Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Abstract

Background:

Pulmonary lymphoepithelioma-like carcinoma (pLELC) is a rare Epstein–Barr virus-associated subtype of non–small cell lung cancer. Although immune checkpoint inhibitors (ICIs) have shown promising activity, robust predictors of benefit remain lacking. Programmed death-ligand 1 (PD-L1) is widely used but has limited accuracy. Routine blood tests and serum tumor markers (STMs) are inexpensive and universally available, yet their prognostic value in pLELC has not been systematically evaluated.

Objectives:

To develop and validate a composite blood-based score integrating hematologic indices and STMs for predicting immunotherapy outcomes in advanced pLELC.

Design:

Multicenter retrospective cohort study.

Methods:

We retrospectively analyzed 254 patients with advanced pLELC treated with ICIs across six tertiary centers in China. Baseline hematologic indices, serum biochemistry, and STMs were collected. A composite Blood Routine & Tumor-Marker Score (BRTS) was constructed using LASSO Cox regression with progression-free survival (PFS) as the endpoint. Patients were stratified into high- and low-BRTS groups, and prognostic value was validated in an independent cohort. Nomograms combining BRTS with clinical variables were developed and internally validated.

Results:

High BRTS was associated with significantly shorter PFS and overall survival (OS) in both training (hazard ratio (HR) for PFS = 4.59; OS = 6.86) and validation cohorts (HR for PFS = 5.37; OS = 3.87; all p < 0.001). In multivariate analyses, BRTS remained an independent predictor alongside treatment line, regimen, and liver metastasis, whereas PD-L1 expression lost significance. Nomograms incorporating BRTS demonstrated good discrimination (C-index ~0.79), calibration, and clinical net benefit. Prognostic utility was consistent across treatment lines.

Conclusion:

The BRTS, derived from widely available laboratory tests, robustly stratified immunotherapy outcomes in advanced pLELC and outperformed PD-L1 alone. This simple, low-cost tool may facilitate individualized treatment decisions and warrants prospective validation.

Plain language summary

A simple blood test score may help predict immunotherapy outcomes in a rare type of lung cancer

What was studied: Pulmonary lymphoepithelioma-like carcinoma (pLELC) is a rare type of lung cancer often linked to Epstein–Barr virus infection. While immunotherapy has become an important treatment option, not all patients benefit equally, and reliable tools to predict treatment response are lacking. What we did: In this study, we analyzed routine blood test results and common tumor markers from patients with advanced pLELC who received immunotherapy at six hospitals in China. Using these standard laboratory measurements, we developed a new score called the Blood Routine and Tumor-marker Score (BRTS). The BRTS combines information about inflammation, nutrition, and tumor activity to help predict how patients respond to immunotherapy. What we found: Patients with higher BRTS values tended to have shorter survival and poorer treatment response, while those with lower scores lived longer and responded better to immunotherapy. These results were consistent across different hospitals and treatment settings. The BRTS showed good accuracy in identifying patients who may not benefit from immunotherapy alone and who might need combination treatment instead. Why it matters: Because the BRTS uses only simple and widely available blood tests, it could easily be applied in everyday clinical practice. This approach may help doctors choose the most suitable treatment for each patient and monitor therapy more effectively, especially in rare cancers like pLELC where personalized guidance is urgently needed.

Keywords

blood routine immunotherapy prognostic model pulmonary lymphoepithelioma-like carcinoma tumor markers

Introduction

Immunotherapy has reshaped the treatment landscape of advanced non–small cell lung cancer (NSCLC), delivering clinically meaningful and durable benefits across multiple pivotal trials.^1–4 Nevertheless, a substantial proportion of patients experience primary resistance or early progression on immune checkpoint inhibitors (ICIs), underscoring the need for reliable and accessible predictors of benefit. Programmed death‑ligand 1 (PD‑L1) remains the most widely used biomarker in routine practice, yet its performance is imperfect and context‑dependent. Other tissue‑ or sequencing‑based indicators—such as tumor mutational burden and circulating tumor DNA (ctDNA)—may provide complementary information, but cost, assay variability, and tissue availability limit their broad adoption in everyday care.⁵ Consequently, there is strong clinical motivation to identify simple, low cost, and widely available biomarkers to refine patient selection for immunotherapy. Consequently, there is strong clinical motivation to identify simple, low cost, and widely available biomarkers to refine patient selection for immunotherapy.

Pulmonary lymphoepithelioma‑like carcinoma (pLELC) is a rare NSCLC subtype that predominantly affects younger, never‑smoking individuals in southern and Southeast Asia, regions with high Epstein–Barr virus (EBV) prevalence.^6,7 pLELC exhibits distinctive clinicopathological and molecular features, including alterations related to cell‑cycle control and the JAK/STAT and NF‑κB pathways.^8,9 A recent study further linked PD‑L1 status with clinical outcomes on ICIs in pLELC, supporting the biological rationale for immunotherapy in this entity.¹⁰

Building on this rationale, our group and others have reported encouraging activity of immunotherapy in pLELC. We previously observed that chemoimmunotherapy yielded a significantly longer 2‑year progression‑free survival (PFS) than chemotherapy alone among patients with pLELC (hazard ratio, 0.38; 95% confidence interval (CI), 0.18–0.78; p = 0.007).¹¹ In our recent research, chemoimmunotherapy significantly improved both PFS and overall survival (OS) compared with chemotherapy alone in first‑ and second‑line settings: in first‑line treatment, median PFS and OS were 17.6 and 26.1 months with chemoimmunotherapy versus 8.7 and 19.2 months with chemotherapy; in second‑line treatment, median PFS and OS were 5.1 and 13.5 months versus 3.3 and 8.9 months, respectively.¹² In a separate retrospective analysis, first‑line immunotherapy was associated with improved PFS and OS in advanced pLELC and in advanced lung squamous cell carcinoma.¹³ However, because pLELC is uncommon, most available data derive from small, retrospective series with heterogeneous treatment lines and limited biomarker characterization.^14,15 Robust, generalizable tools to identify which patients with pLELC are most likely to benefit from ICIs remain lacking.

