Team-based learning and its impact on clinical reasoning in medical education: A review and the unexplored role of artificial intelligence

Abstract

Team-based learning (TBL) is widely adopted in medical education to promote active learning, collaboration, and improved clinical reasoning. Although digital platforms are increasingly integrated into TBL, the use of artificial intelligence (AI) in this context remains largely theoretical. We systematically reviewed empirical studies evaluating the effectiveness of TBL on clinical reasoning skills among medical students. Searches were conducted using PubMed with Boolean operators targeting terms such as “team-based learning,” “clinical reasoning,” “AI,” and “medical education.” After screening 21 studies, 16 met the inclusion criteria. Data were extracted on study design, participants, TBL interventions, assessment methods, and presence of digital or AI tools. Of the 16 studies included, none employed true AI-based interventions. Although three studies incorporated digital platforms such as Tronclass, InteDashboard, or the Learning Activity Management System, these lacked adaptive, intelligent functionalities characteristic of AI. Most studies reported that TBL significantly improved clinical reasoning outcomes, assessed through validated tools such as the Script Concordance Test, readiness assurance tests, or final examinations. However, variability in study design and outcome measures limited cross-comparability. TBL appears effective in enhancing clinical reasoning among medical students, yet no current study has explored its integration with AI. This highlights a critical research gap and the need for future studies evaluating AI-driven, personalized learning tools within TBL frameworks.

Introduction

Team-based learning (TBL) is increasingly adopted in medical education to enhance engagement, knowledge retention, and critical thinking. It provides a structured, student-centered instructional approach that encourages collaboration and active learning. Clinical reasoning, a core competency for medical students, benefits from such structured peer-to-peer interaction, especially when supported by activities such as individual and team readiness assurance tests and clinical application exercises.[1,2]

Evidence from various disciplines—including neurology, pharmacology, and molecular biology—demonstrates that TBL can improve performance outcomes, particularly among low-performing students.[2,3,4] For example, students exposed to TBL often score higher on final examinations or group application tasks compared with those in traditional lecture-based settings.[2,3] These findings suggest that TBL enhances both knowledge retention and students’ confidence in applying clinical reasoning skills. Despite the proliferation of digital platforms to facilitate TBL, such as Tronclass, InteDashboard, and the Learning Activity Management System, none of the reviewed implementations use substantive artificial intelligence (AI) capabilities.[5,6,7] By analyzing existing empirical studies, the review aims to identify current evidence, highlight gaps in AI integration, and propose directions for future research in AI-enhanced medical education. Although AI-driven TBL has been explored in disciplines such as engineering and business education, its application in medical education remains underresearched. This review is focused on exploring this specific gap.

Materials and Methods

Eligibility criteria

We included empirical studies examining TBL in medical students (undergraduate or graduate), with clinical reasoning measured using validated tools. Studies had to involve a comparator (lecture-based or baseline) and assess outcomes beyond student satisfaction.

Search strategy

A comprehensive literature search was conducted using PubMed with the following Boolean strategy: ((artificial intelligence[Title/Abstract]) OR (AI[Title/Abstract])) AND ((TBL[Title/Abstract]) OR (team-based learning[Title/Abstract])) AND (medical education[Title/Abstract]). The search yielded 661 results, which were filtered to include only articles published in the last 10 years, written in English, with full text available, and focusing on individuals aged 19 to 24 years. After applying these filters, 356 records remained. Further screening based on predefined eligibility criteria—such as empirical evaluation of team-based learning among medical students with outcomes related to clinical reasoning—resulted in 21 studies. Of these, 5 were excluded after full-text review, and 16 studies were ultimately included in the systematic review [Figure 1].

Figure 1.

PRISMA flowchart

Selection process

Screening was conducted using predefined criteria evaluating the population, intervention (TBL), AI inclusion, outcome assessments, and study design. The limited number of studies included reflects our strict eligibility criteria to ensure only empirical studies with validated assessments of clinical reasoning among medical students were considered, thus strengthening the internal validity of the review.

Data collection process

Data were extracted on study design, participant demographics, TBL intervention details, clinical reasoning outcomes, student perceptions, and AI use [Table 1].

Table 1.

