Prognostic nomogram for extensive-stage small-cell lung cancer in the immunotherapy era: a multicenter study of first-line platinum-etoposide with or without a PD-1/PD-L1 inhibitor

Introduction

Small cell lung cancer (SCLC) accounts for approximately 13–15% of lung cancers and is characterized by rapid progression, early dissemination and high chemosensitivity followed by early relapse (1). Approximately two-thirds of patients present with extensive-stage small-cell lung cancer (ES-SCLC) and have a dismal prognosis: despite response rates exceeding 60% with platinum–etoposide (EP) chemotherapy, median overall survival (OS) has long remained around 10–12 months and 2-year survival below 10% (2).

Programmed death-ligand 1 (PD-L1) inhibitor-based chemoimmunotherapy has modestly improved outcomes. In IMpower133 and CASPIAN, adding atezolizumab or durvalumab to EP produced statistically significant and clinically meaningful OS benefits, and EP plus a programmed cell death protein 1/PD-L1 (PD-1/PD-L1) inhibitor has become a standard first-line regimen for eligible patients with ES-SCLC (3-7). However, despite the introduction of chemoimmunotherapy, most patients with ES-SCLC still experience early progression, whereas only a minority achieve more durable benefit. This clinically meaningful variability in outcomes among treatment-eligible patients underscores the need for improved baseline risk stratification (8).

In addition to age, performance status and metastatic burden, numerous studies have shown that systemic inflammation and nutritional status strongly influence outcomes in SCLC. Hematologic indices derived from routine tests—including the neutrophil-to-lymphocyte ratio (NLR) (9-12), platelet-to-lymphocyte ratio (PLR), the systemic immune-inflammation index (SII), prognostic nutritional index (PNI) (13), lactate dehydrogenase (LDH), neuron-specific enolase (NSE), and the lung immune prognostic index (LIPI) (14,15)—have been shown to be repeatedly associated with prognosis. These markers are inexpensive and widely available, making them attractive candidates for incorporation into clinical prediction tools. Several prognostic scores and nomograms have been proposed (16-18), including a six-variable model derived from an immunotherapy cohort, and inflammation- or nutrition-based scores.

However, there are important limitations with existing models. Most were developed with small, single-center cohorts from the chemotherapy era, lacking robust external validation and therefore risk overfitting (19-21). Few explicitly incorporate chemoimmunotherapy as a covariate, and most capture only a narrow subset of inflammatory or clinical variables, without jointly considering detailed metastatic topography, systemic inflammatory burden and treatment modality. As a result, their applicability to contemporary ES-SCLC patients treated with PD-1/PD-L1 inhibitor-based regimens is uncertain.

Multicenter retrospective studies conducted in routine clinical practice can complement randomized clinical trials (RCTs) by capturing variations in baseline characteristics and treatment patterns outside strictly controlled trial settings (22). In routine practice, ES-SCLC patients are older, more comorbid and receive heterogeneous therapies, with variable uptake of chemoimmunotherapy across centers. Developing and validating a prognostic model for the immunotherapy era based on a multicenter retrospective cohort could therefore provide more generalizable evidence to support risk-adapted management.

In this study, we analyzed 605 treatment-eligible patients with ES-SCLC treated with first-line EP with or without PD-1/PD-L1 inhibitors at seven tertiary hospitals in China. Using a training/validation design, we constructed a nine-variable Cox model integrating metastatic sites, malignant pleural effusion (MPE), SII, NSE, LIPI and treatment modality, and presented it as a nomogram and integer-based risk score. We assessed discrimination, calibration and clinical net benefit and compared its performance head-to-head with several representative published models. Our aim was to provide a simple, robust and clinically feasible tool for baseline prognostic assessment in treatment-eligible patients with ES-SCLC in the chemoimmunotherapy era. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1445/rc).

Methods

Study design and patient population

This was a multicenter retrospective cohort study. We screened consecutive patients who were first diagnosed with ES-SCLC and who received first-line platinum–etoposide chemotherapy with or without a PD-1/PD-L1 inhibitors after the diagnosis of ES-SCLC between 1 January 2019 and 1 June 2025 at seven tertiary hospitals in China: Tianjin Chest Hospital, Tianjin Haihen Hospital, Tianjin Fourth Central Hospital, Tianjin Fifth Central Hospital, Cangzhou Hospital of Integrated Traditional Chinese and Western Medicine, Fujian Provincial Hospital and Qingdao Municipal Hospital, The overall study design and patient selection process are illustrated in Figure 1.

Figure 1 Study flowchart. The figure illustrates the screening and inclusion process for patients with ES-SCLC across seven participating centers. The numbers of patients excluded at each step, together with the corresponding reasons (e.g., non-ES-SCLC histology, prior malignancies, inadequate baseline laboratory data, incomplete follow-up), are shown. The final study population (n=605) was randomly divided into a training cohort (n=421) and a validation cohort (n=184). ES-SCLC, extensive-stage small-cell lung cancer; ICI, immune checkpoint inhibitor; SCLC, small-cell lung cancer.

Eligible patients had to meet all of the following criteria: (I) histologically or cytologically confirmed SCLC, with initial staging consistent with extensive disease according to the Veterans Administration Lung Study Group classification (23); (II) receipt of first-line platinum-etoposide (EP) with or without a PD-1/PD-L1 inhibitor as the initial systemic treatment for ES-SCLC, with at least one cycle administered; (III) complete baseline hematological and biochemical tests performed at the participating center within 7 days before the start of systemic therapy, allowing calculation of key inflammatory and nutritional indices; (IV) at least one measurable lesion at baseline and tumor response evaluated according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 during follow-up (24); (V) available baseline clinical information, treatment data and follow-up records sufficient to ascertain survival outcomes; and (VI) age ≥18 years at treatment initiation.

Exclusion criteria were: (I) unclear or mixed histology, including combined small-cell and non-small-cell carcinoma; (II) a history of, or concomitant, other active malignant tumors; (III) patients initially diagnosed with limited-stage SCLC who later progressed to extensive-stage disease and subsequently received immune checkpoint inhibitors (ICIs); (IV) missing critical baseline laboratory parameters precluding the calculation of inflammatory or nutritional indices; (V) the presence at diagnosis of active autoimmune disease requiring long-term systemic immunosuppressive therapy, or serious intercurrent conditions such as uncontrolled infection or significant hepatic/renal impairment that would substantially affect the safety or efficacy of immunotherapy; (VI) incomplete medical records with missing key clinical or follow-up information; (VII) exposure to other immunomodulatory agents within 3 months prior to the first dose of ICIs that might interfere with the assessment of peripheral inflammatory status; and (VIII) receipt of combinations of two or more different ICIs as part of first-line therapy; or (IX) receipt of who received ICI monotherapy, non-EP chemotherapy regimens, or other non-standard treatment combinations.

After application of these criteria, a total of 605 patients were included in the final analysis. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Tianjin Chest Hospital (approval No. 2025KY-440010-01). Informed consent was waived in this retrospective study. All participating hospitals were informed and agreed to the study.

