Machine Learning to Predict Mortality in Older Patients With Cancer: Development and External Validation of the...

Area of Science:

• Geriatric Oncology
• Biostatistics
• Machine Learning in Healthcare

Background:

• Prognosis in older cancer patients is complex due to heterogeneity and limitations of current predictive models.
• Accurate prediction is crucial for individualized treatment and care planning in this population.

Purpose of the Study:

• To develop and externally validate the Geriatric Cancer Scoring System (GCSS) for improved mortality prediction.
• To refine individualized prognosis for older cancer patients within one year post-geriatric assessment (GA).

Main Methods:

• Utilized data from two French prospective multicenter cohorts (ELCAPA and ONCODAGE) for training and validation.
• Compared Cox regression, decision tree (DT), and random survival forest (RSF) models using time-dependent area under the receiver operator curve (tAUC).
• Included oncologic, geriatric factors, and routine biomarkers as candidate predictors.

Main Results:

• The random survival forest (RSF) model achieved the highest predictive performance (12-month tAUC: 0.87).
• Key predictors identified included tumor site, metastatic status, weight loss, polypharmacy, impaired function, G-8 score, and inflammatory markers (CRP/albumin).
• The Geriatric Cancer Scoring System (GCSS) was developed based on the RSF model.

Conclusions:

• The GCSS, utilizing a machine learning approach, provides accurate and externally validated mortality prediction for older cancer patients.
• GCSS has the potential to enhance clinical decision-making and patient counseling following geriatric assessment.
• Further validation in international settings is recommended to confirm the generalizability of the GCSS.