The study, led by Kaiyuan Cen of Guidong People’s Hospital of Guangxi Zhuang Autonomous Region and Zhuoqiao He of the First Affiliated Hospital of Shantou University Medical College, was a single-centre retrospective cohort study of adults hospitalized with chronic heart failure between 31 May 2023 and 31 May 2024, with follow-up data locked on 31 May 2025. Of 201 patients screened, 160 met the inclusion criteria, and the clinical stakes of the exercise were immediately apparent: 68 of those 160 patients—42.5 per cent—experienced a major adverse cardiovascular event, or MACE, within a year of their index hospitalization. That figure underscores a persistent problem in cardiology. Chronic heart failure remains a condition of high and unevenly distributed risk, and the field has long lacked tools that can reliably separate the patient who will be readmitted or die within months from the one who will remain stable on guideline-directed therapy.

The technical core of the new approach lies in the transformation of the conventional 12-lead ECG into a vectorcardiogram using the Kors method, a well-established mathematical technique that reconstructs the heart’s electrical dipole as a rotating vector in three-dimensional space. Rather than viewing the heart’s depolarization and repolarization through the fixed projections of twelve surface electrodes, the vectorcardiogram traces the path of the cardiac electrical axis as a loop in the frontal, horizontal, and sagittal planes. From this reconstruction, the researchers extracted features that are difficult or impossible to appreciate on a standard tracing: the spatial angles between the QRS complex, which represents ventricular depolarization, and the T wave, which represents repolarization. Discordance between these two processes—measured as a wide QRS–T angle in any plane—is thought to reflect abnormal ventricular conduction and repolarization heterogeneity, electrophysiological substrate that has been linked in prior studies to arrhythmic death and adverse remodelling.

The prespecified primary model was deliberately parsimonious, incorporating just six predictors: left ventricular end-diastolic diameter (LVEDD) measured by echocardiography, New York Heart Association functional class, the frontal, horizontal, and sagittal QRS–T angles, and a binary morphological feature known as the QRS-loop reversal/U-turn sign. The latter is a qualitative abnormality in which the ventricular depolarization loop reverses its direction of rotation or executes a U-shaped turn, signalling aberrant conduction pathways. Notably, the investigators evaluated whether two of the most celebrated markers in heart failure prognostication—brain natriuretic peptide (BNP) and left ventricular ejection fraction (LVEF)—added value beyond the six-predictor set, embedding them in full-model, comparator-model, and incremental-value analyses. Because the prediction target was a fixed 12-month probability rather than a time-to-event hazard, the team used multivariable logistic regression as the primary modelling framework, an appropriate choice for a binary endpoint observed over a uniform window.

The performance figures reported in the paper are striking. The six-predictor model produced an apparent AUC of 0.946, with a 95 per cent confidence interval of 0.914 to 0.978. Recognizing that apparent performance on the development dataset invariably overstates true predictive ability, the researchers subjected the model to bootstrap internal validation using 1,000 resamples. This procedure yielded an AUC optimism estimate of just 0.012, leaving an optimism-corrected AUC of 0.934. Calibration was assessed with equal rigour: the apparent Brier score of 0.091 rose modestly to 0.106 after bootstrap correction, calibration-in-the-large was −0.009, and the calibration slope was corrected from a perfect 1.000 to 0.852, with the same uniform shrinkage factor applied to the model’s coefficients to guard against overfitting in future applications. A shrinking factor of 0.852 means each predictor’s coefficient is tempered by roughly fifteen per cent, a standard penalty that trades a small loss in apparent fit for improved generalizability.

Perhaps the most clinically persuasive result came from the incremental-value analyses. When the VCG-derived features were added to a conventional base model built on standard clinical predictors, the AUC rose from 0.890 to 0.955—a difference of 0.065 that reached statistical significance at DeLong P = 0.001. The augmented model also achieved a lower Brier score and favourable discrimination and reclassification indices, indicating that it did not merely rank patients differently but genuinely moved them into more accurate risk categories. Decision-curve analysis, a method that evaluates the net clinical benefit of a model across a range of risk thresholds, suggested that the VCG-augmented approach would deliver higher net benefit than selected single-marker comparators within the development cohort. In practical terms, this means that at most clinically meaningful threshold probabilities, acting on the model’s predictions would identify more true events and generate fewer false alarms than relying on any single conventional marker alone.

