Epilepsy affects over 51 million people worldwide, carrying a mortality rate 2–3 times higher than the general population, largely attributable to sudden unexpected death in epilepsy (SUDEP) (21, 22). A previous cohort study of more than 26,000 participants from Taiwan National Health Insurance Research Database demonstrated that patients with epilepsy had 1.8-fold [Hazard Ratio (HR) 1.8, 95% CI 1.5–2.2] increased risk of SCD compared to those without epilepsy during up to 17 years of follow-up (23). A case-control study utilized data from a large scale general practitioner research database with 926 SCD cases and 9,832 controls further revealed that the severity of SCD risk in patients with epilepsy also depends on whether the epilepsy is symptomatic [Odd Ratio (OR) 5.8, 95% CI 2.1–15.6] or stable (OR 1.6, 95% CI 0.7–4.1), the presence of seizure episode triggers, and particularly the use of antiepileptic medications (AEM), which can potentially trigger SCD/SUDEP in certain situations (24). Sodium channel-blocking AEM particularly increase vulnerability (OR 2.8, 95% CI 1.1–7.2), with carbamazepine and gabapentin showing the strongest association (OR 3.2 and 5.7, respectively) (24). Moreover, patients with epilepsy are less likely to have witnessed cardiac arrests or receive bystander CPR, which in part contributes to their significantly lower survival rate compared to non-epileptic subjects (25, 26). Cardiac conduction dysfunction with increased QTc, calcium channel dysfunction, autonomic dysfunction with decreased heart rate variability (HRV), and hypoxemia due to respiratory distress during seizure episodes along with catecholamine surge are potential predisposing and peri-ictal factors that induce SCD/SUDEP in this population (27–29).

Mood disorders represent another critical neuropsychiatric link to SCD. Depression amplifies SCD risk by 1.6-fold (HR 1.6, 95% CI 1.4–1.9) according to a systematic review and meta-analysis of four studies with SCD as the primary outcome, encompassing over 83,000 patients (30). This association is particularly pronounced in the elderly, with a population-based study from Northern Finland demonstrating a 2.7-fold increased risk (HR 2.7, 95% CI 1.4–5.5) in individuals ≥70 years old over an 8-year follow-up period (31). Meanwhile, a cohort study of 1,012 ICD patients demonstrated that individuals with high anxiety levels based on the State-Trait Anxiety Inventory had a 1.9-fold increased risk of ventricular arrhythmia during the first year following ICD implantation compared to those with low anxiety levels (32). The anxiety-SCD association appears particularly prominent in women, as evidenced by a cohort study utilizing the Nurses' Health Study database. Among 72,359 women without baseline history of cardiovascular disease or cancer, those with phobic anxiety (Crown-Crisp index score ≥4) exhibited a 1.6-fold increased risk of SCD (33). Autonomic dysregulation is the key mechanism linking mood disorders to SCD through increased sympathetic and decreased parasympathetic activity, leading to reduced heart rate variability, and increased ventricular arrhythmia (31, 32). Moreover, a population-based study from Denmark reported that antidepressant use is a risk factor for SCD, with risk escalating from 1.6 times for short-term exposure (1–5 years) to 2.2 times for prolonged (6+ years) (34). Some types of antidepressants, particularly tricyclic antidepressants (TCA), have been associated with increased risk of SCD when the dose is high (≥300 mg of amitriptyline or its equivalent), with a 2.5 times increased risk (95% CI, 1.04–6.12) (35). TCAs have been proven to have significant dose-related adverse effects on the cardiovascular system, such as orthostatic hypotension, tachycardia, decreased heart rate variability, and QT prolongation.