Routine peripheral blood tests capture facets of host immunity, systemic inflammation, nutritional status, and tumor burden that may influence ICI responsiveness. Complete blood count–derived indices—such as the neutrophil‑to‑lymphocyte ratio (NLR), platelet‑to‑lymphocyte ratio (PLR), monocyte‑to‑lymphocyte ratio (MLR), and systemic immune‑inflammation index (SII)—together with serum biochemistry markers including lactate dehydrogenase (LDH) and albumin (ALB), have been associated with immunotherapy outcomes in broader NSCLC cohorts.¹⁶ In parallel, baseline serum tumor markers (STMs)—for example carcinoembryonic antigen (CEA), cytokeratin‑19 fragment (CYFRA 21‑1), neuron‑specific enolase (NSE), carbohydrate antigen 19‑9 (CA19‑9), and carbohydrate antigen 125 (CA125)—are routinely measured and have established prognostic value in advanced lung cancer treated with chemotherapy, but their predictive relevance for ICI efficacy is far less defined.^17,18 Given pLELC’s EBV‑driven, immune‑rich microenvironment, inexpensive blood‑based indicators may be particularly informative in this subtype and offer a practical complement to PD‑L1 and sequencing‑based assays. Notably, in a multicenter retrospective study, we preliminarily explored the association of baseline and dynamic STMs with immunotherapy efficacy in pLELC.¹²

Against this backdrop, we conducted a multicenter retrospective study to evaluate whether baseline routine blood tests and STMs can predict the efficacy of immunotherapy in patients with advanced pLELC. Specifically, we assessed the associations of pre‑treatment hematologic and biochemical indices and STMs with short‑term treatment response and survival outcomes (including PFS and OS) among patients receiving ICIs, either as monotherapy or in combination regimens and across treatment lines. Compared with previous hematology-based models such as the Hematological Indices-Based Signature (HIBS),¹⁹ our approach integrates both blood routine indices and STMs to derive a composite Blood Routine & Tumor-Marker Score (BRTS), aiming for broader applicability and stronger prognostic power. We further examined whether integrating multiple readily available parameters improves discriminatory performance compared with single‑marker approaches, deriving a pragmatic, clinic‑ready prediction model for this rare disease.

We hypothesized that routine blood tests, together with baseline STMs, would stratify immunotherapy benefit in pLELC and that a multivariable model combining these features would outperform any single biomarker. Establishing such a tool could help clinicians individualize immunotherapy strategies for pLELC, optimize resource utilization, and prioritize patients for prospective trials aimed at validating blood‑based predictors of ICI response.

Methods

Study reporting

The reporting of this study conforms to the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) Statement.²⁰ The completed TRIPOD checklist is provided in the Supplemental Material.

Study population

This retrospective study included patients diagnosed with pLELC at stages IIIB–IV between January 2017 and June 2024 across six tertiary hospitals in China: Guizhou Provincial People’s Hospital, the Third Affiliated Hospital of Sun Yat-sen University, the Third Xiangya Hospital of Central South University, the University of Hong Kong–Shenzhen Hospital, the General Hospital of Southern Theater Command, and Jiangxi Cancer Hospital (Figure 1).

Figure 1.

Study design for predictive model development in pLELC immunotherapy.

Inclusion and exclusion criteria

Inclusion criteria were as follows: (1) pathologically confirmed diagnosis of pLELC; (2) stage IIIB–IV disease (AJCC 8th edition); and (3) availability of complete baseline clinical and laboratory data, including STMs and blood tests. Exclusion criteria were as follows: incomplete treatment history, Eastern Cooperative Oncology Group performance status (ECOG PS) >2, or receipt of fewer than three cycles of immunotherapy.

Data collection

Baseline clinical and demographic information included sex, age, smoking history, ECOG PS, PD-L1 expression, Tumor–Node–Metastasis (TNM) stage, treatment line, treatment regimen, and metastatic sites.

Laboratory parameters included STMs (CEA, NSE, CYFRA21-1, CA19-9, CA125), biochemical markers (ALB, LDH, C-reactive protein (CRP)), and hematological indices (WBC, neutrophils, lymphocytes, monocytes, platelets, eosinophils, basophils). In addition, systemic inflammation indices were derived from complete blood counts obtained within 14 days before treatment initiation: (1) NLR = neutrophils/lymphocytes, (2) PLR = platelets/lymphocytes, (3) lymphocyte-to-monocyte ratio (LMR) = lymphocytes/monocytes, (4) SII = (platelets × neutrophils)/lymphocytes, (5) derived neutrophil to lymphocyte ratio (dNLR) = neutrophils/(WBC − neutrophils), (6) neutrophil-monocyte-to-lymphocyte ratio (nMLR) = (neutrophils + monocytes)/lymphocytes, and (7) neutrophil-to-platelet ratio (NPR) = neutrophils/platelets.

A small constant (1 × 10⁻⁶) was added to denominators when zero values occurred, and sensitivity analyses were performed without continuity correction. If multiple Complete Blood Count (CBCs) were available, the measurement closest to treatment initiation was selected. Variables with <5% missing values were imputed using multivariate imputation by chained equations to avoid listwise deletion, while features with >20% missingness were excluded from modeling. EBV-DNA results were available only in a subset of patients and were therefore not included in the present analysis to avoid bias; future prospective studies with standardized EBV-DNA testing are warranted.

Definition of endpoints

OS was defined as the time from treatment initiation to death from any cause. PFS was defined as the time from treatment initiation to radiological progression or death. Tumor response was evaluated by CT scans using RECIST v1.1. Best overall response (BOR) categories included complete response, partial response (PR), stable disease (SD), and progressive disease (PD).

PD-L1 expression assessment

PD-L1 expression was assessed using the Dako 22C3 pharmDx antibody, and tumor proportion scores (TPS) were classified as negative (0%–1%), low (1%–49%), or high (⩾50%). In this cohort, no PD-L1-negative (TPS 0%) cases were identified, consistent with prior reports showing high PD-L1 positivity in pLELC.

Blood routine and tumor-marker score

A composite BRTS was developed to integrate laboratory biomarkers into a unified prognostic index. Candidate variables included tumor markers (CEA, NSE, CYFRA21-1, CA19-9, and CA125), biochemical markers (ALB, LDH, and CRP), peripheral blood counts, and derived inflammatory indices (NLR, PLR, LMR, SII, dNLR, nMLR, and NPR). Right-skewed variables were log-transformed, and all continuous variables were standardized.

Feature selection and coefficient estimation were performed using a least absolute shrinkage and selection operator (LASSO) Cox regression implemented in Python (version 3.10; Python Software Foundation, Wilmington, DE, USA) with the scikit-survival and scikit-learn libraries. The optimal penalty parameter (λ) was determined by 10-fold cross-validation using the “1-standard error” rule to balance model simplicity and performance. Variables with nonzero coefficients at the optimal λ were retained to construct the final BRTS formula, developed exclusively in the training cohort to prevent information leakage. Coefficients were directionally aligned so that higher BRTS values consistently indicated a higher risk of progression.

The optimal cutoff value for BRTS was determined using the maximally selected log-rank statistic (lifelines package) based on PFS in the training cohort. This PFS-derived cutoff was then locked and directly applied to OS analyses without re-optimization to avoid endpoint-specific bias and ensure methodological consistency.

Proportional hazards assumptions for all Cox models were verified using Schoenfeld residuals. Model discrimination was assessed via bootstrap resampling with 1000 iterations to obtain optimism-corrected concordance indices (C-index). To explore potential center effects, a sensitivity Cox regression model including hospital site as a stratification factor was performed. Because two major institutions (the Third Affiliated Hospital of Sun Yat-sen University and the General Hospital of Southern Theater Command) contributed more than half of all cases, this analysis was exploratory in nature.