The biological rationale for why these VCG features carry such prognostic weight is grounded in decades of electrophysiological research. Prior work has demonstrated that a wide spatial QRS–T angle predicts cardiac death in the general population, that the frontal QRS–T angle predicts increased morbidity and mortality in chronic heart failure, and that vectorcardiographic findings are associated with recurrent ventricular arrhythmias in patients with implantable cardioverter-defibrillators. The QRS–T angle quantifies the degree of spatial discordance between the sequence of ventricular activation and the sequence of recovery—discordance that widens as conduction disease, ischaemia, and structural remodelling accumulate in the failing myocardium. The QRS-loop reversal/U-turn sign adds morphological information about the activation pathway itself, capturing abnormalities such as conduction delay and scar-related altered depolarization that angle measures alone may miss. Together with echocardiographic ventricular dimensions and symptomatic status, these features sketch a compact portrait of both the structure and the electrophysiology of the failing heart.

The authors are careful, appropriately so, to frame the work as a development and internal validation study rather than a demonstration of clinical readiness. Internal validation by bootstrap quantifies how much a model’s apparent performance is inflated by overfitting to its own data, but it cannot address the questions that matter most before deployment: whether the model generalizes to patients from different hospitals, ethnic backgrounds, and health systems; whether ECG acquisition and processing pipelines elsewhere would yield comparable VCG reconstructions; and whether clinicians acting on the model’s outputs would genuinely alter management in ways that improve outcomes. The de-identified analytic dataset cannot be made publicly available owing to institutional privacy requirements, though statistical code and data excerpts are available from the corresponding author on reasonable request. The study received no external funding, was approved by the Ethics Committee of Guidong People’s Hospital under approval number GDKY202590, and the authors declare no competing interests.

Even so, the appeal of the approach is hard to overstate. It requires no new hardware, no additional blood draws, and no expensive imaging: the raw material is a routine ECG already recorded for virtually every hospitalized heart failure patient, and the transformation to a vectorcardiogram is a computational step that can be automated in seconds. In an era when deep learning models have shown that ECG voltage data alone can predict mortality, the present work occupies a complementary middle ground—using interpretable, physiologically grounded features extracted from the same inexpensive signal, within a transparent logistic regression framework whose coefficients clinicians can inspect rather than a black box they must trust blindly. Each of the six predictors maps onto a familiar clinical concept: ventricular size, symptom severity, ventricular conduction, and repolarization geometry.

The study team, which also included Hong Chen of the University of Tsukuba, Yi Tan of Guidong People’s Hospital, Lixia Lin of Guangxi University of Chinese Medicine, and Xiaojuan Xu of Tongji University, emphasizes that external multicentre validation is required before any clinical implementation. That caveat is the correct one, and it sets the agenda for the next stage of this line of research. If the 0.934 optimism-corrected AUC survives contact with independent cohorts, heart failure teams could gain a near-zero-cost decision support tool for flagging the roughly four in ten hospitalized patients who face a major cardiovascular event within a year—enabling intensified follow-up, earlier escalation of therapy, and closer surveillance of the patients whose electrical signatures betray a heart in the greatest danger.

Heart failure is not a single disease. Clinicians commonly classify it according to the left ventricular ejection fraction, the proportion of blood pumped out of the heart’s main chamber with each contraction. In heart failure with mildly reduced ejection fraction, known as HFmrEF, the ejection fraction lies between 41 and 49 percent. This intermediate category accounts for roughly one-quarter of heart-failure patients, but people in it have historically been underrepresented in clinical trials and large registries. That has left physicians with less specific evidence to guide prognosis and treatment. The heart’s pumping ability in HFmrEF is impaired, but not to the extent seen in heart failure with reduced ejection fraction, while the biological processes affecting patients may overlap with both reduced and preserved ejection-fraction syndromes. Inflammation is one possible link. When the heart is strained, damaged or congested, immune and vascular systems can become activated, potentially amplifying tissue injury, impairing blood-vessel function and accelerating adverse remodeling of the heart. The new findings place an accessible inflammation marker at the center of risk assessment for this intermediate phenotype.

