Application of AI-assisted PBL-CBL combined teaching in the diagnosis and treatment of post-herpetic neuralgia: a randomised controlled trial

Abstract

Background

This study aimed to explore the efficacy of a teaching model integrating artificial intelligence-assisted problem-based learning (PBL) with case-based learning (CBL) in standardized clinical teaching of postherpetic neuralgia (PHN) for medical interns.

Methods

A total of 120 interns at the Affiliated Hospital of Yunnan University were randomly divided into an observation group and a control group (n = 60 each). The control group received conventional lecture-based teaching, while the observation group was taught using the AI-assisted PBL-CBL combined model. After four weeks of teaching, the educational outcomes were evaluated through theoretical examinations, clinical skill assessments, Mini-Clinical Evaluation Exercise (Mini-CEX) scores, and a teaching satisfaction questionnaire.

Results

After four weeks, the theoretical score of the observation group was 87.35 ± 5.66, and the clinical practice score was 90.12 ± 3.47, both significantly higher than those of the control group (84.67 ± 4.05 and 88.23 ± 3.94, respectively; P < 0.01). The Mini-CEX evaluation showed that students in the observation group had significantly improved clinical practice skills (P < 0.001), whereas no significant differences were observed between the two groups in humanistic qualities, professionalism, or organization/efficiency (P > 0.05). In addition, the observation group reported higher teaching satisfaction than the control group.

Conclusions

The AI-assisted PBL-CBL teaching model significantly improves medical interns’ theoretical knowledge and clinical skills related to PHN, enhances their clinical reasoning ability, and increases teaching satisfaction.

Introduction

Post-herpetic neuralgia (PHN) is the most common complication of herpes zoster (HZ) [1, 2]. PHN pain is typically confined to the dermatomal area affected by the original rash, significantly impairing the quality of life, sleep status and daily living capacity of patients [3]. The incidence of PHN is significantly associated with age, with older adults over 50 and immunocompromised patients at higher risk [4]. When applying theory to practice in a clinical scenario, students have yet to become acquainted with PHN. This lack of familiarity often hinders students’ ability to effectively integrate theoretical knowledge into clinical decision-making, posing a barrier to developing sound clinical reasoning. Consequently, it is imperative to reform the educational model by maximising student’s clinical reasoning ability taking into account the limitations of teaching resources.

The ability of clinical reasoning is one of the core competencies required by medical professionals. The development of clinical reasoning ability needs a thorough assimilation of theory and clinical practice. Traditional lecture-based teaching is teacher-centered and realizes one-way knowledge delivery from teachers to students [5]. However, this method fails to mirror the complexities of real-world clinical situations and is therefore said to be limited in its effectiveness to develop medical students’ clinical reasoning skills [6]. An innovative teaching model that has gained popularity recently includes Problem-Based Learning (PBL) and Case-Based Learning (CBL) [7, 8]. PBL employs an approach centered on problems to stimulate inquiry while CBL uses an approach focused on a real-life case to link theory with practice [9, 10]. Combining both approaches yields additional benefits. The artificial intelligence (AI) technology is developing rapidly, and its assisted teaching application in medical education can make use of virtual simulation and personalized guidance to overcome the constraints of time, place, and resources in traditional teaching [11–13]. Therefore, integrating AI-assisted technology with the PBL-CBL combined teaching model may offer a promising approach to address the limitations of traditional lectures, enhance clinical reasoning development, and optimize the use of limited teaching resources in PHN education.

This study aims to explore the AI-assisted PBL-CBL teaching model, investigate its application effect in cultivating students’ clinical reasoning ability in the clinical teaching of PHN, and provide a scientific basis for optimizing clinical teaching strategies in the field of PHN.

Methods

Design and participants

This study adopted a single-center randomized controlled trial (RCT) design in strict accordance with the CONSORT guidelines. Study participants were interns recruited from the Affiliated Hospital of Yunnan University between March 1 and May 30, 2025. The study was approved by the medical ethics committee of the Affiliated Hospital of Yunnan University (ethics approval number: KJK#030-25-5R1), and written informed consent was obtained from all participants. Participants were randomly assigned at a 1:1 ratio using a random number table, with allocation concealment and assessor blinding implemented to control bias. Based on the study design, a total sample size of 120 cases was enrolled, with 60 cases in both the control group and the observation group (Fig. 1).

