Medication counseling within community pharmacies is crucial for managing chronic diseases, yet significant challenges regarding correctness and completeness remain in Jordan. Although generative artificial intelligence (AI) can be utilized for patient education, there is a lack of research on clinical impact and safety of AI in medication counseling conducted by pharmacists in real-world practice. The aim of this study is to evaluate the effect of pharmacist-supervised AI-assisted medication counseling on the correctness and completeness of counseling information and 30-day medication adherence among patients in Jordanian community pharmacies.
Materials and Methods: This pragmatic, two-arm cluster randomized controlled trial enrolled 136 adult patients across 16 community pharmacies in Jordan (8 clusters per arm). Pharmacists in the intervention arm used a standardized prompt strategy with ChatGPT® to generate counseling drafts, which were then verified and edited before delivery. The control arm provided usual counseling. Co-primary outcomes were correctness and completeness of counseling information (percentage scores based on blinded transcript analysis). Secondary outcomes included 30-day medication adherence (General Medication Adherence Scale \[GMAS\]), immediate patient understanding, and satisfaction. Data were analyzed using mixed-effects linear and logistic regression models.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
OTHER
Masking
QUADRUPLE
Enrollment
136
For all eligible patients in the intervention arm, the pharmacist performed the standard patient assessment and determined which medicine(s) needed counselling. Then, the pharmacist input a prompt in a de-identified format into ChatGPT®. The prompt was a request for an easy-to-understand counselling document with information regarding the indications for the medication, dosage, schedule, route, course, missed doses, possible side effects, important precautions, storage, and advice on taking the medicine as prescribed (Appendix A). The pharmacist ensured that the content generated by the AI was accurate and clear, making corrections where necessary, and then gave verbal counselling to the patient. The AI output was never provided to the patients without pharmacist evaluation. It is worth noting that pharmacists could also reject the AI output as inaccurate, insufficient, hazardous, and inappropriate altogether. Reproducibility was ensured through documenting the date and time, prompt te
Petra University
Amman, Jordan
Percentage of Applicable Counseling Domains Provided Correctly
Defined as the proportion of clinically applicable counseling domains communicated accurately during the encounter, compared with a medication-specific reference sheet. Scored on a 0-100 scale, calculated as (Number of applicable domains correctly informed / Total number of applicable domains) x 100.Correctness score= (Number of applicable domains
Time frame: day 0
Percentage of Essential Counseling Domains Addressed
Defined as the proportion of essential counseling domains that were addressed during the encounter. Scored on a 0-100 scale, calculated as (Number of applicable domains addressed / Total number of applicable domains) x 100.
Time frame: Day 0
Number of Counseling Deficiencies Categorized by Clinical Severity
The frequency of omitted or incorrect counseling information, independently assessed by a panel of pharmacists using a 3-point scale: Low Severity (minor wording issues), Moderate Severity (errors leading to sub-therapeutic effects), and High Severity (errors with high potential for significant patient harm).
Time frame: Day 0
Score on the General Medication Adherence Scale (GMAS)
Medication adherence assessed via telephone follow-up using the continuous total score from the General Medication Adherence Scale (GMAS). Higher scores indicate better medication adherence.
Time frame: 30 Days Post-Encounter
Number of Participants Achieving Good Adherence
The number of participants meeting the validated threshold for "good adherence" based on their GMAS survey responses.
Time frame: 30 Days Post-Encounter
Total Score on the Immediate Patient Understanding (Teach-Back) Assessment
A brief interviewer-administered understanding assessment based on teach-back principles. Scores range from 0 to 4, with higher scores indicating a better understanding of the medication.
Time frame: Day 0
Total Score on the Patient Satisfaction Questionnaire
A questionnaire covering clarity, usefulness, confidence, and overall satisfaction. Total scores range from 5 to 25, with higher scores indicating greater patient satisfaction.
Time frame: Day 0
Time Spent on Face-to-Face Counseling
Total face-to-face counseling time measured in minutes using audio timestamps from the start of counseling to completion.
Time frame: Day 0
Number of Encounters Based on AI Output Acceptance Level
The proportion of encounters in which the AI-generated counseling output was fully accepted, edited before delivery, or rejected outright by the pharmacist.
Time frame: Day 0
Number of AI-Related Discrepancies Identified
The frequency of detected AI inaccuracies prior to counseling, such as omitted counseling points, overly technical wording, or incomplete missed-dose advice.
Time frame: Day 0
Number of Clinical Near Misses and Safety Incidents
The number of encounters featuring a "near miss" (an AI error identified and corrected by the pharmacist before reaching the patient) or an "incident" (inaccurate information that actually reached the patient).
Time frame: Day 0
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.