Abstract
Small cell lung cancer (SCLC) is an aggressive malignancy with poor prognosis. No validated prognostic score has been established to guide clinical decisions in the extensive stage (ES). This narrative review critically examines the evolution of prognostic models in SCLC. We aim to highlight current gaps and propose directions for the development of clinically actionable tools. We conducted a comprehensive review of the literature on SCLC prognostic models, focusing on historical context, model design, variables used, validation methods, and real-world applicability. Comparative strengths and limitations were analysed across different model types. We analysed early scoring systems, modern nomograms, inflammation-based and nutritional scores, as well as integrative models. Historical tools are often limited to disease stage, performance status, basic laboratory values, most lack external validation, are retrospective, or were developed on chemotherapy-only cohorts. Recent models incorporate broader clinical data and, in some cases, nomograms or web-based calculators. Yet, few have undergone external validation or demonstrated utility in diverse clinical settings. The absence of dynamic, personalized models prevents integration into contemporary practice. Although numerous prognostic tools have been proposed, a reliable, validated tool is still lacking. Future prognostic models must move beyond static clinical parameters. Incorporating molecular biomarkers, real-world data, and machine learning could enable the development of validated, adaptive tools with true clinical relevance. Collaborative, prospective efforts will be critical to achieve this goal.
Introduction
Small cell lung cancer (SCLC) accounts for roughly 15% of lung cancers and it is characterized by rapid growth, high invasiveness, early metastatic spread, and frequent recurrence. Approximately 70% of patients are diagnosed with extensive-stage (ES-SCLC), defined as tumour not amenable to radical radiotherapy, including malignant pericardial and/or pleural effusions, corresponding to stage IV disease by TNM staging.1,2 Untreated, median survival ranges from 4 to 12 weeks; even treated, prognosis remains poor, with 2-year overall survival (OS) around 7% in ES-SCLC versus 21.1% in limited-stage (LS) disease.3–5
Platinum-based chemotherapy has long been foundational due to SCLC chemosensitivity. Since 2021, immunotherapy with immune checkpoint inhibitors (ICI) such as atezolizumab and durvalumab added modest but significant survival improvements in ES-SCLC, demonstrated in phase III trials.6,7 Despite only modest gains in median progression-free survival (mPFS) and OS, 2- to 5-year survival analyses reveal a small subset of long-term survivors emerging with immunotherapy-based regimens.8,9 Nonetheless, no validated prognostic or predictive biomarkers currently guide ICI use in ES-SCLC: PD-L1 expression is largely absent or non-predictive, tumour mutational burden (TMB) has limited predictive value, and clinical outcomes appear independent of PD-L1 levels or platinum choice.10–12 Previous molecular classifications based on neuroendocrine gene expression or transcription factors offer biological insights but remain complex and impractical for routine clinical application. 10 New data is emerging in unexplored fields of prognostic factors as irAE in SCLC treated with chemoimmunotherapy, although being little yet. 13 Thus, there is an urgent clinical need for practical, cost-effective prognostic and predictive tools that can stratify patients likely to benefit from systemic therapies, enabling optimized treatment selection and surveillance. 14 Current prognostic models, however, suffer from several limitations. This review critically examines established and emerging prognostic tools to identify those best suited to the contemporary ES-SCLC landscape, striving to bridge existing evidence gaps and improve patient management.
