Construction and validation of a prognostic nomogram for advanced esophageal squamous cell carcinoma patients treated with PD-1 inhibitor-based therapy

Abstract

This study aimed to develop a prognostic nomogram to enhance the accuracy of survival prediction and guide individualized treatment decisions for patients with advanced esophageal squamous cell carcinoma (ESCC). A total of 162 patients with advanced ESCC treated with PD-1 inhibitor-based therapy at the First Affiliated Hospital of Anhui Medical University constituted the training set, while 79 patients from the Second Affiliated Hospital formed the validation set. Independent prognostic factors associated with overall survival (OS) were identified using multivariate Cox regression analysis to construct the nomogram. The model’s performance was evaluated in terms of discrimination, calibration, generalizability and clinical utility through the concordance index (C-index), area under the receiver operating characteristic curve (AUC), calibration plots, external validation and decision curve analysis (DCA). Kaplan-Meier analysis with Log-Rank tests was employed to compare OS across different risk strata. The nomogram incorporated five independent prognostic variables. In the training set, the C-index was 0.769, with AUC values of 0.942, 0.850, and 0.658 for predicting 0.5-, 1-, and 2-year OS, respectively. In the validation set, the C-index was 0.786, with corresponding AUC values of 0.916, 0.919, and 0.800. The nomogram demonstrated superior predictive performance compared to the neutrophil-to-lymphocyte ratio (NLR) based on both C-index and AUC metrics. Calibration plots showed good agreement between predicted and observed survival probabilities at 0.5, 1, and 2 years. DCA confirmed the satisfactory clinical utility of the model. Stratification analyses revealed significantly longer OS in low-risk patients compared to high-risk patients in both sets (all P < 0.001). This nomogram provides a reliable tool for predicting prognosis in patients with advanced ESCC undergoing PD-1 inhibitor-based therapy, offering a potential reference for clinical decision-making.

Introduction

Esophageal squamous cell carcinoma (ESCC) is among the most aggressive malignancies of the gastrointestinal tract, with high global incidence and mortality rates [1]. In China, more than half of ESCC cases are diagnosed at locally advanced or metastatic stages due to the lack of specific early clinical manifestations [2]. Conventional treatment modalities, including surgery, chemotherapy, and radiotherapy, have demonstrated limited efficacy in advanced disease, resulting in poor clinical outcomes: median overall survival (OS) typically ranges from 8 to 11 months, with a 5-year survival rate of approximately 5% [3].

In recent years, the increasing use of programmed death receptor-1 (PD-1) inhibitors has marked a new era in the treatment of ESCC. Multiple phase III randomized controlled trials have confirmed the efficacy of immunotherapy in advanced ESCC, with PD-1 inhibitor-based regimens significantly improving survival outcomes [4]. For instance, in the RATIONALE-306 trial, the median OS reached 17.2 months [5]. As a result, immunotherapy has become the standard of care for first-line and second-line treatment in advanced ESCC. However, real-world evidence indicates that only a subset of patients derive clinical benefit from immunotherapy, underscoring the critical need to accurately identify responders as a key clinical priority. Currently available biomarkers for predicting immunotherapy response, including programmed death-ligand 1 (PD-L1) expression, microsatellite instability (MSI), and tumor mutation burden (TMB), are limited by several practical constraints [6]. These include the need for invasive tissue sampling, technical complexity, high costs, and inconsistent predictive performance [7, 8], all of which hinder their widespread clinical application. Although emerging multi-gene signature models show promise [9, 10], their reliance on sophisticated molecular assays and the lack of extensive real-world validation limit their feasibility, reproducibility, and clinical utility.

Prognostic nomograms, which quantify and integrate key prognostic factors, have become widely adopted in oncology prognosis research [11]. In this study, we developed and validated a prognostic nomogram for advanced ESCC patients treated with PD-1 inhibitor-based therapy, based on easily accessible clinical parameters. This nomogram is intended to facilitate precise prognostic prediction and assist clinicians in tailoring individualized treatment strategies for these patients.

