Adopting AI in medical ethics review: A configurational fsQCA study of practitioners’ willingness

Abstract

Objective

The rapid integration of AI in medical research necessitates that Research Ethics Committees (RECs) employ AI tools to oversee AI-driven studies. However, systematic research into the willingness of ethics review practitioners to adopt AI remains scarce. This study addresses this gap by investigating the determinants of AI adoption intentions among REC members and staff.

Methods

Based on the Unified Theory of Acceptance and Use of Technology (UTAUT) and the Technology-Organization-Environment (TOE) frameworks, a context-adapted questionnaire was developed and distributed to 697 ethics review practitioners across 11 provinces in China. Yielding 516 valid responses. Fuzzy-set Qualitative Comparative Analysis (fsQCA) was applied to identify multiple configurational pathways to high AI adoption willingness and examine differences between the two practitioner groups.

Results

The study established a theoretical framework for AI adoption willingness, integrating performance expectancy, perceived risk, social influence, facilitating conditions, technology anxiety, personal innovativeness. No single necessary condition for high adoption intention was identified, with all consistency scores falling below 0.9. For ethics committee members, seven pathways emerged across three models: social influence-driven, personal innovativeness-driven, and organization-technology synergy-driven (solution coverage=0.744). For ethics committee staff, eight pathways were identified, which were categorized into personal innovativeness-driven and organization-individual innovativeness synergy-driven (solution coverage = 0.753). Performance expectancy and personal innovativeness were core conditions in most pathways, their roles differed markedly between groups.

Conclusion

This study constructs a theoretical model to examine AI adoption willingness among ethics review practitioners. Findings reveal that high adoption readiness emerges from multiple equifinal configurations rather than isolated factors. Ethics committee members require organizational empowerment and social influence, while staff motivation depends primarily on personal innovativeness. These divergent pathways provide suggest distinct implementation strategies: providing institutional safeguards for members while fostering individual innovativeness among staff. These insights will enhance AI-assisted ethics review and inform technology governance in similar healthcare contexts.

Keywords

medical ethics review practitioners adopting willingness technology-organization-environment fuzzy-set qualitative comparative analysis

Introduction

The rapid advancement of AI technology is profoundly reshaping the global medical research and innovation, spanning various applications from diagnosis to drug development.¹ This transition necessitates more robust medical ethics governance. Ethics review in medical institutions serves a crucial mechanism to ensure that technological innovations comply with ethical norms while guaranteeing that clinical research respects the rights and well-being of participants and patients.²

Traditional medical ethics reviews rely primarily on manual expertise. However, conventional oversight mechanisms struggle to maintain consistency and scalability in the face of accelerating clinical research. Traditional methods are often inadequate for the rapid iteration of medical technologies. Challenges such as prolonged processing times and individual experiential biases have becoming outstanding.³ This inadequacy is further exacerbated as Research Ethics Committees (RECs) use AI tools to oversee AI-driven protocols. This recursive cycle necessitates a focus on fundamental trust and accountability, making the adoption of AI by practitioners a matter of fundamental trust and accountability, far beyond a mere issue of efficiency.

“AI for Science' provides novel avenues to address these systemic constraints. A notable development is “Mirror,' the first specialized ethics review agent introduced at the 2025 World Artificial Intelligence Conference. This system leverages generative AI to regulatory tracking, risk identification, and report generation (WAIC Proceedings,2025). It is currently undergoing pilot implementation, representing a shift toward AI-augmented ethical governance.⁴ Fukataki et al. developed a ChatGPT-based pre-review tool for institutional review boards, demonstrating its potential to improve review accuracy and reproducibility.⁵ Some institutions in the United States are also exploring AI for initial protocol review and risk classification to reduce the workload of ethics committees.⁶ These practical developments reflect an institutional shift toward transformative review models. By delegating repetitive tasks to AI, ethics committee members can prioritize normative deliberation and complex decision-making.⁷ Ultimately, AI integration will transcend mere operational optimization, catalyzing a paradigm shift from experience-based oversight toward a data-informed, human-centric decision-making framework.⁸

Consequently, medical RECs face the unprecedented challenge of using AI tools to review AI projects. This governance paradox has gained significant scholarly attention. However, AI integration into ethical oversight introduces critical risks: (1) The black-box nature of AI complicates the interpretation of decision-making logic, which may raise concerns about the transparency and accountability of the review process.⁹ Furthermore, biased training datasets may replicate or exacerbate existing ethical inequities.¹⁰ (2) AI processes highly sensitive subject information, where leakage or misuse could lead to severe legal and ethical consequences.¹¹ These concerns regarding reliability and accountability directly influence the willingness of committee members to adopt AI systems.^12,13

Despite technological advancements, the human factor remains the primary bottleneck for effective AI implementation.^14–17 Ethics committee members retain ultimate accountability and legal liability for review decisions, granting them significant authority and discretion.¹⁸ Conversely, ethics committee staff manage operational tasks, such as preliminary screening and procedural coordination.^19,20 This role heterogeneity leads to distinctly adoption motivations. Consequently, the transformation of ethics review models depends on the attitudes and engagement of these distinct practitioner groups. How do they view the introduction of AI tools into medical ethics review? Are they willing to actively learn and use AI to assist in ethics review? How do factors such as trust in AI accuracy, ease of use, and concerns about their own professional value affect their adoption willingness?

This study investigates the willingness of these practitioners to adopt AI and the underlying driving mechanisms. It aims to identify key influencing factors, analyze their configurational pathways and compare pathway differences between the two groups of ethics review practitioners. These findings will provide a theoretical basis and practical guidance for AI application in ethical review work.

UTAUT based AI adoption willingness analysis framework

This study utilizes the UTAUT as its core framework.^21,22 By integrating context-specific factors, the research develops a multidimensional model comprising six key determinants. (1) performance expectancy,²³ (2) social influence,²⁴ (3) facilitating conditions are directly derived from the core UTAUT model²⁵; (4) perceived risk is an extension from technology acceptance literature,²⁶ added due to the high-stakes nature of ethics review; (5) technology anxiety is an extension from research on human-computer interaction and technology rejection²⁷; (6) personal innovativeness is included to account for individual variance in technological receptivity.²⁸ This integrated approach enables a systematic investigation into the mechanisms driving AI adoption among ethics review practitioners.

Technical factors influencing the adoption of AI among ethics review practitioners

Performance expectancy reflects the extent to which ethics review practitioners believe that using an AI system can enhance task performance and review quality.²⁹ This construct encompasses the perceived capacity of AI to improve review efficiency and decisional consistency.³⁰ AI system reliability is a primary consideration, focusing on algorithmic accuracy, decision explainability, and adaptability to complex ethical scenarios. This factors directly influence the authority of ethics review results. At this stage, AI technology holds the potential in assisting ethics review practitioners with standardized processes, providing risk warnings and decision support, and improving work performance,³¹ its efficacy in high-stakes ethical adjudication requires further validation.³² Consequently, performance expectancy remains a critical determinant of adoption intentions.

Perceived risk denotes the subjective assessment of potential adverse outcomes and uncertainties associated with AI adoption.³³ Specifically, AI systems rely on vast amounts of sensitive data, including research protocols and personal subject information. Data leakage or misuse could lead to severe legal and ethical accountability,³⁴ directly increasing perceived risk. In addition, practitioners concern that AI decision-making logic may conflict with established ethical principles or cultural contexts,³⁵ leading to decisions that are efficient but biased, thereby inhibiting adoption willingness.

