Table 5.
True and spurious covariate relationships identified in the simulated data by the automated stepwise covariate modeling, Lasso, and SOHGA approaches and the models fit characteristics
Method | “True” covariates | Spurious covariates | Objective function value | ||
---|---|---|---|---|---|
Clearance | Volume of distribution | Clearance | Volume of distribution | ||
Original model | BMI, CRCL | BSA, Sex | – | – | 6101.2 |
Stepwise covariate modeling (SCM): p value for inclusion, p value for elimination | |||||
0.05, 0.05 | BMI, CRCL | Sex | WT | HT, CV1 | 6085.9 |
0.05, 0.01 | BMI, CRCL | Sex | – | HT, CV1 | 6091.1 |
0.10, 0.01 | BMI, CRCL | Sex | – | HT, CV1 | 6091.1 |
Lasso model | BMI, CRCL | – | – | – | 6254.2 |
Single-objective, hybrid genetic algorithm | |||||
3.84 point penalty per parameter | BMI, CRCL | Sex | BSA | HT, CV1 | 6086.7 |
10 point penalty per parameter | BMI, CRCL | Sex | – | HT | 6097.9 |
BMI body mass index, BSA body surface area, CRCL creatinine clearance, CV1 unrelated covariate 1, HT height, WT weight