Time-dependent S-wave areas by 24-hour ECG are correlated with a high risk of sudden cardiac death: ECG prediction model development and validation for SCD risk

Introduction

Sudden cardiac death (SCD) accounts for nearly 50% of all cardiovascular deaths, representing a critical global health challenge (1). In the United States alone, approximately 360,000 individuals succumb to SCD annually (1). In China, this number rises to an alarming 544,000 cases annually, the highest worldwide (2). Despite its clinical importance, the success rate of out-of-hospital rescue efforts remains below than 1%, underscoring an urgent need for effective preventive strategies (3). Currently, implantable cardioverter defibrillators (ICDs) represent the primary preventive measure; however, their use mainly depends on left ventricular ejection fraction (LVEF). This single parameter fails to capture the heterogeneity of SCD etiologies (4,5). Furthermore, many patients classified as low risk by LVEF criteria still experience SCD, highlighting the necessity for more comprehensive risk prediction tools to enable early identification and intervention (6-8).

SCD predominantly arises from abnormal cardiac electrical activity, with lethal ventricular arrhythmias, such as sustained ventricular tachycardia (VT) and ventricular fibrillation (VF), accounting for over 76% of SCD cases (9). Electrocardiogram (ECG), a non-invasive, low-cost, and widely accessible diagnostic tool, has traditionally been used for SCD risk stratification. Static ECG parameters, including prolonged QRS duration, prolonged QT interval, and fragmented QRS waves, have been extensively studied as potential predictors of SCD (10-12). However, their predictive power remains suboptimal, particularly in reflecting the dynamic nature of cardiac electrical activity.

The S-wave, a key indicator of late ventricular depolarization, has recently gained attention for its association with malignant ventricular arrhythmias. Previous studies have linked deep and/or large S-waves in lead I to malignant arrhythmias in patients with Brugada syndrome (13,14). Nevertheless, prior research has primarily targeted specific patients with individual cardiovascular conditions, overlooking the combined effects of S-wave amplitude and width. Additionally, circadian rhythm variations of cardiac electrical activity have not yet been integrated into existing predictive models.

Holter electrocardiography, which provides continuous 24-hour monitoring, presents a unique opportunity for the analysis of time-dependent ECG parameters. Unlike traditional 15-second ECG recordings, dynamic ECG data collected by Holter monitoring can reveal cardiac electrical temporal patterns. Recent studies have suggested that circadian rhythms may significantly influence the predictive performance of ECG parameters for SCD (15-17). However, comprehensive analyses integrating these temporal characteristics into predictive models remain lacking.

In this study, we utilized 24-hour Holter ECG data to investigate the predictive value of time-dependent ECG parameters, particularly those involving the S-wave. By integrating time-based parameters with morphological features, we identified dynamic ECG markers that reflect diurnal variations. We analyzed ECG data from three participant groups with distinct levels of SCD risk: SCD high-risk group (SCDHR), high-risk heart failure group (HF) and healthy control group (HC). The objective was to identify predictive time-dependent ECG parameters, especially those related to the S-wave, and evaluate their circadian rhythm distribution. We present this article in accordance with the TRIPOD reporting checklist (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-11/rc).

Methods

The clinical data for this study were derived from the project titled “A Prospective, Multiple Center, Cohort Study of Prediction Model on Sudden Cardiac Death and Devices Development by Automatic Analysis From 24 h Electrocardiogram in China (PM-SCD CHINA)” (registered at https://clinicaltrials.gov/, Clinical Trial ID NCT03485079). Study participants were recruited from (I) Sun Yat-sen Memorial Hospital, Sun Yat-sen University, (II) Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, (III) Drum Tower Hospital, Nanjing University Medical School and (IV) Zhongshan Hospital, Fudan University between September 2017 and December 2021. The study protocol was approved by the study team leader unit, Medical Ethics Committee of Sun Yat-sen Memorial Hospital, Sun Yat-sen University (Ethics number: SYSKY-2023-688-01) and strictly adhered to the Declaration of Helsinki and its subsequent amendments. All participating hospitals/institutions were informed and agreed with this study. All patients provided written informed consent.

Participants

Eligible participants aged 18–75 years were enrolled and categorized into three groups:

• SCDHR group: participants with recorded episodes of VT, VF, or cardiac arrest who met at least one of the following criteria:
  • Survivors of VF or hemodynamically unstable sustained VT-induced cardiac arrest due to irreversible causes (including appropriate device therapy with anti-tachycardia pacing and defibrillation);
  • Individuals with spontaneous sustained VT (including appropriate device therapy with anti-tachycardia pacing and defibrillation) in individuals diagnosed with ischemic cardiomyopathy (LVEF ≤35%), dilated cardiomyopathy (LVEF ≤35%) or hypertrophic cardiomyopathy, and classified as New York Heart Association (NYHA) class II or III.
• HF group: participants without evidence of VF or sustained VT, but diagnosed with ischemic cardiomyopathy (LVEF ≤35%), dilated cardiomyopathy (LVEF ≤35%), or hypertrophic cardiomyopathy, and classified as NYHA class II or III.
• HC group: participants without evidence of coronary artery disease, cardiomyopathy, other structural heart diseases, or hereditary arrhythmia diseases, having normal heart function and LVEF.