Materials and methods

Study subjects

The training set consisted of advanced ESCC patients treated at the First Affiliated Hospital of Anhui Medical University between August 2019 and February 2022. The validation set included patients from the Second Affiliated Hospital during the same period. Inclusion criteria were as follows: (a) histologically or cytologically confirmed ESCC; (b) unresectable locally advanced or metastatic disease; (c) receipt of PD-1 inhibitor-based therapy (monotherapy or combination); (d) availability of complete clinical data. Exclusion criteria were: (a) presence of another primary malignancy; (b) participation in clinical trials (due to blinded treatment regimens making it impossible to confirm whether placebo or PD-1 inhibitor was administered); (c) absence of baseline hematological test data; (d) active severe infection, autoimmune disease, or hematologic disorder. All patients received at least two cycles of PD-1 inhibitor-based therapy, and treatment was discontinued upon disease progression, occurrence of intolerable toxicities or adverse effects, or patient refusal of further therapy. This study was approved by the Ethics Committee of Anhui Medical University (Approval No. Quick-PJ 2022-14-35).

Data collection

Patient information was primarily obtained from the electronic medical record system, including: age, gender, primary tumor location, distant metastatic sites, Eastern Cooperative Oncology Group Performance Status (ECOG PS), history of alcohol consumption and smoking, prior radical surgery, treatment line, type of PD-1 inhibitor, and treatment regimen. Baseline peripheral blood routine test results were also collected to calculate inflammatory markers, including the neutrophil-to-lymphocyte ratio (NLR) and systemic immune-inflammation index (SII). The definitions and cutoff values for NLR and SII were based on our previous study [12]; the cutoff values were set at 4.748 and 887.895, respectively.

Follow-up

Follow-up was primarily conducted through the electronic medical records system, supplemented by telephone calls, WeChat messages, and other methods. OS was defined as the interval from the initiation of PD-1 inhibitor-based therapy to death from any cause or the last follow-up. The final follow-up dates for the training and validation sets were August 31, 2022, and June 29, 2023, respectively.

Statistical analysis

Baseline clinical characteristics between the training and validation sets were compared using the chi-square test or Fisher’s exact test. Univariate Cox proportional hazards regression analysis was first performed on the training set to identify clinical variables with P < 0.05 as potential prognostic factors. Multivariate Cox regression analysis was subsequently conducted using stepwise backward elimination to determine independent prognostic factors for OS. Based on these independent predictors, a nomogram was constructed using the “rms” package in R software, assigning weighted scores to predict 0.5-, 1-, and 2-year OS probabilities for advanced ESCC patients treated with PD-1 inhibitor-based therapy. The discriminative ability of the nomogram was assessed using the concordance index (C-index) [13] and the area under the time-dependent receiver operating characteristic(ROC) curve (AUC) [14], calculated at 0.5-, 1-, and 2-year intervals using the “timeROC” package. Calibration was evaluated through calibration plots [15]. External validation was performed in the validation set using the same metrics (C-index, AUC, and calibration plots) to assess the model’s generalizability [11]. The clinical utility of the model was assessed by decision curve analysis (DCA). Patients were stratified into low-risk and high-risk groups based on the median prognostic score from the training set [16], and differences in OS were analyzed using Kaplan-Meier curves and the Log-Rank test.

Statistical analyses were performed using SPSS 26.0 software (SPSS Inc., Chicago, IL, USA), and R software version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria) was employed for graphical visualization. All statistical tests were two-sided, and a P-value < 0.05 was considered statistically significant.

