Data-driven Development of Digital Health Applications on the Example of Dementia Screening

Abstract

Following the paradigm of precision medicine, the combination of health data and Machine Learning (ML) is promising to improve the quality of healthcare services e.g. by making diagnoses and therapeutic interventions as early and precise as possible. The implementation of this approach requires sufficient amounts of data with a high quality along the data life cycle. This goal seems recently achievable through the implementation of several national digital health strategies and the hope of a growing societal acceptance of digital health applications due to the implications of the COVID-19 pandemic. But, a collection of tools and methods is missing, which supports developers to use data as driving force of the development process. Due to the iterative nature of software application development, it allows the continuous improvement through the integration of collected digital data. We refer to this as a data-driven approach and identify steps to take and tools for its implementation. Associated challenges and opportunities of this translational approach are outlined on the example of a self-developed dementia screening application. Using our methodology, we compared multiple ML algorithms based on the data of an observational study (n=55) and achieved models with sensitivity up to 89% for unhealthy participants within this use case.