Organizational factors influencing the adoption of AI among ethics review practitioners

Organizational support provides a critical external environment that shapes the adoption willingness of ethics review practitioners.³⁶ Social influence reflects the expectations of key stakeholders, including peers, superiors, and institutional leadership.³⁷ This is manifested through management support, policy promotion, demonstrations of industry best practices, and positive feedback from colleagues.³⁸ First, strong leadership endorsement, demonstrated through resource allocation and policy, reinforces the legitimacy of AI systems. Such influence encourages practitioners to prioritize AI application strategically. Second, systematic training and technical support are essential for effective technology use. Thus, social influence remains a decisive factor in AI adoption among ethics review practitioners.³⁹

Facilitating conditions denote the extent to which practitioners believe that necessary organizational infrastructure and resources are available to support AI integration.⁴⁰ These conditions include essential hardware, dedicated funding, technical training, reliable support, and compatibility with existing ethics review systems. Furthermore, adequate data and computing resources are critical for effective implementation.Greater accessibility to these support mechanisms increases the likelihood of AI adoption. Therefore, facilitating conditions represent a decisive organizational factor influencing AI adoption willingness.⁴¹

Individual factors influencing the adoption of AI among ethics review practitioners

While the UTAUT framework primarily addresses the cognitive dimensions, affective factors significantly influence adoption intentions.⁴² Technology anxiety denotes the emotional unease or apprehension practitioners experience toward AI. This anxiety often stems from the perceived threat that AI poses to professional relevance, which dampens technological receptivity.⁴³ As AI capabilities increasingly demonstrate the potential to replace human tasks, they evoke stronger anxiety than other technologies. This heightened anxiety exerts a substantial impact on adoption willingness⁴⁴ Consequently, this study incorporates technology anxiety into the analytical framework.

Personal innovativeness refers to an individual’s inherent tendency to try new technologies.⁴⁵ This trait manifests as a receptive attitude toward AI, the proactive workflows optimization through technological integration, and a greater tolerance for systemic uncertainties.⁴⁶ As a positive intrinsic quality, personal innovativeness effectively promotes technology adoption. Prior research has confirmed this positive impact within consumer contexts.⁴⁷ Because AI represents a disruptive innovation, its adoption relies on high innovativeness among ethics review practitioners. Thus, this variable is included in our model.

Research model

This study integrates the UTAUT and TOE framework⁴⁸ to develop a configurational model for analyzing AI adoption willingness among medical ethics review practitioners. The UTAUT identifies key individual factors influencing adoption willingness, while the TOE framework provides broader contextual dimensions for technological innovativeness. This combination captures both micro-level cognitive factors and macro-level contextual influences. We integrated these frameworks by mapping UTAUT constructs onto three adapted TOE domains (Figure 1).

Figure 1.

A model of AI adoption willingness among medical ethics reviewpractitioners.

The Technological dimension includes performance expectancy and perceived risk, reflecting the perceived utility and security of the AI system, which are fundamental to adoption decisions in high-stakes ethics review settings.⁴⁵

The Organizational dimension comprises social influence and facilitating conditions to capture institutional and normative context. These factors determine whether practitioners feel supported and expected to adopt AI, thereby reducing barriers to use.⁴⁹

Finally, the Environmental dimension is adapted as an Individual layer to capture personal traits, specifically technology anxiety and personal innovativeness, including these affective and dispositional factors is critical, as prior research shows they significantly influence technology acceptance beyond cognitive evaluations.⁵⁰

This theoretical integration aligns with established precedents in adoption research. Through configurational analysis, this study aims to identify multiple causal pathways leading to high adoption willingness and to thoroughly examine the key moderating role of ethics review practitioners’ positions, thereby systematically elucidating the differentiated adoption logic across different roles.

Methods

Method selection

This study adheres to the STROBE guidelines for cross-sectional research. The methodology integrates a questionnaire survey method⁵¹ and fsQCA.⁵² By combining qualitative research with quantitative rigor, fsQCA effectively examines complexity in small to medium-sized samples. It identifies multiple conjunctural causation, examining how various antecedent conditions form distinct pathways. It also addresses equifinality, where different condition combinations lead to the same high willingness. Furthermore, fsQCA accounts for causal asymmetry, as conditions fostering adoption may differ fundamentally from those hindering it. These features align with the objective of exploring how configurational conditions influence adoption willingness.

To support the fsQCA, this study employed a questionnaire survey to collect structured data, providing a systematic quantitative foundation for the configurational analysis. The use of scales based on UTAUT ensures theoretical continuity and explanatory power for the research findings.

Questionnaire design

The questionnaire scales were adapted from established UTAUT frameworks. To ensure content validity, a focus group of 11 experts from clinical medicine, artificial intelligence, and medical ethics refined the items through multiple deliberation rounds, about 30 minutes per person, three members of the research team were responsible for data encoding and organization. The expert selection criteria included: (a) at least five years of professional experience; (b) prior involvement in ethics review or AI-related research; and (c) willingness to participate in the two-round review process.

This qualitative phase identified six core themes: performance expectancy (efficiency, accuracy); perceived risk (privacy, bias, accountability); social influence (leadership and peer endorsement); facilitating conditions (infrastructure and training); technology anxiety (fear of displacement); and personal innovativeness (intrinsic willingness to explore). Revisions transformed generic items into statements reflecting specific ethics review scenarios. Before the formal survey, a pilot study with 8 practitioners assessed the clarity and reliability of the questionnaire. Based on these results, minor wording adjustments were made without deleting any items.

The revision involved two stages. In the first round, experts evaluated the comprehensiveness of questionnaire dimensions and item wording. Experts specified the measurement of performance expectancy from improving work efficiency to accelerating the processing speed of ethics review materials. This stage resulted in a 26-item version. During the second round, experts assessed scale applicability. This led to the deletion of 5 scattered functional attitude items and the merging of 4 redundant items. Simultaneously, 5 key items regarding psychological and organizational environment variables. The final questionnaire consists of 30 items (Supplemental File 1). The reporting of this focus group adhered to the Consolidated Criteria for Reporting Qualitative Research (COREQ) checklist, provided in Supplemental File 2.

The questionnaire has two sections. The first collects demographic information, including gender, age, education, hospital level, and years of service. The second includes the main scales, covering six antecedent conditions (performance expectancy, perceived risk, social influence, facilitating conditions, technology anxiety, and personal innovativeness) and the outcome variable (adoption willingness). All items use a 5-point Likert scale (1 = strongly disagree; 3 = neutral; 5 = strongly agree).

Questionnaire survey and implementation

This cross-sectional survey used convenience sampling to distribute electronic questionnaires to RECs across various medical institutions in China in tertiary and secondary/primary hospitals across 11 provinces (including Beijing, Shanghai, Guangdong, Jiangsu, etc.) and municipalities from June 2025 to December 2025. The convenience sampling strategy was necessitated by the practical difficulty of accessing the entire population of REC practitioners in China, as there is no publicly available national registry. To mitigate potential selection bias, we invited practitioners from diverse hospital tiers and regions to mitigate selection bias. We acknowledge that self-selection by AI-interested practitioners may overestimate adoption willingness, a limitation further addressed in the Discussion. Inclusion criteria required active members or staff of a medical Research Ethics Committee, practical ethical review experience, and voluntary participation. The exclusion criteria are to exclude if they were incomplete, logically inconsistent, or completion time significantly below the average threshold.

