Benchmarking the Impact of Noise on Deep Learning-Based Classification of Atrial Fibrillation in 12-Lead ECG

Electrocardiography analysis is widely used in various clinical applications and Deep Learning models for classification tasks are currently in the focus of research. Due to their data-driven character, they bear the potential to handle signal noise efficiently, but its influence on the accuracy of these methods is still unclear. Therefore, we benchmark the influence of four types of noise on the accuracy of a Deep Learning-based method for atrial fibrillation detection in 12-lead electrocardiograms. We use a subset of a publicly available dataset (PTB-XL) and use the metadata provided by human experts regarding noise for assigning a signal quality to each electrocardiogram. Furthermore, we compute a quantitative signal-to-noise ratio for each electrocardiogram. We analyze the accuracy of the Deep Learning model with respect to both metrics and observe that the method can robustly identify atrial fibrillation, even in cases signals are labelled by human experts as being noisy on multiple leads. False positive and false negative rates are slightly worse for data being labelled as noisy. Interestingly, data annotated as showing baseline drift noise results in an accuracy very similar to data without. We conclude that the issue of processing noisy electrocardiography data can be addressed successfully by Deep Learning methods that might not need preprocessing as many conventional methods do.

Analysis of a Deep Learning Model for 12-Lead ECG Classification Reveals Learned Features Similar to Diagnostic Criteria

Despite their remarkable performance, deep neural networks remain unadopted in clinical practice, which is considered to be partially due to their lack of explainability. In this work, we apply explainable attribution methods to a pre-trained deep neural network for abnormality classification in 12-lead electrocardiography to open this "black box" and understand the relationship between model prediction and learned features. We classify data from two public databases (CPSC 2018, PTB-XL) and the attribution methods assign a "relevance score" to each sample of the classified signals. This allows analyzing what the network learned during training, for which we propose quantitative methods: average relevance scores over a) classes, b) leads, and c) average beats. The analyses of relevance scores for atrial fibrillation and left bundle branch block compared to healthy controls show that their mean values a) increase with higher classification probability and correspond to false classifications when around zero, and b) correspond to clinical recommendations regarding which lead to consider. Furthermore, c) visible P-waves and concordant T-waves result in clearly negative relevance scores in atrial fibrillation and left bundle branch block classification, respectively. Results are similar across both databases despite differences in study population and hardware. In summary, our analysis suggests that the DNN learned features similar to cardiology textbook knowledge.

Consistency as a Data Quality Measure for German Corona Consensus Items Mapped from National Pandemic Cohort Network Data Collections

BACKGROUND

 As a national effort to better understand the current pandemic, three cohorts collect sociodemographic and clinical data from coronavirus disease 2019 (COVID-19) patients from different target populations within the German National Pandemic Cohort Network (NAPKON). Furthermore, the German Corona Consensus Dataset (GECCO) was introduced as a harmonized basic information model for COVID-19 patients in clinical routine. To compare the cohort data with other GECCO-based studies, data items are mapped to GECCO. As mapping from one information model to another is complex, an additional consistency evaluation of the mapped items is recommended to detect possible mapping issues or source data inconsistencies.

OBJECTIVES

 The goal of this work is to assure high consistency of research data mapped to the GECCO data model. In particular, it aims at identifying contradictions within interdependent GECCO data items of the German national COVID-19 cohorts to allow investigation of possible reasons for identified contradictions. We furthermore aim at enabling other researchers to easily perform data quality evaluation on GECCO-based datasets and adapt to similar data models.

METHODS

 All suitable data items from each of the three NAPKON cohorts are mapped to the GECCO items. A consistency assessment tool (dqGecco) is implemented, following the design of an existing quality assessment framework, retaining their-defined consistency taxonomies, including logical and empirical contradictions. Results of the assessment are verified independently on the primary data source.

RESULTS

 Our consistency assessment tool helped in correcting the mapping procedure and reveals remaining contradictory value combinations within COVID-19 symptoms, vital signs, and COVID-19 severity. Consistency rates differ between the different indicators and cohorts ranging from 95.84% up to 100%.

CONCLUSION

 An efficient and portable tool capable of discovering inconsistencies in the COVID-19 domain has been developed and applied to three different cohorts. As the GECCO dataset is employed in different platforms and studies, the tool can be directly applied there or adapted to similar information models.

