Search

We investigated a stacking ensemble method that combines multiple base learners within a database. The results on external validation across four large databases suggest a stacking ensemble could improve model transportability.

Feature importance is often used to explain clinical prediction models. In this work, we examine three challenges using experiments with electronic health record data: computational feasibility, choosing between methods, and interpretation of the resulting explanation. This work aims to create awareness of the disagreement between feature importance methods and underscores the need for guidance to practitioners how to deal with these discrepancies.

Context Hyperglycemia and autonomic dysfunction are bidirectionally related. Objective We investigated the association of longitudinal evolution of heart rate variability (HRV) with incident type 2 diabetes (T2D) among the general population. Methods We included 7630 participants (mean age 63.7 years, 58% women) from the population-based Rotterdam Study who had no history of T2D and atrial fibrillation at baseline and had repeated HRV assessments at baseline and during follow-up. We used joint models to assess the association between longitudinal evolution of heart rate and different HRV metrics (including the heart rate–corrected SD of the normal-to-normal RR intervals [SDNNc], and root mean square of successive RR-interval differences [RMSSDc]) with incident T2D. Models were adjusted for cardiovascular risk factors. Bidirectional Mendelian randomization (MR) using summary-level data was also performed. Results During a median follow-up of 8.6 years, 871 individuals developed incident T2D. One SD increase in heart rate (hazard ratio [HR] 1.20; 95% CI, 1.09-1.33), and log(RMSSDc) (HR 1.16; 95% CI, 1.01-1.33) were independently associated with incident T2D. The HRs were 1.54 (95% CI, 1.08-2.06) for participants younger than 62 years and 1.15 (95% CI, 1.01-1.31) for those older than 62 years for heart rate (P for interaction Conclusion Autonomic dysfunction precedes development of T2D, especially among younger individuals, while MR analysis suggests no causal relationship. More studies are needed to further validate our findings.

INTRODUCTION: External validation of prediction models is increasingly being seen as a minimum requirement for acceptance in clinical practice. However, the lack of interoperability of healthcare databases has been the biggest barrier to this occurring on a large scale. Recent improvements in database interoperability enable a standardized analytical framework for model development and external validation. External validation of a model in a new database lacks context, whereby the external validation can be compared with a benchmark in this database. Iterative pairwise external validation (IPEV) is a framework that uses a rotating model development and validation approach to contextualize the assessment of performance across a network of databases. As a use case, we predicted 1-year risk of heart failure in patients with type 2 diabetes mellitus.METHODS: The method follows a two-step process involving (1) development of baseline and data-driven models in each database according to best practices and (2) validation of these models across the remaining databases. We introduce a heatmap visualization that supports the assessment of the internal and external model performance in all available databases. As a use case, we developed and validated models to predict 1-year risk of heart failure in patients initializing a second pharmacological intervention for type 2 diabetes mellitus. We leveraged the power of the Observational Medical Outcomes Partnership common data model to create an open-source software package to increase the consistency, speed, and transparency of this process.RESULTS: A total of 403,187 patients from five databases were included in the study. We developed five models that, when assessed internally, had a discriminative performance ranging from 0.73 to 0.81 area under the receiver operating characteristic curve with acceptable calibration. When we externally validated these models in a new database, three models achieved consistent performance and in context often performed similarly to models developed in the database itself. The visualization of IPEV provided valuable insights. From this, we identified the model developed in the Commercial Claims and Encounters (CCAE) database as the best performing model overall.CONCLUSION: Using IPEV lends weight to the model development process. The rotation of development through multiple databases provides context to model assessment, leading to improved understanding of transportability and generalizability. The inclusion of a baseline model in all modelling steps provides further context to the performance gains of increasing model complexity. The CCAE model was identified as a candidate for clinical use. The use case demonstrates that IPEV provides a huge opportunity in a new era of standardised data and analytics to improve insight into and trust in prediction models at an unprecedented scale.

