This study employed a combined observational derivation and randomized validation approach to establish and evaluate predictive models for the depth of DLT insertion (Figure 1). Initially, a total of 98 patients scheduled for thoracic surgery involving one-lung ventilation were screened for participation. Sixteen patients were subsequently excluded because DLT was not performed, resulting in 82 participants included in the final analysis. This phase led to the development of three predictive formulas for DLT insertion depth: based on standing height (Depth = 1.083 + 0.166 × standing height), sitting height (Depth = 0.32 × SH), and an integrated model combining body height and sternum-carina distance (Ds-c: Depth = 1.543 + 0.155 × standing height + 0.202 × Ds-c).

The subsequent phase involved a triple-arm randomized controlled trial (RCT) with 336 patients to compare these predictive models. Sample size was calculated to detect a 0.3 cm error reduction with 90% power (α=0.05), requiring 102/group; 112/group were enrolled accounting for 10% attrition. The random allocation ratio was 1:1:1, ensuring equal distribution among three groups: standing height (height-based formula), sitting height (sitting height-based formula), and CT (CT-guided formula). Prespecified primary endpoints were: (1) Absolute error (continuous; mean difference in cm), (2) Clinical accuracy (binary; proportion with optimal positioning). Secondary outcomes included tube mispositioning rates.

CT Measurement Methodology Thoracic CT scans were performed according to the hospital's standard diagnostic protocol and analyzed using OsiriX imaging software. Various anatomical measurements obtained from CT scans were used to predict intubation depth, including the distances from the vocal cords to the carina (Dv-c), from the cricoid to the carina (Dc-c), and from the thoracic inlet to the carina (Ds-c). Although the Dv-c measurement demonstrated the highest predictive accuracy at 90%, its clinical applicability was limited in our control group for two key reasons23. First, the standard lung CT protocol did not include imaging of the vocal cords, leading to insufficient data for this measurement in most patients. Second, marking the catheter at the glottis level prior to intubation was essential for observational purposes but did not interfere with the anesthesia induction process. Consequently, Ds-c was ultimately selected as the primary measurement for this study due to its superior predictive accuracy.

DLT Positioning Criteria The optimal positioning of the DLT was defined as having the bronchial cuff located in the main bronchus, with its edge positioned within 0.5 cm below the carina or at the carina level to ensure unobstruction of the opposite main bronchus. Additional criteria specified that the catheter tip should not obstruct the lobar bronchus opening, and the side hole should be situated above the carina, directly facing the main bronchus opening on the contralateral side. Surgeons confirmed satisfactory lung collapse on the operative side. Fiberoptic bronchoscopy was utilized to assess the optimal placement of the DLT, with evaluations conducted in the supine position after induction and again in the lateral position prior to the procedure. In the observational cohort, intubation was performed first, followed by verification. In the validation cohort, an independent physician calculated the insertion depth based on the predictive formulas. The anesthesiologist, unaware of group assignments, inserted the DLT according to the calculated depth, which was subsequently assessed by another independent physician using fiberoptic bronchoscopy. The DLT position was recorded as accurate if it fell within the optimal range; otherwise, it was deemed inaccurate.

