What fosters patients' acceptance of Clinical Decision Support Systems? The influence of knowledge and provider type on popular willingness to use artificial medical intelligence

Download data and study materials from OSF

Principal investigators:

Christopher McLaughlin

United States Air Force

Email: cmmclaughlin@gmail.com

Sue Tolleson-Rinehart

University of North Carolina

at Chapel Hill

Email: suetr@unc.edu

Homepage: http://sph.unc.edu/adv_profile/sue-tolleson-rinehart-phd/

Anthony Viera

University of North Carolina

at Chapel Hill

Email: anthony_viera@med.unc.edu

Homepage: http://www.med.unc.edu/fammed/research/faculty/anthony-viera

Sample size: 1211

Field period: 06/01/2015-08/10/2015

1. Does knowledge of a computer decision aid's accuracy relative to a physician influence perception physicians who use that technology?
2. Do people view the care provided by a physician extender using a computer decision aid to be equivalent to care provided by a physician?

Experimental Manipulations
Vignette 1: A patient-physician encounter for an ankle injury employs a 3 (no CDSS vs. CDSS equal to average doctor vs. CDSS superior to average doctor) x 2 (outcome: accurate assessment and treatment of ankle injury vs. missed ankle fracture diagnosis) x 2 (heed vs. defy CDSS recommendation) between-subject design.
Vignette 2: A combat triage scenario in which one provider had to determine which two out of four casualties would be evacuated to a surgeon by helicopter and which two would be evacuated to a surgeon by ground. The vignette employs a 2 (provider: medic vs. physician) x 3 (CDSS deference: no decision aid used vs. CDSS present and heeded vs. CDSS present and defied) x 2 (outcome: all four casualties survive vs. one of the four casualties dies) between-subject design.

1. Seven-point Likert assessment of provider quality
2. Seven-point Likert assessment of clinical outcome acceptability
3. Four-point Likert assessment of provider responsibility for the clinical outcome.

Summary of Results
Knowledge about the accuracy of a clinical decision support system (CDSS) affects perception of medical providers who use those tools. People support the use of CDSS more when they are explicitly told that the decision aids are more accurate relative to the average physician. Clinical outcomes, regardless of whether they are positive or negative, are rated as less acceptable when physicians defy the recommendation of a highly accurate CDSS than when the recommendation is heeded. Conversely, when the lower quality CDSS is employed, the acceptability of a good or bad outcome is not affected by whether the physician heeded or defied the recommendation.
We found that people hold physicians to be more responsible for clinical outcomes than physician extenders, medics in this case. Alternatively, in assessments of provider quality and clinical outcome acceptability, there was no significant difference between physicians and medics when a decision aid was used. In the combat triage scenario, an outcome deemed negative was one in which a soldier died as a result of the provider's decision. A positive clinical outcome was one in which all of the soldiers survived as a result of the provider's decision. We found that use of a decision aid does not influence assessments of provider responsibility for positive clinical outcomes. If, though, a provider defied a CDSS recommendation and the soldier died, the provider was deemed much more responsible for that outcome than if the provider had heeded the recommendation or used no CDSS at all.
The exception to the similarity of provider quality and outcome acceptability ratings between medics and doctors was when participants were primed to believe that CDSS accuracy was superior to the accuracy of an average physician. In those cases, medics were rated more positively for heeding CDSS recommendations and more negatively for defying the recommendations than in the cases where they had previously been told that CDSS were equivalent to physicians or had no information about CDSS accuracy.