Novel mechanical ventilator weaning predictive model

Abstract

Mechanical ventilation (MV) is a common life support system in intensive care units. Accurate identification of patients who are capable of being extubated can shorten the MV duration and potentially reduce MV‐related complications. Therefore, prediction of patients who can successfully be weaned from the mechanical ventilator is an important issue. The electronic medical record system (EMRs) has been applied and developed in respiratory therapy in recent years. It can increase the quality of critical care. However, there is no perfect index available that can be used to determine successful MV weaning. Our purpose was to establish a novel model that can predict successful weaning from MV. Patients' information was collected from the Kaohsiung Medical University Hospital respiratory therapy EMRs. In this retrospective study, we collected basic information, classic weaning index, and respiratory parameters during spontaneous breathing trials of patients eligible for extubation. According to the results of extubation, patients were divided into successful extubation and extubation failure groups. This retrospective cohort study included 169 patients. Statistical analysis revealed successful extubation predictors, including sex; height; oxygen saturation; Glasgow Coma Scale; Acute Physiology and Chronic Health Evaluation II score; pulmonary disease history; and the first, 30th, 60th, and 90th minute respiratory parameters. We built a predictive model based on these predictors. The area under the curve of this model was 0.889. We established a model for predicting the successful extubation. This model was novel to combine with serial weaning parameters and thus can help intensivists to make extubation decisions easily.

1. INTRODUCTION

Mechanical ventilation (MV) is a common life support system used in the intensive care unit (ICU). MV can provide patients with respiratory failure with appropriate oxygenation and ventilation and give clinicians more time to treat the underlying disease. ¹ Once the cause of respiratory failure is corrected, the most important issue is weaning the patient off MV. Inappropriate early weaning from MV may increase the load of the respiratory and cardiac systems. By contrast, delay in weaning from MV may cause diaphragmatic atrophy and weakness. ² Therefore, accurate identification of patients who are capable of spontaneous breathing can shorten MV duration and potentially reduce MV‐related complications.

Several clinical weaning indices are used, such as the rapid shallow breathing index (RSBI), maximal inspiratory pressure (Pimax), airway occlusion pressure in the first 100 ms (P0.1), and P0.1/Pimax. The most commonly used weaning index is the RSBI. ³ However, no perfect index is available that can be used to determine weaning success. The respiratory system is affected by many systems, such as hemodynamic, neurological, or exchange gas systems. Current evidence has shown that 15% to 20% of patients experience extubation failure and required intubation within 48 to 72 hours. Reintubation increases the risk of pneumonia, in addition to increasing medical cost and mortality. Extubation failure is associated with extremely poor outcomes, including mortality rates of 25% to 50%. ⁴ , ⁵ Therefore, how to predict which patients can successfully be weaned from a mechanical ventilator is an important issue.

With the advancement of information technology, electronic medical records system (EMRs) has developed rapidly in recent years. The EMRs can be used to record, manage, and monitor patients'profiles. ⁶ The EMRs has been applied and developed in respiratory therapy recently. Some studies have confirmed that respiratory therapy EMRs can reduce typing errors and improve the quality of patient care and patient safety. ⁷

The purpose of this research was to explore the predictors and establish a model that can predict successful weaning from MV using EMRs.

2. METHODS

We conducted a retrospective study of patients at Kaohsiung Medical University Hospital (KMUH) in Taiwan. The study was approved by the KMUH Institutional Review Board (IRB, KMUHIRB‐E(I)‐20180293).

Patients' information, from 1 June 2017 to 31 March 2018, was collected from the KMUH respiratory therapy EMRs. This retrospective study collected basic information of patients eligible for extubation. The information gathered included age, gender, past history, Acute Physiology and Chronic Health Evaluation II (APACHE II) score, reasons for using MV, Glasgow Coma Scale (GCS) upon entering the ICU and before spontaneous breathing trial (SBT), blood pressure, heart rate (HR), oxygen saturation measured by pulse oximetry (SpO₂), classic weaning index (eg, RSBI, Pimax), and respiratory parameters during SBT (eg, respiratory rate [RR], tidal volume [V_T], minute ventilation [V_E], peak inspiratory pressure, and mean airway pressure).