ECG equipment

Dynamic ECG data were collected using a single-lead 24-hour ECG recorder (Model: CT-083S; Registration Certificate: Zhejiang Food and Drug Administration Certificate [quasi] 2014 No. 2210778; Manufacturer: Hangzhou Baihui Medical Equipment Co., Ltd., Hangzhou, China), provided by Beijing Yihe Health Management Co., Ltd. (Beijing, China). Recordings lasted at least 24 hours. All participants received assistance with device installation and removal. Backup paper ECG data were recorded at a paper speed of 25 mm/s with a calibration of 10 mm/mV.

ECG data processing and filtering

The 24-hour ECG recordings were segmented into 1,440 one-minute intervals, labeled from 00:00 to 24:00, each further divided into 60 one-second segments. Criteria for excluding one-second segments included:

• Maximum amplitude <0.2 mV.
• Maximum amplitude >5 mV.
• Amplitude range exceeding twice the median difference for that minute.

Segments meeting these exclusion criteria were discarded. If fewer than 30 valid one-second segments remained within a minute, the entire minute was excluded. ECG samples containing fewer than 100 qualifying one-minute segments were removed from analysis. Additionally, time points containing fewer than 100 valid one-minute segments across all samples were also excluded.

Dataset construction

After filtering, the earliest recorded 24-hour ECG data from each participant were retained as the sample data for constructing the training and independent test sets. Samples were randomly assigned to the two sets. The training set comprised 70% of the samples from Sun Yat-sen Memorial Hospital, while the independent test set included the remaining samples from this hospital and all samples from the other three centers. Sample sizes were balanced as closely as possible.

ECG parameters extraction

Parameter extraction and multivariable regression modeling were performed using Python. In order to comprehensively cover the ECG morphological parameters and their related combinations as much as possible. We designed five major categories including: heart rate; maximum and minimum peak values of waves (including R waves, T waves, and S waves), maximum, minimum, and average area under the curve (AUC) of waves and baseline (including R waves, T waves, and S waves), sum of peak values of waves, and the difference in AUC between different waves and the baseline. A total of 42 simple or compound morphological ECG parameters were computed for each one-minute segment based on above five categories. The related area parameters were calculated by integrating ECG amplitude over time within the wave region. Detailed descriptions of all parameters are provided in Table S1.

Statistical analysis

For the descriptive statistics in this study, the continuous variables are expressed as the mean ± (SD) or the mean and its 95% confidence interval (95% CI); the discontinuous variables are expressed as the number (N). For the missing baseline data not related to the major 42 morphological ECG parameters, the mean imputation method was used to process the corresponding missing values. For excluded patients, their related data would be all deleted and not used for any result analysis.

The ECG analysis process is outlined in Figure 1. Student’s t-test was conducted to compare the 42 ECG parameters between the SCDHR, HF, and HC groups. The P values were adjusted for multiple comparisons, with statistical significance set at P<0.0012 (0.05/42).

Figure 1 Schematic of the ECG parameters associated with SCDHR or HF. First, the ECG data of each individual were scattered into a one-minute-long time fragment. Second, a multivariate logistic regression was used to select the critical ECG parameters. Third, the associations of individual ECG parameters with SCD or HF were tested by univariable logistic regression. Finally, certain time-dependent S-wave area-related ECG parameters (i.e., “inteS_mean”, “inteSm”, and “inteST”) were found to be potentially early predictive factors for the risk of SCD. A P value <0.0012 (0.05/42) was considered statistically significant. ECG, electrocardiogram; HC, healthy control; HF, heart failure; OR, odds ratio; SCD, sudden cardiac death; SCDHR, sudden cardiac death high-risk.

Univariable and multivariable logistic regression analyses identified significant parameters previously confirmed by Student’s t-test. K-means clustering (18) was applied to group the 42 parameters in the training set. Within each cluster, the parameter with the lowest Akaike information criterion (AIC) score, determined by univariable logistic regression analysis between groups (SCDHR vs. HC, SCDHR vs. HF, HF vs. HC), was selected for modeling. Selected parameters were subsequently included a multivariable logistic regression model, optimized by automatic backward selection.

