Prediction of 2-Year Mortality in Patients with Advanced Non-Small Cell Lung Cancer Treated with Immune Checkpoint Inhibitors Based on a Machine Learning Algorithm

doi:10.21203/rs.3.rs-2727235/v1

Download PDF

Research Article

Prediction of 2-Year Mortality in Patients with Advanced Non-Small Cell Lung Cancer Treated with Immune Checkpoint Inhibitors Based on a Machine Learning Algorithm

https://doi.org/10.21203/rs.3.rs-2727235/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Immune checkpoint inhibitors (ICIs) extend the survival time in patients with advanced non-small-cell lung cancer (NSCLC), but not all patients respond to the treatment. However, current immune-related biomarkers, such as programmed death receptor ligand 1 (PD-L1) and tumor mutation burden(TMB) are limited. We created a machine learning (ML) based clinical decision support algorithm to evaluate the 2-Year mortality of NSCLC patients treated with ICIs by comprehensively combining the clinical information.

Methods: We collected clinical data, including patient characteristics, mutations and laboratory findings, from the electronic medical records of 345 patients with NSCLC treated with ICIs. Through machine learning and statistical analysis, we constructed effective immune risk score models. Twenty-four clinically meaningful features were used in supervised ML algorithms, including decision trees, random forest and XGBoost , to predict the 2-Year mortality.

Results: Several factors, including driver gene mutation status, derived neutrophil to lymphocyte ratio (dNLR), prognostic nutrition index (PNI), could independently predict the 2-Year mortality.ML platforms based on the random forest algorithm using 24 clinical features showed more significant prediction performance (area under the curve [AUC] 0.736).

Conclusions: Collectively, our machine learning algorithm offers a clinical decision support model to predict the 2-Year mortality in NSCLC treated with ICIs.

non-small cell lung cancer

immune checkpoint inhibitors

machine learning.

Non-small cell lung cancer (NSCLC) represents 85 % of patients diagnosed with lung cancer. Although the overall survival(OS) of NSCLC patients has improved with the continuous development of diagnosis and treatment, the 5-year OS rate is less than 5%(1). Immune checkpoint inhibitors (ICIs) have changed the treatment mode of NSCLC, prolonged survival, and were approved as first- and second-line therapies in NSCLC(2, 3). Nevertheless, only a small number of patients with NSCLC respond to ICIs(4). Therefore, it is critical to find ICI-related biomarkers and to identify patients who can benefit from ICIs.

At present, many tumor-related biomarkers, such as programmed death receptor ligand 1 (PD-L1) and tumor mutation burden (TMB), have been approved as a biomarker for investigating the efficacy of PD-1/PD-L1 inhibitors(5, 6). However, their application is limited by the lack of uniform detection methods and the absence of definite cut-off values. With the increasing use of ICIs, a variety of clinical characteristics have been identified that are associated with treatment response in NSCLC patients, such as smoking history, presentation status, gender, driver mutations, pathology, and biomarkers of the peripheral blood, such as neutrophil/lymphocyte ratio(7-9).However, the prediction ability of a single index is limited, thus, a model integrating these factors is needed.

ML algorithms have several advantages over traditional statistical modeling. Predictive models using machine learning algorithms may facilitate the recognition of clinically important risk among patients with several marginal risk factors that may otherwise not raise clinical concerns. Furthermore, machine learning algorithms can easily incorporate new clinical data to continually update and optimize algorithms with minimal oversight(10). So it could better explain the complex interactions between risk factors. Recently, ML has improved the ability to predict disease progression and treatment response in various diseases(11, 12). However, ML has not been widely studied in NSCLC patients treated with ICIs.

In this work, we conducted a large retrospective analysis of patients with advanced NSCLC receiving ICIs therapy.Our study aimed to develop a predictive model for prescribing PD-1/PD-L1 inhibitors in patients with NSCLC using machine-learning algorithms based on clinicopathological factors.

2.1 Study Population Characteristics

Of 345 patients in the training and validation cohorts, the baseline information of clinical characteristics among these patients was summarized. 141 (40.7%) died during the 24-month follow-up period. Most subjects were male (242,70.1%) and 103 (29.9%) were female. There are slightly more non-smokers (158,45.8%) than smokers (187, 54.2%). Patients with driver genomic alterations ac-counted for 33.6%. All characteristics are presented in Table 1.

Table 1. Patient Characteristics, Stratified by Death Status Within 24 Months of the Index Encounter.

