Predicting the early risk of chronic kidney disease in patients with diabetes using real-world data

Journal: Nature Medicine

Published: 2019-01-07

DOI: 10.1038/s41591-018-0239-8

Affiliations: 5

Authors: 13

Go to article

Research Highlight

A tool for predicting chronic kidney disease

© Hero Images/Getty

© Hero Images/Getty

An algorithm that uses real-world data from electronic health records can more accurately identify patients at early risk for chronic kidney disease than other predictive models based on clinical trial datasets.

A team led by scientists from Roche’s diabetes care unit analysed five years’ worth of medical records from over 400,000 people with diabetes. After extracting seven features contained within the demographic, clinical and laboratory records, they trained an algorithm to predict later development of chronic kidney disease—a common complication of diabetes.

In a head-to-head battle of algorithms, the new one based on real-world evidence outperformed published models originating from study results. This is presumably because the diversity of data generated from everyday visits to the doctor’s office is more representative of the general population than those derived from strictly controlled trials.

Supported content

  1. Nature Medicine 25, 57–59 (2019). doi: 10.1038/s41591-018-0239-8
Institutions FC
IBM Research - Zurich, Switzerland 0.38
Roche Diagnostics GmbH - Mannheim, Germany 0.38
Lilly Research Laboratories (LRL), United States of America (USA) 0.08
Indiana Biosciences Research Institute, United States of America (USA) 0.08
Regenstrief Institute, United States of America (USA) 0.08

Return