A Proposal to Reflect Survival Difference and Modify the Staging System for Lung Adenocarcinoma and Squamous Cell Carcinoma: Based on the Machine Learning

New perspectives on cancer clinical research in the era of big data and machine learning

In the 21st century, the development of medical science has entered the era of big data, and machine learning has become an essential tool for mining medical big data. The establishment of the SEER database has provided a wealth of epidemiological data for cancer clinical research, and the number of studies based on SEER and machine learning has been growing in recent years. This article reviews recent research based on SEER and machine learning and finds that the current focus of such studies is primarily on the development and validation of models using machine learning algorithms, with the main directions being lymph node metastasis prediction, distant metastasis prediction, and prognosis-related research. Compared to traditional models, machine learning algorithms have the advantage of stronger adaptability, but also suffer from disadvantages such as overfitting and poor interpretability, which need to be weighed in practical applications. At present, machine learning algorithms, as the foundation of artificial intelligence, have just begun to emerge in the field of cancer clinical research. The future development of oncology will enter a more precise era of cancer research, characterized by larger data, higher dimensions, and more frequent information exchange. Machine learning is bound to shine brightly in this field.

First-line cemiplimab monotherapy and continued cemiplimab beyond progression plus chemotherapy for advanced non-small-cell lung cancer with PD-L1 50% or more (EMPOWER-Lung 1): 35-month follow-up from a mutlicentre, open-label, randomised, phase 3 trial

BACKGROUND

Cemiplimab provided significant survival benefit to patients with advanced non-small-cell lung cancer with PD-L1 tumour expression of at least 50% and no actionable biomarkers at 1-year follow-up. In this exploratory analysis, we provide outcomes after 35 months' follow-up and the effect of adding chemotherapy to cemiplimab at the time of disease progression.

METHODS

EMPOWER-Lung 1 was a multicentre, open-label, randomised, phase 3 trial. We enrolled patients (aged ≥18 years) with histologically confirmed squamous or non-squamous advanced non-small-cell lung cancer with PD-L1 tumour expression of 50% or more. We randomly assigned (1:1) patients to intravenous cemiplimab 350 mg every 3 weeks for up to 108 weeks, or until disease progression, or investigator's choice of chemotherapy. Central randomisation scheme generated by an interactive web response system governed the randomisation process that was stratified by histology and geographical region. Primary endpoints were overall survival and progression free survival, as assessed by a blinded independent central review (BICR) per Response Evaluation Criteria in Solid Tumours version 1.1. Patients with disease progression on cemiplimab could continue cemiplimab with the addition of up to four cycles of chemotherapy. We assessed response in these patients by BICR against a new baseline, defined as the last scan before chemotherapy initiation. The primary endpoints were assessed in all randomly assigned participants (ie, intention-to-treat population) and in those with a PD-L1 expression of at least 50%. We assessed adverse events in all patients who received at least one dose of their assigned treatment. This trial is registered with ClinicalTrials.gov, NCT03088540.

FINDINGS

Between May 29, 2017, and March 4, 2020, we recruited 712 patients (607 [85%] were male and 105 [15%] were female). We randomly assigned 357 (50%) to cemiplimab and 355 (50%) to chemotherapy. 284 (50%) patients assigned to cemiplimab and 281 (50%) assigned to chemotherapy had verified PD-L1 expression of at least 50%. At 35 months' follow-up, among those with a verified PD-L1 expression of at least 50% median overall survival in the cemiplimab group was 26·1 months (95% CI 22·1-31·8; 149 [52%] of 284 died) versus 13·3 months (10·5-16·2; 188 [67%] of 281 died) in the chemotherapy group (hazard ratio [HR] 0·57, 95% CI 0·46-0·71; p<0·0001), median progression-free survival was 8·1 months (95% CI 6·2-8·8; 214 events occurred) in the cemiplimab group versus 5·3 months (4·3-6·1; 236 events occurred) in the chemotherapy group (HR 0·51, 95% CI 0·42-0·62; p<0·0001). Continued cemiplimab plus chemotherapy as second-line therapy (n=64) resulted in a median progression-free survival of 6·6 months (6·1-9·3) and overall survival of 15·1 months (11·3-18·7). The most common grade 3-4 treatment-emergent adverse events were anaemia (15 [4%] of 356 patients in the cemiplimab group vs 60 [17%] of 343 in the control group), neutropenia (three [1%] vs 35 [10%]), and pneumonia (18 [5%] vs 13 [4%]). Treatment-related deaths occurred in ten (3%) of 356 patients treated with cemiplimab (due to autoimmune myocarditis, cardiac failure, cardio-respiratory arrest, cardiopulmonary failure, septic shock, tumour hyperprogression, nephritis, respiratory failure, [n=1 each] and general disorders or unknown [n=2]) and in seven (2%) of 343 patients treated with chemotherapy (due to pneumonia and pulmonary embolism [n=2 each], and cardiac arrest, lung abscess, and myocardial infarction [n=1 each]). The safety profile of cemiplimab at 35 months, and of continued cemiplimab plus chemotherapy, was generally consistent with that previously observed for these treatments, with no new safety signals INTERPRETATION: At 35 months' follow-up, the survival benefit of cemiplimab for patients with advanced non-small-cell lung cancer was at least as pronounced as at 1 year, affirming its use as first-line monotherapy for this population. Adding chemotherapy to cemiplimab at progression might provide a new second-line treatment for patients with advanced non-small-cell lung cancer.