The inclusion criteria were as follows:

1. Adult (≥20 years).

2. MV time > 48 hours.

3. Endotracheal tube size of 7.0 to 8.0 mm.

4. First episode of using mechanical ventilator.

5. Clinically stable.

6. Met the guidelines for starting a weaning trial.

Exclusion criteria were as follows:

1. Tracheal stenosis.

2. Postoperative patients.

3. Frequent cardiac arrhythmias.

4. Intracranial pressure of >20 mm Hg.

5. Severe brain disease.

6. Sedative drug use.

7. Hemodynamic instability.

8. Poor control of systemic infection.

9. Underwent tracheostomy.

We screened the information of patients who completed an SBT and underwent extubation. After screening, we collected the eligible patients' information regarding SBT (low pressure‐support ventilation) from the KMUH respiratory therapy EMRs. During SBT, patients were ventilated with pressure support of 6 cm H₂O with 5 cm H₂O of positive end‐expiratory pressure and the fraction of inspired oxygen (FiO₂) set as 40%. The SBT protocol is available in Data S1. Decision of extubation was based on whether patients had passed the SBT trial. The criteria of SBT failure is in Data S1. According to the extubation results, patients were divided into the successful extubation group and the extubation failure group. Extubation failure was defined as extubation followed by resumption of mechanical support or respiratory failure occurring within 48 hours. The flowchart of the study designed is shown in Figure 1.

FIGURE 1.

Flowchart of the study design.

Statistical analysis was performed using the IBM SPSS Statistics for Windows version 20 (IBM, Armonk, New York) and JMP version 13 (SAS Institute, Cary, North Carolina) software programs. Nominal data were compared using a chi‐square test or Fisher exact test as appropriate. Categorical variables were expressed as frequencies. Participant characteristics were compared for both groups using Wilcoxon‐Mann‐Whitney test. Continuous variables were expressed as mean ± SD. A P value of <.05 was considered significant in two‐tailed tests.

Multiple logistic regression analyses were used to identify the connection between successful extubation and related parameters, as shown in Table 3. We converted categorical variables to dummy variables into the multiple logistic regression and used stepwise selection method to choose the predictive variables into prediction model.

TABLE 3.

Multivariate analysis for extubation success

	Model 1				Model 2				Model 3				Model 4				Model 5
Variable	OR	95% CI		P value	OR	95% CI		P value	OR	95% CI		P value	OR	95% CI		P value	OR	95% CI		P value
		Lower	Upper			Lower	Upper			Lower	Upper			Lower	Upper			Lower	Upper
Sex	0.48	0.17	1.34	.159	0.50	0.17	1.41	.189	0.43	0.14	1.27	.125	0.27	0.08	0.94	.039*	0.17	0.04	0.72	.016*
Height	1.06	1.00	1.12	.044*	1.05	0.98	1.11	.150	1.07	1.00	1.14	.056	1.05	0.98	1.13	.161	1.08	1.00	1.17	.056
SpO2 before SBT	1.32	1.06	1.66	.014*	1.33	1.06	1.68	.014*	1.37	1.08	1.74	.009*	1.39	1.07	1.80	.012*	1.60	1.17	2.20	.003*
GCS before SBT	1.12	0.87	1.43	.371	1.11	0.86	1.45	.421	1.12	0.85	1.49	.417	1.08	0.79	1.47	.633	1.27	0.88	1.84	.202
APACHE II score	0.96	0.91	1.01	.099	0.95	0.90	1.01	.093	0.95	0.89	1.00	.065	0.94	0.88	1.00	.061	0.93	0.87	1.00	.067
Pulmonary disease	0.79	0.29	2.18	.652	0.75	0.27	2.12	.589	0.76	0.26	2.18	.605	0.67	0.21	2.13	.501	0.43	0.12	1.58	.204
m1VE					1.06	0.82	1.39	.640	1.08	0.83	1.41	.567	1.15	0.81	1.64	.422	0.94	0.64	1.38	.736
m1VT					1.00	1.00	1.01	.252	1.00	1.00	1.01	.078	1.00	1.00	1.01	.098	1.01	1.00	1.02	.007*
m1RR					0.97	0.88	1.06	.502	0.94	0.85	1.05	.266	0.94	0.83	1.06	.303	0.92	0.81	1.05	.224
m30VE									0.79	0.60	1.06	.114	0.66	0.46	0.93	.019*	0.61	0.39	0.95	.028*
m30VT									1.00	1.00	1.00	.989	1.00	0.99	1.00	.960	1.00	1.00	1.01	.475
m30RR									1.12	0.96	1.30	.144	1.27	1.04	1.56	.018*	1.19	0.93	1.51	.164
m60VE													1.13	0.85	1.52	.405	1.16	0.83	1.61	.388
m60VT													1.00	1.00	1.01	.849	1.00	1.00	1.01	.772
m60RR													0.86	0.76	0.96	.009*	0.80	0.68	0.93	.003*
m90VE																	0.73	0.46	1.18	.201
m90VT																	1.00	1.00	1.01	.86 4
m90RR																	1.41	1.07	1.87	.016*
AUC				.737				.748				.772				.820				.889

