JMIR Publications

ABSTRACT

Background:

The rapid adoption of artificial intelligence (AI) in healthcare emphasizes the importance of assessing healthcare professionals' attitudes towards this technology.

Objective:

The aim of this study was to develop and validate a reliable and comprehensive Artificial Intelligence Attitude Scale (AIAS) with a number of subscales for physicians.

Methods:

This qualitative study was conducted cross-sectionally between January and March 2024. For this investigation, three stages were identified. The first stage involves the development of an item pool pertaining to artificial intelligence and informatics. The next stage entails administering the generated items to a primary sample of physicians using a 5-point Likert questionnaire, followed by conducting explanatory and confirmatory factor analyses on the acquired data. The last stage consists of distributing the newly established scale to a secondary sample of physicians, assessing the scores across the scale's subcategories, and investigating their variations concerning certain independent variables.

Results:

A questionnaire comprising 47 items was distributed to clinicians following an expert evaluation. The initial sample consisted of 239 participants who completed the questionnaire. After the outliers were eliminated, the results of 198 participants were analyzed. Exploratory factor analysis (EFA) identified a three-factor structure (Belief, Knowledge, Practice) with robust factor loadings (0.64 and above) and an overall explained variance of 62.64%. The data set was considered appropriate for EFA according to the Kaiser-Mayer-Olkin (KMO) (0.91) and Bartlett's test (χ2=712.63, P<.001), and the principal axis factoring method with varimax rotation was utilized due to the lack of multivariate normality. After confirmatory factor analysis (CFA), the scale's (AIAS) final version comprised 26 items and three subscales emerged: belief (11 items), knowledge (7 items) and practice (8 items). All t values of the loadings of these subscales were significant (P<.001). The AIAS showed strong internal consistency (Cronbach's alpha >0.88 and composite reliability >0.70 for all subscales) and acceptable model fit indices in CFA. The AIAS was administered to a second new sample of 509 physicians in the concluding stage of the study, and 441 responses from this sample were considered to be valid. Findings revealed that physicians generally held positive beliefs about AI (mean belief score: 37.58 ± 9.22), but knowledge (18.71 ± 5.81) and practice (11.31 ± 3.7) scores remained lower. Male physicians (P<.001), those in academic roles (P<.001) and those working in private hospitals (P<.05) generally reported higher scores on some subscales, while higher workload was associated with lower practice scores (P<.05).

Conclusions:

The AIAS provides a robust tool to assess attitudes towards AI in healthcare. While physicians recognize the potential benefits of AI, practical adoption requires targeted training, improved infrastructure and addressing ethical challenges.