No generic approach for sample size estimation can be made in personalised medicine when designing a stratification cohort. The conclusion is drawn from the first stakeholder consultation by the PERMIT project which looked to identify approaches to estimate the number of patients needed to be included in a cohort for different types of stratification studies using machine learning. This workshop was the first in a series organised by the PERMIT project which will be used to elaborate the robust and reproducible recommendations for Personalised Medicine.
There are many specific study factors that must be considered when determining sample size. The approach for sample size estimation needs to be chosen according to the specific study type and goals as well as the parameters that influence the sample size requirements including :
a) Diagnostic vs. prognostic vs. drug-related biomarker discovery
b) Supervised vs. unsupervised setting
c) Categorical vs. quantitative outcomes for supervised studies
d) Model discovery vs. model validation
e) Treatment response prediction vs. patient sub-group stratification
f) Mechanistic model building vs. purely statistics-based model building
g) Benefit maximization vs. benefit/risk maximization
For further details on the factors and their influence on sample size calculation stay tuned for the full PERMIT project recommendation.