The Association Between Family Physicians’ Artificial Intelligence Anxiety Levels and Medical Artificial Intelligence Readiness

Artificial intelligence (AI) refers to computer systems capable of mimicking certain aspects of human intelligence, including learning, reasoning, and problem-solving⁽¹⁾. AI has emerged as one of the most rapidly evolving domains in contemporary science and technology, with goals encompassing simulating human intelligence, solving complex problems, data analytics, making predictions, and decision-making^{(2, 3)}.

The healthcare sector is a significant AI application domain. This technology promises an array of advantages, including enhanced diagnostic accuracy, optimization of treatment plans, and early disease detection^{(4, 5)}. AI-driven systems are adept at processing large volumes of healthcare data, offering clinicians an opportunity for more accurate diagnoses, developing effective treatment modalities, and ultimately improving patients’ quality of life. Furthermore, AI plays a critical role in medical imaging, aiding early diagnosis and treatment in fields such as radiology, pathology, and neurology by analyzing images to identify abnormal structures^{(6, 7)}.

The application of AI in healthcare has brought about a transformative wave, with an extensive range of tools designed to assist diagnostics, treatment planning, and patient management^(8-11). However, seamless and effective integration of these AI applications into clinical settings requires a comprehensive readiness strategy for healthcare providers. The concept of “readiness in medical AI” goes beyond simply installing software or deploying algorithms; it encompasses the clinical, educational, and ethical readiness to utilize these new tools effectively⁽¹²⁾. Healthcare providers, especially primary care physicians, must critically evaluate the reliability and clinical relevance of AI-based tools. Although many of these algorithms may be validated through rigorous trials or studies, physicians should understand the specific conditions under which the AI models are trained and tested^(13-15).

It is crucial for clinicians to understand the fundamental principles of AI technologies. Medical schools and continuing education programs should offer courses and training sessions that delve into the basics of machine learning, data analysis, and types of algorithms most commonly used in healthcare applications. In practice, these foundational technologies support informed decisions. The landscape of medical AI is continuously changing with advancements in machine-learning models and algorithms. To keep healthcare providers up to date, ongoing professional development initiatives should offer workshops, webinars, and updated training modules that focus on the latest innovations.

Although existing studies have largely focused on secondary and tertiary healthcare settings, limited research has addressed physicians’ concerns and readiness for AI in primary care. Notably, the scarcity of studies on the relationship between AI anxiety levels and medical AI readiness in family physicians attracts attention. Given these gaps, this study aimed to investigate the relationship between anxiety levels and readiness for medical AI among family physicians.

Study Design

This descriptive and cross-sectional study was conducted between April and July 2022 in Türkiye, using social media platforms such as WhatsApp and Telegram. The study population consisted of general practitioners and family medicine specialists working in public and private primary healthcare services in Türkiye (n=24.082). The required minimum sample size was calculated as 191, based on a 0.20 effect size for correlation analysis in the G*Power 3.1.9.7 program, with an alpha level of 0.05 and 80% power⁽¹⁶⁾. A total of 199 volunteers were recruited from the target population.

Data Collection

Data were collected by a self-administered method with the help of a survey form. In the study conducted as an e-survey, a Google form was created and social media tools such as WhatsApp and Telegram were used. Participants were reached through simple random sampling. The first question sought informed consent from participants to ensure voluntary participation. When they agreed to participate in the study, they were asked to answer the e-survey. The e-survey included a personal information form, an AI anxiety scale (AIAS), and a medical AI readiness scale for medical students (MAIRS-MS).