Objective This systematic review aims to assess how information from unstructured text is used to develop and validate clinical prognostic prediction models. We summarize the prediction problems and methodological landscape and determine whether using text data in addition to more commonly used structured data improves the prediction performance. Materials and Methods We searched Embase, MEDLINE, Web of Science, and Google Scholar to identify studies that developed prognostic prediction models using information extracted from unstructured text in a data-driven manner, published in the period from January 2005 to March 2021. Data items were extracted, analyzed, and a meta-analysis of the model performance was carried out to assess the added value of text to structured-data models. Results We identified 126 studies that described 145 clinical prediction problems. Combining text and structured data improved model performance, compared with using only text or only structured data. In these studies, a wide variety of dense and sparse numeric text representations were combined with both deep learning and more traditional machine learning methods. External validation, public availability, and attention for the explainability of the developed models were limited. Conclusion The use of unstructured text in the development of prognostic prediction models has been found beneficial in addition to structured data in most studies. The text data are source of valuable information for prediction model development and should not be neglected. We suggest a future focus on explainability and external validation of the developed models, promoting robust and trustworthy prediction models in clinical practice.

Aims In patients with Brugada syndrome (BrS) but without spontaneous Type-1 electrocardiogram, several electrocardiographic characteristics have been studied, including the β-angle. Previous studies suggested that the β-angle might be useful in distinguishing BrS-patients from patients with only suggestive repolarization patterns without performing sodium channel blocker provocation testing. In this study, we aimed to determine the diagnostic value of the β-angle in patients suspected of BrS. Methods and results A large cohort (n = 1430) of consecutive patients who underwent provocation testing was evaluated. β-angles were measured in leads V1, V2, and their corresponding positions over the second and third intercostal space. Receiver-operating characteristic curves were constructed and the diagnostic accuracy of previously reported β-angle cut-offs were calculated and evaluated. The importance of the β-angle for predicting the provocation test outcome was determined using a prediction model constructed with logistic regression. The optimum β-angle cut-off in our cohort for ruling out a positive provocation test was 15°; sensitivities were 80–98% and negative predictive values were 79–96% among the right precordial leads. Previously reported β-angle cut-offs performed less well, indicated by lower Youden indices. In the optimism-corrected prediction model [C-statistic: 0.78 (95% CI: 0.75–0.81)], the β-angle had large value (Z-score: 2.1–10.3) and aided construction of a nomogram to predict test outcome. Conclusion To predict the outcome of provocation testing for BrS, the β-angle alone does not demonstrate strong diagnostic characteristics. However, the β-angle is an important variable to predict provocation test outcome and thus has added value.

Background Vaccine information in European electronic health record (EHR) databases is represented using various clinical and database-specific coding systems and drug vocabularies. The lack of harmonization constitutes a challenge in reusing EHR data in collaborative benefit-risk studies about vaccines. Methods We designed an ontology of the properties that are commonly used in vaccine descriptions, called Ontology of Vaccine Descriptions (VaccO), with a dictionary for the analysis of multilingual vaccine descriptions. We implemented five algorithms for the alignment of vaccine coding systems, i.e., the identification of corresponding codes from different coding ystems, based on an analysis of the code descriptors. The algorithms were evaluated by comparing their results with manually created alignments in two reference sets including clinical and database-specific coding systems with multilingual code descriptors. Results The best-performing algorithm represented code descriptors as logical statements about entities in the VaccO ontology and used an ontology reasoner to infer common properties and identify corresponding vaccine codes. The evaluation demonstrated excellent performance of the approach (F-scores 0.91 and 0.96). Conclusion The VaccO ontology allows the identification, representation, and comparison of heterogeneous descriptions of vaccines. The automatic alignment of vaccine coding systems can accelerate the readiness of EHR databases in collaborative vaccine studies.

Background:Identification of rheumatoid arthritis (RA) patients at high risk of adverse health outcomes remains a major challenge. We aimed to develop and validate prediction models for a variety of adverse health outcomes in RA patients initiating first-line methotrexate (MTX) monotherapy. Methods:Data from 15 claims and electronic health record databases across 9 countries were used. Models were developed and internally validated on Optum®De-identified Clinformatics®Data Mart Database using L1- regularized logistic regression to estimate the risk of adverse health outcomes within 3 months (leukopenia, pancytopenia, infection), 2 years (myocardial infarction (MI) and stroke), and 5 years (cancers [colorectal, breast, uterine] after treatment initiation. Candidate predictors included demographic variables and past medical history. Models were externally validated on all other databases. Performance was assessed using the area under the receiver operator characteristic curve (AUC) and calibration plots. Findings:Models were developed and internally validated on 21,547 RA patients and externally validated on 131,928 RA patients. Models for serious infection (AUC: internal 0.74, external ranging from 0.62 to 0.83), MI (AUC: internal 0.76, external ranging from 0.56 to 0.82), and stroke (AUC: internal 0.77, external ranging from 0.63 to 0.95), showed good discrimination and adequate calibration. Models for the other outcomes showed modest internal discrimination (AUC Interpretation:We developed and validated prediction models for a variety of adverse health outcomes in RA patients initiating first-line MTX monotherapy. Final models for serious infection, MI, and stroke demonstrated good performance across multiple databases and can be studied for clinical use. Funding:This activity under the European Health Data&Evidence Network (EHDEN) has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 806968. This Joint Under- taking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA.

Background Clinical guidelines categorize atrial fibrillation (AF) based on the temporality of AF events. Due to its dependence on event duration, this classification is not applicable to population-based cohort settings. We aimed to develop a simple and standardized method to classify AF patterns at population level. Additionally, we compared the longitudinal trajectories of cardiovascular risk factors preceding the AF patterns, and between men and women. Methods Between 1990 and 2014, participants from the population-based Rotterdam study were followed for AF status, and categorized into ‘single-documented AF episode’, ‘multiple-documented AF episodes’, or ‘long-standing persistent AF’. Using repeated measurements we created linear mixed-effects models to assess the longitudinal evolution of risk factors prior to AF diagnosis. Results We included 14,061 participants (59.1% women, mean age 65.4 ± 10.2 years). After a median follow-up of 9.4 years (interquartile range 8.27), 1,137 (8.1%) participants were categorized as ‘single-documented AF episode’, 208 (1.5%) as ‘multiple-documented AF episodes’, and 57 (0.4%) as ‘long-standing persistent AF’. In men, we found poorer trajectories of weight and waist circumference preceding ‘long-standing persistent AF’ as compared to the other patterns. In women, we found worse trajectories of all risk factors between ‘long-standing persistent AF’ and the other patterns. Conclusion We developed a standardized method to classify AF patterns in the general population. Participants categorized as ‘long-standing persistent AF’ showed poorer trajectories of cardiovascular risk factors prior to AF diagnosis, as compared to the other patterns. Our findings highlight sex differences in AF pathophysiology and provide insight into possible risk factors of AF patterns. Graphical abstract

Background Sex differences and causality of the association between heart rate variability (HRV) and atrial fibrillation (AF) in the general population remain unclear. Methods 12,334 participants free of AF from the population-based Rotterdam Study were included. Measures of HRV including the standard deviation of normal RR intervals (SDNN), SDNN corrected for heart rate (SDNNc), RR interval differences (RMSSD), RMSSD corrected for heart rate (RMSSDc), and heart rate were assessed at baseline and follow-up examinations. Joint models, adjusted for cardiovascular risk factors, were used to determine the association between longitudinal measures of HRV with new-onset AF. Genetic variants for HRV were used as instrumental variables in a Mendelian randomization (MR) analysis using genome-wide association studies (GWAS) summary-level data. Results During a median follow-up of 9.4 years, 1302 incident AF cases occurred among 12,334 participants (mean age 64.8 years, 58.3% women). In joint models, higher SDNN (fully-adjusted hazard ratio (HR), 95% confidence interval (CI) 1.24, 1.04–1.47, p = 0.0213), and higher RMSSD (fully-adjusted HR, 95% CI 1.33, 1.13–1.54, p = 0.0010) were significantly associated with new-onset AF. Sex-stratified analyses showed that the associations were mostly prominent among women. In MR analyses, a genetically determined increase in SDNN (odds ratio (OR), 95% CI 1.60, 1.27–2.02, p = 8.36 × 10–05), and RMSSD (OR, 95% CI 1.56, 1.31–1.86, p = 6.32 × 10–07) were significantly associated with an increased odds of AF. Conclusion Longitudinal measures of uncorrected HRV were significantly associated with new-onset AF, especially among women. MR analyses supported the causal relationship between uncorrected measures of HRV with AF. Our findings indicate that measures to modulate HRV might prevent AF in the general population, in particular in women. Graphical abstract AF; atrial fibrillation, GWAS; genome-wide association study, IVW; inverse variance weighted, MR; Mendelian randomization, MR-PRESSO; MR-egger and mendelian randomization pleiotropy residual sum and outlier, RMSSD; root mean square of successive RR interval differences, RMSSDc; root mean square of successive RR interval differences corrected for heart rate, SDNN; standard deviation of normal to normal RR intervals, SDNNc; standard deviation of normal to normal RR intervals corrected for heart rate, WME; weighted median estimator. aRotterdam Study n=12,334 bHRV GWAS n=53,174 cAF GWAS n=1,030,836

Objectives: To develop and validate classifiers for automatic detection of actionable findings and documentation of nonroutine communication in routinely delivered radiology reports. Methods: Two radiologists annotated all actionable findings and communication mentions in a training set of 1,306 radiology reports and a test set of 1,000 reports randomly selected from the electronic health record system of a large tertiary hospital. Various feature sets were constructed based on the impression section of the reports using different preprocessing steps (stemming, removal of stop words, negations, and previously known or stable findings) and n-grams. Random forest classifiers were trained to detect actionable findings, and a decision-rule classifier was trained to find communication mentions. Classifier performance was evaluated by the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Results: On the training set, the actionable finding classifier with the highest cross-validated performance was obtained for a feature set of unigrams, after stemming and removal of negated, known, and stable findings. On the test set, this classifier achieved an AUC of 0.876 (95% CI 0.854–0.898). The classifier for communication detection was trained after negation removal, using unigrams as features. The resultant decision rule had a sensitivity of 0.841 (95% CI 0.706–0.921) and specificity of 0.990 (95% CI 0.981–0.994) on the test set. Conclusions: Automatic detection of actionable findings and subsequent communication in routinely delivered radiology reports is possible. This can serve quality control purposes and may alert radiologists to the presence of actionable findings during reporting. Key Points: • Classifiers were developed for automatic detection of the broad spectrum of actionable findings and subsequent communication mentions in routinely delivered radiology reports. • Straightforward report preprocessing and simple feature sets can produce well-performing classifiers. • The resultant classifiers show good performance for detection of actionable findings and excellent performance for detection of communication mentions.

Aims: Hydroxychloroquine and chloroquine ([hydroxy]chloroquine) are drugs used to treat malaria and rheumatological disorders and were recently suggested as beneficial for prevention and treatment of patients with coronavirus disease 2019 (COVID-19) due to SARS-CoV-2 infection. However, longitudinal studies to assess the electrocardiographic and cardiotoxic effects of these drugs are limited. In this study, we aimed to investigate the effect of these drugs on QTc-interval and incidence of sudden cardiac death (SCD). Methods We designed a longitudinal follow-up study of individuals within the prospective population-based Rotterdam Study. Eligible individuals had available data on medication and repeated ECG measurements. The study period was between 1 January 1991 and 1 January 2014. We studied on current and past use of [hydroxy]chloroquine as a time-varying exposure; high versus low daily dose of [hydroxy]chloroquine. QTc-interval duration, and the occurrence of SCD were the main outcomes. SCD was defined as an unexpected and sudden death due to cardiac arrhythmia within one hour of the onset of acute symptoms, and in patients without cardiac symptoms within 24 hours before death. Results Among the study population of 14 594 individuals (58.8% women) with an average age of 65 years, 346 patients used [hydroxy]chloroquine at any time during follow-up. The total number of SCD cases was 609. In a multiple linear mixed model analysis, the current use of [hydroxy]chloroquine was associated with a significantly increased duration of the QTc-interval of 8.1 ms (95% CI: 3.6; 12.6) compared with non-users. The association was stronger among current-high daily dosage [15.3 (95%CI: 7.0; 23.6)] compared with current-low daily dosage [5.5 (95%CI: 0.4; 10.7)] users. In a Cox proportional hazard regression analysis, the risk of SCD was significantly higher in participants who were current users of [hydroxy]chloroquine than in non-users [adjusted hazard ratio; 3.7 (95%CI: 1.1; 12.6)]. Conclusions In this longitudinal study, persons who received [hydroxy]chloroquine had an increased QTc-interval duration and the association was dose-dependent. [Hydroxy]chloroquine was associated with a significantly increased risk of SCD. As long as their activity against COVID-19 is controversial, cardiotoxicity is a strong argument against using these drugs to treat COVID-19 infections.

Background: The P-wave duration (PWD) is an electrocardiographic measurement that represents cardiac conduction in the atria. Shortened or prolonged PWD is associated with atrial fibrillation (AF). We used exome-chip data to examine the associations between common and rare variants with PWD. Methods: Fifteen studies comprising 64 440 individuals (56 943 European, 5681 African, 1186 Hispanic, 630 Asian) and ≈230 000 variants were used to examine associations with maximum PWD across the 12-lead ECG. Meta-analyses summarized association results for common variants; gene-based burden and sequence kernel association tests examined low-frequency variant-PWD associations. Additionally, we examined the associations between PWD loci and AF using previous AF genome-wide association studies. Results: We identified 21 common and low-frequency genetic loci (14 novel) associated with maximum PWD, including several AF loci ( TTN , CAND2 , SCN10A , PITX2 , CAV1 , SYNPO2L , SOX5 , TBX5, MYH6, RPL3L ). The top variants at known sarcomere genes ( TTN, MYH6 ) were associated with longer PWD and increased AF risk. However, top variants at other loci (eg, PITX2 and SCN10A ) were associated with longer PWD but lower AF risk. Conclusions: Our results highlight multiple novel genetic loci associated with PWD, and underscore the shared mechanisms of atrial conduction and AF. Prolonged PWD may be an endophenotype for several different genetic mechanisms of AF.

Background and objectivesThere is an increasing interest to use real-world data to illustrate how patients with specific medical conditions are treated in real life. Insight in the current treatment practices helps to improve and tailor patient care. We aim to provide an easy tool to support the development and analysis of treatment pathways for a wide variety of medical conditions.MethodsWe formally defined the process of constructing treatment pathways and developed an open-source R package TreatmentPatterns (https://github.com/mi-erasmusmc/TreatmentPatterns) to enable a reproducible and timely analysis of treatment patterns.ResultsThe developed package supports the analysis of treatment patterns of specific populations of interest. We demonstrate the functionality of the package by analyzing the treatment patterns of three common chronic diseases (type II diabetes mellitus, hypertension, and depression) in the Dutch Integrated Primary Care Information (IPCI) database.ConclusionTreatmentPatterns is a tool to make the analysis of treatment patterns more accessible, more standardized, and more interpretation friendly. This tool can facilitate and contribute to the accumulation of knowledge on real-world treatment patterns across disease domains. We encourage researchers to further adjust and add custom analysis to the package based on their research needs.

The QT interval is an electrocardiographic measure representing the sum of ventricular depolarization (QRS duration) and repolarization (JT interval). Abnormalities of the QT interval are associated with potentially fatal ventricular arrhythmia. We conducted genome-wide multi-ancestry analyses in >250,000 individuals and identified 177, 156 and 121 independent loci for QT, JT and QRS, respectively, including a male-specific X-chromosome locus. Using gene-based rare-variant methods, we identified associations with Mendelian disease genes. Enrichments were observed in established pathways for QT and JT, with new genes indicated in insulin-receptor signalling and cardiac energy metabolism. In contrast, connective tissue components and processes for cell growth and extracellular matrix interactions were significantly enriched for QRS. We demonstrate polygenic risk score associations with atrial fibrillation, conduction disease and sudden cardiac death. Prioritization of druggable genes highlighted potential therapeutic targets for arrhythmia. Together, these results substantially advance our understanding of the genetic architecture of ventricular depolarization and repolarization.

Background Hyperglycemia and autonomic dysfunction are bidirectionally related. The implications from former cross-sectional studies regarding the association of autonomic dysfunction and diabetes are limited due to the potential reverse causation. We aimed to investigate the association of longitudinal evolution of heart rate variability (HRV) with incident type 2 diabetes mellitus (T2D) among the general population. Methods Using data from the prospective population-based cohort study, 7630 participants with available repeated assessments of HRV, who were free of T2D and atrial fibrillation, were included. Measurements of HRV were analyzed from standard 10-s electrocardiograms records, including heart rate and two time-domain HRV markers: the heart-rate corrected standard deviation of the normal-to-normal RR intervals (SDNNc) and heart-rate corrected root mean square of successive RR-interval differences (RMSSDc). Joint models with the Bayes approach were performed to assess the associations between longitudinal evolutions of heart rate and different HRV metrics with incident T2D during follow-up. Models were adjusted for age, sex, body mass index, smoking status, systolic blood pressure, total and high-density lipoprotein cholesterol, use of blood pressure-lowering or lipid-lowering medication, and prevalent cardiovascular disease. Spearmen correlation was used to examine the associations between HRV metrics and glycemic traits (fasting blood glucose, insulin, and homeostatic model assessment of insulin resistance (HOMA-IR) and beta-cell function (HOMA-β)) at baseline. Results During a median follow-up time of 8.6 years, 871 individuals developed T2D. Based on joint models, one standard deviation (SD) increment of heart rate (hazard ratio [HR], 1.21; 95% CI, 1.09–1.34), log(SDNNc) (HR, 1.10; 95% CI, 0.94–1.27), and log(RMSSDc) (HR, 1.15; 95% CI, 1.02–1.31) was associated with the risk of developing T2D in fully-adjusted models. Subgroup analyses suggested stronger associations among men. The HRs of incident T2D per SD increment were 1.25 (1.09, 1.43) for men and 1.16 (0.99, 1.35) for women for heart rate, 1.23 (1.01, 1.51) for men and 0.97 (0.78, 1.20) for women for log(SDNNc), and 1.24 (1.05, 1.48) for men and 1.09 (0.89, 1.31) for women for log(RMSSDc). Results did not change after excluding participants who were prediabetes at baseline or after exclusion of underweight participants. Spearmen correlation indicated that heart rate was significantly associated with baseline glycemic traits, while RMSSDc was only related to fasting blood glucose. Conclusion Our study suggests that both higher heart rate and heart rate variability are associated with an increased risk of T2D development, especially among men. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Erasmus MC and Erasmus University Rotterdam; Netherlands Organization for Scientific Research; Netherlands Organization for Health Research and Development (ZonMw); Research Institute for Diseases in the Elderly; Netherlands Genomics Initiative; Netherlands Ministry of Education, Culture and Science; Netherlands Ministry of Health, Welfare and Sports; European Commission; and Municipality of Rotterdam. We would like to thank the China Scholarship Council for the scholarship to K.W. Forest plot

Pre-competitive data sharing can offer the pharmaceutical industry significant benefits in terms of reducing the time and costs involved in getting a new drug to market through more informed testing strategies and knowledge gained by pooling data. If sufficient data is shared and can be co-analyzed, then it can also offer the potential for reduced animal usage and improvements in the in silico prediction of toxicological effects. Data sharing benefits can be further enhanced by applying the FAIR Guiding Principles, reducing time spent curating, transforming and aggregating datasets and allowing more time for data mining and analysis. We hope to facilitate data sharing by other organizations and initiatives by describing lessons learned as part of the Enhancing TRANslational SAFEty Assessment through Integrative Knowledge Management (eTRANSAFE) project, an Innovative Medicines Initiative (IMI) partnership which aims to integrate publicly available data sources with proprietary preclinical and clinical data donated by pharmaceutical organizations. Methods to foster trust and overcome non-technical barriers to data sharing such as legal and IPR (intellectual property rights) are described, including the security requirements that pharmaceutical organizations generally expect to be met. We share the consensus achieved among pharmaceutical partners on decision criteria to be included in internal clearance procedures used to decide if data can be shared. We also report on the consensus achieved on specific data fields to be excluded from sharing for sensitive preclinical safety and pharmacology data that could otherwise not be shared. This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 777365 (“eTRANSAFE”). This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA.

Patients with Brugada syndrome (BrS) can show a leftward deviation of the frontal QRS-axis upon provocation with sodium channel blockers. The cause of this axis change is unclear. In this study, we aimed to determine (1) the prevalence of this left axis deviation and (2) to evaluate its cause, using the insights that could be derived from vectorcardiograms. Hence, from a large cohort of patients who underwent ajmaline provocation testing (n = 1430), we selected patients in whom a type-1 BrS-ECG was evoked (n = 345). Depolarization and repolarization parameters were analyzed for reconstructed vectorcardiograms and were compared between patients with and without a &gt, 30&deg, leftward axis shift. We found (1) that the prevalence of a left axis deviation during provocation testing was 18% and (2) that this left axis deviation was not explained by terminal conduction slowing in the right ventricular outflow tract (4th QRS-loop quartile: +17 &plusmn, 14 ms versus +13 &plusmn, 15 ms, nonsignificant) but was associated with a more proximal conduction slowing (1st QRS-loop quartile: +12[8, 18] ms versus +8[4, 12] ms, p &lt, 0.001 and 3rd QRS-loop quartile: +12 &plusmn, 10 ms versus +5 &plusmn, 7 ms, p &lt, 0.001). There was no important heterogeneity of the action potential morphology (no difference in the ventricular gradient), but a left axis deviation did result in a discordant repolarization (spatial QRS-T angle: 122[59, 147]&deg, versus 44[25, 91]&deg, p &lt, 0.001). Thus, although the development of the type-1 BrS-ECG is characterized by a terminal conduction delay in the right ventricle, BrS-patients with a left axis deviation upon sodium channel blocker provocation have an additional proximal conduction slowing, which is associated with a subsequent discordant repolarization. Whether this has implications for risk stratification is still undetermined.