Although BRTS was constructed from baseline parameters, all included biomarkers are routinely collected during treatment; thus, BRTS could theoretically be recalculated dynamically, but this was not formally analyzed in the present study.

Treatment regimens

All patients received at least three cycles of systemic therapy. Treatment regimens included the following: (1) platinum-based doublet chemotherapy plus PD-1/PD-L1 inhibitors; (2) non-platinum chemotherapy plus PD-1/PD-L1 inhibitors; or (3) PD-1/PD-L1 inhibitor monotherapy. Both first-line and second-line immunotherapy were analyzed separately in subgroup analyses.

Training and validation cohorts

To enable model development and internal validation, the dataset was randomly divided into a training cohort (70%) and an internal validation cohort (30%) using stratified random sampling based on PFS event status (progression vs censoring). The training cohort was used for variable selection, model construction, and nomogram development, while the validation cohort was reserved for independent performance assessment. A fixed random seed was applied to ensure reproducibility.

Evaluation of BRTS

After calculating BRTS, patients were stratified into high- and low-risk groups. Kaplan–Meier survival curves for PFS and OS were plotted in both training and validation cohorts, with differences compared using the log-rank test. Hazard ratios with 95% CIs were estimated from Cox models.

To further assess clinical relevance, the association between BRTS risk groups and BOR distribution was analyzed descriptively. BOR categories were compared between high- and low-BRTS groups using chi-square or Fisher’s exact tests.

Subgroup analysis by treatment line

Patients were further stratified according to treatment line (first line vs second line). Within each subgroup, Kaplan–Meier survival analyses were performed for PFS and OS, with comparisons by log-rank test and hazard ratios estimated from Cox models. The distribution of BOR categories was also compared between first-line and second-line subgroups using chi-square or Fisher’s exact tests.

Nomogram construction and validation

Independent prognostic factors identified in multivariable Cox regression of the training cohort were incorporated into a final Cox model to construct the nomogram. Each predictor was assigned a weighted score proportional to its regression coefficient, with the largest absolute coefficient scaled to 100 points. The total points for each patient equaled the sum across predictors, and corresponding probabilities of 1-year and 2-year PFS and OS were derived.

Model performance was evaluated as follows: (1) Discrimination: assessed by time-dependent receiver operating characteristic (ROC) curves at 1 and 2 years, with area under the ROC curves (AUCs); (2) calibration: assessed by calibration plots comparing predicted versus observed survival probabilities; (3) risk stratification: patients were divided into high- and low-risk groups according to the median total points in the training cohort, and PFS/OS were compared with Kaplan–Meier curves and log-rank tests in both cohorts; (4) clinical utility: assessed by decision curve analysis (DCA) at 12 and 24 months, comparing the net benefit of the nomogram to “treat-all” and “treat-none” strategies; (5) clinical correlation: the relationship between nomogram-defined risk groups and BOR was examined using chi-square or Fisher’s exact tests.

Statistical analysis

Continuous variables were expressed as mean ± standard deviation (SD) and compared between groups using the Student’s t-test (for normally distributed data) or the Mann–Whitney U test (for non-normally distributed data, as appropriate). Categorical variables were summarized as counts and percentages, and compared using χ² or Fisher’s exact tests. Survival outcomes were analyzed by the Kaplan–Meier method and compared using the log-rank test. Hazard ratios with 95% CIs were estimated using univariate and multivariate Cox proportional hazards models.

All analyses were conducted using Python 3.10 (lifelines, scikit-learn, scikit-survival, matplotlib, and pandas). Two-sided p values <0.05 were considered statistically significant.

Results

Baseline characteristics

A total of 254 patients were included, with 177 assigned to the training cohort and 77 to the validation cohort (Figure 1). The mean age was 52.8 years, and the sex distribution was balanced. Most patients had stage IV disease, were never smokers, and had high PD-L1 expression (⩾50%). PD-L1 expression was assessed using the Dako 22C3 pharmDx antibody, and notably, no PD-L1–negative (0%) tumors were identified; all included patients exhibited PD-L1 TPS ⩾1%. Immunotherapy was administered predominantly in the first-line setting (72.1%) and more often in combination with chemotherapy (64.2%) than as monotherapy. ECOG performance status was 0–1 in over 90% of patients. The lung and bone were the most common metastatic sites, while liver and adrenal involvement were less frequent. Pembrolizumab was the most frequently used PD-1 inhibitor. Baseline tumor markers, biochemical parameters, and inflammatory indices were comparable between the two cohorts. Baseline tumor markers, biochemical parameters, and inflammatory indices were comparable between the two cohorts, and PD-L1 testing was performed uniformly across all centers using standardized protocols, minimizing selection bias. Overall, the training and validation groups were well balanced across clinical and laboratory characteristics (Table 1).

Table 1.

Characteristics of the patients in the training and validation cohort at baseline.

Characteristics	Total (n = 254)	Training (n = 177)	Validation (n = 77)
Age (mean ± SD, year)	52.78 ± 11.97	51.71 ± 11.99	55.25 ± 11.62
Sex
Male	123 (48.43%)	87	36
Female	131 (51.57%)	90	41
Clinical stage
IIIB	29 (11.42%)	15	14
IIIC	14 (5.51%)	10	4
IV	211 (83.07%)	152	59
Smoking history
Never smoker	187 (73.62%)	131	56
Current/former	67 (26.38%)	46	21
PD-L1 expression
1%–49%	90 (35.43%)	62	28
⩾50%	164 (64.57%)	115	49
Treatment line
First-line	183 (72.05%)	128	55
Second-line	71 (27.95%)	49	22
Treatment type
Monotherapy	91 (35.83%)	63	28
Combination therapy	163 (64.17%)	114	49
ECOG PS
0–1	233 (91.73%)	162	71
2	21 (8.27%)	15	6
Metastasis sites
Liver	40 (15.75%)	31	9
Lung	179 (70.47%)	127	52
Brain	7 (2.75%)	7	0
Bone	74 (29.13%)	51	23
Adrenal	37 (14.57%)	27	10
Drug
Pembrolizumab	184 (72.44%)	135	49
Nivolumab	27 (10.63%)	17	10
Sintilimab	19 (7.48%)	13	6
Tislelizumab	24 (9.45%)	12	12
CEA (ng/mL)	10.98 ± 16.54	10.03 ± 14.35	13.18 ± 20.65
NSE (ng/mL)	30.06 ± 25.73	30.16 ± 25.39	29.84 ± 26.68
CYFRA21-1 (ng/mL)	23.18 ± 35.73	24.48 ± 38.43	20.17 ± 28.59
CA19-9 (ng/mL)	48.49 ± 100.29	53.00 ± 116.05	38.11 ± 46.23
CA125 (ng/mL)	60.92 ± 104.89	60.26 ± 103.41	62.43 ± 108.93
ALB (g/L)	38.02 ± 4.54	38.14 ± 4.64	37.73 ± 4.32
LDH (U/L)	248.39 ± 197.03	250.38 ± 218.54	243.83 ± 136.35
CRP (mg/L)	31.84 ± 39.09	29.67 ± 34.57	35.64 ± 47.82
NLR	5.02 ± 3.06	4.78 ± 2.71	5.6 ± 3.69
PLR	221 ± 103.19	215.86 ± 100.14	232.81 ± 109.64
LMR	2.48 ± 1.18	2.52 ± 1.21	2.41 ± 1.11
SII	1640.04 ± 1283.20	1563.43 ± 1153.91	1816.15 ± 1533.90
dNLR	2.96 ± 1.78	2.8 ± 1.51	3.33 ± 2.24
nMLR	3.96 ± 3.66	3.89 ± 3.65	4.12 ± 3.69
NPR	0.02 ± 0.01	0.02 ± 0.01	0.03 ± 0.02

Continuous variables are expressed as the mean ± standard deviation. PD-L1 expression was assessed using the Dako 22C3 antibody. No PD-L1–negative (0%) tumors were identified; all included patients exhibited PD-L1 TPS ⩾1%.

