Clinical decision support needs to integrate knowledge that can follow from both physiologic/mechanistic as well as purely data-driven approaches. It needs to incorporate a wide range of information from different scales, from cell-level to population level, and it needs to use multi-modal data (imaging, lab values, -omics, signals and electronic health records) as well as accommodate a multi-systemic viewpoint. Finally, it needs to provide actionable insights.
As boundary condition for healthcare we need to optimize resource use, calling for a stratified stepwise approach, where increasingly accurate (and complex) assessments are provided for those patients who really need it. Additionally, it needs to be co-created with end-users (healthcare professionals and patients) as well as developers of models and medical device companies.
Hypertrophic cardiomyopathy (HCM) is a multifactorial disease with a wide range of manifestations. Management of HCM requires decision support for individual stratification, risk assessment and treatment planning addressing the abovementioned requirements.
In SMASH-HCM we have developed a decision support solution addresses these requirements. After iterative user requirement elicitations, an Application Programming Interface (API) was developed that provides links between the models and algorithms developed for in-vitro, in-vivo, and cellular to population-level models. The software solution orchestrates the combinations of developed models and algorithms that are needed to address the questions posed by the end-user (e.g., risk assessment, intervention planning, patient state follow-up).
From a software architecture view, it provides a reproducible and scalable digital-twin based platform unifying multimodal data with mechanistic and data-driven models and ensuing virtual human twin. It uses a separation of concerns approach with data processing, analysis, and visualization layers, supporting advanced risk stratification and patient-specific disease progression modelling. Additionally, it supports seamless combinations of models developed by researchers in different programming languages.
The next steps include deployment, formal validation, and preparation for post-project availability and exploitation.