Regarding respiratory diseases, chronic obstructive pulmonary disease (COPD) and obstructive sleep apnea (OSA) emerge as prominent predictors. A population-based study utilizing the Rotterdam Study database (approximately 15,000 participants aged ≥45 years with up to 24 years of follow-up) found that COPD was associated with an increased risk of SCD (HR 1.3, 95% CI 1.1–1.7). Notably, the risk escalated to 3.6-fold (HR 3.6, 95% CI 2.4–5.4) among COPD patients with frequent exacerbations (defined as ≥2 hospitalizations for moderate or severe exacerbations per year) (36). These findings were confirmed by a community-based case-control study in Oregon, USA [the Oregon Sudden Unexpected Death Study (ORSUDS) with catchment population approximately 1 million], which demonstrated a significant association between COPD and SCD (OR 2.2, 95% CI 1.4–3.5) after multivariable adjustment (37). In patients with COPD, chronic hypoxemia depletes cellular energy reserves, predisposing to non-shockable rhythms where cardiomyocytes cannot sustain adequate function, resulting in electromechanical dissociation or pump failure (38). Type 3 pulmonary hypertension drives right ventricular remodeling with increased arrhythmic susceptibility (39), while persistent systemic inflammation—mediated by pulmonary-derived cytokines—directly triggers ventricular dysrhythmias (40).

On the other hand, OSA presents a distinct pathophysiology characterized by repetitive nocturnal hypoxia-arousal cycles and intermittent autonomic activation. In a Minnesota cohort study by Gami et al. involving 10,701 adults who underwent polysomnography with follow-up from 1987 to 2003, OSA severity markers were significant predictors of SCD. Specifically, an apnea-hypopnea index (AHI) >20 conferred an HR of 1.6, while mean nocturnal O₂ saturation <93% was associated with an HR of 2.9 (41). This pattern generates oxidative stress, ventricular remodeling from supply-demand mismatch, and ectopic activity culminating in potentially fatal arrhythmias (42, 43). Additionally, a post hoc analysis of 112 SCD cases from the study above revealed an important finding: patients with severe OSA (AHI ≥40) were significantly more likely to experience cardiac arrest during sleep between midnight and 6 a.m. compared to those without OSA [Relative Risk (RR) 2.6, 95% CI 1.3–5.4]. This nocturnal timing pattern reduces the likelihood of witnessed events and access to bystander resuscitation, thereby further increasing SCD risk (44).

Hepatic pathology constitutes the primary gastrointestinal link to SCD. A recent large-scale study of approximately 5.4 million adults aged 20–39 years from the Korean National Health Insurance Service database (2009–2012, followed through 2020) investigated metabolic dysfunction-associated steatotic liver disease (MASLD) as a predictor of sudden cardiac arrest (SCA). MASLD was determined by the fatty liver index (FLI), and participants with high FLI (≥60) demonstrated a 1.6-fold increased risk of SCA (HR 1.6, 95% CI 1.4–1.7) (45). While the exact mechanisms remain poorly understood, SLD potentially facilitates SCD through metabolic syndrome-induced non-ischemic cardiomyopathy and chronic low-grade inflammation promoting arrhythmogenic ventricular fibrosis (46). Supporting this hypothesis, previous studies have documented significant ECG changes in liver disease patients. One study comparing 94 cirrhosis patients with 37 controls with mild chronic active hepatitis found prolonged QTc intervals (440.3 ± 3.2 ms vs. 393.6 ± 3.7 ms) (47), while another demonstrated QTc prolongation in 69 patients with alcoholic liver disease compared to age- and sex-matched healthy non-drinking controls (450 ms vs. 439 ms) (48). Furthermore, autopsy analysis from an ongoing community-based study in Southern California (2015–2023) [PREdiction of Sudden Death in Multi-ethnic Communities (PRESTO)] of 162 SCD cases revealed that SCD cases with SLD exhibited higher body mass index, greater hepatic mass, and increased alcohol consumption rates, yet paradoxically demonstrated lower coronary artery disease prevalence (49% vs. 71%) compared to those without SLD (49). These findings underscore the importance of non-cardiovascular mechanisms in SCD pathogenesis.

