Artificial Intelligence-Enabled Electrocardiography Detects Hypoalbuminemia and Identifies the Mechanism of Hepatorenal and Cardiovascular Events

Figures

• 

  FIGURE 1. Development, tuning, internal validation, and external validation set generation and ECG labeling of albumin. Schematic of the dataset creation and analysis strategy, which was devised to assure a robust and reliable dataset for training, validating, and testing of the network. Once a patient’s data were placed in one of the datasets, that individual’s data were used only in that set, avoiding “cross-contamination” among the training, validation, and test datasets. The details of the flow chart and how each of the datasets was used are described in “Materials and Methods”.

• 

  FIGURE 2. Predicted albumin (ECG-Alb) and actual Alb. (A) Scatter plots of ECG-Alb compared to the actual Alb. The x-axis indicates the actual lab measured Alb, and the y-axis presents the ECG-Alb. Red points represent the highest density, followed by yellow, green light blue, and dark blue. We presented the mean difference (Diff), Pearson correlation coefficients (COR), and mean absolute errors (MAE) to demonstrate the accuracy of the DLM. The black lines with 95% conference intervals are fitted via simple linear regression. (B) The distributions of Alb in the internal and external validation sets. The color gradient from white to red demonstrates the ECG-Alb from normal to low. The top panel shows the original distribution of each dataset, and the bottom panel shows the distribution of ECG-Alb for each actual Alb value.

• 

  FIGURE 3. The ROC curve of DLM predictions based on ECG to detect mild to severe hypoalbuminemia. Mild, moderate, and severe hypoalbuminemia were defined as an actual albumin (Alb) of ≤ 3.5, ≤ 3.0, and ≤ 2.5, respectively. The operating point was selected based on the maximum Youden’s index in the tuning set and presented using a circle mark, and the area under the ROC curve (AUC), sensitivity (Sens.), specificity (Spec.), positive predictive value (PPV), and negative predictive value (NPV) were calculated based on it. Due to the different distributions of Alb in the internal and external validation sets, we generated a balanced dataset for each set to ensure the same number of cases for different values of Alb.

• 

  FIGURE 4. Relationship between the selected ECG features and predicted albumin (ECG-Alb). The related importance is based on the information gain of the XGB model, and the R-square (R-sq) is the coefficient of determination to use selected ECG features for predicting ECG-Alb. The AI-ECG predictions were classified as normal ECG-Alb, low ECG-Alb, and severe low ECG-Alb based on the operating points, as in the previous ROC curve analysis. The analyses are conducted both in internal and external validation sets (*p for trend < 0.05; ***p for trend < 0.001).

• 

  FIGURE 5. Long-term incidence of developing new-onset malnutrition events in patients with an initially normal albumin (Alb) of > 3.5 g/dL stratified by AI-ECG prediction. The AI-ECG predictions were classified as normal ECG-Alb (yellow line), low ECG-Alb (pink line), and severe low ECG-Alb (burgundy line) based on the operating points, as in the previous ROC curve analysis. The analyses were conducted both in the internal and external validation sets. The table shows the at-risk population and cumulative risk for the given time intervals in each risk stratification.