FUNDING

Regeneron Pharmaceuticals and Sanofi.

Overinterpretation of findings in machine learning prediction model studies in oncology: a systematic review

OBJECTIVES

In biomedical research, spin is the overinterpretation of findings, and it is a growing concern. To date, the presence of spin has not been evaluated in prognostic model research in oncology, including studies developing and validating models for individualized risk prediction.

STUDY DESIGN AND SETTING

We conducted a systematic review, searching MEDLINE and EMBASE for oncology-related studies that developed and validated a prognostic model using machine learning published between 1st January, 2019, and 5th September, 2019. We used existing spin frameworks and described areas of highly suggestive spin practices.

RESULTS

We included 62 publications (including 152 developed models; 37 validated models). Reporting was inconsistent between methods and the results in 27% of studies due to additional analysis and selective reporting. Thirty-two studies (out of 36 applicable studies) reported comparisons between developed models in their discussion and predominantly used discrimination measures to support their claims (78%). Thirty-five studies (56%) used an overly strong or leading word in their title, abstract, results, discussion, or conclusion.

CONCLUSION

The potential for spin needs to be considered when reading, interpreting, and using studies that developed and validated prognostic models in oncology. Researchers should carefully report their prognostic model research using words that reflect their actual results and strength of evidence.

A nomogram for predicting postoperative overall survival of patients with lung squamous cell carcinoma: A SEER-based study

Background

Lung squamous cell carcinoma (LSCC) is a common subtype of non-small cell lung cancer. Our study aimed to construct and validate a nomogram for predicting overall survival (OS) for postoperative LSCC patients.

Methods

A total of 8,078 patients eligible for recruitment between 2010 and 2015 were selected from the Surveillance, Epidemiology, and End Results database. Study outcomes were 1-, 2- and 3-year OS. Analyses performed included univariate and multivariate Cox regression, receiver operating characteristic (ROC) curve construction, calibration plotting, decision curve analysis (DCA) and Kaplan-Meier survival plotting.

Results

Seven variables were selected to establish our predictive nomogram. Areas under the ROC curves were 0.658, 0.651 and 0.647 for the training cohort and 0.673, 0.667 and 0.658 for the validation cohort at 1-, 2- and 3-year time-points, respectively. Calibration curves confirmed satisfactory consistencies between nomogram-predicted and observed survival probabilities, while DCA confirmed significant clinical usefulness of our model. For risk stratification, patients were divided into three risk groups with significant differences in OS on Kaplan-Meier analysis (P < 0.001).

Conclusion

Here, we designed and validated a prognostic nomogram for OS in postoperative LSCC patients. Application of our model in the clinical setting may assist clinicians in evaluating patient prognosis and providing highly individualized therapy.

A machine learning method for improving liver cancer staging

Liver cancer is a common malignant tumor, and its clinical stage is closely related to the clinical treatment and prognosis of patients. Currently, the BCLC staging system revised by the BCLC group of University of Barcelona is the globally recognized staging system for liver cancer. However, with the deepening of related research, the current staging system can no longer fully meet the clinical needs. In this work, we propose a novel machine learning method for constructing an automatic hepatocellular carcinoma staging model that incorporates far more clinical variables than any existing staging system. Our model is based on random survival forests, which generates a unique hazard function for each patient. B-splines are used to embed hazard functions into vectors in low-dimensional space and hierarchical clustering method groups similar patients to form staging cohorts. The resulting staging system significantly outperforms the BCLC system in terms of distinctiveness between patients in different stages.