All participants provided electronic informed consent before the study began. This study involves an anonymous online questionnaire approved by the Ethics Committee of China-Japan Friendship Hospital (Approval No.:[2024-KY-254]). Instead of providing physical signatures, participants reviewed an informed consent statement on the survey’s opening page. This statement detailed the study’s purpose, its voluntary nature, confidentiality protocols, and the right to withdraw. Respondents had to click “I have read and agree to participate' to access the questionnaire. Data collected were anonymized, encrypted, and strictly limited to academic research purposes.

Sample size consideration and data saturation

This study did not perform a sample size calculation based on conventional power analysis or the principle of data saturation as used in purely qualitative research. Instead, the target sample size was guided by methodological recommendations for fsQCA, which typically requires more than 50 cases to generate stable configurations. Out of 697 distributed surveys, 0 people refused to participate and 516 valid responses collected (195 from ethics committee members and 321 from ethics committee staff), yielding a valid response rate of 74.0%. This sample size meets fsQCA requirements. Our data support stable configurations identification and rigorous subgroup analysis. While formal saturation was not assessed, the current scale provides sufficient depth for configurational analysis. The results of the analysis are returned to the participants in a de identified state. Perform data cleaning and analysis using Excel 2026 and AMOS software. Participant demographics are summarized in Table 1.

Table 1.

Basic demographic information of respondents.

Characteristic	Ethics reviewer N=195, n (37.8%)	Assisting ethics reviewer N=321, n (62.2%)
Gender
Male	76(39.0%)	61(19.0%)
Female	119(61.0%)	260(81.0%)
Age Group (years)
≤30	18(9.2%)	50(15.6%)
30-39	43(22.1%)	124(38.6%)
40-49	62(31.8%)	107(33.3%)
≥50	72(36.9%)	40(12.5%)
Hospital Level
Tertiary hospital	177(90.8%)	290(90.3%)
Secondary & Primary hospital	18(9.2%)	31(9.7%)
Educational Level
Bachelor’s degree	90(46.2%)	138(43.0%)
Master’s degree	53(27.2%)	152(47.4%)
Doctoral degree	49(25.1%)	28(8.7%)
Other	3(1.5%)	3(0.9%)
Years of Service
≤3	94(48.2%)	142(44.2%)
4-6	50(25.6%)	90(28.0%)
7-10	29(14.9%)	54(16.8%)
≥11	22(11.3%)	35(10.9%)

Data calibration and analysis of reliability and validity

All variables demonstrated satisfactory reliability and validity, with Cronbach’s α > 0.7, composite reliability (CR) > 0.8, and both average variance extracted (AVE) and all factor loadings > 0.5. These results indicate that the scales effectively measured their respective constructs with good convergent validity. Additionally, the square root of each latent variable’s AVE exceeded its correlations with other variables, demonstrating adequate discriminant validity (Table 2). Thus, the measurement model is robust and suitable for further configurational analysis.

Table 2.

Reliability and effectiveness analysis.

Variable	Item	Factor loading	Cronbach’s α	KMO	CR	AVE
Performance Expectancy	1	0.899	0.924	0.836	0.929	0.767
	2	0.922
	3	0.905
	4	0.769
Perceived Risk	1	0.818	0.896	0.831	0.915	0.730
	2	0.888
	3	0.845
	4	0.864
Facilitating Conditions	1	0.779	0.872	0.818	0.856	0.598
	2	0.824
	3	0.778
	4	0.708
Social Influence	1	0.539	0.862	0.741	0.802	0.508
	2	0.803
	3	0.758
	4	0.723
Technology Anxiety	1	0.892	0.936	0.897	0.942	0.766
	2	0.814
	3	0.918
	4	0.847
	5	0.901
Personal innovativeness	1	0.812	0.872	0.810	0.820	0.534
	2	0.671
	3	0.762
	4	0.669
Adoption Willingness	1	0.881	0.966	0.753	0.919	0.792
	2	0.905
	3	0.883

Following the established fsQCA protocols,⁵³ we calibrated the raw data into fuzzy-set membership scores using three qualitative anchors based on the 5-point Likert scale: (1) Full membership threshold (95%, “strongly agree', coded as 5). This represents the point at which a case is fully in the set of a given condition; (2) Crossover point (50%, “neutral', (coded as 3). This is the point of maximum ambiguity, where a case is neither in nor out of the set; (3) Full non-membership threshold (5%, strongly disagree, coded as 1). This percentile-based approach provides a systematic and robust calibration for Likert data lacking external benchmarks.

The variables perceived risk and technology anxiety were reverse-calibrated so that higher scores consistently represent higher risk and anxiety. Furthermore, core conditions appeared in both the intermediate and parsimonious solutions, indicating a strong causal relationship, while peripheral conditions appeared only in the intermediate solutions.

Result

Necessary condition analysis

Before analyzing combined factors, we tested whether any single antecedent condition was necessary for high adoption willingness. A condition is considered necessary when its consistency score exceeds 0.9. According to the necessity analysis (Table 3), all consistency scores were below the 0.9 threshold in both the ethics committee members and staff. Performance expectancy yielded the highest consistency score at 0.840, which remains insufficient for necessity. This indicates that no single factor drives high adoption willingness, further supporting the adoption of a configurational approach to analyze the synergistic effects of multiple factors.

Table 3.

Necessity condition test.

Condition variable	Ethics committee members		Ethics committee staff
Condition variable	Consistency	Coverage	Consistency	Coverage
Performance Expectancy	0.840	0.866	0.831	0.875
∼Performance Expectancy	0.432	0.781	0.436	0.767
Perceived Risk	0.639	0.832	0.632	0.845
∼Perceived Risk	0.603	0.799	0.630	0.818
Facilitating Conditions	0.662	0.890	0.654	0.870
∼Facilitating Conditions	0.592	0.759	0.572	0.747
Social Influence	0.692	0.938	0.683	0.890
∼ Social Influence	0.563	0.716	0.542	0.721
Technology Anxiety	0.716	0.876	0.695	0.886
∼Technology Anxiety	0.566	0.802	0.597	0.813
Personal Innovativeness	0.727	0.962	0.750	0.966
∼ Personal Innovativeness	0.565	0.736	0.530	0.715

Adequacy analysis of configuration

Separate QCA analyses were conducted for both groups. Among the 64 logically possible configurations of the six variables, the ethics committee members sample contained 28 observed configurations with 36 logical remainders, while the ethics committee staff sample contained 26 observed configurations with 38 logical remainders. The case frequency threshold were set at 5 for ethics committee members and 4 for ethics committee staff. All solution configurations demonstrated consistency scores above 0.8 and PRI consistency scores above 0.75 (Table 4).

Table 4.

The perfected truth table.

