Effectiveness of a standardized weaning protocol in tracheostomized patients with prolonged mechanical ventilation: a prospective single-arm interventional trial

Abstract

Background

Patients requiring prolonged mechanical ventilation (PMV) after tracheostomy represent a severely deconditioned population with high mortality and substantial healthcare burden. Weaning outcomes vary widely across centers, and evidence supporting standardized protocol in dedicated rehabilitation settings remains limited.

Methods

We conducted a prospective, single-arm interventional trial at a specialized weaning unit (SWU) of a tertiary rehabilitation hospital. From January 2025 to October 2025, consecutive tracheostomized PMV patients were enrolled and managed with a standardized stepwise weaning protocol integrating spontaneous breathing trials, high-flow oxygen therapy, and noninvasive ventilator. The primary outcome was the 60-day weaning success rate, which was tested against a pre-specified historical benchmark (p₀ = 65%) using exact binomial, score-based, and Katz log-method analyses. Secondary outcomes included 6-month mortality, analyzed through modified Poisson regression with propensity score matching and overlap weighting. An individualized calculated weaning score was developed using LASSO-Poisson regression with log(p₀) as an offset to elucidate the heterogeneity in treatment response.

Results

Weaning success was achieved in 189 of 250 patients (75.6%; 95% CI 69.9–80.5%), significantly exceeding the historical benchmark (RR, 1.16; 95% CI 1.08–1.25; p < 0.001). Mean weaning duration was 19.9 ± 10.0 days; 30-day reconnection occurred in 2.1% of successfully weaned patients. Six-month mortality was 14.4% and was lower among successfully weaned patients (9.0% vs 31.1%) with confounding and limited event numbers precluding a definitive causal inference. Tertile stratification by the calculated weaning score revealed a monotonic gradient in weaning success (T1, 44.0%; T2, 84.3%; T3, 98.8%; trend p < 0.001). The tool was deployed online at https://jianghongying.shinyapps.io/Weaning-for-PMV/.

Conclusions

A standardized weaning protocol delivered in a dedicated rehabilitation unit was associated with a 60-day weaning success rate that exceeded a pre-specified historical benchmark in PMV patients. The individualized calculated weaning score identifies clinical phenotypes for individualized decision-making; the accompanying web-based calculator is provided as an exploratory research tool and warrants external validation before any routine clinical application.

Trial registration

ClinicalTrials.gov, NCT06642714.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40560-026-00935-x.

Background

Prolonged mechanical ventilation (PMV), generally defined as the requirement for invasive ventilation for more than 6 hours per day over a minimum of 21 consecutive days [1], has emerged as one of the most resource-intensive phenotypes in contemporary critical care. Population aging and sustained improvements in short-term intensive care unit (ICU) survival have progressively shifted severely ill patients from early mortality toward long-term ventilator dependence [2]. A systematic review of 124 cohorts estimated in-hospital and 1-year mortality at 29% and 59%, respectively, among PMV patients [3]. Recent multicenter data indicate that fewer than one in four survivors regain functional independence [4]. Tracheostomized PMV patients typify the chronic critical illness phenotype, characterized by persistent organ dysfunction, diaphragmatic and skeletal-muscle wasting, recurrent infection, and neurocognitive impairment [5, 6]. Their economic burden substantially exceeds that of the general ICU population, reflected in prolonged hospitalization and frequent readmission [4, 7]. Consequently, liberation from ventilator is not merely a technical endpoint but the principal determinant of survival, functional recovery, and healthcare resource utilization.

The modern conceptual framework for ventilator liberation rests on two cornerstones. The WIND study reclassified weaning into simple, difficult, and prolonged categories, and showed that prolonged weaning, although affecting only 10–14% of ventilated patients, accounts for the majority of weaning-related mortality and resource consumption [8]. Subsequently, the ATS/CHEST liberation guidelines standardized screening, spontaneous breathing trials (SBT), and post-extubation support [9]. Recommendations specific to tracheostomized PMV patients, however, rest on low- to moderate-quality evidence. Reported weaning success rates range from 30 to 85% across specialized centers [10, 11], and expert consensus commonly adopts 60–65% as a pragmatic historical benchmark for outcome assessment [12, 13]. The individual components of weaning support have robust but fragmented evidence bases. Recent multicenter randomized trials have refined the choice of SBT protocol for extubation decisions [14], yet their applicability to stepwise pressure-support de-escalation in tracheostomized patients has not been directly tested. High-flow oxygen therapy (HFOT) delivered via tracheostomy improves oxygenation, reduces respiratory work, and enhances secretion clearance in small physiological studies [15, 16], but prospective evaluations with weaning success as endpoint are lacking in the PMV population. Noninvasive ventilator (NIV) has been evaluated as a facilitation strategy for early extubation in intubated patients with a natural airway; systematic reviews and meta-analyses in this population have reported reductions in mortality, ventilator-associated pneumonia, reintubation, and tracheostomy rates [17, 18]. Meanwhile, post-extubation trials have elucidated its comparative efficacy relative to HFOT [19, 20]. However, these findings pertain to short-term intubation and mask-interface NIV, and their direct applicability to tracheostomized PMV patients—who require ventilation through distinct single-limb circuits with different leak characteristics, exhalation-port positioning, and CO₂ clearance dynamics—has not been prospectively established. This distinction underscores the need for protocol-level evidence specific to tracheostomy-delivered ventilatory support in the PMV population.

Institutional responses have coalesced around specialized weaning units (SWU), respiratory intermediate care units, and long-term acute-care hospitals (LTACH). North American LTACH networks report success rates of approximately 54–58%, but modest gains in long-term survival and functional recovery [13, 21]. European regional weaning centers achieve success rates of 60–75% yet display marked heterogeneity in admission criteria, protocol content, and rehabilitation intensity [22, 23]. In Asia, and particularly within the Chinese rehabilitation hospital system, prospective protocolized evidence for SWU-based weaning is virtually absent. SWU has the conceptual advantages, including combined medical and rehabilitative capability, high-dependency monitoring, and treatment horizons long enough to accommodate stepwise respiratory-support transitions alongside systematic pulmonary rehabilitation, early mobilization, and nutritional optimization [10, 24]. These advantages align with accumulating evidence that PMV weaning requires coordinated respiratory support, rehabilitation, and multidisciplinary management rather than isolated technical maneuvers [11, 25]. Although the TEAM trial has cautioned that indiscriminately high-intensity mobilization may not improve outcomes [26], underscoring the importance of phenotype-matched intervention intensity. Moreover, patient-level heterogeneity in weaning response is substantial, yet conventional single-covariate subgroup analyses are prone to false-positive findings; an individualized calculated weaning score (ICWS) anchored to a historical benchmark and adapted to single-arm real-world designs has, to our knowledge, not been developed for PMV weaning in a rehabilitation setting.

