Effectiveness of an Edge AI-based Augmented Reality System for Hand Hygiene Training: a Multi-centre, Mixed-methods Cluster Randomised Controlled Trial

The importance of hand hygiene cannot be underestimated in both healthcare and community settings (Allegranzi et al., 2011; Mathur, 2011). The promotion of hand hygiene is the best practice in preventing health care-associated infection (WHO, 2009). It has been estimated that hand hygiene could reduce over 500 000 attributable deaths per year. It was found that the annual economic impact of health care-associated infection in the US was approximately US$6.5 billion in 2004 (WHO, 2009) and every US$1 spent on hand hygiene promotion could result in a US$23.7 benefit (Chen et al., 2011). Education plays a key role in setting a good practice base in hand hygiene.

The traditional training methods for hand hygiene include in-class lecture and practical, but they seldom receive individualised instructions and feedback due to limited manpower and facilities. Hence, there is an urgent need to design an automated and individualised training system. This proposed project aims to design and test an innovative education tool for hand hygiene, with the following specific objectives:

1. develop a prototype of the AI-empowered AR system for hand hygiene training and assessment;
2. conduct a multi-centre, mixed-methods non-inferiority randomised controlled trial (RCT) to test the effectiveness of this AI-based AR training system on knowledge and practice of hand hygiene in in in different populations in different settings, including undergraduate students of health-related disciplines, healthcare professionals and supporting staff in hospitals;
3. evaluate the effects of the AI-based AR training system and collect feedback from participants.
4. investigate participants' experiences, attitudes, and perceptions of hand hygiene training using user feedback survey and focus group interviews.

If demonstrated effective, this system can be integrated into the curriculum of students of health-related disciplines, healthcare professionals and supporting staff from hospitals as an e-learning approach. In future, this system can also be used at point-of-care for real-time monitoring and audit of healthcare workers in other healthcare settings. In addition to quantitative outcome measures, qualitative interviews will be conducted to explore students' learning experiences, attitudes, and perceived barriers regarding hand hygiene practice and the use of AI-based AR training tools.

Methods

Study design This is a mixed-methods study consisting of a multi-centre non-inferiority two-arm randomised controlled trial (RCT) and a qualitative component. The RCT will evaluate the effectiveness of the AI-based AR training system on students' knowledge and practice of hand hygiene, while the qualitative study will explore students' experiences, attitudes, and perceived barriers related to hand hygiene training through semi-structured interviews. There are three centres for this trial: the Hong Kong Polytechnic University, Chinese University of Hong Kong, and Tongji Hospital in Wuhan, China. We will recruit eligible undergraduates from two universities, and healthcare professionals/supporting staff from Tongji Hospital.

Subject recruitment We will recruit 480 undergraduate students from health-related disciplines from the Hong Kong Polytechnic University, and the Chinese University of Hong Kong (n=240); healthcare professionals and supporting staff in hospitals (n=240). Data collection will be conducted during September 2025 to December 2026. Participants will subscribe a time slot of training sessions (each session has up to 25 participants). All participants who attend the same session will be assigned to the same group by cluster randomisation. A statistician will assign the groups in advance by a random number generator. The participants will not be informed about their group before arriving the study sites.

We calculated the sample size for a cluster randomised controlled trial. We assume that the number of participants in each cluster ranges from 6 to 25, there will be approximately 10-30 clusters in each group. We also assume the standard deviation of subjects is 2.00, the intracluster correlation coefficient is 0.010, and the coefficient of variation of cluster sizes is 0.500 (Ahn et al. 2015). The total sample size of 480 can achieve >91% power to detect a difference between two group means (confidence interval -1.0 and 1.0) in a two-sided t-test with a significance level of 0.050.

Pre-intervention assessment During the site visit, the participants will be invited to sign a consent form first and fill in a questionnaire on knowledge and practice of hand hygiene (attached in Appendix 1).

Participants will need to put fluorescent powder on both of their hands and then perform hand washing using liquid soap without any instructions (1st HW attempt). The camera installed above the hand-wash basins will take videos of both hands (no face nor other parts of the body) during this procedure. After washing hands, participants will need to scan fluorescent powder remained on their hands in a hand scanner (The Semmelweis Scanner™). The recorded hand washing videos will be used to train the AI algorithms for automatic image processing and assessments. Two IPC experts will judge the quality of hand hygiene in these videos, which will be adopted as ground truth in image processing. The percentage of fluorescent gel residual on hands shown in the hand scanner will be used as an objective assessment for efficacy of hand washing in individual participants.

Interventions groups Group A (AI-based AR training system) Participants will use the AI-empowered AR training system to get individualised assessments of their hand wash performance. They will not be informed about the hand scanner results.