Clinical variables and endpoint definitions

For each enrolled patient, we extracted baseline demographic characteristics, clinical features, laboratory parameters and treatment information from the electronic medical records. Demographic and general clinical variables included age, sex, smoking status (never, former, current), smoking exposure quantified by the smoking index (number of cigarettes per day × years of smoking), and Eastern Cooperative Oncology Group performance status (ECOG PS, 0 or 1). The primary tumor location was recorded as left or right lung. Baseline metastatic status was carefully documented, including the presence or absence of liver, bone, brain and adrenal metastases, as well as MPE. Based on these data, the number of involved distant organs (MetSitesCount; range, 0 to ≥2) was calculated for each patient.

Laboratory variables were obtained from blood samples collected within 7 days before the initiation of first-line systemic therapy at each center. These included peripheral white blood cell, neutrophil, lymphocyte and platelet counts; serum albumin (ALB), LDH, alanine aminotransferase (ALT), aspartate aminotransferase (AST), sodium (Na) and potassium (K) levels; and tumor markers such as NSE and carcinoembryonic antigen (CEA). Based on these measurements, we calculated several composite indices reflecting systemic inflammation and nutritional status, including the NLR, derived NLR (dNLR), PLR, SII, PNI, and LIPI. Detailed formulas for these indices are provided in Appendix 1.

LIPI was categorized according to the published definition based on dNLR and LDH (14), and patients were classified as having good, intermediate, or poor LIPI. Detailed definitions are provided in Appendix 1. Electrolyte disturbances were defined as follows: hyponatremia as Na <135 mmol/L; hypokalemia as K <3.5 mmol/L; and hyperkalemia as K >5.5 mmol/L.

Treatment-related variables included the date of first ICI administration, treatment modality (chemoimmunotherapy versus chemotherapy alone), and subsequent treatments after disease progression when available. For the purpose of modelling, chemoimmunotherapy was defined as first-line platinum–etoposide combined with a PD-1 or PD-L1 inhibitor, with ICIs initiated on cycle 1 day 1 whenever feasible. Chemotherapy alone was defined as first-line EP without any ICI throughout the entire course of first-line treatment. Patients who received ICI monotherapy, non-EP regimens, or other non-standard treatment combinations were excluded from the analysis.

The primary endpoint was OS, defined as the interval from the initiation of first-line systemic therapy (i.e., cycle 1 day 1 of an EP-based regimen, with or without a PD-1/PD-L1 inhibitor) to death from any cause. Patients alive at last contact were censored at the date of last follow-up. The secondary endpoint was progression-free survival (PFS), defined as the interval from the same index date (initiation of first-line systemic therapy) to the first documented disease progression per RECIST 1.1 or death from any cause, whichever occurred first. Patients alive without progression at last follow-up were censored at that date. Disease progression was assessed by experienced radiologists or treating physicians based on serial imaging and clinical evaluation. Follow-up data were obtained through outpatient visits, inpatient records and telephone contact. The database was locked on 30 June 2025.

The selection of these composite indices was based on prior literature supporting their prognostic relevance in SCLC and immunotherapy-treated thoracic malignancies.

Model development, validation and risk stratification

All 605 eligible patients were randomly assigned in a 7:3 ratio to a training cohort (n=421) for model development and a validation cohort (n=184) for independent performance assessment, using a computer-generated random sequence without stratification by centre. This split was selected as a commonly used approach in prognostic model development to balance two objectives: preserving sufficient sample size and event numbers for stable model derivation in the training cohort, while retaining an adequately sized independent subset for internal validation of discrimination and calibration.

In the training cohort, candidate predictors were first examined in univariable Cox proportional hazards models for their association with OS. Continuous inflammatory and tumour markers (e.g., SII, LDH, NSE, NLR) were dichotomised at optimal cut-off values identified in the training cohort using maximally selected log-rank statistics, and the same cut-offs were then applied to the validation cohort. Variables with P<0.10 in univariable analysis, together with clinically important covariates, were subsequently entered into a multivariable Cox model. Backward stepwise selection guided by Akaike’s information criterion was used to derive the final parsimonious model. Potential collinearity, particularly between SII and LIPI, was assessed using correlation matrices and variance inflation factors, and no problematic collinearity or strong interactions were detected.

The final model coefficients were used to construct a nomogram and a corresponding integer-based risk score. For each predictor, the absolute value of its regression coefficient was divided by the smallest non-zero coefficient and rounded to the nearest integer to generate a simple points system. The total risk score for each patient was obtained by summing the points across all predictors, and higher scores indicated a worse prognosis. Based on the distribution of total scores in the training cohort, the 25th and 75th percentiles were chosen as cut-off points to define three risk categories; these corresponded to total scores of ≤16 (low risk), 17–89 (intermediate risk) and ≥90 (high risk). The same scoring system and thresholds were applied to the validation cohort and to PFS analyses.

Model performance was evaluated in terms of discrimination, calibration and clinical usefulness. Discrimination was quantified using Harrell’s concordance index (C-index) and time-dependent receiver-operating-characteristic curves with corresponding areas under the curve at 12, 18 and 24 months in both cohorts. Calibration of predicted 1-, 1.5- and 2-year OS probabilities was assessed with calibration plots incorporating bootstrap resampling. Clinical utility was examined with decision-curve analysis based on the net benefit across a range of clinically relevant threshold probabilities. The proportional hazards assumption for the final model was checked using Schoenfeld residuals and global tests.

Statistical analysis

All statistical analyses were performed using R software (version 4.4.3; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are presented as medians with interquartile ranges (IQRs) and were compared between the training and validation cohorts using the Wilcoxon rank-sum test, whereas categorical variables are summarised as counts and percentages and were compared using the χ² test or Fisher’s exact test, as appropriate. OS and PFS were estimated by the Kaplan-Meier method and compared with the log-rank test. Univariable and multivariable Cox proportional hazards models were used to estimate hazard ratios (HRs) with 95% confidence intervals (CIs). Model construction, discrimination and calibration were performed mainly with the survival, rms and timeROC packages, and clinical utility was assessed using decision-curve analysis implemented in rmda. Nomograms, calibration plots, receiver-operating-characteristic curves and other graphical displays were generated with rms and ggplot2. All statistical tests were two-sided and P values <0.05 were considered statistically significant.

