Presentation
HE7 - I, Doctor: Patient Preference for Medical Diagnostic Artificial Intelligence
DescriptionAs the use of automation and artificial intelligence grows, it is important to understand the benefits these tools can bring to healthcare. With the current strain on the healthcare industry, implementing automation and AI may be able to bring some relief to the overworked employees and patients by optimizing diagnosis timelines and streamlining patient appointments for routine procedures. The faster a diagnosis is received, the faster treatment can be implemented, which results in better overall outcomes. However, patients and providers may feel wary of the use of these technologies, as their newness may produce uncertainty and distrust, particularly for some segments of the population in need of healthcare treatment. In order for automation and AI to be a productive tool in the medical field, it is pertinent for all parties involved to be able to work alongside and with the technology. This type of cooperation requires an understanding of the benefits this type of care can provide, such as increased diagnostic accuracy and efficiency, as well as decreased invasiveness compared to traditional approaches. Previous literature has examined attitudes towards automated healthcare, but primarily focused on demographic trends (Stai et al., 2020). While demographic trends may be informative when attempting to enhance the relationship between automated healthcare and specific target populations, it fails to acknowledge the specific factors that the automation can manipulate in order to increase trust. In this study, four variables that could impact an individual’s willingness to choose an AI or automated care application were investigated: the accuracy, efficiency, and invasiveness of the technology compared to the traditional method, and the risk level of the overall health scenario.
In order to investigate these factors, an online survey was implemented amongst 60 psychology undergraduate students at Colorado State University, with the goal of examining levels and variables that may impact preference for diagnostic artificial intelligence and automation. The survey was broken down into four separate hypothetical medical scenarios, in which different versions of both a traditional medical intervention and an AI or automated intervention were presented. The scenarios included a possible skin cancer diagnosis, a routine test for tuberculosis, a heart disease screening, and a routine physical examination. Each scenario was followed by two to three blocks of questions, with each block measuring a different factor that may influence the participant’s likelihood to choose the AI diagnostic: the procedure’s Accuracy, Efficiency, or Invasiveness.
Risk was measured between scenarios, as two scenarios were deemed “high risk” (cancer diagnosis and heart disease screening with symptoms) while the other two were deemed “low risk” (routine tuberculosis test and physical). Accuracy was measured across all scenarios, as well as between variables, as each question indicated a varying level of accuracy of the traditional method (70%, 75%, 80%, or 90%) compared to the automation’s 70%, or indicated a varying level of the automation’s accuracy (70%, 75%, 80%, or 90%) compared to the traditional method’s 70%. The participant was asked to respond to a likert scale question on how likely they would be to choose the automated or traditional intervention (0=extremely unlikely, 5=extremely likely). Invasiveness and efficiency were measured by adding a modifying statement presented after the scenario which suggested that the AI method would be more efficient or less invasive than the traditional method before the above questions were asked.
After analysis, it was found that the variables of accuracy, efficiency, and risk all impacted the participant’s preference for AI diagnostics over the traditional approach. Participants preferred the AI diagnostic over the traditional human diagnostic method when it was said to produce more accurate results at each level of accuracy (70%, 75%, 80%, 90%) compared to the traditional method’s 70%. In particular, the baseline preference, when both methods were 70%, indicated a significant preference for the AI method, which may be partially due to the younger demographics of the participant pool. Participants also significantly chose the AI diagnostic when it was more efficient than the traditional method between and across all scenarios, indicating the importance that individuals place on interventions that may save them time.
In regards to invasiveness, preference for diagnostic AI failed to be overall significantly affected when the automation took a less invasive approach than the traditional method, aside from the Tuberculosis scenario. While less invasiveness is an actual benefit of automation, as it can lead to lower infection rates, this difference does not appear to greatly impact the patient’s perception of the automation. However, it was found that the risk of the health scenario did significantly affect an individual’s preference for automation (Table 4). These findings suggest that high-risk scenarios, such as receiving a cancer or heart disease diagnosis, decrease an individual’s preference for AI or automated diagnostics compared to low-risk scenarios, such as a routine TB test or physical exam. Understanding the components that influence a patient’s perceptions of medical automation and AI is essential for the successful implementation of these tools, which have the capacity to positively impact the field of healthcare.
The intent of this study was to further the understanding of factors that may lead to increased compliance and preference for automated healthcare. However, the current study only depicted four types of medical diagnostic AI, when there are hundreds of similar automated diagnostic tools in existence. Again, the current study found a significant preference for automated healthcare when both intervention methods were at the same accuracy level of 70%. This preference is likely due to the younger population of the study, as the average age was 19.8 years. Future studies should target a wider age range of participants, specifically older populations that may have more hesitations around technology as a whole. These factors are essential to identify, as presenting them to patients may be enough to sway their opinions of these new technologies.
