A56-02 Accuracy and Predictability of Point-of-care Ultrasound in Patients With Cardiogenic Shock: A Bayesian Diagnostic Test Accuracy Meta-analysis

Abstract Background Cardiogenic shock (CS) remains a critical emergency with high mortality, demanding rapid and reliable bedside diagnosis. Point-of-care ultrasound (POCUS) has emerged as a transformative, non-invasive tool for real-time hemodynamic assessment, yet its diagnostic predictability is poorly characterized due to marked heterogeneity across studies. Methods We conducted a systematic review and Bayesian diagnostic test accuracy meta-analysis using a bivariate random-effects model. A comprehensive search was performed across PubMED/MEDLINE, Embase, and Scopus databases, including both randomized controlled trials (RCTs) and observational studies. True positives (TP), false negatives (FN), false positives (FP), and true negatives (TN) were extracted to compute diseased (n1 = TP + FN) and non-diseased (n2 = FP + TN) sample sizes. The model used a logit link, normal random effects, and sensitivity/specificity (Se/Sp) parametrization. Markov Chain Monte Carlo (MCMC) sampling was employed, with convergence assessed via Rhat ≈ 1, effective sample size 1, 000, and deviance information criterion (DIC). A robust scale-mixture model was fitted for sensitivity analysis. Key outcomes included pooled and predictive sensitivity/specificity with 95% credible intervals (CrI). Results This Bayesian diagnostic test accuracy meta-analysis integrates evidence from eight studies (N = 764 participants) to provide clinically actionable estimates of POCUS performance. Pooled sensitivity was 0. 868 (95% CrI: 0. 529 to 0. 981), reflecting moderate ability to detect CS, while pooled specificity was 0. 971 (95% CrI: 0. 951 to 0. 984), indicating excellent rule-in performance. Predictive distributions, which are critical for real-world predictability, showed that in a new study, expected sensitivity would be 0. 774 (95% CrI: 0. 012 to 1. 000) and specificity 0. 968 (95% CrI: 0. 920 to 0. 991). Between-study heterogeneity was substantial in sensitivity (σSe = 2. 447) but minimal in specificity (σSp = 0. 338), suggesting consistent negative test reliability. The Bayesian area under the SROC curve (BAUC), derived analytically from posterior samples of SROC parameters, was 0. 970 (95% CrI: 0. 952 to 0. 984), confirming excellent overall diagnostic accuracy. The robust model yielded nearly identical results (sensitivity: 0. 865, 95% CrI: 0. 507 to 0. 981; specificity: 0. 972, 95% CrI: 0. 953 to 0. 985). Conclusions POCUS is a highly specific, rule-in tool for cardiogenic shock, with predictable negative test performance across settings. However, wide predictive intervals in sensitivity underscore caution when using POCUS to exclude CS, particularly in heterogeneous populations. This Bayesian framework advances beyond traditional frequentist approaches by providing clinically interpretable predictive intervals, enabling evidence-based integration of POCUS into shock protocols. Future research should standardize POCUS protocols and explore operator-dependent variability to enhance diagnostic predictability. This abstract is funded by: None