Model performance was evaluated using the average AUC from five-fold cross-validation, repeated five times. Parameters with superior performance were subsequently analyzed by sequentially linking models developed from different time segments of the training set into a single composite model using a greedy-search method. These individual models were sequentially combined based on their performance rankings (AUC), continuing this process until all 1,440-time segments were integrated. The final composite model, characterized by the highest AUC and comprising more than two segments, was validated in the independent test set. Additionally, the effectiveness of the logistic regression model was compared with the LASSO regression model. All statistical analysis was conducted with R Studio software (2022.07.1+554).

Results

Study population and ECG samples

Between September 2017 and December 2020, 325 participants meeting the inclusion criteria were recruited (Table S2). As depicted in Figure 2, all 325 participants underwent 24-hour ECG recordings on the day of enrollment. After data filtering and excluding non-qualifying samples, 289 ECG recordings remained for the final analysis. These recordings were utilized to evaluate intergroup differences in ECG parameters across the SCDHR, HF, and HC groups and to construct multivariable regression models. The sources of patients and their allocation in training and independent testing set were shown in Table 1.

Figure 2 Overall workflow and main results of this study. ECG, electrocardiogram; HC, healthy control; HF, heart failure; inteS_mean, the average area under the S-wave; inteSm, the minimum area under the S-wave; inteST, the difference in the S wave and T wave; SCDHR, sudden cardiac death high-risk.

Table 2 summarizes the demographic characteristics, basic medical history, clinical data (mean ± SD) and therapeutic history of the participants. Significant differences were identified between the HC group and the other two groups concerning baseline medical history, cardiovascular-related laboratory parameters, echocardiographic findings, and therapeutic regimens (all P<0.05). In contrast, no significant differences were found between the HF and SCDHR groups in terms of demographic characteristics, previous medical history, or most laboratory and echocardiographic parameters (P>0.05). However, notable differences between these two groups were observed regarding the history of coronary artery disease, NT-pro BNP levels, ejection fraction (EF), previous ICD implantation, and certain medications, including beta-blockers, diuretics, statins, antiarrhythmic drugs, and single antiplatelet therapy. Most indicators for baseline characteristics of both training and independent testing sets participants basically consistent with their overall level (Table S3).

Intergroup differences in the ECG parameters

Significant differences in ECG parameters between the groups were identified using Student’s t-tests. A total of 1,440 ECG segments were analyzed.

SCDHR vs. HC group

In the training set, significant differences were observed in 20 ECG parameters (P<0.0012), predominantly within specific time intervals. Most differences occurred during the period from 16:00 to 22:00 (Figure 3A). These significant parameters included:

• Average S-wave area (inteS_mean);
• Total T-wave area in one minute (inteT_sum);
• Average R- and S-wave area (inteRS);
• T-wave interval duration (t_T).

Figure 3 ECG parameters and time fragments in the training set. (A) The number of time points with significant difference for each morphological ECG parameter within a day between the SCDHR patients and HCs (nSCDHR=25 and nHC=75). (B) The number of ECG parameters with significant difference at each time point within a day between the SCDHR patients and HF patients (nSCDHR=25 and nHF=57). (C) The number of time points with significant difference for each morphological ECG parameter within a day between the SCDHR patients and HF patients (nSCDHR=25 and nHF=57). (D) The number of ECG parameters with significant difference at each time point within a day between the HF patients and HCs (nHF=57 and nHC=75). (E) The number of time points with significant difference for each morphological ECG parameter within a day between the HF patients and HCs (nHF=57 and nHC=75). A P value <0.0012 (0.05/42) was considered statistically significant. ECG, electrocardiogram; HC, healthy control; HF, heart failure; SCDHR, sudden cardiac death high-risk.

The distribution of P values across all 1,440 segments for these parameters were shown in Figure S1.

SCDHR vs. HF group

Fourteen ECG parameters exhibited significant differences (P<0.0012) in at least one time segment (Figure 3B,3C). Parameters prominently associated with the S wave included:

• Maximum S-wave amplitude (maxS);
• Average combined area of R and S waves minus the T wave (inteRST);
• Average combined area of S and T waves (inteST);
• Total duration between S-wave onset and offset per minute (t_S_sum).

These parameters demonstrated showed significant differences across more than 10 times segments. The distribution of P values for these parameters across the 1,440 segments is depicted in Figure S2.

HF vs. HC group

Distinct differences between the HF and HC groups were predominantly observed within the 21:00–06:00 time window (Figure 3D). A similar phenomenon was observed in the independent test set (Figure S3). Ten parameters exhibited significant differences (P<0.0012) across more than 20-time segments (Figure 3E), including:

• inteST;
• Minimum S-wave area per minute (inteSm);
• inteS_mean;
• Average S-wave amplitude per minute (mean_S);
• Total S-wave area per minute (inteS_sum);
• maxS;
• Minimum S-wave amplitude per minute (minS);
• inteRST;
• inteRS;
• R-wave onset-to-T-wave offset interval (t_RT).

