A cross-sectional, self-administered questionnaire survey was conducted between December 1, 2024, and December 31, 2024. All public and private hospitals in Jordan that offer O&G training programs were included. Trainees holding a professional degree in IT were excluded.

An English-language hard copy questionnaire was developed by the researchers and was informed by relevant published literature on AI technologies [14-18]. Additionally, ChatGPT was used to suggest some questionnaire items. The research team reviewed these suggestions for relevance and clarity and decided which items to retain, edit, or exclude from the final version of the questionnaire. Content validity was established by 3 gynecologists. Clarity and comprehension were assessed by the results of a pilot study that included 20 trainees [19]. All comments were considered in the final version of the questionnaire that was used for data collection. Moreover, the data collected from the pilot study were not included in the final analysis because the questionnaire was subsequently modified.

The questionnaire consisted of 2 parts. The first part collected data on the participants’ characteristics, including age, gender, residency year, place of training (public or private hospital), and number of years of experience since graduation (fresh graduate: <3 years; senior graduate: ≥3 years). In addition, the participants’ current self-reported knowledge of IT was measured using a 5-point Likert scale (very poor, below average, average, above average, or excellent). Furthermore, participants were asked about their knowledge sources of the medical applications of AI (formal lecture, conference, internet, or a colleague), if they ever had formal training on the medical applications of AI during residency training (yes, no, or not sure), and if they believe that AI should be included in undergraduate and postgraduate medical training (yes, no, or not sure).

The second part of the questionnaire was used to assess trainees’ self-reported knowledge about current applications of AI in O&G, their attitudes toward its introduction into clinical practice, and their perceptions of importance. This part comprised 36 statements across 3 domains: 9 assessing self-reported knowledge, 21 assessing attitudes, and 6 assessing perceptions. Moreover, the unequal number of items in each domain was not intentional but resulted from the number of items generated for each domain during questionnaire development. Of the 36 items, 21 (58.3%) assessed attitude, 9 (25%) assessed self-reported knowledge, and 6 (16.7%) assessed perception. At the beginning of each domain, candidates were asked the following question: “How much do you agree with the following statements?” This was followed by a series of statements to which participants responded using a 5-point Likert scale (strongly agree, agree, undecided, disagree, and strongly disagree). The statements were phrased to reflect high self-reported knowledge, positive attitude, and positive perception of importance (Multimedia Appendix 1).

Members of the research team approached potential participants in the research sites during the morning report meetings. This time was chosen because it was when most members of the team, including those who had been on call the previous day, those on call that day, and trainees assigned to other duties, were expected to be present. The research objectives and the inclusion and exclusion criteria were explained to the participants.

The total number of O&G trainees in Jordan at the time of the study was 497. Therefore, a sample size of 217 was required to achieve a 95% confidence level and a 5% margin of error. Allowing for a 20% rate of ineligible responses, we aimed to recruit 270 participants for the study [20].

Several variables were grouped for better comparisons. These included age in years (<30 and ≥30 years), residency year (junior: years 1 and 2 and senior: years 3, 4, and 5), number of years since graduation (junior graduate: <3 years and senior graduate: ≥3 years), place of training (public or private hospital), current self-reported knowledge of AI and medical applications of AI (below average or poorer, and average or better), and source of knowledge about the medical application of AI (internet or noninternet).

For the calculation of self-reported KAP domain scores, the Likert scale responses were converted into numerical values (strongly agree=5, agree=4, undecided=3, disagree=2, and strongly disagree=1). Then, the maximum attainable scores for each domain were calculated by summing the responses, with maximum scores of 45, 105, and 30 for the self-reported knowledge, attitude, and perception domains, respectively. Additionally, the maximum attainable scores of the 3 domains were summed to generate a study-specific KAP score, with a maximum possible score of 180. In this study, the KAP score was used to define KAP-based readiness.

To allow comparisons across the scores of the 3 domains and the KAP-based readiness, scores were converted into percentages of their respective maximum attainable scores. For descriptive purposes, the total scores of the 3 domains were further grouped using the modified Bloom cut-off points: <60%, 60% to 79%, and ≥80%. Therefore, self-reported knowledge, attitude, perception, and KAP-based readiness scores were grouped as low, moderate, or high [21].

Data normality was tested using the Shapiro-Wilk test. For normally distributed data, descriptive statistics were reported as mean (SD), whereas nonparametric and Likert scale data were reported as median (IQR). Spearman rank correlation was used to study the relationships between the percentage scores of the 3 domains. Bivariable analyses were conducted to examine associations between KAP-based readiness categories and participant characteristics. Additionally, multivariable linear regression analysis was performed using the KAP-based readiness percentage score as a continuous variable. Age group, gender, residency year, place of training, self-reported IT knowledge, source of AI knowledge, and formal AI training during residency were entered simultaneously as independent variables. Multicollinearity was assessed, and sensitivity analyses were performed.

A total of 270 questionnaires were distributed, and all were returned. Of these, 52 (19.3%) were excluded because they had more than 20% missing item-level data. The remaining 218 questionnaires were included in the final analysis. Among the included questionnaires, missing data were <3% and were replaced using the median or modes, as appropriate [22].

Data were analyzed using SPSS (version 26; IBM Corp). The level of significance was set at α<.05.

Ethics approval was granted by the research committee of the Faculty of Medicine, Al-Balqa Applied University (reference number: ب ع ط / 2024 / 25). In reporting this study, we followed STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) [23] and KAP studies [24] reporting guidelines. Participants were informed that no personal identifiable information would be collected, no incentives would be offered to participate in the study, and the mean time needed for the completion of the questionnaire was 7 minutes. Verbal rather than written consents were obtained because of the limited time available during morning report meetings. Members of the research team explained the study aims and the voluntary nature of participation. Questionnaires were distributed to trainees who verbally agreed to participate.

Of the 218 trainees recruited, the median age was 28 (IQR 24-38) years, 180 (82.6%) participants were female, 117 (53.7%) were junior trainees, 128 (58.7%) were senior graduates, and 148 (67.9%) reported training at a public hospital. Additionally, “average or better” self-reported IT knowledge was expressed by 183 (83.9%) participants. The internet was the most commonly reported source of knowledge about the medical applications of AI, reported by 165 (75.7%) participants, and 151 (69.3%) and 160 (73.4%) participants believed that AI training should be introduced to undergraduate and postgraduate medical education, respectively (Table 1).

The results showed that Cronbach α was 0.79 for self-reported knowledge, 0.82 for attitudes, 0.71 for perceptions, and 0.83 for study-specific KAP-based readiness, indicating acceptable to good internal consistency. Data analysis showed statistically significant positive correlations between self-reported knowledge and attitude (r=0.39; 95% CI 0.26-0.51; P<.001), self-reported knowledge and perception of importance (r=0.47; 95% CI 0.34-0.58; P<.001), and attitude and perception of importance (r=0.59; 95% CI 0.47-0.68; P<.001).

Table 2 shows the median (IQR) of the percentages of the maximum attainable scores of the 3 domains and the KAP-based readiness, in addition to the 3 modified Bloom’s categories of the respective domains. The analysis showed that the moderate category was most frequent in all domains: self-reported knowledge: 72.9% (159/218); attitude: 80.7% (176/218); and perception: 65.6% (143/218).

The KAP-based readiness was moderate in 86.2% (188/218) of the participants. The distribution of the 3 KAP-based readiness categories in different subgroups of participants’ characteristics is shown in Table 3. Moderate KAP-based categories were the most common in every subgroup, ranging from 81.8% (18/22) to 97.1% (34/35). While low KAP-based readiness was highest among participants who trained at private hospitals (10/70, 14.3%), high KAP-based readiness was highest among participants who had received formal AI training during residency (3/22, 13.6%).

In the bivariable analyses, no statistically significant associations were found between KAP-based categories and trainee characteristics. Although the association did not reach statistical significance, participants who had received formal AI training were more likely to demonstrate high KAP-based readiness than those who had not received formal AI training (3/22, 13.6% vs 9/196, 4.6%). Similarly, participants supporting postgraduate AI training were more likely to have high KAP-based readiness than those who did not support AI training or were unsure (3/22, 6.9% vs 9/196, 1.7%; Table 4).

The multivariable linear regression model was not statistically significant (F_7,210=1.049; P=.39; R²=0.034; adjusted R²=0.002). None of the participant characteristics included in the analysis was associated with the study-specific KAP-based readiness score. Two sensitivity analyses were conducted. In the first analysis, trainees’ beliefs about incorporating AI training into undergraduate and postgraduate medical education were added to the model, which improved the model fit (ΔR²=0.104; ΔF_2,208=12.573; P<.001; R²=0.138). In the second analysis, years since graduation was added to the primary model, but this did not improve the model fit (ΔR²=0.001; ΔF_1,209=0.136; P=.71). Although not statistically significant, higher adjusted KAP-based readiness scores were observed among trainees who had previously received formal AI training, those who trained at public hospitals, and those who used noninternet sources as their source of AI knowledge (Table 5).

The participants showed moderate levels of AI-related knowledge, positive attitudes and perceptions, and moderate KAP-based readiness. Additionally, formal training in the medical applications of AI during residency was uncommon. The internet was the most frequently reported source of knowledge about the medical application of AI, and the majority of participants supported incorporating AI training into undergraduate and postgraduate medical education. No statistically significant associations were identified between study-specific KAP-based readiness categories and participants’ characteristics. Additionally, multivariable linear regression did not identify independent predictors of the total KAP-based readiness scores.

