Prediction Models: Revolutionary in Principle, But Do They Do More Good Than Harm?

See accompanying article on page 2959; listen to the podcast by Dr Cooperberg at www.jco.org/podcasts It can sometimes seem as though we are drowning in predic-tion models. Every month brings a multitude of newly published risk calculators and nomograms to add to the multitude already in the literature—there are more than 100 prediction models on prostate cancer alone1—and Web sites such as www.nomograms.org, www.nomogram.org, and www.cancernomograms.org continue to proliferate. As such, it is easy to become somewhat inured to predic-tion modeling and thus to forget that it constitutes an important shift in the way that medicine is practiced. Biopsy for prostate cancer is a good illustration of the potential role of prediction models. For many years, the typical approach has been to consider biopsy in men with a prostate-specific antigen (PSA) above 4 ng/mL; a positive clinical exam (the digital rectal examina-tion); or perhaps a low free-to-total PSA ratio, such as less than 15%.2