Results

Patient enrollment and baseline characteristics

Between January 2019 and June 2025, 605 patients with ES-SCLC met the eligibility criteria and were included in the analysis; 421 were allocated to the training cohort and 184 to the validation cohort. The overall cohort was predominantly male (79%) with a median age of 66 years (IQR, 60–71 years), and 70% had a smoking index ≥400, indicating a high cumulative tobacco exposure. Most patients had preserved ECOG PS (0/1 in 67%/33%, respectively), and right-sided primary tumors were slightly more frequent than left-sided ones (56% vs. 44%). Although all patients fulfilled the definition of extensive-stage disease, the number of involved distant organs was generally low: 77% had no solid organ metastasis, 16% had a single metastatic organ and only 7% had ≥2 involved organs. Bone metastases were the most common distant site (10%), followed by liver (3.8%), adrenal (3.6%) and brain (3.3%) involvement, and 10% had MPE. Laboratory parameters and inflammatory indices were mildly to moderately elevated, with median SII of 670 and median NLR of 2.71, while 11% of patients had hyponatremia and 6.3% hypokalemia. Overall, 232 patients (38%) received first-line chemoimmunotherapy and 373 (62%) EP chemotherapy alone, reflecting the gradual uptake of PD-1/PD-L1 inhibitor-based regimens during the study period. Overall, the cohort showed relatively favorable baseline characteristics, including exclusively preserved performance status (ECOG 0–1), a low proportion of brain metastases, and a generally limited solid-organ metastatic burden at diagnosis, which should be considered when interpreting prognosis and generalizability.

Comparability between the training and validation cohorts

Baseline characteristics were generally well balanced between the training and validation cohorts, with no significant differences in age, sex, smoking burden, ECOG PS, metastatic burden, inflammatory markers or treatment modality (all P>0.05; Table 1). A slight imbalance in primary tumour laterality was observed, but this variable was not associated with OS in univariable analysis and was therefore unlikely to bias the results.

Determination of optimal cut-off values for continuous inflammatory and tumor markers

Prior to multivariable modelling, optimal cut-off values for four key continuous biomarkers (SII, LDH, NSE and NLR) were determined in the training cohort using maximally selected log-rank statistics with OS as the endpoint (Figure 2A-2D). The thresholds that maximised between-group separation in OS were 707.3 for SII, 253.4 U/L for LDH, 25.1 ng/mL for NSE and 4.07 for NLR. The corresponding test statistics were as follows:

• SII: 707.3 (χ²=22.48, P=2.12×10⁻⁶);
• LDH: 253.4 U/L (χ²=17.70, P=2.58×10⁻⁵);
• NSE: 25.1 ng/mL (χ²=24.53, P=7.33×10⁻⁷);
• NLR: 4.07 (χ²=27.58, P=1.50×10⁻⁷).

Figure 2 Determination of optimal cut-off values and survival curves for key biomarkers. (A-D) Maximally selected log-rank statistics for SII (A), LDH (B), NSE (C) and NLR (D). The x-axes show candidate cut-off values across the observed range for each marker, and the y-axes show the corresponding −log₁₀(P) from log-rank tests. Vertical dashed lines indicate the optimal cut-off values that maximised separation in overall survival: SII 707.3, LDH 253.4 U/L, NSE 25.1 ng/mL and NLR 4.07. (E-H) Kaplan-Meier curves for overall survival according to low vs. high levels of SII (E), LDH (F), NSE (G) and NLR (H), respectively, defined by the derived cut-offs. In all four markers, high-level groups show significantly shorter overall survival than low-level groups (all log-rank P<0.001). LDH, lactate dehydrogenase; NLR, neutrophil-to-lymphocyte ratio; NSE, neuron-specific enolase; SII, systemic immune-inflammation index.

Using these thresholds, patients were dichotomised into low vs. high groups for each biomarker (SII ≤707.3 vs. >707.3; LDH ≤253.4 vs. >253.4 U/L; NSE ≤25.1 vs. >25.1 ng/mL; NLR ≤4.07 vs. >4.07). Kaplan-Meier analyses consistently showed significantly worse OS in the high-level groups for all four markers (all log-rank P<0.0001; Figure 2E-2H), supporting the prognostic relevance of heightened systemic inflammation and tumour burden in this cohort. Notably, the derived SII threshold was close to the cohort median, yielding reasonably balanced group sizes and facilitating clinical interpretability.

In subsequent analyses, these biomarkers were entered as binary variables (SII_high, LDH_high, NSE_high and NLR_high) into univariable and multivariable Cox models together with metastatic pattern, LIPI and treatment modality to construct the final OS nomogram.

Univariable and multivariable Cox regression analyses

In the training cohort, univariable Cox proportional hazards models identified several baseline characteristics that were significantly associated with OS (Table 2, Figure 3). Inflammatory and tumour-related markers, including high SII, elevated LDH and NSE, as well as an unfavourable LIPI category, were all linked to an increased risk of death, whereas most demographic variables showed only weak or no associations with survival. The presence of liver, bone, brain, adrenal metastases or MPE, worse ECOG PS and a higher number of involved metastatic organs were also associated with shorter OS, while patients who received first-line chemoimmunotherapy tended to have better outcomes than those treated with chemotherapy alone.

Figure 3 Univariable and multivariable Cox regression analyses for overall survival. (A) Univariable analysis showing HRs and 95% CIs for demographic characteristics, metastatic burden, inflammatory and nutritional markers, and treatment modality. (B) Multivariable analysis of the final nine-variable Cox model. The model includes NSE_high, SII_high, liver metastasis, bone metastasis, brain metastasis, adrenal metastasis, malignant pleural effusion, LIPI (three categories) and treatment modality (Immuno + Chemo vs. chemotherapy alone). Variables with HR >1 are associated with shorter overall survival, whereas HR <1 indicates a protective effect. AIC, Akaike information criterion; CI, confidence interval; HR, hazard ratio; Immuno + Chemo, PD-1/PD-L1 inhibitors plus platinum-etoposide chemotherapy; LIPI, lung immune prognostic index; NSE, neuron-specific enolase; NSE_high, serum NSE above the optimal cut-off; SII, systemic immune-inflammation index; SII_high, SII above the optimal cut-off.

Variables with P<0.10 in univariable analysis and/or clear clinical relevance were subsequently entered into multivariable Cox models. In the final parsimonious model, nine predictors remained independently associated with OS: high NSE, high SII, liver metastasis, bone metastasis, brain metastasis, adrenal metastasis, MPE, LIPI category and treatment modality (chemoimmunotherapy vs chemotherapy alone). Together, these factors integrate systemic inflammation, tumour burden, metastatic topography and first-line treatment modality, and formed the basis for the prognostic nomogram and risk score. Detailed HRs and 95% CIs for all candidate variables are provided in Table 2.

Construction of the nomogram and risk scoring system

The final nine-variable Cox model was used to construct a nomogram for predicting 12-, 18- and 24-month OS (Figure 4). Each predictor was assigned a number of points proportional to its regression coefficient, and the total score was obtained by summing the points across all variables. MPE contributed the largest number of points in the nomogram, followed by liver and brain metastases, whereas bone and adrenal metastases, high SII, high NSE, LIPI category and treatment modality had intermediate weights. For any individual patient, the total points can be projected onto the bottom scales to estimate personalised survival probabilities at the pre-specified time points.