Nine of these parameters were related to the S wave. Figure S4 illustrates the P values distribution for these parameters. Specifically, four S-wave area-related parameters (inteST, inteSm, inteS_mean, inteS_sum) showed consistent significant differences across nearly all time segments within the 21:00–06:00 interval.

Logistic regression analysis of ECG parameters

To exclude redundant parameters, before multivariable logistic regression analysis, all 42 ECG parameters were categorized into several parameter lusters based on their similarity using K-means clustering and only one parameter was selected from each cluster for multivariable logistic regression analysis (Appendix 1, Supplementary methods). The clustering results are shown in Figure S5 and the best cluster number is 8, meaning that ECG parameters were divided into 8 lusters for each time segment. Within each cluster, the optimal parameter was selected according to the lowest AIC value.

SCDHR vs. HC group

To evaluate the predictive ability of ECG parameters, a multivariable logistic regression model was developed for each time segment in the training set. These individual models were subsequently combined into a single comprehensive model using a greedy-search method.

We found that a combination of 10 time segments (16:06, 14:37, 21:39, 17:32, 17:28, 17:15, 14:18, 19:23, 13:50, and 9:31) and five ECG parameters—(I) inteS_mean; (II) t_T; (III) average value of R peak value in one minute (mean_R); (IV) minimum absolute value of negative T peak in one minute (T_minus) and (V) t_RT—demonstrated optimal performance. This model achieved an AUC of 0.887 in the training set (Figure S7A and Figure 4A) and 0.747 in the independent test set (Figure 4A and Table 3). Additional performance metrics included an accuracy of 0.755, a precision of 0.800, a recall of 0.444, and F1 score of 0.571 (Table 3).

Figure 4 Logistic regression model analysis for discriminating between SCDHR patients and HCs. (A) The ROC curves of the combined model for discriminating between SCDHR patients and HCs (nSCDHR=18 and nHC=33). (B) The OR values of individual features for SCDHR in the training set at significant time fragments (nSCDHR=25). (C) The OR values of individual features associated with SCDHR in the independent dataset at multiple time points (nSCDHR=18). (D) The difference in these features between the SCDHR patients and HCs at these time fragments was tested by the Student’s t-test (nSCHDR=18 and nHC=33). The OR values indicate the change in risk when the feature changes by 1% unit. HC, healthy control; HF, heart failure; OR, odds ratio; ROC, receiver operating characteristic; SCDHR, sudden cardiac death high-risk; Test, independent test set; Training, training set.

In univariable logistic regression analysis, the five ECG parameters exhibited significant associations with SCDHR group at above 10 identified time points after adjustment of P value by the Benjamini and Hochberg method (19) (Figure 4B). Specifically, significant associations were observed for: (I) inteS_mean at 16:06, 19:23, and 21:39; (II) t_T at 19:23; (III) mean_R at 21:39; (IV) T_minus at 16:06; and (V) t_RT at 13:50, 14:18, 17:28, and 19:23 in the training set. Although, in the independent testing set, inteS_mean and t_T did not reach statistical significance at some of these time points (Figure 4C), inteS_mean at 21:39 and t_T at 19:23 significantly differed between the SCDHR and HC groups in both the training and independent test set (Figure 4D). In addition, the increase in inteS_mean at 16:06 and 19:23 showed potential for distinguishing SCDHR participants from controls.

SCDHR vs. HF group

Using a similar method, a combination of five ECG parameters—(I) maximum area of T peak in one minute (inteTM); (II) inteSm; (III) average area of the T peak in one minute (inteT_mean); (IV) t_T and (V) t_S_sum—at three specific time segments (10:56, 14:38, and 14:51) were identified as the optimal discriminators between the SCDHR and HF groups, yielding an AUC of 0.857 in the training set (Figures S7B,S8) and 0.714 in the independent test set (Figure S8 and Table 3). Other effectiveness indicators inteSm and inteS_mean consistently showed significant differences are summarized in Table 3. Notably, simultaneous increases in both parameters at 23:33, 23:39, 23:40, 23:46, 0:15, 1:06, and 05:55 was strongly associated with HF. However, in univariable analysis, no significant correlation was observed in the independent test set (Figure S6A,S6B), except for inteSm at 14:38, which demonstrated potential as a distinguishing marker (Figure S6C).

HF vs. HC group

A combination of six ECG parameter—(I) total R peak area in one minute (inteR_sum); (II) maximum absolute value of negative T peak in one minute (T_plus); (III) inteSm; (IV) mean_R; (V) inteS_mean and (VI) inteST—across seven time points (0:15, 1:06, 5:55, 23:33, 23:39, 23:40, and 23:46) demonstrated optimal performance in differentiating HF participants from HCs. The model achieved an AUC of 0.965 in the training set (Figure S9 and Figure 5A) and 0.842 with an accuracy of 0.704 in the independent test set. Other effectiveness indicators are presented in Figure 5A and Table 3.