	No. (%)
Characteristic	Alive (n = 204)	Died (n = 141)
Age (years)
<65	127（62.3）	91（64.5）
≥65	77（37.7）	50（35.5）
Gender
Male	148（72.5）	94（66.7）
Female	56（27.5）	47（33.3）
ECOG PS score
0-1	186（91.2）	115（81.6）
2-3	18（8.8）	26（18.4）
Smoking
Never	186（91.2）	115（81.6）
Current/previous	18（8.8）	26（18.4）
Histology
Squamous	58（28.4）	43（30.5）
Non-squamous	146（71.6）	98（69.5）
Driver oncogene alteration
Mutant	52（25.5）	64（45.4）
Wild-type / Unknown	152（74.5）	77（54.6）
Line of immunotherapy
<2	101（19.5）	47（33.3）
≥2	103（50.5）	94（66.7）
Location of metastasis
Brain	27（13.2）	27（13.2）
Liver	28（13.7）	35（24.8）
Bone	55（27.0）	53（37.6）
Lung/pleura	75（36.8）	73（51.8）

2.2 Algorithm Variable Importance

The top 10 variables in terms of variable importance for the 3 algorithms are shown in Table 2. The top predictors shared across all models were mutation patterns, prognostic nutrition index (PNI) and derived neutrophil-to-lymphocyte ratio (dNLR).Lung/pelura metastasis,Inflammatory factors, dNLR, PNI and driver oncogene alteration served as irreplaceable weights at decision trees ranches (Figure 1) .SHAP is used to further explain the impact of features on theXGBoost model, it showed that high dNLR, current/previous smoking and high LMR at baseline, bone metastasis,positive driver gene mutation have remained independent risk aspects for OS. (Figure 2).

Table 2. Variable Importance in Descending Order of Feature Importance for Decision Trees , Random Forest and XGBoost*.

Algorithm	Variable *	Algorithm	Variable *
Decision Trees	dNLR	Random Forest	Driver oncogene alteration
	PNI		dNLR
	Driver oncogene alteration		PNI
	Lung/pelura metastasis		TNM stage
	NLR		ECOG PS score
	White blood cells		Age
	Monocytes		Line of immunotherapy
	LMR		PLR
	A/G		NLR
	Liver metastasis		Smoking
XGBoost	dNLR
	Smoking
	LMR
	Bone metastasis
	Driver oncogene alteration
	Liver metastasis
	PLR
	PNI
	Lung/pelura metastasis
	Age

* Variable importance is ranked by importance rating for the decision trees, random forest and XGBoost models.

Abbreviations: NLR, neutrophil-lymphocyte ratio;dNLR, derived NLR;PNI, Prognostic nutritional

index;LMR, Lymphocyte-Monocyte Ratio;PLR, Platelet-Lymphocyte Ratio.

2.3 Model Performance

Algorithm discrimination and other performance metrics in the validation set are presented for each model in Table 3. The random forest and XGBoost models had higher PPV (70.4% and 60.9%, respectively) than the decision trees model (55.8%). AUC values were calculated for all models. The top two most accurate prediction models were derived from the random forest and XGBoost. The AUC of the two models was 0.736 and 0.724, respectively. Despite hyperparameter tuning, the random forest displayed evidence of overfitting, with AUCs in the training set of 0.99.

Table 3. Performance Metrics of Machine Learning Models

Algorithm	Positive Predictive Value^a	ROC-AUC^a	Accuracy	F1 Score
Decision Trees	0.704^b	0.709	0.709^b	0.559
Random Forest	0.558	0.736^b	0.660	0.624
XGBoost	0.609	0.724	0.699	0.644^b

^a Coprimary performance metric.

^b Refers to the best-performing model(s) for each performance metric.

1.4 Predicted probability of NSCLC overall survival

Given the importance results of the ML algorithm model, mutation patterns, PNI and dNLR were selected as variables for the construction of immune risk model in NSCLC. By integrating these factors, we used nomogram algorithm to predict the probability of OS of 6, 12 and 18 months in NSCLC (Figure 3). The nomogram indicated dNLR contributed slightly less to patients’ prognosis compared with other factors. C-index were conducted for the probability of OS of 6, 12 and 18 months, and the C-index was 0.694. Furthermore, we assessed and validated its effectiveness by time-dependent ROC analysis (Figure 4) and calibration plots (Figure 5). The AUC values for OS of 6, 12 and 18 months were 0.692, 0.672, and 0.679, respectively, in the training cohort, and 0.701, 0.759, 0.834, respectively, in the validation cohort,which highlighted that the predictive model performed well in predicting overall survival of patients with advanced NSCLC receiving ICI therapy.