Readiness for the introduction of AI technologies to clinical practice includes factors related to health care professionals and health care facilities [25,26]. Therefore, successful implementation requires adequate knowledge and positive attitudes toward AI, which may encourage greater engagement with this emerging technology. Moreover, it requires supportive organizational infrastructure and policies [27,28]. Therefore, the KAP-based readiness scores in this study are part of the overall readiness evaluation.

The results showed that participants have moderate self-reported knowledge and moderately positive perception and attitude. While this may indicate that O&G trainees are generally aware of the technology, these results should not be considered as evidence of satisfactory readiness because discrepancies exist between O&G trainees’ self-reported IT and AI skills and their actual interaction with the AI technology in clinical scenarios [10], highlighting the need for structured training to improve readiness.

The significant positive correlations between self-reported knowledge, attitude toward the introduction of AI, and its perceived importance suggest that participants who reported greater self-reported knowledge tend to report higher positive attitude and perception scores (P<.001). While a similar pattern has been reported by another study [29], greater knowledge in the medical applications of AI may not translate to more positive attitudes, probably because health care professionals with greater knowledge of AI may have more awareness of its ethical and privacy-related implications, which may influence their attitudes toward its use [30].

This study showed that most of the participants never received formal training on the medical applications of AI. An earlier report that included internal medicine trainees showed similar results [31]. Additionally, three-quarters of the participants reported that the internet was their main source of knowledge about the medical application of AI. These findings are supported by the results of an earlier report that included medical and dental students [32]. Both findings are relevant and may be interrelated, as the lack of formal training may encourage participants to seek knowledge from informal sources, which may not always be reliable [33]. Moreover, structured training in the medical applications of AI was shown to significantly improve clinical competencies in diagnosis and treatment [34]. Additionally, formal AI knowledge sources such as conferences and scientific literature showed a positive association with better AI knowledge and attitudes [35]. Therefore, the limited formal training in medical AI technologies and reliance on informal sources of knowledge observed in this study highlight an important gap in current O&G training.

The participants in this study supported the integration of AI training into undergraduate and postgraduate medical education. Adding these two variables has improved the model in the sensitivity analysis. However, this improvement should be interpreted with caution, as these two variables are conceptually related to the attitude component of the KAP-based readiness score. Similar educational gaps and support have been reported by previous studies [18,31]. Moreover, limited formal AI training among medical students was associated with lower knowledge and less favorable attitudes toward its introduction [36]. Therefore, incorporating training on the medical applications of AI into undergraduate and postgraduate medical education may improve knowledge and enhance readiness for the introduction of AI technologies.

The current place of training in this study was not significantly associated with KAP-based readiness. However, low readiness was more frequently observed among trainees who trained at private hospitals. Differences in KAP-based readiness across training settings may be related to variations in organizational infrastructure and initiatives to introduce new technologies into clinical practice [37].

In this study, none of the measured trainees’ characteristics were significantly associated with KAP-based readiness. Therefore, further research should examine individual and organizational variables that may contribute to readiness, such as IT and AI literacy and training, health care professionals’ acceptance of the new technology, ethical concerns, and organizational infrastructure [38,39].

Assessing trainees’ knowledge, attitude, and perception before implementing structured training may identify gaps that should be addressed prior to introducing IT to clinical practice. The findings of this study contribute to the limited literature on O&G trainees’ perspectives toward AI and support the need for structured AI training to improve readiness.

This study specifically reported on the perspective of O&G trainees. The sample size was calculated in advance based on the available national number of trainees. Trainees were recruited from public and private hospitals, from different residency years, and from different geographical areas, therefore providing views from a wide range of training settings. The study assessed several related domains, including self-reported knowledge, attitude, perception, and KAP-based readiness. In addition, the questionnaire showed acceptable internal consistency across the different domains.

This study is cross-sectional; therefore, causal relationships between trainees’ characteristics and KAP-based readiness cannot be established. The questionnaire was developed specifically for this study and, therefore, was study-specific and not externally validated. The combined KAP-based readiness score was influenced more by the attitude domain because the 3 domains contained unequal numbers of items. Therefore, the score should not be interpreted as assigning equal weight to the 3 domains. Because of substantial missing item-level data in the excluded questionnaires, we were unable to compare them with the included questionnaires. IT and AI knowledge were not objectively assessed because this was not an aim of the study. Recruitment was based on trainees who were accessible at the time of data collection in the participating hospitals. The small proportion of item-level missing data was imputed, which may have negative impacts on the results.

O&G trainees in Jordan demonstrated moderate self-reported AI-related knowledge, generally positive attitudes and perceptions of its importance, and moderate study-specific KAP-based readiness. Moreover, formal training on AI medical applications was uncommon, and most trainees supported the integration of AI training into medical education. These findings support the need for structured AI education, and future research should evaluate broader and objectively measured individual and organizational determinants of AI-related readiness.