Figure 5 Logistic regression model analysis for discriminating between HF patients and HCs. (A) The ROC curves of the combined model for discriminating between HF patients and HCs (nHF=81 and nHC=33). (B) The OR values of individual features for HF patients in the training set at significant time fragments (nHF=57). (C) The OR values of individual features associated with HF patients and HCs in the independent set at multiple time points (nHC=33 and nHF=81). (D) The difference in these features between the HF patients and HCs at these time fragments was tested by the Student’s t-test (nHC=75 and nHF=57). The OR values indicate the change in risk when the feature changes by 1% unit. HC, healthy control; HF, heart failure; OR, odds ratio; ROC, receiver operating characteristic; Test, independent test set; Training, training set.

In univariable analysis of both the training and independent test sets, inteSm and inteS_mean consistently showed significant differences (Figure 5B,5C). Notably, simultaneous increases in both parameters at 23:33, 23:39, 23:40, 23:46, 0:15, 1:06, and 05:55 was strongly associated with HF. They both showed significant differential expression between HF and HC group in each of seven time point (Figure 5D).

Finally, we evaluated the clinical performance of three logistics models. For the comparison pair of SCDHR and HC, in training set, when threshold probability was in the range of 0.1–0.9, clinical outcomes significantly tended to be more biased towards SCDHR group and suggested a well performance in clinical practice (Figure S10A). Calibration curve for training set showed the predicted value of the model was basically consistent with its measured value and they both closed to the ideal curve, indicating good predictive performance (Figure S10B). In the independent test set, the correlation results of the two curves were similar to those in training set (Figure S10C,S10D). For the comparison pair of SCDHR and HF, decision curve analysis showed significantly tended to be more biased towards SCDHR group when threshold probability larger than 0.1 and calibration curve showed basically consistent the predicted and measured value in training set (Figure S11A,S11B). In independent testing set, the effective threshold range displayed by decision curve analysis was 0.1–0.9 (Figure S11C) and calibration curve also showed basically consistent the predicted and measured value (Figure S11D). The clinical performance of logistics models for HF and HC was shown in Figure S12. Decision curve analysis was significantly tended to be more biased towards HF group when threshold probability larger than 0.1 and calibration curve showed good predictive performance in both training set and independent test set.

Discussion

In this study, we observed a significant increase in the mean S-wave area on ECG in both patients with a history of SCDHR and HF compared to HCs. This observation suggests that an increase in the mean S-wave area is a robust parameter for distinguishing individuals with prior SCD or those at high risk of SCD, with a strong correlation to with SCD incidence. Furthermore, our analysis revealed distinct circadian rhythms in the S-wave area. Specifically, in participants with HF, the mean S-wave area predominantly increased during the late-night to early-morning period (23:30–06:00). In contrast, in the SCDHR group, this increase primarily occurred from late afternoon to bedtime (16:00–22:00). These time-dependent variations may have important implications for understanding and managing the SCD risk in these organic patient populations.

Summary of findings and clinical significance

Circadian rhythms of the S-wave area and underlying physiological mechanisms

Our study highlights the clinical significance of S-wave area-related ECG parameters, particularly inteSm and inteS_mean, in patients at high risk of SCD and HF. We identified significant circadian variations in these parameters, indicating intrinsic physiological rhythms in cardiac electrical activity. Such variation may be influenced by circadian regulation of autonomic nervous system activity (20), fluctuations in sympathetic (21,22) and parasympathetic one, and changes in L-type calcium channel conductance (23). Additionally, circadian shifts in blood volume and venous reflux may also contribute to these observed variations (24).

Previous studies have demonstrated increased the risk of cardiac events, such as myocardial infarction and cardiac arrest, predominantly during the early morning hours (25-27). Our findings align with these observations, highlighting specific periods of increased S-wave area as potential markers of heightened cardiac electrical instability.

An increase in the S-wave area may reflect alterations in the electrophysiological properties of myocardial cell membranes, particularly conduction delays in the right ventricular outflow tract due to abnormal depolarization. Such conduction delays suggest an increased the likelihood of cardiac electrical instability during specific circadian periods (28-31). Current evidence suggests that reactive myocardial interstitial fibrosis, induced by altered expression of gap junction proteins such as decreased expression of SCN5A, may be one possible mechanism underlying these electrophysiological changes and the observed increase in ECG S-wave area (32-36). These insights lay a foundation for further investigations into circadian rhythms of cardiac electrical activity and may help the development of the time-specific cardiac protection strategies.