Results

Baseline clinical characteristics

A total of 162 patients were included in the training set and 79 patients in the validation set based on the predefined inclusion and exclusion criteria. The baseline clinical characteristics of both sets are summarized in Table 1. The majority of patients were male, with most having an ECOG PS ≤ 1. The primary tumor site was predominantly located in the middle and lower esophagus, and lymph nodes were the most common site of distant metastasis. Due to factors including drug availability and cost-effectiveness, Camrelizumab was the most commonly prescribed PD-1 inhibitor in clinical practice. Comparative analysis revealed no significant differences in baseline clinical characteristics between the two sets, except for treatment line (P = 0.045) and treatment regimen (P < 0.001). In the training set, 119 patients (73.5%) received first-line therapy, a significantly higher proportion than the 48 patients (60.8%) in the validation set. Regarding treatment regimen, no patients in the training set received monotherapy with PD-1 inhibitors; all received combination immunotherapy, including 96 patients (59.3%) treated with PD-1 inhibitors plus chemotherapy, 29 patients (17.9%) with PD-1 inhibitors plus targeted therapy, and 37 patients (22.8%) receiving PD-1 inhibitors combined with both chemotherapy and targeted therapy. In contrast, 12 patients (15.2%) in the validation set received PD-1 inhibitor monotherapy, and only 2 patients (2.5%) received combined chemotherapy and targeted therapy.

Table 1.

Baseline clinical characteristics of patients in the training set and validation set

Characteristics	Patients, N(%)		P value
	Training set (N = 162)	Validation set (N = 79)
Age(years)			0.131
≤ 65	74(45.7)	28(35.4)
> 65	88(54.3)	51(64.6)
Gender			0.936
Male	136(84.0)	66(83.5)
Female	26(16.0)	13(16.5)
Tumor location			0.898
Upper	10(6.2)	6(7.6)
Middle	79(48.8)	39(49.4)
Low	73(45.0)	34(43.0)
Metastatic sites
Liver metastasis			0.922
Negative	126(77.8)	61(77.2)
Positive	36(22.2)	18(22.8)
Distant lymph node metastasis			0.685
Negative	94(58.0)	48(60.8)
Positive	68(42.0)	31(39.2)
Lung metastasis			0.061
Negative	125(77.2)	52(65.8)
Positive	37(22.8)	27(34.2)
Bone metastasis			0.257
Negative	147(90.7)	75(94.9)
Positive	15(9.3)	4(5.1)
ECOG PS			0.087
≤ 1	149(92.0)	67(84.8)
≥ 2	13(8.0)	12(15.2)
Alcohol consumption history			0.051
No	111(68.5)	44(55.7)
Yes	51(31.5)	35(44.3)
Smoking history			0.083
No	103(63.6)	41(51.9)
Yes	59(36.4)	38(48.1)
Radical surgery history			0.729
No	70(43.2)	36(45.6)
Yes	92(56.8)	43(54.4)
Treatment line			0.045
1 line	119(73.5)	48(60.8)
≥ 2 lines	43(26.5)	31(39.2)
Type of PD-1 inhibitor			0.080
Camrelizumab	139(85.8)	75(94.9)
Sintilimab	18(11.1)	3(3.8)
Toripalimab	5(3.1)	1(1.3)
Treatment regimen			< 0.001
PD-1 monotherapy	0(0.0)	12(15.2)
PD-1 + Chemotherapy	96(59.3)	45(57.0)
PD-1 + Target therapy	29(17.9)	20(25.3)
PD-1 + Chemotherapy	37(22.8)	2(2.5)
+ Target therapy
NLR			0.071
≤ 4.748	128(79.0)	54(68.4)
> 4.748	34(21.0)	25(31.6)
SII			0.602
≤ 887.895	118(72.8)	55(69.6)
> 887.895	44(27.2)	24(30.4)

ECOG PS Eastern Cooperative Oncology Group Performance Status, PD-1 programmed death receptor-1, NLR neutrophil-to-lymphocyte ratio, SII systemic immune-inflammation index

Construction of a prognostic nomogram

Multivariate Cox regression analysis of the training set identified five independent prognostic factors associated with OS: distant lymph node metastasis (P = 0.001), ECOG PS (P = 0.002), treatment line (P = 0.048), baseline NLR (P = 0.002), and SII (P = 0.016) (Table 2). Based on these variables, a prognostic nomogram was developed to predict survival outcomes in patients with advanced ESCC receiving PD-1 inhibitor-based therapy (Fig. 1). Each independent prognostic factor was assigned a weighted score in the nomogram. Quantitative assessment demonstrated that ECOG PS had the greatest contribution to the prognostic score (Fig. 1), indicating it was the most influential predictor of OS. As a visual predictive tool, this nomogram allows clinicians to integrate the five aforementioned clinical parameters prior to initiating immunotherapy, calculate individualized prognostic scores, and estimate 0.5-, 1-, and 2-year OS probabilities, thereby facilitating more informed treatment decisions.