Level	Conditional variable						Number	Outcome	Raw consist.	PRI consist.	SYM consist
Level	A	B	C	D	E	F	Number	Y	Raw consist.	PRI consist.	SYM consist
Ethics Committee Members	1	0	1	1	0	1	14	1	0.990488	0.977455	0.977456
	1	0	1	1	1	1	17	1	0.989494	0.972657	0.972658
	1	1	1	1	1	1	28	1	0.984818	0.966188	0.966188
	1	1	0	1	1	1	9	1	0.987762	0.96197	0.961971
	1	0	0	1	1	1	4	1	0.988011	0.959669	0.959669
	1	0	0	1	0	1	4	1	0.987605	0.947851	0.947851
	1	1	1	1	0	1	7	1	0.985828	0.946107	0.946107
	1	1	0	1	0	1	5	1	0.986735	0.945932	0.945933
	1	1	0	1	1	0	3	1	0.980528	0.922068	0.922069
	1	1	0	0	1	1	9	1	0.966498	0.884269	0.884269
	1	0	0	0	1	1	3	1	0.969519	0.876774	0.876774
	1	1	1	0	1	1	4	1	0.969357	0.874122	0.874121
	1	1	1	1	0	0	4	1	0.974365	0.869114	0.869114
	1	1	1	1	1	0	6	1	0.967126	0.868663	0.893443
	1	0	1	1	1	0	3	1	0.967186	0.854012	0.877935
	1	1	0	0	0	1	4	1	0.96287	0.850295	0.850295
	1	0	1	1	0	0	3	1	0.960451	0.783041	0.79935
	1	1	1	0	1	0	3	1	0.952266	0.77605	0.796235
	0	0	0	0	0	1	3	1	0.950468	0.767882	0.7822
	1	1	0	0	0	0	6	0	0.935879	0.713007	0.721694
	1	0	0	0	1	0	4	0	0.929881	0.70782	0.730832
	1	0	1	0	0	0	3	0	0.943049	0.650794	0.650794
	1	0	0	0	0	0	7	0	0.916395	0.612732	0.625136
	0	1	0	0	1	0	4	0	0.907088	0.561457	0.570167
	0	1	0	0	0	0	3	0	0.87581	0.450173	0.450173
	0	0	0	0	1	0	9	0	0.867166	0.44156	0.476452
	0	0	0	0	0	0	7	0	0.832192	0.335028	0.335029
Ethics Committee Staff	1	0	1	1	0	1	30	1	0.991669	0.981936	0.981936
	1	0	1	1	1	1	17	1	0.992742	0.981226	0.981225
	1	1	1	1	1	1	44	1	0.990535	0.978991	0.98487
	1	1	1	1	0	1	13	1	0.992832	0.977759	0.978904
	1	1	1	0	1	1	5	1	0.993415	0.976361	0.976361
	1	1	0	1	1	1	9	1	0.991701	0.974139	0.974138
	1	0	0	1	1	1	9	1	0.990525	0.972295	0.972295
	1	1	0	0	1	1	14	1	0.990289	0.965889	0.965889
	1	0	0	0	1	1	11	1	0.984171	0.946928	0.946928
	1	0	0	0	0	1	8	1	0.982787	0.92311	0.934906
	1	0	1	0	0	1	6	1	0.985088	0.922135	0.922136
	1	1	1	0	1	0	5	1	0.972816	0.84881	0.84881
	1	1	0	0	1	0	13	1	0.959841	0.795934	0.802858
	1	1	1	1	1	0	13	1	0.954753	0.79115	0.827101
	1	0	0	0	1	0	6	1	0.952753	0.790689	0.796733
	0	0	0	0	1	1	7	1	0.965259	0.779747	0.797942
	0	0	0	0	0	1	5	1	0.968806	0.770209	0.814471
	1	1	0	0	0	0	7	0	0.947763	0.703927	0.72298
	1	0	1	1	0	0	6	0	0.935576	0.690947	0.711568
	1	0	0	0	0	0	5	0	0.935134	0.689109	0.695299
	0	1	0	0	1	0	5	0	0.932326	0.587178	0.601263
	0	0	0	0	1	0	5	0	0.918017	0.49919	0.506847
	0	1	0	0	0	0	9	0	0.899122	0.405042	0.414608
	0	0	0	0	0	0	13	0	0.87811	0.370963	0.405197

Upon applying standard analysis to the enhanced truth table, three types of solutions were derived: complex, intermediate, and parsimonious. The intermediate solution incorporated counterfactual analysis, assuming that performance expectancy could enhance AI adoption willingness, while the presence or absence of other conditions (perceived risk, social influence, facilitating conditions, technology anxiety, and personal innovativeness) could also contribute to high adoption willingness. By comparing intermediate and parsimonious solutions, core and peripheral conditions were identified for each configuration. As shown in Table 5, performance expectancy acted as a core condition in most paths. In contrast, personal innovativeness frequently functions as a core condition in the paths for ethics committee staff, highlighting the fundamental differences in adoption logic between the two roles.

Table 5.

Configuration analysis of high willingness to adopt AI ethics review agent among different ethical reviewers.

Conditions	Ethics committee members							Ethics committee staff
	S1			S2		S3		N1				N2
	S1a	S1b	S2a	S2b	S2c	S3a	S3b	N1a	N1b	N1c	N1d	N2a	N2b	N2c	N2d
Performance Expectancy	•	•	•	•	◦	●	●		•	•	•	●	•	●	●
Perceived Risk			•		◦	●	●	◦		◦	◦		•
Facilitating Conditions	•		◦	◦	◦	●		◦	•		•	◦	•	◦
Social Influence	●	●			◦		●	◦	•	◦		◦			•
Technology Anxiety				•	◦	●	●			◦	◦	●	•	●	●
Personal innovativeness		●	●	●	●			●	●	●	●			●	●
Consistency	0.965	0.985	0.970	0.972	0.950	0.953	0.976	0.957	0.991	0.984	0.988	0.955	0.955	0.987	0.991
Raw coverage	0.530	0.560	0.348	0.356	0.214	0.399	0.401	0.300	0.479	0.244	0.305	0.361	0.369	0.338	0.390
Unique coverage	0.026	0.022	0.005	0.007	0.020	0.025	0.011	0.033	0.043	0.000	0.000	0.038	0.025	0.006	0.005
Solution consistency				0.948							0.951
Solution coverage				0.744							0.753

●Represents the existence of core causal conditions, •Representing the existence of marginal causal conditions, ◦Representing the absence of causal conditions at the edge, ‘blank’ indicates whether the condition can exist or not in the configuration.

Ethics committee members configuration

The analysis of the ethics committee members sample identified seven distinct pathways leading to high adoption willingness. These were categorized into three configurations: Social Influence-Driven, Personal innovativeness-Driven, and Organization-Technology Synergy-Driven.

(1) Social Influence-Driven Mode. Sub-modes S1a and S1b both feature social influence as a core condition, forming a second-order equifinal configuration. S1a is supplemented by high performance expectancy and high facilitating conditions, while S1b integrates social influence with personal innovativeness as core drivers, supported by high performance expectancy as a peripheral condition. This mode indicates that professional environments favoring AI usage enhance members’ recognition of the technology’s value. Specifically, the synergy between social influence and either technical facilitation or individual innovativeness significantly reinforces adoption willingness.

