Predicting weaning failure from invasive mechanical ventilation: The promise and pitfalls of clinical prediction scores

Table 1 Summary of the various prediction scores to predict weaning failure

Scoring method	Interpretation	Uses	Limitations
1 Morganroth scale	Score of < 55 predicts successful weaning	Applied to patients requiring short-term and long-term mechanical ventilation	Limited data
2 RSBI	Score of < 105 predicts successful weaning	Easy to calculate	Can be confounded by multiple patient factors
3 CROP index	Score of > 13 mL/breath/min predicts successful weaning	Includes respiratory and ventilator parameters	Cannot be used in neuro patients
4 Gluck and Corgian score	Score of < 3 predicts successful weaning	Simple bedside measurement	Not validated in large studies
5 BWAP	Score of > 50 predicts successful weaning	Comprehensive weaning checklist	Limited data
6 Modified Burns Wean Assessment Program	Score of > 60 predicts successful weaning	Useful for long-term mechanical ventilation	Limited data
7 Persian weaning tool	Score of > 50 suggests readiness to wean	Similar to BWAP	Limited data
8 HACOR scoring	Score of > 5 predicts weaning failure	Easy bedside tool	Limited data
9 WEANS NOW	Score of 1 or more predicts weaning failure	Multiple parameters included	Complex
10 ExPreS	Score of > 59 has a high probability of extubation success	Simple tool, shown to reduce extubation failure rates	Not validated in large studies

RSBI: Rapid shallow breathing index; BWAP: Burns Wean Assessment Program.

Table 2 Summary of the various artificial intelligence models to predict weaning failure

Ref.	Sample size	Performance metrics	Technique used	Characteristics involved	Primary outcome
Hsieh et al[51]	3602 patients	AUC-0.85, Accuracy-94%, Precision-0.939, F1-0.867, Recall-0.822	K-fold cross-validation	Age, gender, cause of intubation, MAPs, MIP, APACHE II scores, GCS	Successful extubation from MV
Huang et al[52]	233 patients	AUC-0.97, Accuracy-94%, F1 score-95.8%, Sensitivity-87.5%, Specificity-96.7%	Logistic regression, Random Forest, and support vector machine model	Gender, APACHE II score, hospital stay duration in days, MV duration in days	Successful extubation
Lin et al[53]	Part 1: 2405; Part 2: 131	Try weaning phase: AUC-0.860, Accuracy-0.768, Sensitivity-0.788, Specificity-0.733; Extubation phase: AUC-0.923, Accuracy-0.842, Sensitivity-0.842, Specificity-0.842	Logistic regression, RF, SVM, KNN, Light GBM, MLP, XGBoost	Age, APACHE II score, TISS score, FiO2, PEEP, RR, MV, Ppeak, SpO2, HR, BP	Successful extubation, MV time, ICU LOS, Hospital LOS
Liu et al[54]	Stage 1: 5873 patients; Stage 2: 4172 patients	Stage 1: AUC-0.860, Accuracy-0.768, Sensitivity-0.788, Specificity-0.733; Stage 2: AUC-0.923, Accuracy-0.842, Sensitivity-0.842, Specificity-0.842	Logistic regression, RF, SVM, KNN, Light GBM, MLP, XGBoost	25 features in stage 1, 20 features in stage 2-common variables-APACHE II scores, TISS score, FiO2, PEEP, MV, Ppeak, SpO2, HR, BP	Weaning of MV patients
Xu et al[55]	487 patients	AUC-0.805, Accuracy-0.748, Sensitivity-0.767, Specificity-0.676, Recall-0.888	Logistic regression, RF, SVM, Light GBM, XGBoost	RR, SBT, APACHE II score, GCS, Hb	Weaning of MV patients

AUC: Area under curve; BP: Blood pressure; FiO₂: Fraction of inspired oxygen; GCS: Glasgow coma scale; HR: Heart rate; LOS: Length of stay; MAP: Mean airway pressure; MIP: Maximal inspiratory pressure; MV: Mechanical ventilation; PEEP: Positive end-expiratory pressure; Ppeak: Peak airway pressure; RR: Respiratory RATE; SPO₂: Peripheral oxygen saturation; TISS: Therapeutic intervention scoring system.