Radiomics-clinical integration guides prophylactic cranial irradiation decisions in limited-stage small cell lung cancer: a brain metastasis risk stratification model

Introduction

In limited-stage small cell lung cancer (LS-SCLC), brain metastasis (BM) is a major driver of treatment failure and mortality, with up to 50% of patients developing BM within 2 years of diagnosis (1,2). Despite the proven efficacy of prophylactic cranial irradiation (PCI) in reducing BM incidence, its use is increasingly debated due to potential neurocognitive toxicities (3,4). The optimal strategy for identifying patients who would benefit most from PCI remains an open scientific question, as current clinical criteria fail to accurately stratify BM risk.

The purpose of this investigation is to develop a predictive model for BM in LS-SCLC by integrating chest computed tomography (CT) radiomics with clinical parameters. While prior studies have explored risk factors for BM, they have largely relied on clinical variables alone, which lack the precision needed for personalized treatment decisions (5-7). In the current clinical context, where the balance between treatment efficacy and patient quality of life is paramount, a more accurate predictive model could guide the selective use of PCI, ensuring its benefits are maximized while minimizing unnecessary exposure.

To address this challenge, we hypothesize that combining chest CT radiomics, which captures tumor heterogeneity and microenvironmental features, with relevant clinical parameters will enhance BM risk stratification (8). By leveraging advanced radiomic techniques and machine learning algorithms, we aim to develop and validate a comprehensive predictive model that can effectively identify high-risk patients who may benefit from PCI. This approach seeks to provide a data-driven solution to the ongoing debate surrounding PCI use in LS-SCLC, ultimately improving patient outcomes through more personalized treatment strategies. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-326/rc).

Methods

Patient characteristics

This retrospective analysis evaluated individuals with LS-SCLC treated at Tianjin Medical University Cancer Institute & Hospital from September 2013 to March 2021. The study enrolled subjects meeting the following criteria: (I) histologically confirmed SCLC diagnosis; (II) LS-SCLC staging through imaging with baseline brain magnetic resonance imaging (MRI)/CT; (III) completion of thoracic radiotherapy (TRT). Exclusion parameters included: (I) unavailable clinical records or simulation CT images; (II) prior radiotherapy (RT) history; (III) incomplete TRT course. BM-free survival (BMFS) was calculated from treatment initiation to intracranial progression, death, or last follow-up. Using R software, the 141 eligible cases were randomly allocated into training (n=98) and validation (n=43) cohorts. External validation was conducted on an independent cohort comprising 24 patients from Hebei Province Cangzhou Hospital of Integrated Traditional and Western Medicine—a geographically distinct institution where participants underwent consistent TRT protocols while meeting equivalent eligibility criteria. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Tianjin Medical University Cancer Institute & Hospital (No. bc20240049) and individual consent for this retrospective analysis was waived. Hebei Province Cangzhou Hospital of Integrated Traditional and Western Medicine was also informed and agreed to the study. This multicenter verification strategy enabled robust evaluation of the model’s generalizability across heterogeneous clinical settings.

Clinical factors

The analysis incorporated demographic variables (gender, age at diagnosis, smoking history), functional status (Karnofsky Performance Status), tumor staging (T/N classification), serological biomarkers [neuron-specific enolase (NSE), pro-gastrin-releasing peptide (Pro-GRP)], and immunohistochemical profiles [chromogranin A (CgA), thyroid transcription factor-1 (TTF-1), Ki-67]. Treatment protocols (chemotherapy regimens, PCI administration) were documented, though PCI was excluded from model construction to isolate its predictive value. After developing the model, we compared brain metastasis-free survival (BMFS) between high-risk and low-risk groups with and without PCI.