CRP is produced mainly by the liver after stimulation by inflammatory signaling molecules, particularly interleukin-6. It is part of the body’s acute-phase response and can rise sharply during bacterial infections, tissue injury and other inflammatory conditions. But CRP is not simply an on-off indicator of infection. Small, persistent elevations can reflect a broader state of immune activation associated with obesity, kidney disease, diabetes, vascular disease, chronic lung conditions and cardiovascular stress. In heart failure, inflammatory cytokines such as interleukin-6, tumor necrosis factor-alpha and interleukin-1 beta may contribute to a self-reinforcing cycle. They can influence vascular tone, reduce nitric-oxide availability, alter cardiac muscle function and promote structural changes in the myocardium. CRP may also participate in innate immune signaling, including activation of the classical complement pathway, although it remains uncertain whether CRP directly drives worsening heart failure or primarily serves as a highly useful indicator of underlying biological stress. The distinction matters: a prognostic marker can identify danger without necessarily being an effective drug target.

The researchers analyzed data from the HARMER registry, a retrospective, single-center registry of consecutive adults hospitalized with HFmrEF at University Medical Centre Mannheim in Germany between January 2016 and December 2022. To qualify, patients had to have symptoms or signs of heart failure and a left ventricular ejection fraction of 41 to 49 percent documented by standardized transthoracic echocardiography during the hospitalization. Of 2,228 patients initially identified, 44 were lost to follow-up, leaving 1,978 individuals for the primary analysis. Their median CRP concentration was 13.3 mg/L, with an interquartile range of 3.5 to 43.7 mg/L. The investigators divided patients into four predefined groups: below 5 mg/L, 5 to less than 10 mg/L, 10 to less than 50 mg/L, and 50 mg/L or higher. They then compared outcomes using Kaplan–Meier survival analyses and Cox proportional-hazards models, adjusting for factors including age, sex, body-mass index, coronary artery disease, chronic kidney disease, diabetes, acute decompensated heart failure, ischemic cardiomyopathy and anemia. CRP was also analyzed as a continuous variable after logarithmic transformation, a statistical approach that reduces the influence of its highly skewed distribution.

The raw outcome differences were striking. During the 30-month follow-up, death occurred in 17.0 percent of patients with CRP below 5 mg/L. The corresponding proportions were 30.1 percent among those with CRP between 5 and less than 10 mg/L, 36.4 percent among those with CRP from 10 to less than 50 mg/L, and 46.8 percent among those with CRP of at least 50 mg/L. Before adjustment, the highest CRP category was associated with a hazard ratio of 3.571 compared with the lowest category. After accounting for clinical characteristics, the association weakened but remained statistically robust: hazard ratios were 1.720 for 5 to less than 10 mg/L, 1.813 for 10 to less than 50 mg/L, and 2.275 for at least 50 mg/L. In the continuous analysis, each increase in the natural logarithm of CRP was associated with a hazard ratio of 1.250 for mortality. These results indicate a graded relationship rather than a single threshold at which risk suddenly appears. Even a mild elevation was informative.

Patients with higher CRP also tended to be older and more likely to have chronic kidney disease, diabetes, anemia, worse kidney-function measurements and more severe heart-failure symptoms. Their median NT-proBNP concentrations—a marker of cardiac wall stress—were higher, rising from 1,164 pg/mL in the lowest CRP group to 4,914 pg/mL in the highest. Moderate-to-severe mitral and tricuspid regurgitation were more common, and acute decompensated heart failure increased stepwise across the CRP categories. These patterns illustrate why the researchers used multivariable adjustment: high CRP can be a sign of several coexisting conditions rather than an isolated heart-failure mechanism. Even after these potential confounders were considered, the association with mortality persisted. The relationship was also broadly consistent across men and women, older and younger patients, and people with or without ischemic cardiomyopathy, although the strength of the association varied. In patients aged 70 or older, even the lowest elevated CRP category remained associated with increased risk, while among younger patients the clearest signal appeared at levels of 50 mg/L or above.

