Table 3. Patient disease tagging results for diagnoses assigned during different time intervals in terms of precision-at-k, with k = 1, 3, 5; UppBnd shows the best results achievable (i.e., all the correct diagnoses assigned to all the patients).
Time Interval | Metrics | UppBnd | Patient Representation |
|||
---|---|---|---|---|---|---|
RawFeat | PCA | ICA | DeepPatient | |||
30 days (16,374 patients) | Prec@1 | 1.000 | 0.319 | 0.343 | 0.345 | 0.392* |
Prec@3 | 0.492 | 0.217 | 0.251 | 0.255 | 0.277* | |
Prec@5 | 0.319 | 0.191 | 0.214 | 0.215 | 0.226* | |
60 days (21,924 patients) | Prec@1 | 1.000 | 0.329 | 0.349 | 0.353 | 0.402* |
Prec@3 | 0.511 | 0.221 | 0.254 | 0.259 | 0.282* | |
Prec@5 | 0.335 | 0.199 | 0.216 | 0.219 | 0.230* | |
90 days (25,220 patients) | Prec@1 | 1.000 | 0.332 | 0.353 | 0.360 | 0.404* |
Prec@3 | 0.521 | 0.243 | 0.257 | 0.262 | 0.285* | |
Prec@5 | 0.345 | 0.201 | 0.219 | 0.220 | 0.232* | |
180 days (33,607 patients) | Prec@1 | 1.000 | 0.331 | 0.361 | 0.363 | 0.418* |
Prec@3 | 0.549 | 0.246 | 0.261 | 0.265 | 0.290* | |
Prec@5 | 0.370 | 0.207 | 0.221 | 0.224 | 0.236* |
(*) The difference with the corresponding second best measurement is statistically significant (p < 0.05, t-test).