Development and validation of a serum inflammatory biomarker-driven machine-learning model for prognostic stratification in surgical limited-stage small cell lung cancer

Abstract

Background

Robust prognostic markers for small cell lung cancer (SCLC) are currently lacking, underscoring the need for novel prediction models to optimize individualized treatment and improve patient outcomes. Inflammatory/nutritional indexes have been extensively employed in prognostic investigations of malignant tumors. The study aimed to precisely ascertain the prognosis of SCLC patients undergoing surgery by preoperative serological indexes.

Methods

We included patients with SCLC who underwent surgery at The Affiliated Hospital of Qingdao University. Potential predictors included basic clinical characteristics and preoperative serum inflammatory/nutritional indexes. We employed 10 machine learning algorithms and their 101 combinations to select the superior model and establish a novel nomogram. Follow-up involved regular clinic visits or telephone contact, with imaging and laboratory tests conducted at defined intervals to assess overall survival (OS) and progression-free survival (PFS). The cohort was randomly split into training and validation cohorts in a 7:3 ratio. Harrell’s C-index, Kaplan-Meier curves, log-rank tests, and Cox regression analyses were used for model evaluation and prognostic assessment.

Results

A total of 219 patients were included in this study. Prognostic nutritional index (PNI), lymphocyte-to-monocyte ratio (LMR), platelet-to-neutrophil ratio (PNR), neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammatory index (SII), pan-immune-inflammation value (PIV), and systemic inflammatory response index (SIRI) were correlated with the prognosis of SCLC patients. Smoking status and the tumor-node-metastasis (TNM) stage were independent prognostic indicators of OS. The Random Forest model achieved the highest mean concordance index (C-index) (0.784). Patients classified as high-risk based on this model exhibited a higher prevalence of smoking and more advanced pathological N stage and TNM stage. No significant differences were observed between risk groups regarding age, gender, body mass index (BMI), alcohol history, tumor site, pathological T stage, Ki-67 index, or visceral pleural invasion (VPI). Nomograms based on risk grouping, smoking status, and TNM stage demonstrated high precision and considerable clinical value. Multivariate Cox analysis identified PNI and NLR as the most valuable prognostic markers, with optimal cut-off values of 50.6 and 1.99, respectively.

Conclusions

A machine learning model based on serological inflammatory/nutritional indexes can reasonably estimate the long-term prognosis of SCLC patients and is anticipated to serve as a practical instrument for identifying the ideal candidates for thoracic surgery.

Highlight box.

Key findings

• Higher preoperative prognostic nutritional index (PNI), lymphocyte-to-monocyte ratio (LMR), platelet-to-neutrophil ratio (PNR), lower neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammatory index (SII), pan-immune-inflammation value (PIV), and systemic inflammatory response index (SIRI) were significantly associated with a favorable prognosis of small cell lung cancer (SCLC) patients.

• Preoperative PNI and NLR are independent prognostic indicators for SCLC patients, with optimal thresholds of 50.60 and 1.99, respectively.

What is known and what is new?

• Rare studies have examined the long-term prognostic role of preoperative serologic markers in patients with surgically resected SCLC.

• Ten machine learning algorithms and their one hundred and one combinations to select inflammatory/serum markers associated with patient prognosis.

What is the implication, and what should change now?

• Predictive models applying serologic indexes and machine learning methods have a high prognostic value.