Fig. 1.

Study design and flow chart

Eligibility criteria for participants: (1) Satisfactory completion of core professional coursework (2), voluntary participation in this study and signing of an informed consent form.

Exclusion criteria: (1) Inability to complete the training course (2), Prior clinical experience related to PHN, and (3) Inability to complete the full course of instruction and assessment due to reasons such as leave of absence or withdrawal from studies.

Intervention measures

The two teaching modules on PHN focus on its core cognition, clinical evaluation, treatment plan and rehabilitation management. Every module was handled by the same instructor over a duration of four weeks.

The control group was taught using lecture-based instruction. The instructors delivered systematic lectures on the theoretical knowledge of PHN using PowerPoint presentations. They also provided explanations using typical case images. Medical interns were divided into 10 groups of six students each for discussion and learning. The groups summed up the key takeaways and difficult concepts at the end. This was followed by a question-and-answer session.

The observation group adopted a teaching model that combines AI-assisted PBL with CBL, utilizing Doubao AI (V3.8, https://www.doubao.com, developed by ByteDance) as the core AI tool throughout the process. This tool primarily employs two key modules: the intelligent Q&A module and the AI agent module. Through chat-based interaction combined with structured tasks, students engage with the Doubao AI system across the entire learning process, including pre-class preparation, in-class simulation, and post-class consolidation. The interaction mainly consists of three forms: addressing preview difficulties, free Q&A, and clinical simulation tasks. Examples of pre-class preview tasks include explaining the pathogenesis and key risk factors of PHN, as well as the common clinical manifestations of PHN in elderly patients. Examples of in-class simulation tasks include conducting condition assessment and formulating nursing and treatment plans based on virtual patient symptoms, and evaluating the rationality of PHN nursing interventions. The specific implementation process is as follows:

1. Pre-class preparation: Learning materials and core questions are distributed, and students complete preview tasks with the support of the Doubao AI system.

2. In-class implementation: Teacher-verified simulated patients are generated based on PHN cases. Students conduct simulated diagnosis and treatment in groups, with the Doubao AI system evaluating the diagnosis and treatment plans. Teachers guide discussions and correct biases in clinical reasoning.

3. Post-class summary: Groups present their learning outcomes, which are reviewed and commented on by teachers to strengthen the cultivation of clinical thinking.

The prompts and task examples applied are as follows: ① Pre-class preview prompts: “Explain the pathogenesis of PHN and list key risk factors associated with it” “What nursing methods should be adopted for different clinical manifestations of PHN in elderly patients?” ② In-class simulation tasks: “Conduct a comprehensive condition assessment and put forward a targeted medical and nursing plan based on the symptoms of the simulated patient (burning pain in the right chest, paroxysmal aggravation, and sleep disturbance)” “Assess the rationality and precautions of the following nursing interventions for patients with PHN: oral gabapentin 300 mg three times a day, local cold compress therapy, and psychological intervention measures.”

The output forms of Doubao AI mainly include three types: ① Text responses: explanations to theoretical questions, such as interpretations of the pathogenesis of PHN, nursing principles, and the mechanism of drug action. ② Simulated cases: dynamic and scenario-based virtual PHN patient simulations generated by the AI agent module, which can simulate symptom progression and changes in patient condition according to students’ intervention measures. ③ Feedback and evaluation: objective assessments of students’ nursing and treatment plans, including the rationality, safety, effectiveness and individualization of the plans, along with specific suggestions for improvement.

Teaching effectiveness evaluation

The theoretical test is a closed book test. The test paper jointly developed by the two instructors is as per the fundamental theories of PHN, standardized diagnostic and treatment process and rehabilitation management. The total score was 100 marks.

The assessment of clinical skills used simulations. Both student cohorts receive the same simulated patients presenting with PHN (medical history, physical signs and examination data). The assessment checks students on medical history taking, physical examination, diagnosis and differential diagnosis, and treatment plan development. The scoring is done by three dermatology specialists using a standard rubric with a maximum score of 100.

