BACKGROUND:Patients harboring pathogenic/likely pathogenic (P/LP) variants in the desmoplakin (DSP) gene are at risk of ventricular arrhythmias (VAs). In this population, a risk prediction model estimating the 5-year risk of VAs has been recently developed. OBJECTIVE:This study aimed to provide external validation of this prediction model in a new large, international, multicenter cohort and to test its reliability in patients with and without a history of myocarditis-like episodes. METHODS:All patients with a P/LP pathogenic DSP variant enrolled in the Desmoplakin Specific Effort for a Rare Disease Outcome Study Network with no sustained VA before or at first assessment and who were not used for the development of the DSP-risk score (www.DSP-risk.com) were used to test its performance. Model performance was assessed using the c-statistic in both the overall cohort and stratifying by history of myocarditis-like episodes. RESULTS:450 DSP patients from 30 centers were enrolled (mean age 42.2 ± 17.6; 40.4% female; 18.4% with previous myocarditis-like episode). Over a median of 4.3 years (1.6-10.0), 60 sustained VAs were observed. The DSP-risk score yielded good discrimination both overall (c-statistic, 0.719; 95% confidence interval [CI], 0.706-0.733) and for patients with (c-statistic, 0.719; 95% CI, 0.702-0.737) and without previous myocarditis-like episodes (c-statistic, 0.749; 95% CI, 0.740-0.759). CONCLUSION:In a large independent cohort of DSP patients, this study showed external validity of the DSP-risk score. These findings support the use of the DSP-risk score to facilitate shared decision making regarding implantable cardioverter-defibrillator implantation in the primary prevention of VAs in patients harboring DSP P/LP variants.

Background PRKAG2 cardiomyopathy is a genetic cardiomyopathy that phenotypically resembles hypertrophic cardiomyopathy (HCM) and results from myocardial glycogen accumulation. In addition to left ventricular hypertrophy (LVH), PRKAG2 cardiomyopathy is associated with ventricular pre-excitation, cardiac conduction disease, and ventricular tachyarrhythmias. Case Summary A 34-year-old woman with a family history of HCM and Wolff-Parkinson-White syndrome presented with exertional dyspnea, chest discomfort, and near-syncope. Electrocardiogram revealed LVH, T-wave inversions, and pre-excitation. Echocardiography demonstrated severe LVH with outflow tract obstruction. Based on family history, LVH, and pre-excitation, PRKAG2 syndrome was suspected and later confirmed by genetic testing. Discussion PRKAG2 cardiomyopathy results from glycogen accumulation rather than sarcomeric hypertrophy. This case highlights the diagnostic importance of electrocardiography-detected pre-excitation in a patient with severe LVH, raising suspicion for nonsarcomeric HCM. Take-Home Messages Consider PRKAG2 mutations in HCM patients with pre-excitation or conduction disease. Early genetic diagnosis guides management and informs family screening.

BACKGROUND:Current guidelines recommend regular screening for first-degree relatives of gene-elusive arrhythmogenic right ventricular cardiomyopathy (ARVC) patients using a similar regimen as for genotype-positive/phenotype-negative relatives. However, the multifactorial nature of gene-elusive ARVC may necessitate a different approach. This study aimed to determine the yield of cardiac screening in first-degree relatives of ARVC probands without a validated genetic cause. METHODS:We included all first-degree relatives of probands who (1) met the 2010 Task Force Criteria, (2) underwent next-generation sequencing that included all genes with at least moderate evidence for ARVC causation per Clinical Genome Resource appraisal (validated ARVC genes), and (3) had no pathogenic/likely pathogenic (P/LP) variants identified in these genes. The primary and secondary end points were definite ARVC by the 2010 Task Force Criteria and ventricular arrhythmia, respectively. RESULTS:We included 44 relatives (39.0 [22.3-45.8] years; 36% male) from 24 families. In 4 (17%) families, a P/LP variant was identified in a different cardiomyopathy/arrhythmia gene (SCN5A, LMNA, CDH2, FLNC). Overall, 10 (23%) relatives had definite ARVC at baseline evaluation. Of the 20 relatives without definite ARVC who had follow-up available, 8/20 (40%) relatives progressed to definite ARVC during 9.0 (5.8-14.4) years of follow-up. No statistical difference in the yield of baseline screening or serial evaluation between relatives from families with a P/LP variant and relatives from families without a P/LP variant was observed. Of the 27 relatives who had follow-up available, ventricular arrhythmia was observed in 2/27 (7%) relatives and occurred 6.3 and 13.8 years after definite ARVC diagnosis. Both of those relatives were from families without a P/LP variant. CONCLUSIONS:These findings highlight the importance of managing first-degree relatives of ARVC probands without a validated genetic cause similarly to genotype-positive ARVC relatives. Furthermore, using a broad cardiomyopathy and arrhythmia gene panel in ARVC probands, rather than limiting testing to validated ARVC genes alone, is warranted.