Figure 4 Nomogram for predicting overall survival. The nomogram integrates nine predictors: NSE, SII, liver metastasis, bone metastasis, brain metastasis, adrenal metastasis, pleural effusion, LIPI category (good, medium, poor) and treatment modality (Immuno + Chemo vs. chemotherapy alone). For each variable, a point value is assigned by projecting the patient’s status onto the “Points” scale at the top. The sum of all points corresponds to a “Total points” value, which can then be mapped to predicted probabilities of 12-, 18- and 24-month overall survival on the lower scales. Higher total points indicate higher mortality risk. For binary variables, NSE was coded as 0 for low and 1 for high, SII as 0 for low and 1 for high, and treatment as 0 for chemotherapy alone and 1 for Immuno + Chemo. Immuno + Chemo, PD-1/PD-L1 inhibitors plus platinum-etoposide chemotherapy; LIPI, lung immune prognostic index; NSE, neuron-specific enolase; SII, systemic immune-inflammation index.

To facilitate clinical application, we further converted the nomogram into a simple integer-based risk score using the approach described in the Methods. The total score showed an approximately right-skewed distribution in the training cohort (Table 3). Based on the 25th and 75th percentiles of the score distribution, patients were categorised into low-risk (total score ≤16), intermediate-risk [17–89] and high-risk (≥90) groups. The same scoring algorithm and cut-off values were applied to the validation cohort and were also used for PFS analyses.

Survival analysis according to risk score categories

As expected, higher risk scores were associated with substantially worse outcomes. In the training cohort, Kaplan-Meier curves for OS showed clear separation among the three risk groups, with the best survival observed in the low-risk group and the poorest in the high-risk group (Figure 5). Similar patterns were seen for PFS. The validation cohort confirmed these findings, with a stepwise decrease in both OS and PFS across the low-, intermediate- and high-risk categories.

Figure 5 Survival according to risk score categories. (A) Overall survival in the training cohort stratified by low-, intermediate- and high-risk groups defined by the total risk score (low ≤16, intermediate 17–89, high ≥90 points). (B) Overall survival in the validation cohort using the same risk-score cut-offs. (C) Progression-free survival in the training cohort according to the three risk groups. (D) Progression-free survival in the validation cohort. In both cohorts, survival curves show clear and consistent separation across risk groups, with significantly worse outcomes in the high-risk group and the best outcomes in the low-risk group (all log-rank P<0.001). The numbers of patients at risk at selected time points are shown below each panel. OS, overall survival; PFS, progression-free survival.

When risk groups were entered into Cox models, intermediate-risk patients had an increased risk of death compared with the low-risk group, and high-risk patients had the highest hazard of death. A comparable gradient was observed for PFS. These results indicate that the proposed score-based stratification can effectively discriminate prognosis and identify a subset of ES-SCLC patients at particularly high risk of early progression and death.

Discriminative ability, calibration performance and proportional hazards assumption

After establishing the nomogram and risk score, we assessed its discrimination, calibration and key modelling assumptions. Time-dependent receiver operating characteristic (ROC) curves for 12-, 18- and 24-month OS are presented in Figure 6A,6B. In the training cohort, the area under the curves (AUCs) were approximately 0.75 or slightly higher across the three time points, indicating good discrimination within the clinically relevant 1–2-year window. In the validation cohort, AUCs were modestly lower but remained above 0.70, suggesting that the model retained acceptable performance and did not show marked overfitting.

Figure 6 Discriminative performance of the nomogram. (A) Time-dependent ROC curves for predicting 12-, 18- and 24-month overall survival in the training cohort. (B) Corresponding ROC curves in the validation cohort. (C,D) Calibration plots for predicted vs. observed 12- and 24-month overall survival in the training cohort. The diagonal dashed line represents perfect calibration; the solid line shows bias-corrected estimates obtained by bootstrap resampling. (E) tAUC curves for overall survival in the training and validation cohorts over the follow-up period. AUC values generally remained in the 0.70–0.80 range, without marked decline over time, indicating stable discriminative ability. AUC, area under the curve; OS, overall survival; ROC, receiver operating characteristic; tAUC, time-dependent area under the curve.

Calibration plots for predicted 12- and 24-month OS (Figure 6C,6D) showed good agreement between predicted and observed outcomes in both cohorts, indicating good calibration of the nomogram.

To evaluate whether discrimination was stable over time rather than being confined to a single landmark, we further examined time-dependent AUC (tAUC) curves (Figure 6E). In both cohorts, tAUC increased early after treatment initiation and then remained largely within the 0.70–0.80 range during follow-up, with only mild fluctuations and no sustained decline, supporting consistent risk ranking across the observed time horizon. Deviations were mainly limited to the extreme high-risk region, without evidence of systematic over- or underestimation. Finally, Schoenfeld residuals and the global test suggested no significant violation of the proportional hazards assumption (P=0.16; Figure 7), supporting the appropriateness of the Cox proportional hazards framework for nomogram construction.

Figure 7 Proportional hazards assumption checks. The figure shows the scaled Schoenfeld residuals plotted against follow-up time, with a smooth curve (solid line) and a reference line at zero. No obvious time-dependent trends are observed. The global Schoenfeld test yielded P=0.16, indicating that the proportional hazards assumption is acceptable for the final multivariable model. OS, overall survival.

Calibration performance

We additionally generated bootstrap-corrected calibration curves for 12-, 18- and 24-month OS in the training cohort (Figure 8) to examine the reliability of absolute risk estimates. Across all three time points, the bias-corrected curves closely followed the 45° reference line, with relatively narrow uncertainty bands over most of the predicted-risk range. Minor departures were mainly observed at the highest predicted risks, indicating that overall calibration was satisfactory while estimates at the extreme tail should be interpreted with appropriate caution.

Figure 8 Calibration of the nomogram at 12, 18 and 24 months. The x-axes represent predicted overall survival probabilities at each time point, and the y-axes represent observed overall survival probabilities. The dashed line denotes perfect calibration, and the solid blue line shows the bias-corrected smoothed calibration curve obtained from 200 bootstrap resamples; shaded areas indicate 95% confidence intervals. Across all three time points, the calibration curves closely follow the ideal line, with only minor deviations in the extreme high-risk region, suggesting good agreement between predicted and observed risks. OS, overall survival.

Clinical net benefit of the model

We evaluated clinical utility using decision curve analysis (DCA) (Figure 9). In both the training and validation cohorts, the nomogram-based strategy yielded higher net benefit than the “treat-all” and “treat-none” strategies across a broad range of clinically relevant threshold probabilities for 12-month OS (approximately 0.15–0.60; Figure 9A,9B). This suggests that, within common decision thresholds used to intensify follow-up or management, the nomogram may help identify truly high-risk patients while reducing unnecessary interventions among those at lower risk.