Introduction

Lung cancer (LC) continues to be a malignancy with a high global prevalence (1). Despite the small cell lung cancer (SCLC) subtype constituting approximately 14% of all LC cases (2), it is highly aggressive with a strong propensity for recurrence and metastasis (3). According to cancer progression, SCLC is classified into limited-stage SCLC (LS-SCLC) (1/3) and extensive-stage SCLC (ES-SCLC) (2/3) (4). Although platinum-based chemotherapy combined with chest radiotherapy is a standard treatment regimen (5), the 5-year survival rate for SCLC patients remains below 10% (6,7). Additionally, the safety and efficacy of immunotherapy or chemoimmunotherapy for SCLC are unsatisfactory (8,9). Consequently, markers are urgently needed to predict the prognosis and facilitate individualized and precise treatment. Prognosis prediction models for SCLC have been extensively explored, incorporating clinical (such as smoking status), pathological (such as tumor stage and Ki-67 index), and other molecular predictors [such as CD56, synaptophysin (SYN), chromogranin A (CGA), DLL3, and programmed cell death 1 ligand 1 (PD-L1)] (10,11). While traditional statistical methods like the Cox proportional hazards model have identified key factors, their predictive accuracy remains limited due to oversimplified assumptions (12). Machine learning algorithms, such as random survival forest (RSF) and gradient boosting survival analysis (GBSA), show promise in capturing non-linear relationships and interactions (12). However, significant challenges persist, including SCLC heterogeneity, insufficient integration of genomic and transcriptomic data, and inadequate consideration of intratumoral heterogeneity and dynamic tumor evolution under treatment pressure (10-12).

Inflammatory, immune, and nutritional conditions are closely related to the prognosis of cancer patients (13). Neutrophils, lymphocytes, platelets, and monocytes involved in inflammatory and immune responses can promote cancer spread, invasion, and metastasis (14). Serum albumin, hemoglobin, and cholesterol reflect the nutritional status of cancer patients. These metrics have given rise to numerous novel inflammation/nutrition-related indexes, such as platelet-albumin ratio (PAR), fibrinogen-to-albumin ratio (FAR), prognostic nutritional index (PNI), hemoglobin, albumin, lymphocyte, and platelet (HALP) and albumin-to-alkaline phosphatase ratio (AAPR). These markers are of great significance in predicting the survival of cancer patients. Nevertheless, as most patients with SCLC present at an advanced stage, studies based on these markers have primarily concentrated on internal medicine. Lang et al. indicated that lymphocyte-to-monocyte ratio (LMR) serves as an independent prognostic marker for SCLC patients undergoing surgery (15). Beyond this, few investigations have explored the long-term prognostic effect of these markers in surgical population. These studies are highly significant as they may considerably influence the indications for thoracic surgery.

To our knowledge, this is the study with the most extensive sample size on preoperative serum prognostic markers of SCLC patients. In this study, we employed machine learning algorithms to screen the model and markers rationally and developed a novel nomogram that predicts the long-term survival of patients undergoing surgery. This study may guide clinical surgeons in determining the indications for SCLC surgery in the future. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-182/rc).

Methods

Clinical research design and sample size calculation

This retrospective, observational cohort study was designed to evaluate the impact of basic clinical characteristics and preoperative serologic indicators on the prognosis of SCLC patients who underwent surgical treatment. Primary outcomes were overall survival (OS) and progression-free survival (PFS). The risk score was calculated using selected biomarkers, and a predictive model was subsequently developed, internally validated, and clinically evaluated based on the risk score. The cohort was randomly split into the training and validation cohorts in a 7:3 ratio using R software’s ’sample’ function, with a fixed random seed [set.seed (1234)] to ensure reproducibility. The sample size was calculated using the Schoenfeld formula for survival analysis (16,17). With a two-sided significance level of 0.05, a power of 80%, an estimated 50% mortality incidence, and a hazard ratio (HR) of 3.845, the minimum required sample size was 184 cases.

Patient information

We retrieved clinical (including basic clinical information, blood test results 1 week before surgery, and postoperative pathological results) and follow-up data of LS-SCLC patients receiving radical lung lesion resection at The Affiliated Hospital of Qingdao University from June 2013 to June 2023. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The medical ethics committee of The Affiliated Hospital of Qingdao University has approved this research (approval No. QYFY WZLL 29635), and each participant provided the informed consent. Basic clinical information included gender, age, smoking status, alcohol history, body mass index (BMI), tumor size, and location. We excluded patients with the following characteristics: (I) patients receiving other induction therapies such as radiotherapy or chemotherapy before surgery; (II) patients with other cancers before surgery; (III) patients with distant metastases before surgery; (IV) patients with more than two cancerous lesions in the lung; (V) patients with postoperative pathological components of mixed small cell carcinoma; (VI) palliative surgery; (VII) patients with severe dysfunction of organs such as the heart, liver, lungs, and kidneys.