BACKGROUND:Patients with obstructive hypertrophic cardiomyopathy (oHCM) are at an increased risk of the development of atrial fibrillation (AF) and subsequent AF-related morbidity. The impact of treatment with the novel cardiac myosin inhibitor mavacamten on AF is unknown. OBJECTIVE:This study aimed to evaluate incident and recurrent AF rates in patients with oHCM receiving mavacamten. METHODS:This was a retrospective observational study of patients with oHCM undergoing mavacamten treatment at the Johns Hopkins Hypertrophic Cardiomyopathy (HCM) Center. Predicted AF risk was calculated using the HCM-AF score at the time of mavacamten initiation. Patients were followed longitudinally for the development of new or recurrent clinically significant AF. RESULTS:160 patients with oHCM (mean age 62 years; 58% female) were enrolled, including 30 (18.8%) with a diagnosis of AF before mavacamten treatment. Median HCM-AF score at the time of mavacamten initiation was 23. During a median of 1.4 years of mavacamten treatment (interquartile range 0.6-2.3), new AF occurred at a rate of 6.4%/y in those without a previous AF diagnosis. Although the HCM-AF score had reliable discrimination in this population (concordance statistic 0.78), the observed 2-year risk of symptomatic AF modestly exceeded HCM-AF risk predictions (8.2% vs 6.9%). On time-dependent multivariable Cox regression, mavacamten was associated with an increased risk of the development of new, clinically significant AF (hazard ratio 3.0; P = .01). In patients with a previous AF diagnosis, AF recurred during mavacamten treatment at a rate of 19.3%/y. CONCLUSION:Although patients with oHCM had a high baseline risk of AF, mavacamten treatment was associated with an increased risk of AF development beyond that predicted by the HCM-AF score.

Importance:Mavacamten is highly effective in treating symptomatic obstructive hypertrophic cardiomyopathy (oHCM) and was approved for commercial use with a risk mitigation program (REMS) to monitor the impact on left ventricular systolic function (LVSD). The impact of mavacamten on atrial fibrillation (AF) occurrence is not well characterized. Objective:Determine the real-world incidence of new-onset and recurrent AF among patients with HCM treated with commercial mavacamten. Secondary objectives included assessing the incidences of LVSD, heart failure (HF), and cardiogenic shock. Design:A multicenter cohort study. Center-level data were aggregated for patients who received mavacamten from May 2022 through December 2024. Setting:Twenty-one outpatient HCM centers in the United States. Participants:Consecutive patients ≥18 years of age with oHCM were included; those with permanent AF were excluded. Exposures:At least one dose of commercial mavacamten. Main Outcomes and Measures:Incidence of new-onset and recurrent AF, LVSD, HF, and cardiogenic shock. Results:Among 1,538 patients (median age, 66 years [95% CI, 65.9-66.1]; 57% female [95% CI, 54-59%]), 25% [95% CI, 22-27%] had prior AF. Median mavacamten exposure was 13.4 months [95% CI, 11.8-15.0]. Overall AF incidence was 13% [95% CI, 10-17%], including 5% [95% CI, 4-7%] new-onset and 39% [95% CI, 27-51%] recurrent AF in those with history of AF prior to mavacamten initiation. LVEF <50% occurred in 8% [95% CI, 6-9%], of whom 70% had AF. Symptomatic HF occurred in 1.5% [95% CI, 0.8-2.2%], cardiogenic shock in 0.6% [95% CI, 0.1-1.0%], and an overall permanent discontinuation of mavacamten in 7% [95% CI, 4-10%]. Conclusions and Relevance:In this multicenter cohort receiving commercial mavacamten, we identified an annual incidence of 5% new-onset AF and 39% recurrent AF, while 70% of patients who developed LVEF<50% had concurrent AF. Given this association and the morbidity that can be associated with AF, protocols for LVEF assessment and aggressive rhythm management following AF detection may be warranted to improve patient care. Further studies are needed to improve understanding of the impact of mavacamten on AF and patient outcomes.