Materials and Methods
We conducted a comprehensive literature search through PubMed, Google Scholar, and Embase. Search terms combined keywords and Medical Subject Headings (MeSH) related to ES-SCLC and prognostic tools, using Boolean operators to refine results. Specifically, we used combinations of the terms “small cell lung cancer” OR “SCLC” AND “prognostic” AND (“score” OR “nomogram”). The search was limited to publications from 1987 up to November 2025 with no language restrictions initially applied. We manually screened titles and abstracts for relevance, followed by full-text review of potentially eligible studies. Additional relevant articles were identified through reference lists of selected publications. Studies were prioritized based on the quality of evidence, sample size, and clinical relevance for inclusion in this review. Conversely, exclusion criteria were: articles not reporting original data such as reviews, editorials, and conference abstracts; studies with incomplete or unclear information regarding prognostic models or patient outcomes; studies focusing exclusively on limited-stage SCLC when the focus was on extensive-stage disease; non-human or preclinical studies; publications not available in full text or non-English language articles if language restrictions were applied; and studies with overlapping patient populations to avoid duplication of data. Additionally, we excluded studies lacking sufficient follow-up duration or survival data necessary for prognostic evaluation. These exclusion criteria aimed to ensure inclusion of relevant, high-quality evidence focused on clinically meaningful prognostic tools in SCLC.15–17
Results
A summary of the key prognostic scores and nomograms identified in the literature is presented in Table 1. Table 2 specifies event rates and tools with external validation cohort. Table 3 outlines key methodological and validation data. A conceptual feature importance bar plot is presented in Figure 1.
Conceptual feature importance bar plot. Key variables found to impact prognosis across various SCLC prognostic tools are shown above. The importance scores are estimates based on reported model coefficients and literature emphasis. This aggregated visual synthesis highlights the multifactorial nature of prognostic modeling in SCLC integrating clinical, biochemical, molecular and treatment factors.
Main Prognostic Scores and Nomograms in SCLC.
surgery alone or with adjuvant therapy.
Abbreviations: ALP: Alkalyne phosphatase; ANC: absolute neutrophil count; C: Cyclophosphamide; ChT: chemotherapy; CHRNA6: cholinergic receptor nicotinic alpha 6 subunit; CI: confidence interval; CRP: c-reactive protein; D: epidoxorubicin; DDL-3: delta-like protein 3; dNLR: derived Neutrophil-to-lymphocite ratio; E: etoposide; ECOG PS: Eastern Cooperative Group Performance Status; ES: extensive stage; F: female; GPS: Glasgow Prognostic Score; KP: Karnofsky Performance status; HCO3: serum bicarbonate; HR: hazard ratio; I: Ifosfamide; ICI: immune-checkpoint inhibitor; L: line; LDH: lactate dehydrogenase; LMR: Lymphocyte-to-monocyte ratio; LS: limited stage; M: male; mets: metastases; mGPS: modified Glasgow Prognostic Score; mos: months; mOS: median OS; mos: months; Mtx: methotrexate; N: number; Na: serum sodium; ND: not detailed; neg: negative; NLR: Neutrophil-to-lymphocyte ratio; NR: not reported; P: platinum agent; PD-L1: programmed death ligand-1; PFS: progression free survival; PLR: platelet-to-lymphocyte ratio; pos: positive; OS: overall survival; R: radiotherapy; RR: relative risk; SCLC: small-cell lung cancer; T: treatment; TLC: total lymphocyte count; TRT: thoracic radiotherapy; ULN: upper limit of normal; VALG: Veterans Affairs Lung Study Group.
Overview of External Validation Cohorts in SCLC Prognostic Models.
Abbreviations: LIPI: Lung Immune Prognostic Index; N°: numerosity; MSKPS: Memorial Sloan Kettering Prognostic Score; MPS: Manchester Prognostic Score; NA: Not Applicable; NPS: Naples Prognostic Score: SPS: Simplified prognostic score;
Overview of the Key Methodological and Validation Data for the Main SCLC Prognostic Tools.
Abbreviations: AUC: Area Under the Curbe; CRP: C-reactive protein; DLL3: Delta-like ligand 3; ND: not described; NSE: Neuron Specific Enolase; OS: Overall Survival; PD-L1: Programmed Death-Ligand 1.