ALB, albumin; CA125, carbohydrate antigen 125; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; CRP, C-reactive protein; CYFRA21-1, cytokeratin fragment 19; dNLR, derived neutrophil to lymphocyte ratio; ECOG PS, Eastern Cooperative Oncology Group performance status; LDH, lactate dehydrogenase; LMR, lymphocyte-to-monocyte ratio; NLR, neutrophil–lymphocyte ratio; nMLR, neutrophil-monocyte-to-lymphocyte ratio; NPR, neutrophil-to-platelet ratio; NSE, neuron-specific enolase; PD-L1, programmed death ligand-1; PLR, platelet–lymphocyte ratio; SII, systemic immune inflammation index.

Construction of the BRTS

To integrate baseline laboratory indicators into a unified prognostic index, we applied LASSO-Cox regression with PFS as the outcome (Table S1). Several variables were retained with non-zero coefficients, including CEA, CA19-9, CA125, ALB, CRP, monocyte count, and PLR. Among these, elevated CRP (β = 0.676, HR 1.97, p < 0.001), CEA (β = 0.286, HR 1.33, p = 0.004), and PLR (β = 0.264, HR 1.30, p = 0.029) were significantly associated with increased risk, whereas higher ALB (β = −0.290, HR 0.75, p = 0.002) and monocyte count (β = −0.392, HR 0.68, p < 0.001) were protective factors. Other variables such as NSE, LDH, CA19-9, and CA125 showed borderline or non-significant associations but were retained with small positive coefficients. Variables with coefficients shrunk to zero (e.g., CYFRA21-1, WBC, NLR, SII, dNLR, nMLR) were excluded by penalization (Figure 2 and Table S1).

Figure 2.

Construction of the BRTS. (a) LASSO-Cox coefficient profiles of baseline laboratory indicators across a range of penalty parameters (λ). Each line represents the trajectory of a variable’s coefficient as λ changes. (b) Ten-fold cross-validation curve for the LASSO-Cox model. The red dashed line indicates the optimal λ (0.0386) corresponding to the best C-index. (c) Aggregated variable-level importance of features contributing to the BRTS. Tumor markers, blood routine indices, and other parameters are grouped by category. (d) Feature-level standardized coefficients (β) and HRs of variables included in the final BRTS. Positive coefficients represent risk factors, whereas negative coefficients indicate protective effects.

The resulting risk score, termed the Blood Routine & Tumor-Marker Score (BRTS), was computed as the weighted sum of standardized predictors:

\begin{array}{l} B R T S = + 0.676 Z_{l o g (1 + C R P)} + 0.286 Z_{l o g (1 + C E A)} + \\ 0.264 Z_{l o g (1 + P L R)} + 0.214 Z_{l o g (1 + C A 125)} + \\ 0.167 Z_{l o g (1 + C A 199)} + 0.143 Z_{B A S} + \\ 0.079 Z_{l o g (1 + L D H)} + 0.060 Z_{l o g (1 + N S E)} - \\ 0.392 Z_{M O N} - 0.290 Z_{A L B} - 0.146 Z_{L E U} - \\ 0.101 Z_{E O S} + 0.000093 Z_{l o g (1 + C Y F R A 21 - 1)} + \\ (\begin{array}{l} o t h e r t e r m s w i t h c o e f f i c i e n t s \\ n e a r 0, ~ 1 0^{- 7} t o 1 0^{- 6} \end{array}) \end{array}

where $Z_{X} = \frac{X - μ x}{σ x}$ denotes the standardized value of variable $X,$ and $Z_{\log (1 + X)}$ denotes the standardized value of $\log (1 + X) .$ Means $(μ)$ and standard deviations $(σ)$ were derived from the training cohort. The residual term ε corresponds to negligible coefficients close to zero (10⁻⁷–10⁻⁶) and can be ignored in practical applications. No intercept term was included, consistent with the Cox model framework.

Higher BRTS values consistently indicated worse prognosis, and patients were subsequently stratified into high-BRTS and low-BRTS groups using the median score of the training cohort as the cutoff. Because BRTS was derived exclusively from pre-treatment laboratory values, the present analysis supports its role as a baseline risk stratification tool. Nevertheless, since all included variables are routinely monitored, the feasibility of recalculating BRTS dynamically during therapy remains an important direction for future studies.

Prognostic value of BRTS for PFS and OS

In the training cohort, BRTS stratification effectively separated patients with distinct outcomes. High-BRTS patients had significantly shorter PFS (HR = 4.59, 95% CI: 3.25–6.47, p < 0.001; Figure 3(a)) and OS (HR = 6.86, 95% CI: 4.60–10.23, p < 0.001; Figure 3(c)) compared with low-BRTS patients. Consistent results were observed in the validation cohort, where high-BRTS patients showed markedly worse PFS (HR = 5.37, 95% CI: 2.89–9.97, p < 0.001; Figure 3(b)) and OS (HR = 3.87, 95% CI: 2.27–6.61, p < 0.001; Figure 3(d)). These findings demonstrate the robustness and reproducibility of BRTS as a prognostic marker across cohorts. Notably, the prognostic separation achieved by BRTS was stronger and more consistent than that observed for PD-L1 expression alone, suggesting that BRTS may provide complementary value rather than replace PD-L1 in guiding immunotherapy decisions.

Figure 3.

Kaplan–Meier survival curves stratified by BRTS in the training and validation cohorts. (a) PFS in the training cohort. (b) PFS in the validation cohort. (c) OS in the training cohort. (d) OS in the validation cohort.

Prognostic factors for PFS

Univariate Cox regression analysis in the training cohort identified several factors significantly associated with PFS, including PD-L1 expression (HR 1.52, 95% CI 1.11–2.08, p = 0.009), treatment line (second-line vs first-line, HR 17.84, 95% CI 10.67–29.83, p < 0.001), treatment regimen (monotherapy vs combination, HR 1.91, 95% CI 1.38–2.64, p < 0.001), liver metastasis (HR 2.08, 95% CI 1.40–3.09, p < 0.001), and BRTS (high vs low, HR 3.11, 95% CI 2.21–4.38, p < 0.001).

