Predicting overall survival in anaplastic thyroid cancer using machine learning approaches.

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Anaplastic thyroid carcinoma (ATC) is an aggressive cancer with a poor prognosis.
• Machine learning (ML) offers potential for improved survival prediction in ATC.

Purpose of the Study:

• Develop and validate ML models to predict 3-, 6-, and 12-month overall survival (OS) in ATC patients.
• Utilize the SEER database for model development and validation.

Main Methods:

• Employed five ML algorithms: AdaBoost, support vector machines (SVC), gradient boosting, random forests, and naive Bayes.
• Data from the SEER database (2004-2015) were split into training (70%) and testing (30%) sets.
• Model performance was evaluated using concordance index (C-index) and Brier score, with fivefold cross-validation for tuning.

Main Results:

• Gradient boosting model excelled in 3-month survival prediction (C-index: 0.8197).
• AdaBoost model demonstrated superior performance for 6-month survival (C-index: 0.8473).
• SVC model showed the best results for 12-month survival (C-index: 0.8347).
• Key predictors for 6-month OS included surgery, stage IVC, radiation, chemotherapy, and tumor size, with treatments improving and higher stages reducing survival.

Conclusions:

• ML algorithms can accurately predict short-term survival in anaplastic thyroid carcinoma.
• These models hold potential for guiding clinical decision-making and tailoring treatment strategies for ATC patients.