Background A Brugada syndrome (BrS) diagnosis is made upon documenting its characteristic type-1 ECG abnormality, being either spontaneously present or provoked by drug challenge, in the absence of co-morbidities resulting in similar ECG characteristics. In patients without type-1 ECG but in whom a suspicion of BrS exists, several ECG characteristics have been proposed that could aid in distinguishing BrS from non-BrS, particularly in individuals with incomplete right bundle branch block. Purpose To validate the diagnostic value of ECG characteristics that were previously proposed. Methods For this retrospective study, ECGs were evaluated from consecutive patients suspected of BrS who underwent ajmaline testing (n=1392). Baseline ECG and peak-effect ECG were analyzed with the MEANS software. Parameters considered were P, PR, QRS, S, JT, QT and QTc durations (all in ms), J and S amplitudes (both in mV), and QRS and T axes (both in °). ECG evaluations also included b-angles (in °) and duration of the base of a triangle (DBT) at 0.5 mV from the r' (in ms) in right precordial leads (V1, V2, V1ic3, V2ic3). Results Diagnostic type-1 ECGs were uncovered in 345 subjects and 1047 subjects had a negative test. Patients with BrS had significantly more atrial and ventricular conduction abnormalities (P, PR, QRS, and S duration) at baseline; these differences, except in PR duration, were aggravated upon ajmaline administration. In particular, the b-angle (e.g., V1: 28° [20; 46.5] versus 18° [12; 24], p15°, negative predictive values ranged from 78.6–95.5 and sensitivities ranged from 80–98.1% among different leads. Conclusions Patients with BrS show more baseline ECG abnormalities and display more conduction slowing upon ajmaline infusion. The b-angle and DBT do not demonstrate strong enough diagnostic characteristics to serve as a stand-alone diagnostic tool. Drug provocation tests remain necessary for diagnosing BrS in patients without a spontaneous type-1 ECG. Funding Acknowledgement Type of funding source: None

ObjectivesMajor ECG abnormalities have been associated with increased risk of cardiovascular disease (CVD) burden in asymptomatic populations. However, sex differences in occurrence of major ECG abnormalities have been poorly studied, particularly across ethnic groups. The objectives were to investigate (1) sex differences in the prevalence of major and, as a secondary outcome, minor ECG abnormalities, (2) whether patterns of sex differences varied across ethnic groups, by age and (3) to what extent conventional cardiovascular risk factors contributed to observed sex differences.DesignCross-sectional analysis of population-based study.SettingMulti-ethnic, population-based Healthy Life in an Urban Setting cohort, Amsterdam, the Netherlands.Participants8089 men and 11 369 women of Dutch, South-Asian Surinamese, African Surinamese, Ghanaian, Turkish and Moroccan origin aged 18–70 years without CVD.Outcome measuresAge-adjusted and multivariable logistic regression analyses were performed to study sex differences in prevalence of major and, as secondary outcome, minor ECG abnormalities in the overall population, across ethnic groups and by age-groups (18–35, 36–50 and >50 years).ResultsMajor and minor ECG abnormalities were less prevalent in women than men (4.6% vs 6.6% and 23.8% vs 39.8%, respectively). After adjustment for conventional risk factors, sex differences in major abnormalities were smaller in ethnic minority groups (OR ranged from 0.61 in Moroccans to 1.32 in South-Asian Surinamese) than in the Dutch (OR 0.49; 95% CI 0.36 to 0.65). Only in South-Asian Surinamese, women did not have a lower odds than men (OR 1.32; 95% CI 0.96 to 1.84). The pattern of smaller sex differences in ethnic minority groups was more pronounced in older than in younger age-groups.ConclusionsThe prevalence of major ECG abnormalities was lower in women than men. However, sex differences were less apparent in ethnic minority groups. Conventional risk factors did not contribute substantially to observed sex differences.

