gms | German Medical Science

Background: Heart failure (HF) constitutes a high burden for patients, families, and healthcare systems worldwide [1]. Digital Health (dHealth) has potential to support monitoring and decision-making in patient care and self-care [2]. For artificial intelligence applications, a broad knowledge base of clinical and surrogate endpoints is required to evaluate their effectiveness. Currently, such an extensive knowledge base for HF-care enriched with associations between endpoints does not exist.

Purpose: Patient Self-care uSIng eHealth In chrONic HF (PASSION-HF) is working towards personalized and precise care of HF-patients. PASSION-HF aims to create a holistic Medical Outcome Model (MOM) as a reference model for appropriate endpoints in HF-care, and their use in dHealth applications.

Methods: A systematic review conducted in accordance with PRISMA in PubMed, EMBASE, and CINAHL identified relevant endpoints recorded in HF randomized controlled trials (RCT) between June 2010 and June 2020. Within 656 included articles, we identified 57 different interventions and 399 distinct endpoints that build the knowledge base for the MOM. To identify interventions of HF relevance, i.e., interventions that may improve mortality and hospitalization, medical guidelines and expert reviews were applied. For prioritization of endpoints, the effects per study and intervention were extracted. The MOM will be represented as an ontology developed according to METHONTOLOGY [3], implemented in OWL2 (Web Ontology Language) via Protégé.

Results: Endpoints are systematized according to measurement methods and prioritized according to a combination of p-value, effect size, and study size to derive validity and relevance of an endpoint. The MOM-metamodel developed hitherto contains four main classes: Intervention, Outcome, Endpoint, MeasuringInstrument. Object properties describe relationships: improves, measures, uses, shows_significant_improvement, shows_significant_worsening. The intervention classification includes 16 concepts, e.g., beta blocker, exercise training, cardiac resynchronization therapy (CRT). Different endpoints are used to evaluate the effects of an intervention, e.g., CRT uses creatinine and quality of life. Each endpoint can be measured by at least one measuring instrument e.g., survey, external device, or clinical test by automatic, semi-automatic or manual data input. Whether an endpoint has demonstrated effects in the past, i.e., in RCTs, is indicated by the relations shows_significant_improvement and shows_significant_worsening. Depending on the capabilities of data collection, data use, and algorithmic processing, endpoints can be used as input or feedback in dHealth applications.

Discussion: Development of the MOM is continuing. Determining validity and relevance of an endpoint poses a particular challenge. The question arises whether an endpoint has the same validity for all patients or whether this dependents on patients and/or intervention. Identified data gaps from the review increased the challenge to assess the relevance of endpoints. Thus far, the combination of p-value, effect size, and study size have been deemed a suitable approach.

Conclusion: Implementing the MOM as an ontology seems promising. Description logic queries allow specific queries to easily identify (novel) endpoints that are usable for specific dHealth applications and design of clinical studies. This allows data to be provided more precisely to both patients and health care professionals to improve monitoring, decision-making, and self-care.

Funding: PASSION-HF is funded by INTERREG-NEW VB (NEW 702).

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.