Deep Learning Algorithm for Management of Diabetes Mellitus via Electrocardiogram-Based Glycated Hemoglobin (ECG-HbA1c): A Retrospective Cohort Study

Abstract

Background: glycated hemoglobin (HbA1c) provides information on diabetes mellitus (DM) management. Electrocardiography (ECG) is a noninvasive test of cardiac activity that has been determined to be related to DM and its complications. This study developed a deep learning model (DLM) to estimate HbA1c via ECG.

Methods: there were 104,823 ECGs with corresponding HbA1c or fasting glucose which were utilized to train a DLM for calculating ECG-HbA1c. Next, 1539 cases from outpatient departments and health examination centers provided 2190 ECGs for initial validation, and another 3293 cases with their first ECGs were employed to analyze its contributions to DM management. The primary analysis was used to distinguish patients with and without mild to severe DM, and the secondary analysis was to explore the predictive value of ECG-HbA1c for future complications, which included all-cause mortality, new-onset chronic kidney disease (CKD), and new-onset heart failure (HF).

Results: we used a gender/age-matching strategy to train a DLM to achieve the best AUCs of 0.8255 with a sensitivity of 71.9% and specificity of 77.7% in a follow-up cohort with correlation of 0.496 and mean absolute errors of 1.230. The stratified analysis shows that DM presented in patients with fewer comorbidities was significantly more likely to be detected by ECG-HbA1c. Patients with higher ECG-HbA1c under the same Lab-HbA1c exhibited worse physical conditions. Of interest, ECG-HbA1c may contribute to the mortality (gender/age adjusted hazard ratio (HR): 1.53, 95% conference interval (CI): 1.08-2.17), new-onset CKD (HR: 1.56, 95% CI: 1.30-1.87), and new-onset HF (HR: 1.51, 95% CI: 1.13-2.01) independently of Lab-HbA1c. An additional impact of ECG-HbA1c on the risk of all-cause mortality (C-index: 0.831 to 0.835, p < 0.05), new-onset CKD (C-index: 0.735 to 0.745, p < 0.01), and new-onset HF (C-index: 0.793 to 0.796, p < 0.05) were observed in full adjustment models.

Conclusion: the ECG-HbA1c could be considered as a novel biomarker for screening DM and predicting the progression of DM and its complications.

Keywords: artificial intelligence; deep learning; diabetes mellitus; electrocardiogram; glycated hemoglobin.

Figure 1

The summary of study design in this study. The process of development, validation, and follow-up cohorts with each electrocardiogram (ECG) labeling of HbA1c was indicated. The patients in validation and follow-up cohorts were totally different from development cohort. The development cohort included three subsets (subset-1: outpatient department samples; subset-2: full samples; subset-3: samples with corresponding HbA1c). Abbreviations: DM, diabetes mellitus.

Figure 2

The implementation of our deep learning model. (A) The model architectures of the deep learning model for analyzing ECG. (B) Four training strategies were based on different sampling processes. The matching strategy was to split the sample to multiple blocks based on different conditions. The batch samples were sampled from each block with the same probability.

Figure 3

Analysis of deep learning model and traditional machine learning models. (A) The performance comparison of deep learning model (DLM) trained by 6 different weighting strategies in the validation cohort. The DLM (…) were made by predictions of deep learning models using different strategies. The XGB model and Elastic net demonstrated the corresponding predictions. The gender/age-matching strategy provides the highest correlation between estimated HbA1c and actual HbA1c. (B) Performance comparison for detecting DM and severe DM in the follow-up cohort. ROC curves were created from predictions of the deep learning model trained using a gender/age-matching strategy. Moreover, the performance of XGB model and elastic net were also presented. (C) Scatter plot between DLM predictions and actual HbA1c in the follow-up cohort. The x-axis indicates the true HbA1c from laboratory tests. The y-axis presents the predicted HbA1c from the deep learning model trained using a gender/age-matching strategy. Red points represent the highest density, followed by yellow, green, light blue, and dark blue. (D) Related feature importance ranking in XGB model (information gain) and elastic net (standard coefficient). There are only the top 10 important variables in each model, and the blue color demonstrates the negative relationship between variables and actual HbA1c.

Figure 4

Stratified analysis for detecting DM and severe DM in the follow-up cohort. The DLM’s sensitivity and specificity to detect DM and severe DM are tabulated across a series of stratified analyses. The p-value was the significant test of strength of association, and a significance level was 0.0045 based on the Bonferroni correction.

Figure 5

Characteristics and risk analysis in patients with actual HbA1c and ECG-HbA1c. (A) Patient characteristics in different ECG-HbA1c groups and real HbA1c groups. Bars represent the mean or proportion where appropriate and corresponding 95% conference intervals, which are adjusted by real HbA1c in each group via linear or logistic regression. Significant tests are based on the trend test (*: p for trend < 0.05; **: p for trend <0.01; ***: p for trend <0.001), and the sign represents the correlation direction. (B) Risk matrixes of ECG-HbA1c and HbA1c groups on DM related complications. The hazard ratios (HRs) are based on a Cox proportional hazard model before and after adjusting by gender and age. The color gradient represents the risk of corresponding group.

Figure 6

Additional contributions of ECG-HbA1c on DM related complications. (A) A Cox proportional hazard model and C-index are used as the performance assessment for a series of models. The model 1 includes significant demographic data, the model 2 includes variables in model 1 and additional significant disease histories, and the model 3 includes variables in model 2 and additional significant laboratory tests. Abbreviations: *, p < 0.05; **, p < 0.01; ***, p < 0.001. (B) The multivariable analyses of the models with best performance (model 3 + HbA1c + ECG-HbA1c) described above. The risk score can be calculated based on these coefficients to provide the corresponding C-index as above.