R Geilleit1, Z Q Hen2, C Y Chong3, A P Loh4, N L Pang5, G M Peterson6, K C Ng2, A Huis7, D F de Korne8. 1. Medical Innovation and Care Transformation, KK Women's and Children's Hospital, SingHealth Duke - NUS Academic Medical Centre, Singapore; Radboud Institute for Health Sciences, Scientific Centre for Quality of Healthcare (IQ Healthcare), Radboud University Medical Centre, Nijmegen, The Netherlands. 2. Medical Innovation and Care Transformation, KK Women's and Children's Hospital, SingHealth Duke - NUS Academic Medical Centre, Singapore. 3. Infectious Diseases, Department of Paediatrics, KK Women's and Children's Hospital, Singapore; Paediatrics Academic Medical Program, Duke-NUS Medical School, Singapore; Yong Loo Lin School of Medicine, National University of Singapore, Singapore. 4. Department of Biomedical Engineering, National University Singapore, Singapore. 5. Quality, Safety and Risk Management, KK Women's and Children's Hospital, Singapore. 6. Health Services Innovation, School of Medicine, University of Tasmania, Australia. 7. Radboud Institute for Health Sciences, Scientific Centre for Quality of Healthcare (IQ Healthcare), Radboud University Medical Centre, Nijmegen, The Netherlands. 8. Medical Innovation and Care Transformation, KK Women's and Children's Hospital, SingHealth Duke - NUS Academic Medical Centre, Singapore; Erasmus School of Health Policy and Management, Erasmus University Rotterdam, The Netherlands; Health Services and Systems Research, Duke-NUS Medical School, Singapore. Electronic address: dirk.de.korne@kkh.com.sg.
Abstract
BACKGROUND: Various technologies have been developed to improve hand hygiene (HH) compliance in inpatient settings; however, little is known about the feasibility of machine learning technology for this purpose in outpatient clinics. AIM: To assess the effectiveness, user experiences, and costs of implementing a real-time HH notification machine learning system in outpatient clinics. METHODS: In our mixed methods study, a multi-disciplinary team co-created an infrared guided sensor system to automatically notify clinicians to perform HH just before first patient contact. Notification technology effects were measured by comparing HH compliance at baseline (without notifications) with real-time auditory notifications that continued till HH was performed (intervention I) or notifications lasting 15 s (intervention II). User experiences were collected during daily briefings and semi-structured interviews. Costs of implementation of the system were calculated and compared to the current observational auditing programme. FINDINGS: Average baseline HH performance before first patient contact was 53.8%. With real-time auditory notifications that continued till HH was performed, overall HH performance increased to 100% (P < 0.001). With auditory notifications of a maximum duration of 15 s, HH performance was 80.4% (P < 0.001). Users emphasized the relevance of real-time notification and contributed to technical feasibility improvements that were implemented in the prototype. Annual running costs for the machine learning system were estimated to be 46% lower than the observational auditing programme. CONCLUSION: Machine learning technology that enables real-time HH notification provides a promising cost-effective approach to both improving and monitoring HH, and deserves further development in outpatient settings.
BACKGROUND: Various technologies have been developed to improve hand hygiene (HH) compliance in inpatient settings; however, little is known about the feasibility of machine learning technology for this purpose in outpatient clinics. AIM: To assess the effectiveness, user experiences, and costs of implementing a real-time HH notification machine learning system in outpatient clinics. METHODS: In our mixed methods study, a multi-disciplinary team co-created an infrared guided sensor system to automatically notify clinicians to perform HH just before first patient contact. Notification technology effects were measured by comparing HH compliance at baseline (without notifications) with real-time auditory notifications that continued till HH was performed (intervention I) or notifications lasting 15 s (intervention II). User experiences were collected during daily briefings and semi-structured interviews. Costs of implementation of the system were calculated and compared to the current observational auditing programme. FINDINGS: Average baseline HH performance before first patient contact was 53.8%. With real-time auditory notifications that continued till HH was performed, overall HH performance increased to 100% (P < 0.001). With auditory notifications of a maximum duration of 15 s, HH performance was 80.4% (P < 0.001). Users emphasized the relevance of real-time notification and contributed to technical feasibility improvements that were implemented in the prototype. Annual running costs for the machine learning system were estimated to be 46% lower than the observational auditing programme. CONCLUSION: Machine learning technology that enables real-time HH notification provides a promising cost-effective approach to both improving and monitoring HH, and deserves further development in outpatient settings.
Authors: David W Bates; David Levine; Ania Syrowatka; Masha Kuznetsova; Kelly Jean Thomas Craig; Angela Rui; Gretchen Purcell Jackson; Kyu Rhee Journal: NPJ Digit Med Date: 2021-03-19