Note: Mean ± SD.

Abbreviations: AUC, area under curve; GCS, Glasgow Coma Scale; m1V_E, m30V_E,m60V_E, and m90V_E, the first minute,30th minute,60th minute, and 90th minute ventilation during spontaneous breathing trial; m1V_T, m30V_T,m60V_T, and m90V_T, the first minute,30th minute,60th minute, and 90th minute tidal volume during spontaneous breathing trial; m1 RR, m30 RR,m60 RR, and m90 RR, the first minute,30th minute,60th minute, and 90th minute respiratory rate during spontaneous breathing trial; OR, odds ratio; SBT, spontaneous breathing trial.

^* P < .05

After try and error, we found predictors of successful extubation to be gender; height; SpO₂; GCS; APACHE II score; and pulmonary disease history. According to the studies by Liu et al, observing dynamic changes in ventilation and physiological parameters during SBT may improve the prediction of weaning success. ⁸ Therefore, we combined the predictors and ventilation parameters to establish prediction models.

The performance of the models in predicting weaning outcome was evaluated through the sensitivity, specificity, predictive values, likelihood ratios, and areas under the curve (AUC) of receiver operating characteristic (see Table 4). The cutoff point for the models was calculated using the Youden index. Through combining analytical results, the prediction of successful extubation can be obtained as the logistic regression equation, as shown in Data S2.

TABLE 4.

Quality indicators of the model to predict weaning outcome

	Cutoff point	Sensitivity	Specificity	PPV	NPV	LR+	LR−	AUC
Model 1	73.48	89	50	0.90	0.48	1.88	0.26	0.737
Model 2	84.38	72	68	0.92	0.33	2.79	0.12	0.748
Model 3	90.64	47	93	0.97	0.26	3.73	0.07	0.772
Model 4	77.42	81	71	0.93	0.44	2.11	0.21	0.820
Model 5	85.00	73	96	0.99	0.44	2.25	0.20	0.889

Abbreviations: AUC, area under curve; LR⁺, likelihood ratio of positive test; LR⁻, likelihood ratio of negative test; PPV, positive predictive value; NPV, negative predictive value.