Two specific gaps therefore persist: (i) the prospective effectiveness of a standardized stepwise weaning protocol integrating SBT, tracheostomy-delivered HFOT, and NIV in tracheostomized PMV patients managed in a rehabilitation hospital SWU remains unestablished; and (ii) a single-arm-compatible, benchmark-anchored ICWS capable of identifying clinical phenotypes most likely to benefit is lacking. To address these gaps, we conducted a prospective, single-arm interventional trial in consecutive tracheostomized PMV patients at a tertiary rehabilitation hospital SWU, evaluating 60-day weaning success against a pre-specified benchmark (p₀ = 65%) and developing a LASSO-Poisson ICWS deployed as a publicly accessible web calculator. We pre-specified a historical weaning success benchmark of 65.0% drawn from prospective specialized weaning-center cohorts and international consensus estimates [12, 27, 28].

Methods

Study design and setting

This was a prospective, single-arm, single-center interventional trial conducted at the SWU of the Department of Pulmonary Rehabilitation, Beijing Rehabilitation Hospital of Capital Medical University—a 950-bed tertiary rehabilitation hospital. The SWU operates as a 20-bed high-dependency unit (HDU) with continuous cardiorespiratory monitoring and full mechanical ventilation capability, to which patients requiring PMV are referred from the ICU of acute-care hospitals for structured weaning and rehabilitation.

The study protocol was reviewed and approved by the Ethics Committee of Beijing Rehabilitation Hospital of Capital Medical University (approval no. 2025bkky031) and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from each participant or from a legally authorized representative when the patient was temporarily unable to communicate. Reporting adheres to the extension of the TREND (Transparent Reporting of Evaluations with Nonrandomized Designs) statement, the reporting guideline for non-randomized interventional evaluations [29].

Participants

Between 1 January 2025 and 31 October 2025, all tracheostomized patients with PMV who were consecutively referred to the SWU for weaning rehabilitation were screened for eligibility. The enrollment window was fixed in advance as a 10-month calendar period. Recruitment ended at the pre-specified calendar boundary rather than upon attainment of a fixed sample size; all consecutively referred, eligible patients within the window were enrolled. PMV was defined, in accordance with the National Association for Medical Direction of Respiratory Care (NAMDRC) consensus, as the need for invasive mechanical ventilation for more than 6 h per day for at least 21 consecutive days [30].

Exclusion criteria were (1) referral intended solely for clinical stabilization prior to discharge to home or nursing care rather than for an active weaning attempt; (2) end-stage disease with life expectancy of less than 3 months; (3) death or discharge within 14 days after referral; and (4) refusal to provide informed consent (Fig. 1).

Fig. 1.

Patient flow diagram. NIV, noninvasive ventilator

Exclusion criteria (1), (2) and (3) were determined from the referral documentation and the admission assessment, before any study intervention was delivered. Patients meeting any of these three criteria were not admitted to the study. Patients meeting none of the above criteria provided written informed consent and were admitted to the SWU, and the standardized weaning protocol was started on the day of admission. Criterion (3) could not be determined at admission and was therefore assessed prospectively as each patient reached day 14.

Weaning protocol (intervention)

All enrolled patients received a standardized, stepwise weaning protocol embedded within an intensive pulmonary rehabilitation program (Fig. 2).

Fig. 2.

Standardized stepwise weaning protocol. (1) Step 1, Clinical stability: absence of organ failure, absence of active sepsis, and hemodynamic stability without vasoactive support. (2) Step 2, De-escalation of support: stepwise reduction toward PSV. (3) Step 3, SBT trigger and conduct: once mode is reduced to PSV with PSV + PEEP ≤16 cmH₂O, a 2-h SBT is performed on PSV with the cuff inflated (minimal support, not full disconnection). (4) Step 4, Stratification by SBT outcome: from SBT passers to HFOT via tracheostomy titrated to SpO₂ ≥95%, with defined objective/subjective failure criteria and daily extension of tolerated duration; from SBT failures to gradual PSV down-titration with repeated SBTs, and, if PSV de-escalation fails on 5 consecutive days, transition to a single-limb portable home ventilator. (5) Step 5, Step-down from HDU to the general ward. (6) Step 6, Decannulation per the department speaking-valve protocol. IMV, invasive mechanical ventilation; PSV, pressure-support ventilation; SBT, spontaneous breathing trials; HFOT, high-flow oxygen therapy; NIV, noninvasive ventilator; HDU, high-dependency unit

Step 1: Confirmation of clinical stability: absence of organ failure, absence of active sepsis, and hemodynamic stability without vasoactive support.

Step 2: Progressive de-escalation of ventilator support parameters.

Step 3: SBT: a 2-h SBT was initiated with the tracheal cuff inflated once the ventilator mode had been reduced to pressure-support ventilation (PSV) with PSV + PEEP ≤ 16 cmH₂O, which was consistent with the low-level pressure-support readiness criteria of international weaning recommendations [9, 27]. On reaching this threshold, a 2-h SBT was performed on pressure-support ventilation with the tracheal cuff inflated (minimal support rather than complete ventilator disconnection); pre-defined objective and subjective failure criteria mandated reconnection at prior settings.

Step 4: Stratified management by SBT outcome:

1. Patients passing the SBT underwent titration of HFOT. HFOT was delivered through the tracheostomy tube with active heated humidification. Flow was initiated at 50 L/min and adjusted within a pre-defined range of 30–60 L/min according to arterial blood gas results and comfort, with FiO₂ titrated on the basis of arterial blood gases and pulse oximetry to maintain SpO₂ ≥ 95% [31]. Because high-flow oxygen via tracheostomy has not been shown to accelerate liberation per se [16], it was used primarily for airway humidification and secretion clearance; flow was adjusted only to oxygenation, humidification, and comfort targets and was not reduced to satisfy discontinuation criteria, and weaning success was judged solely on the pre-defined objective and subjective failure criteria, independently of the flow setting. Pre-defined objective failure criteria (tachypnea, tachycardia, hyper- or hypotension, hypoxemia, acidosis, arrhythmia) or subjective failure criteria (agitation, depressed mental status, diaphoresis, increased work of breathing) mandated immediate reconnection to the ventilator at prior settings; this interval defined “basic HFOT tolerance time”, which was then extended daily until tolerance exceeded 16 h, after which continuous 24-h HFOT was attempted.

