Table 4.
Univariate and multivariable logistic regression of factors associated with weaning failure (n = 246)
| Variables | Missing values, n (%) | Absolute standardized differences | Odd ratio (95% confidence interval), p value by logistic regression | |
|---|---|---|---|---|
| Univariate | Multivariable | |||
| Age (per year) | 0 | 47 | 1.03 (1.01–1.05), p = 0.01 | 1.02 (0.997–1.05), p = 0.08 |
| Body mass index (per kg/m2) | 6 (2%) | 32 | 1.06 (1.01–1.11), p = 0.02 | I/NR |
| COPD (yes vs. no) | 0 | 48 | 3.0 (1.6–5.8), p = 0.001 | 2.2 (1.02–4.7), p = 0.045 |
| Cardiac disease (yes vs. no) | 0 | 37 | 2.2 (1.2–4.0), p = 0.01 | I/NR |
| Left ventricle ejection fraction at cardiac ultrasound (%) | 44 (18%) | 27 | 0.98 (0.96–1.0), p = 0.09 | NI |
| Supra–ventricular arrhythmias (yes vs. no) | 0 | 26 | 1.9 (1.01–3.6), p = 0.046 | NI |
| Septic shock (yes vs. no) | 0 | 37 | 2.1 (1.2–3.7), p = 0.01 | I/NR |
| Fluid balance between ICU admission and first SBT (per L) | 15 (6%) | 44 | 1.07 (1.03–1.12), p = 0.002 | NI |
| Acute respiratory failure as cause of intubation (yes vs. no) | 0 | 55 | 3.0 (1.7–5.2), p < 0.001 | NI |
| PaO2/FiO2 ratio (per mmHg) | 3 (1%) | 58 | 0.994 (0.991–0.997), p < 0.001 | 0.996 (0.993–1.0), p = 0.03 |
| Duration of MV before the first SBT (per day) | 0 | 57 | 1.11 (1.06–1.17), p < 0.001 | 1.11 (1.05–1.17), p < 0.001 |
| ARDS before the first SBT (yes vs. no) | 0 | 49 | 3.0 (1.6–5.7), p < 0.001 | NI |
| Neuromuscular blockade before the first SBT (yes vs. no) | 0 | 54 | 3.5 (1.9–6.6), p < 0.001 | NI |
| VAP before the first SBT (yes vs. no) | 0 | 33 | 2.6 (1.2–5.4), p = 0.01 | NI |
| Moderate-to-large pleural effusion (yes vs. no) | 0 | 58 | 3.2 (1.8–5.7), p < 0.001 | 3.0 (1.5–5.8), p = 0.001 |
SAPS II simplified acute physiology score, COPD chronic obstructive pulmonary disease, ARDS acute respiratory distress syndrome, VAP ventilator-associated pneumonia, SBT spontaneous breathing trial, NI not included, I/NR included, but not retained by the final model
Among related univariate factors, only the most statistically robust (yet clinically relevant) was entered into the regression model in order to minimize the effect of colinearity. The selection process was guided by consistency (less than 5% missing values) and maximal imbalances between groups (as estimated by absolute standardized differences), as follows: cardiac disease was selected among supra-ventricular arrhythmias, left ventricle ejection fraction and cardiac disease; septic shock was selected among fluid balance between ICU admission and first SBT and septic shock; PaO2/FiO2 ratio was selected among acute respiratory failure as cause of intubation and PaO2/FiO2 ratio; duration of MV before the first SBT was selected among neuromuscular blockade, duration of MV before the first SBT, VAP, and ARDS. The multivariable model showed a good calibration as assessed by the Hosmer and Lemeshow goodness-of-fit test [χ2 (8 df) = 6.8, p = 0.56] and a fair discrimination as assessed by the receiver operating characteristics curve [area under the curve of 0.76 (0.69–0.82), p < 0.001]