Information Technology Systems for Infection Control in German University Hospitals-Results of a Structured Survey a Year into the Severe Acute Respiratory Syndrome Coronavirus 2 Pandemic

BACKGROUND

 Digitalization is playing a major role in mastering the current coronavirus 2019 (COVID-19) pandemic. However, several outbreaks of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in German hospitals last year have shown that many of the surveillance and warning mechanisms related to infection control (IC) in hospitals need to be updated.

OBJECTIVES

 The main objective of the following work was to assess the state of information technology (IT) systems supporting IC and surveillance in German university hospitals in March 2021, almost a year into the SARS-CoV-2 pandemic.

METHODS

 As part of the National Research Network for Applied Surveillance and Testing project within the Network University Medicine, a cross-sectional survey was conducted to assess the situation of IC IT systems in 36 university hospitals in Germany.

RESULTS

 Among the most prominent findings were the lack of standardization of IC IT systems and the predominant use of commercial IC IT systems, while the vast majority of hospitals reported inadequacies in the features their IC IT systems provide for their daily work. However, as the pandemic has shown that there is a need for systems that can help improve health care, several German university hospitals have already started this upgrade independently.

CONCLUSIONS

 The deep challenges faced by the German health care sector regarding the integration and interoperability of IT systems designed for IC and surveillance are unlikely to be solved through punctual interventions and require collaboration between educational, medical, and administrative institutions.

Severity, predictors and clinical correlates of Post-COVID syndrome (PCS) in Germany: A prospective, multi-centre, population-based cohort study

BACKGROUND

Post-COVID syndrome (PCS) is an important sequela of COVID-19, characterised by symptom persistence for >3 months, post-acute symptom development, and worsening of pre-existing comorbidities. The causes and public health impact of PCS are still unclear, not least for the lack of efficient means to assess the presence and severity of PCS.

METHODS

COVIDOM is a population-based cohort study of polymerase chain reaction (PCR) confirmed cases of SARS-CoV-2 infection, recruited through public health authorities in three German regions (Kiel, Berlin, Würzburg) between November 15, 2020 and September 29, 2021. Main inclusion criteria were (i) a PCR confirmed SARS-CoV-2 infection and (ii) a period of at least 6 months between the infection and the visit to the COVIDOM study site. Other inclusion criteria were written informed consent and age ≥18 years. Key exclusion criterion was an acute reinfection with SARS-CoV-2. Study site visits included standardised interviews, in-depth examination, and biomaterial procurement. In sub-cohort Kiel-I, a PCS (severity) score was developed based upon 12 long-term symptom complexes. Two validation sub-cohorts (Würzburg/Berlin, Kiel-II) were used for PCS score replication and identification of clinically meaningful predictors. This study is registered at clinicaltrials.gov (NCT04679584) and at the German Registry for Clinical Studies (DRKS, DRKS00023742).

FINDINGS

In Kiel-I (n = 667, 57% women), 90% of participants had received outpatient treatment for acute COVID-19. Neurological ailments (61·5%), fatigue (57·1%), and sleep disturbance (57·0%) were the most frequent persisting symptoms at 6-12 months after infection. Across sub-cohorts (Würzburg/Berlin, n = 316, 52% women; Kiel-II, n = 459, 56% women), higher PCS scores were associated with lower health-related quality of life (EQ-5D-5L-VAS/-index: r = -0·54/ -0·56, all p < 0·0001). Severe, moderate, and mild/no PCS according to the individual participant's PCS score occurred in 18·8%, 48·2%, and 32·9%, respectively, of the Kiel-I sub-cohort. In both validation sub-cohorts, statistically significant predictors of the PCS score included the intensity of acute phase symptoms and the level of personal resilience.

INTERPRETATION

PCS severity can be quantified by an easy-to-use symptom-based score reflecting acute phase disease burden and general psychological predisposition. The PCS score thus holds promise to facilitate the clinical diagnosis of PCS, scientific studies of its natural course, and the development of therapeutic interventions.

FUNDING

The COVIDOM study is funded by the Network University Medicine (NUM) as part of the National Pandemic Cohort Network (NAPKON).

