0.45-2.98)

HR: Hazard ratio; CI: Confidence interval; BMI: Body mass index; LVI: Lymphovascular invasion; PNI: Perineural invasion; AJCC: American Joint Committee on Cancer; TNM: Tumor-node-metastasis; CA19-9: Cancer antigen 19-9; CEA: Carcinoembryonic antigen.

Nomogram construction and validation

We constructed a prognostic nomogram incorporating significant covariates identified through Cox proportional hazards and random forest analyses, combined with clinically validated prognostic parameters. TNM stage, exhibiting the largest absolute coefficient value, was allocated a reference scale ranging from 0 points to 57.5 points (Figure 2A). LVI emerged as the strongest predictor, contributing the maximum risk score (36 points). The total nomogram scores were converted to estimated 1-, 3-, and 5-year recurrence probabilities via nonlinear transformation. Bootstrap internal validation (500 iterations) confirmed satisfactory calibration accuracy (Figure 2B).

Figure 2 Prognostic nomogram construction and calibration. A: Multivariable predictive model integrating tumor-node-metastasis stage, lymphovascular invasion, perineural invasion, and body mass index for recurrence-free survival estimation; B: Calibration plots demonstrate agreement between predicted and observed outcomes for 1-year, 3-year, and 5-year recurrence-free survival; C: Time-dependent receiver operating characteristic analysis of the nomogram’s predictive accuracy. pTNM: Pathological tumor-node-metastasis; LVI: Lymphovascular invasion; PNI: Perineural invasion; BMI: Body mass index; RFS: Recurrence-free survival.

The model demonstrated a concordance index (C-index) of 0.804 upon bootstrap internal validation (500 resamples), indicating robust discriminatory performance. Time-dependent receiver operating characteristic analysis yielded area under the curve values of 0.814 (1-year), 0.871 (3-year), and 0.862 (5-year) for recurrence prediction (Figure 2C). Decision curve analysis consistently demonstrated positive net benefits across clinically relevant threshold probabilities for 1-5 years survival predictions (Figure 3).

Figure 3 Decision curve analysis evaluating clinical utility of the nomogram across threshold probabilities. RFS: Recurrence-free survival.

DISCUSSION

Although LVI and PNI are established prognostic biomarkers in colorectal cancer, their significance remains poorly characterized in the specific context of dMMR/MSI-H rectal adenocarcinoma. Our study aimed to address this gap. The unique immune microenvironment and distinct clinical behavior of dMMR/MSI-H tumors may alter the prognostic implications of these traditional pathological features, highlighting the need for dedicated studies in this molecular subset.

Our findings demonstrate that in this retrospective multi-institutional cohort of 107 treatment-naive stage II-III dMMR/MSI-H rectal adenocarcinoma patients, histopathologically confirmed LVI and PNI were observed in 21.5% (23/107) and 13.1% (14/107) of cases, respectively. Critically, multivariable analysis identified LVI as an independent risk factor for RFS (HR = 3.31, P = 0.018), while PNI showed no independent prognostic value (HR = 1.75, P = 0.305). These findings collectively support the integration of LVI/PNI status into prognostic models for dMMR/MSI-H rectal cancer management.

Although current National Comprehensive Cancer Network Guidelines (2024) recommend against adjuvant chemotherapy for stage II dMMR/MSI-H patients (category 2A)[11], emerging evidence demonstrates significant tumor biological heterogeneity within the dMMR/MSI-H molecular subgroup[19]. Of particular relevance, LVI/PNI correlate strongly with occult lymph node micrometastases detectable via enhanced pathology, suggesting a potential high-risk subset that might benefit from adjuvant therapy-a hypothesis warranting prospective validation[20]. Notably, stage III patients with concomitant LVI/PNI positivity exhibited elevated recurrence rates, a phenomenon mechanistically linked to epithelial-mesenchymal transition and autophagic flux suppression in multi-omics studies[21-23]. While oxaliplatin-based regimens may benefit LVI/PNI-positive subgroups, rigorous risk-benefit assessments considering chemotherapy-related neurotoxicity remain imperative.

The significance of this study extends beyond its academic contribution by providing a foundation for refining clinical decision-making. The identification of LVI as an independent prognostic factor, in particular, could aid in risk stratification. BMI was retained because it is readily available and reflects nutritional status; moreover, obesity-related immune modulation may influence cancer outcomes in dMMR/MSI-H disease. However, its contribution in the random forest analysis was small, and BMI should therefore be viewed as exploratory. More refined body composition indicators, including sarcopenia and visceral adiposity, may provide greater prognostic relevance and deserve further investigation. We developed a preliminary prognostic nomogram that combines LVI, PNI, and TNM stage to estimate long-term survival. This tool is intended for exploration only. The number of recurrence events was small, which limits the reliability of the model. For this reason, the nomogram should be seen as hypothesis-generating and not ready to guide clinical practice.

This study has several important limitations. Postoperative adjuvant chemotherapy data were not collected in a sufficiently standardized manner across participating centers, making treatment-related confounding difficult to control. This issue is especially relevant in stage III disease, where adjuvant chemotherapy substantially affects recurrence risk. As a result, the prognostic effect attributed to LVI may have been influenced by treatment heterogeneity and should therefore be interpreted cautiously until validated in prospective cohorts with detailed treatment adjustment. In addition, only 21 recurrence events were observed relative to the number of candidate predictors included in the multivariable analysis, raising concerns regarding model stability and overfitting. Although bootstrap internal validation was applied instead of random data splitting to reduce optimism bias, the current model remains exploratory and hypothesis-generating, particularly given the rarity of dMMR/MSI-H rectal adenocarcinoma. The proposed nomogram also lacks external validation and should not be considered suitable for clinical application at present. Finally, interobserver variability in pathological assessment may still exist, and comprehensive molecular profiling data were unavailable for all patients.

Future prospective studies should record adjuvant therapy regimens in detail and include them in the analysis. In addition, the model was developed within a single cohort, although patients were enrolled from multiple institutions. There is no external validation. There may also be variability between observers when assessing LVI and PNI, and molecular profiling data were not available for all cases. Taken together, these limitations mean that the prognostic nomogram is preliminary and needs validation in larger, prospective cohorts before it can be considered for clinical use.

Therefore, our findings, particularly regarding the independent prognostic value of PNI, should be considered hypothesis-generating, and the nomogram must be viewed as preliminary. Future studies incorporating multi-omics approaches and validation in larger, prospective cohorts are essential to confirm and refine our model for potential clinical application[24-27].

CONCLUSION

In conclusion, this exploratory study demonstrates that LVI is an independent predictor of RFS in patients with stage II-III dMMR/MSI-H rectal adenocarcinoma, whereas PNI showed no independent prognostic value. These markers may be useful for future risk assessment, but the current results are hypothesis-generating. The proposed nomogram is preliminary. It needs careful external validation in larger, prospective cohorts. Such studies should also account for key confounders, including adjuvant therapy, before the model can be considered for clinical use.