Predictive Sensitivity and Concordance of Machine-learning Tools for Diagnosing DFNA9 in a Large Series of p.Pro51Ser Variant Carriers in the COCH-gene.

Objective: In this study we aimed to evaluate the predictive cross-sectional sensitivity and longitudinal concordance of a machine-learning algorithm in a series of genetically confirmed p.(Pro51Ser) variant carriers (DFNA9).

Study Design: Cross-sectional study.

Setting: Tertiary and secondary referral center.

Patients: Audiograms of 111 subjects with the p.(Pro51Ser) mutation in the COCH-gene were analyzed cross-sectionally. A subset of 17 subjects with repeated audiograms were used for longitudinal analysis.

Interventions: All audiological thresholds were run through the web-based AudioGene v4.0 software.

Main Outcome Measures: Sensitivity for accurate prediction of DFNA9 for cross-sectional data and concordance of correct prediction for longitudinal auditory data.

Results: DFNA9 was predicted with a sensitivity of 93.7% in a series of 222 cross-sectionally collected audiological thresholds (76.1% as first gene locus). When using the hearing thresholds of the best ear, the sensitivity was 94.6%. The sensitivity was significantly higher in DFNA9 patients aged younger than 40 and aged 60 years or older, compared to the age group of 40 to 59 years, with resp. 97.6% (p < 0.0001) and 98.8% (p < 0.0001) accurate predictions. An average concordance of 91.6% was found to show the same response in all successive longitudinal audiometric data per patient.

Conclusions: Audioprofiling software can accurately predict DFNA9 in an area with a high prevalence of confirmed carriers of the p.(Pro51Ser) variant in the COCH-gene. This algorithm yields high promises for helping clinicians in directing genetic testing in case of a strong family history of progressive hearing loss, especially for very young and old carriers.