OR WAIT null SECS
The deep learning survival neural network model demonstrated the potential to provide personalized treatment recommendations based on real clinical data in patients with non-small cell lung cancer.
A study published in JAMA Network Open found that the deep learning survival neural network model (DeepSurv) shows possible benefits in prognostic evaluation and treatment recommendation with regard to lung cancer-specific survival.
Additionally, researchers demonstrated the potential for DeepSurv to provide personalized treatment recommendations based on real clinical data in patients with non-small cell lung cancer (NSCLC).
“The results of our pilot study proved that the deep learning network model (DeepSurv) performed better than conventional linear regression modeling (TNM staging model) in postoperative outcome prediction for patients with newly diagnosed NSCLC,” the authors wrote. “Also, this model may serve as a useful analytical tool for treatment recommendation in patients with NSCLC, given its evidence of the significant prognostic benefits of following the treatment recommendation, which clearly outweigh those associated with not following the recommendation.”
Researchers developed and validated DeepSurv using consecutive cases of newly diagnosed patients with stages I to IV NSCLC between January 2010 and December 2015 in a Surveillance, Epidemiology, and End Results (SEER) database. In total, 127 features, including patient characteristics, tumor stage, and treatment strategies, were assessed for analysis. Moreover, the algorithm was externally validated on an independent test cohort made up of 1182 patients with stage I to III NSCLC diagnosed between January 2009 and December 2013 in Shanghai Pulmonary Hospital.
Overall, 17,322 patients with NSCLC were included in the study and the majority of tumors were stage I disease (10,273 [59.3%]) and adenocarcinoma (11,985 [69.2%]). Notably, the median follow-up time was 24 (range, 10-43) months and 3119 patients died from lung cancer-related complications during the follow-up period.
DeepSurv demonstrated more promising results in the prediction of lung cancer-specific survival than the tumor, node, and metastasis stage on the test data set (C statistic = 0.739 vs 0.706). Even further, the population who received the recommended treatments had superior survival rates than those who received treatments not recommended (HR, 2.99; 95% CI, 2.49-3.59; P < 0.001), which was verified by propensity score-matched groups.
“As a new analytic tool, the deep learning network model will likely become more widely applied to support clinical decision-making,” the authors wrote.
In order to facilitate the survival predictions and treatment recommendations of the model, researchers also developed a user-friendly interface. This interface consists of 3 views, including the user input view, the survival prediction view, and the treatment recommendation view.
“To date, identifying patients who are appropriate for initial surgical management and conveying individualized prognostic analyses of postoperative outcomes has been an elusive goal. Instead, most published models are guided by patient characteristics to generate prognostic factors and are influenced by biases for different treatments,” the authors wrote. “The DeepSurv model and its user-friendly graphic interface has the potential to address this clinical dilemma and better share individual outcomes following different surgical procedures.”
Notably, the investigators indicated that external validation is lacking in this current study. Therefore, further study is necessary to validate the advantages of DeepSurv in survival prediction.
She Y, Jin Z, Wu J, et al. Development and Validation of a Deep Learning Model for Non-Small Cell Lung Cancer Survival. JAMA Network Open. doi: 10.1001/jamanetworkopen.2020.5842.