In multivariate analysis, treatment line (HR 16.53, 95% CI 10.09–27.07, p < 0.001), treatment regimen (HR 1.48, 95% CI 1.00–2.18, p = 0.05), liver metastasis (HR 2.07, 95% CI 1.37–3.13, p < 0.001), and BRTS (high vs low, HR 3.22, 95% CI 2.27–4.56, p < 0.001) remained independent prognostic factors for shorter PFS. PD-L1 expression, although significant in univariate analysis, did not retain statistical significance after adjustment (Table 2).

Table 2.

Univariate and multivariate Cox regression analyses for PFS in the training cohort.

Variable	Univariate analysis		Multivariate analysis
Variable	HR (95% CI)	p Value	HR (95% CI)	p Value
Age (>60 vs ⩽60, years)	0.86 (0.58–1.27)	0.439
Sex (male vs female)	0.80 (0.59–1.08)	0.153
Smoking (yes vs no)	0.83 (0.59–1.17)	0.288
PD-L1 (high vs low)	1.52 (1.11–2.08)	0.009	0.90 (0.63–1.29)	0.563
c-Stage IIIC
IIIC vs IIIB	1.27 (0.53–3.03)	0.591
IV vs IIIB	1.66 (0.90–3.07)	0.105
Treatment line
Second line vs first line	17.84 (10.67–29.83)	<0.001	16.53 (10.09–27.07)	<0.001
Treatment regimen
Monotherapy vs combination	1.91 (1.38–2.64)	<0.001	1.48 (1.00–2.18)	0.05
Liver metastasis (yes vs no)	2.08 (1.40–3.09)	<0.001	2.07 (1.37–3.13)	<0.001
Bone metastasis	1.02 (0.73–1.42)	0.913
BRTS (high vs low)	3.11 (2.21–4.38)	<0.001	3.22 (2.27–4.56)	<0.001

BRTS, Blood Routine & Tumor-Marker Score; HR, hazard ratio; PD-L1, programmed death-ligand 1; PFS, progression‑free survival.

Prognostic factors for OS

In the training cohort, univariate Cox analysis identified several variables significantly associated with OS, including sex (male vs female, HR 0.72, 95% CI 0.52–0.98, p = 0.038), smoking history (yes vs no, HR 0.64, 95% CI 0.44–0.92, p = 0.017), PD-L1 expression (high vs low, HR 1.43, 95% CI 1.03–1.98, p = 0.031), treatment line (second line vs first line, HR 14.55, 95% CI 8.88–23.85, p < 0.001), treatment regimen (monotherapy vs combination, HR 1.79, 95% CI 1.29–2.48, p < 0.001), liver metastasis (HR 2.09, 95% CI 1.41–3.11, p < 0.001), and BRTS (high vs low, HR 3.90, 95% CI 2.70–5.63, p < 0.001).

In multivariate analysis, smoking history (HR 0.52, 95% CI 0.34–0.80, p = 0.003), treatment line (HR 19.98, 95% CI 11.50–34.71, p < 0.001), treatment regimen (HR 1.93, 95% CI 1.32–2.82, p < 0.001), liver metastasis (HR 2.41, 95% CI 1.60–3.63, p < 0.001), and BRTS (HR 4.30, 95% CI 2.98–6.19, p < 0.001) emerged as independent prognostic factors for OS. Notably, PD-L1 expression and sex, which were significant in univariate analysis, lost statistical significance after adjustment (Table 3).

Table 3.

Univariate and multivariate Cox regression analyses for OS in the training cohort.

Variable	Univariate analysis		Multivariate analysis
Variable	HR (95% CI)	p Value	HR (95% CI)	p Value
Age (>60 vs ⩽60, years)	0.87 (0.58–1.32)	0.522
Sex (male vs female)	0.72 (0.52–0.98)	0.038	1.00 (0.71–1.42)	0.981
Smoking (yes vs no)	0.64 (0.44–0.92)	0.017	0.52 (0.34–0.80)	0.003
PD-L1 (high vs low)	1.43 (1.03–1.98)	0.031	0.73 (0.47–1.12)	0.145
c-Stage IIIC
IIIC vs IIIB	1.42 (0.56–3.60)	0.462
IV vs IIIB	1.42 (0.75–2.70)	0.283
Treatment line
Second line vs First line	14.55 (8.88–23.85)	<0.001	19.98 (11.50–34.71)	<0.001
Treatment regimen
Monotherapy vs combination	1.79 (1.29–2.48)	<0.001	1.93 (1.32–2.82)	<0.001
Liver metastasis (yes vs no)	2.09 (1.41–3.11)	<0.001	2.41 (1.60–3.63)	<0.001
Bone metastasis	0.98 (0.69–1.39)	0.914
BRTS (high vs low)	3.90 (2.70–5.63)	<0.001	4.30 (2.98–6.19)	<0.001

BRTS, Blood Routine & Tumor-Marker Score; HR, hazard ratio; OS, overall survival; PD-L1, programmed death-ligand 1.

Model validation and sensitivity analyses

Proportional hazards assumptions were satisfied for all Cox models, with no significant violations detected (all p > 0.05; Table S2). Bootstrap resampling with 1000 iterations yielded optimism-corrected C-indices of 0.787 for PFS and 0.797 for OS, confirming strong model discrimination (Table S3). Exploratory stratified Cox analysis by hospital site revealed no significant inter-center heterogeneity (p = 0.41; Table S4), despite the majority of cases originating from the Third Affiliated Hospital of Sun Yat-sen University and the General Hospital of the Southern Theater Command. These results indicate consistent model performance across institutions.

Nomogram construction and validation

A prognostic nomogram was developed to predict PFS based on independent risk factors (treatment line, regimen, liver metastasis, and BRTS group), achieving good discrimination (C-index = 0.787, Figure 4(a)). For OS, a separate nomogram incorporating treatment line, regimen, liver metastasis, BRTS group, and smoking status showed similar performance (C-index = 0.797, Figure 4(f)). Calibration curves demonstrated excellent agreement between predicted and observed 1- and 2-year survival probabilities in the training cohorts and acceptable consistency in the validation cohorts (Figure 4(b)–(e)). DCA further confirmed that both nomograms provided greater net clinical benefit than “treat-all” or “treat-none” strategies (Figures S1 and S2). This advantage was most pronounced for 1-year predictions of PFS and OS, and although attenuated for 2-year PFS, remained consistently favorable for 2-year OS, supporting their robustness and clinical utility for individualized survival estimation.

Figure 4.

Construction and validation of nomograms for predicting PFS and OS. (a) Nomogram for PFS incorporating treatment line, treatment regimen, liver metastasis, and BRTS. (b) Calibration plot of the PFS nomogram in the training cohort at 1 and 2 years. (c) Calibration plot of the PFS nomogram in the validation cohort at 1 and 2 years. (d) Calibration plot of the OS nomogram in the training cohort at 1 and 2 years. (e) Calibration plot of the OS nomogram in the validation cohort at 1 and 2 years. (f) Nomogram for OS prediction integrating treatment line, treatment regimen, liver metastasis, BRTS, and smoking status.