(2) Personal innovativeness-Driven Mode. Sub-modes S2a, S2b, and S2c all feature personal innovativeness as the core condition, forming a third-order equifinal configuration. S2a/S2b show that even under constraints such as high perceived risk, high technology anxiety, or insufficient external support, the combination of high performance expectancy and this intrinsic trait can still lead to high adoption willingness. Notably, S2c reveals that adoption willingness can be stimulated almost independently of traditional external motivators (e.g., performance gains, social pressure, organizational support) or cognitive-emotional factors like risk and anxiety. Here, adoption willingness stems primarily from a desire to explore new technology, highlighting the decisive role of personal innovativeness in the adoption decisions of ethics committee members.

(3) Organization-Technology Synergy-Driven Mode. Sub-modes S3a and S3b both share high perceived risk, high performance expectancy, and high technology anxiety as their core conditions. S3a incorporates facilitating conditions as an additional core element, representing a rationality-support pathway. In this configuration, ethics committee members exhibit high adoption willingness when clear performance expectations and risk awareness are coupled with strong organizational support. In contrast, S3b identifies social influence as a core condition, indicating that when organizational resources are absent, strong social influence can serve as an effective substitute, also leading to high adoption willingness.

Ethics committee staff configuration

The analysis of ethics committee staff identified eight pathways that lead to high adoption willingness, which can be grouped into two patterns: Personal innovativeness-Driven and Organization-Individual Innovativeness Synergy-Driven.

(1) Personal innovativeness-Driven Mode. This mode includes four pathways (N1a, N1b, N1c, N1d), all featuring high personal innovativeness as the core condition. Sub-mode N1a demonstrates that even in an environments characterized by low facilitating conditions and low social influence, high adoption willingness can be driven primarily by intrinsic motivation and low perceived risk, reflecting highly internalized behavioral characteristics. The other three sub-modes (N1b, N1c, N1d) are all supplemented by high performance expectancy, forming a dual core -driven foundation. Specifically, N1b additionally relies on high social influence and high facilitating conditions; N1c and N1d share low perceived risk and low technology anxiety, which eliminates internal psychological barriers, with N1d further supported by high facilitating conditions. The emergence of this mode can be attributed to personal innovativeness that provides the internal foundation for adoption, while high performance expectancy clarifies the external value of the behavior. Their combination with supporting conditions creates a synergistic effect that facilitates technology adoption.

(2) Organization-Individual innovativeness Synergy-Driven Mode. This mode comprises four pathways (N2a, N2b, N2c, N2d). Sub-mode N2a features performance expectancy and technology anxiety as core conditions, supported by low facilitating conditions and low social influence. This demonstrates that the pursuit of performance benefits can overcome the negative impact of technology anxiety even in low-support environments. Sub-mode N2b is driven by high performance expectancy, high perceived risk, high facilitating conditions, and high technology anxiety, where high facilitating conditions and high performance expectancy become key factors balancing perceived risk and technology anxiety. Sub-modes N2c and N2d introduce personal innovativeness as an additional core condition, forming a dual driver with performance expectancy to counteract technology anxiety. The difference is that N2d receives additional support from high social influence, which partially buffers the pressure. Overall, this mode reveals that with synergy between organizational support and individual traits, strong performance expectancy or personal innovativeness can still dominate and facilitate adoption behavior even amid anxiety and risk pressures.

Robustness test

We performed robustness tests by adjusting the consistency and frequency thresholds. First, the raw consistency threshold was raised from 0.80 to 0.85, while the PRI consistency threshold from 0.7 to 0.75. This more stringent criterion reduces the risk of including configurations based on limited empirical evidence. The resulting solutions were largely identical to the original. Second, the frequency threshold was raised to 6 for all analyses. The results showed highly stable configuration patterns in both groups.

Furthermore, an alternative calibration using the 90th, 50th, and 10th percentiles confirmed the initial findings, with only minor variations in coverage scores (e.g., solution coverage changed by less than 0.02). Overall solution consistency remained above the 0.90 threshold, with the number of driving paths and the distribution of core and peripheral conditions were consistent with the original results These findings indicate that the research conclusions are robust.

Discussion

Theoretical value: Constructing an AI adoption willingness model for medical ethics review practitioners

This study moves beyond the conventional research domain of “what factors influence' AI adoption to address the more pivotal and complex question: in the high-stakes context of medical ethics review, “how do different factors combine' to drive adoption among distinct professional groups? Based on this approach, the study achieves three key theoretical advancements.

First, this study extends existing theoretical boundaries by integrating the UTAUT and TOE frameworks. We develop a specialized AI adoption model for medical ethics review practitioners, incorporating performance expectancy, perceived risk, social influence, facilitating conditions, technology anxiety, and personal innovativeness. By systematically identifies key drivers, this approach provides a robust analytical framework for emerging technology governance. This integrated framework complements existing literature by adding context-specific constructs essential for high-stakes ethics review.²⁹

Second, this study adopts a configurational perspective using fsQCA to introduce a necessary paradigm shift beyond traditional linear models. The results demonstrate that high adoption willingness arises from multiple equifinal pathways rather than a single necessary condition. This approach challenges one-size-fits-all explanations and clarifies the causal complexity inherent in technology adoption. For instance, while prior UTAUT-based research emphasizes the main effects of performance expectancy or social influence,⁵⁴ our findings show that these factors operate within interdependent configurations. Our equifinality finding converges with recent fsQCA studies in health technology adoption,³⁶ but diverges from main-effect models common in UTAUT literature.

Third, this study elucidates role heterogeneity by differentiating ethics review practitioners into committee members and staff. The identification of seven pathways for members and eight for staff refutes the homogeneity assumption prevalent in adoption research. Specifically, adoption among committee members is primarily organizationally-mediated, relying on social influence and facilitating conditions to mitigate professional and legal liabilities. Conversely, committee staff adoption is more individually-driven, with personal innovativeness serving as the primary determinant. This divergence confirms that adoption logic is contingent upon professional roles, providing a granular model for adoption behavior.This finding aligns with prior research on role-based differences between clinicians and administrators. Furthermore, this study extends that logic to AI in ethics review by demonstrating qualitative configurational differences rather than mere mean-level variations.⁵⁵

Additionally, by applying technology adoption theory to AI ethics governance, this study explains professional concerns regarding emerging technologies. Findings suggest that AI systems should function as assistive tools rather than decision-making agents. Consequently, AI outputs should serve as references for expert discussion instead of final review opinions. This approach maintains human subjectivity in ethics review and offers a framework for localized human-AI collaboration, particularly in similar institutional settings, such positioning adheres to established human-in-the-loop governance recommendations.⁵⁶

Practical value: Construction of differentiated strategies for system implementation

The primary practical value of this research lies in its ability to provide medical institutions with an actionable, group-specific roadmap for successfully implementing AI-assisted review systems.

(1) Organizational Level: Implementing Differentiated Support Systems. Medical institutions must establish robust organizational safeguards for ethics committees.⁵⁷ This includes creating a top-down atmosphere that encourages active AI integration by providing necessary hardware, funding, and technical infrastructure.A manual override mechanism is essential for high-risk research, ensuring that human reviewers lead critical evaluations.⁵⁸ Provide AI literacy training to help reviewers critically evaluate AI suggestions and identify potential biases.