Assessment of the Consistency of Categorical Features Within the DZHK Biobanking Basic Set

Data quality in health research encompasses a broad range of aspects and indicators. While some indicators are generic and can be calculated without domain knowledge, others require information about a specific data element. Even more complex are indicators addressing contradictions, that stem from implausible combinations of multiple data elements. In this paper, we investigate how contradictions within interdependent categorical data can be identified and if they give additional information about possible quality issues, their cause, and mitigation options. The 19 data elements that represent four biosample types including their pre-analytic states within the DZHK Biobanking basic set are exported to the CDISC Operational Data Model (ODM), transformed and loaded into a tranSMART instance. Through the implementation of a data quality assessment workflow as a SmartR plug-in, statistical information about the domain-specific consistency of interdependent values are retrieved, assessed, and visualized. Data quality indicators have been selected for the assessment according to common recommendations found in the literature. Different contradictions could be discovered in the dataset including mismatch of interdependent values in the pre-analytic states of blood and urine samples, as well as primary and aliquoted samples. The overall assessment rating shows that 99.61% of the interdependent values are free of contradictions. However, measures within the EDC design to avoid contradictions may result in overestimated missing rates in automatic, item-based quality assessment checks. Through consistency checks on interdependent categorical features, we demonstrated that consistency flaws can be found in the categorical data of biobanking metadata and that they can help to detect issues in the data entry process. Our approach underscores the importance of domain knowledge in the definition of the consistency rules but also knowledge about the EDC implementation of such consistency rules to consider the impact on item-based quality indicators.

The German National Pandemic Cohort Network (NAPKON): rationale, study design and baseline characteristics

The German government initiated the Network University Medicine (NUM) in early 2020 to improve national research activities on the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) pandemic. To this end, 36 German Academic Medical Centers started to collaborate on 13 projects, with the largest being the National Pandemic Cohort Network (NAPKON). The NAPKON’s goal is creating the most comprehensive Coronavirus Disease 2019 (COVID-19) cohort in Germany. Within NAPKON, adult and pediatric patients are observed in three complementary cohort platforms (Cross-Sectoral, High-Resolution and Population-Based) from the initial infection until up to three years of follow-up. Study procedures comprise comprehensive clinical and imaging diagnostics, quality-of-life assessment, patient-reported outcomes and biosampling. The three cohort platforms build on four infrastructure core units (Interaction, Biosampling, Epidemiology, and Integration) and collaborations with NUM projects. Key components of the data capture, regulatory, and data privacy are based on the German Centre for Cardiovascular Research. By April 01, 2022, 34 university and 40 non-university hospitals have enrolled 5298 patients with local data quality reviews performed on 4727 (89%). 47% were female, the median age was 52 (IQR 36–62-) and 50 pediatric cases were included. 44% of patients were hospitalized, 15% admitted to an intensive care unit, and 12% of patients deceased while enrolled. 8845 visits with biosampling in 4349 patients were conducted by April 03, 2022. In this overview article, we summarize NAPKON’s design, relevant milestones including first study population characteristics, and outline the potential of NAPKON for German and international research activities.

Trial registrationhttps://clinicaltrials.gov/ct2/show/NCT04768998.https://clinicaltrials.gov/ct2/show/NCT04747366.https://clinicaltrials.gov/ct2/show/NCT04679584

Tracing COVID-19 Infection Chains Within Healthcare Institutions - Another Brick in the Wall Against SARS-CoV-2

Early anticipation of COVID-19 infection chains within hospitals is of high importance for initiating suitable measures at the right time. Infection control specialists can be supported by application systems able of consolidating and analyzing heterogeneous, up-to-now non-standardized and distributed data needed for tracking COVID-19 infections and infected patients' hospital contacts. We developed a system, Co-Surv-SmICS, assisting in infection chain detection, in an open and standards-based way to ensure reusability of the system across institutions. Data is modelled in alignment to various national modelling initiatives and consensus data definitions, queried in a standardized way by the use of OpenEHR as information modelling standard and its associated model-based query language, analyzed and interactively visualized in the application. A first version has been published and will be enhanced with further features and evaluated in detail with regard to its potentials to support specialists during their work against SARS-CoV-2.

Towards a Generic Description Schema for Clinical Decision Support Systems

Many clinical decision support systems (CDSS) have shown good performance in the research context, but only a few have been brought into routine care. Analysis of success or failure factors in the design of CDSS may support translation from development to routine care by guiding CDSS design and development along these factors. In this work, we propose a schema to describe CDSS designs in a consistent way. We focus on design criteria with the aim to investigate the observed translation gap in CDSS. Existing description models on different aspects relevant for CDSS are combined to a comprehensive schema that allows description and comparison of CDSS without limitation to the domain or architecture.

