This was a multicenter, prospective cohort study. Participants were neurosurgical critically ill patients admitted between May 2024 and April 2025 to the general ICUs of six tertiary hospitals in Shanghai. All six hospitals were regional medical centers, ensuring representativeness and heterogeneity of the study population. Each center independently recruited patients using convenience sampling, with screening based on unified inclusion and exclusion criteria.

Inclusion criteria were: (1) admission to ICU with GCS < 12; and (2) age ≥18 years. Exclusion criteria were: (1) factors such as trauma fixation or skin conditions affecting diaphragmatic motion or ultrasound assessment; (2) invasive or noninvasive ventilation before admission; and (3) pregnancy or lactation. Elimination criteria were: (1) duration of invasive ventilation <24 h during the study; and (2) patient or family request to withdraw from the trial.

Patients were allocated sequentially according to admission time into three phases with distinct weaning management processes:

Sequential allocation by time avoided overlap of multiple processes in the same period and ensured integrity and feasibility of each intervention. All centers followed a unified protocol for patient screening, data collection, and process execution to guarantee cross-center comparability and consistency.

First weaning success rate: defined as the proportion of patients who, after their first SBT and extubation, did not require reintubation, invasive or noninvasive ventilation, and did not die within 48 h, used to evaluate the direct impact of optimized processes on weaning success.

To minimize operator-dependent variability across centers, diaphragm ultrasound was performed by ICU nurses certified in critical care ultrasound at each site. Prior to study initiation, all operators completed standardized training on probe positioning, image acquisition, and measurement definitions for DTF and DE, using a unified operating manual with step-by-step instructions. During data collection, measurements were obtained under the same ventilator settings and patient position (semi-recumbent, head of bed ~30°). For each parameter, multiple consecutive breaths were recorded and averaged (DE: three measurements; DTF: 3–5 respiratory cycles), and images were required to meet prespecified quality criteria (clear diaphragm interface and stable M-mode tracing). If image quality was inadequate or measurements were inconsistent across breaths, acquisition was repeated. Each center conducted routine internal checks of stored images and measurements by a designated senior ultrasound-trained nurse to ensure protocol adherence.

All data were analyzed using SPSS 26.0. Continuous variables were tested for normality. Normally distributed data were expressed as mean ± SD and compared among groups using one-way ANOVA; pairwise comparisons were conducted with LSD (homogeneity of variance) or Tamhane’s T2 (heterogeneity). Non-normally distributed data were expressed as median (IQR) and compared among groups using the Kruskal–Wallis test. Categorical variables were presented as n (%) and compared among groups using chi-square or Fisher’s exact test. All tests were two-sided with significance set at p < 0.05.

Group comparisons of first weaning success were performed using chi-square tests. ROC curves were plotted to evaluate the predictive value of diaphragm ultrasound parameters for first weaning and first SBT success, with AUC calculated. Consistency between first SBT and first weaning outcomes was analyzed using Kappa statistics, with significance set at p < 0.05.

This study was approved by the Ethics Committee of Shanghai General Hospital (Approval No. YL-2025-078) and by the institutional review boards of all participating centers. The study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from the patients or their legal representatives prior to enrollment.

There were no statistically significant differences among the three groups in demographic and clinical baseline characteristics, including age (p = 0.572), sex (p = 0.567), and primary diagnosis distribution (p = 0.945). Before SBT, mean arterial pressure (p = 0.989), PEEP (p = 0.109), tidal volume (p = 0.388), heart rate (p = 0.331), and body temperature (p = 0.918) did not differ significantly among the groups. Regarding diaphragm function assessed before SBT, end-inspiratory diaphragm thickness (p = 0.815), end-expiratory diaphragm thickness (p = 0.692), diaphragm thickening fraction (p = 0.386), and diaphragmatic excursion (p = 0.082) were also comparable across the three groups (Table 1).

Within 48 h after SBT, there was no significant difference among the three groups in the first SBT success rate (χ² = 4.133, p = 0.127) or the first extubation success rate (χ² = 0.211, p = 0.900). However, the incidence of reintubation differed significantly, with lower rates in the DTF and DTF + DE groups compared with the control group (χ² = 10.981, p = 0.004). Short-term survival within 48 h showed no significant difference across groups (χ² = 0.151, p = 0.927). Mechanical ventilation (MV) duration within 48 h was significantly shorter in the DTF and DTF + DE groups compared with controls (F = 8.509, p < 0.001).