Conversely, for ethics committee staff, the focus should be on both reducing pressure and enabling capability.⁵⁹ Management can streamline unnecessary administrative procedures, grant necessary system operation permissions, and protect their innovative thinking. Standardizing AI-augmented pre-review protocols for low-risk procedural tasks will further allow staff to focus on complex qualitative assessments while safeguarding individual professional autonomy.

(2) Technical Level: For ethics committee members, technical priorities must shift from simple deployment to building a trustworthy, efficient human-AI ecosystem.⁶⁰ We advocate for the implementation a comprehensive decision-audit trail allows ethics committee members to trace the rationale behind AI recommendations. When integrated with manual verification mechanisms, this approach clarifies black box, fosters critical engagement with AI outputs, and systematically builds institutional trust in AI-assisted review. Additionally, systems should develop automatic risk grading. By flagging high-risk protocol, these functions focus ethics committee member attention and complement professional judgment.⁶¹

For ethics committee staff, technical design must emphasize usability and operational efficiency. This requires optimizing user interfaces, system speed, and administrative workflows.⁶² System design and institutional communication must reinforce that AI as an augmentative tool intended to procedural burdens. This clear positioning addresses technological anxiety and burnout, ultimately fostering a sustainable, collaborative human-AI work environment.

(3) Individual Level: Implementing Targeted Strategies at the Individual Level to Stimulate Intrinsic Motivation and Alleviate Concerns. For ethics committee members, identify and leverage those with high personal innovativeness as role models to influence peer adoption AI.⁶³ For ethics committee staff, assign them the task of testing new features to make them feel valued. Furthermore, incentive mechanisms should be established to encourage early adopters and recognize AI integration. Crucially, establish reasonable error-tolerance mechanisms that clearly define within the “AI pre-review with human verification' workflow, human verifiers bear supervisory and supplementary responsibilities, not full liability for AI errors. This helps alleviate anxiety about accountability and protects their motivation to explore and use the technology.

Furthermore, enhancing AI transparency and explainability is essential across all groups to bolster performance expectancy and mitigate perceived risk.⁶⁴ Ultimately, this study offers insights that may be relevant to AI ethics governance frameworks in other contexts, by demonstrating one approach to implementing a localized human-AI collaboration model within medical institutions in China.

Research limitations

The sample primarily comprised medical ethics committees, potentially limiting the generalizability of findings to broader healthcare contexts. Future research should include ethics review bodies in academic and corporate settings to validate cross-organizational applicability. In addition, the research framework may have overlooked variables like organizational culture, while the fsQCA method offers limited micro-level insights. Furthermore, selection bias may also exist due to convenience sampling, as AI-interested practitioners were likely more inclined to participate. Future studies should utilize probabilistic sampling and qualitative case studies to develop more comprehensive models. Finally, although the sample size satisfies fsQCA requirements, the lack of systematic data saturation testing means further configurational pathways may exist. Subsequent research with larger, diverse cohorts is necessary to confirm these patterns among specific practitioner subgroups.

Conclusion

Based on the UTAUT and TOE framework, this study examined AI adoption willingness among 516 medical ethics review practitioners using fsQCA. The study found that no single necessary condition drives AI adoption willingness and revealed significantly different driving logics between ethics committee members and ethics committee staff.

The finding of causal complexity and equifinality, seven for ethics committee members and eight for staff. Members primarily rely on organizational and social support, categorized into Social Influence, Personal Innovativeness, and Organization-Technology Synergy modes. In contrast, staff pathways depend more heavily on personal innovativeness, forming Core Personal Innovativeness and Organization-Individual Synergy modes. These findings challenge linear adoption models and highlight role-based heterogeneity in high-stakes professional settings.

Based on these findings, we recommend that medical management institutions implement differentiated AI promotion strategies For ethics committee members, strategies should strengthen institutional safeguards and organizational empowerment by improving regulations and adequate resources to build a trustworthy decision-support environment. For ethics committee staff, the focus should be on stimulating individual potential and optimizing user experience, transforming intrinsic innovativeness into proactive adoption behavior. Future research should validate the configurational across different cultures, conduct longitudinal studies to track evolving trust, and employ qualitative case to explore organizational culture and personal innovativeness.

Ultimately, this research argues that in critical domains like ethics review, he future of AI may depend less on universal deployment than on contextually intelligent integration. By acknowledging the complex, role-specific and multifaceted pathways to adoption, institutions can advance human-AI collaboration,thereby enhancing the efficiency of ethical oversight while safeguarding its foundational integrity.

Supplemental material

Supplemental material - Adopting AI in medical ethics review: A configurational fsQCA study of practitioners’ willingness

Supplemental material for Adopting AI in medical ethics review: A configurational fsQCA study of practitioners’ willingness by Aiyi Zhang, Aijuan Sheng, Hu Chen, Hu Ning, Rui Li, Zhongguang Yu in DIGITAL HEALTH

Footnotes

Acknowledgements

The authors sincerely thank all the medical ethics committee members and staff who participated in this study for sharing their valuable time and insights.

ORCID iD

Aiyi Zhang

Ethical considerations

The study was approved by the Ethics Committee of China-Japan Friendship Hospital (Approval No: [2024-KY-254]). Data collected were anonymized, encrypted, and strictly limited to academic research purposes.

Consent to participate

Electronic informed consent was obtained from all individual participants included in the study. Participants were informed about the study purpose, voluntary participation, confidentiality, and right to withdraw before accessing the online questionnaire. Consent was documented through a mandatory click-to-agree mechanism on the survey’s opening page.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Natural Science Foundation of China project (72104255), Chinese Academy of Medical Sciences (CAMS) Innovation Fund for Medical Sciences (Grant No. 2021-I2M-1-046), the High-Level Project Fund of China–Japan Friendship Hospital (Grant No. 2024-NHLHCRF-GL-12), and the National Health Commission Hospital Management Institute Artificial Intelligence Special Project (Grant No. YLXX24AIF001).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Contributorship

The research team comprised two males and four females. Aiyi Zhang: Writing – original draft, Writing –Review & Editing, Formal Analysis; Aijuan Sheng: Investigation; Hu Chen: Supervision; Ning Hu: Validation; Rui Li: Data curation; Yu: Conceptualization, Funding acquisition, Supervision.

Guarantor

Zhongguang Yu.

Supplemental material

Supplemental material for this article is available online.

References

Al Kuwaiti

Nazer

Al-Reedy

, et al. A Review of the Role of Artificial Intelligence in Healthcare. J Pers Med 2023; 13(6): 951. https://doi.org/10.3390/jpm13060951.

Fournier-Tombs

McHardy

. A Medical Ethics Framework for Conversational Artificial Intelligence. J Med Internet Res 2023; 25: e43068. https://doi.org/10.2196/43068.

Ding

Yin

Fang

, et al. A narrative review on the development of the ethical review mode of multicenter clinical trials in China. Ann Transl Med 2022; 10(22): 1251. https://doi.org/10.21037/atm-22-5213.

Yang

Liu

, et al. Development of an evaluation indicator system for the quality of science and technology ethics review committees: a Chinese study based on the structure-process-outcome framework. BMC Med Ethics 2026; 27(1): 44. https://doi.org/10.1186/s12910-026-01392-9.

