Model-assisted predictions on prognosis in HNSCC: do we learn?

Dedicated software packages incorporating prognostic models are meant to aid physicians in making accurate predictions of prognosis. This study concerns 742 predictions of 5-year survival on consecutive newly diagnosed patients with head- and neck squamous cell carcinoma. The 5-year survival predictions made by the physicians are not compared with actual survival, but with a prediction made by OncologIQ, a dedicated software package. We used a linear regression and a linear mixed-effects model to look at absolute differences between both predictions and possible learning effects. Predictions made by the physicians were optimistic and inaccurate. Using the linear regression and linear mixed-effects models, the physicians' learning effect showed little improvement per successive prediction. We conclude that prognostic predictions in general are imprecise. When given feedback on the model's predicted survival, the accuracy increases, but only very modestly.