A mini-clinical evaluation exercise (Mini-CEX) was used to assess students’ abilities in history taking, physical examination, and clinical reasoning. Each student completed one standardized Mini-CEX encounter involving a real patient under supervised clinical conditions. The validated Mini-CEX tool employs a 9-point rating scale across seven domains: (1) medical interviewing skills (2), physical examination skills (3), humanistic qualities/professionalism (4), clinical judgment (5), counseling skills (6), organizational efficiency, and (7) overall clinical competence. Scores are categorized as unsatisfactory (1–3), satisfactory (4–6), or superior (7–9). All assessments were conducted by the same independent evaluator to ensure inter-rater reliability.

The teaching satisfaction questionnaire was created by the author to determine the participants’ level of teaching satisfaction (Cronbach’s α value was 0.823), as well as improved interest in learning, improved self-directed learning ability, improved clinical reasoning and improved teamwork satisfaction. Each of the dimensions is scored on a scale of 1 to 4, where 1 means very dissatisfied and 4 signifies very satisfied. Overall score of greater than or equal to 16 represents satisfaction. Meanwhile, the observation group was individually assessed on their satisfaction with the new teaching model, using a separate survey scale with a 5-point Likert scale.

In the above evaluation of teaching effectiveness, Mini-CEX is used as the primary outcome indicator, while theoretical testing, clinical skills assessment, and satisfaction are used as secondary outcome indicators to measure students’ knowledge mastery, technical operation ability, and satisfaction with the learning process under different teaching measures.

Statistical analysis

Statistical software SPSS 26.0 was used for data analysis. The distribution of continuous variables was described using mean and standard deviation. Prior to analysis, the normality of distribution was assessed using the Shapiro-Wilk test, and homogeneity of variances was tested using Levene’s test to verify the assumptions for parametric testing. Independent samples t-test was used for between-group comparisons of normally distributed continuous variables, with Cohen’s d calculated as the effect size and 95% confidence intervals (CIs) reported for mean differences. Categorical variables were compared using the chi-square test. No formal correction for multiple comparisons was conducted in this study. A statistically significant difference was indicated when P < 0.05.

Results

General information

From March to May 2025, 120 students were randomly assigned to observation group (n = 60) or control group (n = 60). The mean age of students in the observation group was 22.17 ± 0.60 years, while that in the control group was 22.19 ± 0.66 years; the difference between the two groups was not statistically significant (Table 1, P = 0.866). The breakdown of participants by gender was 89 males and 31 females. There was no significant difference in the mean grades of basic medical courses between the two groups.

Table 1.

General baseline characteristics

	Male/Female	Average age (‾x ± s)	Average Grade in Basic Medical Courses (‾x ± s)
Observation Group(n = 60)	47/13	22.17 ± 0.60	88.24 ± 3.82
Control Group (n = 60)	42/18	22.19 ± 0.66	89.10 ± 3.34
t/χ²	1.09	0.17	1.31
P	> 0.05	> 0.05	> 0.05

Comparison of theoretical examination and clinical skill examination between two groups

According to the data shown in Table 2, the observation group scored significantly higher than the control group in both the theoretical examination (t = 2.98, P < 0.01, 95% CI: 0.87–4.49, Cohen’s d = 0.54) and the clinical skill examination (t = 2.79, P < 0.01, 95% CI: 0.55–3.23, Cohen’s d = 0.51).

Table 2.

Comparison of theoretical exam and clinical skill assessment scores between groups (‾x ± s)

	Theoretical examination	clinical skill examination
Observation Group(n = 60)	87.35 ± 5.66	90.12 ± 3.47
Control Group (n = 60)	84.67 ± 4.05	88.23 ± 3.94
t	2.98	2.79
P	< 0.01	< 0.01

Comparison of clinical reasoning scores between two groups

The results of the Mini-CEX evaluation are presented in the table. Compared with the control group, the observation group demonstrated significantly higher scores in most dimensions of clinical competence assessment. The observation group showed statistically significant improvements in medical interviewing skills (t = 6.07, P < 0.0001), physical examination skills (t = 4.46, P < 0.0001), clinical judgment (t = 10.69, P < 0.0001), and communication skills (t = 4.42, P < 0.0001). For the overall clinical competence dimension, the observation group also achieved a significantly higher score than the control group (t = 8.19, P < 0.0001). The total Mini-CEX score of the observation group was significantly higher than that of the control group (t = 4.55, P < 0.0001). In contrast, no statistically significant differences were observed between the two groups in humanistic qualities/professionalism (t = 1.32, P = 0.19) and organization/efficiency (t = 0.49, P = 0.62)(Table 3).