Serum inflammatory/nutritional indexes

A total of 12 serum inflammatory/nutritional indexes were calculated as follows: PNI, albumin (g/L) + 5 × lymphocytes (10⁹/L) (18); neutrophil-to-lymphocyte ratio [NLR, neutrophils (10⁹/L)/lymphocytes (10⁹/L)] (18); systemic immune-inflammatory index [SII, neutrophils (10⁹/L) × platelets (10⁹/L)/lymphocytes (10⁹/L)] (19); LMR [lymphocytes (10⁹/L)/monocytes (10⁹/L)] (15); platelet-to-neutrophil ratio [PNR, platelets (10⁹/L)/neutrophils (10⁹/L)] (20); pan-immune-inflammation value [PIV, neutrophils (10⁹/L) × platelets (10⁹/L) × monocytes (10⁹/L)/lymphocyte (10⁹/L)] (21); systemic inflammatory response index [SIRI, neutrophils (10⁹/L) × monocytes (10⁹/L)/lymphocytes (10⁹/L)] (22); platelet-to-lymphocyte ratio [PLR, platelets (10⁹/L)/lymphocytes (10⁹/L)] (18); AAPR [albumin (g/L)/alkaline phosphatase (U/L)]; FAR [fibrinogen (g/L)/albumin (g/L)] (23); PAR [platelets/albumin (g/L)] (24); HALP score [hemoglobin (g/L) × albumin (g/L) × lymphocytes (10⁹/L)/platelet (10⁹/L)] (25).

Follow-up

Discharged patients undergoing surgery were regularly followed up in the clinic or via telephone. Blood tests, X-rays, and/or chest computed tomography (CT) scans were reviewed every 3 months for the first year and every 6–12 months thereafter. Further specialized examinations (such as cranial magnetic resonance, abdominal CT, positron emission tomography, or histopathology) are performed when there is a suspicion of recurrence or metastasis elsewhere. Tumor recurrence is defined as ipsilateral lung, mediastinal, or hilar lymph node recurrence; metastasis is defined as recurrence at other sites (15). Adjuvant therapy (radiotherapy or chemotherapy) for all relapsed patients was decided by a multidisciplinary team (MDT). The follow-up commenced from the date of surgery. The period from the surgery data to death or the final follow-up was defined as OS; the period from the surgery data to the first recurrence, metastasis, death, or the final follow-up was defined as PFS. The final follow-up date was October 2023.

Statistical analysis

Statistical analyses were performed using R software (4.3.1) and SPSS (26.0). Referring to the previous study, we used 10 algorithms and 101 algorithm combinations to calculate Harrell’s concordance index (C-index) of each model in both training cohorts and validation cohorts (26). A C-index ≥0.7 is commonly considered indicative of a good prognostic model (27,28). We selected the most appropriate model and variables for further analysis considering the highest average C-index. The ‘scaleData’ function was used to normalize the risk scores, and the ‘surv_cutpoint’ function was used to calculate the optimal cut-off value. Kaplan-Meier (K-M) curves were plotted to analyze the relationship between variables and survival outcomes, and log-rank validation was employed for comparisons between subgroups. Univariate and multivariate Cox analyses assessed the prognostic impact of the risk scores and other parameters on survival. Differences between risk subgroups were compared using the Chi-squared/Fisher’s exact test. Statistical significance was defined as a two-tailed P value less than 0.05.

Results

Prognostic capacity of different inflammatory/nutritional indexes

A total of 219 SCLC patients were eligible for our study through rigorous screening. We investigated the prognostic role of 12 preoperative serum inflammatory/nutritional indexes in these patients. The median follow-up time for all patients was 3.5 years. K-M curves showed the predictive role of 12 inflammatory/nutritional indicators in the total cohort. It can be seen that higher PNI (P<0.001), LMR (P=0.009), PNR (P=0.002), lower NLR (P<0.001), SII (P<0.001), PIV (P=0.003), and SIRI (P<0.001) were significantly associated with a favorable prognosis in SCLC patients (Figure 1A-1G). However, PLR (P=0.08), AAPR (P=0.91), FAR (P=0.84), PAR (P=0.92), and HALP (P=0.34) had no significant effect on patient prognosis (Figure 1H-1L).

