Inspiration
GVHD is a major complication after allogeneic HCT. We wanted to make risk prediction more clinically useful by focusing on how patients move through different outcomes over time.
What We Learned
Transplant outcomes are dynamic and competing: GVHD, relapse, and mortality cannot be understood as isolated events. We also learned that prediction and treatment recommendation require different levels of evidence.
How We Built It
Trajectory-GVHD uses a multi-state machine-learning framework to estimate individual GVHD trajectories and support personalized prophylaxis decisions.
[ \text{Event-free} \rightarrow {\text{GVHD},\ \text{Relapse},\ \text{Death}} ]
Challenges
The main challenge was harmonizing heterogeneous clinical variables and avoiding overclaiming treatment effects. We addressed this by separating predictive risk modeling from causal treatment evaluation.
Built With
- bone-marrow-transplant
- clinical-decision-support
- data-science
- deep-learning
- gvhd
- healthcare
- hematology
- machine-learning
- medical-ai
- multicohort
- personalized-medicine
- predictive-analytics
- risk-stratification
- survival-analysis
Log in or sign up for Devpost to join the conversation.