Sam Harkus1, Dr Jessica Monaghan2, Mr Jason Gavrilis2, Ms Jessma Nash2, Ms Meagan Ward2, Dr Isabel O’Keeffe2, Ms Vivienne Marnane2, Ms Letitia Campbell3, Mr Trumaine Rankmore4, Mr Luke Austin1
1Hearing Australia, Macquarie University, Australia, 2National Acoustic Laboratories, Macquarie University, Australia, 3Kalwun Development Corporation, Gold Coast, Australia, 4University of Newcastle, Newcastle, Australia
Biography:
Sam is a non-Indigenous audiologist, living on Gadigal land in Sydney, NSW. She has worked with Hearing Australia in clinical and project roles, and in research at the National Acoustic Laboratories. The common thread in this work is equitable access to ear health and hearing care for young Aboriginal and Torres Strait Islander children.
Abstract
Background
At any time, 50-90% of young Aboriginal and Torres Strait Islander children have OM, 10% of whom have a persistent case. Persistent OM (pOM)-related hearing loss affects wellbeing, development, literacy, and learning. New, primary healthcare Ear Health and Hearing Check consensus recommendations for young Aboriginal and Torres Strait Islander children are now available. These recommendations guide timing and components of the Check. However, identification of persistent OM still involves repeat review in 3 months, potentially prolonging auditory deprivation and delaying remediation.
Methods
We undertook preliminary machine-learning work using retrospective, longer-term service data to investigate how Ear Health and Hearing Check results could be weighted and combined to enhance accuracy of identification of pOM-related hearing loss ≥30dB HL, based on a single set of results. Input variables included categorical results for parent/caregiver concern, PLUM listening skills results, otoscopy and tympanometry.
Results
We identified a set of results important to accurate identification of children with and without the target hearing loss. This approach resulted in better accuracy (Sensitivity 91%, Specificity 91%, PPV 71%) compared to the accuracy data for each of the individual components.
Outcomes and implications.
Using machine learning, we have identified clinical results that are important to identifying children with pOM related hearing loss ≥ 30dB HL, and to minimising misdiagnoses. The first model has high overall accuracy. These findings may inform development of a tool that enhances practitioners’ ability to identify persistent presentations more accurately, alleviating the need for 3-month reviews for a proportion of children.