Development and validation of a nomogram for predicting the probability of successful extubation among patients with severe stroke receiving mechanical ventilation

Abstract

To address the lack of stroke-specific risk prediction tools for extubation failure, this study aimed to develop and validate a multidimensional nomogram integrating neurological function, respiratory parameters, and systemic status, to provide a basis for individualized clinical decision-making regarding extubation. We retrospectively enrolled 324 mechanically ventilated stroke patients admitted to the intensive care unit of Jining Medical University Affiliated Hospital from January 2022 to May 2024 as the training cohort, and 81 patients from June to December 2024 as the temporal validation cohort. The least absolute shrinkage and selection operator regression was used to screen risk factors from 43 candidate predictors. A nomogram was constructed using multivariate logistic regression. Model performance was evaluated using the receiver operating characteristic curve, calibration curve, decision curve analysis, and clinical impact curve. The developed nomogram integrates neurological function and dynamic respiratory parameters and can effectively identify intensive care unit stroke patients at high risk of extubation failure, potentially providing a tool for optimizing respiratory support strategies. The extubation failure rates were 46.6% (151/324) in the training cohort and 48.1% (39/81) in the validation cohort. Baseline data were well matched between cohorts. Least absolute shrinkage and selection operator regression identified 7 independent predictors. Multivariate logistic regression showed that the National Institutes of Health Stroke Scale score (odds ratio [OR] = 1.09, 95% confidence interval [CI]: 1.03–1.15, P = .003), Acute Physiology and Chronic Health Evaluation II score (OR = 1.08, 95% CI: 1.03–1.13, P = .001), duration of mechanical ventilation (OR = 1.06, 95% CI: 1.02–1.10, P = .001), fraction of inspired oxygen (FiO₂) (OR = 1.06, 95% CI: 1.02–1.11, P = .004), hemoglobin level (OR = 0.98, 95% CI: 0.97–0.99, P = .005), ischemic stroke type (OR = 0.47, 95% CI: 0.26–0.85, P = .012), and history of neurological disease (OR = 0.46, 95% CI: 0.27–0.77, P = .004) were independent influencing factors for extubation failure. The nomogram demonstrated areas under the curve of 0.789 (95% CI: 0.740–0.837) and 0.745 (95% CI: 0.639–0.851) in the training and validation cohorts, with sensitivities/specificities of 74.8%/82.1% and 70.5%/77.6%, respectively. The calibration curve showed minimal deviation (Hosmer–Lemeshow test P = .325), and decision curve analysis indicated a clinical net benefit across a threshold probability range of 15%–98%.

1. Introduction

Stroke, also known as cerebrovascular accident, is a disease caused by spasm or rupture of cerebral blood vessels due to lesions in the cerebral arterial system. It is characterized by 5 key features: high incidence, disability, mortality, recurrence rates, and substantial economic burden.^[1–3] China has the highest prevalence of stroke worldwide, and stroke is also the leading cause of death and disability in Chinese adults, accounting for 23% of all disease-related deaths.^[4] With the acceleration of population aging and urbanization, the exposure to risk factors for cerebrovascular diseases has increased, leading to a growing burden of cerebrovascular diseases.^[5]

Stroke has unique clinical characteristics: it often causes central nervous system damage, which in turn leads to respiratory dysfunction. Therefore, patients often require mechanical ventilation (MV) to sustain life during their stay in the Intensive Care Unit (ICU).^[6] Compared with isolated respiratory system diseases, stroke is more likely to cause pump failure or mixed injuries, resulting in prolonged MV duration.^[7] However, long-term dependence on MV may lead to multiple complications, including ventilator-associated pneumonia (VAP), tracheal injury, and reduced patient comfort.^[8] Although MV is crucial for stroke patients, its associated complications can prolong hospital stays, increase mortality rates,^[2] and impose additional economic burdens: data show that the MV-related hospitalization costs for stroke patients are 2.3 times higher than those for non-ventilated stroke patients.^[9]

Therefore, the development of individualized extubation strategies is particularly important. MV extubation refers to the process of gradually reducing or withdrawing ventilator support through systematic assessment, with the core goal of restoring patients’ spontaneous breathing capacity and minimizing ventilator-related complications.^[10,11] However, premature extubation may lead to insufficient oxygen supply and incomplete recovery of airway protective function, thereby resulting in extubation failure.^[10,11] extubation failure is a major challenge in the clinical management of stroke patients receiving MV in the ICU.^[12,13] Multiple studies have confirmed that compared with patients who successfully weaned, those with extubation failure have significantly prolonged total MV duration and ICU stay, increased tracheostomy rate, and higher mortality.^[12,13]

Epidemiological data show that the extubation failure rate is 28.2% in general ICU patients,^[14] while in stroke patients receiving MV, this rate can reach 46%.^[15] However, a unified extubation standard is still lacking in the field of neurocritical care,^[16] making the unique airway management challenges and the judgment of safe extubation timing urgent clinical bottlenecks to be addressed.^[17] Nevertheless, current clinical practice faces dual challenges: first, traditional extubation parameters (such as respiratory rate and tidal volume) have limited predictive efficacy in stroke patients, requiring the integration of neurological function indicators such as consciousness level assessment^[18]; second, although research data on MV extubation continue to accumulate,^[19] evidence specifically targeting extubation in ICU stroke patients receiving MV remains insufficient, and the reliability of existing extubation assessment standards in this population has not been fully verified.^[8,20] These issues require healthcare providers to comprehensively consider patients’ specific conditions and integrate the latest research evidence when formulating extubation plans to reduce the risk of extubation failure and improve success rates.^[21]

In summary, extubation failure not only prolongs MV duration but also is associated with higher mortality and poor prognosis.^[12,13] Therefore, this study focuses on constructing an accurate risk prediction model for extubation failure in ICU stroke patients, identifying the influencing factors of extubation failure in this population. The aim is to provide precise guidance for the timing of extubation implementation and a reference basis for the development of targeted intervention measures in the future.

