Skip to main content
. 2021 Nov 25;13(1):121–133. doi: 10.1002/jrsm.1537

TABLE 2.

Performance of the classifiers on test set A (n = 9368), B (n = 7,85) and C (n = 5812)

Classifier Training criteria a Set RCTs Non‐RCTs Screening reduction %
Precision % Recall % Precision % Recall %
RCT classifier Include RCT in any human health domain A 12.3 99.7 3.4 85.9 43.3
B 8.1 99.7 2.4 83.5 50.9
C 11.1 99.2 2.6 87.5 43.6
Custom 1 Include any studies that are in the health promotion domain (all studies without Ex1 code) A 11.7 99.1 3.8 99.5 40.9
B 7.8 100% 2.8 100% 49.3
C 13.6 99.5 3.5 96.9 54.2
Custom 2 Include any RCTs, non‐RCTs or reviews in the health promotion domain (all studies without Ex1, Ex2 or Ex3 codes) A 16.5 98.8 5.4 98.6 58.4
B 12.2 99.3 4.3 97.1 67.6
C 19.6 98.1 4.9 93.8 68.6
Custom 3 Include any studies that are RCTs or non‐RCTs in the health promotion domain (all studies without Ex1, Ex2, Ex3 or Ex4 codes) A 19.7 98.0 6.3 96.2 65.4
B 14.6 99.0 5.1 97.1 72.9
C 23.2 97.8 5.8 92.7 73.5
a

Ex1, Ex2, Ex3, Ex4 are described in Table 1.