Pre-treatment hematologic indices [red blood cell (RBC), white blood cell (WBC), platelets (PLT), hemoglobin (HB), neutrophils (NEUT), lymphocytes (LY)] and tumor marker levels were obtained from blood tests prior to simulation CT. Based on these results, we calculated the pan-immune-inflammation value [PIV = lymphocyte count/(neutrophil count × platelet count × monocyte count)], systemic immune-inflammation index (SII = platelet count × neutrophil count/lymphocyte count), lymphocyte/monocyte ratio (LMR), platelet lymphocyte ratio (PLR), and neutrophil lymphocyte ratio (NLR). Blood counts before, during, and after RT were also included in the clinical profile. The method for calculating the rate of change of immune cells during RT is detailed in the Appendix 1. Furthermore, albumin (ALB) and lactate dehydrogenase (LDH) levels before and after RT were included in the clinical characteristics.

All patients received localized CT scans (neck, chest, upper abdomen) encompassing tumor targets and relevant organs at risk (OARs) per clinical requirements (9). Gross tumor volume (GTV) delineation integrated post-neoadjuvant chemotherapy imaging and pathological findings, capturing residual disease (primary tumor + involved lymph nodes) prior to radiotherapy. Clinical target volume (CTV) extended beyond GTV to cover potential microscopic spread, while planning target volume (PTV) incorporated internal target expansion accounting for physiological motion and setup uncertainties. OAR classification adhered to established TRT atlases. Lung dosimetry considered bilateral lungs (excluding GTV, hilar structures, and trachea/main bronchus) as a unified organ. Cardiac contouring followed pericardial boundaries from pulmonary artery bifurcation to apex. Treatment planning employed a collapsed cone convolution algorithm (3×3 mm grid resolution), with GTV/PTV parameters and dose-volume metrics [mean lung dose (MLD), mean heart dose (MHD) and mean body dose (MBD)] extracted from the Pinnacle system. Notably, MBD quantification spanned skull base to L1 inferior margin. The estimated dose of radiation to immune cells (EDRIC) was calculated using Jin et al.’s model (10), subsequently optimized by Ladbury et al. (11), incorporating mean lung/heart/body doses and fractionation parameters:

EDRIC=0.12×MLD+0.08×MHD+[0.45+0.35×0.85×((#offractions)/45)(1/2)]×MBD[1]

Radiomic features

All patients underwent contrast-enhanced CT simulation using a Philips Brilliance scanner (120 kV, 200–250 mAs, 3 mm slice thickness) with standardized contrast injection protocols (2.0–3.0 mL/s, total 80–100 mL). Acquired images (512×512 matrix) were reconstructed in Pinnacle v3.2.0.27, where two radiation oncologists independently delineated GTV using dual-window settings: lung window (−150 to −1,150 HU) for parenchymal lesions and mediastinal window (−135 to 215 HU) for nodal involvement.

Subsequent radiomic analysis employed Three-dimensional Slicer software (v4.11) to process the manually defined regions of interest (ROIs). A comprehensive feature set of 1,037 parameters was extracted, encompassing first-order intensity statistics, three-dimensional morphological descriptors, and advanced texture metrics derived from five distinct matrices: gray-level co-occurrence matrix (GLCM), gray-level size zone matrix (GLSZM), gray-level run-length matrix (GLRLM), gray-level dependence matrix (GLDM), and neighborhood gray tone difference matrix (NGTDM). Spatial frequency characteristics were further captured through 3D wavelet transformations, generating eight directional decomposition sets via combinations of low- and high-pass filtering (LLL, LLH, HHL, etc.). Feature selection leveraged the minimum redundancy maximum relevance (mRMR) algorithm, ultimately identifying ten discriminative predictors.

