Multivariate Model for Predicting Overall Survival in Malignant Pleural Mesothelioma After Adjuvant Radiotherapy

Abstract

<h3>Purpose/Objective(s)</h3> Malignant pleural mesothelioma (MPM) is a rare but aggressive cancer arising from the cells of the thoracic pleura. Its anatomic complexity, large surface area, and unique location pose challenges for target coverage while sparing adjacent organs at risk (OARs) during treatment planning for adjuvant radiotherapy after pleurectomy and decortication. We aimed to investigate the associations between patient/target characteristics and achievable dosimetric plan quality with overall survival. <h3>Materials/Methods</h3> Sixty patients with MPM patients who were treated on a tomotherapy unit between 2013-2018 at a single institution were included. All patients received helical IMRT with 45 Gy/25 fractions. Dosimetric quantities for the lungs, heart, and liver were systematically examined for 37 right-sided (RSM) and 23 left-sided mesothelioma (LSM) patients. Patient characteristics and/or planning dosimetric metrics were input into machine learning algorithms, e.g., LASSO regression, support vector machine (SVM), and multiple layer perceptron (MLP), in order to construct predictive models for overall survival. The performance of predictive models was evaluated with area under the curve (AUC) using 3-fold cross validation. Normal tissue complication probabilities (NTCP) for selected OARs were estimated using the Lyman-Kutcher-Burman model based on individual dose volume histograms. <h3>Results</h3> The achieved dosimetric endpoints were significantly different (<b>P</b> < 0.05) between LSM vs. RSM patients, respectively: ipsilateral lung V20_Gy of 91.2 ± 6.7 vs. 85.3 ± 6.5%, ipsilateral mean lung doses of 39.1 ± 3.8 vs. 36.9 ± 2.5 Gy, heart mean doses of 25.2 ± 3.6 vs. 18.8 ± 3.7 Gy, and liver mean doses of 10.6 ± 2.7 vs. 23.4 ± 4.2 Gy. There was no significant difference in total lung – planning target volume (PTV) mean dose, 13.4 ± 2.0 vs. 14.4 ± 1.5 Gy, nor expected lung pneumonitis for the LSM vs. RSM. No differences were observed between PTV coverage of 94.2 ± 1.2 vs. 93.4 ± 2.4%, and average survival intervals were 16.8 ± 14.4 months and 18.0 ± 18.0 months after RT between LSM vs. RSM. The MLP model achieved AUCs of 0.65, 0.67 and 0.71 using patient characteristics, dosimetric features, and combined features. The identified predictors most associated with overall survival were age, PTV volume, ipsilateral lung volume/V20_Gy, and ipsilateral lung – PTV V20_Gy <h3>Conclusion</h3> We demonstrated correlations between patient specific characteristics and achieved planning dosimetric parameters for left- and right-sided MPM treatment planning endpoints. Combining patient characteristics and dosimetric endpoints, the multivariate predictive model achieved decent performance in predicting overall survival. The predictive model may guide treatment planning to achieve optimal planning dosimetry, and to improve clinical outcome via personalized treatment for MPM.