A total of 200 critically ill neurology patients admitted to the Intensive Care Unit of our hospital from January 2020 to December 2022, and who underwent invasive mechanical ventilation for more than 24 h, were selected for this study. This study was approved by the Ethics Committee of Hangzhou Normal University.

N = Z² × [P × (1-P)]/E², the confidence interval Z is set to 1.96, the probability value P is set to 90%, and the allowable error is set to less than 10%. It is concluded that the sample size required for this study is at least 60–80 cases. Considering that the success rate of extubation in ICU patients with traumatic brain injury is approximately 62% (14), the minimum sample size required for modeling is 96–129 cases. In this study, there were 140 patients in the modeling group, meeting the sample size requirement for modeling. At the same time, to ensure that there is a sufficient sample size to obtain the risk factors for extubation, the ratio of the sample size of the modeling group to the validation group was set at 7:3 (18). This not only ensures a sufficiently large modeling sample size but also avoids the problem of excessive sampling error due to the small number of data in the validation group. Therefore, the total sample size for this study is 200 cases, which is sufficient to ensure the reliability and accuracy of the research results. This investigation employed a stratified randomization technique to allocate subjects to diverse treatment cohorts. Stratified randomization is a strategy devised to enhance the integrity of randomization by dividing participants based on recognized pivotal variables (such as age, gender, body mass index) and subsequently randomizing within each stratum to guarantee an equitable distribution of these variables across the treatment groups within each stratum. To ascertain that the randomization remains balanced throughout the process, the distribution of treatment groups within each stratum and collectively is periodically scrutinized to detect any substantial imbalances.

The data collection work includes three aspects: demographic and baseline data, extubation-related data, and prognosis-related data. Demographic and baseline data mainly include age, gender, baseline Glasgow Coma Scale (GCS), BMI, location of brain injury, comorbidities (such as diabetes, hypertension, chronic lung disease, etc.), and the Acute Physiology and Chronic Health Evaluation (APACHE II) within 24 h of ICU admission.

Extubation is a critical procedure within the domain of intensive care. Its successful execution necessitates a meticulous evaluation of various physiological indices, the acuity of the illness, respiratory competence, coughing efficiency, and the presence of airway edema. Throughout the extubation process, a battery of standardized tests and clinical assessments are employed to gage the viability of extubation and to anticipate potential post-extubation complications. These include, but are not limited to:

Spontaneous Breathing Test (SBT): This test evaluates a patient’s capability to breathe independently of mechanical ventilation. It serves to assess the strength and synchronization of respiratory muscles. The success of an SBT is commonly correlated with parameters such as the peak expiratory flow rate of cough (C-PEFR), and diaphragmatic function.

Balloon Leak Test (BLT): This test is utilized to determine the extent of airway edema and to verify airway patency by observing whether the patient is able to breathe normally following deflation of the tracheal intubation balloon.

Diaphragmatic Ultrasound: As the diaphragm is the primary respiratory muscle, its movement and contraction rate can be monitored via ultrasound, thereby providing an indirect assessment of the patient’s respiratory function.

Clinical Scoring Systems including the Visual Inspiratory Sniffing Angle (VISAGE) Score and the Respiratory Insufficiency Intubation Score (RIS-1): These scoring systems integrate multiple clinical variables to offer a holistic evaluation of a patient’s risk profile during extubation. SBT and diaphragmatic function tests primarily reflect the patient’s respiratory muscle function, while the BLT and airway patency assessments are centered on identifying airway-related risks. The VISAGE and RIS-1 scores encapsulate the patient’s general health status and the risks associated with extubation.