Statistical analyses

Baseline characteristics demonstrated comparable distributions between training and validation cohorts (Table 1), verified through chi-square tests for categorical variables and Mann-Whitney U tests for continuous measures (α=0.05). Predictor variables were selected through univariate logistic regression (LR) screening (P<0.2 threshold) combined with radiomic features optimized via mRMR algorithm (https://github.com/smazzanti/mrmr). Multimodal predictive frameworks were engineered using four machine learning architectures: LR, support vector machines (SVM), random forest (RF), and eXtreme Gradient Boosting (XGB), evaluating radiomics-only, clinical-only, and combined models. The performance of the models was evaluated by calculating the area under the receiver operating characteristic curve (ROC) (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) in the training and internal validation groups. ROC-derived optimal thresholds stratified patients into distinct risk cohorts (high vs. low). Subsequent Kaplan-Meier methodology quantified survival divergence between these risk strata, with log-rank testing for significance confirmation implemented in IBM SPSS Statistics version 23. Prognostic visualization tools including nomograms were constructed using R software (version 4.3.1; http://www.Rproject.org), along with Zstats v1.0 (www.zstats.net). Additional statistical analyses were conducted using various Python 3.11.5 packages, including numpy, scikit-learn, lifelines, pandas, matplotlib, xgboost, openpyxl, and SciPy.

Results

The workflow of this study is detailed in Figure 1.

Figure 1 Workflow of the study. AUC, area under the ROC curve; BMFS, brain metastasis-free survival; K-M curve, Kaplan-Meier curve; LS-SCLC, limited-stage small cell lung cancer; LR, logistic regression; mRMR, minimum redundancy maximum relevance; RF, random forest; ROC, the receiver operating characteristic curve; SVM, support vector machine; XGB, eXtreme gradient boosting.

Clinical characteristics

The cohort comprised 141 LS-SCLC patients (109 males, 32 females) with median ages of 62 [24–79] and 59 [40–81] years respectively. Intracranial metastases occurred in 46 cases (32.6%). RT regimens included: 43 patients (30.5%) receiving 60 Gy PTV irradiation, and 98 (69.5%) undergoing combined 60 Gy GTV with 54 Gy PTV irradiation. Current/former smokers accounted for 77.3% (n=109), while 65.3% (n=92) received PCI. Using 7:3 randomization, baseline characteristics showed no significant intergroup differences between training (n=98) and internal validation groups (n=43) (Table 1). PCI was administered at a dose of 25 Gy in 10 fractions for eligible patients who achieved complete response (CR) or partial response (PR) after chemoradiotherapy. The decision to administer PCI was based on clinical judgment, considering factors such as performance status, age, and patient preference. The chemotherapy regimens included: EP (etoposide + cisplatin) used in 52.48% of patients, EC (etoposide + carboplatin) used in 43.26% of patients primarily for those with contraindications to cisplatin, and other personalized regimens used in 4.26% of patients, including combinations with paclitaxel or irinotecan.

Radiomics and clinical feature selection

Simulation CT imaging yielded 1,037 initial radiomic features (Figure S1), with mRMR algorithm selecting 10 discriminative markers: 7 GLCM parameters, along with single features from GLSZM, GLRLM and GLDM domains.

Clinical variables were screened through univariate LR (P<0.20 threshold), identifying 6 predictive factors: pre-RT WBC (WBC1) and LMR; post-RT HB (HB3); along with platelet, HB, and WBC change ratios during treatment (PLTR1, HBR2, WBCR4).

Models for BM

Based on the 10 radiomics features and 6 clinical features screened, we constructed the radiomics model (LR_R), and the clinical factor model (LR_C) using LR. We calculated the RadScore based on LR_R (formulas are in the Appendix 2), and the RadScore and the 6 clinical features were included in the combined model (LR_RC). Furthermore, SVM, RF and XGB were used to construct machine learning models for the radiomics model (SVM_R, RF_R and XGB_R), the clinical factors model (SVM_C, RF_C and XGB_C), the combined model (SVM_RC, RF_RC and XGB_RC). The AUC values of models of this study are displayed in Table 2.

As clearly presented in Table 3, our multifactorial LR analysis, also referred to as the LR_RC model, identified six pivotal clinical factors and RadScore that emerged as independent and statistically significant predictors of BM. Notably, HB3, with a P value of 0.005 and an odds ratio (OR) of 5.33 [95% confidence interval (CI): 1.65 to 17.19], PLTR1 (P=0.03, OR: 3.64; 95% CI: 1.14 to 11.65) and RadScore (P=0.002, OR: 2.04; 95% CI: 1.30 to 3.19) stood out as key indicators, each contributing significantly to the predictive model. However, the tumour serum markers proGRP (P=0.37) and NSE (P=0.51), and the immunohistochemical markers Ki-67 (P=0.75), CgA (P=0.34), and TTF-1 (P=0.82) were not included in the multifactorial logistic analyses, due to the threshold of the univariate logistic analyses set at P=0.20.