2. Material and methods

2.1. Participants and design

A total of 324 MV stroke patients admitted to the ICU of Affiliated Hospital of Jining Medical University from January 2022 to May 2024 were retrospectively enrolled as the training cohort to construct the prediction model. Eighty-one MV stroke patients admitted from June to December 2024 were retrospectively included as the validation cohort to verify the model’s predictive value.

2.1.1. Inclusion criteria

Definite cerebrovascular lesions confirmed by CT or MRI, meeting the diagnostic criteria of Guidelines for the Prevention and Treatment of Cerebrovascular Diseases (2024 Edition),^[4] and requiring MV due to severe condition; meeting the initiation criteria for spontaneous breathing trial (SBT) (refer to international extubation guidelines)^[22]; completing the full extubation process during ICU stay with complete data records.

2.1.2. Exclusion criteria

Death before the first extubation attempt; invasive MV duration < 48 hours; tracheostomy performed before extubation.

2.2. Ethics approval

This study was approved by the Ethics Committee of Jining Medical University (Approval No.: JNMC-YX-2023-057). Informed consent was waived due to the retrospective nature of the study.

2.3. Data collection

Data were extracted from the hospital’s electronic medical record system, including: demographic data: age, gender, body mass index; comorbidities: neurological diseases (history of stroke), hypertension, diabetes, heart disease; Stroke characteristics: type, scope; preextubation scores: modified rankin scale, National Institutes of Health Stroke Scale (NIHSS), Glasgow Coma Scale (GCS), Acute Physiology and Chronic Health Evaluation II (APACHE II); preextubation vital signs: heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure; preextubation laboratory indicators: arterial blood potential of hydrogen (pH), standard bicarbonate (SB), hemoglobin; preextubation ventilator parameters: tidal volume (VT), peak inspiratory pressure (Peak), fractional inspired oxygen (FiO₂), rapid shallow breathing index (RSBI), oxygenation index (PaO₂/FiO₂), pressure control, MV mode; treatment details: surgical history, type of SBT (pressure-support ventilation vs T-piece Trial [T-piece]).

2.4. Definition of extubation success

extubation success was defined as meeting all the following criteria: arterial oxygen partial pressure (PaO₂) ≥ 50 mm Hg, oxygen saturation (SpO₂) ≥ 85%, arterial carbon dioxide partial pressure (PaCO₂) increase < 10 mm Hg, pH > 7.32; systolic blood pressure 90 to 160 mm Hg or change < 20%; heart rate < 120 beats/min or change < 20%; respiratory rate < 35 breaths/min or change < 50%; clear consciousness, good subjective feeling, no chest tightness or shortness of breath; spontaneous breathing without auxiliary ventilation for more than 24 hours, and no reintubation required within 48 hours after extubation.^[23]

2.5. Definition of extubation failure

extubation failure was defined as the presence of any of the following: respiratory distress with respiratory rate > 30 breaths/min; heart rate > 120 beats/min; altered mental status (e.g., restlessness, sweating); SpO₂ < 90%; significant increase in PaCO₂ leading to hypoxemia with pH < 7.25; reintubation or tracheostomy required within 48 hours after extubation.

2.6. Study protocol

This study was based on real-world clinical data. Pneumonia diagnosis was made by treating physicians based on clinical symptoms and imaging evidence (initial chest X-ray confirming infiltration). VAP and other types of pneumonia were collectively categorized as “pulmonary infection.” Thresholds for key indicators were set based on clinical relevance and previous research evidence. All extubation decisions were independently made by the clinical team of the participating center, and researchers only recorded objective outcomes to avoid intervention bias.

2.7. Statistical analysis

Data were analyzed using Fengrui Statistical Software (V2.1). Normally distributed continuous data were presented as mean ± standard deviation (x̄ ± s), and comparisons between groups were performed using independent samples t-test. Non-normally distributed data were presented as median (interquartile range) [M (Q_L, Q_U)], and group comparisons were conducted using Mann–Whitney U test. Categorical data were presented as frequency (percentage), and comparisons between groups were performed using χ² test or Fisher exact test. least absolute shrinkage and selection operator (LASSO) regression was used to screen independent risk factors for extubation failure in MV stroke patients. A predictive model was constructed using logistic regression, and a nomogram was developed for visualization. Receiver operating characteristic (ROC) curves and area under the curve (AUC) were used to evaluate the model’s discriminative ability. Hosmer-Lemeshow test and 500-time bootstrapping were used to generate calibration plots to verify model consistency. Clinical impact curves (CIC) and decision curve analysis (DCA) were further used to comprehensively evaluate the model’s clinical utility. A 2-tailed P < .05 was considered statistically significant.