Figure 1.

Kaplan-Meier curves of SCLC patients’ OS according to 12 inflammatory/nutritional indexes in the total cohort. (A) PNI; (B) NLR; (C) SII; (D) LMR; (E) PNR; (F) PIV; (G) SIRI; (H) PLR; (I) AAPR; (J) FAR; (K) PAR; (L) HALP. AAPR, albumin-to-alkaline phosphatase ratio; FAR, fibrinogen-to-albumin ratio; HALP, hemoglobin, albumin, lymphocyte, and platelet; LMR, lymphocyte-to-monocyte ratio; NLR, neutrophil-to-lymphocyte ratio; OS, overall survival; PAR, platelet-albumin ratio; PNI, prognostic nutritional index; PNR, platelet-to-neutrophil ratio; PIV, pan-immune-inflammation value; PLR, platelet-to-lymphocyte ratio; SCLC, small cell lung cancer; SII, systemic immune-inflammatory index; SIRI, systemic inflammatory response index.

Model selection and markers identification

By integrating 101 combinations of machine learning algorithms and calculating the C-index for each model in the training and validation cohorts, we found that the RSF model was the best choice with the highest average C-index (0.784, with 0.861 for the training cohort and 0.706 for the validation cohort) (Figure 2A). As shown in Figure 2B, the error rate of this model stabilized when the ntree was around 800. The variable importance (VIMP) ranking method identified PNI, NLR, and SII as the top contributors. In addition, the RSF model constructed by excluding the HALP (importance <0) yields a higher C-index. Therefore, we calculated the risk score of each patient based on the remaining 11 indexes.

Figure 2.

Computational framework for machine learning algorithms and survival curves based on risk scores. (A) A synthesis of the computational framework led to the generation of a combination of 101 machine-learning algorithms. The C-index of each model is computed for both the training and validation cohorts and sorted by the average value. (B) Prediction error rate of RSF and importance ranking of features. (C,D) Kaplan-Meier curves of OS (C) and PFS (D) based on the risk scores of the RSF model with 11 variables in the total, training, and validation cohorts. AAPR, albumin-to-alkaline phosphatase ratio; C-index, concordance index; FAR, fibrinogen-to-albumin ratio; GBM, generalized boosted regression modeling; HALP, hemoglobin, albumin, lymphocyte, and platelet; LASSO, least absolute shrinkage, and selection operator; LMR, lymphocyte-to-monocyte ratio; NLR, neutrophil-to-lymphocyte ratio; OS, overall survival; PAR, platelet-albumin ratio; PFS, progression-free survival; PNI, prognostic nutritional index; PNR, platelet-to-neutrophil ratio; PIV, pan-immune-inflammation value; PLR, platelet-to-lymphocyte ratio; plsRcox, partial least squares regression for Cox; RSF, random survival forest; SII, systemic immune-inflammatory index; SIRI, systemic inflammatory response index; SuperPC, supervised principal components; SVM, support vector machine.

Prognostic ability of risk scores and relationship to clinicopathological features

Patients were categorized into high-risk and low-risk groups based on the median risk scores of the training and validation cohorts. Table 1 presents the association between the risk subgroups and the baseline information in the training cohort. We discovered that patients in the high-risk group had a significantly higher proportion of smokers and more advanced pathological N stage and tumor-node-metastasis (TNM) stage. There were no differences in age, gender, BMI, alcohol history, tumor location, pathological T stage, Ki-67 index, and visceral pleural invasion (VPI) between the two subgroups. K-M curves showed significantly better OS and PFS in the low-risk group across total, training, and validation cohorts (Figure 2C,2D). This indicates that our model constructed through the machine learning-based integration procedure can effectively distinguish the prognosis of patients in different risk groups, and the model is anticipated to be employed as a clinical prognostic stratification tool.