To test whether the results were being driven by obvious sources of inflammation, the investigators conducted a second analysis that excluded patients with infectious disease, acute myocardial infarction, stroke, malignancy, cardiogenic shock, cardiac arrest or rheumatic disease at admission. This produced a more narrowly defined cohort of 741 patients. The pattern survived the stricter test. Mortality occurred in 15.6 percent of patients with CRP below 5 mg/L, compared with 34.7 percent of those with CRP from 5 to less than 10 mg/L, 33.3 percent of those with CRP from 10 to less than 50 mg/L, and 38.9 percent of those with CRP at least 50 mg/L. In the adjusted model, the hazard ratios remained elevated at 2.370, 1.684 and 2.212, respectively. The persistence of the signal after removing major inflammatory conditions supports the idea that low-grade inflammation associated with heart failure or its comorbidities may carry prognostic information. However, it does not prove that inflammation causes the deaths. The study was observational and retrospective, so unmeasured differences between patients could still account for part of the association.

CRP did not perform in the same way for every outcome. Higher levels were associated with more heart-failure-related rehospitalizations in unadjusted analyses, but the relationship disappeared after multivariable adjustment. Instead, rehospitalization was independently linked to factors such as age, body-mass index, prior congestive heart failure, acute decompensation, anemia, aortic stenosis and mitral regurgitation. This contrast may be biologically meaningful. Mortality over several years can reflect a broad state of systemic vulnerability involving frailty, organ dysfunction, vascular disease and reduced physiological reserve, whereas hospitalization for worsening heart failure may depend more directly on congestion, valve disease, treatment decisions and access to care. The study also found higher rates of in-hospital death, 12-month mortality and major adverse cardiac and cerebrovascular events among patients with elevated CRP. Still, CRP was measured only once during the index hospitalization, and rehospitalizations were captured only at the researchers’ center, potentially missing events treated elsewhere.

The authors emphasize that CRP could become a practical addition to long-term risk stratification because it is inexpensive, widely available and already measured in routine clinical laboratories. The study does not justify using a CRP value alone to make treatment decisions, nor does it show that lowering CRP would improve survival. Two different assay platforms were used during the study period, and post-discharge treatment changes were not available. The single-center design also limits how confidently the findings can be generalized to other populations. Prospective, multicenter studies will need to determine whether repeated CRP measurements improve prediction beyond established markers such as NT-proBNP and whether combining inflammatory, cardiac and metabolic measurements offers a clearer picture of risk. Most importantly, clinical trials will be required to test whether inflammation is a modifiable driver in HFmrEF. Anti-inflammatory therapies, including interleukin-1 beta inhibitors and colchicine, have reduced cardiovascular events in selected settings, but evidence for direct benefit in heart failure remains limited and inconsistent. For now, the message is simpler—and potentially more useful: in HFmrEF, a CRP result that looks only mildly abnormal may be an early warning of much greater long-term danger.

The research underscores a crucial intersection of healthcare and technology, establishing a framework for future studies involving wearable devices. These portable ECG devices are increasingly becoming integral to personal health monitoring, making the study’s findings both timely and significant. The AI model utilized in the research is designed to adapt and learn from varying noise levels typically associated with ECG data, a common issue that has historically hampered the accuracy of clinical assessments.

One of the study’s fascinating aspects is its emphasis on real-world applicability. By employing lead I ECGs, which are easier to use in non-clinical settings than traditional 12-lead systems, the model proposes that individuals can continuously monitor their heart health with devices readily available on the market. This capacity for ongoing monitoring significantly enhances the prospects for early detection and intervention in heart failure cases, ultimately leading to improved patient outcomes.

The implications of this study extend beyond mere diagnostic capabilities; it also suggests a paradigm shift toward personalized medicine. By employing AI for risk evaluation, clinicians may tailor preventative strategies and treatment regimens to individual patient profiles. This model’s proactive stance could shift the management of cardiovascular diseases from reactive to preventive, thereby alleviating the burden on healthcare systems worldwide.