2. Patients failing the SBT underwent gradual PSV down-titration with repeated SBTs. If PSV de-escalation failed on 5 consecutive days, patients were transitioned from the ICU ventilator to a single-limb portable home ventilator (Stella 150, ResMed) connected directly to the tracheostomy tube through a single-limb circuit with an intentional-leak/exhalation port. Although this device class is most often used for mask-interface noninvasive ventilator, in our protocol it delivered ventilation through the tracheostomy; we therefore describe it as a noninvasive-type home ventilator delivered via the tracheostomy tube, not as NIV in the conventional (mask) sense. For brief description, we still referred to it as noninvasive-type home ventilator (NIV) delivered via the tracheostomy tube (single-limb circuit) delivered via the tracheostomy tube. It differs from the conventional dual-limb ICU ventilator principally in portability, leak-compensated single-limb operation, and its role in bridging patients toward home mechanical ventilator and the decannulation pathway.

Step 5: Step-down from HDU: patients were transferred from the HDU to the general ward after successful weaning or stabilization on NIV, followed by gradual NIV weaning during the ward stay.

Step 6: Decannulation: in patients who were fully weaned or required only nocturnal NIV, tracheostomy decannulation was considered according to our department’s speaking-valve-based protocol (which does not require tracheostomy tube capping), as previously described [32].

The protocol comprised standardized, non-discretionary decision nodes and pre-specified domains of bounded clinical individualization. The former included the sequential step progression, the SBT readiness threshold (PSV + PEEP ≤ 16 cmH₂O), SBT duration and failure criteria, and the transition rule to the portable home ventilator after 5 consecutive days of failed de-escalation. The latter included pre-specified domains of bounded clinical individualization, including HFOT flow titration within 30–60 L/min guided by arterial blood gases and comfort, the daily pace of pressure-support de-escalation, and the rate of HFOT tolerance time extension. Co-interventions (nutrition, mobilization, pharmacotherapy) were managed by the multidisciplinary team according to institutional clinical protocols. All enrolled patients completed the standardized stepwise sequence without deviation from the prescribed step order or mandatory SBT criteria.

Baseline assessments

At admission to the SWU, the following variables were systematically recorded: demographics (age, sex, body mass index); referral history (prior ICU length of stay, duration of invasive mechanical ventilation (IMV), time since tracheostomy); primary etiology of respiratory failure; comorbidities (heart, liver, and renal insufficiency; gastrointestinal bleeding; coagulopathy; sepsis; shock; delirium); prior use of extracorporeal membrane oxygenation (ECMO) or continuous renal replacement therapy (CRRT); sedative/analgesic and antipsychotic drug exposure; ventilator mode at referral; arterial blood gas parameters (pH, PaO₂, PaCO₂); nutritional and biochemical indices (Nutrition Risk Screening 2002 (NRS-2002), albumin, prealbumin, creatinine, blood urea nitrogen); severity-of-illness scores (Sequential Organ Failure Assessment (SOFA), Acute Physiology and Chronic Health Evaluation (APACHE) Ⅱ, Charlson Comorbidity Index (CCI)); and rehabilitation-related measures (Barthel Index, Standardized Five Questions (S5Q) score, diaphragm motility, diaphragm thickness, diaphragm thickening fraction (DTF), peak expiratory flow (PEF), Intensive Care Mobility Scale (IMS), Chelsea Critical Care Physical Assessment tool (CPAx)).

Outcomes

The primary outcome was the 60-day weaning success rate. Weaning was considered successful when a patient sustained unassisted spontaneous breathing, free of invasive ventilatory assistance, for at least 5 consecutive days within 60 days of enrollment. Patients who did not meet this criterion by day 60 were classified as weaning failures. Because eligibility required survival and retention in the unit for at least 14 days after referral, the primary outcome estimates the effectiveness of the protocol among patients who received an evaluable course of it. It is not an intention-to-treat estimate anchored at the point of referral, and it should not be read as such.

Secondary outcomes included weaning duration, defined as the interval from enrollment to the day of successful weaning; requirement for post-weaning NIV in the HDU; reconnection to invasive mechanical ventilation within 30 days after weaning; ongoing NIV use after hospital discharge; length of stay in the HDU; total hospital length of stay; and 6-month all-cause mortality, ascertained through scheduled telephone or WeChat follow-up.

Statistical analysis

A pre-specified historical control weaning success rate of 65% was derived from previously published cohorts of PMV patients undergoing structured weaning and from current clinical practice guidelines. The null hypothesis (H₀: p = 0.65) was fixed a priori in the protocol and trial registration (NCT06642714) before first enrollment. The value 65.0% was selected as the upper bound of the pragmatic 60–65% complete-weaning range reported for dedicated weaning units, so that our benchmark would be conservative (i.e., difficult to exceed) rather than self-serving. Assuming an anticipated absolute improvement of ≥ 15 percentage points, a minimum sample size of 225 patients was required to provide 95% power at a one-sided α of 0.05 (H₀: p = 0.65 vs H₁: p > 0.65). To accommodate potential loss to follow-up and protocol deviations, 250 consecutive eligible patients were scheduled to be enrolled.

All analyses were performed using R software (version 4.5.3; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were summarized as mean ± standard deviation or median (interquartile range) and compared across subgroups using Student’s t test or the Mann–Whitney U test, as appropriate. Categorical variables were summarized as n (%) and compared using Pearson’s χ² test or Fisher’s exact test.

Primary outcome analysis. Because of the single-arm superiority design, the observed 60-day weaning success rate was tested against the pre-specified historical control rate (p₀ = 65%) using a one-sided (H₁: p > p₀) exact binomial test as the principal inferential method. Three pre-specified sensitivity analyses were performed: (1) a continuity-corrected score-based normal-approximation test; (2) the Wilson two-sided 95% confidence interval for the observed rate; and (3) the Katz log-method relative risk (RR) of the observed rate versus p₀, with corresponding 95% confidence interval (CI) and one-sided p value.

Association between weaning success and 6-month mortality. Because the outcome was not rare, risk ratios rather than odds ratios were estimated. A modified Poisson regression model (Poisson log-link with Huber–White robust sandwich variance) was used to derive the crude RR of mortality as a function of weaning success. To address the confounding, two complementary balancing strategies were implemented via the ‘WeightIt’ package: (1) 1:1 nearest-neighbor propensity score matching (PSM) with a caliper of 0.2 SD on the logit of the propensity score; and (2) overlap weighting (OW) targeting the average treatment effect on the overlap population (ATO). Propensity scores were estimated through a generalized linear model that included all pre-specified baseline covariates. Covariate balance was quantified using absolute standardized mean differences (|SMD|), with |SMD|< 0.1 indicating adequate balance, and was visualized via Love plots. Adjusted RRs were obtained from weighted modified Poisson regression with robust standard errors.