Logistic regression equation:

$$\mathit{\text{Prob}}=\left[\exp \left(\text{β0}+\text{β1}\times \text{x1}+\dots +\mathrm{\beta k}\times \mathrm{xk}\right)\right]/\left[1+\exp \left(\text{β0}+\text{β1}\times \text{x1}+\dots +\mathrm{\beta k}\times \mathrm{xk}\right)\right]$$

3. RESULTS

The present retrospective study included 169 patients, 28 (16.6%) of whom experienced extubation failure. One hundred and four patients were men (61.5%) and 65 were women (38.5%). Other demographic results are listed in Table 1. In the successful extubation group (141 patients), 92 patients were men (65.2%) and 49 were women (34.8%). These patients had pulmonary disease (20.6%), cardiovascular disease (58.2%), and neuromuscular disease (17.7%). The mean age was 67.8 ± 16.3 years old. The mean APACHE II score was 21.4 ± 8.2 and the mean body mass index (BMI) was 23.5 ± 5.6 kg/m². The mean GCS was 7.4 ± 3.1 upon entering the ICU and10.0 ± 1.6 before the SBT.

TABLE 1.

Characteristics of the 169 patients

Clinical characteristics
Age, years old	68.4 ± 16.0
Gender, n(men/female)	104/65
Medical history
Pulmonary disease (%)	38 (22.5)
Cardiovascular disease (%)	101 (59.8)
Neuromuscular disease (%)	26 (15.4)
APACHE II score	21.9 ± 8.1
Endotracheal tube diameter	7.4 ± 0.19
7.0# (%)	28 (16.6)
7.5# (%)	141 (83.4)
Height (cm)	162.4 ± 8.6
Body weight (kg)	61.4 ± 14.8
BMI (kg/m2)	23.3 ± 5.3
GCS while entering ICU	7.3 ± 3.1
GCS before SBT	10.0 ± 1.7
Successful, failed extubation, n	141/28
Pimax (cmH2O)	34.7 ± 12.5
Cuff leak test (mL)	245.3 ± 112.5
Total duration of MV(days)	8.8 ± 5.3

Note: Mean ± SD.

Abbreviations: BMI, body mass index; GCS, Glasgow Coma Scale; ICU, intensive care unit; MV, mechanical ventilation; Pimax, peak inspiratory maximum pressure; SBT, spontaneous breathing trial.

The successful extubation group had more men than women, and those in this group were taller and had higher BMI than patients in the extubation failure group. No significant difference in underlying diseases was observed between two groups, except for cardiac arrest, which was more frequent in the extubation failure group (21.4% vs 5.6%; P < .05).

The weaning parameters and vital signs measured during the SBT are shown in Table 2. The successful extubation group had, lower HR after SBT (91.5 ± 15.8 vs 100.6 ± 16.9 bpm; P < .05) and higher SpO₂ (98.3 ± 1.9% vs 97.4 ± 2.2%; P < .05).

TABLE 2.

Weaning parameters and vital signs comparison

Variable	Failed extubation (N = 28)	Successful extubation (N = 141)	P value
Gender (male)	12 (42.8)	92 (65.2)	.027*
Age (y)	71.2 ± 13.7	67.8 ± 16.3	.338
APACHE II score	24.4 ± 7.3	21.4 ± 8.2	.069
Height(cm)	158.3 ± 8.9	163.2 ± 8.3	.004*
Body weight(kg)	57.6 ± 10.5	62.3 ± 15.3	.138
BMI (kg/m2)	23.1 ± 4.5	23.5 ± 5.6	.763
GCS while entering ICU	7.1 ± 3.5	7.4 ± 3.1	.800
GCS before SBT	9.7 ± 1.9	10.0 ± 1.6	.452
Pimax (cmH2O)	33.5 ± 11.7	34.9 ± 12.6	.669
Cuff leak (mL)	222.9 ± 121.4	248.9 ± 111.2	.145
RSBI	51.89 ± 24.3	43.57 ± 18.7	.082
Vital signs before SBT
Systolic blood pressure (mm Hg)	143.9 ± 28.8	140.1 ± 25.6	.606
Diastolic blood pressure (mm Hg)	72.0 ± 18.7	71.7 ± 16.4	.758
Heart rate (bpm)	94.1 ± 17.1	87.1 ± 15.0	.083
SpO2 (%)	97.4 ± 2.2	98.3 ± 1.9	.047*
Vital signs after SBT
Systolic blood pressure (mm Hg)	146.0 ± 24.8	147.7 ± 24.9	.869
Diastolic blood pressure (mm Hg)	79.8 ± 19.4	74.9 ± 15.0	.216
Heart rate (bpm)	100.6 ± 16.9	91.5 ± 15.8	.005*
SpO2(%)	96.9 ± 3.1	98.0 ± 1.9	.059
Outcomes
Duration of MV (d)	12.0 ± 7.7	8.1 ± 4.5	0.001*
ICU stay (d)	14.7 ± 8.7	10.9 ± 6.5	.022*
Hospital stay (d)	28.9 ± 19.8	25.9 ± 15.5	.597
Death (%)	13 (46.4%)	4 (2.8%)	<.001*

