AI knowledge, attitudes, perceptions, and willingness to use AI among oncologists in China: a nationwide cross-sectional study

Abstract

Background

The integration of artificial intelligence (AI) into healthcare, particularly in oncology, has attracted considerable attention in China. Understanding oncologists’ views on AI is essential for developing effective integration strategies in oncology settings. This study aimed to evaluate AI knowledge, attitudes, perceptions, and willingness among nationwide oncologists in China. Additionally, we identified factors influencing oncologists’ willingness to use AI.

Methods

A nationwide cross-sectional survey of 1,538 oncologists from 31 regions in China was conducted between August and November 2023 using a 22-item online questionnaire, launched on the Wenjuan online platform and shared by members of Chinese Anti-Cancer Association via WeChat.

Results

The survey achieved an 87.78% completion rate (1,350/1,538). Among the respondents, 90% had used medical AI products, and 14.59% had used ChatGPT. Respondents demonstrated different levels of understanding of AI terms (mean = 3.524), limitations (mean = 3.256), trust (mean = 3.759), and acceptance (mean = 4.022), with scores ranging from 1 (low) to 5 (high) on 5-point Likert items. Perceptions of AI surpassing doctors’ diagnostic abilities had a mean score of 3.255, whereas perceptions of AI replacing doctors had a mean score of 2.829. Both perceptions, along with willingness to use AI (mean = 4.162) and support for its widespread use (mean = 4.061), were also measured on 5-point Likert scale. Factors associated with high willingness to use AI included senior clinical titles (OR 1.41 [95% CI: 1.05–1.90]), working in the radiation oncology department (OR 1.9 [95% CI: 1.29–2.81]), high understanding AI terms (OR 4.66 [95% CI: 3.46–6.34]), prior experience with ChatGPT (OR 2.28 [95% CI: 1.32–4.07]), and high scores on items such as “AI had already surpassed doctors’ diagnostic abilities” (OR 8.82 [95% CI: 1.82–2.56]) or “AI will replace doctors” (OR 13.32 [95% CI: 2.11–3.13]).

Conclusions

This study unveils a good understanding and positive attitudes toward AI, coupled with a strong willingness to use AI products among nationwide oncologists across China, despite ongoing controversy over whether AI will replace doctors. Those with senior titles or greater AI knowledge and experience show a high willingness to use AI. Considering the findings, targeted AI education and increased exposure are crucial for the successful integration of AI into oncology settings in China.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12909-026-08951-z.

Introduction

In recent years, China’s medical artificial intelligence (AI) market has grown rapidly [1–3]. By 2022, China ranked second in the global medical device market and leads the world in AI-related randomized controlled trials (RCTs) and the development of AI medical products [4–6]. In oncology, AI is extensively used, including methods such as deep learning (DL), machine learning (ML), and large language models (LLMs) [7]. DL extracts complex features from imaging and omics data, enabling more accurate cancer screening, diagnosis, and prognosis prediction [8–10]. Machine learning (ML) is crucial for cancer risk stratification, survival assessment, and optimizing personalized treatment plans, with the added benefit of interpretable models that aid clinical decision-making [11–13]. LLMs, a type of generative AI, are applied in oncology to analyze electronic medical records, generate treatment plans, and provide patient education [14–16]. Additionally, generative AI shows great potential in enhancing primary healthcare [17]. However, despite its progress in various fields, concerns about socio-demographic biases in AI models remain in healthcare. A study evaluated biases in LLMs’ clinical decision-making, showing that marginalized groups, such as Black, unhoused, and LGBTQIA+ individuals, were more likely to receive biased recommendations [18]. Another study provided a comprehensive survey of bias evaluation and mitigation techniques for LLMs, offering insights into how to identify and address these biases through various strategies [19]. As AI advances rapidly in oncology, oncologists are leveraging these technologies to enhance their core competencies, while facing the challenge of ensuring fair and equitable healthcare for patients. Therefore, understanding oncologists’ perspectives on AI is essential for developing strategies to integrate AI into oncology.

