Machine learning methods to improve bedside fluid responsiveness prediction in severe sepsis or septic shock: an observational study.

BACKGROUND: Passive leg raising (PLR) predicts fluid responsiveness in critical illness, although restrictions in mobilising patients often preclude this haemodynamic challenge being used. We investigated whether machine learning applied on transthoracic echocardiography (TTE) data might be used as a tool for predicting fluid responsiveness in critically ill patients. METHODS: We studied, 100 critically ill patients (mean age: 62 yr [standard deviation: 14]) with severe sepsis or septic shock prospectively over 24 months. Transthoracic echocardiography measurements were performed at baseline, after PLR, and before and after a standardised fluid challenge in learning and test populations (n=50 patients each). A 15% increase in stroke volume defined fluid responsiveness. The machine learning methods used were classification and regression tree (CART), partial least-squares regression (PLS), neural network (NNET), and linear discriminant analysis (LDA). Each method was applied offline to determine whether fluid responsiveness may be predicted from left and right cardiac ventricular physiological changes detected by cardiac ultrasound. Predictive values for fluid responsiveness were compared by receiver operating characteristics (area under the curve [AUC]; mean [95% confidence intervals]). RESULTS: In the learning sample, the AUC values were PLR 0.76 (0.62-0.89), CART 0.83 (0.73-0.94), PLS 0.97 (0.93-1), NNET 0.93 (0.85-1), and LDA 0.90 (0.81-0.98). In the test sample, the AUC values were PLR 0.77 (0.64-0.91), CART 0.68 (0.54-0.81), PLS 0.83 (0.71-0.96), NNET 0.83 (0.71-0.94), and LDA 0.85 (0.74-0.96) respectively. The PLS model identified inferior vena cava collapsibility, velocity-time integral, S-wave, E/Ea ratio, and E-wave as key echocardiographic parameters. CONCLUSIONS: Machine learning generated several models for predicting fluid responsiveness that were comparable with the haemodynamic response to PLR.

Accuracy of ultrasound B-lines score and E/Ea ratio to estimate extravascular lung water and its variations in patients with acute respiratory distress syndrome.

Extravascular lung water (EVLW) could increase by permeability pulmonary oedema, cardiogenic oedema, or both. Transthoracic echocardiography examination of a patient allows quantifying B-lines, originating from water-thickened interlobular septa, and the E/Ea ratio, related to pulmonary capillary wedge pressure. The aim of our study was to assess the correlation and the trending ability between EVLW measured by transpulmonary thermodilution and the B-lines score or the E/Ea ratio in patients with ARDS. Twenty-six intensive care unit patients were prospectively included. B-lines score was obtained from four ultrasound zones (anterior and lateral chest on left and right hemithorax). E/Ea was measured from the apical four-chamber view. EVLW was compared with the B-lines score and the E/Ea ratio. A linear mixed effect model was used to take account the repeated measurements. A p value<0.05 was considered significant. A total of 73 measurements were collected. The correlation coefficient between EVLW and B-lines score was 0.66 (EVLW=0.71 B-lines+7.64, R2=0.44, p=0.001), versus 0.31 for E/Ea (p=0.06). The correlation between EVLW changes and B-lines variations was significant (R2=0.26, p<0.01), with a concordance rate of 74%. A B-lines score≥6 had a sensitivity of 82% and a specificity of 77% to predict EVLW>10 ml/kg, with an AUC equal to 0.86 (0.76-0.93). The gray zone approach identified a range of B-lines between four and seven for which EVLW>10 ml/kg could not be predicted reliably. The correlation between ultrasound B-lines and EVLW was significant, but the B-lines score was not able to track EVLW changes reliably.

Integrated use of bedside lung ultrasound and echocardiography in acute respiratory failure: a prospective observational study in ICU.

BACKGROUND: It has been suggested that the complementary use of echocardiography could improve the diagnostic accuracy of lung ultrasonography (LUS) in patients with acute respiratory failure (ARF). Nevertheless, the additional diagnostic value of echocardiographic data when coupled with LUS is still debated in this setting. The aim of the current study was to compare the diagnostic accuracy of LUS and an integrative cardiopulmonary ultrasound approach (thoracic ultrasonography [TUS]) in patients with ARF. METHODS: We prospectively recruited patients consecutively admitted for ARF to the ICU of a university teaching hospital over a 12-month period. Inclusion criteria were age ≥ 18 years and the presence of criteria for severe ARF justifying ICU admission. We compared both LUS and TUS approaches and the final diagnosis determined by a panel of experts using machine learning methods to improve the accuracy of the final diagnostic classifiers. RESULTS: One hundred thirty-six patients were included (age, 68 ± 15 years; sex ratio, 1). A three-dimensional partial least squares and multinomial logistic regression model was developed and subsequently tested in an independent sample of patients. Overall, the diagnostic accuracy of TUS was significantly greater than LUS (P < .05, learning and test sample). Comparisons between receiver operating characteristic curves showed that TUS significantly improves the diagnosis of cardiogenic edema (P < .001, learning and test samples), pneumonia (P < .001, learning and test samples), and pulmonary embolism (P < .001, learning sample). CONCLUSIONS: This study demonstrated for the first time to our knowledge a significantly better performance of TUS than LUS in the diagnosis of ARF. The value of the TUS approach was particularly important to disambiguate cases of hemodynamic pulmonary edema and pneumonia. We suggest that the bedside use of artificial intelligence methods in this setting could pave the way for the development of new clinically relevant integrative diagnostic models.