Renal dysfunction progressively amplifies SCD risk from moderate impairment through end-stage kidney disease (ESKD). A study of 15,792 middle-aged individuals (aged 45–64 years at baseline) from the community-based Atherosclerosis Risk in Communities (ARIC) cohort demonstrated that advanced chronic kidney disease (CKD stage 3b or worse) was independently associated with SCD risk (HR 3.7, 95% CI 1.7–7.9) compared to individuals with eGFR ≥90 ml/min per 1.73 m² (50). This association was further elaborated in a recent case-control study utilizing the ORSUDS cohort for discovery and the PRESTO study for validation. In ORSUDS, which included 2,068 SCA cases and 852 controls without history of ventricular arrhythmia or SCA, moderate CKD (stages 3a and 3b) emerged as an independent risk factor for SCA (OR 1.3, 95% CI 1.0–1.7). Notably, SCA risk increased by 24% for every 10 ml/min per 1.73 m² decrease in eGFR below 90 ml/min per 1.73 m² (51). These findings were validated in the PRESTO study including 817 SCA cases and 3,249 controls, which confirmed that moderate CKD was independently associated with SCA (OR 1.5, 95% CI 1.2–2.0), with a 41% increased risk for every 10 ml/min per 1.73 m² decrease in eGFR below 90 ml/min per 1.73 m². CKD might induce SCD through neurohormonal activation (sympathetic nervous system and renin–angiotensin–aldosterone system hyperactivity), systemic inflammation, volume overload, and electrolyte disorders (52, 53). These pathophysiological disturbances trigger myocardial hypertrophy, fibrosis, and adverse ventricular remodeling while simultaneously inducing both macrovascular atherosclerosis and microvascular arteriolosclerosis, impairing coronary perfusion. The resultant ischemic injury not only precipitates heart failure but also creates an arrhythmogenic substrate characterized by heterogeneous scar tissue that facilitates reentrant circuits and conduction system abnormalities, heightening susceptibility to fatal ventricular arrhythmias (52, 54). Particularly, in patients with ESKD undergoing dialysis, SCD contributes to a large portion of cause of death ranging from 14% to 45% with an overall incidence rate from 4.5 to 7 per 100,000 hemodialysis sessions (55–58). The hemodialysis procedure itself may represent an independent risk factor for SCD. A study by Bleyer et al. analyzing data from five dialysis centers in the southeastern U.S. (1995–2003) demonstrated a 1.7-fold increased risk of SCD during the 12-hour period following dialysis initiation compared to expected mortality rates (59). This temporal relationship was further investigated by a study of 32,065 participants in the End-Stage Renal Disease Clinical Performance Measures Project, a nationally representative sample of U.S. hemodialysis patients. Over 2.2 years of follow-up, the SCD rate following a long interdialytic interval (Mondays and Tuesdays, after the weekend gap) was significantly higher than on other days of the week (60). Similar findings were confirmed in a recent ORSUDS analysis, which reported a significantly higher proportion of SCA cases on dialysis days (nearly 3-fold higher than expected by random distribution), particularly following long interdialytic periods (Mondays and Tuesdays) (61). Although the complete mechanistic pathway from dialysis to SCD requires further elucidation, current literature has identified several arrhythmogenic contributors. Significant variation of intravascular volume, uremic toxic accumulation, coupled with derangements in serum electrolyte concentrations and their subsequent rapid normalization (62, 63).

Anemia independently predicts SCA in the general population, as demonstrated by a Korean National Health Insurance Database Cohort study of approximately 500,000 individuals with a mean follow-up of 5.4 years. This study reported that SCA risk increased progressively across anemia severity categories, with HRs ranging from 1.5 to 6.2 in men and 1.9 to 8.8 in women from mild to severe anemia, respectively (64). Furthermore, SCD risk increased by 21%–24% per 1 g/dl decrease in hemoglobin, with inverse correlations observed between hemoglobin levels and ECG abnormalities such as QTc prolongation. A subsequent study utilizing the Mayo Clinic Aortic Stenosis database (8,357 adults with severe aortic stenosis) with SCD ascertainment through the National Death Index and medical records confirmed that anemia history was associated with significantly increased SCD risk (HR 1.4, 95% CI 1.1–1.9) (65). The main mechanistic consequence of anemia is reduced of oxygen content in blood which subsequently induces or worsens the myocardial ischemia causing increasing compensatory sympathetic signal such as tachycardiac and cardiac contraction. This phenomenon can cause cardiac remodeling with increasing of ventricular fibrosis, hypertrophy, and also electrophysiological alteration (66, 67).