3. Results

3.1. Baseline characteristics of MV stroke patients (Table 1)

Table 1.

Comparison of clinical baseline characteristics of stroke patients with weaning attempts between training and validation cohorts.

Variable		Total (n = 405)	Training cohort (n = 324)	Validation cohort (n = 81)		U/χ2/t value			P
Age (year, x̄ ± s)	65.2 ± 12.3		65.5 ± 12.2		64.0 ± 12.6		0.975	.324
Male (n [%])	262 (64.7)		208 (64.2)		54 (66.7)		0.173	.677
Smoking history (n [%])	144 (35.6)		115 (35.5)		29 (35.8)		0.003	.959
Drinking history (n [%])	133 (32.8)		108 (33.3)		25 (30.9)		0.179	.672
Comorbidities (n [%])
Neurological diseases	163 (40.2)		131 (40.4)		32 (39.5)		0.023	.879
Hypertension	301 (74.3)		244 (75.3)		57 (70.4)		0.828	.363
Diabetes	69 (17.0)		61 (18.8)		8 (9.9)		3.673	.055
Heart disease	103 (25.4)		81 (25)		22 (27.2)		0.159	.69
Stroke type (n [%])
Ischemic	98 (24.2)		77 (23.8)		21 (25.9)		0.165	.921
Hemorrhagic	261 (64.4)		210 (64.8)		51 (63)
Both	46 (11.4)		37 (11.4)		9 (11.1)
Stroke scope (n [%])
Multiple sites	189 (46.7)		149 (46)		40 (49.4)		Fisher	.246
Frontoparietal-temporal lobe	19 (4.7)		16 (4.9)		3 (3.7)
Brainstem	22 (5.4)		21 (6.5)		1 (1.2)
Thalamus	26 (6.4)		22 (6.8)		4 (4.9)
Cerebellum	22 (5.4)		19 (5.9)		3 (3.7)
Left basal ganglia	34 (8.4)		23 (7.1)		11 (13.6)
Right basal ganglia	52 (12.8)		39 (12)		13 (16)
Subarachnoid	41 (10.1)		35 (10.8)		6 (7.4)
Preweaning scores
MRS (M [QL, QU])	4.0 (3.0, 5.0)		4.0 (3.0, 5.0)		4.0 (3.0, 5.0)		0.715	.133
NIHSS (M [QL, QU])	16.0 (12.0, 19.0)		16.0 (12.0, 19.0)		16.0 (12.0, 20.0)		0.04	.841
GCS (M [QL, QU])	6.0 (3.0, 8.0)		5.0 (3.0, 8.0)		6.0 (4.0, 8.0)		2.231	.135
APACHEⅡ (x̄±s)	19.0 ± 5.8		19.2 ± 5.5		19.0 ± 5.8		0.11	.740
Preweaning vital signs (x̄±s)
Heart rate (beats/min)	84.1 ± 15.2		84.5 ± 15.9		82.4 ± 12.0		1.259	.263
Respiratory rate (breaths/min)	16.4 ± 4.0		16.6 ± 4.2		15.6 ± 3.5		4.157	.042
SBP (mm Hg)	138.7 ± 26.0		139.4 ± 26.2		135.5 ± 24.9		1.45	.229
DBP (mm Hg)	82.7 ± 17.2		82.9 ± 17.0		81.6 ± 17.9		0.362	.548
BMI (kg/m2)	24.6 ± 3.7		24.6 ± 3.7		24.5 ± 3.6		0.118	.731
Preweaning laboratory indicators (x̄±s)
PH	7.5 ± 0.1		7.5 ± 0.1		7.4 ± 0.1		4.66	.031
SB (mmol/L)	26.8 ± 3.6		27.0 ± 3.5		26.1 ± 4.1		4.443	.036
Na+ (mmol/L)	139.7 ± 10.1		139.8 ± 10.8		139.0 ± 6.1		0.443	.506
K+ (mmol/L)	4.1 ± 1.0		4.1 ± 1.1		4.1 ± 0.5		0.145	.704
Hemoglobin (g/L)	115.0 ± 20.5		114.4 ± 20.5		117.4 ± 20.4		1.354	.245
Preweaning ventilator parameters
VT (mL, x̄±s)	495.3 ± 81.7		493.7 ± 80.2		501.8 ± 87.5		0.631	.427
Peak (cmH2O, x̄±s)	16.7 ± 4.0		16.5 ± 3.9		17.5 ± 4.0		4.467	.035
FiO2 (%, x̄±s)	40.0 ± 6.4		39.9 ± 6.7		40.4 ± 5.3		0.333	.564
PaO2/FiO2 (mm Hg, M [QL, QU])	300.0 (222.5, 390.0)		302.7 (217.5, 395.6)		285.0 (245.0, 371.4)		0.004	.95
RSBI (f/L, M [QL, QU])	32.1 (26.4, 39.0)		32.4 (27.2, 39.8)		29.4 (25.0, 36.5)		4.582	.032
PC (mL/cmH2O, M [QL, QU])	8.0 (5.0, 12.0)		7.0 (5.0, 12.0)		9.0 (5.0, 12.0)		0.413	.521
Preweaning MV mode (n [%])
SPONT, CPAP	50 (12.3)		40 (12.3)		10 (12.3)		1.639	.65
P-A/C, PCV	254 (62.7)		200 (61.7)		54 (66.7)
P-SIMV, BIPAP	27 (6.7)		24 (7.4)		3 (3.7)
V-SIMV, SIMV	74 (18.3)		60 (18.5)		14 (17.3)
Treatment (n [%])
Spontaneous breathing trial
PSV	269 (66.4)		216 (66.7)		53 (65.4)		0.044	.833
T-piece	136 (33.6)		108 (33.3)		28 (34.6)
Surgery
No	162 (40.0)		132 (40.7)		30 (37)		0.37	.543
Yes	243 (60.0)		192 (59.3)		51 (63)
Nutritional support type
Oral	4 (1.0)		3 (0.9)		1 (1.2)		Fisher	.633
Enteral	376 (92.8)		302 (93.2)		74 (91.4)
Parenteral	25 (6.2)		19 (5.9)		6 (7.4)
Sequential treatment
Spontaneous breathing	58 (14.3)		45 (13.9)		13 (16)		0.915	.633
High-flow nasal cannula oxygenation	225 (55.6)		178 (54.9)		47 (58)
Noninvasive positive pressure ventilation	122 (30.1)		101 (31.2)		21 (25.9)
Antibiotic type
None	76 (18.8)		59 (18.2)		17 (21)		0.64	.887
Cephalosporins	159 (39.3)		130 (40.1)		29 (35.8)
Piperacillin	95 (23.5)		75 (23.1)		20 (24.7)
Combination	75 (18.5)		60 (18.5)		15 (18.5)
Outcome
Weaning success (n [%])
Yes	215 (53.1)		173 (53.4)		42 (51.9)		0.062	.803
No	190 (46.9)		151 (46.6)		39 (48.1)
ICU stay (days, M [QL, QU])	23.0 (15.0, 34.0)		23.0 (15.0, 35.0)		22.0 (14.0, 32.0)		0.252	.616
MV duration (days, M [QL, QU])	13.0 (9.0, 18.0)		13.0 (8.0, 18.0)		12.0 (9.0, 19.0)		0.621	.431
Treatment effect (n [%])
Death	14 (3.5)		10 (3.1)		4 (4.9)		Fisher	.808
Improvement	26 (6.4)		21 (6.5)		5 (6.2)
Transfer	84 (20.7)		69 (21.3)		15 (18.5)
Others	281 (69.4)		224 (69.1)		57 (70.4)
Pulmonary infection (n [%])
No	182 (44.9)		147 (45.4)		35 (43.2)		0.122	.727
Yes	223 (55.1)		177 (54.6)		46 (56.8)