Extubation-related data mainly includes records of successful spontaneous breathing trials (SBT), the date of the first extubation attempt or tracheotomy, and general management data collected on the day of extubation, such as the use of corticosteroids (to prevent wheezing after extubation) or the interruption of enteral nutrition, swallowing function, duration of tracheal intubation, cough reflex, sputum volume, and lung infection status. Standardized clinical examination on the day of extubation: vital signs (body temperature, heart rate, systolic blood pressure), breathing (including the type and time of SBT), physical examination (cough assessment, GCS eye language movement items, vomiting reflex). The definitions of these features were standardized based on previously described data. For example, a 4-level scale was used to assess cough intensity: severe, moderate, weak, and none. Record the time and reason for re-intubation. The swallowing function assessment uses the Standardized Swallowing Assessment (SSA) to screen patients for swallowing disorders before extubation. The scale consists of three parts, ranging from easy to difficult: preliminary clinical examination, 5 mL water test, and 60 mL water test. If any item is abnormal during the test, the test is immediately terminated, and the patient is judged to have a positive swallowing disorder. The score of this scale is negatively correlated with the patient’s swallowing function, with a minimum score of 18 and a maximum score of 46. Prognosis-related data mainly includes ICU length of stay, total length of hospital stay, and clinical outcome (death/transfer, etc.).

The extubation test usually refers to a series of examinations and assessments to assess a patient’s suitability for removal of the tracheal tube, which may include airway patency, the patient’s respiratory function, coughing ability, and swallowing ability. Swallowing grading usually refers to a method of assessing a patient’s ability to swallow, and it can help determine if the patient has dysphagia, which is related to whether the patient can eat and water safely after extubation. Swallowing grading may include observing the patient’s mouth and throat movements while swallowing, as well as assessing whether there is a risk of aspiration of food or fluid after swallowing. The cough reflex refers to a patient’s ability to clear foreign objects or secretions from the airway by coughing. This is an important defense mechanism for maintaining airway patency and is often evaluated before and after extubation. These assessment methods, including SBT, extubation tests, swallowing grading, and cough reflex assessment, are also applicable in neurosurgical patients. Since neurosurgical patients may be affected by surgical trauma, disturbance of consciousness, dysphagia, and other factors, these assessments are even more important because they can help physicians determine the safety of patients after extubation and possible risk of complications. Anesthetics may affect the evaluation of extubation tests. Anesthesia can affect a patient’s respiratory center, muscle strength, and coordination, so extubation tests performed during or immediately after anesthesia withdrawal may not accurately reflect a patient’s ability to breathe on his or her own in a fully awake state. Therefore, when evaluating extubation tests, we need to take into account the patient’s current state of anesthesia and evaluate after the patient is awake and the effects of anesthesia have subsided to obtain more accurate results.

Data were analyzed using SPSS 26.0. For data that followed a normal distribution, mean ± standard deviation (mean ± SD) was used to express the results; for non-normally distributed data, median and interquartile range (IQR) were used. Analysis of variance (ANOVA) was employed to test the differences among groups, Mann–Whitney U test was used for between-group comparisons with unequal variances, and t-test was used for comparisons between groups. Logistic regression analysis and other statistical methods were also applied. A p-value less than 0.05 was considered statistically significant.

In this study, we collected data on 200 ICU patients who met the inclusion criteria. Among them, 61 cases (43.6%) were spontaneous intracerebral hemorrhages, 73 cases (52.1%) were traumatic intracerebral hemorrhages, and 6 cases (4.3%) were ischemic strokes. The median age was 54 years old (ranging from 19 to 91 years old), and the baseline GCS score was 12 points (ranging from 3 to 15 points). Using a random number table method, 60 patients were selected for validation of the extubation risk prediction model (validation group). The construction of the risk prediction model (modeling group) was validated with the remaining 140 patients. Table 1 shows the general information and clinical characteristics of the patients in the derivation group, of which 115 cases (82.1%) were male, 25 cases (17.9%) were female, with an average age of 63 years old (ranging from 19 to 91 years old). 45.7% of patients were aged ≥65 years old; 15.7% of patients had a BMI <18.5, and 20% had a BMI ≥24. The APACHE II score within 24 h of ICU admission was 13.12 ± 3.75, with 33.6% of patients scoring ≥15. 77.1% of patients primarily underwent surgeries such as hematoma removal, decompressive craniectomy, and stereotactic puncture drainage. The length of ICU stay was 10.48 ± 5.26 days, and the total hospital stay was 28.54 ± 6.15 days. Finally, Table 1 also shows the general situation and clinical characteristics of patients in the verification group.

