A novel QTc–RR differential biomarker for the early assessment of autonomic dysfunction in type 2 diabetes

Figures

• 

  Figure 1. Electrocardiogram (ECG) signals and QT intervals. ECG signals from five subjects are presented: (A) (normal T wave), (B) (flattened T waves), (C) (peaked T waves), (D) (inverted T waves), and (E) (others). QT intervals for each subject were determined using MATLAB code based on expert-defined rules.

• 

  Figure 2. Comparison of fiducial point detection between expert annotations and the custom algorithm. A representative ECG segment from the PhysioNet QT Database. Red markers represent expert annotations (Qon, Rpeak, Toff), and blue markers represent detected points by our custom algorithm. The close spatial convergence demonstrates high delineation accuracy.

• 

  Figure 3. QTc interval–RR interval differential (RQTdiff) parameter calculation. ECG signals were captured at a sampling rate of 500 Hz for 2000 heartbeats. To illustrate the RQTdiff parameter, only three consecutive R waves (Ri, Ri+1, Ri+2) and two RR intervals (RRIi, RRIi+1) are depicted. The QT interval (QTi+1) is determined as shown in Figure 3, between R waves Ri+1 and Ri+2. The corrected QT interval (QTci+1) is calculated using Bazett’s formula: QTci+1 = QTi+1/√RRIi+1. The RQTdiff parameter (RQTdiff i+2) is the difference between the RR interval (RRIi+1) and the corrected QT interval (QTci+1): RQTdiff i+2 = RRIi+1 − QTci+1.

• 

  Figure 4. (a) Heart rate variations, corrected QT interval (QTc), and RQTdiff over 1000 beats for four subjects: A–D Subjects A and B, both healthy (subject A: male, age 22, BMI: 22.4, HbA1c: 5.5; subject B: female, age 58, BMI: 20.3, HbA1c: 6.2), show moderate fluctuations in heart rate during the recorded period. In contrast, subjects C and D, both with T2DM (subject C: male, age 28, BMI: 41.2, HbA1c: 8.3; subject D: female, age 56, BMI: 28.3, HbA1c: 8.5), exhibit distinct heart rate profiles. Subject C demonstrates a marked reduction in dynamic variability, whereas subject D presents a significantly elevated HRV. Despite these variations, all subjects display consistent patterns in their RRI and RQTdiff, with RRI consistently greater than RQTdiff. QTc values for all four subjects remain within a narrow range. (b) A strong positive correlation (all p < 0.001) was observed between RRI and RQTdiff for all four subjects shown in (a). The healthy young subject A exhibited the highest positive correlation, whereas the two diabetic subjects C, D showed reduced correlation coefficients, along with a narrower range of variation in both RRI and RQTdiff (with lower RQTdiff than subjects A, B). The…

• 

  Figure 5. Ternary diagrams illustrate the separation ability of entropy-based indices—namely, MSESS(RRI), MSESS(RQTdiff), and BEI(RQTdiff)—between the two groups, highlighting the impact of combined glycosylated hemoglobin (HbA1c) and age. For MSESS, RQTdiff demonstrated superior performance compared to RRI.

• 

  Figure 6. Receiver operating characteristic (ROC) curve analysis is performed to determine the optimal cutoffs for the following HRV indices—BEI, LHR, and SSR—with (a) RRI parameter, and (b) RQTdiff parameter for predicting T2DM. ROC analysis reveals that RQTdiff improved the AUCs for the SSR and LHR indices by 27% and 13%, respectively.

• 

  Figure 7. Comparison of overall classification accuracy in logistic regression analysis for predicting type 2 diabetes mellitus using four HRV indices—Baroreflex Entropy Index (BEI), small-scale Multiscale Entropy (MSEss), Low-to-High Frequency Ratio (LHR), and Poincaré plot index (SSR)—with two different input parameters: R–R interval (RRI) and QTc–RR differential (RQTdiff). Across all indices, the models incorporating RQTdiff consistently achieved higher classification performance than those using the RRI.