APACHE II = acute physiology and chronic health evaluation II, BMI = body mass index, CI = confidence interval, CPAP = continuous positive airway pressure, DBP = diastolic blood pressure, ICU = intensive care unit, MV = mechanical ventilation, PC = pressure control, P-SIMV = pressure-support intermittent mandatory ventilation, PSV = pressure-support ventilation, RSBI = rapid shallow breathing index, SB = serum bicarbonate, SBP = systolic blood pressure, SPONT = spontaneous breathing, VT = tidal volume.

The extubation failure rate was 46.6% (151/324) in the training cohort and 48.1% (39/81) in the validation cohort. Significant differences were observed in respiratory rate, pH, SB, Peak, and RSBI between the training and validation cohorts (all P < .05), while no significant differences were found in other indicators, indicating good comparability.

3.2. Predictive model for extubation Failure in ICU MV stroke patients

Taking 17 predictors with P < .05 in the univariate analysis as independent variables, including drinking history, comorbid neurological diseases, stroke type, 4 preextubation scores modified rankin scale, NIHSS, GCS, and APACHE II, respiratory rate, hemoglobin, Peak, FiO₂, pressure control, type of SBT, type of nutritional support, type of sequential treatment, duration of MV, and pulmonary infection status. A LASSO regression model was established with extubation success or failure as the dependent variable (Fig. 1), and 7 potential predictors of extubation failure in stroke patients were identified, namely NIHSS score, duration of endotracheal intubation and MV, APACHE II score, hemoglobin level, FiO₂, history of neurological diseases (yes), and stroke type (ischemic). Subsequently, logistic regression analysis was performed with extubation success or failure as the dependent variable and the 7 predictors screened by the LASSO regression model as independent variables. Results showed that all 7 variables were independent risk predictors for extubation failure (all P < .05, Table 2), and a nomogram prediction model was constructed based on these variables (Fig. 2).

Figure 1.

Feature selection using the LASSO logistic regression model. (A) The selection process of parameter λ; (B) dynamic process diagram of λ and variables; (C) feature number optimization analysis via Bayesian information criterion. LASSO = least absolute shrinkage and selection operator.

Table 2.

Multivariate logistic regression model of successful weaning of MV in ICU stroke patients.

