Trust & Policy in Predictive Modeling
This project explored trust and policy implications for an algorithm based predictive modeling system in healthcare.
In-depth research for implementing the Knowledge Grid (KGrid), a platform designed to bring predictive modeling into healthcare. Through interviews and a structured governance proposal, we outlined security measures and trust-building processes to support successful adoption in clinical settings.
This project was the result of a partnership between the Michigan Medicine Department of Learning Health Sciences (DLHS) and the University of Michigan School of Public Health (SPH) Innovation Studio.
Overview
In an effort to bridge the gap between medical research and clinical practice, we examined the potential of the Knowledge Grid (KGrid), an online platform designed to house predictive models that support healthcare decision-making. Our objective was to explore the trust factors and policy requirements necessary to support KGrid’s implementation, ensuring it meets the needs of healthcare providers, administrators, and researchers.
Objectives
Our primary goal was to understand the trust and policy considerations for integrating predictive models through KGrid in healthcare settings. By understanding the unique concerns of diverse healthcare stakeholders, we aimed to uncover key elements that would foster confidence in the platform and ensure that its implementation aligns with clinical and ethical standards.
Methodology
We conducted 52 in-depth interviews with stakeholders from various healthcare roles, including clinical providers, predictive model developers, administrators, IT professionals, and ethicists. These discussions explored their workflows, experiences with predictive models, and perspectives on clinical decision-making tools.
During the interview process, we noted that many participants struggled to clearly define their trust criteria for predictive models. To support these conversations, we developed a Model Persona—a generative research tool that helps stakeholders clarify their expectations and trust factors when evaluating predictive models.
Our research highlighted several insights:
-
Clear Trust Protocols for Clinical Providers: Clinical providers have consistent protocols for assessing predictive models, providing a foundation for building trust in KGrid.
-
Model Journey Map: We created a detailed journey map illustrating each step a healthcare institution would need to take to create, validate, and deploy a predictive model. The map, developed using insights from our interviews, outlines each stage’s required actions and roles involved.
-
Governance Expectations: Stakeholders emphasized the importance of governance for KGrid. Early oversight from the KGrid team was recommended, transitioning to a formal multidisciplinary governance structure as the platform grows.
Enhancements
Based on our findings, we developed several key recommendations to support the successful implementation and adoption of KGrid:
-
Platform Security Mechanisms - Establishing clear security measures to protect sensitive healthcare data and ensure reliability in clinical settings.
-
Proposed Governance Model - A phased governance structure, beginning with the KGrid team leading initial oversight and evolving into a multidisciplinary governing body as KGrid scales.
-
Model Journey Mapping - A journey map to guide institutions in the process of creating, validating, and deploying predictive models within KGrid. Each stage of the journey includes specific actions, responsible roles, and verification steps to build trust.
Results
The research findings provide a comprehensive roadmap for implementing KGrid in a way that respects the critical trust and policy needs of healthcare stakeholders. By addressing their concerns and establishing a clear governance model, KGrid can be positioned as a trusted, secure, and impactful tool that accelerates the transition of medical research into everyday healthcare practice.