A novel analytical framework for risk stratification of real-world data using machine learning: A small cell lung...

Area of Science:

• Oncology
• Data Science
• Biostatistics

Background:

• Current small cell lung cancer (SCLC) staging leads to broad prognostic subgroups.
• Tumor, Node, and Metastasis (TNM) staging offers potential for improved patient stratification.
• Real-world data (RWD) can enhance understanding of SCLC heterogeneity.

Purpose of the Study:

• To develop a pipeline integrating machine learning and thoracic oncology expertise.
• To identify and analyze distinct prognostic subgroups within an SCLC patient cohort.
• To leverage RWD for improved SCLC patient classification and treatment strategies.

Main Methods:

• Utilized a development pipeline with thoracic oncologists and machine learning.
• Applied unsupervised learning (partition around medoids) to detect patient subgroups.
• Employed Cox regression and random survival forest to analyze covariate impact on prognosis.

Main Results:

• Identified 7 distinct, well-separated patient clusters from a cohort of 636 SCLC patients (Stage IIIA-IVB).
• Key characterizing factors included performance status, lactate dehydrogenase, metastasis spread, cancer stage, and CRP.
• The chosen clustering method demonstrated superior performance over standard techniques in detecting meaningful subgroups.

Conclusions:

• Machine learning and RWD analysis can refine prognostic subgroup detection in SCLC.
• This approach facilitates a deeper understanding of SCLC disease patterns.
• Potential for developing individualized therapies and improving healthcare decision-making for SCLC patients.