gms | German Medical Science

Audiological databases contain different combinations of audiological measurements, experience of audiological experts, or groups of patients depending on the specialization of a certain clinic. With combined databases, a clinical decision-support system (CDSS) for audiology would become possible that relies on big data. For this purpose, the Common Audiological Functional Parameters (CAFPAs) were introduced as abstract, measurement-independent representation of audiological knowledge, which can be used as interpretable intermediate layer in a CDSS. In previous work, expert knowledge was collected to link CAFPAs to audiological measurements from a pre-clinical data set of Hörzentrum Oldenburg [1], and a classification was performed based on expert-estimated CAFPAs [2].

Furthermore, prediction models for CAFPAs were established by [3], and these CAFPA predictions were included into the existing classification framework to obtain a full-working prototype of a CDSS. Results show a comparable performance of predicted CAFPAs compared to expert CAFPAs. By analyzing weights of CAFPAs contributing to high performance, interpretability of the CAFPA prediction as well as classification was enabled. In summary, the predicted CAFPAs can well be employed in the classification, which provides the potential to apply the CDSS to individual patients.

However, the system is so far limited to one data set including speech audiometry outcomes obtained with the Göttingen sentence test (GÖSA). To extend the approach to additional clinical databases including different speech tests, such as the Freiburg monosyllabic speech test (FMST) which is often used to assess hearing device indication criteria according to clinical guidelines, model-based comparability of different speech tests is investigated. This enables consistency checks across the speech tests, as well as for expert links of different speech tests to CAFPAs.

For this, a model-based framework based on the speech intelligibility index (SII) in combination with Plomps model of attenuation (A) and distortion (D) component is used. By combining measured and modeled speech test conditions with knowledge about the patients audiogram, a hearing loss index is estimated based on which hearing device indication criteria can be compared. The general mechanics of the framework will be presented, along with first results on a data set containing FMST and GÖSA for the same patients, which allows a direct check of the desired translation between speech tests.

In the future, this approach will be extended to more speech tests, including different languages to compare clinical databases across countries. In addition, the hearing loss index will be integrated into the CAFPA concept.