Realising the Potential for ML from Electronic Health Records

EasyChair Preprint 5442

Abstract

The potential for applying Machine Learning (ML) to Electronic Health Records (EHRs) has been widely agreed but practical progress has been slow. One reason why EHR data are not immediately usable for ML is lack of information about the meaning of the data. An improved description of the data would help to close this gap. However, the description needed is of the data journey from the original data capture, not just of data in the final form needed for ML. We use a simplified example to show how typical EHR data has to be transformed in a series of steps to enable the use of ML technology. We outline some of the typical transformations and argue that the data transformation needs to be visible to the users of the data. Finally, we suggest that synthetic data could be used to accelerate the interaction between medical practitioners and the ML community.

Keyphrases: Bayesian networks, Electronic Health Records, Healthcare, machine learning, synthetic data

@booklet{EasyChair:5442,
  author    = {Haoyuan Zhang and D. William R. Marsh and Norman Fenton and Martin Neil},
  title     = {Realising the Potential for ML from Electronic Health Records},
  howpublished = {EasyChair Preprint 5442},
  year      = {EasyChair, 2021}}