BACKGROUND:Apical aneurysms in hypertrophic cardiomyopathy (HCM) have been linked to sudden cardiac death (SCD) and a nidus for thromboembolism. Uncertainty remains regarding the level of risk and significance of aneurysm size. OBJECTIVES:The objective of the study was to determine the rate of SCD events and prevalence of apical thrombus or embolic events by size (maximum transverse dimension). METHODS:Apical aneurysms were identified in 510 patients from 10 centers, followed a median of 4.1 years for SCD events (SCD, appropriate implantable cardioverter defibrillator therapy, and resuscitated SCD) or development of apical thrombus/thromboembolism. Relationship between size and SCD events was analyzed using multivariable Cox proportional hazard models. RESULTS:In 510 HCM patients: 19% had small aneurysms (<10 mm), 39% medium (10-19 mm), 39% large (20-39 mm), and 3% very large (≥40 mm). SCD event rate was 2.1%/year, with risk increasing with increasing aneurysm size: 0.2%/year in small, 2.3%/year in medium, 3.0%/year in large and 8.2%/year in very large (P < 0.001). On multivariable analysis, greater size was associated with SCD events, independent of other risk markers or European Society of Cardiology-SCD score. Either an embolic event (3.6% of patients) or apical thrombus (10% of patients) occurred in 13% of patients and was independently associated with greater aneurysm size, 3% in small to 25% in very large aneurysms (P < 0.001). CONCLUSIONS:In a large cohort of HCM patients with apical aneurysms, rates of SCD events were high, with continuous relationship between size and risk. Small aneurysms (<10 mm) were associated with low risk for SCD events (0.2%/year), whereas aneurysms ≥10 mm with high risk (>2%/year). Although embolic events were uncommon, increasing aneurysm size was associated with the prevalence of apical thrombi.

Arrhythmogenic Right Ventricular Cardiomyopathy (ARVC) is a leading contributor to sudden cardiac death worldwide in young adults, yet its diagnosis remains complex, expensive and time-consuming. Machine-learning (ML) classifiers offer a practical solution by delivering rapid, scalable predictions that can lessen dependence on expert interpretation and speed clinical decision-making. Here, we benchmarked six ML algorithms for ARVC detection using area-under-the-curve (AUC) and accuracy as primary metrics. Gradient Boosted Trees outperformed all other models, achieving a c-statistic of 94.34% after rigorous cross-validation. These results underscore the promise of Gradient Boosted Trees classifier as an effective decision-support tool within the ARVC diagnostic workflow, with potential to streamline evaluation and improve patient outcomes.