Data extraction table

Study		Study design		Participants		AI integration		Clinical reasoning assessment
Shiels et al.[1]		Observational		446 first-year, 206 second-year		None		Likert scale (confidence)
Qu et al.[2]		Randomized controlled trial		109		None		Final examination scores
Huang et al.[3]		Quasiexperimental, mixed-methods		99		None		Readiness assurance tests, final examination
Kim et al.[4]		Quasiexperimental, mixed-methods		49		None		Application test, national examination
Chan et al.[5]		Quasiexperimental		801		Learning Activity Mgmt System		Readiness for interprofessional learning
Xin et al.[6]		RCT, Quasiexperimental		100		None		Pre/post-tests
Johnston et al.[7]		Quasiexperimental, observational		195		InteDashboard		Self-report
Radzi et al.[8]		Cross-sectional, mixed-methods		85		None		Not applicable
Rotgans et al.[9]		Observational		88		None		Readiness assurance, situational interest
Zhou et al.[10]		Quasiexperimental		69		Tronclass		Academic performance
Horst et al.[11]		Mixed-methods, observational		78		None		Reflective questionnaire, empathy scale
Ishizuka et al.[12]		Mixed-methods, quasiexperimental		92		None		Script concordance test
Daou et al.[13]		Randomized controlled trial		107		None		Berlin questionnaire
Obad et al.[14]		Observational, mixed-methods		185		None		Clinical scenarios
Alimoglu et al.[15]		Quasiexperimental		373		None		Readiness and retention tests
Tang et al.[16]		Randomized controlled trial		95		None		Self-assessment

Risk of bias

ROB2 was applied to randomized controlled trials; the Newcastle–Ottawa Scale was used for observational and quasiexperimental studies [Table 2].

Table 2.

Risk of bias assessment

Study [#]		Tool used		Risk of bias summary
Qu et al.[2]		ROB2		Low risk
Daou et al.[13]		ROB2		Low risk
Xin et al.[6]		ROB2		Some concerns
Others[1,3,4,5,7,8,9,10,11,12,14,15,16]		Newcastle–Ottawa		Moderate to high quality

Data synthesis

Narrative synthesis was conducted due to methodological heterogeneity. Key themes were identified from both quantitative outcomes and qualitative reports.

Results

A total of sixteen studies met the inclusion criteria and were reviewed for their investigation into the effectiveness of team-based learning (TBL) on clinical reasoning among medical students. These studies encompassed a variety of research designs, including four randomized controlled trials (RCTs), eight quasiexperimental or mixed-methods studies, and four observational studies. Participant numbers ranged significantly, with sample sizes varying from as few as 49 to as many as 801, and the target population primarily consisted of undergraduate medical students.

Regarding the integration of technology, only three of the included studies reported the use of digital learning platforms. Johnston et al.[7] used InteDashboard, Chan et al.[5] incorporated the Learning Activity Management System, and Hu et al.[10] implemented Tronclass as part of their TBL framework. However, it is critical to note that none of these digital tools featured core functionalities characteristic of artificial intelligence (AI), such as intelligent feedback, machine learning adaptation, or real-time analytics. Consequently, no study in this review can be classified as using true AI-enhanced TBL.

Clinical reasoning was assessed through a variety of methods across the studies. Nine of the sixteen studies used validated tools for outcome assessment. For instance, Ishizuka et al.[12] used the Script Concordance Test to evaluate clinical reasoning, while Daou et al.[13] applied the Berlin Questionnaire. Huang et al.[3] and Kim et al.[4] assessed performance through readiness assurance tests and national or final examinations, respectively. Similarly, Qu et al.[2] measured outcomes using final examination scores, and Xin et al.[6] applied pretest and post-test evaluations. Alimoglu et al.[15] reported the use of readiness and retention tests. On the other hand, several studies relied on more subjective measures such as self-assessment,[7,16] Likert scale–based confidence ratings,[1] reflective questionnaires,[11] or reports of situational interest and interprofessional learning readiness,[5,9]

From the sixteen studies, twelve reported positive outcomes indicating significant improvements in students’ clinical reasoning or related knowledge after undergoing TBL interventions. For example, Ishizuka et al.[12] demonstrated improved reasoning using Script Concordance Testing, and Qu et al.[2] observed enhanced final examination performance. Similarly, Huang et al.[3] and Kim et al.[4] identified gains in both readiness assurance and summative assessments. These outcomes reinforce the potential efficacy of TBL in promoting cognitive development related to clinical decision-making.