Based on predictive scores, the cut-off values were determined by regrouping all subjects in the entire NSCLC cohort, as well as the training and validation cohorts into three subgroups with distinct prognosis. Survival analyses showed that in the entire NSCLC, training, and validation cohorts , the low-risk groups in all cohorts were cor-related with higher OS compared with high-risk groups ((Figure 6,P<0.001 in the entire NSCLC cohort, P<0.001 in the training cohort, and P<0.001 in the validation cohort, respectively).

In this cohort study, various clinical features from patients with NSCLC treated with ICIs were collected to establish a predictive model for immunotherapy, ML models accurately predicted the mortality risk of individuals with NSCLC. We selected 24 clinically relevant features based on our knowledge, including dNLR, PNI, driver oncogene alteration ,smoking years, previous line of therapy,Eastern Cooperative Oncology Group performance status and the presence of brain, liver and bone metastases. dNLR, PNI and driver oncogene alteration were independent features to predict the prognosis of patients with advanced NSCLC in our study consistent with previous reports adding relevance to our study.

Machine learning algorithms can be relatively easily retrained to account for emerging cancer survival patterns. Several recent studies have used ML methods to detect tumors and predict the prognosis of cancer types. For example, Singal et al. utilize a random forest to predict the development of hepatocellular carcinoma in patients with cirrhosis(13).Samuel Peter Heilbroner et al. used XGBoost decision tree model to predict cardiac events in patients taking PD- 1/PD- L1 therapy(14). Our model has comparable performance in predicting 2-year mortality in patients with advanced non-small cell lung cancer receiving immunotherapy.

We found that patients with the wild-type driver mutation had a better prognosis than patients with the positive driver mutation. Multiple clinical trials and meta-analyses showed that advanced NSCLC patients with driving EGFR mutation had poorer responses to ICIs monotherapy(15-17). The activation of oncogenes can lead to the activation of multiple immunosuppressive pathways, which can induce PD-L1 expression and the generation of other immunosuppressive factors to evade the host's antitumor immune response(18). Therefore, patients with positive driver gene mutations are often considered to be "immune exempt" populations. The potential factors driving the poor efficacy of immunotherapy in patients with positive gene mutation have been widely studied. The analysis of PD-L1 mRNA and protein levels in the cancer genome map and internal database showed that the expression of PD-L1 mRNA and protein in EGFR mutant NSCLC samples was lower than that in EGFR wild-type tumor samples(19). Studies have confirmed that patients with EGFR mutations had significantly lower TMB(20, 21). In addition, the tumor microenvironment (TME) of EGFR mutant NSCLC involves many types of immune system cells, such as tumor-associated macrophages, CD4 + T cells, CD8 + T cells, regulatory T cells, and mast cells(22). Previous studies classified TME into four different types based on the expression of tumor-infiltrating lymphocytes (TIL) and PD-L1 expression to predict which patients will respond to PD-1/PD-L1 inhibitors treatment, type I (PD-L1+/CD8+), type II (PD-L1/CD8-), type III (PD-L1+/CD8-) and type IV (PD-L1-/CD8+), respectively(23). Among them, type I is considered to be adaptive PD-L1 expression induced by pre-existing intratumoral T cells, therefore, patients with type I are most likely to benefit from immune checkpoint inhibitor therapy(24). Besides, in advanced NSCLC patients with EGFR mutation, the proportion of PD-L1+/CD8+ patients was very small[33]. These may be the potential reasons for the poor efficacy of PD-1/PD-L1 inhibitors in NSCLC patients with EGFR mutation. This is consistent with our findings.

In our model, pretreatment dNLR was a reliable and independent predictor of OS. dNLR is thought to be a marker of the inflammatory tumor microenvironment. dNLR and its dynamic changes are reported to be correlated with the efficacy of ICIs treatment in NSCLC patients(25). The neutrophil is considered a biomarker of inflammation, and it was reported to have both tumor-promoting and anti-tumor heterogeneity(26). During the progression of cancer, the number of absolute neutrophils usually increases, which contributes to the development of a circulating inflammatory environment associated with immune tolerance(27). Neutrophils adjust their function to promote tumor activity. Both inhibiting lymphocyte amplification and increasing peripheral neutrophil count lead to elevated dNLR, which is associated with poor prognosis at baseline(28). Ferrucci et al. found that higher dNLR at baseline was associated with poorer PFS (P <0.001) and OS (P <0.001) in patients treated with Ipilimumab(29).

