Figure 1.
Brief outline of the augmented backward elimination procedure.
Table 1.
Simulation study: bias and root mean squared error (RMSE) of regression coefficients of a continuous exposure variable
in unselected models, models selected by backward elimination (BE) and models selected by augmented backward elimination (ABE) for linear, logistic and Cox regression.
Table 2.
Urine osmolarity example: demographic and clinical characteristics of all 245 patients at baseline.
Figure 2.
Urine osmolarity example: selection path (left column) of standardized regression coefficients and model stability (inclusion frequencies) in
bootstrap resamples (right column) for backward elimination (BE) and augmented backward elimination (ABE).
First row: BE with ; second row: ABE with
and
; third row: ABE with
and
. Abbreviations: ABE, augmented backward elimination; BE, backward elimination; log2UOsm, log2 of urine osmorality; log2CCL, log2 of creatinine clearance; log2Prot, log2 of proteinuria; BBlock, use of beta-blockers; PKD, presence of polycystic kidney disease; Diur, use of diuretics; Age, age in decades; ACEI, use of angiotensin-converting enzyme inhibitors and Angiotensin II type 1 receptor blockers; MAP, mean arterial pressure.
Figure 3.
Urine osmolarity example: number of selected variables in the final models of bootstrap resamples for backward elimination BE with
and augmented backward elimination ABE with
and
.
The highlighted bars indicate the number of selected variables in the original sample. Abbreviations and symbols: , significance threshold; ABE, augmented backward elimination; BE, backward elimination;
, change-in-estimate threshold;.
Table 3.
Urine osmolarity example: final models selected by backward elimination (BE) with a significance threshold , augmented backward elimination (ABE) with
and a change-in-estimate threshold
, and unselected model (No selection).
Table 4.
Urine osmolarity example: incorporating model uncertainty into standard error (SE) estimates of urine osmolarity UOSM.