Variable	B	Wald	SE	OR	95% CI	P
Constant term	−3.436	−2.665	1.290	0.03	0.003–0.40	.008
APACHE Ⅱ score	0.079	3.337	0.024	1.08	1.03–1.13	.001
MV duration	0.061	3.212	0.019	1.06	1.02–1.10	.001
NIHSS score	0.085	2.936	0.029	1.09	1.03–1.15	.003
FiO2	0.061	2.889	0.021	1.06	1.02–1.11	.004
Hemoglobin	−0.018	−2.787	0.006	0.98	0.97–0.99	.005
Stroke type (hemorrhagic)	−0.759	−2.508	0.303	0.47	0.26–0.85	.012
History of neurological diseases (yes)	−0.782	−2.916	0.268	0.46	0.27–0.77	.004

APACHE II = acute physiology and chronic health evaluation II, CI = confidence interval, ICU = intensive care unit, MV = mechanical ventilation, NIHSS = National Institutes of Health Stroke Scale, OR = odds ratio.

Figure 2.

Nomogram for predicting weaning success in ICU stroke patients. ICU = intensive care unit.

3.3. Multivariate logistic regression model for successful MV extubation in ICU patients with stroke

To identify the independent influencing factors for successful MV extubation in ICU patients with stroke, multivariate logistic regression analysis was performed with extubation success as the dependent variable and the 7 potential predictors screened by LASSO regression (APACHE Ⅱ score, duration of endotracheal intubation and MV, NIHSS score, FiO₂, hemoglobin level, stroke type, and history of neurological diseases) as independent variables. The results are presented in Table 2.

As shown in Table 2, 7 variables were confirmed as independent predictors for MV extubation success in ICU patients with stroke (all P < .05). Specifically, higher APACHE Ⅱ score (OR = 1.08, 95% CI: 1.03–1.13, P = .001), longer duration of endotracheal intubation and MV (OR = 1.06, 95% CI: 1.02–1.10, P = .001), and higher NIHSS score (OR = 1.09, 95% CI: 1.03–1.15, P = .003) were independent risk factors for extubation failure. Similarly, higher FiO₂ was also associated with an increased risk of extubation failure (OR = 1.06, 95% CI: 1.02–1.11, P = .004). In contrast, higher hemoglobin level was an independent protective factor for extubation success (OR = 0.98, 95% CI: 0.97–0.99, P = .005). Regarding categorical variables, hemorrhagic stroke (OR = 0.47, 95% CI: 0.26–0.85, P = .012) and history of neurological diseases (OR = 0.46, 95% CI: 0.27–0.77, P = .004) were associated with a lower risk of extubation failure compared with ischemic stroke and no history of neurological diseases, respectively.

3.4. Model construction and validation

3.4.1. Construction of the nomogram

Based on the aforementioned 7 independent predictors and their regression coefficients (Table 2), a nomogram for predicting the probability of successful MV extubation in ICU patients with stroke was constructed (Fig. 2). This nomogram consists of 11 scales, arranged from top to bottom as follows: Points (single-item score scale), Stroke Type, History of Neurological Diseases, Hemoglobin (g/L), MV Duration (day), APACHE Ⅱ Score, NIHSS Score, FiO₂ (%), Total Points (total score scale), Linear Predictor (linear combination value of the Logistic regression model, used for probability conversion), and Probability of Successful extubation.

Each scale corresponds to 1 predictive factor, and the scale marks are directly linked to the scores assigned to different values of the factor. The score weights were calibrated using regression coefficients, which intuitively reflect the impact intensity of each factor on the probability of successful extubation. The application steps of the nomogram are as follows: First, locate the actual value of each predictive factor on its corresponding scale; second, read the single-item score corresponding to that value; third, sum all single-item scores to obtain the total score (Total Points); finally, based on the position of the total score on the Total Points scale, read the corresponding Linear Predictor value, and ultimately obtain the predicted probability of successful extubation for the patient on the Probability of Successful extubation scale, thereby realizing individualized risk quantitative assessment (Fig. 2).

3.4.2. Evaluation of discriminative ability, goodness-of-fit, and clinical efficacy of the nomogram model

ROC curve analysis (Fig. 3A) showed that the nomogram model had a AUC of 0.789 (95% confidence interval [CI]: 0.740–0.837, P < .001), with a sensitivity of 74.8% and a specificity of 82.1%. This indicates that the model has good predictive ability for the risk of extubation failure in ICU stroke patients receiving MV. The Brier score was 20.1% (95% CI: 16.1–24.1), reflecting a low overall predictive error of the model. The calibration curve (Fig. 4A) showed a small fitting deviation between the predicted probability of the model and the actual observed probability (Hosmer-Lemeshow test, χ² = 12.738, P = .325), suggesting a high consistency between the predicted results and the actual outcomes. DCA (Fig. 5A) indicated that when the threshold probability ranged from 15% to 98%, using this model to guide extubation decisions could significantly improve the clinical net benefit. CIC (Fig. 6A) further verified that when the clinical decision threshold probability was > 0.35, the absolute error between the number of extubation failure cases predicted by the model and the actual number of cases was < 5%, indicating accurate risk stratification performance.

Figure 3.

ROC curve of the prediction model for weaning success in ICU stroke patients receiving mechanical ventilation (B). (A) Development cohort. (B) Validation cohort. ICU = intensive care unit, ROC = receiver operating characteristic.

Figure 4.

