In Phase II and Phase III these models inform the risk of an event by incorporating individual patients’ biomarker evolution, which may be governed by a complex non-linear process. During biomarker selection, these models help quantify biomarkers' impact on clinical endpoints.
Nonlinear process dynamics, predicting for the study arm, individual patients, and biomarker.
Non-parametric and semi-mechanistic biomarker models, hierarchical models with external controls, causal estimands, and post-stratification to adjust for the differences between the sample and the population.
These models are often used in preclinical and Phase I trials in order to establish safe dosing regimens. While safety is the main concern, we strive to detect signs of early efficacy.
Small data and non-linear system dynamics.
We are able to infer both the population and individual level parameters and present it to non-specialists in an easily interpretable form.
This is the wide data of the big data spectrum. Traditional machine learning approaches including clustering and dimensionality reduction are unlikely to uncover meaningful clinical relationships.
Too many possible combinations frequently guide the researchers to confuse signal with noise.
Using better priors and Bayesian non-parametrics increases our chances of finding the best biomarker candidates while retaining inferential uncertainty, and guarding against over-confidence.
.svg)