Table 3. The measurement of our method against 4 baseline methods and their data enhanced version for prediagnosis disease prediction.

Methods	AC	MA precision	WA precision	MA recall	WA recall	MA F1-score	WA F1-score
LR	0.52	0.35	0.51	0.18	0.52	0.21	0.48
LR + data enhancement	0.47	0.34	0.48	0.31	0.47	0.31	0.47
GBDT	0.54	0.38	0.53	0.18	0.54	0.21	0.5
GBDT + data enhancement	0.54	0.44	0.54	0.32	0.53	0.34	0.53
HAN	0.62	0.51	0.61	0.42	0.62	0.45	0.61
HAN + data enhancement	0.63	0.54	0.62	0.42	0.63	0.46	0.62
BERT	0.64	0.54	0.61	0.44	0.65	0.48	0.65
BERT + data enhancement	0.64	0.55	0.62	0.44	0.67	0.47	0.65
Ours (MSCNN)	0.68	0.56	0.67	0.45	0.68	0.5	0.67
Ours (MSCNN) + data enhancement	0.68	0.59	0.67	0.49	0.68	0.51	0.67

Metrics included AC, MA precision, WA precision, MA recall, WA recall, MA F1-score, and WA F1-score. AC, accuracy; MA, macro average; WA, weighted average. Methods include: LR, logistic regression; GBDT, gradient-boosted decision tree; HAN, hierarchical attention networks; BERT, bidirectional encoder representations from transformers; MSCNN, medical-semantic-aware convolution neural network.