Historical Prognostic Models (pre-2000s)
Models like the Manchester Prognostic Score, Kawahara Score, and Maestu Index have demonstrated ability to stratify patients by survival outcomes historically but lack direct evidence showing that their use in clinical decision-making improves treatment outcomes.18,20,21
Modern Prognostic Nomograms (Post-2000s)
These modern nomograms consistently show superior predictive accuracy over traditional staging systems by incorporating clinical, biochemical, and molecular factors. While these models provide individualized survival estimates which can inform prognosis, there is limited prospective evidence that they actually guide treatment stratification that enhances survival or quality of life. They facilitate risk categorization that could be used to tailor treatment intensity or follow-up schedule but require further validation in interventional trials.5,22–25
Inflammatory and Nutritional Scores
The first is calculated through the following formula:
The dynamic changes in PNI are a promising support for personalized immunotherapy. In SCLC, it has a close connection with OS. 31 Jiang et al included 9 studies involving 4164 SCLC patients and found that low PNI led to worse OS (HR = 1.43, 95% CI: 1.24-1.64). PNI ≤ 47.7 could indicate a poor prognosis for patients with SCLC. 31 Li et al included 93 SCLC patients who received radiotherapy to explore the prognostic effect of PNI. The results indicated that the OS of patients with PNI ≤ 47.7 was significantly shorter than that of patients with PNI > 47.7, and PNI ≤ 47.7 could indicate a poor prognosis for patients with SCLC. 32
Secondly, the CONUT score is an immuno-nutritional biomarker derived from serum albumin, TLC and total cholesterol concentration. Yilmaz et al in 2020 33 retrospectively analysed 216 SCLC patients (of which 59 with LS-SCLC) diagnosed between 2010 and 2019. A high CONUT (≥ 2) score, related to both shorter PFS (5 vs 10 months, P < .001) and OS (9 vs 14 months, P < .001). At the multivariate analysis, the CONUT score was found to be an independent factor for PFS and OS (HR: 0.612, P = .018; HR: 0.650, P = .038, respectively), alongside the stage. 33
Contemporary Prognostic Models
Scores like the LIPI, SPS, MSKPS, GPS/mGPS, and Naples Prognostic Score integrate clinical and inflammatory markers and show robust prognostic power in retrospective cohorts. LIPI, for example, stratifies immunotherapy-treated patients into risk groups with distinct survival probabilities. 44 While these tools support improved risk stratification, prospective validation showing that their systematic use to tailor therapies or follow-up schedules improves patient-centered outcomes remains to be demonstrated.
Emerging AI Prognostic Tools
AI models integrating clinical, radiomic, and pathological data have demonstrated high prediction accuracy for survival outcomes in lung cancers, with limited data on SCLC. 45 Models based on clinical and demographic data (eg, the XGBoost model) have demonstrated high predictive performance in estimating survival, with features like treatment modalities influencing outcomes.. 46
AI-driven spatial cell phenomics and tumor microenvironment analyses improve risk stratification beyond standard staging, which could enable better treatment selection. 47
Clinical utility of AI- generated nomograms has been explored in retrospective cohorts and multi-centre studies with proof of good calibration, discrimination and decision curve analysis, which suggested practical benefit in management. 48 Yet, contemporary evidence is mostly observational; definitive proof that AI-guided treatment stratification improves patient survival in prospective settings is still lacking.
Discussion
From the evidence above, it can be noted that MPS stratifies patients into good, intermediate, and poor prognosis categories based on factors such as LDH and tumour stage, with mOS of 12.9 months for good and 5.8 months for poor prognosis. Its simplicity makes it a practical tool, allowing quick assessments. The SPS predictively identifies risk groups with median OS ranging from 26.9 months for good to 6.8 months for poor prognosis. It balances clinical and inflammatory factors, enhancing its practicality in everyday clinical settings. In both cases, main constraints are the limited cohort on which they were tested, raising concerns about generalizability.
As to MSKPS, it predicts OS effectively, with higher scores correlating with worse survival outcomes. Its versatility across various cancers supports its clinical adoption. Nevertheless, its initial development on pancreatic cancer, and later adaptation to SCLC, warrants further validation for the latter.