Using Explainable Artificial Intelligence Models (ML) to Predict Suspected Diagnoses as Clinical Decision Support

The complexity of emergency cases and the number of emergency patients have increased dramatically. Due to a reduced or even missing specialist medical staff in the emergency departments (EDs), medical knowledge is often used without professional supervision for the diagnosis. The result is a failure in diagnosis and treatment, even death in the worst case. Secondary: high expenditure of time and high costs. Using accurate patient data from the German national registry of the medical emergency departments (AKTIN-registry, Home - Notaufnahmeregister (aktin.org)), the most 20 frequent diagnoses were selected for creating explainable artificial intelligence (XAI) models as part of the ENSURE project (ENSURE (umg.eu)). 137.152 samples and 51 features (vital signs and symptoms) were analyzed. The XAI models achieved a mean area under the curve (AUC) one-vs-rest of 0.98 for logistic regression (LR) and 0.99 for the random forest (RF), and predictive accuracies of 0.927 (LR) and 0.99 (RF). Based on its grade of explainability and performance, the best model will be incorporated into a portable CDSS to improve diagnoses and outcomes of ED treatment and reduce cost. The CDSS will be tested in a clinical pilot study at EDs of selected hospitals in Germany.

The COVID-19 Data Exchange Platform of the German University Medicine

COVID-19 has challenged the healthcare systems worldwide. To quickly identify successful diagnostic and therapeutic approaches large data sharing approaches are inevitable. Though organizational clinical data are abundant, many of them are available only in isolated silos and largely inaccessible to external researchers. To overcome and tackle this challenge the university medicine network (comprising all 36 German university hospitals) has been founded in April 2020 to coordinate COVID-19 action plans, diagnostic and therapeutic strategies and collaborative research activities. 13 projects were initiated from which the CODEX project, aiming at the development of a Germany-wide Covid-19 Data Exchange Platform, is presented in this publication. We illustrate the conceptual design, the stepwise development and deployment, first results and the current status.

Development of a Clinical Decision Support System for Smart Algorithms in Emergency Medicine

The development of clinical decision support systems (CDSS) is complex and requires user-centered planning of assistive interventions. Especially in the setting of emergency care requiring time-critical decisions and interventions, it is important to adapt a CDSS to the needs of the user in terms of acceptance, usability and utility. In the so-called ENSURE project, a user-centered approach was applied to develop the CDSS intervention. In the context of this paper, we present a path to the first mockup development for a CDSS interface by addressing Campbell's Five Rights within the CDSS workflow.

Therapy Switches in Fingolimod-Treated Patients with Multiple Sclerosis: Long-Term Experience from the German MS Registry

INTRODUCTIONS

Therapy switches in patients with multiple sclerosis (MS) receiving treatment with fingolimod occur frequently in clinical practice but are not well represented in real-world data. The aim of this study was to identify and characterize treatment switches and reveal sociodemographic/clinical changes over time in fingolimod-treated people with MS (PwMS).

METHODS

Data on 2536 fingolimod-treated PwMS extracted from the German MS Registry during different time periods were analyzed (2010-2019).

RESULTS

Overall, 28.3% of PwMS were treatment-naïve before fingolimod initiation. Interferon beta (30.7%) was the most common pre-fingolimod treatment. Ocrelizumab (19.8%) was the most frequent subsequent treatment in the 944 patients on fingolimod who switched. Between 2010 and 2019, median disease duration at fingolimod initiation decreased from 8.5 to 7.1 years (p < 0.001), and patients taking fingolimod for ≥ 1 year after treatment initiation decreased from 89.6 to 80.5% (p < 0.001). Females (p < 0.001) and young patients (p = 0.003) showed a shorter time on fingolimod. The most frequent reason for switching was disease activity (relapse/MRI) despite treatment. The annualized relapse rate increased from 0.37 in patients on fingolimod to 0.47 after treatment cessation, decreasing to 0.19 after treatment with a subsequent disease-modifying drug (DMD) was initiated.

CONCLUSION

Treatment switches from fingolimod to subsequent DMDs currently occur after shorter treatment durations than 10 years ago, possibly due to the growing treatment spectrum. Planning adequate washout periods is essential and should be done on an individualized basis.

Artefacts in continuous overnight blood pressure assessment based on pulse transit time

Continuous non-invasive blood pressure measurements bear a high potential. Particular in Somnology they allow to derive comfortably the systolic and diastolic blood pressure from an electrocardiogram and a synchronous photoplethysmogram without sleep disruption. In this short article some possible problems of this method are discussed along overnight recordings with a SOMNOtouch NIBP device.

