OHDSI Home | Forums | Wiki | Github

How to develop a dynamic disease prediction model?


(Junjie Cai) #1


I’ve read Design and implementation of a standardized framework to generate and evaluate patient-level prediction models using observational healthcare data and found a great conceptual framework for patient level prediction.

However, it’s limited that each patient have only one index date determined by certain clinical events, so it’s not suitable to use the model to give dynamic prediction (e.g. calculate risk score every hour for each patient)

Since the clinical data of patients update dynamically after admittance to hospital, It’s valuable for a prediction model if it update the risk prediction dynamically.

Do you have advice how to build a dynamic patient prediction model, or can you recommend some material on this topic?

(Christian Reich) #2


Wait. The index date is for the training and validation runs. Nothing prevents you from re-running the actual model, once ready and deployed, every hour or every time a new datum comes in for the patient. But that could probably be built in a more efficient way.