Seattle-based University of Washington researchers used the "model to data" approach, which allows machine-learning research on private biomedical data, to create predictive analytics models without having direct access to patient EHRs.
The researchers described their pilot project in a July 8 study published in the Journal of the American Medical Informatics Association. The group investigated the MTD approach in which developers send models to an isolated environment for training and evaluation of sensitive data. For the analysis, researchers asked model developers to develop a model that predicts the likelihood of patient mortality within 180 days of the patient's last visit.
The researchers chose all UW patients who had at least one visit in the EHR dataset UW Observational Medical Outcomes Partnerships Common Data Model, which represented 1.3 million patients, 22 million visits, 33 million procedures, 5 million drug exposure records, 48 million condition records, 10 million observations and 221 million measurements, according to the report.
The model developer was able to create three mortality prediction models under the MTD framework using demographics, five common chronic diseases and the 1,000 most common features from the EHR's condition, procedure and drug domains. The developer used UW Medicine's EHR data without gaining full access to the dataset or clinical environment.
"We believe this enables future predictive analytics sandboxing activities and the development of new clinical predictive methods safely," the researchers concluded.