Client Profile
Anglo American has always been a pioneer in safeguarding the health of their workforce. For this global company with more than 50,000 mine workers, merely being compliant with labour law and occupational health and safety regulations is not good enough, as they are on a mission to achieve zero harm in the workplace.
The Challenge
Traditionally, hearing tests to screen for noise-induced hearing loss (NIHL) are interpreted manually and focus on lagging indicators such as Standard Threshold Shift (STS) and Percentage Loss of Hearing (PLH). This screening method is costly, time-consuming, and has limited reproducibility as it relies on subjective interpretation. Moving towards more efficient, prevention-focused management of NIHL required a new approach. The challenge was to design and deploy a solution that could be automated while at the same time leveraged the expertise of occupational health experts whose intuition for recognising early signs of NIHL is difficult to distil into a rule-based decision algorithm.
Our Approach
We developed and trained a machine learning algorithm (recurrent neural network) on an external dataset of 2284 audiometric tests that were labelled by experts as either showing bilateral noise-induced hearing loss (418 positive cases) or not (1866 negative cases). To assess performance, we deployed this new detection algorithm on a set of new, previously unseen audiometry data, and compared its accuracy (i.e. correspondence to expert opinion) against that of alternative approaches, including the existence of a unilateral or bilateral notch in the audiogram, a standard threshold shift (STS) and a percentage loss of hearing (PLH) > 3.2%.
The Impact
The recurrent neural network performed better than the alternative methods: its AUC (area under the Receiver-Operating Characteristic curve) was 93%, compared to 60%, 66% and 87% for PLH, STS and bilateral notch respectively. Furthermore, we extended the model to allow the user independent control over the sensitivity and specificity of the model, by creating a third “grey zone” in between NIHL and no NIHL. For large mining companies like Anglo American with thousands of workers at risk of NIHL, our solution enables proactive, prevention-oriented management of NIHL while simultaneously saving substantial amounts of time and money.
“The models developed by the Wimmy team have accelerated the transformation from what used to be a fragmented, reactive and descriptive approach, to an integrated health system that is forward-looking, action-focused and preventative.”