Note: Mean ± SD.

Abbreviations: BMI, body mass index; GCS, Glasgow Coma Scale; ICU, intensive care unit; MV, mechanical ventilation; Pimax, peak inspiratory maximum pressure; RSBI, rapid shallow breathing index; SBT, spontaneous breathing trial; SpO₂, oxygen saturation measured by pulse oximetry.

^* P < .05

In the successful extubation group, the prognosis was significantly better than that of the failure group in terms of duration of MV, length of ICU, and mortality. Four patients in the extubation failure group were reintubated with ventilator use, and two of them received tracheostomy. All four of these patients were discharged. Sixteen patients received noninvasive MV, 11 of whom were discharged and 5 of whom died. The other eight patients did not receive invasive or noninvasive MV and died.

We analyzed the patients' basic data and ventilation parameters through a logistic regression. We found predictors of successful extubation to be sex, height, SpO₂, GCS, APACHE II score, pulmonary disease history, V_E, RR, V_T at the first minute, 30th minute, 60th minute, and 90th minute.

We included the various predictors in the logistic regression model and built predictive models. The case evaluated as positive by the model is successful extubation. For comparison, these estimated are shown in Table 3.

As shown in Table 3, when the predictive factors included only sex, height, SpO₂, GCS, APACHE II score, and pulmonary disease, the AUC was 0.737.

When the first‐minute respiratory parameters (V_E, RR, and V_T) were included, the AUC increased to 0.748. When the 90th‐minute respiratory parameters (V_E, RR, and V_T) were added to the model, the AUC reached 0.889, indicating a better predictive value. The quality indicators of the models are shown in Table 4. Model 5 had the highest AUC value of 0.889. The sensitivity was 73% and the specificity was 96%. The AUC curves of the models are shown in Figure 2. The estimated functions are shown in Data S2.

FIGURE 2.

Receiver operating characteristic (ROC) curves of the model to predict weaning outcome

We validated the model 5 in an independent group with 72 patients (IRB number: KMUHIRB‐E(II)‐20200094). Forty‐six patients were men (63.8%) and 26 were women (36.1%). These patients had pulmonary disease (26.4%), cardiovascular disease (58.3%), and neuromuscular disease (15.2%). The average age was 68.4 ± 15.3 years old. The mean APACHE II score was 22.1 ± 9.2 and the average BMI was 22.9 ± 4.7 kg/m². The mean GCS was 5.6 ± 3.2 upon entering the ICU and 9.9 ± 1.9 before the SBT. Fourteen (19.4%) patients failed extubation. The validation results were that 50 of them were successfully extubated and their model test results were also successful. Thirteen of them whose extubation were failed had the same result in the model test. Eight of them were successfully extubated but the model test results were failed. One of them failed in extubation, but the result of the model test was successful. The results of validation are as follows: sensitivity and specificity is 86% and 92%, respectively. In addition, the positive predictive value is 0.98 and negative predictive value is 0.61. The validation results are similar to our original model's result. Furthermore, we recommend that further studies with larger samples with different kinds of patients are needed to evaluate and validate the model.