Table 3.

Comparison of Mini-CEX scores between two groups (‾x ± s)

	Observation Group (n = 60)	Control Group (n = 60)	t	P
Medical Interviewing Skills	4.53 ± 0.32	4.23 ± 0.21	6.07	< 0.0001
Physical Examination Skills	4.44 ± 0.27	4.22 ± 0.27	4.46	< 0.0001
Clinical Judgment	4.54 ± 0.21	4.13 ± 0.21	10.69	< 0.0001
Communication Skills	4.72 ± 0.36	4.46 ± 0.28	4.42	< 0.0001
Humanistic Qualities / Professionalism	4.92 ± 0.66	5.04 ± 0.25	1.32	0.19
Organization / Efficiency	4.45 ± 0.36	4.50 ± 0.70	0.49	0.62
Overall Clinical Competence	4.57 ± 0.28	4.2 ± 0.21	8.19	< 0.0001
Total Score	32.16 ± 1.87	30.87 ± 1.15	4.55	< 0.0001

Satisfaction analysis

A total of 120 questionnaires were distributed, with 120 valid responses collected, achieving a 100% valid response rate. The total satisfaction rate of the observation group was 95% (57/60) which is higher than the control group 85% (51/60).

Compared with the control group, the observation group showed significantly higher scores in multiple dimensions. The observation group achieved a significantly higher overall teaching satisfaction score (t = 1.98, P < 0.05). The improvement in learning interest was also significantly greater in the observation group (t = 5.14, P < 0.001). Similarly, the enhancement of autonomous learning ability was significantly more pronounced in the observation group (t = 5.06, P < 0.001). The observation group also demonstrated a significant advantage in the enhancement of clinical reasoning skills (t = 4.17, P < 0.001). No statistically significant difference was observed between the two groups in teamwork satisfaction (t = 1.89, P > 0.05) (Table 4).

Table 4.

Comparison of Teaching Satisfaction Between Two Groups of Participants (‾x ± s, 4-point scale)

	Teaching Satisfaction	Learning Interest Improved	Enhancement of autonomous learning ability	Enhanced clinical reasoning skills	Teamwork satisfaction
Observation Group(n = 60)	3.61 ± 0.55	3.75 ± 0.43	3.90 ± 0.30	3.70 ± 0.59	3.63 ± 0.72
Control Group (n = 60)	3.4 ± 0.61	3.26 ± 0.60	3.45 ± 0.62	3.16 ± 0.81	3.35 ± 0.89
t	1.98	5.14	5.06	4.17	1.89
P	< 0.05	< 0.001	< 0.001	< 0.001	> 0.05

Students and other respondents in the observation group were quite satisfied with the new model of teaching overall with a score of 95%. Nonetheless, a high 91.67% users are satisfied with the use of AI. However, 6.67% are not confident enough to devise treatment plans (Table 5).

Table 5.

Questionnaire Analysis of the Observation Group Teaching Model (‾x ± s,5-point scale)

Project	Average Score (5-point scale)	Satisfied/Very Satisfied Ratio (%)
Learning Engagement	4.28 ± 0.58	96.67
Satisfaction with Learning Content	4.08 ± 0.79	98.33
Interest in Course Knowledge	4.31 ± 0.59	98.33
Improvement in Self-Directed Learning Ability	4.36 ± 0.74	98.33
Level of Assistance from New Teaching Models	4.50 ± 0.65	98.33
Ability to Independently Solve New Problems	3.95 ± 0.77	96.67
Comprehensive Clinical Skills	4.01 ± 0.68	96.67
Enhanced Event Management Capabilities	3.90 ± 0.66	96.67
Confidence in Formulating Treatment Plans	3.78 ± 0.69	93.33
Satisfaction with AI Usage	3.63 ± 0.92	91.67
Satisfaction with Learning Environment	4.25 ± 0.63	98.33
Overall Average Satisfaction	3.92 ± 0.56	95

Discussion

Given the complex diagnosis and treatment of PHN, doctors’ critical thinking skills are required to a greater extent while conventional teaching models are still inadequate for instilling clinical reasoning in medical students. This paper integrates AI platform with PBL-CBL methods. The aim of this study is to improve the quality of clinical education regarding PHN and stimulate students’ clinical reasoning.

