Scalable, Shareable, and Computable Clinical Knowledge for AI-Based Processing of Hospital-Based Nursing Data - Artificial intelligence (AI) is increasingly being used in healthcare. For AI to be clinically relevant in health care the data being used in models (model inputs) need to be accurate and easily explainable. Unfortunately, there are a lack of data scientists who are developing AI models that understand data documented by nurses and can explain its clinical meaning and best ways it should be used as AI model input. Our team has been working together for over 10 years and are leading experts in how to use nursing data in AI models. We developed the healthcare process modeling (HPM) framework that guides data scientists on how to best use nursing data in AI models. Our team developed and tested the CONCERN (Communicating Narrative Concerns Entered by RNs) Early Warning Score, shows a significant reduction in mortality risk among patients in the hospital for nurses and physicians who use the predictive score compared to those who do not use the score. This study, called Scalable, Shareable, and Computable Clinical Knowledge for AI-Based Processing of Hospital-Based Nursing Data (SC2K), will investigate how we can combine knowledge from nurses with computational AI model to improve the use of nursing data in AI algorithms. Specific Aims are: Aim 1.Test and validate different computational approaches for 2 AI-based use cases (1. classifying missing data versus missed care, and 2. classifying implicit biases) that use nursing data.; Aim 2. Generate graphs that illustrate important nursing knowledge that is useful for 2 AI-based use cases (1. classifying missing data versus missed care, and 2. classifying implicit biases) that leverage inpatient nursing and multi-modal data.; Aim 3. Extend computational and knowledge graph approaches for additional types of nursing data, such as from mobile devices and images, applied to 5 additional AI-based use cases.; and Aim 4. Share computational models, knowledge graphs and other resources publicly online for any person to as an Open-Source pipeline. Upon completion of the first 2 years, we will make the complete set of knowledge graphs integrated with computational models for the first 2 AI-based use cases openly available. In year 3-5, we will scale and share optimized computational models integrated with knowledge graphs for 5 new use cases and make these resources available on an open-source platform.
