1123. Recognizing Hand Hygiene Technique from Accelerometry Data: a First Step Towards a New Standard of Hand Hygiene Measurement
Session: Oral Abstract Session: Improving Healthcare Worker Hand Hygiene Adherence
Friday, October 4, 2013: 3:00 PM
Room: The Moscone Center: 200-212

Many interventions focus on increasing hand hygiene (HH) rates, that is, whether HH is performed or not. Yet HH technique (or how HH is performed) may dramatically impact the effectiveness of hand sanitizer.  Unfortunately, HH technique is difficult to measure in real-world settings. The goal of this pilot project is to develop an approach to studying HH technique using wrist-based 3 dimensional accelerometers (3DAs). 


To measure HH technique we used motes (small programmable, battery-powered wireless devices with 3DAs). Motes were worn on the wrists of healthcare professionals (HCPs). Another mote was affixed to a hand sanitizer dispenser. This mote instructs the wrist-based motes to record accelerometer data for 10 seconds after a dispensing event.

We gathered measurements of two types of hand movements from 22 different HCPs: (1) wild-type movement – what HCPs do normally when practicing HH – without a finger nail scrubbing action; (2) a specific motion designed to distribute hand sanitizer to the HCPs finger nails as per WHO guidelines.

Recorded data consist of acceleration readings along the x, y, and z axes for each hand. Data are processed to create features used to train a neural network (NN). Specifically, we examined the raw readings, switches (when one axis has a higher value than another in one reading and a lower value than the other in the next), and peaks (when readings change by more than a prespecified threshold). Our features were the mean, standard deviation, discrete Fourier transform, number of switches, and distance between peaks for each axis and the number of peaks, average distance between them, total number of switches, and mean and standard deviation of the distance between switches for all axes.


Using a NN and 10-fold cross validation to choose between the two types we were able to distinguish between wild-type HH and the fingertip scrub with a precision of 91.8% and recall of 91.8%. We correctly identify 92.6% of fingertip scrub samples and 90.9% of wild-type HH technique samples.


HH technique is likely to be an important component of HH effectiveness. However, without a reliable method to distinguish between specific HH movements, it is difficult to study HH technique. We have developed a novel approach using motes to measure and differentiate HH technique.

Valerie Galluzzi1, James Arbogast, PhD2, Ted Herman, PhD1, Danny Parker, PhD3, Alberto Segre, PhD1, David Shumaker4 and Philip M. Polgreen, MD5, (1)University of Iowa, Iowa City, IA, (2)GOJO Inductries, Inc, Akron, OH, (3)GTD Unlimted, Oxford, MS, (4)GOJO Industries, Akron, OH, (5)Division of Infectious Diseases, Department of Internal Medicine, University of Iowa Carver College of Medicine, Iowa City, IA


V. Galluzzi, None

J. Arbogast, GOJO Industries: Employee, Salary

T. Herman, GOJO Industries: Grant Investigator, Grant recipient
Medline Industries: Grant Investigator, Grant recipient

D. Parker, None

A. Segre, GOJO Industries: Grant Investigator, Grant recipient
Medline Industries: Grant Investigator, Grant recipient

D. Shumaker, GOJO Industries: Employee, Salary

P. M. Polgreen, GOJO Industries: Grant Investigator, Grant recipient
Medline Industries: Grant Investigator, Grant recipient

Findings in the abstracts are embargoed until 12:01 a.m. PST, Oct. 2nd with the exception of research findings presented at the IDWeek press conferences.