Table 5.
Trends of model performance with increasing feature sets.
| Variables included in the model | Number of features | AUCa | ACCb |
| SBPc_last, Age, Gender, Years_After_Hypertension (in the last CHDd-free record) | 4 | 0.7547 | 0.6941 |
| + Diabetes diagnosis | 5 | 0.7766 | 0.7090 |
| + Hyperlipidemia diagnosis | 6 | 0.8111 | 0.7339 |
| +Inpatient diagnosis flag +Total in-hospital days + Total in-hospital visit number |
9 | 0.8134 | 0.7341 |
| + Diagnosed symptoms (eg, hypertension level, cerebral disease, dizziness, nephropathy, gout, hyperuricemia, palpitation) |
19 | 0.8289 | 0.7460 |
| + Multipoint SBP statistics (SBP_max, SBP_min, SBP_mean) |
22 | 0.8589 | 0.7766 |
| + Dynamic SBP trends (SBP_ min(max.mean)_1st(2nd)_half) |
28 | 0.8752 | 0.7929 |
| + Medical activities trends (N_visits_1st_half, N_visits_2nd_half, Visit_trend_ratio) |
31 | 0.9195 | 0.8350 |
| + Medical activities trends (N_visits_last_3me, N_visits_last_6mf) |
33 | 0.9427 | 0.8686 |
aAUC: area under the receiver operating characteristic curve.
bACC: accuracy.
cSBP: systolic blood pressure.
dCHD: coronary heart disease.
e3m: 3 months.
f6m: 6 months.