Figure 9 Decision curve analysis of the nomogram. (A,B) Net benefit curves for the nomogram compared with two extreme strategies—treating all patients (treat-all) and treating no patients (treat-none)—in the training (A) and validation (B) cohorts. The x-axes represent threshold probabilities for 12-month overall survival-based risk classification, and the y-axes represent net benefit. Across a wide range of clinically relevant thresholds (approximately 0.15–0.60), the nomogram provides higher net benefit than both treat-all and treat-none in both cohorts. (C,D) Incremental net benefit of the nomogram compared with a model based on SII alone for predicting 12-month (C) and 24-month (D) overall survival in the training cohort. In most threshold ranges, the nomogram curve lies above the SII-only curve and the two extreme strategies, indicating additional clinical value from integrating metastatic burden, systemic inflammation/nutritional status and treatment modality. OS, overall survival; SII, systemic immune-inflammation index.

We further compared the comprehensive nomogram with a single-marker model based on SII alone (Figure 9C,9D). Across most thresholds (approximately 0.25–0.60), the nomogram consistently provided greater net benefit than the SII-only strategy, indicating incremental value from integrating metastatic pattern, systemic inflammation and treatment modality. Notably, net benefit decreased at longer horizons (e.g., 24 months), which is expected as prediction becomes more challenging; nevertheless, the nomogram generally maintained an advantage over SII alone.

Comparison with previously reported prognostic models

To benchmark our model against existing tools, we reconstructed three representative published prognostic models (Dang, Gao, and Ge) (16-18) within the same training and validation cohorts using the variables and cut-offs reported in the original studies. Time-dependent ROC analyses were performed to estimate AUCs for 12-, 18- and 24-month OS for each model (Figure 10A,10B).

Figure 10 Comparison with previously published prognostic models. (A) tAUCs at 12, 18 and 24 months for overall survival in the training cohort for the proposed nine-variable nomogram and three representative models (Dang, Gao and Ge). (B) Corresponding AUCs in the validation cohort. In both cohorts, the proposed nomogram consistently achieves the highest AUCs or remains at the upper end of the range across all time points, whereas previously published models show lower AUCs and a greater decline in discrimination when moving from the training to the validation cohort. AUC, area under the curve; OS, overall survival; tAUC, time-dependent area under the curve.

In the training cohort, the published models generally achieved AUCs in the range of approximately 0.65–0.75 across time points, whereas our nine-variable nomogram showed consistently higher discrimination, with AUCs approaching ~0.80 at 12 months and remaining around or above ~0.75 at 18 and 24 months. In the validation cohort, all models exhibited some reduction in AUC; however, our nomogram retained comparatively stable performance, with AUCs remaining around ~0.75 and showing less attenuation than the comparator models.

Taken together, these head-to-head comparisons indicate that, within the same ES-SCLC study population and evaluation framework, the proposed nomogram provides more robust discrimination across cohorts and time horizons than several representative previously reported scoring systems, supporting its potential utility for risk stratification and individualised follow-up planning.

Discussion

Clinical context and rationale

ES-SCLC remains one of the most aggressive solid tumours, characterised by rapid systemic dissemination, short-lived responses to cytotoxic chemotherapy, and limited long-term survival. For decades, EP was the default first-line regimen, with most patients experiencing early relapse. The introduction of ICIs has produced the first sustained survival gains in the first-line setting: IMpower133 demonstrated that adding atezolizumab to carboplatin-etoposide improved OS, and CASPIAN confirmed an OS benefit with durvalumab plus EP, with durable survival tails reported in longer follow-up analyses (3,4).

Beyond these global standards, additional phase III trials in Asian populations have strengthened the evidence base for combining ICIs with EP-based chemotherapy, including CAPSTONE-1 (adebrelimab, a PD-L1 inhibitor) and ASTRUM-005 (serplulimab, a PD-1 inhibitor), both reporting clinically meaningful OS improvements versus chemotherapy alone (5,7).

Current guideline frameworks therefore generally recommend chemoimmunotherapy as preferred first-line therapy for eligible ES-SCLC patients, most commonly with atezolizumab or durvalumab in international guidelines, while PD-1-based regimens are supported by robust regional trial data (3,4,6,7).

However, the net benefit of first-line chemoimmunotherapy is heterogeneous. A substantial subset of patients progresses early despite ICI exposure, and intensification strategies have not consistently yielded incremental benefit (e.g., SKYSCRAPER-02 did not show improvement when adding tiragolumab to atezolizumab plus carboplatin-etoposide) (25).

In KEYNOTE-604, pembrolizumab plus EP improved PFS but did not meet the prespecified statistical threshold for OS, illustrating persistent uncertainty regarding who truly benefits most (26).

Meanwhile, widely applicable predictive biomarkers remain limited in routine SCLC practice due to tissue scarcity, inconsistent PD-L1 performance, and uneven access to comprehensive molecular profiling. This is particularly relevant when clinicians must make baseline prognostic assessments before treatment initiation, creating a need for a bedside-ready tool that can support risk stratification using routinely available variables.

Accordingly, a key gap is the lack of front-loaded, baseline-only prognostic models developed in the contemporary chemoimmunotherapy era that jointly capture metastatic topography, systemic inflammatory/nutritional status, tumour markers, and treatment strategy using routinely available variables. Many earlier models were derived in the chemotherapy era, lacked granularity in metastatic patterns, or relied on on-treatment landmarks that are unavailable when first-line decisions are made. We therefore developed and internally validated a nine-variable nomogram and integer-based score using variables obtainable before the first cycle, with the aim of supporting baseline prognostic assessment, follow-up planning, and supportive-care prioritization from day 1.

Key findings

In this multicenter retrospective cohort of treatment-eligible patients with ES-SCLC, we developed and internally validated a nine-variable prognostic model that integrates metastatic topography, systemic inflammatory/nutritional status, tumor markers, and first-line treatment modality in patients receiving EP-based therapy in routine clinical practice. Importantly, because the derivation cohort consisted exclusively of patients with preserved ECOG PS and relatively limited baseline disease burden, the intended scope of this model is primarily baseline prognostic stratification in such treatment-eligible patients rather than direct extrapolation to all ES-SCLC populations. The nomogram and integer-based score demonstrated consistently good discrimination (C-index around 0.75; tAUC generally 0.70–0.80 across 12–24 months) with satisfactory calibration and meaningful net benefit on decision-curve analysis.

Importantly, the model did not merely reproduce a ‘generic’ inflammation-only risk score: by explicitly incorporating organ-specific metastatic involvement and MPE, it captured clinically intuitive but under-modelled anatomic risk. Notably, MPE carried the greatest point weight in the nomogram, exceeding liver and brain metastases, underscoring that intrathoracic metastatic biology and symptom burden can dominate prognosis in routine practice.

From a practical standpoint in the chemoimmunotherapy era, the model is intended to support baseline prognostic assessment at treatment initiation, because all variables are available before therapy starts. Importantly, this score is prognostic rather than a validated predictive biomarker and should not be used to deny guideline-supported chemoimmunotherapy in eligible patients. Instead, its clinical utility lies in identifying patients who may require intensified supportive care, closer surveillance, and earlier clinical-trial discussion from day 1 (4-7,27).