ICWS and subgroup exploration. Given the modest sample size and the well-documented inflation of type-I error in conventional single-covariate subgroup analyses, a pre-specified data-driven strategy was used to characterize heterogeneity in weaning response. This approach is analogous to the personalized weaning score framework proposed in previous cohort studies but was adapted to the single-arm design by anchoring the model directly to the historical reference rate p₀. Because the study lacked a concurrent control group, the resulting score estimates the individualized calculated probability of weaning success under the study protocol relative to the historical benchmark p₀, rather than the causal treatment benefit; it was therefore termed the “ICWS” to avoid implying a treatment effect contrast that the single-arm design cannot support.

1. Variable selection was performed using LASSO-penalized Poisson regression with a fixed offset of log(p₀), so that the linear predictor Xβ directly quantifies the log-relative-risk of weaning success relative to the historical benchmark. The optimal penalty λ was selected by minimum cross-validated deviance over tenfold cross-validation.

2. Model refitting. The variables retained by LASSO were refit in an offset-Poisson regression [log(μ) = log(p₀) + Xβ] with Huber–White robust sandwich variance to obtain adjusted RRs and 95% CIs.

3. ICWS. For each patient, the ICWS was defined as the linear predictor Xβ (i.e., predicted link minus offset), such that exp(Xβ) represents the patient-specific RR of weaning success relative to p₀ = 65%. A score of 0 on the log scale (RR = 1) indicates an expected rate equal to p₀, positive values indicate above-benchmark score, and negative values indicate below-benchmark likelihood of success.

4. Tertile stratification. Patients were classified into tertiles of the ICWS (T1, lowest score; T3, highest score). The Cochran–Armitage trend test assessed the monotonic trend of weaning success across ordered tertiles. Within-tertile success rates were compared against p₀ by exact binomial tests, and corresponding RRs were derived by the Katz log-method.

5. Tool deployment. To support clinical implementation and external validation, the final weaning score algorithm was deployed as a publicly accessible web-based calculator.

All sensitivity analyses and the weaning score development were reported transparently; no formal adjustment for multiplicity was applied because the primary inference was based on a single pre-specified test.

Results

Patient disposition and baseline characteristics

A total of 301 tracheostomized patients with PMV were consecutively referred to the SWU; 250 met the eligibility criteria and were enrolled in the study. Follow-up at 6 months was complete for all 250 participants (Fig. 1).

Baseline demographic and clinical characteristics are summarized in Table 1. The mean age was 66.72 ± 14.87 years and 186 patients (74.4%) were male. The mean body mass index was 23.09 ± 4.01 kg/m². The cohort represented an advanced, late-referral PMV population: mean pre-referral ICU stay 57.38 ± 29.54 days, prior invasive ventilation 55.52 ± 27.60 days, and time since tracheostomy 43.70 ± 28.16 days. Common comorbidities included heart failure (46.8%), liver insufficiency (38.8%), gastrointestinal bleeding (38.4%), sepsis (34.8%), coagulopathy (33.6%), and shock (27.6%). The mean SOFA score was 4.74 ± 2.48, mean APACHE Ⅱ score 15.95 ± 5.66, and mean CCI 6.11 ± 2.62. ECMO and CRRT had been used during the prior ICU stay in 8.0% and 12.0% of patients, respectively.

Table 1.

