Background: There is a huge variability in the way individuals with tinnitus respond to interventions. These experiential variations together with a range of associated etiologies, contribute to tinnitus being a highly heterogeneous condition. Despite this heterogeneity, a “one size fits all” approach is taken when making management recommendations. Although there are various management approaches, not all are equally effective. Psychological approaches such as cognitive behavioral therapy (CBT) have the most evidence-base.
Objectives: Managing tinnitus is challenging due to the significant variations in tinnitus experiences and treatment success. Tailored interventions based on individual tinnitus profiles may improve outcomes. Predictive models of treatment success are, however, lacking. The current study aimed to used exploratory data mining techniques (i.e., decision tree models) to identify the variables associated with treatment success for an Internet-based cognitive behavioral therapy (ICBT) for tinnitus.
Methods: Individuals (n = 228) who underwent ICBT in three separate clinical trials were included in this analysis. The primary outcome variable was reducing 13 points in tinnitus severity as measured by the Tinnitus Functional Index following the intervention. Predictor variables included demographic characteristics, tinnitus, and hearing-related variables, and clinical factors (i.e., anxiety, depression, insomnia, hyperacusis, hearing disability, cognitive function, and life satisfaction). Analyses were undertaken using various exploratory machine learning algorithms to identify the most suitable variable. Six decision tree models were implemented, namely Classification and decision trees (CART), C5.0, Gradient Boosting, AdaBoost algorithm, eXtreme Gradient Boosting and Random Forest. The SHapley Additive exPlanations (SHAP) framework was applied to the two best models to identify the relative predictor importance.
Results: Of the six decision tree models, CART [accuracy of 70.7% (SD=2.4) sensitivity of 74.0% (SD=5.5), specificity of 64% (SD=3.7), and area under the receiver operating characteristic curve (AUC) )] and Gradient boosting [accuracy of 71.8% (SD=1.5), sensitivity of 78.3% (SD=2.8), specificity of 58.7% (SD=4.2), and AUC 0.68 (SD= were found to be the best predictive models. Although the other models had an acceptable accuracy (ranged between 56.3 to 66.7%) and sensitivity (varied between 68.6 to 77.9%), they all had relatively weak specificity (varied between 31.1 to 50.0%) and AUC (varied between .52 to .62). Higher education level was the most influencing factors in the ICBT outcome. The CART decision tree model identified three participant groups who had at least 85% success probability following undertaking ICBT.
Conclusions: In this study, decision tree models, especially the CART and Gradient Boosting models, appear to be promising in predicting the ICBT outcomes. Their predictive power may be improved by using larger sample sizes and including a wider range of predictive factors in future studies.