We determined that a higher PNI at baseline was an important predictor of a better prognosis. PNI is combined with serum albumin and lymphocyte levels and can reflect the immune nutritional status of patients(30). This suggests that it has potential predictive value for patients with advanced NSCLC treated with ICIs. The count level of lymphocytes can be used as an alternative index to represent the nutritional status of the patients and chronic inflammation, lymphocytes play an important role in tumor defense by inducing cytotoxic cell death and inhibiting the proliferation and migration of cancer cells(31). The predictive role of PNI in a variety of malignancies has been extensively studied. Shoji et al. retrospectively analyzed the pretreatment PNI levels of 102 NSCLC patients treated with ICIs, the results showed that PNI was significantly correlated with the objective remission rate(ORR), disease control rate(DCR) and PFS of these patients(32). Yakup Bozkaya et al. explored the prognostic and predictive significance of PNI for patients with metastatic NSCLC undergoing first-line chemotherapy and showed that high PNI was an independent prognostic factor for OS and PFS(33). A meta-analysis of 15 articles discussed the role of PNI in the prognosis of NSCLC patients and showed that low PNI was a reliable indicator of poor OS(34). Another retrospective assessment of PNI in patients with advanced NSCLC treated with PD-1 inhibitors showed that low PNI was associated with poorer PFS and OS(35).

we further constructed a nomogram model which provided a personalized system for patients with NSCLC to predict prognostic. Good performance was determined in both the training set and validation set. All sets including the entire set, training set, and validation set showed that low-risk groups were correlated with longer OS compared with high-risk groups.

Our research does have some limitations. The factors affecting the nutritional status and inflammatory response are not completely controllable and are dynamic changes. We did not furtherly explore the relationship between the changes in peripheral hematological indicators and prognosis before and after the use of ICIs. With limited data available, the relationship between PD-L1 expression, TMB , and the prognosis was not analyzed in this study. Finally, although our study can provide certain clinical insights for patient selection, more prospective clinical research is needed to verify our conclusions.

5.1 Patients

A total of 345 advanced NSCLC patients treated with ICIs in the First Affiliated Hospital of Xi'an Jiaotong University from December 2019 to January 2020 were in-cluded. Data, including patient characteristics, laboratory results, genetic mutational status, metastatic sites, line of therapy, PD-1 inhibitor type and response, were col-lected.The study was approved by the ethics committee of our hospital (xjtu1af2021lsk-108). The need for informed consent was waived.

5.2 Clinical features and ML models

Through a literature review, we extracted the clinical factors known to be associ-ated with ICIs treatment. To avoid compounding effects, features dependent on other features were removed. These selected features were used as variables for establishing an ML model. The study population was randomly split into a training cohort, in which the mortality risk algorithms were derived, and a validation cohort, in which the algorithms were applied and tested. The training cohort consisted of 70% of the total study population, and the validation cohort consisted of the remaining 30%. In this study, we repeated 1000 times for cohort division and model construction to make full use of the sample information. The predictive models decision trees, random forest and XGBoost were compared and used for analyses.We adopted a machine learning algorithm for importance evaluation and immune risk model construction. Common features were selected to construct an immune risk assessment system for patients with NSCLC. Then, we used the calibration set to validate the model performance and performed receiver operating characteristic(ROC) curve analysis for visualizing its predictive value.

5.3 Statistical analysis

We used descriptive statistics to compare the characteristics of the study popula-tion, stratified by death status at 2 years. Algorithms were developed from the training cohort and assessed on the independent validation cohort, which played no role in model development, by calculating the positive predictive value (PPV) and the receiver operating characteristic curve (AUC). To further describe model performance, we also constructed model calibration plots and calculated secondary metrics of clinical pre-diction models, including accuracy and specificity. In order to evaluate the prognosis condition, we also used Kaplan-Meier for survival analysis. The Pearson correlation test was used for continuous variables. p<0.05 was defined as statistical significance. Data analysis and visualization were conducted based on statistical tools SPSS V.26.0, GraphPad Prism V.7, and R Programming Language (R V.3.6.1).