4. DISCUSSION

MV is an important life support system for patients with respiratory failure, but its use also induced many complications. Therefore, confirming that patients are suitable for weaning from MV is crucial. In our study, we mainly analyzed ventilator parameters during the SBT and explored those that could accurately predict the success of extubation.

Most conventional weaning indices have high sensitivity but low specificity. The specificity was approximately 40% to 70%. Our models overcame this problem. Our predictive model was based on sex, height, SpO₂, GCS, APACHE II score, history of pulmonary disease, V_T, RR, and V_E. The AUC value was 0.889, sensitivity was 73%, and specificity was 96%. The predictive model we have developed provides more precise prediction of successful extubation than traditional weaning parameters (eg, RSBI).

An SBT is a routine test used to confirm whether patients are suitable for weaning from MV. ⁹ Some studies had shown that patient's breathing pattern during SBT can be used as an indicator to predict weaning success. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ During SBT, patients in whom extubation has failed may have irregular breathing patterns due to insufficient respiratory load over time. ¹¹ Imbalance between the respiratory muscle load and capacity may cause alveolar hypoventilation. It will lead to weaning failure. Furthermore, worsening of pulmonary mechanics may cause further difficulty of the respiratory muscles and can lead to marked alterations of oxygen use by the peripheral tissues. ¹⁴ Therefore, data centered on respiratory physiology alone at a single time point may not be a reliable guide for making extubation decisions. Observing dynamic changes in ventilation and physiological parameters during SBT may improve the prediction of weaning success. ⁸ Therefore, we attempted to incorporate serial ventilation parameters during the SBT to predict weaning success. According to the studies by Kuo et al and Yu, ¹⁵ , ¹⁶ we set the time points of analysis with the interval set as 30 minutes.

Breathing patterns and variability may be affected by different SBT methods. ¹⁰ We used low‐pressure‐support ventilation for the SBT. Previous studies had shown that the T‐piece SBT is more likely to have a negative effect on the adjustment of HR, ¹⁷ whereas low‐pressure‐support ventilation SBT is more effective than T‐tube SBT. ¹⁸ Ouellette et al indicated that low‐pressure‐support ventilation SBT has higher rate of extubation success. ¹⁹ In addition, the EMRs can be used to monitor and record a patient's ventilation parameters in low‐pressure‐support ventilation SBT, and it is more convenient to use than a pneumotachograph used for these measurements in previous studies.

The weaning index most commonly used in clinical settings is the RSBI, with a cutoff value of 105 originally used to predict weaning failure. ²⁰ In our study, all patients had values that met the RSBI cutoff value, but the extubation failure rate was still 16%. This result indicated that this commonly used weaning index is still not perfect. In addition, other studies have shown that RSBI may have different predictive abilities, depending on the method, the classification, and the ethnic group studied. ²¹ , ²² , ²³ , ²⁴

Thille et al indicated that potential risk factors for extubation failure included an age of >65 years, cardiovascular disease, APACHE II score, pneumonia as the reason for intubation, RSBI of >105, GCS <8, excessive endotracheal secretions, cough function absent or weak, partial pressure of carbon dioxide of >44 mm Hg during SBT, and hemoglobin of <10 g/dL. ⁴ In our present study, we chose two of the variables, APACHE II and GCS, to include in estimated functions. The APACHE II scoring system was originally designed for use in the ICU to assess the severity of disease and related morbidity. It can provide a guide for treatment intervention. ²⁵ In the past decade, the APACHE II score has been shown to be a predictor of weaning outcome. Several studies demonstrated that the APACHE II score was statistically higher in extubation failure groups than in the successful extubation groups. ²⁶ , ²⁷ Yu et al previously reported that APACHE II score is an independent factor for predicting extubation failure at a cutoff point of greater than 17.5. ²⁸

Although several studies have found that the GCS cannot be used to predict the failure of extubation, ²⁹ several studies have shown that patients' consciousness level and mental status affect weaning success. ³⁰ , ³¹ , ³² Delirium and anxiety may affect the likelihood of weaning success. ³³ Therefore, adequate assessment and treatment of patients' consciousness level and mental status may contribute to successful weaning from MV.

