Structural and electrophysiological determinants of atrial cardiomyopathy identify remodeling discrepancies between paroxysmal and persistent atrial fibrillation

Figures

• 

  FIGURE 1. Study Flowchart. The study comprised three major steps. Step 1 for parameter acquisition (A) was performed in all enrolled participants with extraction of data from LA electro-anatomical mapping (LA structure, LVS extent, and LA voltage) and non-invasive parameters (ECG and clinical baseline characteristics, echocardiography). Extracted data were categorized into four categories/dimensions [(A): structural dimension (green), LVS dimension (brown), voltage dimension (blue) and non-invasive parameters (red)]. The measurement of APWD in comparison to standard PWD from the same ECG at the same beat is illustrated in the red box in (A). In contrast to standard setting (scale at 10 mm/mV and sweeping speed at 25 mm/s), the duration of P-wave in the current study (APWD) is measured at an amplified scale (60–120 mm/mV and 100 mm/s), which enables an improved accuracy in quantifying bi-atrial conduction time. Step 2: model development (B) was performed in the training set using LASSO and logistic regression for selecting the most relevant parameters for AF-persistency from each of the four different categories/dimensions. The selected parameters were used to construct diagnostic models…

• 

  FIGURE 2. Difference in selected parameters from each dimension between paroxysmal (blue) and persistent (red) AF cohorts in both training and validation sets. (A) Illustrates the difference in LAV, (B) illustrates the difference in LVS extent at < 1.0 mV threshold during sinus rhythm, (C) illustrates the difference in GMV, (D) illustrates the difference in APWD. LAV, left atrial volume; LVS, low voltage substrate; GMV, global mean voltage; APWD, duration of the amplified digital P-wave.

• 

  FIGURE 3. LASSO regression was used to select the relevant factors for AF persistency. (A) LASSO coefficient curves for the 62 clinical variables; (B) tuning parameter (λ) selection in the LASSO model used 10-fold cross-validation via minimum criteria. The partial likelihood deviance (binomial deviance) curve was plotted vs. log (λ). Dotted vertical lines were drawn at the optimal values by using the minimum criteria (λ value of 0.0438) and the one SE of the minimum criteria (the 1-SE criteria, λ value of 0.0765). The optimal value of λ (the vertical dashed line) for 10-fold cross-validation was used to select optimal variables. LASSO, least absolute shrinkage and selection operator.

• 

  FIGURE 4. ROC curves of models for identification of AF persistency in training (left) and validation set (right). Results of LAV (green), LVS extent (brown), GMV (blue) and APWD (red) represent the discriminative performance in single-dimension. The integrated (bi-dimensional) model (APWD + LAV) is illustrated in yellow curve. The HATCH score and C2HEST score were presented in purple and pink curves, respectively. ROC, receiver operating characteristics; AUC, area under the curve; LAV, left atrial volume; LVS, low voltage substrate; GMV, global mean voltage; APWD, duration of the amplified digital P-wave.

• 

  FIGURE 5. Comparison between the estimated probability for persistent AF by each model (seven outer layers) and the actual AF types (inner layer, ground truth) in training (left) and validation (right) sets. The figure consists of eight layers, the inner layer represents the ground truth AF types [paroxysmal (blue) or persistent AF (red)] of each individual (small rectangle), and the total number of rectangles in blue and red in the inner layer represents the number of patients in paroxysmal and persistent AF cohort, respectively. As illustrated by the orange arrows and labels at each layer (LAV, LVS extent, GMV, APWD, APWD + LAV, HATCH score, and C2HEST score) in an outward direction, the estimated probability for persistent AF of each individual (small rectangle) by each model was demonstrated in different grades of blue and red, indicating the probability for persistent AF ranging from “0” to “1.0.” The fifth (APWD) and sixth (APWD + LAV) layers which share a higher level of resemblance in color distribution to the inner layer, indicating a better calibration performance. LAV, left atrial volume; LVS, low voltage substrate; GMV, global mean voltage; APWD, duration of the amplified…

• 

  FIGURE 6. Calibration plots of integrated model (APWD + LAV), APWD, HATCH score and C2HEST score in training (upper) and validation (lower) sets. The diagonal gray line indicates perfect prediction of the ideal model. The dashed line represents the performance of the model, the dark solid line represents the calibrated model performance and being closer to the diagonal gray line indicated that the model has better prediction ability. In the left upper corner, different parameters are listed to provide more information of models including Dxy, ROC (equals to 1/2Dxy + 0.5), R2 and Brier Score. LAV, left atrial volume; APWD, duration of the amplified digital P-wave.

• 

  FIGURE 7. Decision curve analysis (DCA) of each model in training (left) and validation (right) sets. The vertical (Y) axis measured the net benefit, the horizontal (X) axis represents the range of potential risk threshold for persistent AF from diagnostic models. The “All” dashed dark line represented the assumption that all patient has persistent AF. The “None” dark dashed line represented the assumption that no patient has persistent AF. The clinical usefulness of each model across the range of risk thresholds is illustrated in different colors: green for LAV, brown for LVS extent, blue for GMV, red for APWD, yellow for APWD + LAV. The HATCH score and C2HEST score were presented in purple and pink curves, respectively. The model which positions closer to the right upper corner indicates better clinical usefulness. LAV, left atrial volume; LVS, low voltage substrate; GMV, global mean voltage; APWD, duration of the amplified digital P-wave.

• 

  FIGURE 8. Diagnostic performance of models using respective optimal thresholds in training (A,C,E) and validation (B,D,F) sets. In the left panel (A,B), a pentagon is generated for training and validation sets, respectively, to illustrate the diagnostic performance of each model using respective thresholds as listed on the legend at bottom. Every pentagon contains three reference levels (dashed circles), representing 50, 75, and 100%, respectively. The middle panel (C,D) illustrates the percentage of individuals above (orange) or below (emerald) respective AF-persistency thresholds of each model in paroxysmal and persistent cohorts. For LAV, LVS extent, and APWD, individuals above respective thresholds are diagnosed as positive for persistent AF, whereas for GMV, those below the threshold are diagnosed as positive for persistent AF. In the right panel (E,F), the diagnostic performance of HATCH score and C2HEST score is illustrated as stacked barplots, the percentage of different scores is displayed as bars from blue (lowest risk) to purple (highest risk) with 1-score increment, as the legends at the bottom. LAV, left atrial volume; LVS, low voltage substrate; GMV, global mean voltage;…

• 

  FIGURE 9. Nomogram for identification of individuals with persistent AF. The nomogram was developed using integrated model (APWD + LAV) in the training set. The total score of the nomogram for individual is the sum of the corresponding score assigned to each risk factor, and the total score corresponds to the individual risk for persistent AF. The optimal cut-off value for the estimated risk for persistent AF in the nomogram is 0.53 (highlighted by the red box), individuals who are estimated with risk higher than 0.53 are diagnosed as high probability for AF persistency and those below 0.53 as high probability for paroxysmal Af type. LAV, left atrial volume; duration of the amplified digital P-wave.