An empirical analysis of dealing with patients who are lost to follow-up when developing prognostic models using a cohort design

Table 3 AUROC results when predicting the simulated outcome within 3 years, when loss to follow-up is based on Charlson comorbidity index

Percentage censored (thres) (%)	Number in training Target Pop (64,770) censored	Training Outcome count (10,104) lost to follow-up		Logistic keep lost to follow-up	Logistic remove lost to follow-up	Logistic remove lost to follow-up non-outcomes	Cox keep lost to follow-up
Percentage censored (thres) (%)	Number in training Target Pop (64,770) censored	Before loss to follow-up date	After loss to follow-up date	Test AUROC (train AUROC)
~ 10	6488	527	901	0.685 (0.697)	0.684 (0.702)	0.675 (0.735)	0.685 (0.693)
~ 20	12,946	1024	1606	0.680 (0.695)	0.683 (0.711)	0.654 (0.754)	0.684 (0.687)
~ 30	19,371	1422	2294	0.678 (0.692)	0.681 (0.710)	0.636 (0.778)	0.682 (0.680)
~ 40	25,834	1925	2847	0.677 (0.692)	0.679 (0.707)	0.621 (0.800)	0.682 (0.675)
~ 50	32,313	2289	3450	0.677 (0.692)	0.676 (0.706)	0.607 (0.837)	0.681 (0.671)
~ 70	45,271	2973	4387	0.681 (0.708)	0.671 (0.693)	0.592 (0.865)	0.678 (0.674)
~ 90	58,274	3726	5394	0.684 (0.714)	0.654 (0.723)	0.590 (0.916)	0.676 (0.689)

ISSN: 1472-6947