Table 2.

Univariate and multivariate Cox regression analyses of overall survival in the training set

Characteristics	Univariate analysis		Multivariate analysis
	HR (95% CI)	P value	HR (95% CI)	P value
Age(years)
≤ 65	1
> 65	0.888(0.574–1.374)	0.593	-	-
Gender
Male	1
Female	0.689(0.372–1.279)	0.235	-	-
Tumor location		0.846
Upper	1
Middle	1.313(0.515–3.347)	0.326	-	-
Low	1.295(0.505–3.321)	0.591	-	-
Metastatic sites
Liver metastasis
Negative	1
Positive	1.484(0.902–2.441)	0.118	-	-
Distant lymph node metastasis
Negative	1
Positive	2.163(1.393–3.357)	< 0.001	2.153(1.364–3.398)	0.001
Lung metastasis
Negative	1
Positive	1.360(0.847–2.182)	0.201	-	-
Bone metastasis
Negative	1
Positive	1.058(0.528–2.120)	0.873	-	-
ECOG PS
≤ 1	1
≥ 2	6.529(3.206–13.299)	< 0.001	3.454(1.600-7.457)	0.002
Alcohol consumption history
No	1
Yes	0.990(0.603–1.627)	0.969	-	-
Smoking history
No	1
Yes	1.049(0.656–1.674)	0.839	-	-
Radical surgery history
No	1
Yes	0.991(0.638–1.538)	0.967	-	-
Treatment line
1 line	1
≥ 2 lines	2.228(1.391–3.570)	0.001	1.665(1.005–2.760)	0.048
Type of PD-1 inhibitor		0.948
Camrelizumab	1
Sintilimab	1.062(0.560–2.014)	0.853	-	-
Toripalimab	1.183(0.370–3.779)	0.777	-	-
Treatment regimen		0.172
PD-1 + Chemotherapy	1
PD-1 + Target therapy	1.660(0.976–2.821)	0.061	-	-
PD-1 + Chemotherapy	1.153(0.678–1.959)	0.600	-	-
+ Target therapy
NLR
≤ 4.748	1
> 4.748	4.947(3.081–7.943)	< 0.001	2.736(1.451–5.159)	0.002
SII
≤ 887.895	1
> 887.895	3.924(2.515–6.122)	< 0.001	2.068(1.144–3.739)	0.016

HR hazard ratios, 95% CI 95% confidence intervals, ECOG PS Eastern Cooperative Oncology Group Performance Status, PD-1 programmed death 1, NLR neutrophil-to-lymphocyte ratio, SII systemic immune-inflammation index.

Fig. 1.

A nomogram for prognosis of advanced esophageal squamous cell carcinoma patients treated with programmed death receptor-1 inhibitor-based therapy

Evaluation and validation of a prognostic nomogram

In the training set, the nomogram achieved a C-index of 0.769, with time-dependent AUC values of 0.942, 0.850, and 0.658 for predicting 0.5-, 1-, and 2-year OS, respectively (Fig. 2a–c). In the validation set, the nomogram demonstrated a C-index of 0.786, with corresponding AUC values of 0.916, 0.919, and 0.800 (Fig. 2d–f). Calibration plot analysis revealed good agreement between observed and predicted survival probabilities at all aforementioned time points in both the training (Fig. 3a) and validation (Fig. 3b) sets. These findings confirm the nomogram’s strong discriminatory capacity and calibration accuracy, with its robustness further substantiated by external validation. We subsequently evaluated the prognostic performance of the NLR. The C-index of NLR was 0.682 in the training set and 0.684 in the validation set, both significantly lower than the corresponding values for the nomogram (P < 0.001, P < 0.001). ROC curve analysis further demonstrated that the time-dependent AUC values of NLR were significantly inferior to those of the nomogram at all time points in both the training (Fig. 2a–c) and validation (Fig. 2d–f) sets. These results indicate that the nomogram provides superior discriminatory performance compared to this single inflammatory biomarker. Given that the median OS of the training set and validation set in this study were 15.4 months and 15.2 months, respectively, we performed DCA of the nomogram for predicting 1-year OS, and the results are presented in Fig. 4. The findings demonstrated that this model exhibited good clinical utility.

