Using keywords to predict the need for an audiogram: an analysis of referral letters using logistic regression

Otolaryngology clinics are often booked without considering the distribution of work for doctors and audiologists. This causes inefficiencies of time and human resources. This may be due to clinics being booked before referrals have been triaged to identify whether a hearing test, known as a pure tone audiogram (PTA), is indicated. A model that can predict the need for PTA without clinician-led triage could be useful to address these booking issues.

To establish if it is possible to predict whether a referred patient requires a PTA based on occurrences of words in the referral letter.

Binary logistic regression analysis of 500 letters of referral for otolaryngology outpatients. The derived model was then tested on 50 referral letters. All the referral letters were reviewed by clinicians and classified according to whether or not a PTA would be required.

The regression model correctly predicted requirement for a PTA in 42 of 50 referral letters (84%), with a sensitivity of 91% and specificity of 82%.

The model is able to predict requirement for a PTA from referral letters with a high degree of accuracy. This method may have a role in assisting administrative/clerical staff or non-specialist clinicians to book appropriate ear, nose and throat (ENT) clinic appointments, with or without a PTA. As a result, workload would be distributed more evenly, through the clinic for both otolaryngologists and audiologists, increasing efficiency.