Automated covariate selection: SAMBA and COSSAC algorithms

Authors: Magnard C
Conference: ACoP
Keywords: ACoP2023, COSSAC, covariate modeling, Lixoft, machine learning, monolix, MonolixSuite Playlist, SAMBA
Software: Monolix®
Division: Clinical Pharmacology and Pharmacometrics (CPP)

SAMBA Key Idea

  • First idea: add all covariates having a low p-value
  • Problem: some covariates might be correlated with each other and carry redundant information e.g weight and BMI
  • Solution: For each parameter,
    • linear regression between individual parameters and combinations of covariates
    • Add covariates used in the best regression model (based on a BIC)

By Claude Magnard

ACoP 14 American Conference on Pharmacometrics, National Harbor, MD, November 5th – 8th, 2023