Prognostic value and clinical net benefit of global longitudinal strain in sepsis: a prospective study with internal bootstrap validation

Figure 1. Violin and box plots illustrating the distribution of key clinical and echocardiographic parameters between the survival and non-survival groups. The plots depict the distribution densities of (A) GLS_AVG, (B) LVEF, (C) SOFA score, and (D) peak NT-proBNP (log-scaled) in sepsis patients, stratified by 28-day mortality. The internal box plots show the median and interquartile ranges. P-values were calculated using the Wilcoxon rank-sum (Mann–Whitney U) test. Non-survivors exhibited significantly worse (less negative) GLS_AVG and higher SOFA scores compared to survivors.

Figure 2. Bland-Altman plot assessing the inter-observer reproducibility of global longitudinal strain (GLS_AVG) measurements. The solid red line represents the mean difference (bias) between the two independent echocardiographic measurements, and the dashed gray lines indicate the 95% limits of agreement (± 1.96 standard deviations). The minimal bias and narrow limits demonstrate the excellent reliability of the speckle-tracking imaging technique.

Figure 3. Spearman rank correlation heatmap of GLS_AVG and key clinical variables. Colors represent the direction and magnitude of the Spearman correlation coefficients (r). Blue indicates a positive correlation, while red indicates a negative correlation; darker shades signify stronger correlations. Asterisks denote the level of statistical significance (*P < 0.05, **P < 0.01, ***P < 0.001). GLS_AVG demonstrated significant correlations with critical disease severity scores (SOFA, APACHE II) and myocardial injury biomarkers.

Figure 4. Receiver operating characteristic (ROC) curves for predicting 28-day mortality. The predictive performance of GLS_AVG (red line) is compared against conventional LVEF (blue line) and the clinical SOFA score (green line). GLS_AVG yielded a higher Area Under the Curve (AUC = 0.796) compared to LVEF (AUC = 0.706). DeLong's test was utilized to compare the statistical difference between the ROC curves.

Figure 5. Clinical nomogram and internal calibration curve for predicting 28-day mortality. (A) A predictive nomogram developed by integrating GLS_AVG and the SOFA score. Clinicians can obtain the total points by drawing a vertical line from each variable to the “Points” axis, which corresponds to the estimated probability of 28-day mortality on the bottom axis. (B) Calibration curve of the nomogram utilizing 1,000 bootstrap resamples. The x-axis represents the nomogram-predicted probability, and the y-axis represents the actual observed probability. The dashed 45-degree line represents perfect ideal prediction, while the solid line indicates the bias-corrected performance of the model, demonstrating excellent internal consistency.

Figure 6. Decision curve analysis (DCA) of the predictive models for 28-day mortality. The DCA evaluates the clinical net benefit of different predictive strategies across a wide range of threshold probabilities. The combined GLS_AVG + SOFA model (Prob_Combined) and GLS_AVG alone demonstrated a consistently higher net benefit compared to relying solely on LVEF, or the extreme clinical strategies of assuming all patients will die (“Treat All”) or no patients will die (“Treat None”). This highlights the tangible clinical utility of incorporating strain imaging into early emergency triage.

Figure 7. Forest plot of exploratory subgroup analyses for the association between GLS_AVG and 28-day mortality. The plot displays the Odds Ratios (OR) and 95% Confidence Intervals (CI) for GLS_AVG in predicting mortality across various clinical subgroups, including age stratifications, the presence of septic shock, and the requirement for mechanical ventilation. The vertical line at OR = 1 indicates no effect. The directionality of risk remained generally consistent across all subgroups, confirming the broad prognostic applicability of GLS in heterogeneous septic populations.