Diabetes consistently appears as an independent risk factor for SCD across multiple studies. A systematic review and meta-analysis of 19 population-based and 10 patient-based prospective studies encompassing over 6,000 SCD cases from more than 300,000 participants demonstrated a summary RR of 2.0 (95% CI 1.8–2.3) for SCD in individuals with diabetes compared to those without diabetes (68). This association was confirmed in two large Danish nationwide studies using death certificate-defined SCD from 2000 to 2010. The first study focused on younger individuals (aged 1–49 years) and reported incidence ratios of 8.6 (95% CI 5.8–12.8) in the 1–35 age group and 6.1 (95% CI 4.7–7.8) in the 36–49 age group for SCD in those with diabetes compared to non-diabetic persons (69). The second, more recent study investigated the entire Danish population in 2010 (approximately 5.5 million individuals) and revealed differential effects by diabetes type: type 1 diabetes conferred an HR of 2.4 (95% CI 2.0–3.0) for SCD, while type 2 diabetes showed an HR of 1.7 (95% CI 1.6–1.9) (70). This differential risk between diabetes types was previously reported by a community-based case-control study using the ORSUDS database, which included 2,771 SCA cases and 8,313 controls matched by sex, age, race/ethnicity, and geography. This analysis showed that the odds of SCA were 2.4-fold higher in type 1 diabetes compared to type 2 diabetes (71). Evidence is also accumulating regarding prediabetes and SCD risk. The aforementioned meta-analysis reported an RR of 1.2 (95% CI 1.1–1.4) for SCD in prediabetic individuals (68). Additionally, a population-based study in Eastern Finland of 2,641 men aged 42–60 years with 19 years of follow-up demonstrated that nondiabetic men with impaired fasting plasma glucose had a 1.5-fold increased risk of SCD (RR 1.5, 95% CI 1.1–2.1) compared to those with normal fasting glucose levels (72). Diabetes can contribute to SCD both by causing coronary artery disease, and by acting as a risk factor through various mechanisms. These include abnormal myocardial metabolism due to impaired insulin signaling, autonomic nervous system dysfunction as a diabetic complication, increased ventricular fibrosis leading to structural remodeling, and the risk of hypoglycemia (73).

An analysis from the Rotterdam Study, a prospective population-based cohort, including 10,318 participants with a median follow-up of 9.2 years, revealed that elevated free T4 levels were associated with increased SCD risk (HR 2.3 per 1 ng/dl increase, 95% CI 1.3–4.0) (74). Additionally, a similar population-based cohort study in Finland (5,211 participants aged ≥30 years, enrolled 2000–2001, median follow-up 13.2 years) demonstrated that elevated baseline TSH levels were also associated with higher SCD risk (HR 2.3, 95% CI 1.1–4.6) compared to TSH within the normal range (75). Thyroid hormones are known to increase heart rate, conduction velocity, and myocardial contractility by increasing the number of beta-adrenergic receptors, thereby heightening susceptibility to arrhythmias (76, 77).

While immunologic mechanisms are well-established contributors to SCD risk through direct cardiovascular involvement in systemic conditions—particularly myocarditis, cardiac sarcoidosis, cardiac amyloidosis, systemic lupus erythematosus, and systemic sclerosis—this review focuses on non-cardiovascular immune-mediated conditions such as rheumatoid arthritis (RA), for which evidence remains limited. The association between RA and SCD was demonstrated in a population-based cohort study from Rochester, Minnesota, which included 603 residents diagnosed with RA using the American College of Rheumatology 1987 criteria and 603 age- and sex-matched non-RA controls. Beginning 2 years after RA diagnosis, patients with RA had a 1.9-fold increased risk of SCD (95% CI 1.1–3.6) compared to those without RA (78). The pathogenic mechanisms likely involve persistent high-grade systemic inflammation with elevated cytokine levels, which may induce cardiac fibrosis and autonomic cardiac dysfunction. Supporting this hypothesis, 76% of RA patients demonstrated prolonged corrected QT interval at baseline, which rapidly normalized following anticytokine therapy with tocilizumab in a case series of 17 RA patients receiving long-term treatment (79).