Nomogram-based risk stratification for PFS and OS

Risk stratification using the nomogram-derived total points effectively separated patients into high- and low-risk groups in both the training and validation cohorts. For PFS, high-risk patients exhibited significantly shorter survival compared with low-risk patients, with an HR of 5.16 (95% CI: 3.60–7.40, p < 0.001) in the training cohort and 7.34 (95% CI: 4.07–13.24, p < 0.001) in the validation cohort. Time-dependent ROC analysis demonstrated strong predictive accuracy, with AUCs of 0.875 and 0.797 for 1- and 2-year PFS in the training cohort, and 0.908 and 0.780 in the validation cohort, respectively. BOR distributions further supported the discriminative ability of the PFS-nomogram, as low-risk patients had higher rates of PR and fewer PD cases, while high-risk patients were enriched for PD (p < 0.001 in both cohorts, Figure 5).

Figure 5.

Prognostic performance and clinical relevance of the PFS-nomogram model in the training and validation cohorts. (a) Kaplan–Meier survival curves of PFS stratified by the nomogram-derived risk groups in the training cohort. (b) Kaplan–Meier survival curves of PFS stratified by nomogram risk groups in the validation cohort. (c) Time-dependent ROC curves for 1- and 2-year PFS prediction in the training cohort. (d) Time-dependent ROC curves for 1- and 2-year PFS prediction in the validation cohort. (e) Distribution of BOR by nomogram-defined risk groups in the training cohort. (f) Distribution of BOR in the validation cohort.

For OS, the nomogram similarly achieved clear separation between risk groups. High-risk patients had markedly worse OS than low-risk patients, with HRs of 8.58 (95% CI: 5.47–13.44, p < 0.001) in the training cohort and 8.68 (95% CI: 4.63–16.24, p < 0.001) in the validation cohort. ROC analysis confirmed excellent predictive performance, with AUCs of 0.910 and 0.861 for 1- and 2-year OS in the training cohort, and 0.921 and 0.853 in the validation cohort. BOR distributions again demonstrated significant differences between risk groups (p < 0.001), with low-risk patients showing higher PR rates and high-risk patients demonstrating increased PD rates across both cohorts (Figure 6).

Figure 6.

Prognostic performance and clinical relevance of the OS-nomogram model in the training and validation cohorts. (a) Kaplan–Meier survival curves of OS stratified by the nomogram-derived risk groups in the training cohort. (b) Kaplan–Meier survival curves of OS stratified by nomogram risk groups in the validation cohort. (c) Time-dependent ROC curves for 1- and 2-year OS prediction in the training cohort. (d) Time-dependent ROC curves for 1- and 2-year OS prediction in the validation cohort. (e) Distribution of BOR by nomogram-defined risk groups in the training cohort. (f) Distribution of BOR in the validation cohort.

Prognostic value of BRTS in first- and second-line treatment subgroups

In the first-line treatment subgroup, patients with a high BRTS had significantly shorter PFS and OS compared with those with a low BRTS (Figure S3(A) and (B), log-rank p < 0.001), indicating the strong prognostic value of BRTS. Similarly, in the second-line treatment subgroup, high BRTS was associated with significantly worse PFS (p = 0.005) and OS (p < 0.001), confirming its prognostic relevance across treatment settings (Figure S3(C) and (D)). Consistent with these findings, BOR distributions differed significantly by BRTS level in both first-line (p = 0.039) and second-line (p = 0.029) subgroups (Figure S3(E) and (F)). Patients with low BRTS exhibited higher proportions of PR, whereas high BRTS was associated with increased rates of SD and PD.

Baseline characteristics between the first-line and second-line cohorts are summarized in Table S5. As expected, second-line patients showed higher systemic inflammation and tumor burden (e.g., elevated CRP, NLR, SII, and lower albumin; all p < 0.001), whereas most demographic and clinical parameters—including age, sex, stage distribution, and CA125 levels—were comparable between groups. Importantly, BRTS retained its prognostic significance after adjustment for treatment line, regimen, and metastatic profile in multivariate analyses, underscoring its independence from baseline imbalance. These results collectively support the robustness of BRTS across different therapeutic contexts.

Discussion

In this multicenter retrospective study, we demonstrated that baseline hematologic indices and STMs can be integrated into a composite BRTS that robustly stratifies immunotherapy outcomes in patients with pLELC. Patients with high BRTS had consistently shorter PFS and OS, independent of treatment line, regimen, and metastatic profile. The score was reproducible across training and validation cohorts and retained prognostic value in both first- and second-line subgroups. When incorporated into a nomogram together with clinical variables, BRTS improved discriminatory accuracy and net clinical benefit, underscoring its potential to guide individualized treatment strategies in this rare NSCLC subtype. Importantly, the BRTS framework could be readily implemented in routine oncology practice, as it relies solely on standard laboratory assays, allowing clinicians to stratify patients before initiating immunotherapy and to tailor follow-up or combination strategies accordingly.

Beyond its clinical performance, the present study also addressed key methodological rigor points often lacking in prior biomarker reports. Specifically, variable selection was performed using a Python-based LASSO-Cox model with 10-fold cross-validation and the 1-standard-error rule to avoid overfitting. The BRTS cutoff was discovery-locked based on PFS and applied unchanged to OS analyses, ensuring independence between endpoints. Proportional hazards assumptions were systematically verified with Schoenfeld residuals, and no violations were detected. Internal validation was conducted through 1000-bootstrap resampling to obtain optimism-corrected C-indices, while a stratified Cox model assessed potential center effects, confirming model consistency across hospitals. These quality-control steps collectively enhance the robustness, reproducibility, and credibility of our findings.

To further address potential confounding by treatment line, we conducted stratified analyses and a comprehensive baseline comparison between first-line and second-line cohorts (Table S5). Although second-line patients exhibited higher inflammatory indices (CRP, NLR, and SII) and lower albumin, BRTS remained significantly associated with both PFS and OS within each subgroup (Figure S3), and retained its independent prognostic value after multivariable adjustment. These findings reinforce that BRTS captures intrinsic host–tumor interactions rather than merely reflecting treatment intensity or clinical baseline imbalances.

