Method for evaluating prediction models that apply the results of randomized trials to individual patients

Table 4 Results for the ACCENT data for a treatment threshold (T) of 0.05.

Strategy	Treat no patients, regardless of model	Treat all patients, regardless of model	Treat according to prediction model
			Prediction model recommends treatment	Prediction model recommends no treatment	Prediction model total
Number of patients	1397	1474	889	564	1453
Number of events	568 (40.66%)	472 (32.02%)	343 (38.6%)	139 (24.6%)	482 (33.2%)
Decrease in event rate		8.64%			7.49%
Number of treatments		1474 (100%)			889 (61.2%)
Net benefit		8.64% – 100% × 0.05 = 0.0364			7.5% – 61.2% × 0.05 = 0.0443

The results in the first and second columns give the results from the treatment and control groups. The third column shows that there were 889 patients in the treatment group for whom the prediction model estimated a benefit of 5% or more and that 343 of these died. The fourth column shows that there were 564 patients in the control group of the randomized trial with predicted benefit of less than 5%, of whom 139 died. The final column shows the overall results for patients given an individualized prediction.

ISSN: 1745-6215