References
Stai, B., Heller, N., McSweeney, S., Rickman, J., Blake, P., Vasdev, R., Edgerton, Z., Tejpaul, R., Peterson, M., Rosenberg, J., Kalapara, A., Regmi, S., Papanikolopoulos, N., & Weight, C. (2020). Public perceptions of artificial intelligence and robotics in medicine. Journal of Endourology, 34(10), 1041–1048. https://doi.org/10.1089/end.2020.0137
In order to investigate these factors, an online survey was implemented amongst 60 psychology undergraduate students at Colorado State University, with the goal of examining levels and variables that may impact preference for diagnostic artificial intelligence and automation. The survey was broken down into four separate hypothetical medical scenarios, in which different versions of both a traditional medical intervention and an AI or automated intervention were presented. The scenarios included a possible skin cancer diagnosis, a routine test for tuberculosis, a heart disease screening, and a routine physical examination. Each scenario was followed by two to three blocks of questions, with each block measuring a different factor that may influence the participant’s likelihood to choose the AI diagnostic: the procedure’s Accuracy, Efficiency, or Invasiveness.
Risk was measured between scenarios, as two scenarios were deemed “high risk” (cancer diagnosis and heart disease screening with symptoms) while the other two were deemed “low risk” (routine tuberculosis test and physical). Accuracy was measured across all scenarios, as well as between variables, as each question indicated a varying level of accuracy of the traditional method (70%, 75%, 80%, or 90%) compared to the automation’s 70%, or indicated a varying level of the automation’s accuracy (70%, 75%, 80%, or 90%) compared to the traditional method’s 70%. The participant was asked to respond to a likert scale question on how likely they would be to choose the automated or traditional intervention (0=extremely unlikely, 5=extremely likely). Invasiveness and efficiency were measured by adding a modifying statement presented after the scenario which suggested that the AI method would be more efficient or less invasive than the traditional method before the above questions were asked.
After analysis, it was found that the variables of accuracy, efficiency, and risk all impacted the participant’s preference for AI diagnostics over the traditional approach. Participants preferred the AI diagnostic over the traditional human diagnostic method when it was said to produce more accurate results at each level of accuracy (70%, 75%, 80%, 90%) compared to the traditional method’s 70%. In particular, the baseline preference, when both methods were 70%, indicated a significant preference for the AI method, which may be partially due to the younger demographics of the participant pool. Participants also significantly chose the AI diagnostic when it was more efficient than the traditional method between and across all scenarios, indicating the importance that individuals place on interventions that may save them time.
In regards to invasiveness, preference for diagnostic AI failed to be overall significantly affected when the automation took a less invasive approach than the traditional method, aside from the Tuberculosis scenario. While less invasiveness is an actual benefit of automation, as it can lead to lower infection rates, this difference does not appear to greatly impact the patient’s perception of the automation. However, it was found that the risk of the health scenario did significantly affect an individual’s preference for automation (Table 4). These findings suggest that high-risk scenarios, such as receiving a cancer or heart disease diagnosis, decrease an individual’s preference for AI or automated diagnostics compared to low-risk scenarios, such as a routine TB test or physical exam. Understanding the components that influence a patient’s perceptions of medical automation and AI is essential for the successful implementation of these tools, which have the capacity to positively impact the field of healthcare.
The intent of this study was to further the understanding of factors that may lead to increased compliance and preference for automated healthcare. However, the current study only depicted four types of medical diagnostic AI, when there are hundreds of similar automated diagnostic tools in existence. Again, the current study found a significant preference for automated healthcare when both intervention methods were at the same accuracy level of 70%. This preference is likely due to the younger population of the study, as the average age was 19.8 years. Future studies should target a wider age range of participants, specifically older populations that may have more hesitations around technology as a whole. These factors are essential to identify, as presenting them to patients may be enough to sway their opinions of these new technologies.
References
Stai, B., Heller, N., McSweeney, S., Rickman, J., Blake, P., Vasdev, R., Edgerton, Z., Tejpaul, R., Peterson, M., Rosenberg, J., Kalapara, A., Regmi, S., Papanikolopoulos, N., & Weight, C. (2020). Public perceptions of artificial intelligence and robotics in medicine. Journal of Endourology, 34(10), 1041–1048. https://doi.org/10.1089/end.2020.0137
Event Type
Poster Presentation
TimeMonday, March 254:45pm - 6:15pm CDT
LocationSalon C
Digital Health
Simulation and Education
Hospital Environments
Medical and Drug Delivery Devices
Patient Safety Research and Initiatives