BACKGROUND:Cardiac amyloidosis (CA) and hypertrophic cardiomyopathy often manifest with overlapping clinical features, making diagnosis and management challenging. CASE SUMMARY:We describe a 79-year-old man presenting with exertional dyspnea and presyncope who was diagnosed with transthyretin CA with left ventricular outflow tract (LVOT) obstruction and was treated with concurrent mavacamten and tafamidis. DISCUSSION:There is substantial overlap in the clinical phenotypes of hypertrophic cardiomyopathy and CA, and the presence of hemodynamically significant LVOT does not exclude the possibility of CA as an underlying diagnosis. Endomyocardial biopsy may be required when noninvasive testing is ambiguous or in the presence of overlapping disease features. TAKE-HOME MESSAGES:Hemodynamically significant LVOT obstruction is a rare but clinically relevant manifestation of CA that may lead to diagnostic ambiguity and difficult management decisions. Tafamidis and mavacamten can be used safely in combination to provide symptom relief and slow disease progression in obstructive transthyretin CA.

BACKGROUND:Atrial fibrillation (AF) is the most common sustained arrhythmia in patients with hypertrophic cardiomyopathy (HCM). The 2024 American Heart Association/American College of Cardiology guidelines recommend validated clinical tools such as the HCM-AF score for individualized assessment of AF risk. To date, these tools have been validated only in predominantly white HCM patient populations. OBJECTIVE:This study aimed to compare the performances of published AF risk prediction tools in a racially diverse cohort of patients with HCM. METHODS:This was a retrospective study of patients with HCM without previous AF evaluated at the Johns Hopkins HCM Center. Assessments of AF risk were generated using the HCM-AF score and other non-HCM-specific risk scores (C2HEST, HARMS2-AF, CHA2DS2-VASc, and CHARGE-AF). Patients were followed longitudinally for the development of new-onset, clinically significant AF. Discrimination was assessed using concordance-based c-statistics. RESULTS:A total of 631 patients with HCM were included, with a mean age of 55.9 ± 15.3 years; 49.7% were women, 64.7% were white, 24.1% were black, and 11.2% identified with other nonwhite race. During a median follow-up of 3.1 years, new AF was diagnosed in 18.9% of patients. The HCM-AF score demonstrated better risk discrimination (c-statistic 0.72) than other non-HCM-specific risk scores (c-statistics 0.56-0.67) and effectively stratified patients into low-risk (0.9% AF/year), medium-risk (3.4% AF/year), and high-risk groups (7.4% AF/year). Discrimination of AF risk by the HCM-AF score was similar for white (c-statistic 0.71) and nonwhite patients (c-statistic 0.74). CONCLUSION:The HCM-AF score demonstrated good AF risk discrimination in a diverse cohort of patients with HCM, outperforming alternative non-HCM-specific AF risk scores and validating its use in nonwhite HCM populations.

Background:Pathogenic variants in the desmin gene (DES) are implicated in diverse cardiomyopathy and skeletal myopathy phentoypes with high rates of major adverse cardiac events (MACE). Although pediatric cases have been reported, the genetic profile, clinical features and outcomes in affected children remain poorly defined. This study aimed to comprehensively characterize these aspects in the pediatric population. Methods:We conducted a systematic review and individual patient data meta-analysis of Medline (PubMed) and Embase, including patients with pathogenic or likely pathogenic DES variant(s) diagnosed before age 21. MACE were defined as cardiac conduction disease (CCD) requiring device implantation, sustained ventricular arrhythmias (VA), and heart failure (HF) events. Results:Fifty-one pediatric patients were included (58.8% male; 64.7% probands; median age at first evaluation: 14.0 years [IQR 11.0-17.0], median follow-up: 5.0 years [1.0-11.5]). Among probands, 60.6% (20/33) had a heterozygous variant (20 non-frameshift/non-truncating and 2 frameshift/truncating), while the remaining 11 (33.3%) harbored biallelic genotypes (homozygous (n=8), compound heterozygous (n=3)). At presentation, 54.9% (28/51) patients had cardiomyopathy and 54.9% had skeletal myopathy. CCD occurred in 39.2% (median age: 19.0 years), sustained VA in 21.6% (median age: 18.0 years), and HF events in 37.3% (median age: 19.0 years). In total, 64.7% experienced MACE (median age: 17.0 years [13.0-19.0]), and 19.6% died during follow-up (median age: 23.5 years [14.8-28.0]). Rates of cardiomyopathy diagnosis, CCD, VA and HF events as well as composite MACE were similar between those with a heterozygous and biallelic variants (p>0.05 for all), but the latter group had nearly twice higher rate of skeletal myopathy (43.6% vs 83.3%, p=0.016). Conclusion:Pediatric DES-associated disease is characterized by heterogeneous cardiomyopathy phenotypes and substantially high MACE burden. Probands were more likely than genotype-positive relatives to develop CCD and HF events, while sustained VA rates were similar and substantial in both groups. Cardiomyopathy rates and outcomes were comparable between patients with heterozygous and biallelic variants.

