The rapidly growing quantity of health data presents researchers with ample opportunity for innovation. At the same time, exploitation of the value of Big Data poses various ethical challenges that must be addressed in order to fulfil the requirements of responsible research and innovation (Gerke et al. 2020; Howe III and Elenberg 2020). Data sovereignty and its principles of self-determination and informed consent are central goals in this endeavor. However, their consistent implementation has enormous consequences for the collection and processing of data in practice, especially given the complexity and growth of data in healthcare, which implies that artificial intelligence (AI) will increasingly be applied in the field due to its potential to unlock relevant, but previously hidden, information from the growing number of data (Jiang et al. 2017). Consequently, there is a need for ethically sound guidelines to help determine how data sovereignty and informed consent can be implemented in clinical research. Using the method of a narrative literature review combined with a design thinking approach, this paper aims to contribute to the literature by answering the following research question: What are the practical requirements for the thorough implementation of data sovereignty and informed consent in healthcare? We show that privacy-preserving technologies, human-centered usability and interaction design, explainable and trustworthy AI, user acceptance and trust, patient involvement, and effective legislation are key requirements for data sovereignty and self-determination in clinical research. We outline the implications for the development of IT solutions in the German healthcare system.

Currently, the medical sector sees a lot of digitization efforts. Besides the already available and growing amount of data there is still a need to receive data from patients. A special form of data donation are medical questionnaires. They come with a fixed set of questions regarding medical conditions and can be performed regularly or once. Such data donations show promising potential for research but come with open questions regarding privacy. There are many examples that show that re-identification is even possible with anonymized data. To mitigate such risks privacy enhancing technologies like Differential Privacy (DP) can be used. This paper presents a procedure for privacy preserving questionnaires using an implementation of DP called RAPPOR. Through its Randomized Response mechanism RAPPOR is well suited to be used with questionnaires. To proof feasibility of this procedure various experiments are conducted that show that this approach can improve privacy while still retaining a good level of utility.

Numerous business workflows involve printed forms, such as invoices or receipts, which are often manually digitalized to persistently search or store the data. As hardware scanners are costly and inflexible, smartphones are increasingly used for digitalization. Here, processing algorithms need to deal with prevailing environmental factors, such as shadows or crumples. Current state-of-the-art approaches learn supervised image dewarping models based on pairs of raw images and rectification meshes. The available results show promising predictive accuracies for dewarping, but generated errors still lead to sub-optimal information retrieval. In this paper, we explore the potential of improving dewarping models using additional, structured information in the form of invoice templates. We provide two core contributions: (1) a novel dataset, referred to as Inv3D, comprising synthetic and real-world high-resolution invoice images with structural templates, rectification meshes, and a multiplicity of per-pixel supervision signals and (2) a novel image dewarping algorithm, which extends the state-of-the-art approach GeoTr to leverage structural templates using attention. Our extensive evaluation includes an implementation of DewarpNet and shows that exploiting structured templates can improve the performance for image dewarping. We report superior performance for the proposed algorithm on our new benchmark for all metrics, including an improved local distortion of 26.1 %. We made our new dataset and all code publicly available at https://felixhertlein.github.io/inv3d .

Background Personalized mRNA vaccines are promising new therapeutic options for patients with cancer. Because mRNA vaccines are not yet approved for first-line therapy, the vaccines are presently applied to individuals that received prior therapies that can have immunocompromising effects. There is a need to address how prior treatments impact mRNA vaccine outcomes.Method Therefore, we analyzed the response to BioNTech/Pfizer’s anti-SARS-CoV-2 mRNA vaccine in 237 oncology outpatients, which cover a broad spectrum of hematologic malignancies and solid tumors and a variety of treatments. Patients were stratified by the time interval between the last treatment and first vaccination and by the presence or absence of florid tumors and IgG titers and T cell responses were analyzed 14 days after the second vaccination.Results Regardless of the last treatment time point, our data indicate that vaccination responses in patients with checkpoint inhibition were comparable to healthy controls. In contrast, patients after chemotherapy or cortisone therapy did not develop an immune response until 6 months after the last systemic therapy and patients after Cht-immune checkpoint inhibitor and tyrosine kinase inhibitor therapy only after 12 months.Conclusion Accordingly, our data support that timing of mRNA-based therapy is critical and we suggest that at least a 6-months or 12-months waiting interval should be observed before mRNA vaccination in systemically treated patients.

Screening mammography is a widely used approach for early breast cancer detection, effectively increasing the survival rate of affected patients. According to the Food and Drug Administration's Mammography Quality Standards Act and Program statistics, approximately 39 million mammography procedures are performed in the United States each year. Therefore, breast cancer screening is among the most common radiological tasks. Interpretation of screening mammograms by a specialist radiologist includes primarily the review of breast positioning quality, which is a key factor affecting the sensitivity of mammography and thus the diagnostic performance. Each mammogram with inadequate positioning may lead to a missed cancer or, in case of false positive signal interpretation, to follow-up activities, increased emotional burden and potential over-therapy and must be repeated, requiring the return of the patient. In this study, we have developed deep convolutional neuronal networks to differentiate mammograms with inadequate breast positioning from the adequate ones. The aim of the proposed automated positioning quality evaluation is to assist radiology technologists in detecting poorly positioned mammograms during patient visits, improve mammography performance, and decrease the recall rate. The implemented models have achieved 96.5% accuracy in cranio-caudal view classification and 93.3% accuracy in mediolateral oblique view regarding breast positioning quality. In addition to these results, we developed a software module that allows the study to be applied in practice by presenting the implemented model predictions and informing the technologist about the missing quality criteria.

To date, targeted tyrosine kinase inhibitors have been approved for FGFR2 and FGFR3 fusions (pemigatinib and erdafitinib, respectively), but the importance of FGFR2 mutations for transformation activity and as a druggable gene variant with response to different FGFR inhibitors is poorly understood. FGFR2 inhibitors present a mainstay of treatment for locally advanced or metastatic intrahepatic cholangiocellular carcinoma (iCCA). A 74-year-old male was diagnosed with iCCA in liver segments seven and eight with infiltration of the hepatic veins and inferior vena cava revealed a C382R mutation of the intramembrane domain of FGRR2 receptor. We performed an in-silico study to understand the potential mode-of-action of the mutant FGFR2 targets. Based on experimentally determined structures we then used a structure generated by AlphaFold2 as the variation in question is located at a position not determined well in the experiments. This revealed that the C382R mutation is located in the trans-membranal domain at a position crucial for signal transduction, both for activation and inhibition of downstream-signaling. The Molecular Tumor Board decided to start the treatment with 13.5 mg pemigatinib once daily for 14 days, followed by 7 days of free therapy interval resulting in a sustained partial response. The patient continues to be treated of 13.5 mg as described above. In our case report, we were able to show that the patient in whom an C382R mutation was detected responded to the therapy with pemigatinib. This shows that real-world scenarios differ from the data of the approval studies, thereby illustrating how complex data on patients with FGFR mutations is. One of the main problems of large approval studies is that the functionality of the respective alterations is often disregarded. Our results suggest that respective mutation may be successfully targeted by FGFR-selective tyrosine-kinase inhibitors, demonstrating the importance of the functional characterization of mutations.

Mit wachsenden Erkenntnissen zu grundlegenden molekularen Mechanismen werden zunehmend Zielstrukturen für neuartige personalisierte Therapien für das Endometriumkarzinoms (EC) identifiziert. Diese Therapien haben das Potenzial, das Langzeitüberleben von Krebspatientinnen mit geeigneten Biomarkern zu verbessern. In der vorliegenden Studie wurde bei 11 Patientinnen mit EC, ein Panel Hybrid Capture-basiertes Next Generation Sequencing (FoundationOne CDx, Penzberg, Deutschland) durchgeführt. Bei allen untersuchten Patientinnen wurden genetische Veränderungen festgestellt, die potenzielle Ziele für personalisierte Therapien darstellen. Signifikante Veränderungen wurden bei 7 (63,6%) Patientinnen in TP53, bei 4 (36,4%) in PIK2CA, bei 3 (27,3%) in ERBB2, bei 3 (27,3%) in ARIDA1A, bei 2 (18,2%) in NF1, bei 2 (18,2%) in PTEN und bei einer Patientin in PIK3R1, festgestellt. TMB konnte bei 10 von 11 Patientinnen analysiert werden und war bei 5 (45,5%) niedrig, bei 3 (27,3%) intermediate und bei 2 Patientinnen (18,2%) hoch. Die Mikrosatelliten waren bei 2 Patientinnen (18,2 %) instabil (MSI) und bei 7 Patientinnen (63,6 %) stabil (MSS). Außerdem wurden die Ergebnisse der Immunhistochemie (IHC) mit den NGS-Daten vergleichen. Drei Patientinnen waren in der IHC Her2 positiv, obwohl keine ERBB2-Amplifikation nachweisbar war. Im Gegensatz dazu identifizierte die NGS-Analyse eine Patientin, die in der IHC negativ für Her2 war jedoch eine ERBB2-Amplifikationen aufwies und sich somit für Her2/neu-gerichtete Therapien qualifizierte. Diese Daten unterstreichen die Bedeutung der Identifizierung molekularer Muster, da 100 % der analysierten Patientinnen potenziell therapierelevante Alterationen aufwiesen. Alle Patientinnen mit EC sollten eine NGS-Analyse, sowie eine Untersuchung auf Her2/neu und PD-L1 mittels IHC erhalten.