Calibration curve of the prediction model for weaning success in ICU stroke patients receiving mechanical ventilation (A). (A) Development cohort. (B) Validation cohort. ICU = intensive care unit.

Figure 5.

DCA of the prediction model for successful weaning in ICU stroke patients receiving mechanical ventilation (A). (A) Development cohort. (B) Validation cohort. DCA = decision curve analysis, ICU = intensive care unit.

Figure 6.

CIC of the prediction model for successful weaning in ICU stroke patients receiving mechanical ventilation (A). (A) Development cohort. (B) Validation cohort. CIC = clinical impact curve, ICU = intensive care unit.

3.4.3. Model validation

In the validation cohort, ROC curve analysis (Fig. 3B) showed that the AUC of the nomogram model for predicting extubation failure in ICU stroke patients receiving MV was 0.745 (95% CI: 0.639–0.851, P < .01), with a sensitivity of 70.5% and a specificity of 77.6%. The calibration curve (Fig. 4B) showed a high consistency between the predicted risk probability of the model and the actual observed outcomes. Both the DCA curve (Fig. 5B) and CIC curve (Fig. 6B) indicated that the model had stable predictive performance and clinical transformation value.

4. Discussion

4.1. Significance of constructing a extubation prediction model for ICU stroke patients receiving MV

MV extubation is a key link in the clinical management of patients with severe stroke, typically accounting for 40% to 50% of the total MV treatment time.^[24] Studies have shown that extubation failure not only prolongs the ICU stay and increases medical costs but also has a significant positive correlation with patient mortality. However, traditional parameters have low specificity, and the existing extubation assessment system has insufficient predictive efficacy in stroke patients^[25,26]; moreover, it fails to integrate multi-dimensional clinical features, physiological data, medical imaging, and other factors, resulting in 10%–29% of patients requiring reintubation even if they meet the extubation criteria.^[27]

Notably, the COVID-19 pandemic has further exacerbated these challenges, thereby highlighting the clinical utility of our nomogram: Bagi et al^[28] showed resource constraints raised COVID-19 patient readmissions, highlighting needs for precise stroke extubation tools; Shahsavarinia et al^[29]’s review linked pandemic drugs to cardiovascular risks, destabilizing stroke patients. These challenges amplify extubation uncertainty. Therefore, identifying factors affecting extubation failure in patients with severe stroke receiving MV and formulating effective preventive measures based on different influencing factors are of great significance for improving the extubation success rate.

4.2. Analysis of main outcomes of extubation in ICU stroke patients receiving MV

Among 405 stroke patients receiving MV, the extubation failure rate was 46.9% (190/405) and the success rate was 53.1% (215/405), which is consistent with the findings of Yao et al.^[30] This study adopted the widely accepted definition of extubation failure, i.e., requiring reintubation within 48 hours after extubation^[31] or tracheostomy.^[32] However, existing studies have significant heterogeneity in the definition of MV extubation failure; for example, the time threshold ranges from 48 hours to 2 weeks, and outcome indicators include reintubation or tracheostomy, leading to a failure rate ranging from 6% to 77%.^[33]

4.3. Analysis of secondary outcomes of extubation in ICU stroke patients receiving MV

The median ICU stay was significantly prolonged in the extubation failure group (27.0 [15.0–37.5] days vs 21.0 [15.0–32.0] days, P = .051), and the duration of MV was also significantly higher than that in the success group (15.0 [10.0–20.0] days vs 12.0 [7.0–16.0] days, P < .001). This is consistent with the conclusions of global multi-center studies, suggesting a significant positive correlation between MV duration and medical resource consumption.^[34] In terms of prognostic outcomes, the mortality rate (4.6% vs 1.7%) and incidence of pulmonary infection (64.2% vs 46.2%) in the extubation failure group were significantly increased (P < .01), while the clinical improvement rate (2.6% vs 9.8%) was significantly decreased. This difference may be related to the increased risk of VAP and diaphragmatic dysfunction in the extubation failure group.^[35]

4.4. Key variables of the prediction model for extubation failure in ICU stroke patients receiving MV and their guiding value for nursing practice

4.4.1. Impact of NIHSS score on extubation failure and implications for nursing practice

The results of this study showed that the NIHSS score is not only a core quantitative indicator of neurological deficit but also an important parameter for predicting respiratory complications. This is consistent with the research results of Chen Qianxi et al^[33] on patients with acute ischemic stroke: patients with an admission NIHSS score ≥ 15 had a significantly increased risk of extubation failure (OR = 2.2, 95% CI: 1.8–10.9, P = .001). This phenomenon is closely related to the injury of the brainstem and medullary respiratory centers. As a key region for respiratory rhythm generation, ischemic injury of the medulla oblongata leads to weakened respiratory drive and abnormal phrenic nerve regulation, manifested as decreased diaphragmatic contraction coordination and central respiratory pattern disorder.^[36] In clinical nursing practice, nurses need to dynamically monitor changes in the NIHSS score to provide an important basis for selecting the timing of extubation. Studies have confirmed that patients with extubation failure have a significantly lower diaphragmatic contraction efficiency index.^[37] In such cases, precise assessment using bedside ultrasound is required: when the diaphragm excursion < 1.2 cm or the diaphragm thickening fraction < 30%, it indicates diaphragmatic dysfunction,^[38] and extubation should be delayed while initiating targeted respiratory muscle training.^[39]