Relevant Aspects for Sustainable Open Source Pandemic Apps and Platform Deployment with Focus on Community Building

The global COVID-19 pandemic revealed the necessity for mobile and web-based solutions for a variety of medical processes, e.g., individual risk calculation, communication of health information and contact tracing. Many such solutions are provided in form of open source software. However, there are major obstacles to the sustainable long-term continuation of such projects. As the topic of sustainability strategies is complex, a classification would be useful to help new projects to identify relevant sustainability factors. Based on a literature review a classification for long-term success of open source software was created. This paper presents a classification focusing on five unique categories: (1) structural decision, (2) revenue generation, (3) user focus, (4) openness and (5) community building. It was developed within the NUM-COMPASS project, focusing content-wise on pandemic apps and structure-wise on open-source provision. We provide some insights into the community building dimension by discussing factors that go into building sustainable communities.

Corona Health-A Study- and Sensor-Based Mobile App Platform Exploring Aspects of the COVID-19 Pandemic

Physical and mental well-being during the COVID-19 pandemic is typically assessed via surveys, which might make it difficult to conduct longitudinal studies and might lead to data suffering from recall bias. Ecological momentary assessment (EMA) driven smartphone apps can help alleviate such issues, allowing for in situ recordings. Implementing such an app is not trivial, necessitates strict regulatory and legal requirements, and requires short development cycles to appropriately react to abrupt changes in the pandemic. Based on an existing app framework, we developed Corona Health, an app that serves as a platform for deploying questionnaire-based studies in combination with recordings of mobile sensors. In this paper, we present the technical details of Corona Health and provide first insights into the collected data. Through collaborative efforts from experts from public health, medicine, psychology, and computer science, we released Corona Health publicly on Google Play and the Apple App Store (in July 2020) in eight languages and attracted 7290 installations so far. Currently, five studies related to physical and mental well-being are deployed and 17,241 questionnaires have been filled out. Corona Health proves to be a viable tool for conducting research related to the COVID-19 pandemic and can serve as a blueprint for future EMA-based studies. The data we collected will substantially improve our knowledge on mental and physical health states, traits and trajectories as well as its risk and protective factors over the course of the COVID-19 pandemic and its diverse prevention measures.

Motivating Developers to Use Interoperable Standards for Data in Pandemic Health Apps

The COVID-19 pandemic has brought along a massive increase in app development. However, most of these apps are not using interoperable data. The COMPASS project of the German COVID-19 Research Network of University Medicine ("Netzwerk Universitätsmedizin (NUM)") tackles this issue, by offering open-source technology, best practice catalogues, and suggestions for designing interoperable pandemic health applications (https://www.netzwerk-universitaetsmedizin.de/projekte/compass). Therefore, COMPASS conceived a framework that includes automated conformity checks as well as reference implementations for more efficient and pandemic-tailored app developments. It further aims to motivate and support developers to use interoperable standards.

Multiple Sclerosis Data Alliance - A global multi-stakeholder collaboration to scale-up real world data research

The Multiple Sclerosis Data Alliance (MSDA), a global multi-stakeholder collaboration, is working to accelerate research insights for innovative care and treatment for people with multiple sclerosis (MS) through better use of real-world data (RWD). Despite the increasing reliance on RWD, challenges and limitations complicate the generation, collection, and use of these data. MSDA aims to tackle sociological and technical challenges arising with scaling up RWD, specifically focused on MS data. MSDA envisions a patient-centred data ecosystem in which all stakeholders contribute and use big data to co-create the innovations needed to advance timely treatment and care of people with MS.

Digital teaching as an instrument for cross-location teaching networks in medical informatics: opportunities and challenges

The increasingly digitized healthcare system requires new skills from all those involved. In order to impart these competencies, appropriate courses must be developed at educational institutions. In view of the rapid development of new aspects of digitization, this presents a challenge; suitable teaching formats must be developed successively. The establishment of cross-location teaching networks is one way to better meet training needs and to make the necessary spectrum of educational content available. As part of the Medical Informatics Initiative, the HiGHmed consortium is establishing such a teaching network, in the field of medical informatics, which covers many topics related to the digitization of the health care system. Various problem areas in the German education system were identified that hinder the development of the teaching network. These problem areas were prioritized firstly according to the urgency of the solution from the point of view of the HiGHmed consortium and secondly according to existing competencies in the participating societies. A workshop on the four most relevant topics was organized with experts from the German Society for Medical Informatics, Biometry and Epidemiology (GMDS), the Society for Medical Education (GMA) and the HiGHmed consortium. These are: recognition of exam results from teaching modules that are offered digitally and across locations, and their integration into existing curricula; recognition of digital, cross-location teaching in the teachers' teaching load; nationwide uniform competencies for teachers, in order to be able to conduct digital teaching effectively and with comparable quality; technical infrastructure to efficiently and securely communicate and manage the recognition of exam results between educational institutions. For all subject areas, existing preliminary work was identified on the basis of working questions, and short- and long-term needs for action were formulated. Finally, a need for the redesign of a technologically supported syntactic and semantic interoperability of learning performance recording was identified.

