717: MULTIORGAN ULTRASOUND MODEL PREDICTS 48-HOUR EXTUBATION FAILURE: MULTICENTER PROSPECTIVE STUDY

Le résumé fourni par la source

Introduction: Extubation fails in ~15–35% of ventilated ICU patients, and re-intubation worsens outcomes. Single-organ ultrasound studies exist, but integrated models are scarce. We hypothesized that a multiorgan ultrasound model would predict 48-hour extubation failure. Methods: Prospective multicenter study in 15 hospitals across 4 countries (ClinicalTrials.gov NCT06609798) with local IRB approvals and surrogate consent. Outcome: extubation failure within 48 hours. Centers were split by hospital into derivation (74%; n=255) and external validation (26%; n=86). Predictors included modified Lung Ultrasound Score (LUS); TAPSE, MAPSE, E/e′; diaphragmatic excursion and thickening fraction; BMI; and prespecified interactions. We fitted mixed-effects logistic regression (random patient intercept) and a tuned random forest. For the logistic model, bootstrap validation (300 replicates) preceded external validation. Discrimination was summarized by c-statistic; calibration by Cox slope and calibration-in-the-large (CITL). Clinical benefit was expressed as observed failure proportions in predefined high- versus low-risk strata. Results: We analyzed 341 extubations in 326 patients; 18.2% failed within 48 h. The final logistic model retained four interactions with explicit thresholds: TAPSE < 20 mm × diaphragmatic excursion < 14 mm (OR 2.38; 95% CI 1.21–4.69); MAPSE < 14 mm × E/e′ >15 (OR 3.59; 95% CI 0.82–15.77); LUS >6 × BMI (OR 1.01 per unit; 95% CI 0.99–1.04); MAPSE < 14 mm × BMI (OR 1.75; 95% CI 0.85–3.61). Bootstrap c-statistic was 0.65 (95% CI 0.56–0.74) with slope ≈1.02 and CITL -0.1; external c-statistic was 0.74 (0.59–0.90) with slope 1.05 (0.13–1.97) and CITL −0.51 (−1.17 to 0.14). With a 0.34 threshold (rule-in to defer extubation), 8.5% were high risk with 45% failures, versus 16% in low risk. The tuned random forest (30 trees; depth 1) achieved c-statistics of 0.70 (0.62–0.78) in derivation and 0.73 (0.59–0.88) in validation; at 0.24, 11.4% were high risk with 49% failures versus 14% in low risk. Conclusions: A multiorgan ultrasound model predicted 48-hour extubation failure with good calibration and discrimination. Risk stratification concentrated failures in the high-risk group while keeping failure rates lower in the low-risk group, supporting ultrasound-guided, risk-based extubation decisions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.