At 14 days after SBT, survival remained comparable among the three groups (χ² = 0.151, p = 0.927). By contrast, the total MV duration differed significantly, with markedly shorter times observed in the DTF and DTF + DE groups than in the control group (F = 9.025, p < 0.001) (Table 2).

In the control group, DE and DTF were used as test variables, with weaning failure as the state variable, to construct ROC curves. As shown in Figure 1, the AUC for DE was 0.760 (p < 0.01, 95% CI: 0.639–0.881), with a standard error of 0.062. The AUC for DTF was 0.825 (p < 0.01, 95% CI: 0.711–0.938), with a standard error of 0.058. The combined ROC curve of DE and DTF yielded an AUC of 0.830 (p < 0.01, 95% CI: 0.722–0.939), with a standard error of 0.056 (Figure 1).

In the control group, among patients who passed SBT, 41 achieved successful weaning while 9 failed; among those who did not pass SBT, 12 did not achieve successful weaning. In the DTF group, among patients who passed SBT, 40 achieved successful weaning and 2 failed; among those who did not pass SBT, 22 did not achieve successful weaning. In the DTF + DE group, among patients who passed SBT, 40 achieved successful weaning and only 1 failed; among those who did not pass SBT, 20 did not achieve successful weaning.

Consistency analysis of first SBT results and final weaning outcomes is shown in Table 1. The Kappa value was 0.638 (p < 0.01) in the control group, 0.932 (p < 0.01) in the DTF group, and 0.963 (p < 0.01) in the DTF + DE group. All groups demonstrated strong consistency between first SBT results and weaning outcomes, with higher Kappa values in the DTF and DTF + DE groups compared with the control group (Table 3).

ROC analysis in this study showed that DTF achieved an AUC of 0.825 for predicting weaning failure, superior to DE (AUC = 0.760), while their combination yielded an AUC of 0.830 with limited incremental value. This finding suggests that DTF alone provides strong predictive efficacy, whereas DE, though valuable, offers only marginal additive benefit. From the perspective of bedside feasibility, a DTF-based stratification strategy is simpler, maintaining discriminative power while allowing rapid implementation by nurses in routine workflows.

Previous studies have mostly focused on single parameters. DiNino et al. first reported high sensitivity and specificity of DTF, with cutoffs commonly between 20 and 30% (9). The systematic review and meta-analysis conducted by Elfeky et al. further demonstrated that DTF achieved a higher overall AUC for predicting extubation success compared with traditional respiratory mechanics parameters (10). Using 20% as a cutoff in this study produced consistent results, supporting its applicability in nurse-led practice within neurosurgical ICUs. In contrast, DE showed lower predictive accuracy, likely influenced by imaging windows, probe angles, and patient body habitus, thus requiring greater operator experience. Jesus et al. reported lower reproducibility of DE compared with thickness measurements, aligning with the limited added value of DE observed here (11).

It is noteworthy that although DE alone is inferior to DTF, combined assessment may strengthen confidence in borderline cases. For instance, when DTF approaches 27%, DE ≥ 1.3 cm may indicate sufficient diaphragmatic motion to support extubation (12). Thus, despite limited statistical gain, DE may still serve as supplementary information in uncertain individual cases.

Physiologically, DTF reflects relative fiber thickening during diaphragmatic contraction, closely approximating endurance and workload, which explains its stable predictive performance (13). DE reflects displacement amplitude, influenced by chest wall compliance, lung volume, and inspiratory drive, making interpretation more complex in patients with impaired compliance or altered neural drive (14). This may explain why DE was less predictive than DTF in neurosurgical populations. Previous studies also suggested that DTF provides independent discrimination of weaning tolerance, whereas DE requires integration with other indices (15). Therefore, DTF is better suited as the primary stratification tool, while DE may serve as an adjunct in complex cases, especially for identifying insufficient drive or restricted motion.

This study indicates that diaphragm-based stratification can shift weaning from maximizing early SBT “pass rates” to improving post-extubation stability. First SBT success was numerically highest in the control group, although the between-group difference was not statistically significant, consistent with a more permissive conventional 30 min SBT. In contrast, reintubation was markedly lower in both diaphragm-guided groups (<10%) than in controls (25.81%; p < 0.05), suggesting fewer delayed failures after an apparently successful SBT. Although first weaning success did not differ significantly, agreement between initial SBT assessment and subsequent weaning outcome improved (Kappa 0.638 in controls vs. 0.932 and 0.963 in the DTF and DTF + DE groups), supporting better decision consistency. Importantly, total mechanical ventilation duration was shorter in the diaphragm-guided groups, implying that preventing downstream setbacks (e.g., reintubation and repeated weaning attempts) outweighed any delays from selective prolonged observation.