In conclusion, ML algorithms based on clinical factors had adequate performance in identifying patients with cancer who had a high risk of 2-years mortality. Our findings suggest that ML tools hold promise for integration into clinical workflows to guide treatment decisions.At the same time, we constructed an immune risk scoring system and explored a clinical prediction model for patients with advanced NSCLC receiving ICIs , which has good predictive accuracy.

7.1 Ethical Approval

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee of The First Affiliated Hospital of Xi’an Jiaotong University (protocol code xjtu1af2021lsk-108 and 2021.4.15).The research was carried out in accordance with a named standard.

7.2 Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

7.3 Authors' contributions

RZ and XCL: study conception and planning, statistical analysis,data interpretation, manuscript drafting, andfinal approval of manuscript. WJX , RCW , FRL,JZ and LY: data interpretation and-final approval of the manuscript. CHS: study conception and planning, statistical analysis, data interpretation, manuscript drafting, andfinal approval of the manuscript. All authors have read and agreed to the published version of the manuscript.

7.4 Availability of data and materials

The data that support the findings of this study are available from the corresponding author, CHS, upon reasonable request.

7.5 Acknowledgments

We thank The First Affiliated Hospital of Xi’an Jiaotong University for research equipment and technical support for this research.

Gridelli C, Rossi A, Carbone DP, Guarize J, Karachaliou N, Mok T, et al. Non-Small-Cell Lung Cancer. Nat Rev Dis Primers (2015) 1:15009. Epub 2015/01/01. doi: 10.1038/nrdp.2015.9.
Mok TSK, Wu YL, Kudaba I, Kowalski DM, Cho BC, Turna HZ, et al. Pembrolizumab Versus Chemotherapy for Previously Untreated, Pd-L1-Expressing, Locally Advanced or Metastatic Non-Small-Cell Lung Cancer (Keynote-042): A Randomised, Open-Label, Controlled, Phase 3 Trial. Lancet (2019) 393(10183):1819-30. Epub 2019/04/09. doi: 10.1016/S0140-6736(18)32409-7.
Zhou C, Chen G, Huang Y, Zhou J, Lin L, Feng J, et al. Camrelizumab Plus Carboplatin and Pemetrexed Versus Chemotherapy Alone in Chemotherapy-Naive Patients with Advanced Non-Squamous Non-Small-Cell Lung Cancer (Camel): A Randomised, Open-Label, Multicentre, Phase 3 Trial. Lancet Respir Med (2021) 9(3):305-14. Epub 2020/12/22. doi: 10.1016/S2213-2600(20)30365-9.
Garon EB, Rizvi NA, Hui R, Leighl N, Balmanoukian AS, Eder JP, et al. Pembrolizumab for the Treatment of Non-Small-Cell Lung Cancer. N Engl J Med (2015) 372(21):2018-28. Epub 2015/04/22. doi: 10.1056/NEJMoa1501824.
Gandara DR, Paul SM, Kowanetz M, Schleifman E, Zou W, Li Y, et al. Blood-Based Tumor Mutational Burden as a Predictor of Clinical Benefit in Non-Small-Cell Lung Cancer Patients Treated with Atezolizumab. Nat Med (2018) 24(9):1441-8. Epub 2018/08/08. doi: 10.1038/s41591-018-0134-3.
Li F, Dong X. Pembrolizumab Provides Long-Term Survival Benefits in Advanced Non-Small Cell Lung Cancer: The 5-Year Outcomes of the Keynote-024 Trial. Thorac Cancer (2021) 12(23):3085-7. Epub 2021/10/14. doi: 10.1111/1759-7714.14193.
Johannet P, Sawyers A, Qian Y, Kozloff S, Gulati N, Donnelly D, et al. Baseline Prognostic Nutritional Index and Changes in Pretreatment Body Mass Index Associate with Immunotherapy Response in Patients with Advanced Cancer. Journal for ImmunoTherapy of Cancer (2020) 8(2). doi: 10.1136/jitc-2020-001674.
