Description
We used UK primary care EHR data to discover a minimum set of best features which could identify the onset of dementia. After specifying 88 possible predictive features, we used random forest classifiers to identify the top features in the 5 years prior to diagnosis, in a dataset of > 90,000 dementia cases and matched controls. Classification performance improved significantly within the last year before diagnosis, when memory loss was evidently recognized by clinicians.
Learning Objective: Understand challenges and possible solutions in predicting dementia onset using data from primary care electronic health records.
Authors:
Elizabeth Ford (Presenter)
Brighton and Sussex Medical School
Johannes Starlinger, Charité – Universitätsmedizin Berlin
Presentation Materials: