Search

Wearables are commercially available devices allowing continuous monitoring of users' health parameters. Their easy availability, increasing accuracy and functionality render them relevant for medical practice, specifically for longitudinal monitoring. There are clear benefits for the health care system, such as the opportunity of timely interventions by monitoring a patient during his daily life, resulting in a cost reduction in medical care and improved patient well-being. However, some tools are essential to enable the application of wearables in medical daily practice. For example, there is a need for software solutions that allow clinicians to quickly and easily analyze data from devices of their patients. The goal of this study was to develop a dashboard for physicians, which allows rapid data interpretation of longitudinal data from the Apple Watch. The prototype dashboard is an interactive web-based visualization platform utilizing Plotly. The dashboard displays the most important parameters like heart rate, steps per day, activity, exercise collected by the Apple Watch in a user-friendly and accessible way. Clear visualization makes it easy to identify trends or deviations in the data and see how these changes in daily behaviour affect patients' health. Our software is a key component to monitor patients with heart failure who participate in the HiGHmed use case cardiology project.

The HiGHmed consortium aims to create a shared information governance framework to integrate clinical routine data. One challenge is the replacement of unstructured reporting (e.g. doctoral letters) with structured reporting in clinical routine. The Heidelberg cardiology department evaluates dynamic PDF forms for structured data reporting of heart failure (HF) patients. In this use case, we aim to identify potential caveats or shortcomings in data processing at an early stage. We employed data mining strategies to detect patterns related to incomplete or false data, which we found to be present among all data types. We then discuss the characteristics of the baseline patient cohort in Heidelberg to find out about specific peculiarities and potential biases, which may be site-specific. Briefly, our patient population is predominantly male (67%), NYHA III are the most common severity classes, NYHA IV is missing entirely. Most patients have a dilated cardiomyopathy (DCM) or coronary heart disease (CHD) diagnosed as their cause of HF. Finally, we also analyzed how comorbidities and risk factors relate to specific disease entities of heart failure patients. Family anamnesis was more frequent among cardiomyopathy patients than among CHD patients, who show a more dominating presence of dyslipidemia instead. Generally, the most dominant risk factor was arterial hypertension, while at the other end of the scale alcoholism appears to be underreported.

One of the major obstacles for research on German medical reports is the lack of de-identified medical corpora. Previous de-identification tasks focused on non-German medical texts, which raised the demand for an in-depth evaluation of de-identification methods on German medical texts. Because of remarkable advancements in natural language processing using supervised machine learning methods on limited training data, we evaluated them for the first time on German medical reports using our annotated data set consisting of 113 medical reports from the cardiology domain. We applied state-of-the-art deep learning methods using pre-trained models as input to a bidirectional LSTM network and well-established conditional random fields for de-identification of German medical reports. We performed an extensive evaluation for de-identification and multiclass named entity recognition. Using rule based and out of domain machine learning methods as a baseline, the conditional random field improved F2-score from 70 to 93% for de-identification, the neural approach reached 96% in F2-score while keeping balanced precision and recall rates. These results show, that state-of-the-art machine learning methods can play a crucial role in de-identification of German medical reports.

Translational research in the medical sector is dependent on clear communication between all participants. Visualization helps to represent data from different sources in a comprehensible way across disciplines. Existing tools for clinical data management are usually monolithic and technically challenging to set up, others require a transformation into specific data models while providing mostly non-interactive visualizations or being specialized to very particular use cases. Statistical programming languages (R, Julia) on the other hand offer great flexibility in data analytics, but are harder to access for clinicians with little to no programming expertise. Our software, the Medical Data Explorer (MedEx), aims to fill this gap as light-weight, intuitive, web-based solution with simple data import routes. We couple a modern dynamic web interface with an in-memory database solution for near real-time responsiveness. MedEx provides multiple visualization options (Scatterplot, correlation heatmap, bar chart, grouped boxplot, grouped histogram, coplot) to get an easy overview on the loaded data as well as to perform pattern discovery and elementary statistics. We demonstrate the utility of MedEx, by example, on data from the cardiology research warehouse of Heidelberg University Hospital. In summary, our tool empowers clinicians to conduct their own interactive exploratory data analysis.

Medical texts are a vast resource for medical and computational research. In contrast to newswire or wikipedia texts medical texts need to be de-identified before making them accessible to a wider NLP research community. We created a prototype for German medical text de-identification and named entity recognition using a three-step approach. First, we used well known rule-based models based on regular expressions and gazetteers, second we used a spelling variant detector based on Levenshtein distance, exploiting the fact that the medical texts contain semi-structured headers including sensible personal data, and third we trained a named entity recognition model on out of domain data to add statistical capabilities to our prototype. Using a baseline based on regular expressions and gazetteers we could improve F2-score from 78% to 85% for de-identification. Our prototype is a first step for further research on German medical text de-identification and could show that using spelling variant detection and out of domain trained statistical models can improve de-identification performance significantly.