As depicted in Table S1, the multifactorial LR analysis conducted within LR_C encompassed six clinical factors, among which pre-RT HB [P=0.003, with an OR (95% CI) of 5.02 (1.75 to 14.40)] emerged as statistically independent predictors of BM.

Model evaluation

Models with an AUC greater than 0.7 for both the training and validation group were considered to have predictive value. In the training set, the internal validation set and the external validation set, the combined model constructed using LR met these requirements, but the SVM, RF and XGB machine learning models did not (Table 2). The combined model constructed using LR achieved good AUC in the training set (0.831), the internal validation set (0.831) and the external validation set (0.863). In the SVM machine learning model, although all three models achieved high AUCs in the training group, they performed poorly in the validation group, including only 0.638 in the internal validation set. In RF and XGB machine learning models, all 6 models achieved better AUC in the training group, which may have problems such as overfitting, while in the internal validation group, the AUC underperformed. The AUC, accuracy, sensitivity, specificity, PPV, and NPV of the models are summarized in Table S2. The ROC curves, calibration curves and decision curve analysis (DCA) curves of the LR combined model in the training set, the internal validation set and the external validation are shown in Figure 2. We created nomograms for the LR combined model, as shown in Figure 3 and Figure S2.

Figure 2 ROC curves (A-C), calibration curves (D-F) and DCA curves (G-I) of LR combined model in training set, internal validation set, external validation set. AUC, area under the ROC curve; DCA, decision curve analysis; LR, logistic regression; ROC, the receiver operating characteristic curve.

Figure 3 Nomogram of the LR_RC model. HB3, hemoglobin post radiotherapy; HBR2, ratio of change in hemoglobin before and after radiotherapy; LMR, lymphocyte-to-monocyte ratio; PLTR, ratio of change in platelet before and after radiotherapy; RadScore, Radiomics scores; WBC1, white blood cell before radiotherapy; WBCR4, ratio of change in WBC before and after radiotherapy; WBC, white blood cell.

PCI and BMFS

Using the LR_RC model, we identified distinct high-risk and low-risk groups for BM. To evaluate whether the groups with and without PCI were comparable, a new comparative analysis has been conducted to assess baseline characteristics and key prognostic variables (Table S3). Figure 4 presents Kaplan-Meier curves plotted independently for these two groups. Our findings revealed that in the high-risk cohort, the implementation of PCI led to a marked improvement in BMFS (P<0.001). Conversely, within the low-risk group, no statistically significant difference was observed in BMFS between patients who underwent PCI and those who did not (P=0.13).

Figure 4 K-M curves for brain metastasis-free survival. (A) In the high-risk population, PCI significantly improved BMFS. (B) In the low-risk population, there was no statistically significant difference in the effect of whether or not PCI had an effect on BMFS. The values “0” and “1” indicate whether PCI was performed, with “0” representing no PCI and “1” representing PCI performed. BMFS, brain metastasis-free survival; CI, confidence interval; HR, hazard ratio; K-M curve, Kaplan-Meier curve; PCI, prophylactic cranial irradiation.

Discussion

To the best of our knowledge, this is the first study to utilize chest CT radiomics in combined with clinical factors to predict BM in LS-SCLC. Notably, we have innovatively incorporated indicators of immune cell alterations before and after RT into our clinical factors. We constructed LR, SVM, RF, and XGB models to identify the model with the highest predictive efficacy, which is anticipated to guide clinicians in selecting appropriate treatment options.