Baseline characteristics overall and by weaning status/6-month mortality

	Total N = 250	Weaning No, N = 61	Weaning Yes, N = 189	p value	Mortality No, N = 214	Mortality Yes, N = 36	p value
Age (years)	66.72  ±  14.87	70.62 ± 15.98	65.46 ± 14.31	0.027	65.30 ± 14.57	75.17 ± 14.00	< 0.001
Male	186 (74.4%)	45 (73.8%)	141 (74.6%)	0.897	153 (71.5%)	33 (91.7%)	0.010
Body mass index (kg/m2)	23.09 ± 4.01	23.22 ± 4.29	23.04 ± 3.93	0.771	23.10 ± 3.82	23.03 ± 5.07	0.938
Heart failure	117 (46.8%)	40 (65.6%)	77 (40.7%)	< 0.001	87 (40.7%)	30 (83.3%)	< 0.001
Liver insufficiency	97 (38.8%)	31 (50.8%)	66 (34.9%)	0.027	74 (34.6%)	23 (63.9%)	< 0.001
Renal insufficiency	67 (26.8%)	29 (47.5%)	38 (20.1%)	< 0.001	44 (20.6%)	23 (63.9%)	< 0.001
Gastrointestinal bleeding	96 (38.4%)	32 (52.5%)	64 (33.9%)	0.009	69 (32.2%)	27 (75.0%)	< 0.001
Abnormal coagulation	84 (33.6%)	30 (49.2%)	54 (28.6%)	0.003	57 (26.6%)	27 (75.0%)	< 0.001
Sepsis	87 (34.8%)	31 (50.8%)	56 (29.6%)	0.003	58 (27.1%)	29 (80.6%)	< 0.001
Shock	69 (27.6%)	31 (50.8%)	38 (20.1%)	< 0.001	45 (21.0%)	24 (66.7%)	< 0.001
Delirium	41 (16.4%)	15 (24.6%)	26 (13.8%)	0.007	25 (11.7%)	16 (44.4%)	< 0.001
Primary disease*				0.497			0.235
Lung disease	75 (30.0%)	17 (27.9%)	58 (30.7%)		61 (28.5%)	14 (38.9%)
Heart disease	20 (8.0%)	6 (9.8%)	14 (7.4%)		19 (8.9%)	1 (2.8%)
Neuromuscular disease/spinal cord injury	45 (18.0%)	15 (24.6%)	30 (15.9%)		40 (18.7%)	5 (13.9%)
Brain injury	81 (32.4%)	18 (29.5%)	63 (33.3%)		72 (33.6%)	9 (25.0%)
Thoracoabdominal surgery	29 (11.6%)	5 (8.2%)	24 (12.7%)		22 (10.3%)	7 (19.4%)
Ventilator mode				0.023			0.356
AC	48 (19.2%)	19 (31.1%)	29 (15.3%)		38 (17.8%)	10 (27.8%)
SIMV+PSV	135 (54.0%)	27 (44.3%)	108 (57.1%)		117 (54.7%)	18 (50.0%)
PSV	67 (26.8%)	15 (24.6%)	52 (27.5%)		59 (27.6%)	8 (22.2%)
Pre-ICU-days	57.38 ± 29.54	56.66 ± 27.53	57.61 ± 30.23	0.818	58.07 ± 29.83	53.31 ± 27.79	0.352
Pre-IMV-days	55.52 ± 27.60	55.49 ± 28.65	55.52 ± 27.34	0.994	56.27 ± 28.08	51.06 ± 24.47	0.253
Pre-tracheostomy-days	43.70 ± 28.16	44.39 ± 28.49	43.48 ± 28.13	0.828	44.58 ± 28.80	38.47 ± 23.69	0.172
pH (unitless)	7.45 ± 0.06	7.44 ± 0.07	7.45 ± 0.06	0.140	7.45 ± 0.06	7.45 ± 0.07	0.871
PaO2 (mmHg)	94.00 (71.00,128.00)	92.00 (68.00,122.00)	94.00 (74.00,129.00)	0.475	95.00 (72.00,125.00)	85.50 (61.50,130.50)	0.518
PaCO2 (mmHg)	44.12 ± 11.46	44.70 ± 12.71	43.93 ± 11.05	0.669	44.36 ± 11.64	42.64 ± 10.32	0.367
Albumin (g/L)	34.78 ± 7.83	34.42 ± 7.50	34.90 ± 7.94	0.673	34.88 ± 8.02	34.18 ± 6.64	0.573
Prealbumin (g/L)	0.19 ± 0.11	0.20 ± 0.12	0.18 ± 0.11	0.277	0.19 ± 0.11	0.14 ± 0.07	< 0.001
Creatinine (μmol/L)	48.40 (37.70,65.00)	50.70 (36.90,69.30)	48.30 (37.90,62.90)	0.776	48.60 (37.90,64.10)	45.45 (36.40,71.45)	0.872
Blood urea nitrogen (μmol/L)	7.10 (5.20,10.50)	9.10 (5.80,12.90)	6.80 (4.80,9.50)	0.007	7.05 (5.00,10.20)	8.50 (6.05,12.15)	0.074
NRS-2002	4.67 ± 1.25	4.99 ± 1.14	4.57 ± 1.27	0.017	4.62 ± 1.24	4.97 ± 1.30	0.139
SOFA	4.74 ± 2.48	5.79 ± 2.66	4.41 ± 2.33	< 0.001	4.62 ± 2.44	5.47 ± 2.66	0.079
APACHE II	15.95 ± 5.66	18.57 ± 6.54	15.10 ± 5.09	< 0.001	15.68 ± 5.76	17.56 ± 4.85	0.041
CCI	6.11 ± 2.62	6.85 ± 3.08	5.87 ± 2.41	0.026	6.01 ± 2.66	6.72 ± 2.26	0.095
Diaphragm motility (cm)	0.83 ± 0.73	0.85 ± 1.28	0.83 ± 0.43	0.876	0.83 ± 0.77	0.85 ± 0.52	0.825
Diaphragm thickness (cm)	1.73 ± 0.71	1.67 ± 0.65	1.75 ± 0.73	0.389	1.71 ± 0.71	1.84 ± 0.67	0.287
DTF (%)	14.52 ± 8.12	13.57 ± 8.05	14.83 ± 8.13	0.292	14.56 ± 8.19	14.31 ± 7.77	0.860
PEF (L/min)	79.11 ± 40.89	65.72 ± 36.25	83.43 ± 41.45	0.002	78.35 ± 39.48	83.61 ± 48.81	0.543
Barthel Index	0.00 (0.00,0.00)	0.00 (0.00,0.00)	0.00 (0.00,0.00)	0.302	0.00 (0.00,0.00)	0.00 (0.00,0.00)	0.061
S5Q	3.06 ± 1.88	2.41 ± 1.98	3.28 ± 1.80	0.003	3.10 ± 1.86	2.83 ± 2.02	0.458
IMS	1.68 ± 1.86	1.21 ± 1.61	1.84 ± 1.91	0.014	1.68 ± 1.88	1.69 ± 1.72	0.969
CPAx	10.65 ± 8.04	8.13 ± 7.13	11.46 ± 8.17	0.003	10.64 ± 8.07	10.67 ± 7.97	0.988
Sedatives analgesics	121 (48.4%)	31 (50.8%)	90 (47.6%)	0.664	98 (45.8%)	23 (63.9%)	0.044
Antipsychotics	126 (50.4%)	30 (49.2%)	96 (50.8%)	0.827	102 (47.7%)	24 (66.7%)	0.035
ECMO	20 (8.0%)	14 (23.0%)	6 (3.2%)	< 0.001	4 (1.9%)	16 (44.4%)	< 0.001
CRRT	30 (12.0%)	15 (24.6%)	15 (7.9%)	< 0.001	13 (6.1%)	17 (47.2%)	< 0.001

AC assist/control ventilation, SIMV synchronized intermittent mandatory ventilation, PSV pressure-support ventilation, ICU intensive care unit, IMV invasive mechanical ventilation, NRS-2002 Nutrition Risk Screening 2002, SOFA Sequential Organ Failure Assessment, APACHE Acute Physiology and Chronic Health Evaluation, CCI Charlson Comorbidity Index, DTF diaphragm thickening fraction, PEF peak expiratory flow, S5Q Standardized Five Questions, IMS Intensive Care Mobility Scale, CPAx Chelsea Critical Care Physical Assessment tool, ECMO extracorporeal membrane oxygenation, CRRT continuous renal replacement therapy

*Brain disease/injury included hemorrhagic stroke, ischemic stroke, central nervous system tumor ( ±  post-opera), encephalitis, hypoxic-ischemic encephalopathy, Parkinson’s disease/parkinsonism, and traumatic brain injury; lung disease included pneumonia (bacterial/community-acquired pneumonia/hospital-acquired pneumonia/severe), aspiration pneumonia, viral pneumonia (incl. COVID-19), chronic obstructive pulmonary disease/acute exacerbation of chronic obstructive pulmonary disease, interstitial lung disease, lung cancer ( ±  post-opera), fungal/opportunistic pneumonia, and asthma; neuromuscular disease/spinal cord injury included spinal cord injury (incl. cervical), motor neuron disease/amyotrophic lateral sclerosis, Guillain–Barré syndrome, myopathy/myositis, multiple system atrophy, neuromyelitis optica spectrum disorders, and cranial neuritis; thoraco-abdominal/other surgery included orthopedic/spine surgery, cardiac surgery, thoracic surgery, head and neck surgery, trauma, and abdominal/other surgery; heart disease included ischemic/structural heart disease, heart failure, and post-cardiac-arrest

Primary outcome

Of the 250 enrolled patients, 189 (75.6%; Wilson 95% CI 69.9–80.5%) achieved successful weaning within 60 days, substantially exceeding the pre-specified historical control rate of 65% (Table 2). The superiority of the observed rate over p₀ was confirmed by all pre-specified analytic approaches. They included exact binomial test, 75.6% vs p₀ = 65%, one-sided p < 0.001; continuity-corrected score test, one-sided p < 0.001; Katz log-method RR of observed rate vs p₀ = 1.16 (95% CI 1.08–1.25; one-sided p < 0.001).