Comparison of logistic regression with traditional methods

We compared logistic regression models with traditional methods, specifically LASSO regression to evaluate their effectiveness in analyzing dynamic ECG parameters associated with SCD risk. The logistic regression model demonstrated superior performance compared with LASSO regression in differentiating SCD risk groups (Appendix 1, Supplementary methods). Specifically, logistic regression achieved higher AUC values than LASSO regression in distinguishing between the groups:

• SCDHR vs. HC: logistic regression AUC =0.887 vs. LASSO regression AUC =0.767.
• SCDHR vs. HF: logistic regression (AUC =0.857) vs. LASSO (AUC =0.714).
• HF vs. HC: logistic regression (AUC =0.965) vs. LASSO (AUC =0.849).

These results validate the robustness of logistic regression for analyzing the dynamic relationship between ECG parameters and SCD risk.

Clinical application

Personalized monitoring and intervention strategies

The identification of circadian rhythm changes in the S-wave area-related parameters offers opportunities for designing personalized monitoring and intervention programs in high-risk patients. By focusing on periods of heightened electrical instability, clinicians can implement targeted preventive measures to reduce the risk of SCD in high-risk patients.

Integration with wearable devices and remote monitoring systems

Advancements in wearable devices and remote monitoring systems enable the real-time collection of dynamic ECG data. These systems can autonomously detect abnormal signals in ECG, and alert both patients and healthcare professionals to take immediate action. Personalized risk assessment tools derived from dynamic ECG data have the potential to improve disease management and optimize treatment plans, ultimately reducing SCD incidence.

Study limitations and future research directions

Study limitations

Despite the significant findings, this study had several limitations. First, the relatively small sample size, particularly in the independent test set, may limit the generalizability of the results. Additionally, the study population, recruited from four major medical centers in China, may not represent broader geographic and racial diversity, further restricting the generalizability of the findings. Therefore, larger, more diverse cohorts are required to validate these findings. Second, since only cross-sectional data was utilized, this limited the ability to evaluate causality between S-wave area changes and the development or progression of SCD. Longitudinal studies are needed to confirm the temporal relationship. Third, this study primarily focused on ECG parameters and their circadian rhythms without examining clinical outcomes such as actual incidence rates of SCD, hospitalization, or other cardiac events, which would strengthen the practical implications of the findings. Last but not least, other clinical factors, such as the presence of arrhythmias, medication use (e.g., beta-blockers), or underlying structural heart disease, were not thoroughly examined for their potential impact on S-wave area parameters. Future research addressing these potential confounders is encouraged to solve this limitation.

Future research directions

To translate these research findings into clinical practice, future research should prioritize multicenter, randomized controlled trials to validate the applicability of these ECG parameters across diverse populations. Our research team has incorporated the identified ECG parameters into analysis modules for wearable ECG devices (e.g., wearable ECG detection vests and 24-hour dynamic ECG recorders). These tools will undergo prospective validation to evaluate their effectiveness and reliability in predicting SCD events in real-world clinical settings.

Conclusions

This study demonstrated that S-wave area-related ECG parameters exhibited significant circadian variations, reflecting the intrinsic physiological rhythms of cardiac electrical activity. These parameters displayed robust discriminative capability in distinguishing high-risk populations for SCD from HCs. Moreover, circadian rhythms appeared closely related to previously reported peak periods of cardiac events, suggesting a potential intrinsic connection between ECG changes and circadian timing. After further validation in larger, diverse cohorts, these time-dependent ECG parameters—particularly those involving the S-wave area—may enhance personalized, time-targeted strategies for predicting and preventing SCD.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-11/rc

Data Sharing Statement: Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-11/dss

Peer Review File: Available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-11/prf

