Skip to main content
. 2020 May 6;27(6):877–883. doi: 10.1093/jamia/ocaa032

Table 1.

Test set performance of labeling heuristic requiring multiple disease-specific code mentions compared to phenotype classifiers trained with data labeled using this multiple mentions approach

Phenotype Prevalence of cases in test set Multiple mentions of SNOMED code
APHRODITE classifier
Recall boost using classifier Precision loss using classifier
No. of mentions Recall Precision Recall Precision
Appendicitis 0.05 2 0.31 1.00 0.97 0.99 0.66 0.01
T2DM 0.14 4 0.24 0.99 0.60 0.91 0.36 0.08
Cataracts 0.17 4 0.07 0.97 0.63 0.93 0.56 0.04
HF 0.02 4 0.33 0.94 0.99 0.56 0.66 0.38
AAA 0.04 4 0.22 0.99 0.53 0.97 0.31 0.02
Epileptic seizure 0.02 4 0.06 1.00 0.22 0.94 0.17 0.06
PAD 0.05 4 0.18 0.98 0.91 0.91 0.72 0.07
Adult onset obesity 0.36 4 0.20 1.00 0.29 0.91 0.09 0.09
Glaucoma 0.01 4 0.08 1.00 0.22 0.88 0.14 0.12
VTE 0.01 4 0.03 1.00 0.69 0.22 0.66 0.78

Abbreviations: AAA, abdominal aortic aneurysm; HF, heart failure; T2DM, type 2 diabetes mellitus; PAD, peripheral arterial disease; VTE, venous thromboembolism.