Therefore, the main evolution from historical to contemporary models is represented by the overcome of some of these limits, such as small cohorts, outdated staging systems, lack of external testing, and lack of consensus on the first line and supportive care treatments.
As to the inflammatory and nutritional markers, a PNI ≤ 47.7 indicates a poor prognosis, with significant associations with OS. It reflects the nutritional and immune status of patients. Although it may play a relevant role in treatment planning, it might not fully account for confounding factors affecting survival outcomes. Secondly, higher CONUT scores correlate with shorter OS and PFS. It offers a straightforward nutritional assessment, aiding in treatment decisions. The requirement for routine lab tests may limit its implementation in some clinical settings.
Eventually, the LIPI score shows promise in predicting outcomes in patients receiving immunotherapy, enhancing its relevance in current treatment paradigms. It integrates immunological factors, aligning with contemporary treatments. It is limited by small sample sizes and further validation is needed.
Significant heterogeneity exists in the cutoff values for inflammation-nutrition indices such as NLR and PLR across SCLC studies, limiting comparability and clinical application. To address this, large-scale multicentre cohort studies with harmonized protocols are essential to empirically derive robust, generalizable cutoffs.14,27 Individual patient data meta-analyses can further calibrate these thresholds across diverse populations, reducing centre-specific biases, as well as expert consensus through multidisciplinary panels. 14 Additionally, standardizing biomarker measurement timing and laboratory methods will minimize pre-analytical variability. 14 Emerging approaches using continuous modelling or machine learning could reduce dependence on arbitrary cutoff points, improving prognostic precision. 41 Combining these strategies will enhance reproducibility and clinical utility of inflammation-nutrition scores in SCLC prognostication.
With regards to nomograms, main limitations were not stratifying the risk according to staging, the inclusion of some unreproducible parameters (eg the health insurance plan), the lack of routinely available biochemical data (eg NLR and PLR), and neglecting the sequence of treatment as a prognostic element. Some models were powered on OS only, becoming vulnerable to non-cancer-specific deaths and unknown complications. Furthermore, nomograms can be a very delicate tool with some fine limitations, related to their calibration or statical description of variable data. 49
Frail Categories
Prognostic tools in SCLC exhibit variable performance across important, frail subgroups. In elderly patients tailored models incorporating geriatric assessment scales, such as the Charlson comorbidity index or Cumulative Illness Rating Scale for Geriatrics (CIRS-G), have shown improved prognostic accuracy and facilitate individualized treatment decisions in LS-SCLC. 50 For patients with significant comorbidities, integrating comorbidity burden into prognostic models better stratifies survival risk and may influence treatment tolerance and choice, with cardiovascular and cerebrovascular diseases notably impacting outcomes. 51 Regarding patients receiving second-line therapy, prognostic considerations must include prior treatment response, relapse timing, and PS; although specific tools validated in this subgroup are limited, incorporating these variables enhances patient selection and may optimize outcomes of subsequent therapies. 52 Overall, adapting prognostic models to account for subgroup-specific clinical features improves their predictive precision and supports more nuanced, patient-centered management in diverse SCLC populations.
Prognostic Tools with Predictive Potential
Some tools reviewed also have predictive potential beyond simple prognosis. They generally involve biomarkers or factors that correlate with response to specific treatments or could guide treatment selection.