Before conducting regression analysis, continuous variables in this study were transformed into categorical variables according to literature review and clinical experience to make the predictive model more convenient for clinical use. Factors that may be significantly correlated with successful extubation include general characteristics of the patients such as age, APACHE II score, BMI, etc.; GCS score, oxygenation index on the day of extubation; airway characteristics including cough reflex, frequency of tracheal suctioning, and swallowing function (Table 2).

For whether the extubation was successful or not as the dependent variable, single-factor analysis was conducted on the general and clinical data that may affect critically ill neurologic patients and mechanically ventilated patients undergoing extubation. The results are shown in Tables 3, 4. Among the patients, 113 cases (80.7%) attempted extubation at least once, and 5 cases (3.6%) required reintubation within 48 h after extubation. In patients with extubation failure, there were fewer cases of spontaneous intracerebral hemorrhage [22 cases (52.4%) vs. 39 cases (39.8%), p = 0.02], older age [61 years old (19–91 years old) vs. 54 years old (20–89 years old), p = 0.001], and lower GCS scores on the day of extubation [7 points (5–8 points) vs. 7 points (5–9 points), p = 0.003]. Details are shown in Table 3.

The data of 140 patients was used to construct the model. The single-factor analysis results of the general data of the two groups showed that age, APACHEII score, and whether there was a combination of COPD had statistical significance (all p < 0.05), as shown in Table 3. Age ≥ 65 years old (2 = 11.685, p = 0.001), APACHEII score ≥ 15 points (2 = 6.734, p = 0.009), and the combination of chronic lung disease (2 = 4.879, p = 0.027) are potential risk factors for extubation failure. As age increases, the possibility of extubation failure tends to increase. As the APACHEII score increases, especially in patients with scores ≥15, the incidence rate of extubation failure increases. Patients with combined chronic lung disease have a higher proportion of eventual extubation failure clinically. The single-factor analysis results of the two groups with different characteristics showed that the GCS score on the day of extubation, sputum volume, cough strength, and swallowing function were statistically significant (all p < 0.05), as shown in Table 4.

Tolerance and variance inflation factor (VIF) were used to test for multicollinearity among variables prior to logistic regression analysis to determine if there is multicollinearity among variables. Based on the standard proposed by Kleinbaum DG, the criterion for the presence of multicollinearity is a tolerance of <0.1 or a VIF of >10.0 (19). In this study, there was no multicollinearity among the variables. The dependent variable is whether the extubation is successful, with an entry standard of 0.05 and a removal standard of 0.10. The meaningful variables in the univariate analysis: age ≥ 65 years (p = 0.001), APACHE II score ≥ 15 points (p = 0.009), combined chronic pulmonary disease (p = 0.027), etc., were included in the equation by the stepwise forward method (Likelihood Ratio, LR: forward) of the maximum likelihood ratio estimate. Table 5 shows the results of the binary logistic regression analysis. The equation itself is meaningful (χ² = 46.806, p < 0.001), NagelkerkeR2 = 0.401. Seven risk factors affecting whether the patient is successfully extubated entered the Logistic regression equation, namely age ≥ 65 years, high APACHE II score (≥ 15 points), chronic pulmonary disease, GCS score on the day of extubation, sputum volume (sputum suction frequency), cough reflex, and swallowing function. As age increases, the risk of extubation failure in ICU patients increases; simultaneously, as the severity of the patient’s disease increases, such as an increase in the APACHE II score, the incidence of extubation failure also increases, that is, the higher the severity of the disease, the higher the incidence of extubation failure; if chronic pulmonary disease is combined, the possibility of extubation failure increases; on the day of extubation, the results of the blood gas analysis of the SBT test, if the oxygenation index is less than 300, then re-intubation or tracheostomy is required after extubation. In addition, if the patient is in a deeper coma, with a GCS score of <8, and at the same time has a large amount of sputum, poor cough reflex, and swallowing dysfunction, it will greatly increase the failure rate of extubation.