The adoption of AI in healthcare is met with both optimism and skepticism worldwide. In Poland, general medicine physicians are enthusiastic about AI integration but cautious about its practical applications, especially in radiology and pathology [20]. In Jordan, medical and pathology students recognized AI’s potential in healthcare but expressed concerns about the lack of formal training and its ethical implications in clinical decisions. Most students agreed that AI should be included in medical curricula to bridge these gaps [21]. A survey of medical students across nine Arab countries revealed a significant knowledge gap in AI, though students were optimistic about its potential to revolutionize healthcare, particularly in radiology, and emphasized the need for AI integration into medical education [22]. Among U.S. oncologists, AI’s usefulness in clinical decision-making was acknowledged, but concerns about its reliability, ethics, and legal accountability were prominent [23]. Although some studies did not focus on oncologists, their findings provide general insights toward AI in healthcare, which are also relevant to oncology.

In China, a national survey highlighted the need for AI training in medical education, although concerns about AI’s potential biases and lack of ethical guidelines persist [24]. Another research revealed that, although Chinese medical students showed a strong interest in learning about AI, a significant knowledge gap remains [25]. Recent studies have shown that AI-based tools, such as customized GPT models, are increasingly seen as valuable in enhancing clinical knowledge acquisition [26]. Furthermore, a survey conducted among radiologists in southeastern China, highlighted their optimism about AI in medical imaging, while also pointing out significant gaps in AI education [27]. Despite these advancements, a survey on nationwide oncologists’ views of AI in China remains limited. This study aimed to (1) assess AI knowledge, attitudes, perceptions, and willingness to use AI among oncologists nationwide in China, and (2) identify factors influencing their willingness to use AI.

Methods

The Unified Theory of Acceptance and Use of Technology (UTAUT), proposed in 2003 [28], is widely used to analyze user behavior and intentions. This model has been applied to medical information technology [29, 30]. Drawing on the UTAUT framework, we developed a 22-item questionnaire tailored to assess Chinese oncologists’ knowledge, attitudes, perceptions, and willingness to adopt AI in healthcare. The development process included a pilot study, which involved interviews with 11 oncologists from various oncology specialties. Their feedback was critical in refining the questionnaire, ensuring content validity and enhancing its face validity and relevance to the target population. It should be noted that the “knowledge” referred to in the questionnaire reflects respondents’ perceptions of AI knowledge or familiarity rather than actual knowledge. Since the data were deidentified, the Ethical Review Committee of Chongqing University Cancer Hospital approved the study for exemption from ethical review. Digital informed consent was obtained on the first page of the questionnaire. Data were collected from August to November 2023 across 31 regions in China, ensuring that all responses were confidential and anonymous. The questionnaire was distributed online via the online platform and shared through the WeChat app by members of the Chinese Anti-Cancer Association (CACA) on Artificial Intelligence in Oncology. The target respondents were physicians with “Oncology” specialization on their Medical Practitioner Qualification Certificate or those working in oncology-related departments in China.

The first nine questions gathered demographic information, including gender, age, education level, clinical title, work experience, clinical department, university affiliation, hospital grade, and category. Resident doctors and attending physicians were categorized as junior titles, while associate chief physicians and chief physicians were considered senior titles. Respondents’ IP addresses were collected to determine their location, which was categorized into four economic zones of China: Northeast, East, Central, or West [31]. AI concept was assessed in question 10, while questions 11 and 12 focused on experience with AI products. Questions 13–16 used 5-point Likert scales to evaluate four dimension included AI knowledge (AI terms and limitations), attitudes (trust and acceptance), perceptions (“AI has surpassed the average diagnostic ability of doctors” and “AI will replace doctors”), and willingness to use AI. Scores of 1–2 indicated low levels, 3 indicated medium level, and 4–5 indicated high levels. Question 17 explored AI’s impact on the doctor-patient relationship, while questions 18–19 covered IT and ChatGPT experience. Here, ChatGPT is used as a representative example of AI models. Questions 20 and 21 examined potential positive impacts and concerns related to AI, and question 22 was open-ended.

Descriptive statistics were used to summarize the demographic data and questionnaire items. The internal consistency of the questionnaire was evaluated using Cronbach’s α, which showed good (0.6 to 0.8) to excellent (0.8 to 1) reliability across the four dimensions: AI knowledge, attitudes, perceptions, and willingness to use. Univariable logistic regression identified factors significantly associated with willingness to use AI, while Pearson correlation analysis assessed the relationships between the four dimensions (Supplementary Table 1). Due to the high correlation between the dimensions of AI knowledge, attitudes, perceptions, and willingness to use AI, individual items were included one by one as independent variables in the multivariable logistic regression to examine their associations with willingness to use AI. In the main analysis, ‘Understanding of AI-related terms’ from the knowledge dimension was used as the dependent variable, with a sensitivity analysis replacing it with ‘Awareness of AI limitations’ to create an alternative model. Additionally, sensitivity analyses were conducted for two items from the perceptions dimension of “AI has surpassed the average diagnostic ability of doctors” and “AI will replace doctors”. Each item is the dependent variable in separate multivariable logistic regression models. All P-values were two-sided, with P < 0.05 considered significant. Statistical analyses were performed using Decision Linnc software (V1.0.8.6) [32].

Results

Demographic characteristics

A total of 1,538 oncologists from 31 regions in China participated in the survey, yielding a 69% response rate (1,538/2,201) and an 87.8% completion rate (1,350/1,538). The participants represented 5.13% of the registered oncologists in China, according to the 2022 China Health Statistics Yearbook [33]. Of the respondents, 46% were female, and 54% were male. Age distribution was as follows: 26.74% aged 21–30 years, 38.67% aged 31–40 years, 23.63% aged 41–50 years, 10.22% aged 51–60 years, and 0.74% aged over 60 years. In terms of education, 30.81% held a bachelor’s degree, 46.07% held a master’s degree, and 23.11% held a doctoral degree. Half respondents had less than 10 years of work experience (50.81%). Respondents were mainly employed in medical oncology (32.3%), radiation oncology (25.4%), and surgical oncology (24.2%). Geographically, most respondents were from the Eastern region (40.5%), followed by the Western region (32.5%), Central region (14.8%), and Northeast region (12.2%). A majority (90%) had experience using medical AI products, while 20.9% were aware of ChatGPT, and 14.6% had used it (Table 1).

Table 1.

Demographic characteristics

Variable names	No. (%), N = 1350
Gender
Female	621 (46.00)
Male	729 (54.00)
Age
21 ~ 30 y	361 (26.74)
31 ~ 40 y	522 (38.67)
41 ~ 50 y	319 (23.63)
51 ~ 60 y	138 (10.22)
≥61 y	10 (0.74)
Education level
Bachelor	416 (30.81)
Master	622 (46.07)
Doctorate	312 (23.11)
Clinical title
Resident doctor	417 (30.89)
Attending physician	385 (28.52)
Associate chief physician	289 (21.41)
Chief physician	259 (19.19)
Working experience
1 ~ 5 y	455 (33.70)
6 ~ 10 y	231 (17.11)
11 ~ 15 y	230 (17.04)
16 ~ 20 y	161 (11.93)
≥21 y	273 (20.22)
Clinical department
Medical oncology	436 (32.30)
Radiation oncology	343 (25.41)
Surgical oncology	326 (24.15)
Other	245 (18.15)
University affiliated hospital
Yes	801 (59.33)
No	549 (40.67)
Hospital grade
Grade II	147 (10.89)
Grade III	1203 (89.11)
Hospital category
General hospital	823 (60.96)
Cancer hospital	527 (39.04)
Economic zone
Northeast	164 (12.15)
Eastern	547 (40.52)
Central	200 (14.81)
Western	439 (32.52)
Have any experience using medical AI products?
Yes	1215 (90.00)
No	135 (10.00)
Heard of or used ChatGPT?
Never heard	282 (20.89)
Heard, but no use experience	871 (64.52)
Have used ChatGPT or related products	197 (14.59)

AI Knowledge, attitudes, perceptions, and willingness to use AI

Table 2 presents the responses to the 5-point Likert scale questions, where scores of 1–2 indicate low level, a score of 3 indicates middle level, and scores of 4–5 indicate high level.

Table 2.

AI knowledge, attitudes, perceptions, and willingness to use AI

Dimension	Variable names	No. (%), N = 1350	Mean	Cronbach’s α
Knowledge	Understanding of AI related terms			0.839
	high	680 (50.37)	3.524
	middle	415 (30.74)
	low	255 (18.89)
	Awareness of the limitations of AI
	high	530 (39.26)	3.256
	middle	473 (35.04)
	low	347 (25.70)
Attitudes	Trust in AI technology			0.879
	high	827 (61.26)	3.759
	middle	388 (28.74)
	low	135 (10.00)
	Acceptance of AI technology
	high	976 (72.30)	4.022
	middle	277 (20.52)
	low	97 (7.19)
Perceptions	AI has surpassed the average diagnostic ability of doctors			0.779
	high	588 (43.56)	3.255
	middle	356 (26.37)
	low	406 (30.07)
	AI will replace doctors
	high	435 (32.22)	2.829
	middle	339 (25.11)
	low	576 (42.67)
Willingness	Willingness to use AI products			0.913
	high	1032 (76.44)	4.162
	middle	219 (16.22)
	low	99 (7.33)
	Willingness to use AI products as much as possible
	high	966 (71.56)	4.061
	middle	272 (20.15)
	low	112 (8.30)

Chinese oncologists demonstrated an understanding of AI terms, with a mean score of 3.524. Specifically, 680 respondents (50.37%) demonstrated high understanding, 415 (30.74%) exhibited middle understanding, and 255 (18.89%) showed low understanding. The awareness of AI limitations scored a mean of 3.256, with 530 respondents (39.26%) showing high awareness, 473 (35.04%) demonstrating middle awareness, and 347 (25.70%) indicating low awareness.

Regarding attitudes, trust in AI technology had a mean score of 3.759. Specifically, 827 respondents (61.26%) demonstrated high trust, 388 (28.74%) displayed middle trust, and 135 (10.00%) showed low trust. Acceptance of AI in healthcare had a mean score of 4.022, with 976 respondents (72.30%) expressing high acceptance, 277 (20.52%) showing middle acceptance, and 97 (7.19%) indicating low acceptance.

The perception that AI surpasses the average diagnostic ability of doctors had a mean score of 3.255. Among the respondents, 43.56% rated this aspect as high, 26.37% rated it as middle, and 30.07% rated it as low. Regarding AI replacing doctors, the mean score was 2.829. Specifically, 32.22% rated this aspect as low, 25.11% rated it as middle, and 42.67% rated it as low.

The willingness to use AI products was high, with a mean score of 4.162; 76.44% had high willingness, 16.22% had middle willingness, and 7.33% were low. Similarly, the willingness for extensive AI use was high, with a mean score of 4.061.

Factors associated with high willingness to use AI

Although Chinese oncologists demonstrate high understanding and trust in AI, their willingness to use it is the most critical factor for its practical implementation. Therefore, we further analyzed the key factors influencing willingness to use AI, using regression analyses where “high willingness” was defined as a score of 4 or 5 on a 5-point Likert scale.

Univariable logistic regression identified that male oncologists, those with doctorate degrees or senior clinical titles, and those working in radiation or surgical oncology department were more likely to demonstrate high willingness to use AI. Additionally, employment in a grade III hospital, high understanding of AI terms, high awareness of AI limitations, high trust in AI, and prior experience with AI products and ChatGPT were also significant factors (Table 3).

Table 3.

Univariable and multivariable logistic regression for high willingness to use AI

	Univariable OR (CI_95)	P-value	Multivariable OR(CI_95)	P-value
Demographics
Gender
Male	1.53 (0.17, 0.68)	0.001	1.15 (0.86, 1.53)	0.33
Age			··	··
31 ~ 40 y	0.92 (-0.39, 0.23)	0.62
41 ~ 50 y	1.36 (-0.06, 0.68)	0.10
51 ~ 60 y	1.03 (-0.43, 0.49)	0.91
≥61 y	2.9 (-0.63, 3.99)	0.32
Education level			··	··
Doctorate	2.23 (0.43, 1.18)	<0.001
Master	1.28 (-0.04, 0.52)	0.09
Clinical title
Senior titles	1.58 (0.2, 0.73)	<0.001	1.41 (1.05, 1.90)	0.02
Working experience			··	··
6 ~ 10 y	0.82 (-0.56, 0.16)	0.26
11 ~ 15 y	1.18 (-0.21, 0.55)	0.40
16 ~ 20 y	1.32 (-0.16, 0.73)	0.22
≥21 y	1.21 (-0.17, 0.56)	0.3
Clinical department
Radiation oncology	2.3 (0.48, 1.20)	<0.001	1.9 (1.29, 2.81)	0.001
Surgical oncology	1.5 (0.07, 0.74)	0.02	1.24 (0.86, 1.8)	0.25
Others	0.95 (-0.40, 0.29)	0.76	0.89 (0.6, 1.31)	0.55
University affiliated hospital
Yes	0.65 (-0.69, -0.18)	<0.001	0.95 (0.7, 1.28)	0.72
Hospital grade
Grade III	1.93 (0.29, 1.02)	<0.001	1.18 (0.74, 1.89)	0.48
Hospital category			··	··
Cancer hospital	1.31 (0.01, 0.53)	0.05
Economic zone			··	··
Eastern	1.04 (-0.36, 0.42)	0.86
Northeast	1.43 (-0.17, 0.90)	0.19
Western	0.67 (-0.80, -0.01)	0.05
Knowledge
Understanding of AI terms
high	5.24 (1.37, 1.96)	<0.001	4.66 (3.46, 6.34)	<0.001
Awareness of the limitations of AI			··	··
high	7.19 (1.62, 2.36)	<0.001
Attitudes
Trust in AI technology			··	··
high	11.54 (2.14, 2.76)	<0.001
Acceptance of AI technology			··	··
high	18.64 (2.62, 3.24)	<0.001
Perceptions
AI has surpassed the average diagnostic ability of doctors			··	··
high	11.35 (1.72, 2.43)	<0.001
AI will replace doctors			··	··
high	7.87 (1.96, 2.95)	<0.001
Experience
Have any experience using medical AI products?
Yes	2.08 (0.35, 1.1)	<0.001	1.49 (0.98, 2.25)	0.06
Heard of or used ChatGPT?
Heard, but never use	1.65 (0.2, 0.79)	<0.001	1.21 (0.87, 1.68)	0.25
Have used ChatGPT or related products	4.19 (0.93, 1.97)	<0.001	2.28 (1.32, 4.07)	0.004

In the multivariable logistic regression model, senior titles (OR 1.41, 95% CI:1.05–1.90), working in the radiation oncology department (OR 1.9, 95% CI:1.29, 2.81), high understanding of AI terms (OR 4.66, 95% CI: 3.46, 6.34), and prior experience with ChatGPT (OR 2.28, 95% CI:1.32, 4.07) were independently associated with high willingness to use AI (Table 3).

Sensitivity analyses, in which ‘Understanding of AI terms’ was replaced with ‘Awareness of AI limitations’ in an alternative regression model, yielded results consistent with the main analysis (Supplementary Table 2). Additionally, sensitivity analyses showed that respondents who scored 4 or 5 on the questions “AI has surpassed the average diagnostic ability of doctors” or “AI will replace doctors” were independently associated with a high willingness to use AI (Supplementary Tables 3 & 4).

Discussion

With a sample of 1,350 respondents across 31 regions, this study provides the first comprehensive analysis of nationwide oncologists’ AI knowledge, attitudes, perceptions, and willingness to use AI in China. The findings reveal that Chinese oncologists possess a good understanding of AI knowledge and generally exhibit positive attitudes toward AI. Although perceptions of AI’s potential to surpass doctors’ diagnostic capabilities are moderately favorable, some skepticism remains, and there is ongoing debate about whether AI will replace doctors. Nevertheless, the willingness to use AI remains high. Significant predictors included a high understanding of AI terms and limitations, senior clinical titles, working in the radiation oncology department, and prior experience with ChatGPT.

Surveys of 4,492 medical students from nine countries, 718 dermatology professionals from 91 countries, and 669 medical students and physicians in Korea indicate relatively low familiarity with AI [22, 34, 35]. However, our study reveals that most respondents have a high understanding of AI terms and limitations. This may be attributed to the extensive application of AI in oncology, particularly in areas such as cancer diagnostics, medical imaging, and radiotherapy in recent years [36, 37]. In addition, the growing familiarity with AI may be driven by the increasing popularity of generative AI tools, such as ChatGPT, which has rapidly gained traction since late 2022. The widespread use of these tools has likely enhanced the understanding and acceptance of AI. As generative AI tools evolve and gain further momentum across various fields, their influence on healthcare professionals’ views on AI is expected to grow, further shaping perspectives towards its application in medical practice.