AIAS

AIAS was developed by Wang and Wang⁽¹⁷⁾. Terzi⁽¹⁸⁾ conducted the Turkish validity and reliability study with teachers in 2020. The scale consists of 21 items and four sub-dimensions: learning, job replacement, socio-technical blindness, and AI configuration. Scoring ranges from 21 to 147, with higher scores indicating increased levels of anxiety towards AI. Terzi⁽¹⁸⁾ found, Cronbach’s alpha reliability coefficient of the scale to be 0.96 for the total score, and reported Cronbach’s alpha reliability coefficients of the learning, job replacement, socio-technical blindness, and AI configuration subscales as 0.89, 0.95, 0.89, and 0.95, respectively. The scale was applied to family physicians and family medicine specialists by Kolcu et al.⁽¹⁹⁾. and the validity and reliability of the scale was re- evaluated Cronbach’s alpha reliability coefficient for the total score of the scale was found to be 0.95.

MAIRS-MS

MAIRS-MS developed by Karaca et al.⁽²⁰⁾. The scale consists of 22 items and four sub-dimensions: cognition, ability, vision, and ethics. Scoring ranges from 22 to 110, with higher scores indicating higher levels of readiness. Karaca et al. ⁽²⁰⁾ found Cronbach’s alpha reliability coefficient of the scale 0.87 for the total score, and reported Cronbach’s alpha reliability coefficients of the cognition, ability, vision, and ethics subscales as 0.83, 0.77, 0.72, and 0.63, respectively.

Ethics Approval

The study was approved by the Kırklareli University Local Ethics Committee (approval no: E-35523585-302.99-45759, date: 11.04.2022) and was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the volunteers. Permission to use these scales was obtained from the authors.

Statistical Analysis

Descriptive statistics, including frequency (n), percentage (%), mean, and standard deviation, were used for data analysis. Reliability analysis was conducted for the AIAS and MAIRS, and the results were evaluated with Cronbach’s alpha coefficient. The normality of distribution was tested with the Shapiro-Wilk test. Differences between two independent groups in parametrically distributed AIAS and MAIRS total scores were analyzed using the independent t-test, and differences between three or more independent groups were examined with one-way analysis of variance. The difference between two independent groups in the non-parametric distributed AIAS and MAIRS subscale scores was examined with the Mann-Whitney U test, and the difference between three or more independent groups was examined with the Kruskal-Wallis H test. The relationship between two numeric variables was examined with Pearson correlation analysis in parametric distributions and with Spearman’s correlation analysis in non-parametric distributions. Both unadjusted and adjusted multivariate linear regression analyses were performed to identify predictors associated with AIAS and MAIRS. Models were adjusted for age and sex. Since the sub-dimensions of AIAS and MAIRS were not distributed normally, Z transformation was performed before they were included in the model. The change in variance in the created models was explained by adjusted R-square. Data were analyzed using the Statistical Package for the Social Sciences version 26.0 (SPSS 26.0, SPSS, Inc., Chicago, IL, USA); p<0.05 was considered statistically significant.

The mean age of the participants is 45.93±9.19 years (min: 27, max: 63), and their average professional experience is 20.88±9.55 years (min: 2, max: 39). Among the participants, 63.8% were male, and of all participants, 44.2% worked in provincial centers. In the research group, 19.1% had specialized training in family medicine and 12.6% had received education in information technology or computer science. Of the family physicians, 30.2% are interested in information technology or computer science, while 21.1% are interested in the field of AI. Table 1 presents the distribution of participants’ descriptive characteristics.

Reliability analysis confirmed the internal consistency of both scales in this study population. For the AIAS, Cronbach’s alpha coefficients were 0.96 for the total scale, and 0.94, 0.94, 0.93 and 0.98 for the learning, job replacement, socio-technical blindness, and AI configuration subscales, respectively. For the MAIRS, Cronbach’s alpha coefficients were 0.96 for the total scale, and 0.94, 0.91, and 0.83 for the cognition, ability, vision, and ethics subscales, respectively (with 0.94 for both cognition and ability).

This study investigated the distribution of total and sub-dimension scores on the AIAS and MAIRS based on participants’ descriptive characteristics (Table 2 and Table 3).

Table 4 presents the correlation analysis between AI anxiety and medical AI readiness. Significant negative correlations were found between overall medical AI readiness and most anxiety dimensions, with the strongest relationships observed for learning anxiety (r=-0.311, p<0.001) and AI configuration anxiety (r=-0.256, p<0.001). Learning anxiety showed consistent negative correlations with all readiness dimensions, particularly ability (r=-0.337, p<0.001) and vision (r=-0.286, p<0.001). Job replacement anxiety demonstrated weaker associations, while socio-technical blindness showed minimal correlation with most readiness dimensions except vision, (r=-0.145, p=0.041).

Table 5 presents a multivariate linear regression analysis of predictors associated with AIAS and MAIRS. Learning anxiety and AI configuration anxiety were significant negative predictors of overall AI readiness, while socio-technical blindness showed a positive association (Table 5).

Our study found that female family physicians consistently felt more anxious about AI technology compared to their male colleagues, especially when it came to configuring AI systems (p=0.004). This pattern isn’t unique to medicine, researchers have been documenting similar gender differences with technology for decades⁽²¹⁾. Park⁽²²⁾ found that women often feel less in control when using digital tools, contributing to increased anxiety. Mayall's⁽²³⁾ research showed that women tend to doubt their computer abilities more than men, even when their actual skills are similar. This trend has historical roots; as observed by Shashaani⁽²⁴⁾, these differences in computer attitudes were evident during the 1990s. Fetler⁽²⁵⁾ found that male students consistently scored higher on computer literacy tests. Interestingly, Kaya et al.’s⁽²⁶⁾ more recent work suggested these gender gaps might be shrinking, though our findings show they are still very much present among practicing physicians.

The higher anxiety levels we observed among female doctors likely stem from less exposure to computing during their medical training, simply working in environments where they felt less supported when learning new tech skills. What this means for hospitals and clinics is clear: the typical one-size-fits-all approach to AI training simply won’t work. Medical organizations need to create learning programs that meet people where they are technologically, offering safe spaces where doctors can ask questions without feeling judged and take time to experiment with AI tools at their own pace. We also need research that follows physicians over time to see whether these anxiety differences actually affect how they use AI with patients and whether thoughtful, supportive training can help close these gaps in real-world practice.

Contrary to the initial hypotheses, having specialized training in family medicine did not result in a noteworthy decrease in AI anxiety. Conversely, a noticeable reduction in AI anxiety was discerned in participants with an educational background or vested interest in computing or IT, thus substantiating previous studies^{(26, 27)}. This revelation directs us toward a pivotal intervention point: the enhancement of technological familiarity and proficiency may play a crucial role in mitigating anxiety related to AI, irrespective of specialized medical training. Thus, the findings imply that the incorporation of specialized training and computer literacy into medical education or specialty training curricula is necessary, warranting thoughtful deliberation.

A decrease in learning and AI configuration anxiety was positively correlated with enhanced medical AI readiness. This finding emphasizes the potential efficacy of meticulous and targeted educational interventions in alleviating AI anxiety, subsequently fortifying AI readiness, and aligning with Shinners et al.⁽²⁸⁾ propositions. These findings have substantial implications for the refinement of training curricula, suggesting that the incorporation of AI-related knowledge and ability development could be crucial for cultivating competencies in prospective healthcare professionals^{(29, 30)}. While the utility of AI in medicine is largely acknowledged, most healthcare practitioners do not possess comprehensive knowledge of AI’s fundamentals and express concerns about the possible repercussions of its extensive implementation in clinical settings⁽³¹⁾. Given these findings, it is imperative to further investigate whether there is a causal relationship between learning anxiety, AI configuration anxiety, and medical AI readiness, to delineate the underlying dynamics more precisely and develop more tailored and effective interventions.

