A Preliminary Study of Hand Hygiene Compliance Characteristics with Machine Learning Methods

Increasing hospital re-admission rates due to Hospital Acquired Infections (HAIs) are a concern at many healthcare facilities. To prevent the spread of HAIs, caregivers are expected to comply with recommended hand hygiene guidelines, which require reliable and timely hand hygiene compliance measurement systems. The current standard practice of monitoring compliance involves the direct observation of caregivers’ hand cleaning as they enter or exit a patient room by a trained observer, which can be time-consuming, resource-intensive, and subject to bias. To alleviate the heavy manual effort and reduce errors, this paper studies the characteristics of compliance that could be used to assist the direct observation approach by deciding when and where to station manual auditors and to improve compliance by providing just-in-time alerts or potentially recommending training materials to non-compliant staff. The paper analyzes location and handwashing station activation data from a 30-bed intensive care unit (ICU) and uses machine learning to assess if location, time-based factors, or other data can be used to predict handwashing non-compliance events. The results show that a care provider’s entry compliance is highly indicative of the same provider’s exit compliance and that compliance of the most recent patient room visit can also predict entry compliance of a staff member’s current patient room visit