Ten principles to strengthen prognosis research

Abstract

This chapter provides a set of ten principles for ensuring high-quality prognosis research. There are three general principles for strengthening prognosis research: the need for study registration and protocols, use of reporting guidelines, and importance of replication and validation studies. The seven other principles concern study analysis and presentation: use of estimation and confidence intervals rather than statistical hypothesis testing; use of interaction estimates when analysing subgroups; avoidance of categorization of continuous predictor and outcome variables; multiple imputation of missing values; adjustment of new prognostic factor estimates for established factors; avoidance of univariable estimates for predictor selection when developing prognostic models; use of penalization techniques within prognostic model development to reduce overfitting and overly extreme predictions for new individuals; and use of competing risk models in frail populations.