Constraint based temporal event sequence mining for Glioblastoma survival prediction

Area of Science:

• Oncology
• Bioinformatics
• Medical Informatics

Background:

• Glioblastoma (GBM) is a rare cancer with significant treatment challenges.
• Developing personalized treatment plans for GBM patients is crucial for improving outcomes.

Purpose of the Study:

• To predict which Glioblastoma (GBM) patients will survive longer than the median survival time.
• To assess the predictive power of treatment patterns in conjunction with clinical and genomic factors.

Main Methods:

• Developed a predictive model using clinical and genomic data from approximately 300 newly diagnosed GBM patients.
• Employed sequential mining algorithms with 'exact-order' and 'temporal overlap' constraints to extract treatment patterns.
• Utilized logistic regression and Cox regression models to predict patient survival outcomes.

Main Results:

• Identified key predictive features including gene mRNA expression levels, age, Karnofsky performance score, and prescribed therapeutic agents.
• Achieved a c-statistic of 0.85 with the logistic regression model and 0.84 with the Cox regression model.
• Demonstrated the significant impact of diverse feature sources on survival prediction accuracy.

Conclusions:

• Diverse data sources, including clinical, genomic, and treatment patterns, are vital for predicting GBM patient survival.
• The developed predictive model represents a foundational step towards personalized treatment strategies for GBM.
• The methodology can be extended to other cancer types for broader application in personalized oncology.