Table 4.
The application of AI in prediction of events
| Study | Aim | AI method | n | Accuracy results | Conclusions |
|---|---|---|---|---|---|
| Prediction of Hypotension | |||||
| Gratz 2020 [45] | Predict the likelihood of post spinal hypotension from arterial stiffness | ANN | 45 | AUC, 0.89; Sn, 0.84; Sp, 0.91 | This study demonstrated that arterial stiffness variability is an effective predictor of postinduction hypotension. |
| Lin 2008 [46] | Identify patients with high risk of hypotension during spinal anesthesia | ANN | 375 | AUC, 0.796; Sn, 75.9%; Sp, 76.0% | The model should be useful in increasing vigilance in those patients most at risk for hypotension during spinal anesthesia. |
| Kang 2020 [7] | Predict postinduction hypotension from intraoperative data | Random Forest; ANN | 222 | AUC of Random Forest model 0.842; Accuracy 76.28%; | Models can predict hypotension occurring during the period between tracheal intubation and incision. |
| Kendale 2018 [47] | Prediction for the risk of postinduction hypotension | Gradient boosting machine | 13.323 | AUC 0.74 | The model can forecast postinduction hypotension, with performance dependent on model choice and proper tuning. |
| Lin 2011 [48] | Identify patients at high risk for postinduction hypotension | ANN | 294 | Accuracy 82.3%; AUC 0.893; Sn 76.4%; Sp 85.6%; | The model has good discrimination of risk of postinduction hypotension. |
| Wijnberge 2020 [24] | Early warning system of hypotension during noncardiac surgery. | Machine Learning | 68 | Median time of hypotension 8.0 min intervention group vs. 32.7 in control group | The use of AI early warning system compared with standard care resulted in less intraoperative hypotension. |
| Prediction of Hypoxemia | |||||
| Geng 2019 [49] | Prediction of hypoxemia during sedation for gastrointestinal endoscopy | ANN | 220 | Accuracy 90%; AUC 0.80; Sn 14%; Sp 98% | The model was useful for prediction of hypoxemia. |
| Lundberg 2018 [26] | Predic the risk of hypoxemia and provides explanations of the risk factors. | Machine Learning | 53.126 | For initial prediction, AUC 0.83; For real-time prediciton AUC 0.81 | The system can help improve the clinical understanding of hypoxemia risk during anaesthesia care. |
| Sippl 2017 [50] | Model perioperative hypoxia | ANN | 124 | Sn 74%; Sp 93% | The model is able to classify oxygen desaturation on a level similiar to the mutual agreement between human experts. |
| Prediction of different events | |||||
| Huang 2022 [51] | Prediction of surgery and anesthesia emergence duration | ANN | 4.285 | Accuracy, 0.9552 | Prediction accuracies of the proposed serial prediction systems are acceptable in comparison to separate systems. |
| Huang 2003 [52] | Predict response during isoflurane anaesthesia from time series of EEGs | ANN | 98 | Accuracy 91.84% | The technique outperforms competing techniques, is computationally fast, and offers acceptable real-time clinical performance. |
| Knorr 2006 [53] | Distinguish between normal breathing and obstructed airway events. | ANN | 10 | Accuracy 86.1%; Sn 72.9%; Sp 93.0% | The model has potential to distinguishing between normal and obstructed airway events. |
| Mansoor Baig 2013 [54] | Detection of absolute hypovolaemia | Fuzzy Logic | 20 | Kappa value of the best FL model, 0.75 | FLMS-2 model has shown to accurately detect differences between the levels of hypovolaemia |
| Peng 2007 [27] | Predict postoperative nausea and vomiting in patients who received general anaesthesia. | ANN | 430 | Accuracy, 83.3%; AUC, 0.814; Sn, 77.9%; Sp 85.0% | The ANN model appears to be a suitable model for clinicians to use cost-effective antiemetic treatments. |
| Ren 2022 [55] | Predict the amount of blood transfusion during cesarean section. | XGB classifier | 150 | Accuracy 0.953: AUC 0.881 | The XGB model has a strong prediction performance, can offer precise individual predictions for patients, and has a promising future in clinical use. |
| Santanen 2003 [56] | Predict the recovery of a neuromuscular block during general anaesthesia | ANN | 66 | CC 0.91; Mean absolute prediction error 6.75 | Model could predict individual recovery times significantly better than the average-based method. |
| Zhang 2018 [57] | Predicts a patient’s ASA using the patient’s home medications and comorbidities. | RF | 41.932 | AUC 0.884; Cohen’s Kappa 0.456; | RF algorithm can predict ASA with agreement identical to that of anesthesiologists described in literature. |
EEG electroencephalography, ANN artificial neural networks CNN convolutional neural network, SVM support vector machine, RF Random Forest, AUC area under curve, Sn sensitivity, Sp specificity, CC correlation coefficient