Methodological conduct of prognostic prediction models developed using machine learning in oncology: a systematic review

BACKGROUND

Describe and evaluate the methodological conduct of prognostic prediction models developed using machine learning methods in oncology.

METHODS

We conducted a systematic review in MEDLINE and Embase between 01/01/2019 and 05/09/2019, for studies developing a prognostic prediction model using machine learning methods in oncology. We used the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement, Prediction model Risk Of Bias ASsessment Tool (PROBAST) and CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) to assess the methodological conduct of included publications. Results were summarised by modelling type: regression-, non-regression-based and ensemble machine learning models.

RESULTS

Sixty-two publications met inclusion criteria developing 152 models across all publications. Forty-two models were regression-based, 71 were non-regression-based and 39 were ensemble models. A median of 647 individuals (IQR: 203 to 4059) and 195 events (IQR: 38 to 1269) were used for model development, and 553 individuals (IQR: 69 to 3069) and 50 events (IQR: 17.5 to 326.5) for model validation. A higher number of events per predictor was used for developing regression-based models (median: 8, IQR: 7.1 to 23.5), compared to alternative machine learning (median: 3.4, IQR: 1.1 to 19.1) and ensemble models (median: 1.7, IQR: 1.1 to 6). Sample size was rarely justified (n = 5/62; 8%). Some or all continuous predictors were categorised before modelling in 24 studies (39%). 46% (n = 24/62) of models reporting predictor selection before modelling used univariable analyses, and common method across all modelling types. Ten out of 24 models for time-to-event outcomes accounted for censoring (42%). A split sample approach was the most popular method for internal validation (n = 25/62, 40%). Calibration was reported in 11 studies. Less than half of models were reported or made available.

CONCLUSIONS

The methodological conduct of machine learning based clinical prediction models is poor. Guidance is urgently needed, with increased awareness and education of minimum prediction modelling standards. Particular focus is needed on sample size estimation, development and validation analysis methods, and ensuring the model is available for independent validation, to improve quality of machine learning based clinical prediction models.

[Predicting the risk of tumor progression in patients with early stages of adenocarcinoma and squamous cell lung carcinoma based on laboratory parameters]

Non-small cell carcinoma (NSCLC) prevails in the structure of the incidence of lung cancer. In patients with I stage NSCLC, only 60-70% overcome the 5-year survival barrier, and at II stage it decreases to 35-40%. The reason for such a high mortality rate is almost always a relapse of the disease. The main histological forms of NSCLC - adenocarcinoma (AC) and squamous cell carcinoma (SCLC) - differ in the course, protocols and effectiveness of the treatment. Comparative survival data for AK and PCLC are controversial, and reliable biomarkers for determining the risk of tumor progression are lacking. In thus study we have investigated the possibility of using laboratory parameters characterizing the level of some blood proteins involved in carcinogenesis in patients with early stages of AC and SCLC to determine the risk of disease progression. We retrospectively analyzed the duration of the relapse-free period after surgical treatment for one year in 1250 patients (816 with stages I and II of adenocarcinoma, G1-3 and 434 with early stages of SCLC, G1-3). In 81 patients with AC and 36 - with SCLC (stages I-II, G1-3) the level of CYFRA 21-1 and SCC by electrochemiluminescent method, chemokines CXCL5, CXCL8, TPA, pyruvate kinase M2, HIF-1α and hyaluronic acid by enzyme immunoassay, receptors CXCR1, CXCR2, CD44v6 by flow cytometry were determined. Using the Kaplan-Meier graphical analysis, groups of low (stage I G1-2 + stage II G1) and high (stage I G3 + stage II G2-3) risk of tumor progression were identified. In the case of the one-year survival rate of patients with AC was higher than with SCLC. In patients with AC and a high risk of tumor recurrence, compared with a low one, the level of CYFRA 21-1, the mean intensity of fluorescence (MFI) of the CXCR1 receptor in granulocytes, and the relative content of the CXCR2 receptor in lymphocytes were higher. In the case of rapid progression of SCLC in patients, the relative content of the CXCR2 receptor in lymphocytes, the proportion of monocytes equipped with the CD44v6 receptor, and the SCC level were higher than with slow progression. Regression equations, including combinations of the above parameters (threshold value for AC - 0,512, for SCLC - 0,409, sensitivity - 91,9% and 90,0%, specificity - 90,0% and 87,5%, respectively), allow to predict the probability of tumor recurrence.

