Time to compare impact and feasibility of prediction models in real life.

Siontis and colleagues’ conclusion that comparison of risk prediction models for cardiovascular disease “would benefit from standardised reporting and formal, consistent statistical comparisons” is only part of the story. The other urgent need is for feasibility and impact studies comparing the use of risk prediction models in real clinical practice: Can local computer systems handle the model? Can GPs and others be trained to use it? What are the resource implications? We recently performed a similar review of type 2 diabetes risk prediction models and scores. In addition to collating statistical evaluations of the models, we reviewed the impact in real clinical practice.We discovered that it wasminimal. Researchers have spent too much time making minor adjustments to in silico statistical models, and too little considering who would use the model, on whom, and in what circumstances.