Background: Tinnitus is a heterogeneous condition. The aim of this study as to compare the online and hospital responses to the Spanish version of European School for Interdisciplinary Tinnitus Research screening-questionnaire (ESIT-SQ) in tinnitus individuals by an unsupervised age clustering. Methods: A cross-sectional study was performed including 434 white Spanish patients with chronic tinnitus to assess the demographic and clinical profile through the ESIT-SQ, with 204 outpatients and 230 individuals from an online survey; a K-means clustering algorithm was used to classify both responses according to age. Results: Online survey showed a high proportion of Meniere’s disease (MD) patients compared to both the general population and the outpatient cohort. The responses showed statistically significant differences between groups regarding education level, tinnitus-related hearing disorders (MD, hyperacusis), sleep difficulties, dyslipidemia, and other tinnitus characteristics, including duration, type of onset, the report of mitigating factors and the use of treatments. However, these differences were partially confirmed after adjusting for age. Conclusions: Self-reported tinnitus surveys are a low confidence source for tinnitus phenotyping. Additional clinical evaluation is needed for tinnitus research to reach the diagnosis. Age-based cluster analysis might help to better define clinical profiles and to compare responses in ESIT-SQ among subgroups of patients with tinnitus.

Chronic tinnitus, the perception of a phantom sound in the absence of corresponding stimulus, is a condition known to affect patients’ quality of life. Recent advances in mHealth have enabled patients to maintain a ‘disease journal’ of ecologically-valid momentary assessments, improving patients’ own awareness of their disease while also providing clinicians valuable data for research. In this study, we investigate the effect of non-personalised tips on patients’ perception of tinnitus, and on their continued use of the application. The data collected from the study involved three groups of patients that used the app for 16 weeks. Groups A & Y were exposed to feedback from the start of the study, while group B only received tips for the second half of the study. Groups A and Y were run by different supervisors and also differed in the number of hospital visits during the study. Users of Group A and B underwent assessment at baseline, mid-study, post-study and follow-up, while users of group Y were only assessed at baseline and post-study. It is seen that the users in group B use the app for longer, and also more often during the day. The answers of the users to the Ecological Momentary Assessments are seen to form clusters where the degree to which the tinnitus distress depends on tinnitus loudness varies. Additionally, cluster-level models were able to predict new unseen data with better accuracy than a single global model. This strengthens the argument that the discovered clusters really do reflect underlying patterns in disease expression.

Missing data is a common occurrence in the time series domain, for instance due to faulty sensors, server downtime or patients not attending their scheduled appointments. One of the best methods to impute these missing values is Multiple Imputations by Chained Equations (MICE) which has the drawback that it can only model linear relationships among the variables in a multivariate time series. The advancement of deep learning and its ability to model non-linear relationships among variables make it a promising candidate for time series imputation. This work proposes a modified Convolutional Denoising Autoencoder (CDA) based approach to impute multivariate time series data in combination with a preprocessing step that encodes time series data into 2D images using Gramian Angular Summation Field (GASF). We compare our approach against a standard feed-forward Multi Layer Perceptron (MLP) and MICE. All our experiments were performed on 5 UEA MTSC multivariate time series datasets, where 20 to 50% of the data was simulated to be missing completely at random. The CDA model outperforms all the other models in 4 out of 5 datasets and is tied for the best algorithm in the remaining case.

Some mHealth apps record user activity continuously and unobtrusively, while other apps rely by nature on user engagement and self-discipline: users are asked to enter data that cannot be assessed otherwise, e.g., on how they feel and what non-measurable symptoms they have. Over time, this leads to substantial differences in the length of the time series of recordings for the different users. In this study, we propose two algorithms for wellbeing-prediction from such time series, and we compare their performance on the users of a pilot study on diabetic patients - with time series length varying between 8 and 87 recordings.Our first approach learns a model from the few users, on which many recordings are available, and applies this model to predict the 2nd, 3rd, and so forth recording of users newly joining the mHealth platform. Our second approach rather exploits the similarity among the first few recordings of newly arriving users. Our results for the first approach indicate that the target variable for users who use the app for long are not predictive for users who use the app only for a short time. Our results for the second approach indicate that few initial recordings suffice to inform the predictive model and improve performance considerably.

Traditional active learning tries to identify instances for which the acquisition of the label increases model performance under budget constraints. Less research has been devoted to the task of actively acquiring feature values, whereupon both the instance and the feature must be selected intelligently and even less to a scenario where the instances arrive in a stream with feature drift. We propose an active feature acquisition strategy for data streams with feature drift, as well as an active feature acquisition evaluation framework. We also implement a baseline that chooses features randomly and compare the random approach against eight different methods in a scenario where we can acquire at most one feature at the time per instance and where all features are considered to cost the same. Our initial experiments on 9 different data sets, with 7 different degrees of missing features and 8 different budgets show that our developed methods outperform the random acquisition on 7 data sets and have a comparable performance on the remaining two.