The use and misuse of risk prediction tools for clinical decision-making

Risk prediction tools are of great value in supporting clinical decision-making: they can identify, for example, the potential benefit of a treatment on prognosis of an individual patient, the risk of (late) health complaints, or the risk of further spread of a disease. These individual estimates can consequently be used to weigh the benefits of a certain treatment against its harms, which is crucial in the current era of personalised medicine. Additionally, more and more diseases are known to be very heterogeneous which is reflected in the necessity for prediction tools to incorporate the multifactorial influence of a wide range of different patientand disease-related variables to provide accurate estimates. A very well-known example of a clinical risk prediction tool in breast cancer is PREDICT [1], which has very recently been updated to a version including progesterone receptor expression [2]. PREDICT estimates individual overall survival probabilities for breast cancer patients, in combination with additional survival benefits of several adjuvant treatment modalities. The underlying model uses expected treatment effects obtained from randomised controlled trials and is specifically designed for treatment decision-making. A similar approach was used to develop the PORTRET tool – an alternative prediction tool specifically for older breast cancer patients – since the PREDICT tool was shown to inaccurately predict survival in this patient group [3].