[Skip to Content]
[Skip to Content Landing]
Comment & Response
November 3, 2015

Risk Prediction for Individuals—Reply

Author Affiliations
  • 1Royal Victoria Hospital, McGill University, Montreal, Quebec, Canada
  • 2Department of Mathematics and Statistics, Boston University, Boston, Massachusetts
  • 3Duke Clinical Research Institute, Duke University, Durham, North Carolina
JAMA. 2015;314(17):1875-1876. doi:10.1001/jama.2015.12221

In Reply We agree with Dr Van Calster and colleagues that risk assessment is valuable in the care of individuals. We stated in the conclusion of our Viewpoint that “Predictive algorithms are an essential component of guideline recommendations.” The phrase about probability not being meaningful in the individual context was misunderstood and not intended to imply that calculating individual risks was not useful.

Instead, we wanted to direct the reader to the excellent discussion by Cohen1 of the limitations of the frequentist notion of probability understood as an estimate of the frequency of a particular event in a total sequence of events. Basically, we meant that frequentist probability is not verifiable in an individual context because a person either experiences or does not experience the event. Consequently, we agree that calibration is important in the assessment of predictive model performance.