Application of Data Mining to a Large Hearing-Aid Manufacturer's Dataset to Identify Possible Benefits for Clinicians, Manufacturers, and Users.

The dataset was heavily redacted for anonymity and commercial confidentiality, so age, gender, and which ear was fitted were not available. The audiogram values were those entered into fitting software, not necessarily the wearer's measured audiogram. Left and right devices were not directly linked, so the number of unique wearers could only be estimated as in excess of 150,000. Some devices appeared to have been used by more than one person, and the authors excluded records with large audiogram changes as possible loan devices. The analyses searched many combinations, so some findings may exist purely by chance. No patient benefit, satisfaction, or communication outcome was measured; the paper presents the work as a proof-of-concept.

The Medical Research Council funded the work. The dataset came from a hearing-aid manufacturer; the authors state their analysis was largely blind and not directed by the manufacturer.

Do not read the identified links between device style, audiogram values, and logged sound levels as proof that data mining improves hearing aid fitting or patient benefit. The paper is a proof-of-concept secondary analysis of redacted manufacturer logs, with no measured patient outcomes.