Funding: This work was partly funded by the National Key R&D Program of China (No. 2020YFB0204803 to J.W.). This work was also supported by the General Program of National Natural Science Foundation of China (NSFC) (Nos. 82070237 to J.W., 81870170 to J.W., 81970200 to Y.C., 81770229 to Y.C., 81970388 to Y.Z., 81903299 to Q.C.), the Guangzhou Health and Medical Collaborative Innovation Major Project (No. 201803040010 to J.W.), the Guangdong Provincial Laboratory of Regenerative Medicine and Health (No. 198F041814 to J.W.), and the Natural Science Foundation of Guangdong (No. 2019A1515011682 to Y.Z.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cdt.amegroups.com/article/view/10.21037/cdt-2025-11/coif). Q.C. reports funding support from The General Program of National Natural Science Foundation of China (NSFC) (No. 81903299). Y.C. reports funding support from the General Program of National Natural Science Foundation of China (NSFC) (Nos. 81970200 and 81770229). Y.Z. reports funding support from The General Program of National Natural Science Foundation of China (NSFC) (No. 81970388) and The Natural Science Foundation of Guangdong (No. 2019A1515011682). J.W. reports funding support from The National Key R&D Program of China (No. 2020YFB0204803), The General Program of National Natural Science Foundation of China (NSFC) (Nos. 82070237 and 81870170), The Guangzhou Health and Medical Collaborative Innovation Major Project (No. 201803040010), and The Guangdong Provincial Laboratory of Regenerative Medicine and Health (No. 198F041814). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study protocol was approved by the study team leader unit, Medical Ethics Committee of Sun Yat-sen Memorial Hospital, Sun Yat-sen University (Ethics number: SYSKY-2023-688-01) and strictly adhered to the Declaration of Helsinki and its subsequent amendments. All participating hospitals/institutions were informed and agreed with this study. All patients provided written informed consent.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Tsao CW, Aday AW, Almarzooq ZI, et al. Heart Disease and Stroke Statistics-2022 Update: A Report From the American Heart Association. Circulation 2022;145:e153-639. [Crossref] [PubMed]
2. Hu S. Report On Cardiovascular Diseases In China (2018). Beijing: Encyclopedia of China Publishing House; 2019.1.
3. Shi HT, Ge JB. Improving public defibrillator use in China. Lancet 2016;388:1156-7. [Crossref] [PubMed]
4. Zeppenfeld K, Tfelt-Hansen J, de Riva M, et al. 2022 ESC Guidelines for the management of patients with ventricular arrhythmias and the prevention of sudden cardiac death. Eur Heart J 2022;43:3997-4126. [Crossref] [PubMed]
5. Al-Khatib SM, Stevenson WG, Ackerman MJ, et al. 2017 AHA/ACC/HRS Guideline for Management of Patients With Ventricular Arrhythmias and the Prevention of Sudden Cardiac Death: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines and the Heart Rhythm Society. Circulation 2018;138:e272-391. [Crossref] [PubMed]
6. Waks JW, Sitlani CM, Soliman EZ, et al. Global Electric Heterogeneity Risk Score for Prediction of Sudden Cardiac Death in the General Population: The Atherosclerosis Risk in Communities (ARIC) and Cardiovascular Health (CHS) Studies. Circulation 2016;133:2222-34. [Crossref] [PubMed]
7. Di Marco A, Brown PF, Bradley J, et al. Improved Risk Stratification for Ventricular Arrhythmias and Sudden Death in Patients With Nonischemic Dilated Cardiomyopathy. J Am Coll Cardiol 2021;77:2890-905. [Crossref] [PubMed]
8. Theerasuwipakorn N, Chokesuwattanaskul R, Phannajit J, et al. Impact of late gadolinium-enhanced cardiac MRI on arrhythmic and mortality outcomes in nonischemic dilated cardiomyopathy: updated systematic review and meta-analysis. Sci Rep 2023;13:13775. [Crossref] [PubMed]
9. Albert CM, Chae CU, Grodstein F, et al. Prospective study of sudden cardiac death among women in the United States. Circulation 2003;107:2096-101. [Crossref] [PubMed]
10. Vera-Sarmiento HL, Tanriverdi T, Hurtado-de-Mendoza D, et al. QRS prolongation is associated with associated with adverse cardiac remodeling in hypertrophic cardiomyopathy. J Electrocardiol 2024;87:153818. [Crossref] [PubMed]
11. Sarquella-Brugada G, García-Algar O, Zambrano MD, et al. Early Identification of Prolonged QT Interval for Prevention of Sudden Infant Death. Front Pediatr 2021;9:704580. [Crossref] [PubMed]
12. Attachaipanich T, Attachaipanich S, Kaewboot K. Fragmented QRS complex as a predictor of in-hospital life-threatening arrhythmias in myocardial infarction patients: A systematic review and meta-analysis. Indian Heart J 2025;77:142-50. [Crossref] [PubMed]
