Dynamic Mortality Risk Prediction in Myelodysplastic Syndromes Using Longitudinal Clinical Data.

Area of Science:

• Hematology
• Machine Learning in Medicine
• Oncology

Background:

• Myelodysplastic syndromes (MDS) have varied patient outcomes, requiring tailored treatment strategies.
• Current risk stratification tools (e.g., IPSS-R) are static and do not account for disease progression.
• There is a need for dynamic risk assessment to guide clinical decisions throughout the patient's journey.

Purpose of the Study:

• To develop a data-driven, dynamic model for predicting short-term mortality in MDS patients.
• To continuously assess 1-year mortality risk across the disease course.
• To create a tool that supplements existing static risk scores.

Main Methods:

• A machine learning model using gradient-boosted decision trees was developed.
• The model incorporated longitudinal blood parameters and diagnosis-based features.
• Training was performed on a large MDS registry cohort (n=1,024) and validated on independent cohorts (n=317).

Main Results:

• The model achieved an area under the ROC curve of approximately 0.8 in validation cohorts, outperforming static models.
• Accurate mortality risk prediction was achieved within 90 days of diagnosis.
• Feature importance analysis confirmed clinical relevance and interpretability.

Conclusions:

• A dynamic risk model offers continuous, individualized 1-year mortality risk assessment for MDS patients.
• This approach enhances risk stratification beyond static diagnostic scores.
• Incorporating longitudinal data is crucial for accurate MDS risk assessment.