Fukataki

Hayashi

Nishimoto

, et al. Developing artificial intelligence tools for institutional review board pre-review: A pilot study on ChatGPT's accuracy and reproducibility. PLOS Digit Health 2025; 4(6): e0000695. https://doi.org/10.1371/journal.pdig.0000695.

Vardas

Marketou

Vardas

. Medicine, healthcare and the AI act: gaps, challenges and future implications. Eur Heart J Digit Health 2025; 6(4): 833–839. https://doi.org/10.1093/ehjdh/ztaf041.

Elgin

. Ethical implications of AI-driven clinical decision support systems on healthcare resource allocation: a qualitative study of healthcare professionals' perspectives. BMC Med Ethics 2024; 25(1): 148. https://doi.org/10.1186/s12910-024-01151-8.

Gumus

Alan

. Perspectives of physicians, nurses, and patients on the use of artificial intelligence and robotic nurses in healthcare. Int Nurs Rev 2025; 72(2): e70017. https://doi.org/10.1111/inr.70017.

Freyer

Groß

Lipprandt

. The ethical requirement of explainability for AI-DSS in healthcare: a systematic review of reasons. BMC Med Ethics 2024; 25(1): 104. https://doi.org/10.1186/s12910-024-01103-2.

10.

Gichoya

Thomas

Celi

, et al. AI pitfalls and what not to do: mitigating bias in AI. Br J Radiol 2023; 96(1150): 20230023. https://doi.org/10.1259/bjr.20230023.

11.

Wang

Marivate

, et al. On the transparency of large AI models. Patterns (N Y) 2023; 4(7): 100797. https://doi.org/10.1016/j.patter.2023.100797.

12.

Guan

Dong

Zhao

. Ethical Risk Factors and Mechanisms in Artificial Intelligence Decision Making. Behav Sci (Basel) 2022; 12(9): 343. https://doi.org/10.3390/bs12090343.

13.

Ismatullaev

UVU

Kim

. Review of the Factors Affecting Acceptance of AI-Infused Systems. Hum Factors 2024; 66(1): 126–144. https://doi.org/10.1177/00187208211064707.

14.

Agafonov

Babic

Sousa

, et al. Editorial: Trustworthy AI for healthcare. Front Digit Health 2024; 6: 1427233. https://doi.org/10.3389/fdgth.2024.1427233.

15.

Kowald

Scher

Pammer-Schindler

, et al. Establishing and evaluating trustworthy AI: overview and research challenges. Front Big Data 2024; 7: 1467222. https://doi.org/10.3389/fdata.2024.1467222.

16.

Drabiak

Kyzer

Nemov

, et al. AI and machine learning ethics, law, diversity, and global impact. Br J Radiol 2023; 96(1150): 20220934. https://doi.org/10.1259/bjr.20220934.

17.

Zhang

. Ethics and governance of trustworthy medical artificial intelligence. BMC Med Inform Decis Mak 2023; 23(1): 7. https://doi.org/10.1186/s12911-023-02103-9.

18.

Kong

. A study on the normative path of ethics review in China: based on the perspective of Panopticism. Front Med (Lausanne) 2023; 10: 1268046. https://doi.org/10.3389/fmed.2023.1268046.

19.

Ansari

. Supporting a Member Secretary of an Ethics Committee. Asian Bioeth Rev 2023; 15(4): 361–363. https://doi.org/10.1007/s41649-023-00255-4.

20.

Schuppli

Fraser

. Factors influencing the effectiveness of research ethics committees. J Med Ethics 2007; 33(5): 294–301. https://doi.org/10.1136/jme.2005.015057.

21.

Mohamad

Azlan

, et al. Utilization of artificial intelligence in clinical practice: A systematic review of China's experiences. Digit Health 2025; 11: 20552076251343752. https://doi.org/10.1177/20552076251343752.

22.

Abdekhoda

Dehnad

Zarei

. Factors influencing adoption of e-learning in healthcare: integration of UTAUT and TTF model. BMC Med Inform Decis Mak 2022; 22(1): 327. https://doi.org/10.1186/s12911-022-02060-9.

23.

Wulf

Lewthwaite

. Optimizing performance through intrinsic motivation and attention for learning: The OPTIMAL theory of motor learning. Psychon Bull Rev 2016; 23(5): 1382–1414. https://doi.org/10.3758/s13423-015-0999-9.

24.

Al-Rahmi

Shamsuddin

Wahab

, et al. Integrating the Role of UTAUT and TTF Model to Evaluate Social Media Use for Teaching and Learning in Higher Education. Front Public Health 2022; 10: 905968. https://doi.org/10.3389/fpubh.2022.905968.

25.

Thanthrige

Sako

, et al. Determinants of Health Care Technology Adoption Using an Integrated Unified Theory of Acceptance and Use of Technology and Task Technology Fit Model: Systematic Review and Meta-Analysis. J Med Internet Res 2025; 27: e64524. https://doi.org/10.2196/64524.

26.

Wang

Liu

, et al. Investigating the factors influencing users' adoption of artificial intelligence health assistants based on an extended UTAUT model. Sci Rep 2025; 15(1): 18215. https://doi.org/10.1038/s41598-025-01897-0.

27.

Hoque

Sorwar

. Understanding factors influencing the adoption of mHealth by the elderly: An extension of the UTAUT model. Int J Med Inform 2017; 101: 75–84. https://doi.org/10.1016/j.ijmedinf.2017.02.002.

28.

Dai

Yang

, et al. Attitudes, Perceptions, and Factors Influencing the Adoption of AI in Health Care Among Medical Staff: Nationwide Cross-Sectional Survey Study. J Med Internet Res 2025; 27: e75343. https://doi.org/10.2196/75343.

29.

Dingel

Kleine

Cecil

, et al. Predictors of Health Care Practitioners' Intention to Use AI-Enabled Clinical Decision Support Systems: Meta-Analysis Based on the Unified Theory of Acceptance and Use of Technology. J Med Internet Res 2024; 26: e57224. https://doi.org/10.2196/57224.

30.

Duan

Liu

Rong

, et al. Integrating AI in medical education: a comprehensive study of medical students' attitudes, concerns, and behavioral intentions. BMC Med Educ 2025; 25(1): 599. https://doi.org/10.1186/s12909-025-07177-9.

31.

Jindal

Lungren

Shah

. Ensuring useful adoption of generative artificial intelligence in healthcare. J Am Med Inform Assoc 2024; 31(6): 1441–1444. https://doi.org/10.1093/jamia/ocae043.

32.

Cao

Liu

. Artificial intelligence-assisted psychosis risk screening in adolescents: Practices and challenges. World J Psychiatry 2022; 12(10): 1287–1297. https://doi.org/10.5498/wjp.v12.i10.1287.

33.

Herrmann

. What's next for responsible artificial intelligence: a way forward through responsible innovation. Heliyon 2023; 9(3): e14379. https://doi.org/10.1016/j.heliyon.2023.e14379.

34.

Lambert

Madi

Sopka

, et al. An integrative review on the acceptance of artificial intelligence among healthcare professionals in hospitals. NPJ Digit Med 2023; 6(1): 111. https://doi.org/10.1038/s41746-023-00852-5.

35.