A plausible mechanism is a reduction of false positives. Conventional SBT criteria mainly reflect short-term respiratory and hemodynamic tolerance and may not capture respiratory muscle endurance; therefore, some borderline patients can tolerate 30 min yet fail after extubation because of delayed fatigue, representing false-positive SBT results (7). Mezidi et al. similarly noted that short SBTs may overestimate weaning capacity, whereas longer observation can unmask fatigue that emerges later (16). In our protocol, prolonged observation was applied selectively to patients below diaphragm thresholds, rather than universally, aiming to identify delayed intolerance without imposing delays on low-risk patients.

Stratification may, however, increase false negatives: extubation-capable patients may show transient instability during prolonged observation and be deferred unnecessarily (17). This trade-off likely contributes to the absence of a significant increase in first weaning success. From a safety standpoint, the clinical penalty of false negatives is mainly delayed liberation, whereas false positives lead to extubation failure and reintubation. Consistent with this risk asymmetry, Cheng et al. reported that extubation failure carries a worse prognostic impact than moderate prolongation of ventilation (18). Finally, Mahmoodpoor et al. found DTF to outperform conventional respiratory mechanics in predicting weaning outcomes (19); our prospective, multicenter findings provide convergent evidence that a DTF-based, nurse-performed bedside assessment can support a targeted, safety-oriented weaning strategy.

In neurosurgical ICUs, the organization of weaning differs from general ICUs. Physicians are often occupied with rounds, surgery, and outpatient duties, limiting their bedside availability, while readiness assessment and SBT initiation primarily fall to nurses. Following sedation interruption, rapid changes in consciousness, hemodynamics, and respiration require immediate nursing judgment. Incorporating diaphragm ultrasound into routine nursing assessments allows simultaneous functional monitoring and SBT screening within daily workflow, fitting the neurosurgical ICU model and addressing the demand for precision weaning.

This integration shifts the nursing role. Traditionally, nurses focused on monitoring and execution, relying on vital signs and experience, with limited ability to quantify respiratory muscle function (20). This study showed that nurse-led DTF measurement achieved predictive AUC (0.825) comparable to physician-led reports, and Kappa analysis demonstrated significantly improved agreement between SBT and extubation (21). This indicates that with standardized training, nurses can provide objective evidence equivalent to physicians, shifting weaning management from “experience-based” to “data-driven.”

The value of nursing also lies in advancing readiness assessments and improving efficiency. By performing diaphragm assessment and SBT immediately after sedation interruption, nurses minimized delays from waiting for physician rounds. Although stratification lowered first SBT success, it reduced reintubation and shortened cumulative ventilation, reflecting that early nursing intervention improved both stability and safety.

The nurse–physician collaboration model also evolved. Nurses now provide objective diaphragm parameters rather than solely reporting vital signs, forming a “nurse-trigger, physician-confirm” workflow that streamlined decision-making and strengthened team trust. Hirzallah et al. emphasized that nurses should play an active rather than passive role in weaning, and the findings here support this positioning (22).

From a disciplinary perspective, diaphragm ultrasound as a skill independently mastered by nurses provides new technical advantages in critical care. Historically physician-led, this evaluation was shown here to be reliably performed by nurses, producing reproducible data with direct clinical relevance. Many studies have proposed dynamic diaphragm thickness as a reliable endurance marker (10, 23). This study further demonstrates that when performed by nurses, such assessment directly influenced extubation outcomes. This practice strengthens nursing authority in respiratory monitoring and weaning, highlighting the unique value of nursing within multidisciplinary collaboration.

This study has limitations. Group allocation was time-period based rather than randomized, so temporal bias (including seasonal variation such as respiratory infection prevalence) cannot be excluded. The diaphragm-guided phases used an extended observation window for patients below ultrasound thresholds; therefore, some benefit may reflect longer observation itself rather than DTF/DE stratification alone. Sedative exposure was not quantified across groups; although assessments were initiated only after patients reached a comparable arousal level (RASS ≥ − 2), residual confounding may remain.