The LIPI stratifies patients treated with ICI or chemotherapy into risk groups with distinct survival outcomes, suggesting potential predictive value for immunotherapy response. LIPI has also been extensively studied for treatment-by-risk interaction and heterogeneity of treatment effect, particularly in relation to immunotherapy and chemotherapy. However, the predictive role—meaning differential benefit or harm from immunotherapy versus chemotherapy by LIPI risk group—has not been definitively confirmed in prospective studies yet.53,54
Inflammatory and nutritional scores like the NLR have been linked with immune checkpoint blockade outcomes, implying they may predict response to ICIs. No formal treatment-by-risk interaction or heterogeneity-of-treatment-effect analyses have generally been conducted on NLR, PLR, and GPS/mGPS, limiting claims of true predictive utility beyond prognostic stratification.14,27 In contrast, older historical scores (Manchester Score, Kawahara Score, Maestu Index) and most pure prognostic inflammation indexes primarily stratify survival risk but lack evidence of guiding treatment choices that improve outcomes.18–21
The prognostic nomograms incorporating molecular markers such as PD-L1 and DLL3 (eg, Zhao et al nomogram) may have predictive implications as these markers are targets or modulators of therapy response. 25
AI-driven models that integrate clinical, radiomic, and pathological data show high survival prediction accuracy and potentially guide treatment selection, although prospective evidence is lacking. AI prognostic tools integrating multi-modal data show promise for treatment selection but lack published prospective evidence or interaction analyses confirming predictive treatment stratification. 55
In summary, predictive potential beyond prognosis is most apparent in models related to immunotherapy response markers (LIPI, NLR), molecular markers included in modern nomograms (PD-L1, DLL3), and emerging AI tools integrating multi-modal data aimed at guiding treatment stratification. Nonetheless, prospective trials establishing their clinical utility for treatment decisions remain limited.
Case Vignettes
Here are two illustrative case vignettes that integrate prognostic tool elements for SCLC, showing how different patient profiles may be stratified and what insights the prognostic models might provide:
Case Vignette 1: Good Prognosis SCLC Patient
Patient: Mr A., a 62-year-old male, newly diagnosed with LS-SCLC.
Clinical data:
ECOG PS: 0 Stage: Limited (VALG) Serum LDH < ULN NSE < ULN Albumin: normal (>3.5 g/dL) NLR: low (2.0) PD-L1 expression: low No liver or adrenal metastases visible on imaging Comorbidity score: Low Charlson index (0)
Planned treatment: concurrent chemoradiotherapy
Prognostic assessment:
According to the Pan et al or Zhao et al nomograms, the combination of limited stage, good PS, normal LDH and low inflammatory markers corresponds to a favourable risk group.
The LIPI would likely classify this patient as low risk considering low derived NLR and normal LDH.
Predicted median overall survival is above average for SCLC, potentially exceeding 20 months.
The SPS would indicate a “good prognosis” group.
Clinical implications:
The patient is expected to respond well to standard concurrent chemoradiotherapy.
Prognostic tools support a less intensive follow-up strategy with a focus on curative intent.
Regular monitoring of inflammatory markers like NLR could assist in detecting early relapse or treatment complications.
Case Vignette 2: Poor Prognosis SCLC Patient
Patient: Ms B., a 70-year-old female with ES-SCLC.
Clinical data:
ECOG PS: 2 Stage: Extensive (VALG) Serum LDH > 2x ULN NSE: elevated Albumin: low (<3.0 g/dL) NLR: high (5.5) PD-L1 expression: high Liver metastases present; adrenal metastases absent Charlson comorbidity index: 3
Planned treatment: systemic chemotherapy with immunotherapy
Prognostic assessment:
The Maestu index and MPS place this patient in the poor prognosis group due to high LDH, poor PS, and advanced stage. LIPI likely classifies her as intermediate or poor risk, reflecting elevated dNLR and LDH.
Nomograms from Zhao et al or Ni et al incorporating molecular and inflammatory markers predict shorter overall survival (<7-8 months).
The Naples Prognostic Score would also indicate high risk given inflammation and nutritional deficiencies.
Clinical implication:
Prognostic models suggest limited survival benefit from aggressive treatment; focus may shift to symptom control and quality of life. The high PD-L1 expression suggests some potential sensitivity to immunotherapy, warranting its inclusion.
These models guide discussion on realistic goals of care with patient and caregivers.
Close monitoring of nutritional status and early supportive interventions are advised.