Biological and clinical interpretation of key predictors

The independent effects of SII, LIPI and NSE highlight the intertwined roles of host inflammatory state, tumour burden/metabolic activity and neuroendocrine differentiation in ES-SCLC. SII integrates neutrophil, platelet and lymphocyte counts and therefore reflects both pro-tumour inflammation and impaired antitumour immunity; high SII plausibly represents a myeloid- and platelet-skewed milieu together with lymphopenia that limits cytotoxic responses. In the context of PD-1/PD-L1 blockade, such baseline inflammatory states may also act as pragmatic surrogates of an immunosuppressive host milieu that can blunt effective antitumour immunity and contribute to the wide inter-patient variability in chemoimmunotherapy outcomes (11,28).

LIPI, combining dNLR and LDH, complements SII by linking myeloid predominance with tumour metabolic load and hypoxia (14).

The simultaneous retention of SII and LIPI suggests related but non-redundant biological axes: SII emphasises platelet-associated inflammatory burden, whereas LIPI adds LDH-driven information that may capture aggressive disease biology not fully explained by SII alone.

NSE preserved prognostic value after adjustment for metastatic pattern and inflammatory indices. Elevated baseline NSE is associated with greater tumour burden, rapid proliferation and a ‘classical’ neuroendocrine phenotype, and may partly proxy molecular subtypes of SCLC that are not captured by anatomic staging alone. Thus, serum neuroendocrine markers still convey independent prognostic information in the PD-1/PD-L1 inhibitor era (29).

The inclusion of site-specific metastases (liver, bone, brain and adrenal) together with MPE underscores that metastatic topography may convey prognostic information beyond a simple count of involved organs. In model construction, we did not simultaneously retain the number of involved distant organs and individual metastatic locations in the final multivariable model because these variables reflect overlapping dimensions of disease extent and could introduce redundancy while reducing model interpretability.

Notably, MPE received the highest point weight in the nomogram. This finding should be interpreted cautiously. On the one hand, MPE may represent an adverse clinical phenotype that is not fully captured by organ-count metrics alone, potentially reflecting pleural compartment tumour biology, substantial symptom burden, impaired respiratory reserve, nutritional deterioration, repeated pleural interventions, and reduced tolerance to systemic therapy. On the other hand, in a retrospective observational cohort, we cannot exclude the possibility that MPE also acts partly as a surrogate for more advanced intrathoracic disease and greater overall disease burden. Accordingly, the prognostic importance of MPE in our model likely reflects a combination of independent clinical impact and correlation with disease extent.

Contemporary pleural disease guidance emphasises that symptomatic MPE commonly requires definitive palliative strategies (e.g., indwelling pleural catheters and/or pleurodesis), reflecting its substantial clinical burden (30,31). Beyond symptom burden, MPE may represent a distinct immunologically ‘cold’ compartment. Multiple studies demonstrate enrichment of immunosuppressive cell populations and mediators in malignant pleural fluid, including regulatory T cells and chemokine-driven recruitment pathways (e.g., CCL22-associated Treg trafficking), as well as macrophage programmes that impair effector T-cell function (e.g., TGF-β-linked dysfunction) (32-34).

These data provide a mechanistic rationale for why MPE may portend particularly poor outcomes even compared with certain distant metastatic sites, and why its prognostic ‘weight’ in this treatment-eligible cohort can exceed traditional expectations (35,36).

Finally, treatment modality remained independently prognostic after accounting for baseline risk, consistent with phase III trials supporting PD-1/PD-L1 blockade added to EP-based chemotherapy in ES-SCLC (3-5,7,26).

Comparison with previous prognostic models

Several prognostic models for SCLC have been published in recent years, many centred on inflammatory/nutritional indices or conventional clinical variables, but most were developed in smaller, single-centre cohorts and/or in the chemotherapy era, with limited representation of chemoimmunotherapy and minimal granularity in metastatic patterns. In this study, we reconstructed and refitted representative published ES-SCLC models (e.g., Dang, Gao and Ge) within the same dataset and compared tAUCs directly (16-18).

Across 12-, 18- and 24-month OS in both cohorts, our nine-variable model generally achieved higher AUCs and showed less attenuation from training to validation. Two additional points strengthen the fairness and interpretability of this head-to-head comparison. First, some immunotherapy-era models incorporate early on-treatment response milestones to improve discrimination; while clinically meaningful, these approaches are less applicable for baseline-only prognostication at treatment initiation. By contrast, our model is intentionally ‘front-loaded’ with routinely available baseline variables, enabling risk stratification at the time of initial clinical assessment before treatment begins (16). Second, recently proposed ‘clinlabomics’ nomograms have screened large panels of laboratory indicators to derive compact models; however, many do not explicitly incorporate immunotherapy exposure or detailed metastatic topography (37).

Taken together, our results suggest that integrating metastatic topography, inflammation, and treatment context yields robust discrimination in the immunotherapy era, rather than simply adding another inflammation-based score. Future work should further examine treatment-risk interactions and assess whether model performance remains stable across different therapeutic subgroups.

Methodological considerations and potential biases

Several methodological aspects merit attention. First, we used the maximally selected log-rank statistic to derive outcome-oriented cut-offs for SII, LDH, NSE and NLR in the training cohort. Although this approach can inflate optimism if repeatedly tuned, the thresholds were fixed a priori and applied unchanged to the validation cohort, and the observed validation performance supports reasonable robustness.

Second, SII and LIPI share overlapping components (neutrophil and lymphocyte counts), raising concern for multicollinearity. We explicitly assessed collinearity and found no evidence of problematic inflation; the retention of both indices suggests that platelet-associated inflammation (SII) and LDH-driven metabolic load (LIPI) capture complementary information.

Third, as an observational study conducted in treatment-eligible patients, treatment selection for chemoimmunotherapy versus chemotherapy alone was non-random and may reflect age, comorbidities, access and physician preference. The coefficient for chemoimmunotherapy in our multivariable model should therefore be interpreted as an adjusted association within a prognostic framework rather than a causal treatment effect. Residual confounding cannot be excluded; nevertheless, the directionality is concordant with phase III evidence supporting PD-1/PD-L1 blockade added to EP in ES-SCLC (3-5,7).

Fourth, our cohort included both PD-1 and PD-L1 inhibitors in the chemoimmunotherapy group, reflecting routine clinical practice in treatment-eligible patients. Potential heterogeneity across ICI classes, dosing schedules, and maintenance strategies may introduce additional variability; future analyses with larger samples should assess whether model performance and associations remain stable when stratifying by ICI class and regimen.

An additional modelling decision concerned metastatic burden representation: because the number of involved distant organs and individual metastatic sites were strongly related descriptors of disease extent, we prioritized site-specific variables in the final model to favor anatomical granularity and clinical interpretability over inclusion of a summary count variable.

Finally, we excluded patients with missing key baseline laboratory parameters, which may introduce selection toward fitter individuals. Inter-centre variation in laboratory assays and reference ranges may also add noise; these issues should be considered when applying the model to other healthcare settings.