Table 1. Differences in clinicopathologic characteristics between high- and low-risk subgroups in the training cohort.

Characteristics	High-risk group (n=76)	Low-risk group (n=77)	P value
Age (years)			0.95
<60	30	30
≥60	46	47
Gender			0.12
Male	59	51
Female	17	26
BMI (kg/m2)			0.21
<24	28	36
≥24	48	41
Smoking status			0.04*
Yes	49	37
No	27	40
Alcohol			0.88
Yes	16	17
No	60	60
Tumor site			0.83
RUL	16	20
RML	5	6
RLL	21	22
LUL	11	12
LLL	23	17
T stage			0.67
1	33	39
2	33	29
3 & 4	10	9
N stage			0.002*
N0	28	46
N1	12	15
N2	36	16
TNM stage			<0.001*
Stage 1	21	40
Stage 2	16	19
Stage 3	39	18
Ki-67			0.29
<80	40	34
≥80	36	43
VPI			0.44
Yes	23	19
No	53	58

*, statistically significant two-sided P value <0.05. BMI, body mass index; LUL, left upper lobe; LLL, left lower lobe; N, lymph node; RLL, right lower lobe; RML, right middle lobe; RUL, right upper lobe; T, tumor; TNM, tumor-node-metastasis; VPI, visceral pleural invasion.

To further confirm the efficacy and sensitivity of risk scores in predicting OS and PFS, we depicted receiver operating characteristic (ROC) curves and calculated time-dependent C-indexes of risk scores, gender, age, BMI, TNM stage, and smoking status and alcohol history. For OS in the training cohort, the area under the curve (AUC) values at 1, 3, 5, and 7 years were 0.904, 0.922, 0.936, and 0.944, respectively (Figure 3A). In the validation cohort, the AUC values at 1, 3, 5, and 7 years were 0.672, 0.724, 0.749, and 0.805, respectively (Figure 3B). For PFS in the training cohort, the AUC values at 1, 3, 5, and 7 years were 0.799, 0.844, 0.873, and 0.898, respectively (Figure 3C). In the validation cohort, the AUC values were 0.760, 0.701, 0.755, and 0.703, respectively (Figure 3D). We were astonished to discover that the AUC value tended to rise over time, indicating considerable validity and robustness, particularly in forecasting the long-term prognosis of patients. Furthermore, compared with other clinical characteristics, the time-dependent C-index of the risk score was the greatest for both OS and PFS (Figure 3E-3H).

Figure 3.

Performance and consistency assessment of the RSF model. ROC curve and AUC value of OS and PFS in the training cohort and validation cohort (A-D). Time-dependent concordance index of risk scores and other clinical characteristics of OS and PFS in the training cohort and validation cohort (E-H). AUC, area under the curve; BMI, body mass index; OS, overall survival; PFS, progression-free survival; ROC, receiver operating characteristic; RSF, random survival forest.

Development, assessment, and validation of the nomogram

To construct a risk-score-based prediction model, we first screened three prognostic correlates by univariate Cox regression analysis: risk score [HR: 3.845, 95% confidence interval (CI): 2.985–4.952, P<0.001], smoking status (HR: 1.855, 95% CI: 1.158–2.971, P=0.01), and TNM stage (HR: 3.088, 95% CI: 1.974–4.830, P<0.001) (Figure 4A). Multivariate Cox regression confirmed that they were both independent prognostic factors [risk score (HR: 3.617, 95% CI: 2.736–4.782, P<0.001), smoking status (HR: 2.180, 95% CI: 1.023–4.643, P=0.043), and TNM stage (HR: 2.199, 95% CI: 1.237–3.910, P=0.007)] for OS (Figure 4B), while risk score (HR: 2.335, 95% CI: 1.861–2.930, P<0.001) and TNM stage (HR: 1.891, 95% CI: 1.137–3.146, P=0.01) were independent prognostic factors for PFS (Figure 4C,4D). To further predict the OS of patients, we established a nomogram by employing risk score, TNM stage, and smoking status as clinical features. The nomogram presents the scores corresponding to the characteristics of the diverse variables. The total score can be used to predict OS’s probability at 1, 3, 5, and 7 years (Figure 5A). The calibration curves demonstrated a satisfactory concordance between the survival probabilities projected by the nomogram and the actual observed values, suggesting that the model possesses a remarkably high predictive accuracy (Figure 5B). Furthermore, the decision curve analysis (DCA) curves indicated that the predictive model possesses a high utility in clinical decision-making (Figure 5C). In the validation cohort, we arrived at similar conclusions (Figure 5D-5F). In conclusion, the prognostic model integrating risk scores with smoking status and TNM stage can effectively predict the OS of SCLC patients and is anticipated to offer valuable information for clinical decision-making.