By contrast, one study (Radzi et al.[8]) did not directly assess clinical reasoning and thus could not contribute evaluative data on TBL’s effectiveness in that domain. Three other studies (Shiels et al.,[1] Johnston et al.,[7] and Tang et al.[16]) used tools such as confidence scales and self-assessment without robust quantitative analysis, leaving their findings less conclusive regarding actual reasoning improvement.

The risk of bias was evaluated using the ROB2 tool for RCTs and the Newcastle–Ottawa Scale for other designs. Of the four RCTs, three (Qu et al.,[2] Daou et al.,[13] and Xin et al.[6]) were found to be of low risk or had some concerns, suggesting generally acceptable methodological rigor. The remaining studies were classified as moderate to high quality based on the Newcastle–Ottawa criteria, indicating that while not randomized, they maintained reasonable standards of design and reporting.

Discussion

This review evaluated the effects of AI-enhanced team-based learning (TBL) on clinical reasoning among medical students. While TBL alone has shown positive impacts on knowledge retention and reasoning skills, our findings reveal a notable gap in the integration of artificial intelligence within these educational frameworks.

Twelve of the sixteen studies included reported statistically significant improvements in clinical reasoning or knowledge retention using TBL strategies. For instance, Ishizuka et al.[12] demonstrated improved scores on Script Concordance Tests and self-assessed reasoning. Similarly, Kim et al.[4] and Huang et al.[3] found superior group performance on readiness assurance tests compared with individual efforts. These results suggest that TBL may foster constructive peer interaction and collaborative cognitive processing that enhances clinical decision-making.

However, a major finding from our review is the absence of genuine AI integration in any of the included studies. The term “AI-enhanced” was not operationalized in any included work, indicating a substantial research gap. The heterogeneity of assessment methods also limits our ability to draw strong conclusions. While some studies employed validated tools such as the Script Concordance Test[12] or Berlin Questionnaire,[13] others relied on subjective self-assessments,[1,16] confidence scales,[1] or nonspecific academic tests. Only a subset of studies directly measured clinical reasoning, highlighting the need for standardized, validated tools in future evaluations. AI can enhance TBL by providing intelligent feedback, tracking learner progress, adapting task difficulty based on performance, and automating formative assessments. These applications can amplify the strengths of TBL in improving diagnostic reasoning, hypothesis generation, and decision-making skills. Expected AI features within TBL include real-time learning analytics, NLP-enabled analysis of group discussions, AI-based grading rubrics for formative tasks, and intelligent team allocation based on learning styles or prior performance. Our search strategy focused solely on PubMed and specific Boolean logic, potentially omitting relevant literature from education databases or grey literature sources. Key TBL components—such as the readiness assurance process, peer discussion, and application tasks—promote metacognition and reasoning. Technology can reinforce these by facilitating interactive modules and digital simulations. Studies in nursing and engineering education support the efficacy of AI-assisted group work in fostering higher-order thinking.

The lack of AI integration represents the most critical gap across all included studies. None of the interventions employed machine learning models, natural language processing, or intelligent tutoring systems to personalize instruction or assess reasoning in real time. Future research should focus on developing and piloting AI-enhanced TBL modules that align with clinical reasoning frameworks and incorporate validated assessment tools. Limitations of this review include reliance on prescreened sources without broad database querying and the exclusion of grey literature. Our search was limited to PubMed-indexed articles within a specific age group and educational level. Expanding to include grey literature, allied health education, and broader AI-related keywords may reveal further insights. Future research should explore AI integration in TBL through tools such as adaptive learning modules, intelligent tutoring systems, and real-time analytics to enhance both learner engagement and diagnostic accuracy in clinical reasoning.

Conclusion

This systematic review demonstrates that team-based learning (TBL) is an effective pedagogical approach for enhancing clinical reasoning skills among medical students. However, despite the growing interest in educational technology, none of the reviewed studies integrated genuine artificial intelligence (AI) features into their TBL interventions. The digital platforms used—such as Tronclass, InteDashboard, and the Learning Activity Management System—lacked adaptive, intelligent functionalities typically associated with AI. This significant gap highlights an untapped opportunity for innovation in medical education. Future research should focus on the design and evaluation of AI-driven TBL frameworks that incorporate personalized feedback, adaptive content delivery, and real-time performance analytics. Such integration has the potential to further optimize clinical reasoning development and transform the learning experience for medical students.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.