Strengths and limitations

Strengths of this study include the multicenter design, focus on contemporary systemic therapy, and a modelling strategy that jointly captures anatomic disease distribution, host inflammatory biology and treatment modality, followed by comprehensive evaluation (discrimination, calibration, proportional hazards assessment and clinical net benefit). Benchmarking against previously published models in the same dataset also provides a transparent, like-for-like comparison of prognostic tools.

Limitations should be acknowledged. The retrospective design is susceptible to selection and information biases, and validation was internal via random splitting rather than external validation in fully independent cohorts. Key biomarkers such as PD-L1 expression, tumour mutational burden, DNA damage repair alterations and molecular SCLC subtypes were unavailable and could further refine risk. Moreover, we did not incorporate dynamic on-treatment milestones (e.g., early response at predefined weeks) that may enhance prediction in some frameworks (16), and extreme high-risk strata were relatively small, warranting cautious interpretation at the distribution tails.

In addition, the cohort consisted exclusively of patients with ECOG PS 0–1 and relatively limited baseline solid-organ metastatic burden, including a low proportion of brain metastases. This likely reflects treatment selection in routine practice and represents an important limitation of the present study.

Accordingly, the nomogram should be interpreted primarily in treatment-eligible patients with preserved functional status, and its applicability to patients with poorer performance status or heavier disease burden remains uncertain and requires further external validation.

Clinical implications and future directions

Because all required variables are routinely collected, the nomogram and score can be applied at treatment initiation to stratify ES-SCLC patients into risk groups and support baseline prognostic assessment, follow-up planning, and supportive-care prioritization. In particular, patients with MPE and other high-risk metastatic patterns may benefit from intensified surveillance, early symptom-directed pleural management, proactive nutritional/supportive care and prioritisation for clinical trials (30,31). Conversely, lower-risk patients may be managed with standard schedules while avoiding unnecessary escalation.

From a clinical perspective, the score may be most useful for baseline prognostic stratification at the start of first-line therapy. In higher-risk patients, it may support closer surveillance, earlier symptom-directed interventions, proactive pleural management when relevant, nutritional/supportive care prioritization, and clinical-trial discussion. Conversely, in lower-risk patients, it may facilitate more individualized prognostic counselling and follow-up planning (3-5,7).

Importantly, current evidence supports chemoimmunotherapy as the preferred first-line strategy for eligible patients. Therefore, this tool should be interpreted as a prognostic aid rather than a validated predictive biomarker for selecting or withholding immunotherapy. Prospective evaluation is required before any treatment-regimen decisions could be linked to baseline risk strata.

Future work should include prospective registries and truly external validation to confirm transportability, alongside integration of molecular, imaging and immune microenvironment features. An important next step is to test whether model-informed surveillance, supportive optimisation, or treatment intensification improves outcomes, rather than merely predicting prognosis.

Conclusions

In this multicenter retrospective cohort, we developed and internally validated a 9-variable nomogram for baseline prognostic stratification in treatment-eligible patients with ES-SCLC receiving contemporary EP-based therapy in routine clinical practice. The model demonstrated consistently good discrimination (C-index ≈0.75; tAUCs generally 0.70–0.80) and satisfactory calibration in both the training and validation cohorts, and provided stable clinical net benefit across a broad range of decision thresholds. Compared with several representative published models (Dang, Gao and Ge) (16-18), the proposed nomogram showed superior and more robust predictive performance when rebuilt and tested in the same study population. Because all required variables are routinely available in daily practice, the resulting nomogram and integer-based risk score can be readily applied to stratify ES-SCLC patients into low-, intermediate-, and high-risk groups, thereby supporting baseline prognostic assessment, determination of follow-up intensity, and supportive-care planning in the immunotherapy era. The model should be interpreted as a prognostic tool rather than a predictive biomarker for treatment selection. Prospective external validation in independent cohorts and integration with molecular and imaging biomarkers are warranted to further refine and extend the clinical utility of this model.

Acknowledgments

We thank all investigators and clinical staff from the participating institutions for their assistance in patient identification, data collection and follow-up. We also acknowledge the contributions of the data management teams and statisticians who provided technical and methodological support.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1445/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1445/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1445/prf