Sha D, Jin Z, Budczies J, Kluck K, Stenzinger A, Sinicrope FA. Tumor Mutational Burden as a Predictive Biomarker in Solid Tumors. Cancer discovery (2020) 10(12):1808-25.
Zhai B, Chen J, Wu J, Yang L, Guo X, Shao J, et al. Predictive Value of the Hemoglobin, Albumin, Lymphocyte, and Platelet (Halp) Score and Lymphocyte-to-Monocyte Ratio (Lmr) in Patients with Non-Small Cell Lung Cancer after Radical Lung Cancer Surgery. Ann Transl Med (2021) 9(12):976. Epub 2021/07/20. doi: 10.21037/atm-21-2120.
Richter AN, Khoshgoftaar TM. A Review of Statistical and Machine Learning Methods for Modeling Cancer Risk Using Structured Clinical Data. Artif Intell Med (2018) 90:1-14. Epub 2018/07/19. doi: 10.1016/j.artmed.2018.06.002.
Agius R, Brieghel C, Andersen MA, Pearson AT, Ledergerber B, Cozzi-Lepri A, et al. Machine Learning Can Identify Newly Diagnosed Patients with Cll at High Risk of Infection. Nat Commun (2020) 11(1):363. Epub 2020/01/19. doi: 10.1038/s41467-019-14225-8.
Kong Y, Yu T. A Deep Neural Network Model Using Random Forest to Extract Feature Representation for Gene Expression Data Classification. Scientific Reports (2018) 8(1). doi: 10.1038/s41598-018-34833-6.
Singal AG, Mukherjee A, Elmunzer JB, Higgins PDR, Lok AS, Zhu J, et al. Machine Learning Algorithms Outperform Conventional Regression Models in Predicting Development of Hepatocellular Carcinoma. American Journal of Gastroenterology (2013) 108(11):1723-30. doi: 10.1038/ajg.2013.332.
Heilbroner SP, Few R, Neilan TG, Mueller J, Chalwa J, Charest F, et al. Predicting Cardiac Adverse Events in Patients Receiving Immune Checkpoint Inhibitors: A Machine Learning Approach. Journal for Immunotherapy of Cancer (2021) 9(10).
Herbst R. Baas P Kim Dw Et Al. Pembrolizumab Versus Docetaxel for Previously Treated, Pd-L1-Positive, Advanced Non-Small-Cell Lung Cancer (Keynote-010): A Randomised Controlled Trial. Lancet (2016) 387(10027):1540-50.
Lee CK, Man J, Lord S, Links M, Gebski V, Mok T, et al. Checkpoint Inhibitors in Metastatic Egfr-Mutated Non–Small Cell Lung Cancer—a Meta-Analysis. Journal of Thoracic Oncology (2017) 12(2):403-7.
Su S, Dong Z, Zhi X, Su J, Chen Z, Yang J, et al. Ma15. 01 Strong Pd-L1 Expression Predicts Poor Response and De Novo Resistance to Egfr Tkis among Non-Small Cell Lung Cancer Patients with Egfr Mutation. Journal of Thoracic Oncology (2018) 13(10):S407.
Akbay EA, Koyama S, Carretero J, Altabef A, Tchaicha JH, Christensen CL, et al. Activation of the Pd-1 Pathway Contributes to Immune Escape in Egfr-Driven Lung Tumorsimmune Escape of Egfr-Mutant Tumors through Pd-1. Cancer discovery (2013) 3(12):1355-63.
Dong Z-Y, Zhang J-T, Liu S-Y, Su J, Zhang C, Xie Z, et al. Egfr Mutation Correlates with Uninflamed Phenotype and Weak Immunogenicity, Causing Impaired Response to Pd-1 Blockade in Non-Small Cell Lung Cancer. Oncoimmunology (2017) 6(11):e1356145.
Offin M, Rizvi H, Tenet M, Ni A, Sanchez-Vega F, Li BT, et al. Tumor Mutation Burden and Efficacy of Egfr-Tyrosine Kinase Inhibitors in Patients with Egfr-Mutant Lung Cancerstmb and Efficacy of Egfr-Tki in Egfr-Mutant Lung Cancers. Clinical Cancer Research (2019) 25(3):1063-9.
Spigel DR, Schrock AB, Fabrizio D, Frampton GM, Sun J, He J, et al. Total Mutation Burden (Tmb) in Lung Cancer (Lc) and Relationship with Response to Pd-1/Pd-L1 Targeted Therapies. American Society of Clinical Oncology (2016).
Whiteside T. The Tumor Microenvironment and Its Role in Promoting Tumor Growth. Oncogene (2008) 27(45):5904-12.
O’Donnell JS, Teng MW, Smyth MJ. Cancer Immunoediting and Resistance to T Cell-Based Immunotherapy. Nature reviews Clinical oncology (2019) 16(3):151-67.