Future Directions
Overall, MSKPS, mGPS, SPS and NPS seem amenable for use in daily practice, since being balanced between inflammatory and clinical variables,3,35,38 with SPS having an external validation cohort (n = 216).
Nevertheless, an ideal, modern prognostic score should integrate clinical, biological, and technological advancements. In the context of advancing prognostic tools for SCLC, it is essential to consider how emerging methodologies such as multi-omics, real-world data integration, and AI-based approaches can shape the next generation of predictive models. Ideal prognostic models should:
Integrate comprehensive data:
Include clinical variables (TNM staging, PS), presence of liver, brain or other metastases, and treatment modalities,35,56,57 as well as blood-based data (eg, NLR, PLR, LDH, sodium, albumin) and molecular biomarkers (eg, mutations of TP53, UNC13A, and mutational signatures, immune infiltration patterns).57–59 Integration of multi-omics strategies: genomics, proteomics, and metabolomics can provide a comprehensive view of tumour biology, enabling the identification of novel biomarkers that are not captured by traditional scoring systems.
60
Use advanced analytics:
Ensure models are internally and externally validated using independent cohorts and robust statistical metrics such as Harrell's C-index and Brier score to assess discrimination and calibration.35,56 AI-driven methodologies can analyse vast datasets to identify patterns that traditional statistical methods might overlook. Machine learning algorithms can enhance the predictive accuracy of prognostic models by integrating diverse data types (clinical, genomic, and imaging). AI can facilitate the development of dynamic models that adapt to new data over time. Machine learning aids the integration of multi-omics to optimize prognostic accuracy. These tools should be transparently reported and independently validated in multicentre cohorts.44,55,56 Data showing that AI-based treatment stratification leads to improved clinical outcomes in SCLC patients remains scarce, underscoring a need for rigorous prospective clinical trials assessing AI tools’ impact on therapy choice and outcomes.
45
Be practical and accessible:
Scores should balance ease of use with predictive power. Simplified but precise scores (eg, based on stage, PS, liver metastases, and NLR) can effectively stratify patients into risk groups.35,57 User-friendly tools (eg, nomograms, web-based calculators, or software) allow clinicians to estimate individualized survival probabilities.56,57 The validated models can be converted into accessible platforms such as web-based calculators or smartphone apps, designed for routine clinical use.
44
Develop and implement standardized protocols for laboratory assays, biomarker cut-offs, and timing of sample collection. This harmonization facilitates consistent biomarker measurement across centers and studies, reducing variability that undermines prognostic model performance.
61
Allow continuous updating:
Regular update to catch emerging biomarkers, novel therapies (such as immunotherapy), and evolving clinical practices is crucial.58,62 Embodying dynamic and longitudinal biomarkers into prognostic models, including serial measurements of immune-inflammatory markers (eg, NLR, CRP), circulating tumor DNA, and molecular profiles can better reflect temporal changes in tumor biology and host response, enabling real-time adjustment of risk stratification during treatment.44,61 Incorporating real-world data (RWD) from various clinical settings enhances the generalizability of prognostic models.
56
Conduct large-scale, prospective multicentre studies to validate existing and emerging prognostic models and nomograms. Rigorous validation can confirm reproducibility and reliability, addressing a key limitation of many current models that have been developed using small or single-center cohorts.63,64 Establish consortia of oncologists, pathologists, bioinformaticians, and statisticians to facilitate data sharing, model development, and validation efforts. Such collaboration accelerates translation from research to practice and supports regulatory approval pathways.
44
Focus on clinical relevance:
Conclusion
In summary
Footnotes
Acknowledgments
The authors would like to thank their colleagues for the support provided during the preparation of this review.
Ethics Consideration
Ethics statement is not applicable.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
AI Declaration Statement and AI-Assisted Technologies
During the preparation of this work the author(s) used AI tools and specifically ChatGPT, Gemini and Perplexity in order to enhance the literature search and optimize the contents. No scientific data has been generated or modified using AI.