Predicting SCD risk represents only the initial step in prevention efforts. Equally critical is identifying which patients will present with shockable vs. non-shockable rhythms, as our primary preventive device – the ICD – provides benefit exclusively for shockable presentations. However, the current indicator for ICD implantation, left ventricular ejection fraction (LVEF), demonstrates poor performance in predicting SCD, particularly among patients with prior myocardial infarction (MI) with c-statistic ranging from 0.50 to 0.56, as reported by a pooled analysis of 20 datasets encompassing more than 140,000 post-MI patients (80). Non-cardiovascular comorbidities may substantially improve our ability to predict initial cardiac arrest rhythm and identify appropriate ICD candidates.

Emerging evidence from community-based research reveals those non-cardiovascular comorbidities significantly drive arrest patterns toward non-shockable rhythms (asystole or pulseless electrical activity – PEA) (81, 82). Specifically, Neurological disorders confer 1.6-fold elevated risk for non-shockable presentations (95% CI 1.4–1.9) (82). Epilepsy patients demonstrate particularly striking differences, with 74% presenting in non-shockable rhythms compared to 56% of controls (p < 0.05) (26). Depression shows similarly pronounced effects: 73.2% vs. 51.2% non-shockable presentations (p < 0.05), translating to an odds ratio of 2.7 (95% CI 2.5–3.0) (83–85). Psychotropic medications—including antidepressants, antipsychotics, and anxiolytics—further amplify this vulnerability (82). COPD severity correlates directly with non-shockable risk, escalating from mild disease (OR 1.5, 95% CI 1.3–1.7) to severe obstruction (OR 2.0, 95% CI 1.8–2.2) (82, 86). Hepatic pathology demonstrates comparable patterns: cirrhotic patients present with non-shockable rhythms in 75% of cases vs. 42% without cirrhosis (p < 0.001) (87, 88). CKD associates with predominantly non-shockable presentations across multiple community studies (51, 89). Particularly, Implantable loop recorder data from hemodialysis patients reveal bradycardia and asystole—rather than ventricular arrhythmias—as the dominant terminal rhythms (90). Machine learning analyses additionally identify anemia as a predictor of pulseless electrical activity (89). Finally, Diabetes increases non-shockable risk (OR 1.3, 95% CI 1.2–1.4), with type 1 diabetes potentially conferring greater vulnerability than type 2 (71, 82). While diverse pathways may explain these associations, a shared mechanism centers on cardiomyocyte energy depletion leading to electromechanical dissociation. Hypoxemia (from anemia, COPD, seizure-related respiratory distress), uremic toxicity (CKD), and impaired glucose utilization (diabetes) converge on cellular energy failure—a state where electrical activity persists despite insufficient contractile function to generate effective cardiac output.

Among non-cardiac conditions, only diabetes, CKD (or eGFR-estimated renal function), COPD, and seizure disorders appear in existing algorithms. Diabetes emerges as the most frequently included, featuring in 7 of 9 models—highlighting its critical pathophysiologic role in SCD. The VFRisk score, developed by Chugh et al. from the ORSUDS and PRESTO (subsequently validated by Truyen et al. in the Framingham Heart Study cohort), stands alone as the only algorithm incorporating three non-cardiac conditions (diabetes, COPD, seizure disorder) (98, 99). Notably, VFRisk was also uniquely designed to differentiate shockable from non-shockable initial rhythms. Recent machine learning approaches have further advanced rhythm prediction, distinguishing PEA from ventricular fibrillation using novel predictors including anemia, CKD, OSA, cancer, and mood disorders (89). Despite abundant evidence linking non-cardiovascular conditions to SCD, their limited incorporation into clinical prediction tools remains concerning. Several factors explain this disconnect:

