Introduction

Purpose and Scope

This Responsible AI content for Sepsis Risk Prediction aligns to the CHAI Responsible AI Guide (RAIG) (formerly known as, Assurance Standards Guide) by creating a set of best practice guidance applicable for developers and implementers of predictive AI sepsis risk prediction solutions used in healthcare organizations. This Responsible AI content includes:

  1. Testing and Evaluation (T&E) Framework: a consensus-defined set of methods, metrics, measures, and benchmarks for developers and implementers to more concretely evaluate the responsible use of predictive AI solutions for sepsis risk prediction. The T&E Framework varies based on use case type.

Individuals involved in the process to develop, deploy, and monitor an AI-enabled sepsis risk prediction solution may leverage the content within this consensus-driven, member-driven Responsible AI content to evaluate their solution for Responsible AI guidance. Teams using this Responsible AI content should identify the AI lifecycle stage, as described in CHAI’s RAIG, that is most applicable to their organizational scenario. They should then work with all relevant teams and stakeholders to align and leverage the methods, metrics, measures, and benchmarks defined in the Testing & Evaluation Framework to evaluate and assess the responsible use of AI.

Audience

This Responsible AI content is for developers and implementers of sepsis risk prediction AI solutions used in healthcare organizations:

  • Developers: individuals involved in the software development process, including requirements gathering, design, coding, testing, and maintenance of software applications (derived from IEEE, 12207:2017)

  • Implementers: individuals responsible for the procurement, deployment, and/or overall realization of a system or component in accordance with a specified design (derived from IEEE 829 and IEEE 730)

  • Healthcare organizations: “a health care organization is a purposefully designed, structured social system developed for the delivery of health care services by specialized workforces to defined communities, populations, or markets.”

Use Case Description

Sepsis risk prediction use case: a machine learning-based early warning system designed to detect potential sepsis cases in real time, enabling timely intervention and improved patient outcomes. Integrated into hospital clinical workflows, the system generates alerts for healthcare providers, including physicians and clinical staff, to identify high-risk patients who would benefit from prompt treatment. The system’s effectiveness is assessed based on provider interactions, particularly whether alerts are confirmed within three hours, and the subsequent initiation of sepsis treatments such as antibiotic therapy. Key stakeholders include patients, healthcare providers, hospital administrators, and researchers evaluating its impact.