Table 2.

Primary and secondary outcomes overall and by weaning status/6-month mortality

	Total N = 250	Weaning No, N = 61	Weaning Yes, N = 189	p value	Mortality No, N = 214	Mortality Yes, N = 36	p value
Weaning	189 (75.6%)	–	–	–	172 (80.4%)	17 (47.2%)	< 0.001
Weaning days	–	–	19.86 ± 10.00	–	19.49 ± 9.86	23.59 ± 10.93	0.154
Post-NIV	–	–	90 (47.6%)	–	80 (46.5%)	10 (58.8%)	0.322
Post-IMV 30 days	–	–	4 (2.1%)	–	–	–	–
Post-NIV out	–	–	54 (28.6%)	–	47 (27.3%)	7 (41.2%)	0.131
Mortality	36 (14.4%)	19 (31.1%)	17 (9.0%)	< 0.001	–	–	–
HDU days	28.24 ± 18.74	47.52 ± 23.62	22.02 ± 11.28	< 0.001	26.02 ± 16.74	41.44 ± 24.14	< 0.001
Hospital days	51.02 ± 21.22	47.89 ± 23.36	52.04 ± 20.44	0.217	50.10 ± 19.99	56.50 ± 27.12	0.183

NIV noninvasive ventilator, IMV invasive mechanical ventilation, HDU high-dependency unit

The convergence of exact, asymptotic, interval-based, and ratio-based approaches supports a statistically robust and clinically meaningful improvement over the historical benchmark, equivalent to an absolute increase of 10.6 percentage points and a relative increase of approximately 16%.

Weaning success rates varied numerically across the five primary disease categories but without statistically significant heterogeneity (p = 0.497; Table 1), indicating that the protocol was broadly effective irrespective of the underlying etiology of respiratory failure.

Secondary outcomes

Among the 250 enrolled patients, the mean HDU length of stay was 28.24 ± 18.74 days and the total hospital length of stay was 51.02 ± 21.22 days (Table 2). Among the 189 successfully weaned patients, the mean weaning duration was 19.86 ± 10.00 days. Post-weaning NIV was required in 90 patients during the HDU stay (47.6% of successes); 54 patients (28.6%) continued NIV after hospital discharge. Only 4 weaned patients (2.1%) required reconnection to invasive mechanical ventilation within 30 days, demonstrating that the great majority of successful weans were durable.

Six-month all-cause mortality was 14.4% (36/250). Among these 36 patients, eight had formal treatment-limitation orders (do-not-resuscitate, do-not-escalate, or comfort-measures-only) documented during the study period or follow-up. Treatment-limitation orders were more prevalent in the weaning failure group (5/61, 8.2%) than in the weaning success group (3/189, 1.6%). The primary causes of death were severe pneumonia (n = 13), sepsis (n = 8) and cardiac events (n = 7), and other (n = 8). The mortality was lower among patients with weaning success than failure (9.0% vs 31.1%) (Table 2). Although the crude modified Poisson risk ratio was protective (RR 0.29, 95% CI 0.15–0.56), this association was substantially attenuated and no longer statistically significant after propensity score matching (RR 0.78, 95% CI 0.27–2.13) and after overlap weighting (RR 0.74, 95% CI 0.21–2.44) (Supplementary Fig. 1, Table 3). The relationship between weaning success and survival is therefore uncertain and should be regarded as hypothesis-generating.

Table 3.

Association between weaning success and 6-month mortality

	Number	Risk ratio (95% confidence interval)
		Crude	PSM	OW
Non-weaning	61	Reference (1.0)	Reference (1.0)	Reference (1.0)
Weaning	189	0.29 (0.15, 0.56)	0.78 (0.27, 2.13)	0.74 (0.21, 2.44)

PSM propensity score matching, OW overlap weighting

Individualized ICWS and heterogeneity of weaning success

To explore patient-level heterogeneity and identify clinical phenotypes most likely to benefit, LASSO-Poisson regression with log(p₀) as offset was applied to the 39 pre-specified baseline characteristics (Supplementary Fig. 2). At the minimum cross-validated deviance, 13 variables were retained in the model: ventilator mode at referral, NRS-2002 score, SOFA, APACHE Ⅱ, primary disease category, heart failure, renal insufficiency, shock, delirium, prealbumin, ECMO, PEF, and S5Q score.

The refit offset-Poisson model with robust sandwich variance yielded the individualized weaning score algorithm. Stratification by tertiles of the ICWS revealed a pronounced and monotonic gradient in weaning success (Fig. 3). The Cochran–Armitage trend test confirmed a highly significant monotonic increase in weaning success across ordered tertiles (p < 0.001). Notably, the weaning success rate approached universal success (98.8%) in the highest-benefit stratum but fell to 44.0%—well below the historical benchmark—in the lowest-benefit stratum, identifying a patient phenotype for whom the present protocol alone may be insufficient and in whom intensified or alternative strategies warrant prospective evaluation.

Fig. 3.

Forest plot of weaning success rate across tertiles of the individualized calculated weaning score, with the historical control (p₀ = 65%) as reference. N number, RR relative risk, CI confidence interval

To enhance the clinical interpretability of the ICWS tool, baseline information was compared among three groups of populations based on tertiles. Compared to the T1 population, those with higher ICWSs had a lower proportion of heart failure, heart insufficiency, liver insufficiency, renal insufficiency, gastrointestinal bleeding, sepsis, shock, delirium, and neuromuscular disease/spinal cord injury (primary disease); lower proportion of AC (ventilator mode), sedative/analgesic, ECMO and CRRT; and lower level of age, pre-ICU days, pre-IMV days, pre-tracheostomy days, creatinine, blood urea nitrogen, NRS-2002, SOFA, APACHE Ⅱ, and CCI. However, they exhibited a higher proportion of lung disease (primary disease) and SIMV+PSV (ventilator mode), and higher level of PaO₂, albumin, PEF, S5Q, and CPAx (Supplementary Fig. 3).

To facilitate exploratory clinical use and future external validation, the weaning score algorithm was implemented as a web-based calculator at https://jianghongying.shinyapps.io/Weaning-for-PMV/ (Supplementary Fig. 4). This tool was intended for research and hypothesis-generating purposes only and should not be used for clinical decision-making until externally validated.