Figure 4.

Forest plot of univariate and multivariate Cox analysis of risk score and other clinical parameters related to OS (A,B) and PFS (C,D). BMI, body mass index; CI, confidence interval; OS, overall survival; PFS, progression-free survival; PI, pleural invasion.

Figure 5.

Nomogram, calibration curves, and the DCA curves of the model based on risk group, smoking status, and TNM stage at 1, 3, 5 and 7 years. (A,D) The nomogram in the training cohort (A) and validation cohort (D). (B,E) The calibration curves in the training cohort (B) and validation cohort (E). (C,F) The DCA curves in the training cohort (C) and validation cohort (F). **, P<0.01; ***, P<0.001. DCA, decision curve analysis; OS, overall survival; TNM, tumor-node-metastasis.

Identification of PNI and NLR as independent prognostic indicators

Since the risk scores calculated by machine learning encompassed more variables, we conducted further analysis to determine the most valuable independent prognostic inflammatory/nutritional indexes based on the importance ranking. The top three ranked indexes (PNI, NLR, and SII) were selected for multivariate Cox analysis. The results showed that PNI (HR: 0.955, 95% CI: 0.916–0.995, P=0.03) and NLR (HR: 1.383, 95% CI: 1.027–1.862, P=0.03) independently predicted OS in SCLC patients receiving surgery (Figure 6A,6B). Finally, we depicted histograms of the distribution of patients with diverse values and discovered that the optimal thresholds for PNI and NLR were 50.60 and 1.99, respectively (Figure 6C,6D).

Figure 6.

Cox forest maps and PNI/NLR threshold histograms. Forest plot of univariate (A) and multivariate (B) Cox analysis of PNI, NLR, SII, smoking status, and TNM stage related to OS. The histogram of patient distribution showed that the optimal thresholds for PNI and NLR were 50.60 and 1.99 (C,D), respectively. Red and blue boxes/lines indicate high and low groups divided by the optimal cutoff value for the variable. NLR, neutrophil-to-lymphocyte ratio; OS, overall survival; PNI, prognostic nutritional index; SII, systemic immune-inflammatory index; TNM, tumor-node-metastasis.

Discussion

To our knowledge, this is the first study to comprehensively explore the prognostic role of multiple serologic markers in surgically treated SCLC patients. PNI, LMR, PNR, NLR, SII, PIV, and SIRI were identified as prognostic indicators. These indexes are readily obtainable from clinical tests, so they could potentially be routine prognostic predictors for thoracic surgeons. Considering the “overlap” of the “mathematical formulas” of these indicators (for instance, lymphocytes were utilized to calculate several inflammation indicators), to minimize multiple covariates and enhance clinical utility, we calculated the importance ranking of these variables. Subsequently, we selected the top three variables in the multivariate Cox analysis and found that PNI and NLR are independent prognostic markers for surgically treated SCLC patients.

Though the indications for SCLC surgical resection remain ambiguous, much research suggests that SCLC patients at various stages may potentially benefit from surgery (29,30). Low-dose computed tomography (LDCT) screening programs can detect early-stage lung cancer and may increase the pool of operable SCLC patients (31,32). A recent study demonstrated that endobronchial elastography with specific tumor markers can accurately determine pathological subtypes (33). Consequently, the clinical significance of surgery for SCLC patients will rise, and thoracic surgery will be more inclined to perform operations with the aim of “curing”.

