Background: Statistical methods are often considered as mere tools to address research questions. The lack of critical understanding can make their use sometimes highly questionable if not inappropriate. Biostatistics should be seen more as a paradigm than a set of tools. Knowledge of methods means a flexible utilization of them, in which modeling and prediction correspond more to an art than to a routine use dictated by circumstances and habits.
Summary: Tree-based methods (or tree-growing techniques) are discussed here as a flexible statistical framework for modeling and prediction to address key questions such as prognostic stratification and treatment effects heterogeneity in both randomized clinical trials and observational studies.
Key messages: We provide some examples in neurology and possible future extensions in which tree-based methods are shown to be crucial for the assessment of the best available therapy for a patient. We show how trees can represent a clinically interpretable and easy-to-implement approach for stratified medicine and treatment tailoring based on responsiveness, as well as for selecting populations for new studies.
© 2016 S. Karger AG, Basel.