Fig. 2.

Time-dependent receiver operating characteristic (ROC) curves of the nomogram and neutrophil-to-lymphocyte ratio (NLR) for predicting 0.5- (a, d), 1- (b, e), and 2- (c, f) year overall survival (OS) in the training and validation set

Fig. 3.

Calibration plots of the nomogram for predicted versus observed 0.5-, 1-, and 2-year overall survival (OS) in the training set (a) and validation set (b)

Fig. 4.

Decision curve analysis (DCA) curves for predicting 1-year overall survival (OS) in the training set (a) and validation set (b)

Prognosis of advanced ESCC patients with different risk stratification

In the training set, the median prognostic score was 61.3, serving as a cutoff to stratify patients into low-risk and high-risk groups for evaluating the nomogram’s prognostic discriminative ability. Kaplan-Meier survival analysis showed that the low-risk group had significantly better OS compared to the high-risk group (median OS: 19.1 months, 95% CI 15.5–22.7 vs. 11.3 months, 95% CI 9.5–13.0; P < 0.001) (Fig. 5a). In the validation set, As shown in Fig. 5b, the low-risk group had a median OS of 21.2 (95% CI 17.4–25.0) months, whereas the high-risk group had a median OS of 8.0 (95% CI 6.6–9.3) months, representing a statistically significant difference (P < 0.001). These results indicate that the cutoff also effectively discriminates survival outcomes across the validation set, supporting the nomogram’s clinical utility.

Fig. 5.

Kaplan-Meier curves of overall survival (OS) for patients stratified into low-risk and high-risk groups in the training set (a) and validation set (b)

Discussion

Despite the significant improvements in OS among patients with advanced ESCC receiving PD-1 inhibitor therapy [17], tumor heterogeneity and inter-patient variability in clinical characteristics result in a considerable proportion of patients deriving limited clinical benefit. Pre-treatment prognostic prediction is essential for guiding clinical decision-making, optimizing the allocation of medical resources, and facilitating treatment planning for patients and their families. Previous studies have demonstrated that single biomarkers offer limited prognostic utility [18], and inadequate evaluation may lead to misguided therapeutic strategies or even delay optimal interventions. In this study, we developed a prognostic nomogram for patients with advanced ESCC treated with PD-1 inhibitor-based therapy, integrating five clinically relevant parameters: status of distant lymph node metastasis, ECOG PS, treatment line, baseline NLR, and SII. These parameters can be obtained through routine tumor assessments and non-invasive laboratory tests, ensuring clinical feasibility, practicality, and cost-effectiveness. The nomogram enables rapid and accurate risk stratification for OS, thereby supporting individualized immunotherapy decision-making.

