Table 2.

Results of the simulated recruitment for model configurations with the highest recruitment score. For each model the proportions of recruited patients in non-progressive category (N), and three progressive categories related to pain (P), structure (S), and combined pain and structure (P ​+ ​S) are shown. Additionally, the value and the corresponding rank amongst all configurations (in brackets) of the F₁ score and the area under the ROC curve is given, together with the recruitment score (RS) using progressive weights.

selection model	r.fun.	N	only P	only S	P ​+ ​S	F1 score	AUC(P)	AUC(S)	RS
MUST, 3y shift, top 60	sum	45%	37%	12%	7%	0.658 (32)	0.696 (50)	0.429 (70)	0.216
MUST, 5y shift, top 60	sum	50%	38%	7%	5%	0.669 (15)	0.803 (11)	0.514 (65)	0.164
HOSTAS, no shift, top 60	sum	23%	52%	7%	18%	0.458 (83)	0.628 (43)	0.570 (62)	0.428
HOSTAS, 2y shift, top 60	sum	42%	32%	15%	12%	0.478 (83)	0.642 (12)	0.482 (44)	0.336
HOSTAS, 3y shift, top 60	sum	47%	37%	10%	7%	0.597 (77)	0.718 (1)	0.449 (72)	0.196
HOSTAS, 5y shift, top 60	sum	67%	27%	5%	2%	0.584 (76)	0.696 (10)	0.548 (5)	0.068
DIGICOD, no shift, top 30	z-score	27%	40%	7%	27%	0.422 (71)	0.735 (28)	0.596 (8)	0.379
DIGICOD, 2y shift, top 30	z-score	43%	30%	10%	17%	0.468 (82)	0.643 (45)	0.597 (16)	0.292
DIGICOD, 3y shift, top 30	sum	50%	37%	10%	3%	0.596 (77)	0.680 (28)	0.353 (43)	0.113
PROCOAC, 2y shift, top 30	z-score	23%	30%	27%	20%	0.510 (79)	0.695 (12)	0.592 (11)	0.438
PROCOAC, 3y shift, top 30	sum	40%	47%	7%	7%	0.626 (75)	0.677 (72)	0.392 (35)	0.200
PROCOAC, 5y shift, top 30	sum	50%	40%	7%	3%	0.615 (77)	0.679 (35)	0.476 (19)	0.128
CHECK, 2y shift, top 150	z-score	35%	31%	21%	14%	0.507 (59)	0.697 (20)	0.621 (2)	0.449
screening model, top 150	z-score	29%	30%	18%	23%	0.544 (47)	0.667 (60)	0.617 (36)	0.537
uninformed selection		63%	12%	20%	5%	—	—	—	—