Web-based prognostic calculators are commonly used by both clinicians and patients to predict who is at-risk for various diseases, including diabetes, heart failure, and common cancers, among others. However, most calculators are developed using data from a group of “average” patients. As a result, these calculators can give patients and clinicians misleading results as no patient is an “average” patient.
In a new study, lead author Farrokh Alemi and Arthur Williams, both of the Department of Health Administration and Policy, along with Cari Levy of Denver Veteran Administration Medical Center, Bruce Citron and Allison Williams of Bay Pines Veteran Administration Medical Center, and Pracht Etienne of the University of South Florida, explain how the results of these web-based prognostic calculators can be misleading. The study is published in the Journal of Palliative Medicine.
As an example, the authors examined data from the electronic health records of 140,699 nursing home residents in 126 medical centers in the Veterans Administration Informatics and Computing Infrastructure. They used the data to demonstrate the misleading results in which hypertension increased mortality risk for patients without stomach cancer, but decreased mortality risk for patient with stomach cancer.
“We outline several approaches and methods to address these problems of misleading results,” Alemi said. “By using propensity matching or other similar methods, in addition to data balancing, we can remove confounding from the results, which would allow for more accurate results based on an individual’s health data.”