The nomogram was evaluated for its discriminatory and calibration performance. In the training set, the model demonstrated a C-index of 0.769. Time-dependent AUC values for predicting 0.5-, 1-, and 2-year OS were 0.942, 0.850, and 0.658, respectively, indicating strong discriminatory ability. In contrast to conventional biomarkers such as PD-L1 expression, MSI, and TMB, blood-based biomarkers eliminate the need for tumor biopsy and offer advantages including minimal invasiveness, ease of repeated sampling, and cost-efficiency. Consequently, peripheral blood inflammatory markers have increasingly been investigated for their potential in predicting response to immunotherapy and clinical outcomes [19]. Among these, NLR has been the most extensively studied; a meta-analysis incorporating 100 studies confirmed that baseline NLR is significantly associated with OS across multiple malignancies [20], including advanced ESCC treated with PD-1 inhibitors [21, 22]. Therefore, we further assessed the predictive performance of NLR. As shown in Fig. 2, the AUC of NLR was significantly lower than that of the nomogram at all three time points, highlighting the superior predictive accuracy of the nomogram. This enhanced performance may be attributed to the inclusion of ECOG PS, a key determinant of OS. Previous studies have established ECOG PS as a critical prognostic factor in advanced malignancies, including ESCC, with predictive value comparable to that of more complex models [23]. A recent meta-analysis encompassing 60 studies and 35,020 patients showed that patients with ECOG PS ≤ 1 had significantly lower mortality risk and prolonged OS, irrespective of treatment type—monotherapy, combination immunotherapy, or treatment line [24]. Notably, clinical trials evaluating approved immunotherapies predominantly enroll patients with ECOG PS ≤ 1, whereas real-world populations of patients with advanced ESCC, such as elderly individuals or those receiving second-line or later therapy, often have poorer performance status (PS ≥ 2), thereby limiting the generalizability of trial findings. By incorporating ECOG PS into the nomogram, our model addresses this evidence gap and provides clinically relevant prognostic information for patients undergoing immunotherapy in real-world settings. Calibration plots demonstrated good agreement between observed and predicted 0.5-, 1-, and 2-year OS probabilities in the training set, confirming the model’s strong calibration performance.

Systematic validation is essential for assessing the generalizability of a nomogram. Although internal validation techniques, such as cross-validation and bootstrapping, can reduce the risk of overfitting, they do not account for the inherent heterogeneity present in real-world patient populations. In contrast, external validation using independent datasets, particularly those derived from multicenter sources, is widely regarded as the gold standard [25]. To evaluate the generalizability of our nomogram, we performed external validation using data from 79 patients with advanced ESCC treated at the Second Affiliated Hospital of Anhui Medical University. The nomogram demonstrated a C-index of 0.786, with AUC values of 0.916, 0.919, and 0.800 for 0.5-, 1-, and 2-year OS prediction, respectively. The calibration plots also showed good agreement with the reference line. These results indicate that the nomogram exhibits strong discriminative ability and accurate calibration in the validation set, supporting its robust generalizability.

Patients were categorized into low-risk and high-risk groups based on the median prognostic score derived from the training set. Kaplan-Meier analysis revealed statistically significant differences in OS between the two risk groups in both sets, suggesting its potential clinical utility in guiding treatment decisions. Specifically, low-risk patients may derive greater survival benefit from immunotherapy, whereas high-risk patients may be better managed with best supportive care to avoid unnecessary treatment-related declines in quality of life.

This study has several limitations. First, the relatively small sample sizes in both the training and validation sets may have introduced selection bias. Second, there was a certain degree of heterogeneity in the treatment regimens across this study, which may have introduced confounding effects on the evaluation of patient prognosis. Hence, future studies could perform stratified analyses and model development according to different treatment regimens to further enhance the accuracy of prognostic prediction. In addition, other key prognostic factors associated with immunotherapy, such as PD-L1 expression, gut microbiota composition [26], tumor genomic profiles [27], and immunological features of the tumor microenvironment [28], were not incorporated into the nomogram, which may also limit its predictive precision.Third, the external validation was conducted using data from a single center, thereby limiting the strength of the evidence and the broader applicability of the nomogram. Therefore, multicenter, large-scale prospective studies are warranted to further validate its clinical utility.

Conclusion

We developed and validated a prognostic nomogram based on five readily available clinical parameters, which serves as a practical tool for predicting prognosis in patients with advanced ESCC undergoing PD-1 inhibitor-based therapy. This nomogram enables risk stratification and supports the implementation of personalized immunotherapy strategies.

Author contributions

WP conceived and designed the study. LD collected and analyzed the data and drafted the manuscript. ZQ assisted with data analysis. JD assisted with data collection. All authors reviewed and approved the final manuscript.

Funding

No funding sources were reported.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of Anhui Medical University (Approval No. Quick-PJ 2022-14-35).

Informed consent

Informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.