The primary aim of this study was to determine the factors influencing the intention to use an automated burn diagnosis tool among first contact clinicians. An additional aim was to assess whether the views of those clinicians are like those of burns specialists, in particular, how they perceive the likelihood and severity of certain clinical risks.

This was a questionnaire-based cross-sectional study. The CHERRIES checklist for reporting results of e-surveys was used [24], with results presented in Table S1 in Multimedia Appendix 1.

In 2019, two consecutive international congresses took place in South Africa and gathered clinicians involved in burn injury care at different levels, both first-contact care and specialized care delivery. Those were judged good opportunities to reach out to several professionals from different parts of Sub-Saharan Africa and other parts of the world that, otherwise, would have been difficult to reach. The survey took place in the context of the development of an algorithm for image-based automated burn diagnosis [23]. The meetings were the Pan African and South African Burn Congress 2019 and the EMSSA (7th Emergency Medicine Society of South Africa) International Conference. At both events, participants were informed about this study taking place when they registered, and pamphlets containing information about the survey were distributed in sessions. Using a QR code provided, those who wished to participate could go on and fill in the survey on their own device or using one of the tablets provided by the research team.

A questionnaire developed built on previous publications [20,21,25] was used to assess the intention to use an automated burn diagnosis tool among the participants based on the Automation Acceptance Model (AAM) [25]. For that purpose, the questionnaire was divided into 3 parts. Following some information on the context of automated burn diagnosis tool development, and with a specific description of what such a tool would consist of, the first section included demographic questions about the participant and its practice in burn care. The second part included 23 questions representing the 7 constructs from AAM [25], as well as anxiety [21,26,27] to reflect the intention to use an automated burn diagnosis tool in resource-poor settings in Africa (Table 1). The 7 constructs were behavioral intention (1 measurement item), attitude (3 items), perceived usefulness (8 items), perceived ease of use (2 items), trust (2 items), and compatibility (2 items) all directly inspired from the AAM and adapted to burns diagnosis and anxiety (5 items) which was shown to be predictive in similar populations for the introduction of telehealth. All constructs were measured on a 7-point Likert scale from 1 “strongly disagree” to 7 “strongly agree.” The third part of the questionnaire evaluated the participant’s perceptions of the risks linked with the use of an automated burn diagnosis tool. Seven risks were specifically evaluated: (1) that severe burns are missed, leading to undermanagement; (2) that minor burns are diagnosed as severe, leading to overmanagement; (3) that clinicians feel less responsible for the care they provide; (4) that the emotional well-being of the patient will be neglected; (5) that the patient confidentiality will be breached; (6) that patients will be less empowered in decisions around their care; (7) that some complex burns may not be recognized by the algorithm. For each risk, the participant had to evaluate the likelihood of happening on a 3-point scale from low or very low to high or very high; and the severity of the risk would it happen on a 4-point scale (insignificant, minor, moderate, or major).

The questionnaire was pilot-tested before the data collection to ensure comprehensibility, content, and length of time.

This study was approved in South Africa by the Human Research Ethics Committee of the University of Cape Town (Dnr 823/2019) and in Sweden by the Ethics Committee (Etikprövningsmyndigheten [Dnr 2019‐05122]). Informed consent was provided by participants prior to filling in the online survey. The data are deidentified and no compensation was provided for participating.

This study is an attempt to bridge the gap between what is known about the potential performance of automated diagnosis for burn care and the readiness of first contact clinicians to endorse this as a practice. We showed that a majority of first contact clinicians from resource-poor African settings would be willing to use automated diagnosis for burns should the technology be made available in their departments. The concepts embedded in the AAM contributed moderately to explaining what the intention to use such a technology rests on. Of special interest is the association between the intention to use and perceived usefulness, but not with attitudes toward the technology. It is also of note that 5 of the 8 hypotheses derived from the model were supported.

When reflecting upon the risks that such technologies could imply, first contact clinicians showed more concerns than burns specialists. They were typically concerned with the consequences of poor diagnostic accuracy (ie, undermanagement of severe burns and overmanagement of minor burns) and, like their specialized peers, feared that clinicians’ sense of responsibility would be reduced and that complex burns might not be identified. The latter was regarded as the risk with the most severe consequences across all clinicians, alongside that of severe burns not being identified. By contrast, first contact clinicians were less concerned with the risk of minor burns being diagnosed as minor or insignificant.