Our findings are consistent with prior evidence that systemic inflammation and nutritional status influence immunotherapy benefit in lung cancer.^21,22 Elevated CRP, PLR, and CEA emerged as risk factors, while albumin and monocyte count were protective, reflecting the interplay between host systemic status and tumor–immune interactions in pLELC despite its distinct EBV-related biology.^12,18 The incorporation of tumor markers such as CEA, CA125, and CA19-9 is particularly notable, as these are commonly measured but seldom investigated in the context of immunotherapy.^17,18 Their retention in the final model suggests that they provide non-redundant information on tumor burden and microenvironmental modulation. The observation that higher circulating monocyte count correlated with better outcomes, while counterintuitive compared with some NSCLC studies, may reflect unique features of pLELC immune ecology, such as monocyte–dendritic differentiation or EBV-driven immune modulation, but requires further validation.^23,24

Beyond its statistical robustness, the biological plausibility of BRTS is further supported by recent evidence from a large pan-cancer analysis published in Nature Medicine, which demonstrated that hematologic and inflammatory biomarkers derived from routine blood tests—such as CRP, NLR, and PLR—strongly correlated with clinical outcomes of immune checkpoint inhibitor therapy across tumor types.¹⁶ These findings provide indirect yet compelling validation for our blood-based model, reinforcing the rationale that easily accessible laboratory indicators can capture the host–tumor immune balance and thereby predict immunotherapy efficacy.¹⁶ Collectively, both internal and external evidence point toward the translational reliability of hematology-based predictors such as BRTS.

This work also extends previous efforts to develop hematology-based models for pLELC. A recent single-center study proposed a HIBS derived solely from routine blood counts and biochemistry, which showed predictive value for PFS and response.¹⁹ Direct head-to-head comparison between BRTS and HIBS was not feasible in our study because raw patient-level data from the HIBS cohort were unavailable, and the feature spaces of the two models differed substantially.¹⁹ Direct head-to-head comparison between BRTS and HIBS was not feasible because raw patient-level data from the HIBS cohort were unavailable, and the variables constituting the two models differed substantially. Nevertheless, by including STMs and overall survival as endpoints, our study complements and extends the earlier work, providing additional clinical applicability. Future multi-institutional collaborations may enable external benchmarking between such models. Interestingly, we found liver metastasis to be a stronger adverse prognostic factor than bone metastasis, which was highlighted in the earlier study.^25–27 This discrepancy may reflect differences in metastatic spectra, treatment regimens, and cohort composition, and highlights the need for further prospective confirmation. Notably, PD-L1 expression, while associated with outcomes in univariate analyses and in prior reports, did not retain significance after adjustment, echoing broader NSCLC literature where PD-L1 provides imperfect predictive resolution.²⁸ Our findings thus suggest that BRTS serves as a complementary rather than competing biomarker to PD-L1, potentially refining patient stratification when the two markers are discordant.

Regarding the timing of BRTS assessment, our model was based on baseline laboratory data before ICI initiation. However, because all variables are routinely collected, BRTS could theoretically be recalculated during therapy to capture dynamic changes. Whether such on-treatment monitoring improves prediction of response or resistance remains to be tested, and future studies with serial sampling should investigate its potential to guide treatment adjustment in real time.

Another important limitation relates to generalizability. All participating centers were in China, where pLELC prevalence is highest, and EBV association is well recognized.^12,29,30 The prognostic value of CA125, CA19-9, and related markers may differ in Western populations with different genetic backgrounds and viral epidemiology. Thus, BRTS should not be immediately generalized to global practice without validation in diverse cohorts. Prospective, international studies will be needed to confirm its utility across different healthcare settings.

From a clinical perspective, the implications of BRTS are straightforward: patients with high BRTS, particularly those with liver metastasis, may not be ideal candidates for ICI monotherapy and should be considered for intensified approaches such as chemoimmunotherapy, antiangiogenic combinations, or clinical trial enrollment. Conversely, patients with low BRTS appear more likely to derive durable benefit from immunotherapy. Beyond prognostic stratification, BRTS could significantly modify clinical practice by enabling pre-treatment risk assessment, facilitating early identification of poor responders, and guiding individualized follow-up intensity. Because it is derived from universally available laboratory parameters, BRTS can be seamlessly incorporated into routine workflows without additional cost or testing burden, providing a practical tool for real-world decision-making. The nomograms generated in this study provide individualized estimates of survival probabilities, which may support shared decision-making and follow-up planning in real-world practice.

The strengths of our study include its multicenter design, relatively large sample size for this rare disease, and rigorous statistical approach using LASSO penalization, internal validation, and DCA. Nevertheless, several limitations warrant consideration. First, the retrospective nature introduces potential selection bias, and residual confounding cannot be excluded. Second, although multicenter, the sample size remained modest for rare subtypes such as pLELC, and most cases originated from two high-volume centers; while we performed stratified Cox analyses confirming no significant inter-center heterogeneity, this warrants prospective verification. Third, laboratory assays, while standardized within centers, may vary across institutions, and cutoff values may not be universally transferable. Fourth, missing values were handled through imputation rather than exclusion, minimizing bias but potentially introducing minor estimation uncertainty. Fifth, to mitigate potential bias from missing data, EBV-DNA results—available in only a subset of patients—were excluded from model construction. We acknowledge that EBV-DNA load has been associated with prognosis in prior reports, and its absence represents a limitation of the current work. Future prospective studies with systematic EBV-DNA collection at baseline and during treatment will be crucial to explore its integration with BRTS for more comprehensive risk stratification. Finally, our findings require external validation in prospective cohorts before clinical adoption.

Future work should focus on prospective and multi-regional validation of BRTS, integration with molecular and imaging biomarkers such as PD-L1, EBV-DNA, ctDNA, and radiomics signatures, and assessment of dynamic changes during treatment to capture emerging resistance. Mechanistic studies dissecting the roles of monocytes and platelet-mediated immune regulation may yield biological insights and therapeutic opportunities. Taken together, our results support the use of BRTS as a pragmatic, blood-based predictor that complements existing biomarkers and may contribute to individualized immunotherapy strategies in pLELC and related rare NSCLC subtypes.

Conclusion

This multicenter study demonstrates that a composite BRTS provides robust prognostic stratification for patients with pLELC receiving immunotherapy. High BRTS was independently associated with inferior PFS and OS, and incorporation of BRTS into nomograms alongside clinical factors significantly improved predictive performance and clinical utility. Given its reliance on inexpensive and widely available laboratory tests, BRTS represents a practical and accessible tool to complement existing biomarkers and support individualized treatment decisions.

Supplemental Material

sj-docx-1-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-docx-1-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-docx-2-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-docx-2-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-docx-3-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-docx-3-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-docx-4-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-docx-4-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-docx-8-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-docx-8-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-pdf-5-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-pdf-5-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-pdf-6-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-pdf-6-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Supplemental Material

sj-pdf-7-tam-10.1177_17588359251403409 – Supplemental material for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study

Supplemental material, sj-pdf-7-tam-10.1177_17588359251403409 for Prediction of immunotherapy efficacy for pulmonary lymphoepithelioma-like carcinoma using baseline routine blood tests and serum tumor markers: a multicenter retrospective study by Xiongwen Yang, Yuanwei Liang, Hao Hu, Yubin Zhou, Huiyin Deng, Jian Huang, Maoli Liang, Zihao Yuan, Longyan Dong and Yi Xiao in Therapeutic Advances in Medical Oncology

Footnotes

Acknowledgements

The authors thank all patients and staff of the participating centers for their contributions to this multicenter study.