Construction of methylation-associated nomogram for predicting the recurrence-free survival risk of stage I-III lung adenocarcinoma

Aim: The aim of our study was to investigate a methylation-associated predictor for prognosis in patients with stage I-III lung adenocarcinoma (LUAD). Methods: A DNA methylation-based signature was developed via univariate, least absolute shrinkage and selection operator and multivariate Cox regression models. Results: We identified a 14-site methylation signature that was correlated with recurrence-free survival of stage I-III lung adenocarcinoma patients. By receiver operating characteristic analysis, we showed the high ability of the 14-site methylation signature for predicting recurrence-free survival. In addition, the nomogram result showed a satisfactory predictive value. Conclusion: We successfully identified a DNA methylation-associated nomogram which can predict recurrence-free survival in patients with stage I-III lung adenocarcinoma.

Long-term outcomes following neoadjuvant or adjuvant chemoradiotherapy for stage I-IIIA non-small cell lung cancer: a propensity-matched analysis

Background

This study aimed to evaluate the long-term survival outcomes of patients undergoing neoadjuvant chemoradiotherapy or adjuvant chemoradiotherapy for T1-4N0-1M0 disease.

Methods

Patients with pT1-4N0-1M0 between 2010 and 2015 who received pre- or postoperative (R0 resection) chemoradiotherapy were identified. The exclusion criteria included N2 or M1 disease, immunotherapy, and targeted therapy. The staging was recalculated according to the new 8th edition TNM classification. Survival and predictors were assessed using Kaplan-Meier and multivariate Cox proportional-hazards model. Propensity-score matching with a ratio of 2:1 was performed to reduce bias in various clinicopathological factors.

Results

Of the 1,769 patients who met the inclusion criteria, 407 and 814 were included in the neoadjuvant and adjuvant chemoradiotherapy group, respectively, after propensity-score matching. The 5-year overall survival (OS) and cancer-specific survival (CSS) were 38.1% and 40.0% for neoadjuvant chemoradiotherapy and 26.3% and 26.5% for adjuvant chemoradiotherapy, respectively [P<0.0001, hazard ratio (HR): 0.7418, 95% confidence interval (CI): 0.6434-0.8553; P<0.0001, HR: 0.7444, 95% CI: 0.6454-0.8587)]. When stratified by stage, stage IIA (P=0.4166, HR: 0.8575, 95% CI: 0.5917-1.243) and IIIA (P=0.0740, HR: 0.7687, 95% CI: 0.5748-1.028) did not show improved 5-year OS in patients receiving neoadjuvant chemoradiotherapy. When stratified by age, similar trends were observed for patients aged more than 75 years. The multivariable analysis showed a significant association of neoadjuvant chemoradiotherapy with better survival.

Conclusions

Neoadjuvant chemoradiotherapy might improve the long-term survival of patients with stage I-IIIA non-small cell lung cancer (NSCLC). For patients aged more than 75 years, neoadjuvant chemoradiotherapy was not associated with an improvement in survival.

A Robust Signature Based on Autophagy-Associated LncRNAs for Predicting Prognosis in Lung Adenocarcinoma

Objective

To construct a predictive signature based on autophagy-associated lncRNAs for predicting prognosis in lung adenocarcinoma (LUAD). Materials and Methods. Differentially expressed autophagy genes (DEAGs) and differentially expressed lncRNAs (DElncRNAs) were screened between normal and LUAD samples at thresholds of ∣log₂Fold Change∣ > 1 and P value < 0.05. Univariate Cox regression analysis was conducted to identify overall survival- (OS-) associated DElncRNAs. The total cohort was randomly divided into a training group (n = 229) and a validation group (n = 229) and a validation group (

Results

A total of 30 DEAGs and 2997 DElncRNAs were identified between 497 LUAD tissues and 54 normal tissues; however, only 1183 DElncRNAs were related to the 30 DEAGs. A signature consisting of 13 DElncRNAs was built to predict OS in lung adenocarcinoma, and the survival analysis indicated a significant OS advantage of the low-risk group over the high-risk group in the training group, with a 5-year OS AUC of 0.854. In the validation group, survival analysis also indicated a significantly favorable OS for the low-risk group over the high-risk group, with a 5-year OS AUC of 0.737. Univariate and multivariate Cox regression analyses indicated that only positive surgical margin (vs negative surgical margin) and high-risk group (vs low-risk group) based on the predictive signature were independent risk factors predictive of overall mortality in LUAD.

Conclusions

This study investigated the association between autophagy-associated lncRNAs and prognosis in LUAD and built a robust predictive signature of 13 lncRNAs to predict OS.