The brain is a frequent site of metastasis in SCLC, and patients with SCLC and BM generally have a poor prognosis. PCI, a widely used clinical technique, is becoming increasingly controversial in SCLC (4). The EORTC study confirmed that PCI reduces the incidence of symptomatic BM while prolonging disease-free and overall survival (12). However, recent studies indicate that in patients with ES-SCLC, PCI does not lead to improved overall survival compared to regular MRI surveillance and may cause neurocognitive dysfunction (3,13). This discrepancy may stem from the EORTC study’s failure to screen patients for BM prior to PCI.

Our predictive models for BM demonstrated robust performance, with the LR_RC model achieving AUC values of 0.831, 0.831, and 0.863 in the training, internal validation, and external validation sets, respectively. Using this model, we identified distinct high-risk and low-risk groups for BM, as illustrated by the Kaplan-Meier curves in Figure 4. In the high-risk cohort, PCI significantly improved BMFS (P<0.001), while in the low-risk group, no statistically significant difference was observed between patients who received PCI and those who did not (P=0.13). However, a strong trend towards benefit from PCI was noted in the low-risk group, with similar hazard ratio (HR) =0.270 for high-risk vs. HR =0.225 for low-risk. Given the relatively small sample size in the low-risk group, our study may have been underpowered to detect a true treatment effect. Therefore, it would be premature to definitively conclude that PCI does not alter BMFS in low-risk patients. The similarity in HRs suggests potential clinical benefits across both risk groups, warranting further validation in larger cohorts. Additionally, potential correlations between PCI use and other prognostic variables should be considered. Our findings highlight the need for additional research to refine personalized PCI strategies and caution against making definitive clinical recommendations based on current data.

Steven Paget’s ‘seed-and-soil’ theory suggests that successful metastasis to distant organs requires not only tumor cells with metastatic potential (the “seed”) but also the appropriate innate characteristics of the host organ (the “soil”) (14). In this context, we aimed to predict the occurrence of BM using chest CT radiomics. Several studies have been published on this topic in non-small cell lung cancer (NSCLC), highlighting the utility of chest CT radiomics in predicting BM. For instance, Xu et al. (15) investigated the predictive capacity of chest CT radiomics for BM in patients with stage III/IV anaplastic lymphoma kinase (ALK)-positive NSCLC. This study included 132 patients, 27 of whom developed BM prior to treatment. One imaging histological feature was identified to predict pre-treatment BM, achieving AUCs of 0.687 and 0.642 for the training and test group, respectively, demonstrating good predictive performance for BM. In another study, Sun et al. (16) analyzed 124 patients with locally advanced NSCLC post-radical resection, using preoperative chest CT radiomics to predict the first occurrence of BM. The model attained a C-index of 0.889in the training cohort and 0.853 in the validation cohort. Zheng et al. (17) employed PET/CT chest radiographic histology to predict the first occurrence of BM in patients with resected I-IIIA NSCLC. This study included 203 surgically treated patients, with the C-index of the PET/CT radiomic model being 0.911, 0.825, and 0.800 in the training, internal validation, and external validation cohorts, respectively. Additionally, Jiang et al. (18) developed a chest CT radiomics model to predict subsequent BM in 75 patients with ALK-rearranged NSCLC receiving crizotinib. Their nomogram, which combined smoking history and radiomic features, demonstrated strong BMFS estimation performance, yielding C-indices of 0.762 and 0.724 in the training and test group. However, despite many studies on NSCLC BM prediction, few studies currently predict the occurrence of SCLC BM, and our model achieved a comparable predictive efficacy. In a recently published study, Zheng et al. (19) investigated the relationship between radiomic score (RS) and the benefit of PCI. Their findings indicated that patients with a high RS derived greater benefit from PCI. In contrast, the radiomic-clinical BM prediction model developed in our study integrates several clinical factors, including changes in blood cell values before and after RT, to more effectively identify patients who may benefit from PCI. Our results demonstrate that when our model differentiates between high-risk and low-risk populations, PCI significantly improves BMFS in the high-risk group. However, our radiomic model alone showed limited predictive performance in external validation (AUC =0.663), which may be attributed to several factors. Although radiomic features reflect tumor heterogeneity and microstructural characteristics on CT images, they may not fully capture the complex biological mechanisms underlying BM in LS-SCLC. Additionally, limitations such as a relatively small external validation cohort, imaging heterogeneity across institutions, and potential noise in feature extraction may have affected the model’s reproducibility and generalizability.