That first contact clinicians show a widespread acceptance of the intention to use automated burn diagnosis is well in line with previous research on the use of clinical AI in general, where 77% of the respondents from different parts of the world indicated being willing to use clinical AI if needed [33]. Perceived usefulness has already been suggested as a strong predictor of both behavioral intention and technology acceptance. This is true for technologies such as smart health care [34,35], whereas in the context of burns, it has been shown for the acceptance of mHealth [21].

As the model used does not pay attention to risk perception per se in the determination of intention to use, our results can shed light on it being a potentially underlying source of influence on the intention to use. Indeed, first contact clinicians’ appraisal of the risk is more multifaceted than that of their specialized peers. A first explanation could be self-confidence, doubting their ability to interpret the output of an automated burn diagnosis [33]. Another one could be minimal experience of mHealth for burn diagnosis, while experience of the like is known to be associated with a decrease in perceived risks in the use of smart health care, thanks to technology transfer knowledge [35]. A third one could be a sense of loss in personal usefulness, first contact clinicians being more prone to answer positively to the fact that AI would replace them.

This is the first study performed in resource-poor settings specifically on the topic of acceptance of automated diagnostic acceptance among health care professionals. This can complement the work done on a similar population on the acceptance of mHealth for burn injuries [21], but also results in the acceptance of AI-based diagnosis among medical students, also performed in resource-poor settings [36]. Studies on intention to use are important to be performed for specific contexts and populations of interests as it has been shown previously that physicians from resource-poor settings may have different opinions regarding clinical AI than their counterparts from other regions of the world [33]. The perceptions might also depend on the qualification of the surveyed professional [21,33,35]. The development of new methodologies to analyze the obtained data (such as machine learning techniques) might also assist in the future in targeting more specifically what might influence the intention to use new technologies by survey participants. The study also focused on 7 different risks that are relevant for the diagnosis of burn care and for which the knowledge around clinicians’ perceptions is important both for the development of algorithms as well as for implementation. It is likely that burns specialists, sharing a familiarity with the condition and aware of the progress made in assisted diagnosis are more inclined to trust the technology [37] and that, by contrasts to first contact clinicians, are more heterogeneous as a group as regards their knowledge and experience about burns, their skills and competences [21]. The sample was conveniently selected among clinicians attending 2 specialized conferences in South Africa for their interest in burn care and likelihood of using automated diagnosis would it be developed. This, however, comes with the limitation that the pool of people who can be included is relatively small, and the results cannot be generalized to other specialties or settings where the viewpoints of the professionals might differ. Furthermore, there might have been respondent bias with those accepting to fulfill the questionnaire having stronger opinions in favor of or against automated diagnosis for burns than those who did not respond. This study was cross-sectional and only assessed the intention to use a potential tool rather than effectively measuring the clinicians’ behaviors once automated diagnosis was implemented. While this information will be interesting to investigate in the future, the knowledge gathered in this study can further assist in the development of the algorithms themselves to respond most appropriately to the stakeholders’ demands. It could also be of interest to assess users’ perceptions with regard to different methodologies with varying performances, in particular when current advances in the technological domain are extremely promising.

This study has implications for future research as the AAM model contributes poorly to informing what the determinants of the intention to use are beyond the perceived usefulness. There is a need to assess the input of clinical risk perceptions: this should be an integral part of the model in the future, and it should receive attention during the evaluation and implementation process. Furthermore, with the body of evidence growing on the potential for automated diagnosis in burn care and its potential benefit in resource-poor settings, research on the attitudes of medical professionals with regard to this technology is necessary for a successful implementation. Finally, it would be interesting to investigate, when implementing such a technology in practice, whether it would enhance the clinical abilities of first-contact clinicians, as well as improve diagnostic efficiency, given the current state of resources and the lack of specialist clinicians.

The results show that most first contact clinicians were inclined to seek automated advice for burn diagnosis if it was available. The conceptual model proposed contributes moderately to explaining what influences intention to use beyond perceived usefulness, with implications for clinical practice where the perceived usefulness makes little doubt. This is encouraging as there are good reasons to believe that the performance of automated diagnostic support will continuously improve. When seeking additional determinants, greater attention should be placed on clinical risk perceptions as well as negative attitudes due to bad experiences or unrealistic expectations. These dimensions should also naturally belong to any AI implementation process in practice, not least to ensure sustainability.