Year: 2019
Type: letter
Abstract: To the Editor—Chen and colleagues [1] suggest that our prediction model of hearing loss among patients receiving aminoglycosides for the treatment of multidrug-resistant tuberculosis (MDR-TB) [2] should undergo validation in other cohorts, achieve a higher level of accuracy, and undergo additional analyses prior to clinical application. We agree with this assessment, but would like to clarify these concerns further. First, we acknowledge that ultrahigh-frequency audiometry is more clinically useful for the early detection of ototoxicity because typical manifestations of cochleotoxicity begin with ultrahigh-frequency hearing loss and are often undetected by standard audiometry testing [3, 4]. However, due to the small sample size of ultrahigh-frequency data in our patient sample, we were unable to develop a prediction model specifically for ultrahigh-frequency hearing loss. We do feel that validation of our prediction model for ultrahigh-frequency hearing loss provides important ancillary data indicating that our model performs reasonably well in detecting this form of hearing loss. But we acknowledge that there is a need for external validation and refinement of the prediction model to achieve better generalizability and practicality using both regular- and ultrahigh-frequency audiometric data. As Chen and colleagues [1] suggest, external validation would ideally use different population sets; this is an important direction for future research. With respect to accuracy, we agree that an area under the curve (AUC) of 0.69 does not suggest high predictive accuracy, but would argue against an arbitrary cutoff of 0.7 for “moderate” versus “poor” predictive accuracy. Especially in contexts such as MDR-TB treatment in sub-Saharan Africa—characterized by extreme limitations of both data and resources—even modest predictive accuracy can add important value. Furthermore, the AUC is a global measure, whereas the clinical value of a prediction rule is better described by the sensitivity and specificity of specific cutoffs that might be used. Specifically, as below (and as argued in the original publication), we believe our prediction rule can be useful in identifying patients with high predicted probabilities of hearing loss. Finally, in response to the helpful recommendation by Chen and colleagues [1], we performed a decision curve analysis to further evaluate the clinical utility of our prediction rule. The decision curve shows that the prediction model adds net benefit when using a threshold probability of 50% or higher to identify patients who might qualify for aminoglycoside-sparing regimens when access to those regimens is limited (Figure 1). Decision curve for prediction of aminoglycoside-induced hearing loss. Abbreviation: Pr, probability. In summary, although the corresponding AUC suggests only modest predictive accuracy, our prediction model may add value in identifying patients with MDR-TB who are at high risk of developing ototoxicity during treatment in clinical settings where access to aminoglycoside-sparing regimens is restricted. Future analyses to externally validate this model in other populations would be welcomed prior to implementing this prediction rule into clinical practice in other settings. Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding organizations/institutions. Financial support. This work was supported by the National Institute of Nursing Research (grant number F31 NR016910 to H. H.) of the National Institutes of Health; a Sigma Theta Tau International Global Nursing Research Grant; a Sigma Theta Tau International/Association of Nurses in AIDS Care Grant; a Global Korean Nursing Foundation Scientific Award; a Dr Scholl Foundation Dissertation Scholarship; and the Johns Hopkins Center for Global Health Established Field Placements Grant. Potential conflicts of interest. The authors: No reported conflicts of interest. Both authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.
Source: Clinical Infectious Diseases
DOI: 10.1093/cid/ciz447
Authors: Hyejeong Hong, David W Dowdy
Institution: Johns Hopkins University
Language: English
Subfield: Sensory Systems
Field: Neuroscience
Domain: Life Sciences
Open Access status: bronze
OpenAlex ID: W2947990371