Aim: Left ventricular systolic dysfunction (LVSD) carries an increased risk for overt heart failure and mortality, yet treatable to mitigate disease progression. An artificial intelligence (AI)-enabled 12-lead electrocardiogram (ECG) model demonstrated promise in LVSD screening, but the performance dropped unexpectedly in external validation. We thus sought to train de novo models for LVSD detection and investigated their performance across multiple institutions and across a broader set of patient strata.
Methods and results: ECGs taken within 14 days of an echocardiogram were obtained from four academic hospitals (three in the United States and one in Japan). Four AI models were trained to detect patients with ejection fraction (EF) <40% using ECGs from each of the four institutions. All the models were then evaluated on the held-out test data set from the same institution and data from the three external institutions. Subgroup analyses stratified by patient characteristics and common ECG abnormalities were performed. A total of 221 846 ECGs were identified from the 4 institutions. While the Brigham and Women's Hospital (BWH)-trained and Keio-trained models yielded similar accuracy on their internal test data [area under the receiver operating curve (AUROC) 0.913 and 0.914, respectively], external validity was worse for the Keio-trained model (AUROC: 0.905-0.915 for BWH trained and 0.849-0.877 for Keio-trained model). Although ECG abnormalities including atrial fibrillation, left bundle branch block, and paced rhythm-reduced detection, the models performed robustly across patient characteristics and other ECG features.
Conclusion: While using the same model architecture, different data sets produced models with different performances for detecting low-EF highlighting the importance of external validation and extensive stratification analysis.
Aims: Familial hypercholesterolaemia (FH) is a disorder of LDL cholesterol clearance, resulting in increased risk of cardiovascular disease. Recently, we developed a Dutch Lipid Clinic Network (DLCN) criteria-based algorithm to facilitate FH detection in electronic health records (EHRs). In this study, we investigated the sensitivity of this and other algorithms in a genetically confirmed FH population.
Methods and results: All patients with a healthcare insurance-related coded diagnosis of 'primary dyslipidaemia' between 2018 and 2020 were assessed for genetically confirmed FH. Data were extracted at the time of genetic confirmation of FH (T1) and during the first visit in 2018-2020 (T2). We assessed the sensitivity of algorithms on T1 and T2 for DLCN ≥ 6 and compared with other algorithms [familial hypercholesterolaemia case ascertainment tool (FAMCAT), Make Early Diagnoses to Prevent Early Death (MEDPED), and Simon Broome (SB)] using EHR-coded data and using all available data (i.e. including non-coded free text). 208 patients with genetically confirmed FH were included. The sensitivity (95% CI) on T1 and T2 with EHR-coded data for DLCN ≥ 6 was 19% (14-25%) and 22% (17-28%), respectively. When using all available data, the sensitivity for DLCN ≥ 6 was 26% (20-32%) on T1 and 28% (22-34%) on T2. For FAMCAT, the sensitivity with EHR-coded data on T1 was 74% (67-79%) and 32% (26-39%) on T2, whilst sensitivity with all available data was 81% on T1 (75-86%) and 45% (39-52%) on T2. For Make Early Diagnoses to Prevent Early Death MEDPED and SB, using all available data, the sensitivity on T1 was 31% (25-37%) and 17% (13-23%), respectively.
Conclusions: The FAMCAT algorithm had significantly better sensitivity than DLCN, MEDPED, and SB. FAMCAT has the best potential for FH case-finding using EHRs.
Aims: To investigate the impact of coronavirus disease 2019 lockdown on trajectories of arterial pulse-wave velocity in a large population of users of connected smart scales that provide reliable measurements of pulse-wave velocity.
Methods and results: Pulse-wave velocity recordings obtained by Withings Heart Health & Body Composition Wi-Fi Smart Scale users before and during lockdown were analysed. We compared two demonstrative countries: France, where strict lockdown rules were enforced (n = 26 196) and Germany, where lockdown was partial (n = 26 847). Subgroup analysis was conducted in users of activity trackers and home blood pressure monitors. Linear growth curve modelling and trajectory clustering analyses were performed. During lockdown, a significant reduction in vascular stiffness, weight, blood pressure, and physical activity was observed in the overall population. Pulse-wave velocity reduction was greater in France than in Germany, corresponding to 5.2 month reduction in vascular age. In the French population, three clusters of stiffness trajectories were identified: decreasing (21.1%), stable (60.6%), and increasing pulse-wave velocity clusters (18.2%). Decreasing and increasing clusters both had higher pulse-wave velocity and vascular age before lockdown compared with the stable cluster. Only the decreasing cluster showed a significant weight reduction (-400 g), whereas living alone was associated with increasing pulse-wave velocity cluster. No clusters were identified in the German population.
Conclusions: During total lockdown in France, a reduction in pulse-wave velocity in a significant proportion of French users of connected smart bathroom scales occurred. The impact on long-term cardiovascular health remains to be established.