BACKGROUND:Penetrance and risk of ventricular arrhythmias (VAs) in arrhythmogenic right ventricular cardiomyopathy (ARVC) are increasingly recognized as being genotype specific. Therefore, genotype-informed family screening protocols may lead to safer and more personalized recommendations than the current one-size-fits-all screening recommendations. We aimed to develop a safe, evidence-based plakophilin-2 (PKP2)-specific longitudinal screening algorithm. METHODS:We included 295 relatives (41% male; age 30.9 years [18.0-47.7 years]) with a pathogenic or likely pathogenic PKP2 variant from 145 families. Phenotype was ascertained with ECG, Holter monitoring, and cardiac imaging and classified by the 2010 Task Force Criteria. VA was defined as a composite of sudden cardiac arrest or death, spontaneous sustained ventricular tachycardia, ventricular fibrillation, or appropriate implantable cardioverter defibrillator intervention. We performed Cox regression to determine predictors of ARVC development and multistate modeling to assess the probability of ARVC development and occurrence of VA. RESULTS:At baseline, 110 relatives (37%) had definite ARVC. During 8.5 years (4.2-12.9 years) of follow-up, 62 of 185 relatives (34%) without definite ARVC at baseline progressed to definite ARVC diagnosis, and 35 of 295 of all relatives (12%) had VA. VAs occurred only in relatives who previously fulfilled definite ARVC diagnosis. Relatives with borderline ARVC (fulfillment of one minor criterion plus the major family history criterion) progressed 5 times faster in the multistate model to definite ARVC diagnosis and compared with genotype-positive/phenotype-negative (G+/P-) relatives (ie, major family history criterion alone). Relatives 20 to 40 years of age had increased risk for developing definite ARVC (hazard ratio, 2.23; P=0.012) compared with those ≥40 years of age. New Task Force Criteria fulfillment most commonly occurred first on ECGs, followed by Holter monitoring and cardiac imaging. Consequently, 3 risk profiles were identified, and appropriate screening protocols were derived: relatives with borderline ARVC (annual ECG and Holter monitoring; complete evaluation [ie, ECGs, Holter monitoring, and imaging] every 2 years), younger (<40 years of age) or symptomatic G+/P- relatives (every 2 years an ECG and Holter monitoring; complete evaluation every 4 years), and older (≥40 years of age) and asymptomatic G+/P- relatives (complete evaluation every 5 years). CONCLUSIONS:An evidence-based longitudinal screening algorithm that integrates age, symptoms, and baseline clinical phenotype may improve patient care and improve efficiency of clinical resource allocation.

