Authors: David Allen1, Jessica Monaghan1
1National Acoustic Laboratories, Sydney, Australia
Background: The high reliance on bone conduction for the detection of ear disease, coupled with the difficulty of obtaining bone conduction thresholds remotely, is a barrier to the use of remote assessment.
Methods: Audiograms of 867316 patients assessed between 1956 and 2020 were labelled for significant bone conduction asymmetry between the ears (a difference of >15dB at one octave or >10dB at two adjacent octaves) or air-bone gap (ABG) in one or both ears (>10dB at one octave). Missing air conduction frequencies were imputed using 5-nearest neighbour imputation. Imputed air conduction thresholds of a randomly selected 90% training set were used as the input to a three-layer feed-forward neural network to predict the presence of bone conduction asymmetry or ABG in either ear.
Results: Of the 780,584 audiograms in the training set and the 86,732 in the test set, 48% in each set showed no asymmetry and no ABG. The neural network was able to detect bone conduction asymmetry (AUC = 0.82, TPR = 69.5%, Χ2 = 4661, p < .001), left ABG (AUC = 0.81, TPR = 66.3% Χ2 = 9500, p < .001) and right ABG (AUC = 0.81, TPR = 66.5% Χ2 = 9199, p < .001) significantly better than chance with a false positive rate of 20%.
Conclusions: Automated methods have some utility in screening for patients who would benefit from additional diagnostic audiometric assessment. However, the relatively low sensitivity of these methods suggests that the use of additional methods, such as client history questionnaires, is also likely to be beneficial.
A recent review stated that only about 2% of older adults in a large sample with hearing loss had conductive hearing loss. I was therefore very surprised to see in this huge sample that about half had asymmetry or an ABG. If almost all of those were asymmetry, I think the reader should be aware of that
I was surprised at the high incidence as well! In our sample, the rate of air-bone gap sufficient to warrant further audiological investigation was higher that that of a bone conduction asymmetry (around 17% had a bone conduction asymmetry, and around 30% had an air-bone gap in each ear). We note that these findings were not necessarily indicative of a clinically relevant conductive hearing loss – we are also identifying people who might have gaps that might be ruled inconsequential after further audiological investigation.