In this cohort, 140 patients were available for constructing a scoring system to predict successful extubation. The simplified score retained 8 predictive factors (Supplementary material 1): age, APACHE II score, comorbid chronic pulmonary disease, oxygenation index, severe cough, swallowing function, suction frequency, and GCS score total. Based on the partial regression coefficients of the above variables, the ICU post-extubation dysphagia risk prediction model formula was constructed as follows: Risk of extubation failure = e a 1 + e a × 100 % , where e is the exponential function. a = 1.374 × age (= 0 or 1) + 1.518 × APACHE II score (= 0 or 1) + 0.971 × comorbid chronic pulmonary disease (= 0 or 1) + 1.038 × GCS score (= 0 or 1) + 2.020 × oxygenation index (= 0 or 1) + 1.225 × airway sputum volume [hourly suction frequency (= 1 or 2 or 3)] + 0.873 × cough intensity (= 1 or 2 or 3) + 1.170 × swallowing (SSA score) (= 1 or 2 or 3) − 4.330.

Based on the logistic regression analysis model, we assigned corresponding scores to each indicator’s partial regression coefficient, making the ICU extubation risk scoring model for severe neurological patients easier to use in clinical practice. Specifically, for each variable, we divided its partial regression coefficient by the smallest regression coefficient value in the logistic regression analysis model (0.971), then multiplied it by a constant 2, and rounded to the nearest integer (20). The final scores for each variable are shown in Table 6. The scores for age < 65, APACHE II score < 15, no history of chronic pulmonary disease, GCS score ≥ 8, oxygenation index ≥300, suction frequency, cough intensity, and swallowing function are 3, 3, 2, 3, 3, 2, 3, and 4, respectively.

The scoring criteria of the prediction model were substituted into the regression equation, and the ROC curve was drawn. The sensitivity and specificity at different scores can be obtained from this curve, and Youden’s index (sensitivity + specificity − 1) can also be calculated from this curve. By maximizing Youden’s index, the critical point can be determined for risk assessment. Table 7 shows that the highest Youden’s index total score is 18 points. Therefore, the critical value of this study model is set at 18 points. A score of less than 18 suggests that the risk of extubation failure may be higher, and a tracheotomy may be clinically recommended. Patients with a score of 18 or more have a relatively higher success rate of extubation and may try to wean off the ventilator.

To evaluate the established risk prediction model for extubation failure, the ROC curve method and the H-L goodness-of-fit test were used to test the discrimination and consistency of the model. The results of the H-L goodness-of-fit test showed that there was no statistical difference between the predicted incidence rate and the actual incidence rate of the model (x² = 5.241, p = 0.732, p > 0.05), indicating that the prediction model has good consistency. Figure 1 is the ROC curve of the extubation risk prediction model. The area under the curve (AUC), or C-statistic, is 0.832 (95% CI, 0.770–0.894), which is close to 1, indicating that the model can well distinguish whether the patients can be successfully extubated. At a score of 18, the specificity of the ROC curve of this model is 81.6%, and the relative sensitivity is 84.1%. The overall prediction accuracy rate is 81.5%, and the negative predictive value is 90.8%, which is relatively high, while the positive predictive value is relatively low at 66.3%, which may be due to the small number of positive samples (only 42 cases). Therefore, to further verify the predictive performance of the model, it is necessary to increase the sample size.

Table 8 shows the distribution of clinical variables in the two groups. There was no statistically significant difference in clinical variables between the two groups.

The variables of the validation group were input into the established risk prediction model of post-extubation dysphagia to calculate the risk of occurrence of dysphagia. The discriminative power and consistency of the prediction for the patients in the validation group were tested using the ROC curve method and the H-L goodness-of-fit test. The area under the curve (AUC) was 0.763 (95% CI, 0.652–0.875), as shown in Figure 2. The H-L goodness-of-fit test chi-square was 5.372, p = 0.717 (p > 0.05), suggesting that the established risk prediction model for extubation failure had good discriminative power and consistency. As shown in Table 9, the prediction results of the validation group indicate that the model has a high specificity of 84.8%, but a low sensitivity of 76.0%. The model’s negative predictive value is at a high level of 88.20%, while the positive predictive value is at a lower level of 59.20%. Overall, the model has a prediction accuracy of 79.8%. This indicates that the model can accurately predict the occurrence rate of extubation failure in ICU patients with severe neurological conditions who require tracheal intubation.