For years, limited knowledge and experience with AI may have contributed to doctors’ lack of trust in AI and even fears of being replaced. A survey of 331 top medical journal authors expressed doubts about AI replacing radiologist [38]. Another study highlighted that, while AI’s accuracy exceeds that of experts, the ultimate decision in Japan must still be made by endoscopists [37]. In our study, 61.26% of respondents expressed high trust in AI, yet fewer than half (42.67%) believed there was a high likelihood of being replaced by AI. This suggests that oncologists are beginning to recognize both the strengths and limitations of AI, but their confidence in not being replaced by AI may be diminishing.

Several studies have reported a high willingness to use AI across various medical fields, consistent with our findings [39, 40]. As for the specific factors affecting willingness, a systematic review of 758 respondents from 39 countries found that accuracy, ease of use, and efficiency were the top factors affecting willingness to use clinical AI [39]. Another systematic review reported that safety is an important factor that could affect the willingness to use AI [30]. User experience significantly influences attitudes toward behavior, which may extend to adoption of AI in healthcare [41]. However, these studies do not analyze the relationship between oncologists’ demographics and their willingness to use AI. In our study, factors such as senior clinical titles, working in the radiation oncology department, a strong understanding of AI terms, and experience with ChatGPT were key drivers of the willingness to use AI. Senior clinical titles (OR 1.41) and working in radiation oncology (OR 1.9) were significant factors, with oncologists in these roles more likely to use AI. AI knowledge (OR 4.66) highlighting its crucial role in promoting willingness to use AI. Experience with ChatGPT (OR 2.28) also positively influenced AI use willingness. However, no significant differences were found in AI use willingness across the economic regions in China. These findings suggest that increasing AI knowledge and support from senior oncologists in specific fields could effectively promote AI adoption in clinical settings across China.

Notably, according to our findings, whether respondents highly believed that AI had already surpassed doctors in diagnostic capabilities or that AI would replace doctors, they both demonstrated a high willingness to use AI. This further suggests that oncologists have recognized the advantages of AI in the medical field far outweigh its potential risks. Embracing and learning AI technology is the necessary path to equipping their core competencies in the future clinical practice.

Limitations

This study has several limitations. First, data were collected in 2023, which may limit the relevance of the findings due to rapid developments in AI. Additionally, sampling bias may be present due to the underrepresentation of oncologists from smaller hospitals or rural areas in China. Finally, response bias could exist due to self-reporting, as participants may overestimate their AI knowledge or willingness to use it.

Conclusion

This study unveils a good understanding and positive attitudes toward AI, coupled with a strong willingness to use AI products among oncologists across China, despite ongoing controversy over whether AI will replace doctors. Oncologists with senior titles, better AI knowledge, and prior experience with AI show an even higher willingness to use it. Considering these findings, targeted AI education and increased exposure are crucial for the successful integration of AI into oncology settings in China.

Abbreviations

AI

Artificial intelligence

DL

Deep learning

ML

Machine learning

LLMs

Large language models

GPT

Generative pretrained transformer

UTAUT

Unified theory of acceptance and use of technology

CACA

Chinese Anti-cancer Association

Authors’ contributions

Xiaomin Xiong led the conceptualization and design of the study, analyzed the data, and contributed to writing the manuscript and creating data visualizations. Ming Li contributed to shaping the research questions and design and drafted the original manuscript. Mingming Xiao focused on data collection, ensuring the accuracy of the empirical evidence. Haixia Liu collected key data, contributing to the stud’ s validity and reliability. Chunling Jiang also contributed to data collection, ensuring alignment with the research objectives. Yongmei Song supervised the research, guided the investigation, and participated in data collection. Jing Zhou supervised the project, contributed to the investigation, and revised the manuscript. Bo Xu supervised the project, contributed to the investigation, and secured funding as the corresponding author.

Funding

This work was supported by the Teaching Reform Research Project of Chongqing University (2021Y55).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethical Committee of Chongqing University Cancer Hospital (Approval number: CZLS2022244-A). Digital informed consent was obtained from all participants before they completed the questionnaire.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.