While one study reports that adjuvant therapy improves prognosis in early-stage SCLC (34), another finds limited benefit (35). In our cohort, since there was no consistent indication for adjuvant therapy in patients with early-stage SCLC, variations in the overall treatment regimens might have influenced clinical outcomes. However, in patients without a definite pathological diagnosis, it is challenging to ascertain the requirement for postoperative adjuvant therapy through preoperative imaging. For LS-SCLC patients treated with surgery, the indication for postoperative adjuvant therapy remains ambiguous. Consequently, this study did not concentrate on the prognostic influence of postoperative adjuvant therapy. The conclusions derived from this approach might prompt some patients to reevaluate their treatment choices (such as opting for puncture or direct surgery) to achieve a more rational treatment plan and anticipated survival.

Though we found several inflammatory markers associated with prognosis in surgically treated SCLC patients, machine-learning-based importance ranking highlighted PNI and NLR as the most valuable. PNI reflects the organism’s nutritional status (36), while NLR is closely associated with the inflammatory response degree (37). Studies regarding their prognostic role in SCLC are not infrequent, yet a consensus has not been attained. For instance, the predictive effectiveness of PNI in SCLC patients who only receive chemotherapy is debatable (38,39). Although numerous studies have discovered that NLR is linked to a poor prognosis in LS-SCLC, the optimal threshold remains indistinct (40-42). There were also conflicting reports on the prognostic value of NLR in ES-SCLC (43,44).

Given the scarcity of SCLC and the ambiguous indication for surgery, few studies have been based on surgical cohorts. Considering the requirement of postoperative adjuvant therapy in SCLC, combining preoperative and pre-adjuvant therapy biomarkers might be of great significance. Nowadays, PD-L1-based immunotherapy has been extensively investigated. The IMpower133 and CASPIAN studies have verified that combining PD-L1 inhibitors with chemotherapy in the first-line treatment of ES-SCLC can attain a higher OS with manageable adverse effects compared to chemotherapy alone (45,46). The ADRIATIC study demonstrated that, compared with the placebo, durvalumab significantly prolonged the OS and PFS of LS-SCLC patients undergoing concurrent chemoradiotherapy (47). PNI functions as a predictor for immune-related adverse events (IRAE) and OS in SCLC patients undergoing immunotherapy, suggesting a close association with an active immune status (18,48,49), similar to the derived NLR (dNLR) (50). In light of our findings, future studies should investigate whether postoperative adjuvant immunochemotherapy offers greater survival benefits for patients with high PNI or low NLR.

Of note, the study did not find age, gender, BMI, tumor location, and pathological data such as the Ki-67 proliferation index and VPI to be prognostic factors for long-term survival. It implies that these factors may not limit the indication for surgery in early-stage SCLC and prognostic judgments based on our model.

There are several limitations. Firstly, this retrospective study conducted at a single institution with a large sample size inevitably introduces population selection bias, affecting the findings’ generalizability and external validity. The medical care level at different times may affect the survival rate of patients. In addition, although this study identified significant predictive effects of several peripheral blood markers, we may have overlooked some clinically relevant factors. Thirdly, the robustness and accuracy of the machine learning models need further validation in an independent cohort. Prospective studies and basic research on the tumor microenvironment are required to uncover more complex pathways.

Conclusions

This study identifies robust, accessible prognostic biomarkers in surgically treated SCLC patients. Incorporating these cost-effective, routine inflammatory and nutritional biomarkers into preoperative risk stratification demonstrates significant clinical utility.

Supplementary

The article’s supplementary files as

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. The medical ethics committee of The Affiliated Hospital of Qingdao University has approved this research (approval No. QYFY WZLL 29635), and each participant provided the informed consent.

Data Sharing Statement

Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-182/dss

tlcr-14-08-2983-dss.pdf ^{(71.5KB, pdf)}

DOI: 10.21037/tlcr-2025-182