Table 1.
Model characteristics.
| Overall | Statistical model | Machine learning model | |||||
|---|---|---|---|---|---|---|---|
| 280 models | Overall 202 models | Mortality 158 models | Readmission 44 models | Overall 78 models | Mortality 47 models | Readmission 31 models | |
| HF type | |||||||
| Acute HF | 73 (26%) | 66 (33%) | 62 (39%) | 4 (9%) | 7 (9%) | 6 (13%) | 1 (3%) |
| Chronic HF | 32 (11%) | 32 (16%) | 25 (16%) | 7 (16%) | 0 (0%) | 0 (0%) | 0 (0%) |
| Not specified | 175 (62%) | 104 (51%) | 71 (44%) | 33 (75%) | 71 (91%) | 41 (87%) | 30 (97%) |
| LVEF | |||||||
| HFrEF | 65 (23%) | 62 (31%) | 55 (35%) | 7 (16%) | 3 (4%) | 3 (6%) | 0 (0%) |
| HFpEF | 23 (8%) | 13 (6%) | 11 (7%) | 2 (5%) | 10 (13%) | 10 (21%) | 0 (0%) |
| Not specified | 192 (69%) | 127 (63%) | 92 (58%) | 35 (80%) | 65 (83%) | 34 (72%) | 31 (100%) |
| Admission type | |||||||
| Inpatient | 172 (61%) | 127 (63) | 98 (62%) | 29 (66%) | 45 (58%) | 25 (53%) | 20 (65%) |
| Outpatient | 45 (16%) | 37 (18%) | 32 (20%) | 5 (11%) | 8 (10%) | 8 (17%) | 0 (0%) |
| Other * | 63 (22%) | 38 (19%) | 28 (18%) | 10 (23%) | 25 (32%) | 14 (30%) | 11 (35%) |
| Region | |||||||
| North America | 116 (41%) | 72 (36%) | 47 (30%) | 25 (57%) | 44 (56%) | 27 (57%) | 17 (55%) |
| Europe | 88 (31%) | 78 (39%) | 67 (42%) | 11 (25%) | 10 (13%) | 2 (4%) | 8 (26%) |
| East Asia | 61 (22%) | 40 (20%) | 35 (22%) | 5 (11%) | 21 (27%) | 18 (38%) | 3 (10%) |
| Others | 15 (5%) | 12 (6%) | 9 (6%) | 3 (7%) | 3 (4%) | 0 (0%) | 3 (10%) |
| Algorithm | |||||||
| Cox regression | 64 (23%) | 64 (32%) | 58 (37%) | 6 (14%) | / | / | / |
| LR | 61 (22%) | 61 (30%) | 31 (20%) | 30 (68%) | / | / | / |
| Score | 77 (28%) | 77 (38%) | 69 (44%) | 8 (18%) | / | / | / |
| RF | 11 (4%) | / | / | / | 11 (14%) | 7 (15%) | 4 (13%) |
| Boosting | 17 (6%) | / | / | / | 17 (22%) | 11 (23%) | 6 (19%) |
| SVM | 7 (3%) | / | / | / | 7 (9%) | 5 (11%) | 2 (6%) |
| Neural network ** | / | / | / | ||||
| Multi-layer perceptron | 7 (3%) | / | / | / | 7 (9%) | 5 (11%) | 2 (6%) |
| Deep learning | 8 (3%) | / | / | / | 8 (10%) | 2 (4%) | 6 (19%) |
| Decision tree | 10 (4%) | / | / | / | 10 (13%) | 8 (17%) | 2 (6%) |
| Others | 18 (6%) | / | / | / | 18 (23%) | 9 (19%) | 9 (29%) |
| Year of publication | |||||||
| 2010–2015 | 69 (25%) | 65 (32%) | 50 (32%) | 15 (34%) | 4 (5%) | 2 (4%) | 2 (6%) |
| 2016–2021 | 211 (75%) | 137 (68%) | 108 (68%) | 29 (66%) | 74 (95%) | 45 (96%) | 29 (94%) |
| Study type | |||||||
| Model development | 179 (64%) | 101 (50%) | 71 (45%) | 30 (68%) | 78 (100%) | 47 (100%) | 31 (100%) |
| Model validation | 101 (36%) | 101 (50%) | 87 (55%) | 14 (32%) | 0 (0%) | 0 (0%) | 0 (0%) |
Values are presented as numbers.
*
“Others” indicates that studies did not specify the origin of patients, or patients have mixed origins.
**
The deep learning model refers to recently proposed neural network-based models (e.g., recurrent neural nets and autoencoder) apart from simple multi-layer perceptron.
HF, heart failure; HFpEF, heart failure with preserved ejection fraction;HFrEF, heart failure with reduced ejection fraction; LVEF. left ventricular ejection fraction.