A Qualitative Exploratory Study of Emergency Medicine Clinician Perspectives on Clinical Decision Support Systems (CDSS) Rooted in Machine Learning in England
Hayes, Tyler F.
MetadataShow full item record
CitationHayes, Tyler F. 2020. A Qualitative Exploratory Study of Emergency Medicine Clinician Perspectives on Clinical Decision Support Systems (CDSS) Rooted in Machine Learning in England. Doctoral dissertation, Harvard Medical School.
AbstractPurpose: To map the landscape of clinical decision support systems (CDSS) rooted in machine learning techniques in emergency medical care through a scoping literature review, and to explore the perspectives, experiences and attitudes of emergency medicine clinicians in England related to CDSS, particularly those rooted in machine learning.
Methods: A comprehensive scoping literature review was performed in order to capture and organize the diversity, and extent of technological validation and integration of point-of-care machine learning-based clinical decision support systems (CDSS) in emergency departments worldwide. Subsequently, semi-structured interviews with emergency medicine clinicians in England familiar with CDSS were conducted. Resulting qualitative data was analyzed to synthesize and critically compare experiences, attitudes and perspectives towards these tools.
Results: 107 publications were identified, with 50% (n = 51) coming from the US and eight from England. 3% (n = 3) were developed on training dataset without subsequent validation. 82% (n = 88) were internally validated, of which 14% (n = 12) were external validated. 3% (n = 3) were prospectively implemented in clinical practice in EDs. The most common applications were diagnosis, followed by outcome prediction, mortality, disposition planning and triage. While each of thirteen clinician interviewees was familiar with artificial intelligence and most respondents (n = 9) were familiar with machine learning and potential applications to emergency medical care, none had used a CDSS rooted in machine learning in their own practice. Themes arising from the interviews were summarized into four different factor families perceived to affect acceptance of CDSS (‘facilitators’ and ‘barriers’) – provider, practice, tool and institutional factors.
Conclusions: There exist few published machine learning-based CDSS for use at point-of-care in emergency medicine in England relative to the United States. Correspondingly, emergency medicine providers in England are not familiar with any such tools. However, they do appreciate the potential for these digital technologies to improve the provision of emergency medical care in England with advantages not only for patients but also for providers. Further work should examine the barriers to machine learning-based CDSS development and validation in England and the aspects of emergency medical care that would be most supported using these technologies. Toward this end, the value of clinicians’ experiences and perspectives towards computerized support tools at the intersection of artificial intelligence should not be underestimated.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37364934