Supervised machine learning model to predict oncotype DX risk category in patients over age 50.

Area of Science:

• Oncology
• Genomics
• Machine Learning

Background:

• The Oncotype DX Recurrence Score (RS) is crucial for guiding treatment in early-stage, estrogen receptor-positive, HER2-negative (ER+/HER2-) breast cancer.
• International use of RS testing is limited by cost and accessibility.
• Predicting RS risk category using accessible clinicopathologic variables is needed, especially for patients over 50 years old.

Purpose of the Study:

• To develop and validate a supervised machine learning model to predict the RS risk category in patients over 50 years old with ER+/HER2- breast cancer.
• To identify patients for whom chemotherapy can be safely omitted based on predicted low RS risk.
• To assess the model's performance in a real-world validation cohort.

Main Methods:

• A random forest model was developed using clinicopathologic data (age, tumor size, histology, PR expression, LVI, grade) from a training cohort.
• The model predicted low (RS ≤ 25) versus high (RS > 25) risk categories.
• Model performance was evaluated on a separate validation cohort of 1293 patients aged over 50 with T1-2, ER+/HER2-, node-negative tumors.

Main Results:

• The model demonstrated high specificity (96.3%) and negative predictive value (92.9%) for identifying patients with a low RS (≤ 25).
• This indicates high confidence in identifying patients unlikely to benefit from chemotherapy.
• Sensitivity and positive predictive value for identifying high RS were lower (48.3% and 65.1%, respectively).

Conclusions:

• The developed machine learning model is highly specific in identifying older patients (≥50 years) with ER+/HER2- breast cancer who can potentially omit chemotherapy.
• The model shows promise for triaging patients for RS testing in resource-limited settings.
• External validation is recommended before widespread clinical implementation.