Establishment of Prediction Formulas In the derivation cohort, standing height, sitting height, and Ds-c were documented for all participants prior to surgery. Following the induction of general anesthesia, patients were positioned for surgery, and one-lung ventilation commenced. The investigator recorded the catheter type and depth after the surgeon approved lung collapse on the operated side, and the anesthesiologist confirmed the catheter's optimal position using fiberoptic bronchoscopy. Regression analysis was employed to derive the prediction formulas based on height and CT measurements Validation Phase Protocol The RCT phase randomized 336 eligible patients equally into three groups using a computer-generated allocation: standing height (height-based formula), sitting height (sitting height-based formula), and CT (CT-guided formula). Preoperative standing height/sitting height measurements, as well as CT analysis, were completed before randomization. Anesthesiologists accessed group assignments through sealed opaque envelopes after patient positioning. All intubations were performed by senior airway specialists using manufacturer-recommended DLT sizes, with initial placement at the formula-predicted depth. After anesthesia induction, patients were positioned laterally to ensure alignment of the head and neck with the spine. An additional observer, blinded to group assignments, used fiberoptic bronchoscopy through the side hole of the bronchial tube to assess the location of the bronchial cuff. The target was to achieve the 'predicted depth.' Data were considered accurate if the depth fell within the 'optimal position' range. If the 'predicted depth' was not attained but a safe critical point was reached, adjustments to the catheter position were halted, and the data were recorded as inaccurate. Depths were classified as shallow if below the 'optimal position' and deep if exceeding it. All intubating anesthesiologists were board-certified specialists with >200 documented DLT placements and current certification in advanced airway management.

Blinding Procedures Participants, outcome assessors, data analysts, and data collectors were fully blinded to group allocation, while intubating clinicians were masked to group assignment and prediction formulas. Independent outcomes assessment was ensured through a three-level blinding system: (1) intubating clinicians received only numerical insertion depths from an independent coordinator without access to prediction methods or group information; (2) fiberoptic bronchoscopy verification used video recordings anonymized by removing group and patient identifiers; and (3) data collection was performed by research nurses unaware of allocations while data analysts processed datasets stripped of randomization codes. Identical interventions were maintained across all groups through standardized DLT models, uniform intubation equipment, and consistent procedural protocols.

Harm Monitoring Procedure-related adverse events specifically attributable to DLT placement (including mucosal injury and hypoxemia defined as SpO₂ <90% persisting >30 seconds) were assessed during standardized bronchoscopic verification. No study-related complications occurred.

Statistical Analysis No interim analyses were planned or performed given low procedural risk and fixed enrollment target. Stopping rules would trigger if >10% participants experienced protocol-defined serious adverse events (e.g., bronchial rupture). All statistical analyses were conducted using R (version 4.1.1; R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was defined as a two-sided P value <0.05.

Descriptive statistics were used to summarize baseline characteristics. Continuous variables were assessed for normality using skewness, kurtosis, and visual inspection of histograms. Normally distributed variables were reported as means with standard errors (SE) and compared using independent samples t-tests. Non-normally distributed data were summarized as medians with interquartile ranges (IQR) and compared using the Mann-Whitney U test. Categorical variables were presented as counts and percentages and analyzed using Pearson's χ² test.

To evaluate the association between anthropometric measurements (standing height, sitting height, sternum-carina distance) and actual double-lumen tube (DLT) depth, Spearman correlation coefficients were calculated.

The primary analysis compared the predictive accuracy of three models (standing height-based, sitting height-based, and CT-guided) using a one-way analysis of variance (ANOVA). When ANOVA indicated statistical significance and homogeneity of variances (tested using Levene's test), Tukey's honestly significant difference (HSD) post hoc test was applied to identify pairwise differences between groups.

Prediction error (absolute difference between predicted and actual DLT depth) was treated as a continuous outcome and analyzed as described above. Clinical accuracy (defined as correct initial DLT positioning without repositioning) was analyzed as a binary outcome. Between-group comparisons were made using χ² tests, and subgroup analyses were conducted stratified by sex and torso proportions (sitting height <85 cm vs. ≥85 cm).

Missing data were minimal and handled using complete-case analysis. No imputation was performed. All analyses were conducted according to the intention-to-treat principle.

Inclusion Criteria:

1. Voluntary participation in research;
2. Age ≥18 years and ≤80 years;
3. Patients receiving general anesthesia and DLT tracheal intubation.

Exclusion Criteria:

1. Patients with scoliosis or deformity;
2. Patients who are unable to stand upright or sit;
3. Patients with maxillary malformations or misaligned incisors;
4. The researchers judged that it was not suitable to participate in this study.

The patients and the personnel measuring the depth did not know the grouping