Fig. 4

A-C, Comparison of diagnostic accuracy for screening for DVT of different ROCs in three ARDS cohorts. Patients were split by generating random numbers to produce a training data set (n*0.7) and a validation data set (n*0.3) in the overall, direct, and indirect groups respectively. A, the ACI model which including age, serum creatinine level, and IMV shows satisfactory forecasting ability for DVT (AUC = 0.786; 95% CI: 0.673–0.898; sensitivity: 74.2%; specificity: 78.4%; P < 0.001) significantly higher than that of the Padua prediction score (AUC = 0.587; P = 0.005 for these two curves) and the Caprini score (AUC = 0.558; P = 0.001 for these two curves). B, the ACI model shows a satisfactory ability to predicting DVT (AUC = 0.783; 95% CI: 0.612–0.953; sensitivity: 81.8%; specificity: 69.6%; P = 0.004) significantly surpassed the Padua prediction score (AUC = 0.521; P = 0.001 for these two curves) and the Caprini score (AUC = 0.492; P = 0.006 for these two curves). C, the ACI model shows satisfactory ability for predicting DVT (AUC = 0.712; 95% CI: 0.519–0.905; sensitivity: 63.6%; specificity: 83.3%; P = 0.024) has no obvious difference compared with the Padua prediction score (AUC = 0.644; P = 0.551 for these two curves) and the Caprini score (AUC = 0.627; P = 0.451 for these two curves). Abbreviations: ACI = age + creatinine + IMV; ARDS, acute respiratory distress syndrome; AUC, area under the curve; CI, confidence interval; DVT, deep vein thrombosis; IMV, invasive mechanical ventilation; ROC, receiver operating characteristic