Objective: Provide guidance on sample size considerations for developing predictive models by empirically establishing the adequate sample size, which balances the competing objectives of improving model performance and reducing model complexity as well as computational requirements. Materials and Methods: We empirically assess the effect of sample size on prediction performance and model complexity by generating learning curves for 81 prediction problems in three large observational health databases, requiring training of 17,248 prediction models. The adequate sample size was defined as the sample size for which the performance of a model equalled the maximum model performance minus a small threshold value. Results: The adequate sample size achieves a median reduction of the number of observations between 9.5% and 78.5% for threshold values between 0.001 and 0.02. The median reduction of the number of predictors in the models at the adequate sample size varied between 8.6% and 68.3%, respectively. Discussion: Based on our results a conservative, yet significant, reduction in sample size and model complexity can be estimated for future prediction work. Though, if a researcher is willing to generate a learning curve a much larger reduction of the model complexity may be possible as suggested by a large outcome-dependent variability. Conclusion: Our results suggest that in most cases only a fraction of the available data was sufficient to produce a model close to the performance of one developed on the full data set, but with a substantially reduced model complexity.

Artificial intelligence (AI) has huge potential to improve the health and well-being of people, but adoption in clinical practice is still limited. Lack of transparency is identified as one of the main barriers to implementation, as clinicians should be confident the AI system can be trusted. Explainable AI has the potential to overcome this issue and can be a step towards trustworthy AI. In this paper we review the recent literature to provide guidance to researchers and practitioners on the design of explainable AI systems for the health-care domain and contribute to formalization of the field of explainable AI. We argue the reason to demand explainability determines what should be explained as this determines the relative importance of the properties of explainability (i.e. interpretability and fidelity). Based on this, we propose a framework to guide the choice between classes of explainable AI methods (explainable modelling versus post-hoc explanation; model-based, attribution-based, or example-based explanations; global and local explanations). Furthermore, we find that quantitative evaluation metrics, which are important for objective standardized evaluation, are still lacking for some properties (e.g. clarity) and types of explanations (e.g. example-based methods). We conclude that explainable modelling can contribute to trustworthy AI, but the benefits of explainability still need to be proven in practice and complementary measures might be needed to create trustworthy AI in health care (e.g. reporting data quality, performing extensive (external) validation, and regulation).

Background Early prehospital recognition of critical conditions such as ST‐segment–elevation myocardial infarction (STEMI) has prognostic relevance. Current international electrocardiographic STEMI thresholds are predominantly based on individuals of Western European descent. However, because of ethnic electrocardiographic variability both in health and disease, there is a need to reevaluate diagnostic ST‐segment elevation thresholds for different populations. We hypothesized that fulfillment of ST‐segment elevation thresholds of STEMI criteria (STE‐ECGs) in apparently healthy individuals is ethnicity dependent. Methods and Results HELIUS (Healthy Life in an Urban Setting) is a multiethnic cohort study including 10 783 apparently healthy subjects of 6 different ethnicities (African Surinamese, Dutch, Ghanaian, Moroccan, South Asian Surinamese, and Turkish). Prevalence of STE‐ECGs across ethnicities, sexes, and age groups was assessed with respect to the 2 international STEMI thresholds: sex and age specific versus sex specific. Mean prevalence of STE‐ECGs was 2.8% to 3.4% (age/sex‐specific and sex‐specific thresholds, respectively), although with large ethnicity‐dependent variability. Prevalences in Western European Dutch were 2.3% to 3.0%, but excessively higher in young (45% STE‐ECGs. Conclusions The accuracy of diagnostic tests partly relies on background prevalence in healthy individuals. In apparently healthy subjects, there is a highly variable ethnicity‐dependent prevalence of ECGs with ST‐segment elevations exceeding STEMI thresholds. This has potential consequences for STEMI evaluations in individuals who are not of Western European descent, putatively resulting in adverse outcomes with both over‐ and underdiagnosis of STEMI.