.jpg)
PRP003: A Collaborative Approach to Implementing Artificial Intelligence in Primary Care.
David Rushlow, MD, MBOE; Xiaoxi Yao, PhD, MPH, MS; Thomas Thacher, MD; Randy Foss, MD; Artika Misra, MD; Matthew Bernard, MD, FAAFP; Steven Rosas; Peter Noseworthy, MD; Ivana Croghan, PhD, MS; Jonathan Inselman, MS; Barbara Barry, PhD; Xuan Zhu, PhD
Abstract
Title: A Collaborative Approach to Implementing Artificial Intelligence in Primary Care.
Context: Artificial intelligence (AI) in medicine has been available for decades; however, adoption in primary care has been fraught with difficulty. Objective: To assess the pragmatic application of an AI-assisted ECG screening tool in detecting asymptomatic heart failure in a primary care practice, through a partnership between investigators in Cardiology and Family Medicine. Study Design: Randomized controlled trial. Setting: Primary care practices in the upper Midwest spanning multiple clinical settings from an urban academic medical center to rural community clinics. Population Studied: 358 primary care providers (MD, DO, PA, NP). Intervention: An AI-assisted tool was used to evaluate routine ECG’s determining the likelihood of underlying reduced left ventricular ejection fraction. Those randomized into the intervention group were given access to an AI-enhanced ECG report within the EHR, showing the AI-derived results after each ordered ECG (positive/negative). In addition to the AI-results, clinicians were encouraged to order a transthoracic echocardiogram (TTE) and would receive email alerts and reminders if no TTE was ordered. Those in the control group were not given these results, and usual care was expected. Surveys and focus groups were used to gauge provider adoption and understand attitudes toward AI tools in practice. Outcome Measure: Characteristics of providers who were more likely to order a follow-up echocardiogram on patients whose ECG’s screened positive for heart failure. Results: Study still in progress; however, we hypothesize that providers who were more likely to order a follow-up echocardiogram are younger, more comfortable with computer-assisted diagnostic technology, and have less complex patient panels. Conclusions: For AI algorithms to be effective in improving the health and well-being of the majority of the population, they must be a standard tool for the primary care provider. Primary care practices provide a real-world laboratory for the critical work of understanding AI implementation at the population level. Our study sheds light on the barriers and enabling provider characteristics to help improve the future application of AI technology in primary care practices.
Context: Artificial intelligence (AI) in medicine has been available for decades; however, adoption in primary care has been fraught with difficulty. Objective: To assess the pragmatic application of an AI-assisted ECG screening tool in detecting asymptomatic heart failure in a primary care practice, through a partnership between investigators in Cardiology and Family Medicine. Study Design: Randomized controlled trial. Setting: Primary care practices in the upper Midwest spanning multiple clinical settings from an urban academic medical center to rural community clinics. Population Studied: 358 primary care providers (MD, DO, PA, NP). Intervention: An AI-assisted tool was used to evaluate routine ECG’s determining the likelihood of underlying reduced left ventricular ejection fraction. Those randomized into the intervention group were given access to an AI-enhanced ECG report within the EHR, showing the AI-derived results after each ordered ECG (positive/negative). In addition to the AI-results, clinicians were encouraged to order a transthoracic echocardiogram (TTE) and would receive email alerts and reminders if no TTE was ordered. Those in the control group were not given these results, and usual care was expected. Surveys and focus groups were used to gauge provider adoption and understand attitudes toward AI tools in practice. Outcome Measure: Characteristics of providers who were more likely to order a follow-up echocardiogram on patients whose ECG’s screened positive for heart failure. Results: Study still in progress; however, we hypothesize that providers who were more likely to order a follow-up echocardiogram are younger, more comfortable with computer-assisted diagnostic technology, and have less complex patient panels. Conclusions: For AI algorithms to be effective in improving the health and well-being of the majority of the population, they must be a standard tool for the primary care provider. Primary care practices provide a real-world laboratory for the critical work of understanding AI implementation at the population level. Our study sheds light on the barriers and enabling provider characteristics to help improve the future application of AI technology in primary care practices.
Jaky Kueper
jkueper@uwo.ca 11/24/2020I am excited to see an RCT to evaluate AI in PC! I am curious how the particular AI-tool was selected? Was it developed through your research program, industry, etc?