Research findings show that the observation group scored significantly higher on theoretical tests and clinical skill tests than the control group. The study proved the AI-assisted PBL-CBL teaching model effectively enhances mastery of PHN and the clinical skills of medical students. Traditional lecture-based teaching primarily focuses on theoretical instruction, where students passively receive knowledge, hence theory cannot be effectively integrated with practice [6, 14]. The AI-supported PBL-CBL model, which employs problems and utilizes case studies, significantly promotes student engagement. Through personalized pre-study resources and real-time Q&A abilities provided by the AI platform, students can organize their theoretical frameworks systematically [15, 16]. Consistent with these findings, other studies on AI-assisted instruction have also confirmed the positive role of AI in enhancing students’ conceptual understanding and knowledge structuring [17]. AI can generate simulations of the progression of symptoms and diagnosis analysis, helping students experience the clinical diagnosis and treatment process in a more intuitive way so that they can apply their theoretical knowledge [17, 18]. This approach to learning through active inquiry breaks through the restrictions of time and space that limit traditional teaching so as to improve learning efficiency and effectiveness.

One of the main goals of medical education is the cultivation of clinical skills [19]. The results of the Mini-CEX assessment showed that the observation group scored significantly higher than the control group in core clinical skills such as medical history taking, physical examination, and clinical judgment, while there were no significant differences between the two groups in humanistic qualities/professionalism and organizational efficiency. This indicates that the teaching model can effectively improve medical students’ core clinical skills, which is consistent with the core needs of clinical teaching. The core advantage of this teaching model lies in the complementary advantages of AI technology combined with PBL-CBL teaching methods. AI can provide standardized clinical simulation scenarios and real-time accurate feedback, which, together with the problem-oriented inquiry and thinking training characteristics of PBL and the immersive teaching advantage of real cases in CBL, effectively enhances the standardization of medical students’ clinical operations and their clinical thinking abilities [20, 21]. In comparison with previous studies on artificial intelligence-assisted teaching, Gizem Ergezen Sahin and colleagues reported higher Mini-CEX scores in the AI-PBL group than in the conventional PBL group, though no statistically significant difference was identified between the two cohorts [21]. In another relevant study conducted by Zeng Hui and team, the PBL-ChatGPT group presented remarkable improvements in medical interview skills, clinical judgement and overall clinical competence relative to the lecture-based learning group. Nevertheless, no obvious intergroup differences were found in humanistic professionalism and organizational efficiency [17]. Such findings are highly consistent with the results observed in the present research. However, there was no significant difference in the scores of humanistic qualities / professionalism and organizational / efficiency between the two groups, mainly because the teaching design focused on the training of core clinical skills, and the formation of such comprehensive qualities requires long-term clinical edification and practical accumulation, making it difficult for short-term intervention to show effects [20, 21]. At the same time, AI technology is insufficient in the refined design of humanistic quality guidance and organizational efficiency training.

According to the teaching satisfaction survey results, the observation group’s satisfaction was significantly higher than that of the control group, indicating that the AI-assisted PBL-CBL teaching model enjoys better reception among students. The model is fully student-centered, inspiring students through problems and case analysis. The personalized advice and resource recommendations from the AI platform can cater to different students’ learning needs while group discussions and presentation sessions enrich their engagement and feeling of accomplishment. On the contrary, conventional lectures are limited in diversity, less engaging for learners, and less motivating to learn. There were no significant differences in team cooperation satisfaction between the two groups, which suggests that implementing group learning in the teaching process is feasible. In fact, team cooperation satisfaction may be more dependent on team dynamics than on AI tools. Customized satisfaction questionnaires were distributed specifically to students in the observation group, and the results showed that some students expressed dissatisfaction with the application of AI in teaching. The main reason for this was the lack of control over data quality and stability of the current AI platform [22, 23].