Alqurni

. Sustainable adoption of artificial intelligence and the Metaverse in higher education: an environmental, social, and governance-based analysis of pedagogical innovation and perceived student learning outcomes. Front Artif Intell 2026; 9: 1738730. https://doi.org/10.3389/frai.2026.1738730.

36.

Zhao

Zhang

, et al. Current status and influencing factors of telehealth readiness among nurses: A multicenter cross-sectional study in Ningxia, China. PLoS One 2026; 21(2): e0343263. https://doi.org/10.1371/journal.pone.0343263.

37.

Cornelissen

Egher

van Beek

, et al. The Drivers of Acceptance of Artificial Intelligence-Powered Care Pathways Among Medical Professionals: Web-Based Survey Study. JMIR Form Res 2022; 6(6): e33368. https://doi.org/10.2196/33368.

38.

Ibrahim

Münscher

Daseking

, et al. The technology acceptance model and adopter type analysis in the context of artificial intelligence. Front Artif Intell 2024; 7: 1496518. https://doi.org/10.3389/frai.2024.1496518.

39.

Kottler

. Artificial Intelligence: A Private Practice Perspective. J Am Coll Radiol 2020; 17(11): 1398–1404. https://doi.org/10.1016/j.jacr.2020.09.029.

40.

Safi

Danzer

Schmailzl

. Empirical Research on Acceptance of Digital Technologies in Medicine Among Patients and Healthy Users: Questionnaire Study. JMIR Hum Factors 2019; 6(4): e13472. https://doi.org/10.2196/13472.

41.

Zhang

Luo

Nie

, et al. Technical attributes, health attribute, consumer attributes and their roles in adoption intention of healthcare wearable technology. Int J Med Inform 2017; 108: 97–109. https://doi.org/10.1016/j.ijmedinf.2017.09.016.

42.

Menon

Shilpa

. Chatting with ChatGPT: Analyzing the factors influencing users' intention to Use the Open AI's ChatGPT using the UTAUT model. Heliyon 2023; 9(11): e20962. https://doi.org/10.1016/j.heliyon.2023.e20962.

43.

Choudhury

Shamszare

. The Impact of Performance Expectancy, Workload, Risk, and Satisfaction on Trust in ChatGPT: Cross-Sectional Survey Analysis. JMIR Hum Factors 2024; 11: e55399. https://doi.org/10.2196/55399.

44.

Brouwer

Barry

de Groot

. The dual-role clinician teacher: A human bridge for socially responsible AI use. Med Educ 2025; 59(12): 1266–1268. https://doi.org/10.1111/medu.70035.

45.

Zhao

, et al. The Willingness of Doctors to Adopt Artificial Intelligence-Driven Clinical Decision Support Systems at Different Hospitals in China: Fuzzy Set Qualitative Comparative Analysis of Survey Data. J Med Internet Res 2025; 27: e62768. https://doi.org/10.2196/62768.

46.

Vijayakumar

Lee

Leong

, et al. Physicians' Perspectives on AI in Clinical Decision Support Systems: Interview Study of the CURATE.AI Personalized Dose Optimization Platform. JMIR Hum Factors 2023; 10: e48476. https://doi.org/10.2196/48476.

47.

Knop

Weber

Mueller

, et al. Human Factors and Technological Characteristics Influencing the Interaction of Medical Professionals With Artificial Intelligence-Enabled Clinical Decision Support Systems: Literature Review. JMIR Hum Factors 2022; 9(1): e28639. https://doi.org/10.2196/28639.

48.

PML

Lit

Cheung

CTY

. Remote work as a new normal? The technology-organization-environment (TOE) context. Technol Soc 2022; 70: 102022. https://doi.org/10.1016/j.techsoc.2022.102022.

49.

Jacob

Sanchez-Vazquez

Ivory

. Understanding Clinicians' Adoption of Mobile Health Tools: A Qualitative Review of the Most Used Frameworks. JMIR Mhealth Uhealth 2020; 8(7): e18072. https://doi.org/10.2196/18072.

50.

Allen

Towne

Jr Maxwell

, et al. Meausures of organizational characteristics associated with adoption and/or implementation of innovations: A systematic review. BMC Health Serv Res 2017; 17(1): 591. https://doi.org/10.1186/s12913-017-2459-x.

51.

Stone

. Design a questionnaire. Bmj 1993; 307(6914): 1264–1266. https://doi.org/10.1136/bmj.307.6914.1264.

52.

, et al. Study on the digital divide in the continuous utilization of Internet medical services for older adults: Combination with PLS-SEM and fsQCA analysis approach. Int J Equity Health 2025; 24(1): 71. https://doi.org/10.1186/s12939-025-02418-1.

53.

Wang

Zhang

, et al. The willingness to continue using wearable devices among the elderly: SEM and FsQCA analysis. BMC Med Inform Decis Mak 2023; 23(1): 218. https://doi.org/10.1186/s12911-023-02336-8.

54.

Cialdini

Goldstein

. Social influence: compliance and conformity. Annu Rev Psychol 2004; 55: 591–621. https://doi.org/10.1146/annurev.psych.55.090902.142015.

55.

Hsieh

Bali

. Untangling pathways from digital instructional leadership to AI pedagogy: a diffusion of innovation theory perspective. BMC Psychol 2025; 14(1): 63. https://doi.org/10.1186/s40359-025-03792-9.

56.

Mohsin Khan

Shah

Shaikh

, et al. Towards secure and trusted AI in healthcare: A systematic review of emerging innovations and ethical challenges. Int J Med Inform 2025; 195: 105780. https://doi.org/10.1016/j.ijmedinf.2024.105780.

57.

Goktas

Grzybowski

. Shaping the Future of Healthcare: Ethical Clinical Challenges and Pathways to Trustworthy AI. J Clin Med 2025; 14(5): 1605. https://doi.org/10.3390/jcm14051605.

58.

Zhu

. AI ethics with Chinese characteristics? Concerns and preferred solutions in Chinese academia. AI Soc 2022; 39: 1–14. https://doi.org/10.1007/s00146-022-01578-w.

59.

Grady

. Institutional Review Boards: Purpose and Challenges. Chest 2015; 148(5): 1148–1155. https://doi.org/10.1378/chest.15-0706.

60.

Gao

Chen

Wang

, et al. Optimising the paradigms of human AI collaborative clinical coding. NPJ Digit Med 2024; 7(1): 368. https://doi.org/10.1038/s41746-024-01363-7.

61.

Sedeeq

Arman

. Human-Centric AI Governance: An Adaptive Public International Law Framework for Ethical and Inclusive AI Regulation in Public Health. J Law Med Ethics 2025; 53(4): 563–574. https://doi.org/10.1017/jme.2025.10175.

62.

David

Choung

Seberger

. Who is responsible? US Public perceptions of AI governance through the lenses of trust and ethics. Public Underst Sci 2024; 33(5): 654–672. https://doi.org/10.1177/09636625231224592.

63.

Funer

Schneider

Heyen

, et al. Impacts of Clinical Decision Support Systems on the Relationship, Communication, and Shared Decision-Making Between Health Care Professionals and Patients: Multistakeholder Interview Study. J Med Internet Res 2024; 26: e55717. https://doi.org/10.2196/55717.

64.

Reddy

. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. Implement Sci 2024; 19(1): 27. https://doi.org/10.1186/s13012-024-01357-9.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.88 MB

0.00 MB