Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study

Our plain-English summary of Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study, by M. Yakar, D. Etiz, M. Metintas, et al (Technology in Cancer Research & Treatment, 2021).

The study

• Authors: M. Yakar, D. Etiz, M. Metintas, G. Ak, O. Celik
• Journal: Technology in Cancer Research & Treatment
• Published: 2021;20:15330338211016373
• Cited by: 22 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Background: Radiation pneumonitis (RP) is a dose-limiting toxicity in lung cancer radiotherapy (RT). As risk factors in the development of RP, patient and tumor characteristics, dosimetric parameters, and treatment features are intertwined, and it is not always possible to associate RP with a single parameter. This study aimed to determine the algorithm that most accurately predicted RP development with machine learning.

Methods: Of the 197 cases diagnosed with stage III lung cancer and underwent RT and chemotherapy between 2014 and 2020, 193 were evaluated. The CTCAE 5.0 grading system was used for the RP evaluation. Synthetic minority oversampling technique was used to create a balanced data set. Logistic regression, artificial neural networks, eXtreme Gradient Boosting (XGB), Support Vector Machines, Random Forest, Gaussian Naive Bayes and Light Gradient Boosting Machine algorithms were used. After the correlation analysis, a permutation-based method was utilized for as a variable selection.

Results: RP was seen in 51 of the 193 cases. Parameters affecting RP were determined as, total(t)V5, ipsilateral lung Dmax, contralateral lung Dmax, total lung Dmax, gross tumor volume, number of chemotherapy cycles before RT, tumor size, lymph node localization and asbestos exposure. LGBM was found to be the algorithm that best predicted RP at 85% accuracy (confidence interval: 0.73-0.96), 97% sensitivity, and 50% specificity.

Conclusion: When the clinical and dosimetric parameters were evaluated together, the LGBM algorithm had the highest accuracy in predicting RP. However, in order to use this algorithm in clinical practice, it is necessary to increase data diversity and the number of patients by sharing data between centers.

In plain English

This pilot study from Eskişehir Osmangazi University in Turkey tested whether machine-learning methods could predict radiation pneumonitis, lung inflammation caused by radiotherapy, in people with stage III lung cancer. Of 197 patients treated with radiotherapy and chemotherapy between 2014 and 2020, 193 were analysed, and 51 (26.4%) developed radiation pneumonitis. Of 77 candidate variables, the authors selected 9 as important for prediction, including asbestos exposure, recorded in 18 patients (9.3%). The article does not describe this exposure or say whether it raised or lowered predicted risk. Of seven methods compared, the authors reported that LGBM performed best, with 85% accuracy (confidence interval 0.73-0.96), 97% sensitivity and 50% specificity. The article's figures disagree: it also reports that LGBM correctly predicted 5 of 6 pneumonitis cases and 28 of 33 other test-set cases, which do not match those percentages, and elsewhere describes the test set as 10 cases with pneumonitis and 29 without. The authors concluded that LGBM was most accurate when clinical and dose factors were combined, but that more patients and more varied data, shared between centers, are needed before clinical use.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Single-center pilot study that built and compared machine-learning prediction models for radiation pneumonitis: logistic regression, artificial neural networks, XGB, SVM, random forest, Gaussian Naive Bayes and LGBM
• Who was studied: 197 patients with stage III non-small cell or small cell lung cancer treated with radiotherapy and chemotherapy at Eskişehir Osmangazi University, Turkey, from 2014 to 2020. Four were excluded for missing data, leaving 193 (11 female, 182 male; median age 63).
• Exposure: Thoracic radiotherapy (45-68 Gy) with chemotherapy; the outcome was radiation pneumonitis. Asbestos exposure (present or absent, with no fiber type or dose given) was one of 77 candidate predictor variables and was present in 18 of 193 patients (9.3%).
• Main results:
  • Radiation pneumonitis occurred in 51 (26.4%) of the 193 patients: 28 had grade 1, 22 grade 2 and 1 grade 3. No grade 4 cases were seen.
  • Of 77 variables, 9 were selected as important for predicting radiation pneumonitis: total lung V5; ipsilateral, contralateral and total lung maximum dose; gross tumor volume; number of chemotherapy cycles before radiotherapy; tumor size; lymph node localization; and asbestos exposure. The direction and size of the asbestos effect are not reported.
  • The authors reported LGBM as the best-performing algorithm, with 85% accuracy (confidence interval 0.73-0.96), 97% sensitivity and 50% specificity. Its ROC AUC in Table 5 was 0.73. These sensitivity and specificity figures do not agree with the test-set counts reported in the same article.
  • Accuracy of the other algorithms in Table 5: logistic regression 0.51, artificial neural networks 0.51, XGB 0.79, SVM 0.54, random forest 0.77, Gaussian Naive Bayes 0.46.
  • The text reports that, in the test data set, LGBM correctly predicted 5 of 6 radiation pneumonitis cases and 28 of 33 non-pneumonitis cases. These counts conflict with the reported 97% sensitivity and 50% specificity. They also conflict with the stated test-set composition of 10 cases with pneumonitis and 29 without.
  • Advanced age and pulmonary comorbidities, reported as risk factors in an earlier meta-analysis, were evaluated in this study but did not contribute to radiation pneumonitis.
• The authors' conclusion: The authors concluded that when clinical and dosimetric parameters were evaluated together, LGBM was the algorithm with the highest accuracy for predicting radiation pneumonitis.
• Limitations the authors state:
  • The authors state that more patients and more diverse data, gathered by sharing data between centers, are needed before the algorithm can be used in clinical practice. They describe the work as a pilot study.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Lung cancer.

Medical terms in this study, explained in our Lexicon: Pneumonitis, Lung Cancer, Chemotherapy, Asbestos Exposure, confidence interval.

Related studies

• The use of a next-generation sequencing-derived machine-learning risk-prediction model (OncoCast-MPM) for malignant pleural mesothelioma: a retrospective study — M.G. Zauderer, A. Martin, et al. (2021)
• Combining a gravitational search algorithm, particle swarm optimization, and fuzzy rules to improve the classification performance of a feed-forward neural network — M.L. Huang, Y.C. Chou (2019)
• Radiation pneumonitis in a patient exposed to asbestos — R.F. Ashford, J. Maher, et al. (1981)
• Dosimetric Correlates of Pulmonary Toxicity in Patients with Malignant Pleural Mesothelioma Receiving Radiation Therapy to the Intact Lungs — M.R. Thompson, V.A. Dumane, et al. (2019)
• Differential diagnosis of pleural mesothelioma using Logic Learning Machine — S. Parodi, R. Filiberti, et al. (2015)
• Computational prediction of diagnosis and feature selection on mesothelioma patient health records — D. Chicco, C. Rovelli (2019)

Browse return to medical research.