The AI-assisted PBL-CBL model encourages active knowledge construction by engaging interns in clinical problem-solving and simulated case practice. Such interactive learning deepens the understanding of PHN core knowledge and improves information retrieval efficiency, which is far more effective than passive lecture reception. The AI platform can deliver targeted real-time feedback on diagnostic and therapeutic decision-making. Interns are able to correct cognitive biases in a timely manner and consolidate systematic clinical reasoning through continuous deliberate practice. AI-generated virtual patient scenarios allow repeated clinical training without the restrictions of limited teaching resources. Such repeated simulation effectively strengthens clinical pattern recognition and standardizes the clinical thinking framework for PHN comprehensive management. The integration of problem-based inquiry, authentic clinical cases and personalized AI guidance rationalizes cognitive resource distribution during learning. It cuts down invalid cognitive interference and concentrates learning resources on the absorption of key clinical competencies. All these intrinsic educational strengths facilitate high-efficiency transformation and flexible application of professional knowledge. These synergistic mechanisms fully explain why the combined teaching intervention yielded better theoretical scores and more competent clinical performance among medical interns.

Although the previous article discussed many advantages of AI platform in teaching, the ethical and practical issues of integrating it into medical education still deserve attention. First, the AI system itself has inherent technical limitations, and the model may extend or amplify the existing differences in medical knowledge or language expression [24]. In the context of pre-class learning resource organization, these limitations may lead students to internalize inaccurate knowledge framework, or ignore the typical manifestations of diseases in minority groups. Secondly, AI platforms usually need to continuously input students’ questions, case discussions and even simulated patient data, which may inadvertently lead to the disclosure of sensitive information. Third, excessive dependence on AI in the early stage of clinical training will bring the risk of weakening basic cognitive ability. If students are used to accepting AI generated explanations or real-time answers directly without critical thinking, their diagnostic reasoning ability, comprehensive information analysis ability, and interpersonal communication inquiry ability may not be fully developed. This phenomenon is particularly worrying in clinical decision-making, because independent judgment is irreplaceable. Therefore, although AI provides scalable educational support, its deployment must be supplemented by structured measures, including transparent auditing of algorithmic bias, data processing protocols to protect privacy, and positioning AI as an auxiliary tool.

This study inevitably has several limitations. First, the single center design with small sample size may lead to selection bias and limit the extrapolation of research results to a wider clinical environment. Second, the short duration of intervention failed to reflect the sustained effects and long-term value of the teaching model. Thirdly, the entire teaching process is conducted solely under simulated patient conditions, lacking the complexity found in real-life environments. Fourth, there are inherent AI-related limitations in this research: the current AI platform still needs continuous optimization, with insufficient flexibility and poor adaptability to diversified clinical teaching demands. In addition, only short-term outcomes were analyzed in the present study; therefore, long-term efficacy and impacts still require prolonged follow-up observation. Future research will expand the sample size, conduct multicenter prospective studies, extend the follow-up period, further optimize and upgrade the AI-assisted teaching system, and continuously improve the overall quality of clinical teaching.

The AI-assisted PBL-CBL teaching mode has improved students’ theoretical and clinical performance, enhanced students’ learning interest and autonomous learning ability, and is welcomed by students. These results show that the integration of AI into clinical teaching can effectively improve students’ learning effect and participation, and provide feasible strategies for strengthening clinical teaching practice.

Conclusions

The findings of this study suggest that the AI-assisted PBL-CBL teaching model could improve medical students’ theoretical knowledge and clinical skills regard PHN. Additionally, this teaching model significantly enhances students’ clinical reasoning ability and achieves higher teaching satisfaction. This teaching method seeks to overcome the shortcomings of traditional teaching methods through an integrated form of training of theory and practice with a view to generating new ideas and methods for clinical teaching of PHN. It is worthy of widespread use.

Authors’ contributions

Yang F and He LP jointly proposed the research direction and designed the study content. Yang F completed the clinical teaching work. Wang JM performed data collation. Yang F conducted the statistical analysis. Yang F and Wang JM drafted the manuscript. He LP reviewed the manuscript. All authors read and approved the final version.

Funding

This research was supported by the Yunnan Provincial Education Department Fund (2025J0024 and 2025J0025).

Data availability

All data generated or analyzed in this study are included in the previously published article.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and approved by the Medical Ethics Committee of Yunnan University Affiliated Hospital. The approval number is KJK#030-25-5R1. All participants provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.