Dr. Rohan Khera, the corresponding author of the study, emphasizes the necessity for further exploration in a prospective study setting. As promising as the findings are, he notes that extensive validation through real-world applications is essential before incorporating such AI-driven assessments into routine clinical practice. The rapid pace of technological advancement necessitates continued research, especially as we explore the ethical, practical, and clinical implications of integrating AI into everyday healthcare.

In this prospective study setting, wearable ECG devices could play a pivotal role in the future of heart health monitoring. The ability to gather real-time data significantly augments the traditional model of intermittent check-ups and provides a more comprehensive picture of an individual’s cardiovascular status. Importantly, the study highlights that these wearable devices are not merely consumer gadgets but may evolve into essential components of preventive cardiology.

Furthermore, the potential to analyze vast datasets garnered from these devices allows for more robust AI training. The model’s adaptability to various data inputs, including noise, positions it as a frontrunner in cardiovascular predictive analytics. This adaptability can enhance its reliability, making it a favorable option in diverse patient populations, particularly in low-resource settings where access to extensive clinical evaluations may be limited.

The study’s focus on heart failure is particularly critical, considering the escalating rates of cardiovascular diseases globally. Effective management and early detection of heart failure can significantly curb progression to severe disease stages, reducing hospitalization rates and associated healthcare costs. Notably, the escalating prevalence of heart failure underscores the urgency for innovative strategies, making this AI-ECG model a compelling contender in the future of cardiology.

In addition to its health implications, this study may catalyze further interest in the convergence of technology and medicine. As more researchers explore AI’s capabilities in diagnosing other medical conditions, the healthcare industry could witness a transformational shift in how various diseases are assessed and managed. The ongoing collaboration between tech developers and healthcare professionals could yield groundbreaking innovations that enhance patient care delivery.

As the medical community continues to embrace AI in healthcare applications, educating practitioners becomes paramount. Each new advancement must come with thorough training for healthcare providers to ensure effective implementation and patient safety. The robust nature of the AI-ECG model necessitates an understanding of its intricacies, including the handling of noise in data inputs and interpreting AI-generated risk scores.

In summary, this study has opened new avenues for heart failure risk stratification, indicating a significant step forward in patient-centered care. The potential to leverage wearable devices anchored by advanced AI models could reshape the healthcare landscape, fostering a proactive approach to managing cardiovascular health. The future holds promise, and as research endeavors continue, the integration of these technologies may lead to optimally tailored interventions for individuals at varying risk levels of heart failure.

The need for further studies and collaborations among stakeholders in this domain cannot be overstated, especially as acceptance of AI in clinical practice transitions from skepticism to standardization. As advancements unfold, a collective commitment to ethical standards and patient safety must guide the application of AI technologies in healthcare. The quest for knowledge, driven by a desire to understand better and mitigate health risks, remains the cornerstone of advancing healthcare solutions.

Effective communication about such studies, including their implications and future prospects, is essential to bridging the gap between theory and practice. As the scientific community analyzes these findings, engaging in robust discussions regarding their applicability, methodologies, effectiveness, and potential challenges ensures that the transition to AI-driven tasks will benefit from collective insights and expertise.

Through this lens, the study sheds light on a monumental shift in cardiovascular health assessment, further emphasizing the need for a holistic approach. As wearable technology garners traction in modern medicine, we may soon see the dawn of a new era in heart health that combines advanced technology with data science, ultimately aiming for optimized, personalized patient care in the realm of cardiology.

Subject of Research: Heart failure risk stratification using AI-adapted ECG models
Article Title: AI-Enhanced ECG Models for Heart Failure Risk Assessment: Revolutionizing Healthcare
News Publication Date: October 2023
Web References: www.jamacardio.com
References: doi:10.1001/jamacardio.2025.0492
Image Credits: JAMA Cardiology

Keywords

Heart failure, Artificial intelligence, Cohort studies, Wearable devices, Electrocardiography, Cardiology, Heart, Noise control, Risk management.