Currently, blood-based inflammatory biomarkers are increasingly utilized in prognostic modeling, as RT inevitably influences systemic immune dynamics. In our LR_RC model, we identified post-RT HB and PLTR as independent predictors of BM. These findings highlight the potential role of physiological changes following RT in influencing metastatic risk. Importantly, accumulating evidence supports a pro-metastatic role for platelets in cancer progression. Feinauer et al. demonstrated that platelets facilitate the arrest of tumor cells within brain capillaries, thereby promoting the formation of BM (20). This mechanism underscores the relevance of PLTR as a surrogate marker of platelet-mediated tumor dissemination. Additionally, our model revealed that elevated post-RT HB levels—not pre-treatment values—were significantly associated with increased BM risk. The underlying mechanisms remain incompletely understood but may involve alterations in blood rheology, such as increased viscosity, which could create a favorable microenvironment for metastatic seeding. Higher HB levels might also enhance angiogenesis, further supporting tumor cell proliferation and spread. It is worth noting that clinical T and N staging, although not selected for further analysis in our model due to their p-values exceeding 0.2 in univariate analysis, may still hold potential significance in the broader context of prognostic modeling for small cell lung cancer (Table S4). Notably, although we included EDRIC—an immune-related prognostic index previously validated in patients with locally advanced esophageal cancer, stage III NSCLC, locally advanced NSCLC, and LS-SCLC treated with definitive concurrent chemoradiotherapy—in our analysis, no significant association was found between EDRIC and BM risk (6,21-23). This may reflect the highly aggressive nature of SCLC, in which EDRIC fails to fully capture RT-induced immune modulation or its impact on metastatic potential. These observations emphasize the importance of integrating both radiomic features and treatment-related physiological changes into predictive models. Further mechanistic studies are warranted to validate these associations and explore therapeutic strategies aimed at modulating these pathways to reduce the risk of BM.

Considering the highly malignant and highly aggressive nature of SCLC, serum tumour markers including NSE and Pro-GRP, as well as immunohistochemical markers such as CgA, TTF-1 and Ki-67, are also considered clinical factors. However, these markers were excluded from the prediction model due to the univariate analysis revealing no significant correlation with BM. Previous study has shown no significant correlation between NSE, pro-GRP levels and BM in NSCLC patients (24). However, in SCLC, several studies confirmed that pro-GRP was considered to be a predictor of BM incidence (25,26). Our study failed to identify a significant correlation between tumour markers and BM, potentially due to the fact that all patients in our cohort underwent TRT, whereas not all of them received PCI, unlike the patient population in the aforementioned study. Liu et al. (7) demonstrated that mLDH levels during treatment could predict the occurrence of BM in LS-SCLC patients undergoing TRT and PCI (5). However, in the current study, the univariate LR analysis conducted to compare LDH and BM yielded a P value of 0.28, failing to surpass the established significance threshold of P<0.2. Consequently, this variable was excluded from the subsequent modeling process.

It is worth noting that our study population was treated prior to the widespread adoption of consolidative immunotherapy with durvalumab, as recommended by recent guidelines following the ADRIATIC trial (27). Emerging data suggest that immunotherapy may alter metastatic patterns and potentially reduce the incidence of BM. While our model was developed in a pre-immunotherapy setting, it may still provide a useful baseline for identifying patients at low risk of BM who could benefit from de-escalated prophylactic strategies. Future studies incorporating patients treated with immunotherapy are warranted to evaluate the generalizability of our model in modern treatment paradigms.

Inevitably, there are some limitations to this study. Firstly, due to the retrospective study design, our results may be affected by selection bias and potential confounders. Secondly, the model constructed in this study was constructed based on single-center data and validated with only a small amount (24 cases) of independent center data, which has some limitations, and we will conduct a multi-center retrospective study in the future to further optimize the model for better prediction.