Esteban et al reported that monitoring SpO₂ did not allow them to predict which patients will extubation failure within 48 hours. ³⁴ But in our study, there was a significant difference between the successful extubation group and the failure group. Therefore, SpO₂ is included in our model.

Previous studies reported reduced lung volumes (total lung capacity and vital capacity), which is associated with both lower body height and female gender. ³⁵ It may explain why gender and height are included in our model by statistical analysis.

According to statistical analysis results, pulmonary disease was included in our model. There were 38 patients had pulmonary disease in our study. The most of the pulmonary diseases were chronic obstructive pulmonary disease (COPD) and pneumonia. Previous study had described that COPD is the most common underlying cause leading into long‐term MV. ³⁶ COPD had several characteristics: primarily bronchial obstruction accompanied by lung hyperinflation, the increased respiratory rates leading to dynamic hyperinflation, and the functional residual capacity moves to the flat upper margin of the pressure‐volume curve. This result in increased work of breathing, together with reduced respiratory muscle capacity. ³⁷ It ultimately leads to increase weaning failure rate. Pneumonia was the most serious inflammatory disease of the respiratory system and also the most common infectious disease. ³⁸ A study by Yu et al found that the rate of extubation failure was as high as 34.4% in patients with severe pneumonia, and the total mortality reached up to 58.4%. ²⁸ Frutos‐Vivar et al also indicated that pneumonia at the initiation of ventilation were the best predictors of extubation failure. ³⁹ As stated above, pulmonary disease was associated with weaning failure.

Tu et al conducted a retrospective study and explored three major successful extubation predictors: GCS, V_E, and RSBI. ⁴⁰ In contrast to our study, GCS and V_E were also included in our model. However, the RSBI is calculated from the RR and the V_T. Therefore, we chose RR and V_T for inclusion in our model.

Our study explored the factors that can predict successful extubation and established a model for predicting extubation, considering not only the patients' ventilation parameters but also patient's medical conditions.

We hope the model can provide clinicians with an accurate assessment tool for determining the probability of successful extubation and that can increase the rate of successful extubation.

4.1. Limitations

Our present study has the following limitations:

1. The study participants were patients in medical ICU. Therefore, whether the model can be applied to all critical ill patients remains to be supported by more research evidence.

2. The number of patients in the successful extubation group and the extubation failure group in our study was different. However, a previous study indicated that a reasonable weaning failure rate is 10% to 20%. The weaning failure rate in our study was 16%, which is within this range of weaning failure rates.

3. May overoptimistic estimate of the predictive performance is a problem because a small size of in the failure group.

4. Not all clinical weaning predictors were included; those not considered included diaphragm movement, endotracheal secretions, lung compliance, cough function, P0.1, arterial blood gas parameters, and ratio of partial pressure of arterial oxygen to FiO₂.

5. Due to the observational retrospective study and the introduction time of EMRs, we have collected the maximum number of patients. We had planned to include more and different kinds of patients to evaluate and validate the model in the further.

5. CONCLUSION

We established a model for predicting extubation based on the predictive factors of successful extubation. The prediction accuracy was high, and the AUC value was 0.889.This model can accurately predict whether a patient can be successfully extubated and can thus improve outcomes. In addition, the results of this study can provide direction for predicting the factors behind successful extubation that can be used as a cornerstone for future research.

CONFLICT OF INTEREST

The authors declare no potential conflict of interest.

Supporting information

Data S1 The Logistic Regression Functions for Predicting Successful Extubation

Data S2 Kaohsiung Medical University Hospital Medical intensive care unit weaning protocol

Chung W‐C, Sheu C‐C, Hung J‐Y, Hsu T‐J, Yang S‐H, Tsai J‐R. Novel mechanical ventilator weaning predictive model. Kaohsiung J Med Sci. 2020;36:841–849. 10.1002/kjm2.12269