Page 121 - AIH-2-1

Artificial Intelligence in Health Does improving diagnostic accuracy increase AI adoption?

and prediction of disease risks with greater accuracy and such as patients, exhibit a degree of AI adoption hesitancy,
timeliness. This combination of clinical and behavioral particularly in its utilization in diagnostics.
data could enable more personalized diagnostic models. In contrast to previous studies, which primarily
However, despite the substantial potential AI offers rely on surveys, this study aims to broaden the existing
in transforming health care, there remains significant literature on AI adoption hesitancy by testing AI adoption
hesitation toward its widespread adoption, particularly in through randomized scenario-based experiments. This
AI-assisted diagnostics. Much of this reluctance stems from approach allows for a more nuanced understanding of how
concerns about patient privacy and the risks associated with individuals respond to AI in varied controlled contexts.
data surveillance. Health-care professionals and patients
3
alike worry that the use of AI in clinical settings could 2. Methods
lead to breaches of sensitive information, unauthorized This study evaluated the public acceptability of AI-based
data access, and misuse of personal health data, which are diagnostic tools and the accuracy trade-offs required
the main factors undermining users’ trust in AI-driven to integrate EHRs and digital data in the domain of
systems. This aversion to AI, fueled by privacy concerns, neurodegenerative diseases (NDs). A survey was conducted
continues to be a major obstacle to its full acceptance in on a representative sample of the French adult population
the health-care field. (n = 1017) using a quota non-probability sampling method
By examining the resistance to AI, previous research (quotas were on age, gender, socio-professional status, and
found that generative chatbot AI faces a hesitant adoption. living area). This collection of data was part of the larger
4
A review of 7912 articles aimed at identifying predictors Discrete Choice Experiment 9-11 aiming at unveiling the
of AI adoption revealed that perceived usefulness, trade-offs surrounding the decision-making by individuals
performance expectancy, trust, and effort were key factors about neurodegenerative testing. Before agreeing with
influencing the willingness to use AI in health care. The study participation, all subjects were given comprehensive
4
review also emphasized that no amount of AI could fully information regarding the study’s purpose, procedures,
replace the value of human interaction or ensure cultural potential risks, and benefits. The study protocol was
sensitivity. In another study related to AI use in health reviewed and approved by the Ethics Committee of Aix-
care, this reluctance was shown to be more pronounced Marseille University (approval number: 2022-10-20-009).
5
among individuals with limited proficiency in Internet or Written consent was obtained from each of the subjects to
computer technologies. A noted source of concern stems participate in this study.
from the uncertainty surrounding the data sources that The 1017 participants were exposed to a set of
power these AI models, leading to skepticism about the alternative scenarios of testing methods to predict the
reliability and accuracy of the health information they hypothetical 10-year risk of developing an ND that affects
generate. In addition, users express unease over the lack of an average of 7% of the population after the age of 65.
12
transparency in how these models operate and the inherent Through the pool institut ViaVoice, participants were
complexity of AI systems. These factors contribute to confidentially randomized to scenarios depicting various
fears of miscommunication, misinterpretation of health levels of AI-based diagnostic integration and non-AI
symptoms, and the potential for inaccurate diagnoses. In traditional laboratory saliva test. The researchers were
another related survey, trust in AI adoption was found to blinded to participants’ identities. The three scenarios
be closely linked to regulatory oversight, with performance of tests included: (1) non-AI diagnostics using a
and communication also playing critical roles in users’ laboratory test with a salivary sample, (2) AI diagnostics
willingness to embrace AI applications in health care. 6 incorporating EHRs, defined as “AI,” and (3) AI diagnostics
A survey conducted in Sweden showed that only 20% incorporating EHR and digital consumer data from mobile
of health-care professionals used AI-based systems in their devices, thereafter, defined as “AI+.” To assess the impact
work, with “trust” emerging as the most critical factor in of diagnostic accuracy on participants’ preferences, the
their willingness to adopt these technologies. A review of attributes of sensitivity (true positive rate) at 60%, 70%,
7
42 studies examining health-care professionals’ acceptance or 95%, and 1-specificity (false positive rate) at 5%, 30%,
of AI revealed widespread concerns, particularly regarding or 40% were also varied. An example of the randomized
AI’s potential for errors, sensitivity, and timely access. In scenario is shown in Figure 1.
addition, the perceived loss of professional autonomy and
challenges in integrating AI into existing clinical workflows 3. Statistical analysis
were consistently identified as a significant barrier to Of the 5085 scenarios randomly proposed, we selected the
adoption. These findings highlight that healthcare workers, pairs (3225) that display a comparison between AI (or AI+)
8

Volume 2 Issue 1 (2025) 115 doi: 10.36922/aih.3561