Conclusions

Combined clinical factor model based on LR model achieved good predictive performance, and PCI significantly reduced BMFS in the high-risk group of BM, which helps clinical decision-making.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-326/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-326/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-326/prf

Funding: This study was funded by National Natural Science Foundation of China (No. 82460580), Natural Science Foundation of Xinjiang Uygur Autonomous Region of China (No. 2023D01A55), and Tianjin Key Medical Discipline (Specialty) Construction Project (No. TJYXZDXK-009A).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-326/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. The study was approved by the Ethics Committee of Tianjin Medical University Cancer Institute & Hospital (No. bc20240049) and individual consent for this retrospective analysis was waived. Hebei Province Cangzhou Hospital of Integrated Traditional and Western Medicine was also 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. Megyesfalvi Z, Tallosy B, Pipek O, et al. The landscape of small cell lung cancer metastases: Organ specificity and timing. Thorac Cancer 2021;12:914-23. [Crossref] [PubMed]
2. Megyesfalvi Z, Gay CM, Popper H, et al. Clinical insights into small cell lung cancer: Tumor heterogeneity, diagnosis, therapy, and future directions. CA Cancer J Clin 2023;73:620-52. [Crossref] [PubMed]
3. Takahashi T, Yamanaka T, Seto T, et al. Prophylactic cranial irradiation versus observation in patients with extensive-disease small-cell lung cancer: a multicentre, randomised, open-label, phase 3 trial. Lancet Oncol 2017;18:663-71. [Crossref] [PubMed]
4. Gaebe K, Erickson AW, Li AY, et al. Re-examining prophylactic cranial irradiation in small cell lung cancer: a systematic review and meta-analysis. EClinicalMedicine 2024;67:102396. [Crossref] [PubMed]
5. Ueki K, Matsuo Y, Kishi N, et al. Usefulness of pro-gastrin-releasing peptide as a predictor of the incidence of brain metastasis and effect of prophylactic cranial irradiation in patients with limited-stage small-cell lung cancer. J Radiat Res 2022;63:636-45. [Crossref] [PubMed]
6. Yu Y, Fu P, Jin JY, et al. Impact of effective dose to immune cells (EDIC) on lymphocyte nadir and survival in limited-stage SCLC. Radiother Oncol 2021;162:26-33. [Crossref] [PubMed]
7. Liu J, Wu D, Shen B, et al. Serum lactate dehydrogenase predicts brain metastasis and survival in limited-stage small cell lung cancer patients treated with thoracic radiotherapy and prophylactic cranial irradiation. Strahlenther Onkol 2022;198:1094-104. [Crossref] [PubMed]
8. Paget S. The distribution of secondary growths in cancer of the breast. 1889. Cancer Metastasis Rev 1989;8:98-101.
9. Ganti AKP, Loo BW, Bassetti M, et al. Small Cell Lung Cancer, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2021;19:1441-64. [Crossref] [PubMed]
10. Jin JY, Hu C, Xiao Y, et al. Higher Radiation Dose to Immune System is Correlated With Poorer Survival in Patients With Stage III Non–small Cell Lung Cancer: A Secondary Study of a Phase 3 Cooperative Group Trial (NRG Oncology RTOG 0617). Int J Radiat Oncol Biol Phys 2017;99:S151-2.
11. Ladbury CJ, Rusthoven CG, Camidge DR, et al. Impact of Radiation Dose to the Host Immune System on Tumor Control and Survival for Stage III Non-Small Cell Lung Cancer Treated with Definitive Radiation Therapy. Int J Radiat Oncol Biol Phys 2019;105:346-55. [Crossref] [PubMed]
12. Slotman B, Faivre-Finn C, Kramer G, et al. Prophylactic cranial irradiation in extensive small-cell lung cancer. N Engl J Med 2007;357:664-72. [Crossref] [PubMed]
