The team aims to implement an existing web-based IT platform (Evidencio), initiated in 2015 by physicians aiming to facilitate the translation of scientific literature towards clinical practice. Together with a group of software engineers, they developed a platform that allowed researchers to create and share user-friendly risk prediction tools online in just a few steps. Over the years, the number of prediction algorithms hosted on the platform has grown substantially, with contributions being made by researchers and healthcare professionals from all over the world.
Integration of the web-based platform into the electronic health record (EHR) system (EPIC) could give instant access to hundreds of prediction algorithms. “Algorithms hosted on the platform are created in a standardized manner, increasing transparency, efficiency and scalability. In addition, the algorithms can be validated semi-automatically using anonymized patient data, enabling rapid assessment of algorithm performance prior to their actual use in clinical practice. The algorithms can than be safely applied to promote patient-tailored healthcare and guide clinical decision-making”, Rick Pleijhuis, resident Allergology and Clinical Immunology at the department of Internal Medicine, explains.
“Many examples exist in which multiple prediction algorithms are available that predict more or less the same outcome. For physicians and their patients, it can be confusing when such algorithms result in different predicted probabilities while being fed the same input variables. In addition, objective information regarding which algorithm performs best given certain circumstances is often lacking. The web-based platform is capable of identifying the best performing prediction algorithms by comparing them head-to-head using local anonymized patient data. The algorithms that performs best could than be integrated in the EHR system to guide clinical decision-making”, Pleijhuis explains.
The growing interest in risk stratification using prediction algorithms has resulted in the publication of tens of thousands of prediction algorithms in medical literature. Yet, only a few are applied in clinical practice on a regular basis. The same accounts for machine learning algorithms, which hold great promise to revolutionize healthcare in the years to come. To ensure successful clinical implementation and true adoption of prediction algorithms by healthcare professionals, however, advances are needed on algorithm validation, user interface, and EHR integration. The project team aims to overcome these challenges through implementation of the aforementioned platform.
“Implementation of the web-based prediction platform itself poses new challenges that exceed the scope of clinical care and scientific research, including ethical, legal and technological issues”, Pleijhuis says. “For example, prediction algorithms are considered medical devices, which means that they require CE-certification. The team also needs support with the extraction of anonymized data for validation purposes, optimization of the data infrastructure, and integration in the EHR system.”
“DASH already has played an important role in connecting our team with experts in the field and identifying key people in the organization to help the project move forward”, says Pleijhuis. “In a previous project, DASH assisted the team in writing a proposal for a big EU call, specifically on the part regarding the applicable laws and regulations for data management.” The current project will receive dedicated support from DASH to speed up the integration process that should allow prediction algorithms to be accessed directly from the EHR system. “So far, we had several useful discussions with the DASH team to clarify our needs regarding relevant legal, technical and data-related challenges”, Pleijhuis says. “If this project is successful, healthcare professionals in the UMCG could have hundreds of pr ediction algorithms at their disposal in the workflow of the EHR system.”
Existing IT solutions used in this project are available for researchers within the UMCG as well.
Anonymization service: Processing data with the aim of preventing the identification of the person to whom it relates.
Virtual Research Workspace: An environment for secure collaboration amongst different researchers (working within and outside the UMCG) for analyzing data simultaneously and/or sharing privacy-sensitive data.