Network physiology in insomnia patients: Assessment of relevant changes in network topology with interpretable machine learning models

Network physiology describes the human body as a complex network of interacting organ systems. It has been applied successfully to determine topological changes in different sleep stages. However, the number of network links can quickly grow above the number of parameters that are typically analyzed with standard statistical methods. Artificial Neural Networks (ANNs) are a promising approach as they are successful in large parameter spaces, such as in digital imaging. On the other hand, ANN models do not provide an intrinsic approach to interpret their predictions, and they typically require large training data sets. Both aspects are critical in biomedical research. Medical decisions need to be explainable, and large data sets of quality assured patient and control data are rare. In this paper, different models for the classification of insomnia-a common sleep disorder-have been trained with 59 patients and age and gender matched controls, based on their physiological networks. Feature relevance evaluation is employed for all methods. For ANNs, the extrinsic interpretation method DeepLift is applied. The results are not identical across methods, but certain network links have been rated as relevant by all or most of the models. While ANNs show less classification accuracy (0.89) than advanced tree-based models (0.92 and 0.93), DeepLift provides an in-depth ANN interpretation with feature relevance scores for individual data samples. The analysis revealed modifications in the pulmonar, ocular, and cerebral subnetworks that have not been described before but are consistent with known findings on the physiological impact of insomnia.

A Performant Web-Based Visualization, Assessment, and Collaboration Tool for Multidimensional Biosignals

Biosignal-based research is often multidisciplinary and benefits greatly from multi-site collaboration. This requires appropriate tooling that supports collaboration, is easy to use, and is accessible. However, current software tools do not provide the necessary functionality, usability, and ubiquitous availability. The latter is particularly crucial in environments, such as hospitals, which often restrict users' permissions to install software. This paper introduces a new web-based application for interactive biosignal visualization and assessment. A focus has been placed on performance to allow for handling files of any size. The proposed solution can load local and remote files. It parses data locally on the client, and harmonizes channel labels. The data can then be scored, annotated, pseudonymized and uploaded to a clinical data management system for further analysis. The data and all actions can be interactively shared with a second party. This lowers the barrier to quickly visually examine data, collaborate and make informed decisions.

Feature relevance in physiological networks for classification of obstructive sleep apnea

OBJECTIVE

Physiological networks (PN) model couplings between organs in a high-dimensional parameter space. Machine learning methods, in particular artifical neural networks (ANNs), are powerful on high-dimensional classification tasks. However, lack of interpretability of the resulting models has been a drawback in research. We assess relevant PN topology changes in obstructive sleep apnea (OSA) by novel ANN interpretation techniques.

APPROACH

ANNs are trained to classify OSA based on the PNs of 48 patients and 48 age and gender matched healthy controls. The PNs consisting of 2812 links are derived from overnight biosignal recordings. The interpretation technique DeepLift is applied to the resulting ANN models, enabling the determination of the relevant features for classification decisions on individual subjects. The mean relevance scores of the features are compared to other machine learning methods (decision tree and random forests) and statistical tests on group differences.

MAIN RESULTS

The ANN interpretation results show good agreement with the compared methods and 87% of the samples could be correctly classified. OSA patients show a significantly higher coupling (p [Formula: see text] 0.001) in light sleep (N2) between breathing rate and EEG [Formula: see text] power in all electrode locations and to chin and leg muscular tone. In deep sleep (N3), OSA leads to significantly lower coupling (p [Formula: see text] 0.01) in lateral connections of EEG [Formula: see text] and [Formula: see text] power in central and frontal positions. Misclassified OSA patients had all mild/moderate AHIs and did not show PN topology changes. Both nights of these patients have been consistently misclassified as healthy. This may indicate, that the impact of respiratory events differs in subjects, thus forming different phenotypes.

SIGNIFICANCE

The proposed PN analysis provides a powerful and robust method to quantify a broad range of physiological interactions. Interpretability of the ANN make them a promising tool to identify new diagnostic markers in data-driven approaches.