13. Wolfson AH, Bae K, Komak R, et al. Primary Analysis of a Phase II Randomized Trial Radiation Therapy Oncology Group (RTOG) 0212: Impact of Different Total Doses and Schedules of Prophylactic Cranial Irradiation on Chronic Neurotoxicity and Quality of Life for Patients with Limited Disease Small Cell Lung Cancer. Int J Radiat Oncol Biol Phys 2011;81:77-84. [Crossref] [PubMed]
14. Chu X, Gong J, Yang X, et al. A "Seed-and-Soil" Radiomics Model Predicts Brain Metastasis Development in Lung Cancer: Implications for Risk-Stratified Prophylactic Cranial Irradiation. Cancers (Basel) 2023;15:307. [Crossref] [PubMed]
15. Xu X, Huang L, Chen J, et al. Application of radiomics signature captured from pretreatment thoracic CT to predict brain metastases in stage III/IV ALK-positive non-small cell lung cancer patients. J Thorac Dis 2019;11:4516-28. [Crossref] [PubMed]
16. Sun F, Chen Y, Chen X, et al. CT-based radiomics for predicting brain metastases as the first failure in patients with curatively resected locally advanced non-small cell lung cancer. Eur J Radiol 2021;134:109411. [Crossref] [PubMed]
17. Zheng Z, Wang J, Tan W, et al. (18)F-FDG PET/CT radiomics predicts brain metastasis in I-IIIA resected Non-Small cell lung cancer. Eur J Radiol 2023;165:110933. [Crossref] [PubMed]
18. Jiang Y, Wang Y, Fu S, et al. A CT-based radiomics model to predict subsequent brain metastasis in patients with ALK-rearranged non-small cell lung cancer undergoing crizotinib treatment. Thorac Cancer 2022;13:1558-69. [Crossref] [PubMed]
19. Zheng X, Liu K, Gao Z, et al. Predicting overall survival and prophylactic cranial irradiation benefit in small cell lung cancer patients: a multicenter cohort study. BMC Cancer 2024;24:1507. [Crossref] [PubMed]
20. Feinauer MJ, Schneider SW, Berghoff AS, et al. Local blood coagulation drives cancer cell arrest and brain metastasis in a mouse model. Blood 2021;137:1219-32. [Crossref] [PubMed]
21. McCall NS, McGinnis HS, Janopaul-Naylor JR, et al. Impact of radiation dose to the immune cells in unresectable or stage III non-small cell lung cancer in the durvalumab era. Radiother Oncol 2022;174:133-40. [Crossref] [PubMed]
22. Xu C, Jin JY, Zhang M, et al. The impact of the effective dose to immune cells on lymphopenia and survival of esophageal cancer after chemoradiotherapy. Radiother Oncol 2020;146:180-6. [Crossref] [PubMed]
23. Yin X, Luo J, Xu C, et al. Is a higher estimated dose of radiation to immune cells predictive of survival in patients with locally advanced non-small cell lung cancer treated with thoracic radiotherapy? Radiother Oncol 2021;159:218-23. [Crossref] [PubMed]
24. Li BT, Lou E, Hsu M, et al. Serum Biomarkers Associated with Clinical Outcomes Fail to Predict Brain Metastases in Patients with Stage IV Non-Small Cell Lung Cancers. PLoS One 2016;11:e0146063. [Crossref] [PubMed]
25. Yonemori K, Sumi M, Fujimoto N, et al. Pro-gastrin-releasing peptide as a factor predicting the incidence of brain metastasis in patients with small cell lung carcinoma with limited disease receiving prophylactic cranial irradiation. Cancer 2005;104:811-6. [Crossref] [PubMed]
26. Hou Q, Sun B, Yao N, et al. Construction of Brain Metastasis Prediction Model and Optimization of Prophylactic Cranial Irradiation Selection for Limited-Stage Small-Cell Lung Cancer. Cancers (Basel) 2022;14:4906. [Crossref] [PubMed]
27. Spigel DR, Cheng Y, Cho BC, et al. ADRIATIC: Durvalumab (D) as consolidation treatment (tx) for patients (pts) with limited-stage small-cell lung cancer (LS-SCLC). J Clin Oncol 2024;42:LBA5.