Group B (Hand scanner + Video training) Participants will use an existing education program that has been demonstrated effective in our previous study (unpublished). In brief, the participants will be informed about their hand scanner results during the first HW attempt. They will also watch a training video about the 7 steps of hand washing.

Post-intervention assessment The participants will take the second HW attempt by putting fluorescent powder on both of their hands again and then perform hand wash with videos recorded. After washing hands, participants will scan fluorescent powder remained on their hands again in a hand scanner.

Participants will also be asked to fill in a questionnaire on knowledge and practice of hand hygiene as part of post-intervention assessment.

Qualitative Study In addition to the randomised controlled trial, a qualitative component will be conducted to explore participants' experiences, attitudes, and perceived barriers toward hand hygiene training (see Appendix 2). A purposive subsample of participants from both intervention and control groups will be invited to join semi-structured interviews guided by the Theoretical Domains Framework (Cane et al., 2012). The interviews will cover topics such as learning experience, motivation, usability, knowledge, skills, and social influences. All interviews will be audio-recorded and transcribed verbatim. Data will be analysed using thematic analysis, with initial codes generated inductively and then organised into broader categories and themes. Themes will be iteratively reviewed and refined to ensure credibility and consistency. Written informed consent will be obtained prior to participation, and confidentiality and voluntary participation will be strictly observed. Each participant will get a 100 HKD coupon. The sample coupon signature form can be found in Appendix 3.

Randomisation, allocation concealment, and blinding Participants will be randomly assigned to the intervention groups through a process of randomisation performed by a statistician who will not be involved in subject recruitment. The participants will be blinded. The RAs involved in subject recruitment and the IPC experts who judge the quality of hand hygiene will be blinded to the groupings. The RAs and student assistants involved in hand hygiene video recording at the study site will not be blinded because they will be giving instructions to the participants.

Ethical consideration

It is anticipated that the education program will not pose any physical or psychological harm to the participants. There is a very low chance of skin allergy to fluorescent powder. If a participant shows any sign of skin allergy, she or he will instantly stop the trail. The contact details of either the ethical or the research in-charge personnel are clearly written in the information sheet.

Explanation of the study will be given to participants who are interested in participating in the study. Participants will also be given the opportunity to ask our project personnel questions. Furthermore, the information sheet including the purpose of the study, confidentiality of the data, and rights of the participants, will also be provided to the participants. The contact details of either the ethical or the research in-charge personnel are clearly written in the information sheet. If they agree to participate in the study, they are required to sign the consent form. Their participation is on a voluntary basis; they have the right to withdraw from the study at any time if they want to without consequence.

All the collected data will be kept confidentially, and only the research personnel can access and handle the data. No personal information of the participants will be collected except the student ID number. For the qualitative component, participants will also be asked for consent to participate in interviews. All interview data will be anonymised and handled confidentially.

Confidentiality and Security

Data collection will be anonymous, only analysed for specified purposes and based on the consent or other legitimate basis laid down by law. No personal information will be linked to image/video data. All the image/video data will be deleted one year after the project is completed. The Project will not involve any activities or results raising security issues.

References

Allegranzi, B., Bagheri Nejad, S., Combescure, C., Graafmans, W., Attar, H., Donaldson, L., \& Pittet, D. (2011). Burden of endemic health-care-associated infection in developing countries: systematic review and meta-analysis. Lancet, 377(9761), 228-241. https://doi.org/10.1016/S0140-6736(10)61458-4

Ahn, C., Heo, M., \& Zhang, S. (2015). Sample Size Calculations for Clustered and Longitudinal Outcomes in Clinical Research. CRC Press. New York.

Chen, Y. C., Sheng, W. H., Wang, J. T., Chang, S. C., Lin, H. C., Tien, K. L., Hsu, L. Y., \& Tsai, K. S. (2011). Effectiveness and limitations of hand hygiene promotion on decreasing healthcare-associated infections. PLoS One, 6(11), e27163. https://doi.org/10.1371/journal.pone.0027163

Mathur, P. (2011). Hand hygiene: back to the basics of infection control. Indian J Med Res, 134(5), 611-620. https://doi.org/10.4103/0971-5916.90985

WHO. (2009). In WHO Guidelines on Hand Hygiene in Health Care: First Global Patient Safety Challenge Clean Care Is Safer Care. https://www.ncbi.nlm.nih.gov/pubmed/23805438

Yang, L., Lu, Y., Cao, J., Huang, J., \& Zhang, M. (2020). E-Tree Learning: A Novel Decentralized Model Learning Framework for Edge AI. IEEE Internet of Things Journal.

Cane, J., O'Connor, D., \& Michie, S. (2012). Validation of the theoretical domains framework for use in behaviour change and implementation research. Implementation science, 7(1), 37.