The cardiac-centric approach reflects long-established associations between cardiovascular conditions and SCD, whereas non-cardiac relationships emerged more recently with incompletely understood mechanisms. Many foundational studies excluded these variables from initial design, making them unavailable for multivariable analysis—particularly in post-hoc analysis. Cardiologists' relative unfamiliarity with non-cardiac conditions may also contribute to their omission from study protocols.

SCD low incidence creates significant statistical challenges. Traditional multivariable logistic regression requires 10–20 events per variable, meaning 15 covariates demand 150–300 SCD cases—beyond most institutions' capacity. Additionally, clinical utility demands simplicity; overly complex algorithms fail adoption in daily practice. These pragmatic constraints prioritize established cardiac factors over emerging non-traditional predictors.

The complicated relationships between non-cardiac and cardiovascular conditions remain poorly characterized, with overlapping effects potentially diluting individual contributions after multivariable adjustment. For instance, CKD achieved statistical significance in VFRisk univariable analysis but lost predictive value in the multivariable model (98). The combinatorial complexity of comprehensively modeling all potentially relevant cardiac and non-cardiac conditions creates prohibitive statistical burdens unfeasible for most investigations.

Incorporating numerous covariates substantially increases overfitting risk, compromising model reproducibility across populations and limiting external validation success. This technical limitation creates systematic bias favoring traditional, well-validated cardiac factors over newer non-cardiac predictors that require larger datasets to demonstrate stable, generalizable effects.

Cross-specialty partnerships in study design and analysis can address inherent limitations of cardiac-centric approaches. Engaging neurologists, nephrologists, pulmonologists, psychiatrists, and endocrinologists provides essential expertise for accurately measuring systemic diseases and interpreting their complex interactions with cardiovascular pathophysiology. Such collaboration reduces specialty bias while enhancing mechanistic understanding of how non-cardiovascular conditions contribute to SCD risk.

Large-scale collaborative networks substantially increase SCD event number, overcoming sample size constraints of single-center investigations. Pooling data across institutions can achieve thousands of SCD cases, enabling robust multivariable models that accommodate numerous predictors. This expanded statistical capacity creates room for both traditional cardiac factors and emerging non-cardiac variables without compromising model stability.

Supervised ML using tree-based algorithms (random forests, gradient boosting) automatically identify non-linear relationships and complex interactions without requiring prior specification. Feature importance metrics systematically rank predictors, enabling development of parsimonious models that retain crucial non-cardiac variables while maintaining clinical practicality. This approach balances comprehensiveness with the simplicity demanded for bedside implementation, potentially enhancing discrimination in large datasets with intricate patterns. Moreover, clustering techniques—including K-means, hierarchical clustering, and model-based methods—uncover hidden patient patterns and novel phenotypes. These approaches can reveal how specific combinations of cardiac and systemic conditions create distinct risk profiles, generating hypotheses about disease synergy that traditional regression frameworks cannot detect. Such insights deepen mechanistic understanding beyond additive risk factor models. Advanced regularization techniques (LASSO, elastic net) and dimensionality reduction methods (principal component analysis) might address collinearity among clustered conditions—hypertension-diabetes-kidney disease or depression-anxiety-cardiac arrest—that confound conventional regression. These methods extract meaningful risk signals from correlated predictor sets, resolving a fundamental challenge in multimorbidity modeling. ML capacity to handle high-dimensional data, capture non-linear relationships, and identify complex interaction patterns positions it as a transformative tool for SCD risk stratification. However, realizing this potential requires rigorous external validation, attention to overfitting prevention, and preservation of clinical interpretability. Success ultimately depends not only on statistical sophistication but also on meaningful collaboration between data scientists, clinicians, and implementation specialists to bridge the gap between algorithmic performance and bedside utility.