Introduction Desmoplakin cardiomyopathy (DSP-CMP) is a genetic cardiomyopathy characterized by high risk of ventricular arrhythmias (VAs), heart failure (HF), and recurrent myocarditis. Data regarding the use and impact of immunosuppression on those myocarditis episodes is scarce. Hypothesis Immunosuppression treatment in DSP-CMP associated myocarditis may improve clinical outcomes. Methods Patients (pts) with DSP-CMP and 1+ episode of myocarditis in the worldwide DSP-ERADOS Network (26 institutions across 9 countries in 3 continents) were enrolled. For each treated episode, immunosuppressive medication regimen and duration (acute vs chronic immunosuppression) were collected. The primary combined outcome was sustained VA and/or HF episodes during follow up. Outcomes of DSP-CMP patients receiving immunosuppression at the time of first myocarditis episode were compared with DSP-CMP patients with untreated myocarditis using a log-rank test. Results Of 815 patients in the DSP registry, 153 patients (age at first presentation 31.1±8.7 yo, 72.5% female, 71.2% proband, LVEF 46.7±8.6%) experiencing a total of 260 myocarditis episodes (age at first episode 32.5±7.4 yo; n=60/153 had myocarditis as initial DSP-CMP presentation; 1.7 episodes/pt; n=54 pts with 2+ episodes,) were enrolled in the current study (Table 1). Of these episodes, 79 (30.4%) were treated with immunosuppression, with 23 treated with multiple agents. Seven (4.6%) patients received chronic immunosuppression. Over a median follow up of 4.6 [1.6-8.2] years, 96 primary outcome events (HF: n=31, VA: n=65) where observed in 73 (47.7%) patients. Patients whose first episode was treated with immunosuppression were less likely to meet the combined endpoint of HF/VA than those whose first episode was not treated (6/31 (19.4%) vs 67/122 (54.9%) p<0.001). Conclusions In a large, multicenter DSP-CMP registry, we demonstrate that approximately one-third of all myocarditis episodes were treated with immunosuppression, most commonly corticosteroid-based strategies. Patients whose first episode of myocarditis was treated with immunosuppression had a lower rate of subsequent VA and HF events. These findings provide preliminary data to support a prospective trial to confirm the role of immunosuppression in DSP-CMP presenting with myocarditis.

Background:Desmoplakin (DSP) cardiomyopathy, caused by variants in the gene DSP, is a unique subtype of cardiomyopathy distinct from typical dilated or arrhythmogenic right ventricular cardiomyopathies. Specific diagnostic and disease staging criteria have yet to be developed for DSP cardiomyopathy. Objective:Utilizing a large cohort of DSP cardiomyopathy patients and their genotype-positive family members, this study aims to develop diagnostic and disease staging criteria for DSP cardiomyopathy. Methods:Patients from the DSP-ERADOS Network with complete rhythm monitoring, electrocardiogram and cardiac magnetic resonance imaging were enrolled. Diagnostic criteria were assessed in initially-presenting patients (probands) and their genotype-positive family members. Early disease criteria (with preserved left ventricular ejection fraction, LVEF) were integrated into standard LVEF-based classifications. Diagnostic and staging criteria were assessed by time-event analyses (major ventricular arrhythmia and heart failure events). Results:A total of 605 patients with complete diagnostic testing were included (mean age 40 yr, 60% female, 40% probands). The most prevalent disease features in probands were premature ventricular contractions (PVCs) >500/24hr (66%), nonsustained ventricular tachycardia (NSVT, 29%), LV late gadolinium enhancement (LGE, 53%), and reduced LVEF (44%). The presence of any one of these features was 97% sensitive for diagnosis (along with a DSP pathogenic variant) and were therefore considered as diagnostic criteria. Using these criteria, 77% of genotype-positive family members were considered clinically affected. Isolated right ventricular (RV) involvement occurred in only 0.7%. The absence of diagnostic criteria identified a low-risk group (composite event rate 0.8%/year, p<0.001). Integration of these criteria into LVEF-based classification improved identification of composite arrhythmia/heart failure events (early: diagnostic criteria with LVEF ≥50%, HR 2.7, p=0.04; intermediate: LVEF 41-49%, HR 3.7, p=0.009; advanced: LVEF ≤40% HR 10.3, p<0.001). LGE was mostly subepicardial (87%). Circumferential (ring-like) LGE was more frequent in intermediate or advanced vs early disease (66% vs 48%, p<0.001). Conclusion:This study identifies genotype-specific diagnostic and disease staging criteria for DSP cardiomyopathy that improve identification of risk for both heart failure and sustained ventricular arrhythmias. This work highlights how gene-specific criteria may be used to refine diagnosis and staging for cardiomyopathy subtypes - a critical step as gene-targeted treatments move toward clinical trials.

