Generating and using "big data" to identify hearing aid patterns of usage in order to optimise and personalise fitting.

The successful use of a hearing aid depends on many factors. Conventionally, a simple measure is taken of the hearing loss, the audiogram. The aid is then set up automatically using assumptions based on an "average" listener. Finally, fine tuning of the settings is performed mostly by verbal feedback to the clinician from the aid user in response to simulated listening situations, or, on repeat visits to the clinic, on verbal feedback from the user, based on their experience in "real-world" listening environments.

The "average" listener is a rare commodity. The pattern of hearing loss is nearly unique to each listener as it involves damage accumulated over many years through very different lifestyles. The audiogram is commonly recognised as being insufficient to define that pattern. Additionally, each listener varies in the range of environments from the quiet living room to the loud rock concert in which they expect their hearing aid to operate and restore their hearing to "normal". Other factors beyond the hearing loss also contribute to the success, or otherwise, of the device. For example, the ability of the brain to make sense of the resulting acoustic input depends on factors such as experience with the acoustic scenario as well as the residual cognitive ability. Cognitive ability declines with a long-term factor such as age and a short-term factor such as fatigue. Personalising a hearing aid to reflect the disabilities, abilities and demands of the wearer is therefore unlikely to be accomplished in a session of "fine tuning" in clinic time.

Modern hearing aids possess the ability to record the general characteristics of the acoustic environment in which the aid operates as well as the pattern of use by the wearer, such as what settings were used and for how long. These recordings can be made for periods of up to about one month. For reasons of privacy they do not record the actual acoustic signal. Along with knowledge of the hearing loss, this data provides a valuable tool for fine tuning of the hearing aid to the requirements of the wearer and monitoring of the aid's effectiveness.

When hearing aids are tested on a wearer in a clinical setting, it is common for the wearer to be seated while test sounds are presented from loudspeakers in fixed positions, either directly ahead, or distributed around the room. Real world use of hearing aids is far more dynamic: in practice both the wearer and the sound sources are moving relative to each other, and the wearer can additionally use head movements to re-direct their attention. Miniature gyroscopes and measures of position and inclination are routinely included in consumer electronic products such as tablets or "smartphones". These data, available in real time, are potentially useful measures of how the aid wearer is interacting with their acoustic environment.

The proposal here is to build a network of academics, clinicians and representatives from manufacturers of hearing aids and clinical diagnostic equipment to identify how to use this data to build a fuller picture of the aid user. This picture should take into account the nature of the hearing loss, the capabilities of the wearer, the situations in which the aid is to be used and the behaviour of the wearer when using the device. With a more accurate profile of the user, it will be possible to provide a more systematic but personalised fitting.

The proposal is to use a network of academics, clinicians and manufacturers of hearing aids and clinical diagnostic equipment to characterise a hearing aid user by recording suitable measures of
(1) the pattern and site of hearing loss
(2) the cognitive function of the aid wearer
(3) the experience of the wearer with using the hearing aid

in order to produce a complex picture of the user and their hearing loss in the form of a data set. Using "data mining" tools from the field of computer science, the intention is to generate novel inferences from individual as well as collections of data sets which will drive (a) personalised fittings and (b) improved characterisation of the real-world rather than laboratory performance of hearing aids. These inferences will vary in their complexity and initially should verify results that have been reported from laboratory studies, such as acclimatisation to the processing of the device. As a test of the sufficiency of the algorithms employed, there should also be new inferences generated that are beyond current thinking.
The network also will meet to discuss methods to incorporate future data streams that will be capable of being recorded by the aid. One such stream of potential is the dynamic behaviour of the aid wearer as they interact in a particular acoustic environment
In order to provide a "proof of concept" using the existing recording capabilities of a hearing aid, a small project is proposed that will generate inferences from a collection of data sets provided by some of the partner manufacturers.

The beneficiaries of this work will ultimately be users of hearing prostheses (hearing aids and cochlear implants). By acquiring more data about the user's hearing loss, cognitive abilities, acoustic environments and their dynamic behaviour within the environment, the aid fitting software, as well as aid operating software, can build up a bigger picture of both the abilities and disabilities of the user. Future development could lead to the aid dynamically varying its fitting according to these measures. The process of interpreting how these datasets characterise the demands of the user is crucial to improving the fitting of the hearing prosthesis. The use of computer science data-mining techniques is not only to generate interpretations of the data, but to add value by generating additional inferences about otherwise unseen relationships within the data. The intended impact is that a user is comfortable with using a piece of technology, but is unaware of the complexity that underpins it. In common with other "successful" technologies, such as tablet computers, this should lead to greater uptake of hearing prostheses from their currently low rate of only 15-20%. Improved uptake rates are seen as important to reduce social isolation and forestall the development of mental health problems and their associated costs.