Table 1 The study group’s manual evaluation of the algorithm (1 = disagree, 2 = unclear, 3 = agree)
From: Using machine learning to predict subsequent events after EMS non-conveyance decisions
|  | I agree with algorithm | The key words are relevant | The text gives clues of the algorithm’s result |
|---|---|---|---|
The patient had subsequent event and the model predicted there will be one (n = 17) | 1 = 0% (n = 0) 2 = 23.5% (n = 4) 3 = 76.5% (n = 13) | 1 = 41.2% (n = 7) 2 = 17.7% (n = 3) 3 = 41.2% (n = 7) | 1 = 0% (n = 0) 2 = 29.4% (n = 5) 3 = 70.6% (n = 12) |
The patient had subsequent event, but the model did not predict one (n = 20) | 1 = 55.0% (n = 11) 2 = 15.0% (n = 3) 3 = 30.0% (n = 6) | 1 = 90.0% (n = 18) 2 = 10.0% (n = 2) 3 = 0% (n = 0) | 1 = 40.0% (n = 8) 2 = 30.0% (n = 6) 3 = 30.0% (n = 6) |
The patient didn’t have subsequent event, but the model predicted that there will be one (n = 20) | 1 = 20.0% (n = 4) 2 = 20.0% (n = 4) 3 = 60.0% (n = 12) | 1 = 50.0% (n = 10) 2 = 25.0% (n = 5) 3 = 25.0% (n = 5) | 1 = 10.0% (n = 2) 2 = 25.0% (n = 5) 3 = 65.0% (n = 13) |
The patient didn’t have subsequent event and the model did not predict one (n = 20) | 1 = 15.0% (n = 3) 2 = 40.0% (n = 8) 3 = 45.0% (n = 9) | 1 = 95.0% (n = 19) 2 = 5.0% (n = 1) 3 = 0% (n = 0) | 1 = 5.0% (n = 1) 2 = 35.0% (n = 7) 3 = 60.0% (n = 12) |