Discussion

In this prospective single-arm interventional trial, a standardized stepwise protocol was associated with a 60-day weaning success rate that exceeded a pre-specified historical benchmark. Because the study lacked a concurrent control group, differences in case mix, referral criteria, institutional capacity, rehabilitation intensity, and outcome definitions cannot be fully excluded, and the results should not be interpreted as demonstrating superiority over usual care. Successful weaning was linked to lower 6-month mortality, although this association was attenuated after causal-inference adjustment. An ICWS resolved the cohort into phenotypes with widely divergent probabilities of liberation. Collectively, these findings extend prior work in three respects: they provide prospective, protocolized evidence from a Chinese rehabilitation hospital SWU; situate the weaning-mortality relationship within a formal causal-inference framework; and introduce a single-arm-compatible, benchmark-anchored metric of individualized benefit.

The observed 60-day success rate of 75.6% lies at the upper end of international PMV experience, against substantial between-center heterogeneity. North American LTACHs report consecutive-cohort success of approximately 54% [13], and Jubran et al.’s randomized trial yielded 43–57% at a median weaning time of 15 days [33]. European outcomes are higher but variable: the German WeanNet cohort of more than 11,000 patients achieved ~62%, stronger in COPD-predominant and weaker in neuromuscular phenotypes [23]; Italian centers integrating rehabilitation and home-ventilation pathways report 63–78% [34]; and UK referral populations approximate 64.0% [28]. Damuth et al.’s systematic review across 124 cohorts documented a 30–85% range, attributing heterogeneity chiefly to case mix, rehabilitation integration, and outcome definitions [3]. Three considerations apply. First, the cohort’s severity profile matched or exceeded that of Jubran’s LTACH population, rendering a lighter case mix an unlikely explanation. Second, the pre-specified benchmark p₀ = 65% aligns with the pragmatic range endorsed by the ERS/ATS task force on prolonged weaning [27] and the pooled estimates of Damuth et al. [3], and is therefore not a conservative reference. Third, success did not differ significantly across primary disease categories, in contrast to the etiology-dependent gradient of WeanNet [23]; this likely reflects cohort homogeneity in referral timing and tracheostomy status, with residual patient-level variation absorbed by the ICWS. Integrated respiratory, rehabilitative, and nutritional care over a long treatment horizon (mean total stay 51 days) may further account for the favorable performance. Furthermore, comparison against a historical benchmark is limited by between-study differences in patient selection, referral thresholds, weaning unit organization, decannulation practice, and health-system context. The 65% benchmark should be read as a pragmatic reference point rather than a precise counterfactual for our specific setting. Meanwhile, our study was conducted in a single 20-bed high-dependency weaning unit within a 950-bed tertiary rehabilitation hospital, staffed by a dedicated multidisciplinary weaning and pulmonary-rehabilitation team. The relatively long average ICU stay before referral, the longer intervals after invasive ventilation and after tracheostomy place our patients at the chronic-critical-illness end of the PMV spectrum. Our findings therefore apply to patients who have survived the acute phase and are referred for dedicated weaning and rehabilitation; they should not be extrapolated to acute-ICU patients earlier in their course or to those with shorter ventilation, in whom baseline weaning probability and competing risks differ, pending multicenter evaluation. Because prolonged mechanical ventilation and chronic critical illness constitute a large and resource-intensive population, defining the subgroup to which effectiveness estimates apply is essential. Within this context, our data thus address an Asian evidence gap previously dominated by retrospective single-center reports.

Durability matters because weaning success is vulnerable to premature declaration. Only 2.1% of successfully weaned patients required reconnection to invasive ventilation within 30 days, well below the 10–20% reintubation rates of general ICU populations [12, 35], the 72-h rates of 22.8% (HFNC) and 19.1% (NIV) after high-risk extubation [19], the 7-day rates of 11.8% (HFNC+NIV) and 18.2% (HFNC alone) in HIGH-WEAN [20], and the 13.6% post-SBT rate of WEAN SAFE [12]. This comparison warrants caution: the retained tracheostomy provides a safety corridor absent after extubation, and the event here is reconnection rather than reintubation. Accordingly, direct numerical comparison between post-extubation reintubation rates in natural-airway populations and post-weaning reconnection rates in tracheostomized patients should be interpreted with caution, as the populations, airway access, and failure mechanisms differ fundamentally. The mean weaning duration of 19.9 ± 10.0 days falls within the international mid-range defined by Jubran (median 15 days) [33], WeanNet (18–22 days) [23], and Italian weaning centers (16–35 days) [36]. Noninvasive support was often still required after liberation: 47.4% of successfully weaned patients received transitional NIV during the HDU stay and 28.0% continued NIV after discharge. Rather than protocol incompleteness, this reflects the contemporary conception of weaning as staged de-escalation. Liberation in PMV may therefore be better characterized along a gradient from freedom from invasive ventilation, to freedom from all positive-pressure support, to functional respiratory independence, strata with distinct implications for quality of life and long-term care.

Six-month all-cause mortality was 14.4%, well below the 40–60% medium-term mortality typically reported in PMV and LTACH cohorts [3, 4, 21], and differed substantially between successful and failed weaning (9.0% vs 31.1%). The crude modified Poisson RR of 0.29 attenuated to 0.78 after PSM and 0.74 after OW, with point estimates remaining protective but confidence intervals crossing unity. Two caveats constrain causal interpretation: the limited event count restricts the stability of matched and weighted adjustment, and weaning success is time-dependent, since patients must survive long enough to be classified as successful, introducing healthy-survivor and immortal-time asymmetries. An additional consideration is the role of treatment-limitation decisions in shaping the observed mortality difference. Patients who fail to wean from prolonged mechanical ventilation face a trajectory of chronic critical illness with poor functional prognosis, which may prompt clinicians and families to transition goals of care from curative to palliative or comfort-oriented management. The treatment-limitation orders, do-not-resuscitate directives, or the withdrawal of life-sustaining therapies was disproportionately implemented in the weaning failure group; the observed higher mortality in this group may partly reflect these decisions. This finding does not invalidate the association between weaning failure and mortality, but it adds an important layer of complexity: the relationship is likely bidirectional, with disease severity driving both weaning failure and treatment-limitation decisions, which in turn influence mortality. The mortality analysis is therefore biologically plausible but should be interpreted cautiously and regarded as hypothesis-generating rather than a definitive causal estimate. A concurrently controlled study with time-appropriate modeling of exposure is needed to test this hypothesis.