Declarations

ORCID iD

Xiongwen Yang

Supplemental material

Supplemental material for this article is available online.

References

Gandhi

Rodríguez-Abreu

Gadgeel

, et al. Pembrolizumab plus chemotherapy in metastatic non-small-cell lung cancer. N Engl J Med 2018; 378: 2078–2092.

Paz-Ares

Luft

Vicente

, et al. Pembrolizumab plus chemotherapy for squamous non-small-cell lung cancer. N Engl J Med 2018; 379: 2040–2051.

Gadgeel

Rodríguez-Abreu

Speranza

, et al. Updated analysis from KEYNOTE-189: pembrolizumab or placebo plus pemetrexed and platinum for previously untreated metastatic nonsquamous non-small-cell lung cancer. J Clin Oncol 2020; 38: 1505–1517.

Rodríguez-Abreu

Powell

Hochmair

, et al. Pemetrexed plus platinum with or without pembrolizumab in patients with previously untreated metastatic nonsquamous NSCLC: protocol-specified final analysis from KEYNOTE-189. Ann Oncol 2021; 32: 881–895.

Zhou

Zhang

Ren

, et al. Quantitative and dynamic ctDNA as a biomarker of response and survival in patients with advanced lung squamous cell carcinoma receiving immunochemotherapy or chemotherapy alone. J Thorac Oncol 2025; 20: 1475–1488.

Travis

Brambilla

Nicholson

, et al. The 2015 World Health Organization classification of lung tumors: impact of genetic, clinical and radiologic advances since the 2004 classification. J Thorac Oncol 2015; 10: 1243–1260.

Gómez-Román

Martínez

Fernández

, et al. Epstein-Barr virus-associated adenocarcinomas and squamous-cell lung carcinomas. Mod Pathol 2009; 22: 530–537.

Hong

Liu

Luo

, et al. The genomic landscape of Epstein-Barr virus-associated pulmonary lymphoepithelioma-like carcinoma. Nat Commun 2019; 10: 3108.

Chen

Zhang

Dai

, et al. Molecular characteristics of primary pulmonary lymphoepithelioma-like carcinoma based on integrated genomic analyses. Signal Transduct Target Ther 2021; 6: 6.

10.

Hong

H-Z

J-K

Zhang

J-T

, et al. Neoadjuvant immunotherapy in patients with pulmonary lymphoepithelioma-like carcinoma. Lung Cancer 2023; 181: 107220.

11.

Xiao

Luo

, et al. Comparison of immunotherapy, chemotherapy, and chemoimmunotherapy in advanced pulmonary lymphoepithelioma-like carcinoma: a retrospective study. Front Oncol 2022; 12: 820302.

12.

Yang

Xiao

Zhou

, et al. Efficacy and safety of immunotherapy in locally advanced or metastatic pulmonary lymphoepithelioma-like carcinoma: a multicenter retrospective study. Ther Adv Med Oncol 2025; 17: 17588359251316099.

13.

Zhou

Huang

Lan

, et al. Comparison of first-line immunotherapy efficacy between advanced lung squamous cell carcinoma and pulmonary lymphoepithelioma-like carcinoma: a propensity score matching multicenter study. J Cancer Res Ther 2023; 19: 1011–1018.

14.

Zhang

Zhou

Chen

, et al. PD-1 inhibition plus platinum-based chemotherapy (PBC) or PBC alone in the first-line treatment of locally advanced or metastatic pulmonary lymphoepithelioma-like carcinoma. Front Immunol 2022; 13: 1015444.

15.

Pang

Liao

Huang

, et al. Exploration of immunotherapy in advanced pulmonary lymphoepithelioma-like carcinoma. Int J Cancer 2023; 152: 2338–2350.

16.

Yoo

Fitzgerald

Cho

, et al. Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data. Nat Med 2025; 31: 869–880.

17.

Yang

Xiao

Zhou

, et al. Dynamic monitoring of serum tumor markers as prognostic factors in patients with advanced non-small-cell lung cancer treated with first-line immunotherapy: a multicenter retrospective study. Ther Adv Med Oncol 2023; 15: 17588359231206282.

18.

Huang

Xiao

Zhou

, et al. Baseline serum tumor markers predict the survival of patients with advanced non-small cell lung cancer receiving first-line immunotherapy: a multicenter retrospective study. BMC Cancer 2023; 23: 812.

19.

Sheng

Chen

, et al. Development of a haematological indices-based nomogram for prognostic prediction and immunotherapy response assessment in primary pulmonary lymphoepithelioma-like carcinoma patients. Transl Lung Cancer Res 2024; 13: 453–464.

20.

Collins

Moons

KGM

Dhiman

, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 2024; 385: e078378.

21.

Golonko

Pienkowski

Swislocka

, et al. Dietary factors and their influence on immunotherapy strategies in oncology: a comprehensive review. Cell Death Dis 2024; 15: 254.

22.

Stares

Ding

Stratton

, et al. Biomarkers of systemic inflammation predict survival with first-line immune checkpoint inhibitors in non-small-cell lung cancer. ESMO Open 2022; 7: 100445.

23.

Kim

Axelsson

Moll

, et al. Associations of monocyte count and other immune cell types with interstitial lung abnormalities. Am J Respir Crit Care Med 2022; 205: 795–805.

24.

Huangfu

Chang

Zhang

, et al. Monocytes-to-lymphocytes ratio increases the prognostic value of circulating tumor cells in non-small cell lung cancer: a prospective study. Transl Cancer Res 2024; 13: 3589–3598.

25.

Yang

, et al. Prevalence and risk factors of bone metastasis and the development of bone metastatic prognostic classification system: a pan-cancer population study. Aging (Albany NY) 2023; 15: 13134–13149.

26.

Zhang

Xiang

Shao

, et al. Different risk and prognostic factors for liver metastasis of breast cancer patients with de novo and relapsed distant metastasis in a Chinese population. Front Oncol 2023; 13: 1102853.

27.

Ming

Zhao

, et al. Assessing the relationship between liver metastases and the survival of patients with non-small cell lung cancer after immune checkpoint inhibitors treatment: a systematic review and meta-analysis. Integr Cancer Ther 2023; 22: 15347354231164584.

28.

Passiglia

Bronte

Bazan

, et al. PD-L1 expression as predictive biomarker in patients with NSCLC: a pooled analysis. Oncotarget 2016; 7: 19738–19747.

29.

Zhang

Wang

, et al. Genomic landscapes of Epstein-Barr virus in pulmonary lymphoepithelioma-like carcinoma. J Virol 2022; 96: e0169321.

30.

Ngan

Yip

Cheng

, et al. Circulating Epstein-Barr virus DNA in serum of patients with lymphoepithelioma-like carcinoma of the lung: a potential surrogate marker for monitoring disease. Clin Cancer Res 2002; 8: 986–994.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB

5.17 MB

0.30 MB

2.77 MB