Page 712 - ISC PROCEEDINGS 21.4

APPLICATION OF GENERATIVE AI CHATBOTS IN PHARMACY EDUCATION: A
CASE STUDY AT A PRIVATE UNIVERSITY IN VIETNAM

3
2
1
4
Tran Thi Thu Trang* , Nguyen Thuy Van , Vu Thi Minh Hoa , Vu Thi Tram ,
Nguyen Huynh Minh Long , Nguyen Thuy Nga , Duong Trong Thanh 7
5
6
1, 2 Thanh Do University, Hanoi, Vietnam.
3 CMC University, Hanoi, Vietnam.
4, 5 Hoa Binh University, Hanoi, Vietnam.
6 Nam Dinh University of Nursing, Ninh Binh, Vietnam.
7 Military University of Culture and Arts, Hanoi, Vietnam.
(*E-mail: tranthutrang235@gmail.com)

ABSTRACT
Background: Interpreting pharmacoepidemiological indicators is an essential
graduation skill in evidence-based pharmacy education, yet few studies in Vietnam have
systematically evaluated tools to bridge this gap. Objective: This study aimed to evaluate
the effectiveness of a structured generative AI chatbot intervention on the
pharmacoepidemiological indicator interpretation competency of fourth-year pharmacy
students at Thanh Do University, and to compare outcomes between ChatGPT and
Gemini groups. Methods: A one-group pretest-posttest design (no parallel control group)
was applied to 37 fourth-year pharmacy students (8 male, 29 female; age 22.1 ± 0.8 years)
in Semester I/2025-2026. The two-week intervention comprised two structured sessions
using pre-designed conversational task protocols applied to three Open Access original
articles (SCCS, meta-analysis, RCT). Students were randomly assigned to ChatGPT (n = 19)
or Gemini (n = 18). Assessment used parallel- form questionnaires (15 items each)
administered at T0 and T1. Results: All seven outcome indicators improved significantly (p
< 0.001), with large effect sizes observed across all measures. Correct interpretation of
aOR/IRR increased from 27% to 86%; NNT/ARR from 11% to 73%; I² in meta-analysis from
8% to 68%. Task completion time decreased by 83.7% and self-confidence improved from
1.8 to 4.1 out of 5.0. No statistically significant difference was detected between ChatGPT
and Gemini on any indicator (Fisher’s exact: p = 0.71-1.00; Mann - Whitney U: p = 0.54).
Conclusion: This study provides initial evidence that a structured, instructor-guided
generative AI chatbot protocol can significantly improve fourth-year pharmacy students’
ability to interpret pharmacoepidemiological indicators. No statistically significant
differences were found between the ChatGPT and Gemini groups. These findings suggest
that well-designed prompts and instructional scaffolding play a more important role than
the choice of AI platform. The approach appears feasible for wider implementation in
Vietnamese pharmacy education.
Keywords: Generative AI; pharmacoepidemiology; chatbot; prompt engineering;
Vietnam.

1. Introduction
Evidence-based pharmacy practice requires not only familiarity with clinical
guidelines but also the capacity to read, critically appraise, and apply findings from
pharmacoepidemiological studies, a competency increasingly recognized as foundational

711