The ICWS moves beyond population-averaged conclusions. Because single-covariate subgroup analyses are prone to false positives and poor reproducibility, the PATH Statement advocates risk- or benefit-based multivariable stratification [37, 38]. The 13 retained variables partition into four biologically coherent domains: disease severity, cardio-renal reserve, respiratory mechanics and airway clearance, and cognition-nutrition-engagement. Clinically, the T1 stratum identifies a phenotype for whom the standard protocol appears insufficient and in whom intensified or alternative strategies warrant evaluation, including prolonged or home NIV [36], diaphragm-protective ventilation or pacing [39, 40], and early palliative-oriented shared decision-making [41]. Conversely, near-ceiling T3 performance raises whether stay duration, decannulation timing, and step-down criteria can be optimized in high-benefit patients. To facilitate translation, the tool has been deployed as an accessible calculator. Three time windows appear informative: referral, for SWU triage and anticipatory discussion in very low-benefit patients; the 72-h reassessment, for aligning rehabilitation intensity, SBT frequency, HFOT/NIV sequencing, and nutritional targets; and weekly re-evaluation, for tracking modifiable variables such as PEF, S5Q, prealbumin, and SOFA. Unlike APACHE or ProVent, which predict mortality, this tool targets weaning success in tracheostomized PMV patients and is anchored to a historical benchmark. To facilitate translation, the tool has been deployed as a publicly accessible calculator for research use. We emphasize that the ICWS has undergone internal evaluation only and has not been subjected to external validation, formal calibration assessment, or decision-curve analysis. In accordance with TRIPOD+AI and PROBAST quality standards [42, 43], it is presented as an exploratory research-support instrument. Clinicians and investigators are encouraged to test and refine it in independent, multicenter cohorts before considering integration into clinical workflows.

Strengths and limitations

Strengths of this study include prospective consecutive enrollment, complete 6-month follow-up, a pre-specified historical benchmark with triangulated statistical testing, a severely deconditioned population comparable to or heavier than international PMV cohorts, and dual reporting of protocol effectiveness and patient-level heterogeneity within one analytical framework.

Several limitations should be acknowledged. First, because this was a single-arm interventional study benchmarked against a pre-specified historical control rather than a concurrent randomized comparator, our findings should be interpreted as evidence of feasibility and of an association between protocolized care and weaning success—not as proof that the protocol is superior to usual care. Residual confounding, case-mix differences, secular or referral-pattern differences between our cohort and the historical benchmark population, and the single-center rehabilitation setting all constrain causal and generalizable inference. Especially, the 65% historical benchmark was derived from a synthesis of studies conducted predominantly in North American LTACHs, European regional weaning centers, and international consensus panels. These settings differ from the Chinese tertiary rehabilitation hospital model in several respects, including referral pathways and timing, payer structure and length-of-stay constraints, multidisciplinary team composition, rehabilitation intensity and modalities, decannulation practice, and post-discharge support infrastructure. Consequently, the 65% rate may not constitute an equally valid benchmark for all healthcare systems or for all types of specialized weaning units. The benchmark should therefore be read as a pragmatic, internationally anchored reference point rather than a precise counterfactual for any specific setting. Confirmation in a multicenter, concurrently controlled study is required before the protocol can be recommended for routine adoption. Second, patients who died or were discharged within 14 days of referral were excluded under a pre-specified criterion. The analyzed cohort had therefore, by construction, survived and remained in the unit for at least 14 days. The primary outcome should accordingly be read as the effectiveness of the protocol among patients who received an evaluable course of it, rather than as an estimate anchored at referral. Because patients who died or were discharged within 14 days of referral were excluded before the final enrollment, our cohort is inherently survivor-selected relative to the full population referred for weaning consideration; this selection likely favors a higher observed weaning success rate and lower observed mortality than would be seen in an unselected referral population, a consideration relevant to comparison with the historical benchmark. Third, male patients comprised 74.4% of the cohort, so generalizability to female-predominant populations requires confirmation. Fourth, the mortality analysis is constrained by limited events, unmeasured functional-reserve confounders, and healthy-survivor asymmetry. Fifth, the ICWS has undergone internal tertile-based evaluation but not temporal or external assessment. Finally, because the protocol was evaluated as an integrated package, the incremental contributions of SBT, HFOT, and NIV cannot be disentangled without factorial or stepped-wedge designs.

Conclusions

A standardized stepwise weaning protocol delivered in a tertiary rehabilitation hospital SWU was feasible in tracheostomized patients with prolonged mechanical ventilation and was associated with a 60-day weaning success rate that exceeded a pre-specified historical benchmark. The ICWS delineated phenotypes ranging from near-universal success to sub-benchmark performance. These single-arm findings are promising but require confirmation in multicenter, concurrently controlled studies before the protocol can be recommended for routine practice.

Supplementary Information

Additional file1 ^{(33MB, tif)}

Additional file2 ^{(33MB, tif)}

Additional file3 ^{(33MB, tif)}

Additional file4 ^{(1.9MB, tif)}

Abbreviations

PMV

Prolonged mechanical ventilation

ICU

Intensive care unit

SBT

Spontaneous breathing trials

HFOT

High-flow oxygen therapy

NIV

Noninvasive ventilator

SWU

Specialized weaning units

LTACH

Long-term acute-care hospitals

ICWS

Individualized calculated weaning score

HDU

High-dependency unit

NAMDRC

National Association for Medical Direction of Respiratory Care

PSV

Pressure-support ventilation

AC

Assist/control ventilation

SIMV

Synchronized intermittent mandatory ventilation

IMV

Invasive mechanical ventilation

ECMO

Extracorporeal membrane oxygenation

CRRT

Continuous renal replacement therapy

NRS-2002

Nutrition Risk Screening 2002

SOFA

Sequential Organ Failure Assessment

APACHE

Acute Physiology and Chronic Health Evaluation

S5Q

Standardized Five Questions

CCI

Charlson Comorbidity Index

DTF

Diaphragm thickening fraction

PEF

Peak expiratory flow

IMS

Intensive Care Mobility Scale

CPAx

Chelsea Critical Care Physical Assessment tool

RR

Relative risk

CI

Confidence interval

PSM

Propensity-score matching

OW

Overlap weighting

SMD

Standardized mean difference

Author contributions

CZ, TZ, and HJ designed the research. CZ, RZ, JG, QL, BY, BZ, and JW collected the data and conducted the data analysis and interpretation. CZ, RZ, and HJ drafted the initial manuscript. TZ and HJ participated in the review and revision of the manuscript. All the authors approved the final draft and agreed to be accountable for all aspects of the work.

Funding

No external funding was received for this work.

Data availability statement

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval

The study protocol was reviewed and approved by the Ethics Committee of Beijing Rehabilitation Hospital of Capital Medical University (approval no. 2025bkky031) and was conducted in accordance with the Declaration of Helsinki.

Consent to participate

Written informed consent was obtained from each participant or from a legally authorized representative when the patient was temporarily unable to communicate.

Consent for publication

Written informed consent for publication was obtained from all the participants.

Competing interest

The authors declare that they have no competing interests.