Machine Learning Methods for Predicting Success in Spontaneous Breathing Trial
| Model | Study Variable | Accuracy | Sensitivity | Specificity | PPV | NVP |
|---|---|---|---|---|---|---|
| k-means | SBT* training | 64,0 | 72,6 | 31,5 | 79,9 | 23,5 |
| SBT * test | 63,0 | 72,0 | 35,7 | 72,7 | 38,5 | |
| Hierarchical Clustering | SBT * training | 52,7 | 53,3 | 50,7 | 80,2 | 22,4 |
| SBT * test | 60,9 | 54,9 | 64,3 | 79,2 | 40,9 | |
| Decision Trees | SBT * training | 77,3 | 99,9 | 1,1 | NI | 1,0 |
| SBT * test | 69,6 | 99,9 | 1,0 | NI | 1,0 | |
| Support Vector Machines | SBT * training | 77,3 | 99,9 | 1,1 | NI | 1,0 |
| SBT * test | 69,6 | 99,9 | 1,1 | NI | 1,0 | |
| Neural Networks | SBT * training | 77,3 | 99,9 | 1,0 | NI | 1,0 |
| SBT * test | 69,6 | 99,9 | 1,0 | NI | 1,0 | |
| Qualification | Description |
|---|---|
| 0 | No cough |
| 1 | Audible movement of air through the endotracheal tube, but no audible cough |
| 2 | Strong cough with movement of secretions into the endotracheal tube |
| 3 | Strong cough with movement of secretions out (expulsion) of the endotracheal tube |
Etiology of Respiratory Failure and Reason for Admission to Intensive Care
| Variables | Values |
|---|---|
| Shock, n(%) | 52 (14,9) |
| Hypercapnia (pH < 7,25, CO2 elevated), n(%) | 23 (6,6) |
| Hypoxemia (PaO2 < 60, usual FiO2), n(%) | 261 (75) |
| Neuromuscular, n(%) | 2 (0,6) |
| Perioperative, n(%) | 10 (2,9) |
| Reason for ICU Admission, n (%) | |
| Medical | 345 (94) |
| Surgical (post-surgical only) | 22 (6) |
General Characteristics of the Population_
| Variables n (%) | Values |
|---|---|
| Male n (%) | 219 (59,7) |
| Age, median (Range) | 61 (18 – 88) |
| Weight in kg, median (IQR) | 70 (60 – 80) |
| Height in cm, mean (SD) | 163,6 (10) |
| Body Mass Index (BMI) in kg/m2, | |
| median (IQR) | 25,3 (21,7 – 29,1) |
| Active smoking, n (%) | 33 (9) |
| Alcoholism n (%) | 22 (6) |
| Comorbidities, n (%) | |
| Diabetes Mellitus | 113 (30,8) |
| Hypertension | 173 (47,1) |
| Asthma | 8 (2,2) |
| Pulmonary Fibrosis | 6 (1,6) |
| Chronic Kidney Disease | 69 (18,8) |
| Chronic Liver Disease | 17 (4,6) |
Machine Learning Methods for Predicting Extubation Success
| Model | Study Variable | Accuracy | Sensitivity | Specificity | PPV | NVP |
|---|---|---|---|---|---|---|
| k-means | SBT* training | 63,4 | 74,4 | 35,1 | 74,7 | 34,7 |
| SBT * test | 63,0 | 76,7 | 37,5 | 69,8 | 46,2 | |
| Hierarchical Clustering | SBT * training | 66,4 | 91,6 | 8,0 | 69,7 | 31,4 |
| SBT * test | 65,2 | 90,0 | 18,8 | 67,5 | 50 | |
| Decision Trees | SBT * training | 89,8 | 98,3 | 70,4 | 94,6 | 68,7 |
| SBT * test | 95,7 | 99,9 | 87,5 | 99,9 | 68,7 | |
| Support Vector Machines | SBT * training | 85,9 | 99,0 | 56,0 | 95,9 | 55 |
| SBT * test | 93,5 | 99,9 | 81,3 | 99,9 | 81,3 | |
| Neural Networks | SBT * training | 85,9 | 99,0 | 56,0 | 95,9 | 55 |
| SBT * test | 93,5 | 99,9 | 81,3 | 99,9 | 81,3 | |