The results of the study revealed that an increase in socio-technical blindness is linked to better readiness in abilities, a phenomenon that is unique and seemingly contradictory. This implies a complicated relationship between understanding technological repercussions, and readiness to adopt them. Such complexity prompts critical reflections on whether heightened cognizance of AI’s restrictions and potential adverse impacts might obstruct the eagerness or perceived capability to apply AI applications proficiently. The potential of AI innovations to enhance care and reduce workload is compelling for organizations. However, skepticism exists among healthcare entities regarding the feasibility of integrating technologies such as AI, owing to the substantial resources needed for infrastructure, training, and policy formulation^{(30, 32)}. Socio-technical systems theory advocates balanced attention to individuals, organizations, and technology prior to introducing new technology. It is vital to delve into the perceptions of healthcare staff to ascertain the hurdles in technology integration and tailor education and training programs based on their requirements, ensuring the optimal utilization of emerging technologies^{(33, 34)}.

The relation between reduced learning and AI configuration anxieties, and enhanced readiness for ethical challenges, especially when coupled with increased socio-technical blindness, calls for more nuanced and multifaceted approaches to AI education in healthcare. This emphasizes the necessity for robust training frameworks that go beyond technical competencies to address ethical considerations and ensure responsible and mindful incorporation of AI technologies into healthcare^{(27, 30)}.

There is an urgent need to adequately equip the healthcare workforce for the integration of AI within healthcare settings. Gaining insights into workforce perceptions of AI is crucial, as it can reveal potential challenges and barriers that organizations might encounter while using this transformative technology. In this study, we examined the levels of AI anxiety and medical AI readiness among family physicians to identify and elucidate existing gaps and disparities. Similarly, amplifying such studies across the entire spectrum of healthcare providers is paramount. Identifying strategic measures to prepare healthcare professionals for the integration of emerging technologies and AI, which are poised to become integral components of their lives, is crucial. This preparation is essential for the seamless adaptation and maximization of the benefits that these innovations can bring to healthcare outcomes.

Study Limitations

A limitation of the study is that it does not represent all family physicians because the research was conducted via social media. Despite the initial studies providing noteworthy insights, follow-up studies are needed to corroborate results, considering the varied geographical, cultural, and institutional contexts. Additionally, future research should delve deeper into exploring causal relationships between learning anxiety, AI configuration anxiety, and medical AI readiness, while aiming to understand the underlying mechanisms between socio-technical blindness and ability readiness.

In conclusion, this study illuminates how diminishing learning and AI configuration anxieties can lead to an elevation in medical AI readiness across cognition, ability, vision, and ethics dimensions. The elucidated gender differences, the impact of computer literacy and specialized training, and novel and intricate relationships between socio-technical blindness and ability readiness are critical areas that require further exploration and validation. Above all, these insights emphasize the need for holistic approaches to AI education in healthcare, incorporating technical, ethical, and socio-technical components to ensure the responsible and proficient use of AI in family medicine, thereby bridging the existing gap in primary healthcare research on AI readiness and anxiety.

Future research could benefit from longitudinal studies that track changes in AI anxiety and readiness over time, especially as family physicians gain more exposure to AI technologies in their practice. This could help in understanding how attitudes and readiness evolve with increased familiarity and use of AI in medical contexts. Studies should be designed to assess the effectiveness of targeted educational interventions aimed at reducing AI anxiety and enhancing readiness. This could include developing specific training modules or simulation-based learning environments that address the identified cognition, ability-based, and ethical concerns. Comparing AI anxiety and readiness between family physicians and other medical specialties could provide insight into specialty-specific concerns and readiness levels. This might also reveal unique educational needs or resistance points that are specific to different areas of practice. Research should also focus on how the integration of AI into clinical practice affects patient outcomes, physician decision-making, and workflow efficiency. This could help to understand the tangible benefits and potential pitfalls of AI in healthcare. By addressing these recommendations, future research could contribute to a deeper understanding of the factors that influence the adoption of AI in healthcare and support the development of strategies to optimize its integration, benefiting healthcare providers and patients.