From data to the decision: A software architecture to integrate predictive modelling in clinical settings

Area of Science:

• * Computational health informatics
• * Applied statistics in medicine

Background:

• * Healthcare systems utilize statistical and mathematical tools for disease screening and management.
• * Current data-driven models face technical and clinical limitations due to dataset specificities.
• * Existing tools often exhibit data-dependency, impacting their real-world applicability.

Purpose of the Study:

• * To propose a novel architecture for a validation framework for discrimination and prediction models.
• * To enhance the reliability of Type 2 diabetes screening tools.
• * To overcome the inherent data-dependency limitations of current predictive models.

Main Methods:

• * Development of a centralized architecture controlled by an 'Orchestrator' component.
• * Integration of diverse data sources into a unified data structure.
• * Design of controlled interaction flows between data sources, models, and user interfaces.

Main Results:

• * The proposed architecture facilitates a robust validation framework for predictive models.
• * It enables the management of diverse data sources under a common structure.
• * The Orchestrator component centralizes control over data-model-interface interactions.

Conclusions:

• * The novel architecture offers a solution to the data-dependency of predictive models in healthcare.
• * It provides a framework for validating models within actual clinical settings.
• * This approach enhances the trustworthiness and utility of disease screening tools, particularly for Type 2 diabetes.