13. Calò L, Giustetto C, Martino A, et al. A New Electrocardiographic Marker of Sudden Death in Brugada Syndrome: The S-Wave in Lead I. J Am Coll Cardiol 2016;67:1427-40. [Crossref] [PubMed]
14. Giustetto C, Nangeroni G, Cerrato N, et al. Ventricular conduction delay as marker of risk in Brugada Syndrome. Results from the analysis of clinical and electrocardiographic features of a large cohort of patients. Int J Cardiol 2020;302:171-7. [Crossref] [PubMed]
15. Batur MK, Aksöyek S, Oto A, et al. Circadian variations of QTc dispersion: is it a clue to morning increase of sudden cardiac death? Clin Cardiol 1999;22:103-6. [Crossref] [PubMed]
16. Kusama Y, Kodani E, Nakagomi A, et al. Variant angina and coronary artery spasm: the clinical spectrum, pathophysiology, and management. J Nippon Med Sch 2011;78:4-12. [Crossref] [PubMed]
17. Aizawa Y, Takatsuki S, Kaneko Y, et al. Comparison of circadian, weekly, and seasonal variations of electrical storms and single events of ventricular fibrillation in patients with Brugada syndrome. Int J Cardiol Heart Vasc 2016;11:104-10. [Crossref] [PubMed]
18. Hartigan JA, Wong MA. Algorithm AS 136: A K-Means Clustering Algorithm. Appl Stat 1979;28:100-8. [Crossref]
19. Haynes W. Benjamini–Hochberg Method. In: Dubitzky W, Wolkenhauer O, Cho K-H, Yokota H, editors. Encyclopedia of Systems Biology. New York, NY: Springer New York; 2013. p. 78.
20. Fukuda K, Kanazawa H, Aizawa Y, et al. Cardiac innervation and sudden cardiac death. Circ Res 2015;116:2005-19. [Crossref] [PubMed]
21. Behrens S, Ehlers C, Brüggemann T, et al. Modification of the circadian pattern of ventricular tachyarrhythmias by beta-blocker therapy. Clin Cardiol 1997;20:253-7. [Crossref] [PubMed]
22. Huang TC, Lin SJ, Chen CJ, et al. Skin sympathetic nerve activity and ventricular arrhythmias in acute coronary syndrome. Heart Rhythm 2022;19:1613-9. [Crossref] [PubMed]
23. Diekman CO, Wei N. Circadian Rhythms of Early Afterdepolarizations and Ventricular Arrhythmias in a Cardiomyocyte Model. Biophys J 2021;120:319-33. [Crossref] [PubMed]
24. Pasqualetti P, Casale R. Daily distribution of episodes of acute cardiogenic pulmonary edema. Cardiology 1997;88:509-12. [Crossref] [PubMed]
25. Yao PC, Li MH, Chen M, et al. Circadian variation pattern of sudden cardiac arrest occurred in Chinese community. Open Heart 2024;11:e002904. [Crossref] [PubMed]
26. Tang Y, Tertulien T, Bhonsale A, et al. Comparison of Circadian Variation for In-Hospital Versus Out-of-Hospital Sudden Cardiac Arrest Survivors. Am J Cardiol 2021;160:1-7. [Crossref] [PubMed]
27. Tran DT, St Pierre Schneider B, McGinnis GR. Circadian Rhythms in Sudden Cardiac Arrest: A Review. Nurs Res 2021;70:298-309. [Crossref] [PubMed]
28. Cortez D, Svensson A, Carlson J, et al. The S-wave angle identifies arrhythmogenic right ventricular cardiomyopathy in patients with electrocardiographically concealed disease phenotype. J Electrocardiol 2018;51:1003-8. [Crossref] [PubMed]
29. Peréz-Riera AR, Ferreira Filho C, de Abreu LC, et al. Do patients with electrocardiographic Brugada type 1 pattern have associated right bundle branch block? A comparative vectorcardiographic study. Europace 2012;14:889-97. [Crossref] [PubMed]
30. Yoon N, Jeong HK, Lee KH, et al. Right Ventricular Longitudinal Conduction Delay in Patients with Brugada Syndrome. J Korean Med Sci 2021;36:e75. [Crossref] [PubMed]
31. Monaco C, Eltsov I, Del Monte A, et al. Assessment of activation delay in the right ventricular outflow tract as a potential complementary diagnostic tool for Brugada Syndrome. Europace 2025;27:euaf093. [Crossref] [PubMed]
32. Coronel R, Casini S, Koopmann TT, et al. Right ventricular fibrosis and conduction delay in a patient with clinical signs of Brugada syndrome: a combined electrophysiological, genetic, histopathologic, and computational study. Circulation 2005;112:2769-77. [Crossref] [PubMed]
33. Miles C, Asimaki A, Ster IC, et al. Biventricular Myocardial Fibrosis and Sudden Death in Patients With Brugada Syndrome. J Am Coll Cardiol 2021;78:1511-21. [Crossref] [PubMed]
34. Nademanee K, Raju H, de Noronha SV, et al. Fibrosis, Connexin-43, and Conduction Abnormalities in the Brugada Syndrome. J Am Coll Cardiol 2015;66:1976-86. [Crossref] [PubMed]
35. Chen GX, Barajas-Martínez H, Ciconte G, et al. Clinical characteristics and electrophysiologic properties of SCN5A variants in fever-induced Brugada syndrome. EBioMedicine 2023;87:104388. [Crossref] [PubMed]
36. Chaloemthanetphong A, Choowongkomon K, Worrapitirungsi W, et al. SCN5A missense variants and their contribution to deaths in Sudden Unexplained Nocturnal Death Syndrome (SUNDS). Forensic Sci Int Genet 2025;76:103237. [Crossref] [PubMed]