Funding: This work was supported by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (project No. 2025ZD0548500 and subproject No. 2025ZD0548504), the Tianjin Key Medical Discipline (Specialty) Construction Project (grant No. TJYXZDXK-018A), the International Science and Technology Innovation Cooperation Program under the National Key R&D Program of China (grant No. 25YFGKHZ00180), and the Tianjin Municipal Health Commission General Program (grant No. TJWJ2023QN064).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-1-1445/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Tianjin Chest Hospital (approval No. 2025KY-440010-01). Informed consent was waived in this retrospective study. All participating hospitals were informed and agreed to the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Fasano R, Serratì S, Rafaschieri T, et al. Small-Cell Lung Cancer: Is Liquid Biopsy a New Tool Able to Predict the Efficacy of Immunotherapy? Biomolecules 2024;14:396. [Crossref] [PubMed]
3. Horn L, Mansfield AS, Szczęsna A, et al. First-Line Atezolizumab plus Chemotherapy in Extensive-Stage Small-Cell Lung Cancer. N Engl J Med 2018;379:2220-9. [Crossref] [PubMed]
4. Paz-Ares L, Dvorkin M, Chen Y, et al. Durvalumab plus platinum-etoposide versus platinum-etoposide in first-line treatment of extensive-stage small-cell lung cancer (CASPIAN): a randomised, controlled, open-label, phase 3 trial. Lancet 2019;394:1929-39. [Crossref] [PubMed]
5. Wang J, Zhou C, Yao W, et al. Adebrelimab or placebo plus carboplatin and etoposide as first-line treatment for extensive-stage small-cell lung cancer (CAPSTONE-1): a multicentre, randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Oncol 2022;23:739-47. [Crossref] [PubMed]
6. Kalemkerian GP, Loo BW, Akerley W, et al. NCCN Guidelines Insights: Small Cell Lung Cancer, Version 2.2018. J Natl Compr Canc Netw 2018;16:1171-82. [Crossref] [PubMed]
7. Cheng Y, Han L, Wu L, et al. Effect of First-Line Serplulimab vs Placebo Added to Chemotherapy on Survival in Patients With Extensive-Stage Small Cell Lung Cancer: The ASTRUM-005 Randomized Clinical Trial. JAMA 2022;328:1223-32. [Crossref] [PubMed]
8. Zhao W, Jiang W, Wang H, et al. Impact of Smoking History on Response to Immunotherapy in Non-Small-Cell Lung Cancer: A Systematic Review and Meta-Analysis. Front Oncol 2021;11:703143. [Crossref] [PubMed]
9. Kang MH, Go SI, Song HN, et al. The prognostic impact of the neutrophil-to-lymphocyte ratio in patients with small-cell lung cancer. Br J Cancer 2014;111:452-60. [Crossref] [PubMed]
10. Zhou C, Hu Q, Song X, et al. Prognostic value of baseline LIPI, LDH and dNLR in ES-SCLC patients receiving immune checkpoint inhibitors: a systematic review and meta-analysis. Front Immunol 2025;16:1640066. [Crossref] [PubMed]
11. Hamakawa Y, Hirahara A, Hayashi A, et al. Prognostic value of systemic immune-inflammation index in patients with small-cell lung cancer treated with immune checkpoint inhibitors. BMC Cancer 2025;25:17. [Crossref] [PubMed]
12. Yu Y, Zeng D, Ou Q, et al. Association of Survival and Immune-Related Biomarkers With Immunotherapy in Patients With Non-Small Cell Lung Cancer: A Meta-analysis and Individual Patient-Level Analysis. JAMA Netw Open 2019;2:e196879. [Crossref] [PubMed]
13. Jiang AM, Zhao R, Liu N, et al. The prognostic value of pretreatment prognostic nutritional index in patients with small cell lung cancer and it's influencing factors: a meta-analysis of observational studies. J Thorac Dis 2020;12:5718-28. [Crossref] [PubMed]
14. Mezquita L, Auclin E, Ferrara R, et al. Association of the Lung Immune Prognostic Index With Immune Checkpoint Inhibitor Outcomes in Patients With Advanced Non-Small Cell Lung Cancer. JAMA Oncol 2018;4:351-7. [Crossref] [PubMed]
15. Li L, Pi C, Yan X, et al. Prognostic Value of the Pretreatment Lung Immune Prognostic Index in Advanced Small Cell Lung Cancer Patients Treated With First-Line PD-1/PD-L1 Inhibitors Plus Chemotherapy. Front Oncol 2021;11:697865. [Crossref] [PubMed]
16. Dang J, Xu G, Guo G, et al. Construction of a prognostic model for extensive-stage small cell lung cancer patients undergoing immune therapy in northernmost China and prediction of treatment efficacy based on response status at different time points. J Cancer Res Clin Oncol 2024;150:255. [Crossref] [PubMed]
17. Gao H, Dang Y, Qi T, et al. Mining prognostic factors of extensive-stage small-cell lung cancer patients using nomogram model. Medicine (Baltimore) 2020;99:e21798. [Crossref] [PubMed]
18. Ge Y, Liu X, Xu Y, et al. Combined systemic immune-inflammatory index and prognostic nutritional index predicts the efficacy and prognosis of ES-SCLC patients receiving PD-L1 inhibitors combined with first-line chemotherapy. Front Oncol 2024;14:1485849. [Crossref] [PubMed]
19. Steyerberg EW, Harrell FE Jr, Borsboom GJ, et al. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774-81. [Crossref] [PubMed]
20. Hou Q, Liang Y, Yao N, et al. Development of a novel nomogram for patients with SCLC and comparison with other models. BMC Cancer 2024;24:1257. [Crossref] [PubMed]
21. Jiang X, Wang Z, Zhang X, et al. A SEER prognostic nomogram analysis for extensive-stage small cell lung cancer: comparing the immune checkpoint inhibitors (ICIs) and non-ICIs eras. J Thorac Dis 2025;17:2866-87. [Crossref] [PubMed]
22. Damiano P, Stefani A, Avancini A, et al. Real-world evidence in extensive disease small cell lung cancer: The missing piece of the puzzle. Crit Rev Oncol Hematol 2025;207:104618. [Crossref] [PubMed]
23. Micke P, Faldum A, Metz T, et al. Staging small cell lung cancer: Veterans Administration Lung Study Group versus International Association for the Study of Lung Cancer--what limits limited disease?. Lung Cancer 2002;37:271-6. [Crossref] [PubMed]
24. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228-47. [Crossref] [PubMed]
25. Rudin CM, Liu SV, Soo RA, et al. SKYSCRAPER-02: Tiragolumab in Combination With Atezolizumab Plus Chemotherapy in Untreated Extensive-Stage Small-Cell Lung Cancer. J Clin Oncol 2024;42:324-35. [Crossref] [PubMed]
26. Rudin CM, Awad MM, Navarro A, et al. Pembrolizumab or Placebo Plus Etoposide and Platinum as First-Line Therapy for Extensive-Stage Small-Cell Lung Cancer: Randomized, Double-Blind, Phase III KEYNOTE-604 Study. J Clin Oncol 2020;38:2369-79. [Crossref] [PubMed]
27. Liu SV, Reck M, Mansfield AS, et al. Updated Overall Survival and PD-L1 Subgroup Analysis of Patients With Extensive-Stage Small-Cell Lung Cancer Treated With Atezolizumab, Carboplatin, and Etoposide (IMpower133). J Clin Oncol 2021;39:619-30. [Crossref] [PubMed]
28. Zhang M, Hao J, Wu Y, et al. Value of the lung immune prognostic index in patients with advanced small cell lung cancer treated with programmed death-ligand 1 and programmed death-1 inhibitors in the Chinese alpine region. Front Oncol 2024;14:1411548. [Crossref] [PubMed]
29. Tian Z, Liang C, Zhang Z, et al. Prognostic value of neuron-specific enolase for small cell lung cancer: a systematic review and meta-analysis. World J Surg Oncol 2020;18:116. [Crossref] [PubMed]
30. Feller-Kopman DJ, Reddy CB, DeCamp MM, et al. Management of Malignant Pleural Effusions. An Official ATS/STS/STR Clinical Practice Guideline. Am J Respir Crit Care Med 2018;198:839-49.
31. Roberts ME, Rahman NM, Maskell NA, et al. British Thoracic Society Guideline for pleural disease. Thorax 2023;78:1143-56. [Crossref] [PubMed]
32. Chen YQ, Shi HZ, Qin XJ, et al. CD4+CD25+ regulatory T lymphocytes in malignant pleural effusion. Am J Respir Crit Care Med 2005;172:1434-9. [Crossref] [PubMed]
33. Qin XJ, Shi HZ, Deng JM, et al. CCL22 recruits CD4-positive CD25-positive regulatory T cells into malignant pleural effusion. Clin Cancer Res 2009;15:2231-7. [Crossref] [PubMed]
34. Li L, Yang L, Wang L, et al. Impaired T cell function in malignant pleural effusion is caused by TGF-β derived predominantly from macrophages. Int J Cancer 2016;139:2261-9. [Crossref] [PubMed]
35. Principe N, Kidman J, Lake RA, et al. Malignant Pleural Effusions-A Window Into Local Anti-Tumor T Cell Immunity?. Front Oncol 2021;11:672747. [Crossref] [PubMed]
36. Murthy P, Ekeke CN, Russell KL, et al. Making cold malignant pleural effusions hot: driving novel immunotherapies. Oncoimmunology 2019;8:e1554969. [Crossref] [PubMed]
37. Peng Q, Yang F, Xu K, et al. Development and Validation of a Clinlabomics-Based Nomogram for Predicting the Prognosis of Small Cell Lung Cancer in China: A Multicenter, Retrospective Cohort Study. Cancer Med 2025;14:e71180. [Crossref] [PubMed]