4.4.2. Impact of APACHE Ⅱ score on extubation failure and implications for nursing practice

APACHE Ⅱ score, as a core indicator for assessing the multi-organ function status of critically ill patients, has been verified in this study to have a positive correlation with the risk of extubation failure. A study by Cheng et al^[40] on stroke patients found that an APACHE Ⅱ score ≥ 15 was an independent risk factor affecting extubation success. A study by Liu et al^[41] also found that a higher APACHE Ⅱ score indicated a higher risk of extubation failure (OR = 1.181, 95% CI: 1.052–1.336, P = .006). In addition, a study by Mei et al^[42] found that the APACHE Ⅱ score was significantly correlated with extubation failure and 30-day mortality in patients with aneurysmal subarachnoid hemorrhage. Furthermore, the incidence of VAP increased by 57% in weaned patients with an APACHE Ⅱ score ≥ 20.^[43] The nursing team should assess this score daily, focusing on cardiovascular function, renal function, and infection control (such as VAP prevention). Especially for patients with sepsis or multiple organ dysfunction, multidisciplinary collaboration is required to achieve organ function rebalancing, and the extubation process should be initiated only after prioritizing the optimization of systemic status.^[44]

4.4.3. Impact of MV duration on extubation failure and implications for nursing practice

The data of this study showed a significant positive correlation between the duration of MV and the risk of extubation failure in ICU stroke patients. A cohort study by Taran et al^[45] showed that a longer MV duration in patients with acute brain injury (OR = 3.47; 95% CI: 1.68–7.19) was associated with a higher extubation failure rate, which is consistent with the results of this study. Further analysis revealed that patients with MV duration exceeding 7 days had a 2.3-fold increased risk of diaphragmatic atrophy and ICU-acquired weakness,^[46,47] suggesting that prolonged positive pressure ventilation may indirectly lead to extubation difficulties by inducing respiratory muscle dysfunction and systemic inflammatory response.^[48] In addition, a study by Krueger et al^[49] found that MV duration, infratentorial lesion localization, and NIHSS score > 15 were the 3 core predictors of extubation failure in neurocritical patients. The team suggested initiating a multidisciplinary extubation assessment for patients who could not be separated from respiratory support 7 days after ICU admission. If the predicted extubation success rate is <30%, tracheostomy should be performed in a timely manner to reduce the risk of VAP. This strategy is consistent with the principle of “early intervention and dynamic assessment” in recent MV management guidelines.^[50] The nursing team needs to establish an MV duration tracking system, initiate a red alert for stroke patients with MV duration ≥ 7 days, and collaborate with respiratory therapists to perform daily diaphragmatic ultrasound assessments and record dynamic changes in the RSBI. Referring to the mechanical power-guided prediction model, patients are divided into low-risk (mechanical power < 12 J/min), medium-risk (12–17 J/min), and high-risk (> 17 J/min) groups. For high-risk groups, an individualized extubation strategy of gradually reducing pressure-support levels combined with neurally adjusted ventilatory assist is adopted. Nurses need to check ventilator parameters and patient tolerance every 4 hours.^[51]

4.4.4. Impact of hemoglobin level on extubation failure and implications for nursing practice

This study confirmed a significant association between hemoglobin level and extubation outcomes in ICU stroke patients receiving MV. A study by Chen Qianxi et al^[33] also showed that factors such as large-area infarction and low hemoglobin level were correlated with MV duration in patients with extubation failure. Another study^[52] pointed out that the hemoglobin level of patients who received red blood cell transfusion during extubation was significantly lower than that of the non-transfusion group (86.3 ± 5.3 vs 95.8 ± 10.5 g/L), and multivariate regression showed that transfusion increased the risk of extubation failure by 3.24 times (P = .045), which may be related to transfusion-related inflammatory responses and microcirculatory disorders. This provides the following guiding directions for nursing practice:

1. Establish a dynamic hemoglobin monitoring system: conduct screening before extubation (target hemoglobin ≥ 90g/L) and detect arterial oxygen content every 6 hours during extubation^[53];

2. Implement an individualized transfusion strategy: consider transfusion when hemoglobin < 85g/L, and monitor C-reactive protein levels after transfusion to assess inflammatory responses^[52];

3. Carry out intensive nutritional intervention: supplement iron (200mg/d) and vitamin C (500 mg/d) for patients with low hemoglobin, combined with a high-protein diet (1.5–2.0 g/kg/d)^[37];

4.4.5. Impact of FiO₂ on extubation failure and implications for nursing practice

FiO₂ is a key parameter for oxygenation management in mechanically ventilated patients, and its setting level is significantly correlated with extubation outcomes. This study found that FiO₂ and MV duration (> 24 hours) are key indicators for extubation risk assessment, which is consistent with the research results of Jia et al.^[54] A study showed that in patients with acute ischemic stroke after endovascular treatment, maintaining arterial partial pressure of oxygen ≥ 150 mm Hg significantly increased the risk of poor neurological function recovery at 3 months (OR = 1.89, 95% CI: 1.21–2.95).^[55] This indicates that supraphysiological oxygen partial pressure caused by excessively high FiO₂ may exacerbate brain injury through oxidative stress. It is worth noting that in patients with high-flow nasal cannula failure, a SpO₂/FiO₂ ratio ≤ 214 increases the intubation risk by 4.3 times,^[56] and this indicator can be used as an important reference for FiO₂ adjustment during extubation. This may be because FiO₂ levels can not only assess the severity of hypoxemia but also indirectly reflect the compensatory capacity of the circulatory system and oxygen delivery efficiency.^[57] Compared with the traditional single threshold method, specific nursing operation points are recommended as follows:

1. Establish a dynamic FiO₂-oxygenation status monitoring form, and record the SpO₂/FiO₂ ratio and neurological assessment results every hour^[58];

2. Under the support of rapid bedside blood gas analyzer testing, accurately control the target PaO₂ within the range of 80 to 110 mm Hg^[59];

3. Collaborate with respiratory therapists to implement lung protection strategies, and initiate a prone position ventilation plan when FiO₂ > 60% for 4 consecutive hours;

4. Conduct multidisciplinary extubation assessment meetings, use FiO₂ requirement > 40% as an early warning indicator for delayed extubation, and construct a “monitoring-adjustment-early warning” trinity FiO₂ management system.^[59]

4.4.6. Impact of history of neurological diseases on extubation failure and implications for nursing practice

This study found that a history of neurological diseases is an independent risk factor for extubation failure (OR = 0.46, 95% CI: 0.27–0.77, P = .004), which is consistent with the research conclusion of Battaglini et al.^[60] The MV management for patients with a history of different neurological diseases should be adjusted according to specific subtypes. A history of neurological diseases may cause neurogenic respiratory muscle dysfunction, thereby affecting extubation outcomes. For example, patients with a previous cerebrovascular event or neurodegenerative diseases (such as dementia) may experience weakened central respiratory drive and decreased respiratory muscle coordination.^[61] Based on this, the following implications for nursing practice can be drawn: standardized tools should be used within 24 hours of admission to conduct early risk assessment and screen high-risk patients. For these high-risk patients, phrenic nerve electrical stimulation should be initiated, combined with a goal-directed sedation protocol, to reduce the extubation failure rate by enhancing respiratory drive.^[45]

4.4.7. Impact of ischemic stroke type on extubation failure and implications for nursing practice

This study confirmed that compared with hemorrhagic stroke, patients with ischemic stroke have a significantly higher risk of extubation failure (OR = 0.47, 95% CI: 0.26–0.85, P = .012). A cohort study by Thibaut et al^[62] provided a pathophysiological explanation: patients with ischemic stroke are more likely to develop central hypoventilation syndrome due to brainstem or bilateral hemisphere involvement, and their proportion of MV needs is 37% higher than that of hemorrhagic stroke (76.8% vs 23.2%). Implications for nursing practice:Brain-lung interaction management: adopt goal-directed intracranial pressure monitoring for patients with ischemic stroke, and it is recommended to maintain cerebral perfusion pressure > 60 mm Hg, which can reduce the risk of respiratory drive inhibition (OR = 0.41, 95% CI: 0.22–0.76).^[63]

4.5. Innovation of extubation in ICU Stroke Patients Receiving MV

Compared with the international prospective ENIO study on neurocritical patients,^[27] this study achieved higher predictive efficacy in the stroke population (AUC increased by 4.9%). The Brier score (20.1%) and calibration consistency verified the reliability of clinical risk stratification, and the stable net benefit was verified by DCA and CIC curves, providing a more accurate tool for individualized extubation decisions in stroke patients.

4.6. Limitations of extubation in ICU Stroke Patients Receiving MV

This study has some limitations. First, the retrospective design and single-center data may lead to insufficient sample representativeness and the risk of selection bias. Second, although internal validation was performed using time-series segmentation, the generalization ability of the model has not been verified by multi-center external datasets. Finally, the existing predictor system needs to be improved, and more multi-dimensional clinical indicators should be integrated. Future studies are recommended to adopt a prospective cohort design, establish a large-scale sample database in collaboration with multiple medical institutions, and include new biomarkers and other predictive variables to improve the external validity and clinical application value of the prediction model.

5. Conclusions

In summary, this study is the first to construct and validate a nomogram for predicting extubation risk in ICU stroke patients receiving MV. Based on 7 independent predictors (including NIHSS score, APACHE II score, and MV duration) screened by LASSO regression, the model showed excellent predictive efficacy in both the training set (AUC = 0.789) and the validation set (AUC = 0.745). This tool can accurately identify patients at high risk of extubation failure, guide the formulation of individualized respiratory support strategies, and reduce the reintubation rate and the incidence of ventilator-related complications, thereby providing a theoretical basis for extubation decisions in neurocritical patients.

Author contributions

Conceptualization: Ranran Yan.

Data curation: Yuanqing Li, Ranran Yan.

Formal analysis: Chunyuan Zhao.

Investigation: Anhao Liu.

Methodology: Xiuping Zhang.

Resources: Ranran Yan, Dongmei Wu.

Supervision: Xuehui Zhang.

Validation: Xuehui Zhang.

Writing—original draft: Meiqi Liu.

Writing—review & editing: Meiqi Liu, Xuehui Zhang, Dongmei Wu.