Teng MW, Ngiow SF, Ribas A, Smyth MJ. Classifying Cancers Based on T-Cell Infiltration and Pd-L1. Cancer research (2015) 75(11):2139-45.
Yang T, Hao L, Yang X, Luo C, Wang G, Cai CL, et al. Prognostic Value of Derived Neutrophil-to-Lymphocyte Ratio (Dnlr) in Patients with Non-Small Cell Lung Cancer Receiving Immune Checkpoint Inhibitors: A Meta-Analysis. BMJ open (2021) 11(9):e049123.
Benitez JC, Recondo G, Rassy E, Mezquita L. The Lipi Score and Inflammatory Biomarkers for Selection of Patients with Solid Tumors Treated with Checkpoint Inhibitors. The Quarterly Journal of Nuclear Medicine and Molecular Imaging: Official Publication of the Italian Association of Nuclear Medicine (AIMN)[and] the International Association of Radiopharmacology (IAR),[and] Section of the Society of (2020) 64(2):162-74.
Prelaj A, Tay R, Ferrara R, Chaput N, Besse B, Califano R. Predictive Biomarkers of Response for Immune Checkpoint Inhibitors in Non–Small-Cell Lung Cancer. European journal of cancer (2019) 106:144-59.
Mezquita L, Auclin E, Ferrara R, Charrier M, Remon J, Planchard D, et al. Association of the Lung Immune Prognostic Index with Immune Checkpoint Inhibitor Outcomes in Patients with Advanced Non–Small Cell Lung Cancer. JAMA oncology (2018) 4(3):351-7.
Tavakkoli M, Wilkins CR, Mones JV, Mauro MJ. A Novel Paradigm between Leukocytosis, G-Csf Secretion, Neutrophil-to-Lymphocyte Ratio, Myeloid-Derived Suppressor Cells, and Prognosis in Non-Small Cell Lung Cancer. Frontiers in Oncology (2019) 9:295.
Onodera T, Goseki N, Kosaki G. Prognostic Nutritional Index in Gastrointestinal Surgery of Malnourished Cancer Patients. Nihon Geka Gakkai Zasshi (1984) 85(9):1001-5.
Mantovani A, Allavena P, Sica A, Balkwill F. Cancer-Related Inflammation. nature (2008) 454(7203):436-44.
Shoji F, Takeoka H, Kozuma Y, Toyokawa G, Yamazaki K, Ichiki M, et al. Pretreatment Prognostic Nutritional Index as a Novel Biomarker in Non-Small Cell Lung Cancer Patients Treated with Immune Checkpoint Inhibitors. Lung Cancer (2019) 136:45-51.
Bozkaya Y, Köstek O, Sakin A, Özyükseler DT, Şakalar T, Çil İ. Is the Prognostic Nutritional Index a Prognostic and Predictive Factor in Metastatic Non-Small Cell Lung Cancer Patients Treated with First-Line Chemotherapy? Supportive Care in Cancer (2020) 28:2273-82.
Hu Y, Shen J, Liu R, Feng Z, Zhang C, Ling L, et al. Prognostic Value of Pretreatment Prognostic Nutritional Index in Non-Small Cell Lung Cancer: A Systematic Review and Meta-Analysis. The International journal of biological markers (2018) 33(4):372-8.
Liu N, Jiang A, Zheng X, Fu X, Zheng H, Gao H, et al. Prognostic Nutritional Index Identifies Risk of Early Progression and Survival Outcomes in Advanced Non-Small Cell Lung Cancer Patients Treated with Pd-1 Inhibitors. Journal of Cancer (2021) 12(10):2960.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Prediction of 2-Year Mortality in Patients with Advanced Non-Small Cell Lung Cancer Treated with Immune Checkpoint Inhibitors Based on a Machine Learning Algorithm

Status:

Version 1

Abstract

Figures

1 Introduction

2 Results

2.1 Study Population Characteristics

2.2 Algorithm Variable Importance

2.3 Model Performance

1.4 Predicted probability of NSCLC overall survival

3 Discussion

4 Limitations

5 Methods

5.1 Patients

5.2 Clinical features and ML models

5.3 Statistical analysis

6 Conclusion

Declarations

7.1 Ethical Approval

7.2 Competing interests

7.3 Authors' contributions

7.4 Availability of data and materials

7.5 Acknowledgments

References

Additional Declarations

Status:

Version 1