BACKGROUND:Arrhythmogenic right ventricular cardiomyopathy (ARVC) is an inherited cardiomyopathy associated with life-threatening ventricular arrhythmias. Currently, ARVC diagnosis is defined by the 2010 modified task force criteria. OBJECTIVE:This study aimed to evaluate the implementation of updated cardiac magnetic resonance (CMR) imaging thresholds used to define pathologic right ventricular (RV) dilation and systolic dysfunction in the diagnosis of ARVC. METHODS:This was a retrospective, observational cohort study that enrolled patients with ARVC with available quantitative CMR evaluated at the Johns Hopkins ARVC Center. Reclassification of RV diagnostic criteria was reevaluated using updated standards for RV structure and function. RESULTS:430 patients with ARVC were enrolled. Mean age was 35.1 years, 47.7% were female, and 66.3% had pathogenic variants in ARVC-associated genes. Mean RV end-diastolic volume index was 114 mL/m2 (±39), and mean RV ejection fraction was 41% (±11%). Using updated standards, 18.3% of patients previously fulfilling major RV structure and function criteria were downgraded to minor criteria, and 10.5% were downgraded to no criteria fulfillment, whereas 35.0% patients fulfilling minor RV structure and function criteria were downgraded to no criteria fulfillment. 1 patient was downgraded to a borderline diagnosis, and 3 patients were downgraded to no ARVC diagnosis. 1 patient in whom ARVC diagnosis was reversed developed ventricular arrhythmia during follow-up, but had a clear alternative explanatory diagnosis of LMNA-related cardiomyopathy. CONCLUSION:Incorporation of modern standards for normal RV structure and function improved the specificity of the ARVC modified task force criteria without reducing sensitivity. These findings support the use of updated CMR-derived criteria for defining the presence or absence of pathologic RV phenotype during ARVC diagnostic evaluation.

Background Deep-learning models designed to assist with clinical decision making abound in cardiology. However, the "black box" nature of these models limits physicians' ability to use them to cross-check clinical gestalt when evaluating model predictions. Analytical techniques such as the popular gradient-weighted class activation mapping (Grad-CAM) may provide insight into model explainability, but the reliability and reproducibility of these techniques have not been studied. Objective To perform a rigorous assessment of the explainability offered by Grad-CAM, with comparison to alternative saliency methods provided by intrinsicly explainable deep-learning models. Methods We examined a well-phenotyped cohort of 1930 patients with hypertrophic cardiomyopathy (HCM) and available electrocardiographic waveform data. Novel deep-learning models were developed for the prediction of 2 high-risk HCM features: left ventricular (LV) apical aneurysm and massive LV hypertrophy. Saliency analysis was performed using (1) Grad-CAM and (2) latent-space variable decoding (LSVD). Results Deep-learning models amenable to Grad-CAM- and LSVD-based saliency analysis demonstrated comparable performances in the identification of LV apical aneurysm (C statistic 0.95 vs 0.93) and massive LV hypertrophy (C statistic 0.82 vs 0.83) during holdout testing. However, while Grad-CAM produced highly variable visual assessments of model attention and offered little insight into the models' underlying decision-making processes, LSVD allowed direct visualization of those electrocardiographic characteristics that differentiated patients with and without the high-risk HCM features of interest. In addition, Kolmogorov-Smirnov goodness-of-fit testing of latent-space variables offered a method for prospectively assessing the likelihood of deep-learning model overfitting. Conclusion Deep-learning models amenable to LSVD analysis offered more robust explainability than did models amenable to the popular Grad-CAM analytical technique while offering comparable predictive performance.