Table 2.
Performance of automated heart failure detection algorithms versus reference standard
| Algorithm | N | Ref. HF+ n = 222 | Ref. HF− n = 820 | Precision (%) | Sensitivity (%) | F1 score (%) | |
|---|---|---|---|---|---|---|---|
| M ICD | HF+ | 117 | 110 (TP) | 7 (FP) | 94 | 50 | 65 |
| (for comparison) | HF− | 925 | 112 (FN) | 813 (TN) | |||
| M Expert | HF+ | 253 | 193 (TP) | 60 (FP) | 76 | 87 | 81 |
| (expert specified) | HF− | 789 | 29 (FN) | 760 (TN) | |||
| A Precision | HF+ | 140 | 134 (TP) | 6 (FP) | 96 | 60 | 74 |
| (precision optimized) | HF− | 902 | 88 (FN) | 814 (TN) | |||
| A Sensitivity | HF+ | 286 | 204 (TP) | 82 (FP) | 71 | 92 | 80 |
| (sensitivity optimized) | HF− | 756 | 18 (FN) | 738 (TN) | |||
| A F1 | HF+ | 209 | 186 (TP) | 23 (FP) | 89 | 84 | 86 |
| (F1 score optimized) | HF+ | 833 | 36 (FN) | 797 (TN) |
Ref reference standard defined by a heart failure specialist inspecting the documents, HF+ heart failure present, HF− heart failure absent, TP true positive, FP false positive, FN false negative, TN true negative. Precision: TP/(TP + FP), sensitivity: TP/(TP + FN), F1: 2 × (precision × sensitivity)/(precision + sensitivity). For details refer to “Methods”