Identifying prognostic groups using machine learning tools following chemoradiation for NSCLC.

The objective of this research paper was to improve the prediction of treatment outcomes in patients with unresectable stage III Non-Small Cell Lung Cancer (NSCLC) by utilizing Cluster Analysis (CA), a machine learning tool capable of identifying complex interactions among variables. The study analyzed treatment outcomes of 92 patients who underwent chemoradiation for inoperable locally advanced NSCLC between 2012 and 2018. Multivariate analysis was performed to identify variables significantly impacting survival, which were then used in a two-step CA to identify a cluster of patients with a better prognosis.

With a median follow-up of 18 months, the median overall survival (OS) was 14 months. Using four input variables (Overall Treatment Time, total radiation dose, technique of radiation, and stage), the CA generated three clusters within the primarily stage III NSCLC group. Cluster 2, characterized by an Overall Treatment Time of up to 80 days, a mean radiation dose of 63.5 Gy, and the use of IGRT, had a median OS of 36 months. In contrast, Cluster 1, with an Overall Treatment Time of up to 96 days, a mean radiation dose of 58 Gy, and conventional techniques, had a median OS of 18 months (p < 0.000).

In conclusion, a cluster with a relatively good prognosis was identified within the stage III NSCLC group. The two-step CA approach attempted to create a model that could provide more specific prognostic information beyond what is provided by TNM-based models.