Point mutations of the fibroblast growth factor receptor (FGFR)2 receptor in intrahepatic cholangiocarcinoma (iCC) are mainly of unknown functional significance compared to FGFR2 fusions. Pemigatinib, a tyrosine kinase inhibitor, is approved for the treatment of cholangiocarcinoma with FGFR2 fusion/rearrangement. Although it is hypothesized that FGFR2 mutations may cause uncontrolled activation of the signaling pathway, the data for targeted therapies for FGFR2 mutations remain unclear. In vitro analyses demonstrated the importance of the p.C382R mutation for ligand-independent constitutive activation of FGFR2 with transforming potential. The following report describes the clinical case of a patient diagnosed with an iCC carrying a FGFR2 p.C382R point mutation which was detected in liquid, as well as in tissue-based biopsies. The patient was treated with pemigatinib, resulting in a sustained complete functional remission in fluorodeoxyglucose-positron emission tomography/computed tomography over 10 months to date. The reported case is the first description of a complete functional remission under the treatment with pemigatinib in a patient with p.C383R mutation.

The amount of data in the medical field is constantly increasing. But it is not only the sheer amount of information that is important, but also its quality and type of representation. While nomenclatures such as SNOMED CT (Systematized Nomenclature of Medicine and Clinical Term) are suited for finegrained documentation and modern analysis, much information is also bound in classifications. This fact is often historical, as billing systems are typically based on classifications such as ICD-10 and then also had been used for documentation. Leveraging this information automatically is subject of this paper – i.e. enabling an automatic mapping from ICD-10 to SNOMED CT. Because this mapping provides a large set of SNOMED codes for each ICD-10 concept, the approach is non-trivial. In order to pick the best possible code, we propose to take advantage of the hierarchical structure of the SNOMED system to find the concept which lies closer to all candidates in the target system. In other words, our algorithm searches the lowest common ancestor (LCA) of all candidates. For evaluation, we studied 1692 codes from a real-world dataset. The results are promising and show that the proposed approach achieves good results in the majority of cases.

With the digital transformation in medicine, enormous amounts of data are being generated and are available for analysis. Process mining techniques can be utilized to extract process models from this data. On the one side these process models typically provide detailed, fine-granular activitiy descriptions. But on the other hand these models become increasingly less recognizable. Therefore, in this contribution we explore the use of Self Organizing Maps for an event abstraction in the medical context. Our approach achieved promising results on a publicly available sepsis data set.

The increasing availability and use of sensitive personal data raises a set of issues regarding the privacy of the individuals behind the data. These concerns become even more important when health data are processed, as are considered sensitive (according to most global regulations). Privacy Enhancing Technologies (PETs) attempt to protect the privacy of individuals whilst preserving the utility of data. One of the most popular technologies recently is Differential Privacy (DP), which was used for the 2020 U.S. Census. Another trend is to combine synthetic data generators with DP to create so-called private synthetic data generators. The objective is to preserve statistical properties as accurately as possible, while the generated data should be as different as possible compared to the original data regarding private features. While these technologies seem promising, there is a gap between academic research on DP and synthetic data and the practical application and evaluation of these techniques for real-world use cases. In this paper, we evaluate three different private synthetic data generators (MWEM, DP-CTGAN, and PATE-CTGAN) on their use-case-specific privacy and utility. For the use case, continuous heart rate measurements from different individuals are analyzed. This work shows that private synthetic data generators have tremendous advantages over traditional techniques, but also require in-depth analysis depending on the use case. Furthermore, it can be seen that each technology has different strengths, so there is no clear winner. However, DP-CTGAN often performs slightly better than the other technologies, so it can be recommended for a continuous medical data use case.

e18750 Background: Data on SARS-CoV-2 infections in oncological patients in the outpatient settings are scarce. Methods: During the spread of the delta variant between April 2021 and September 2021, a total of 10.677 patients were tested for SARS-CoV-2 infection by RT-qPCR in seven outpatient clinics in Bavaria, Germany. Results: Within the tested patient cohort, 4.960 patients (46.5%) suffered from a malignant disease (74% solid tumors and 26% malignant hematological diseases). This group was compared with 5.717 patients (53.5%) without a malignant disease (33.1% with other hematological diseases and 66.9% patients without a hematological or oncological disease). During the observation period, 119 (2.4%) patients with malignancies were tested positive (88 patients with solid tumors; 31 patients with malignant hematological diseases) compared to 115 positive patients (2.0%) in the control group. 32 of 119 positively tested patients (26.9%) suffering from malignant disease required hospitalization and 9/32 patients (28.1%) died during the clinical course. Conclusions: These observations are in clear contrast to data from patients we evaluated during the pre-delta variants period between 15 and 26 April 2020 in the same seven outpatient clinics. In this period, a total of 1.227 patients were tested for SARS-CoV-2 by RT-qPCR. 78/1227 patients (6.3%) were tested positive in RT-qPCR and most showed mild symptoms of infection. None of the SARS-CoV-2 infected patients died. These data were analyzed when no vaccination was available. These data were evaluated during a period where no vaccine was available. Vaccination of patients with malignancies with BiontechPfizer's mRNA vaccines was started in April 2021. The response to the vaccine was tested by an antibody assay (Elecsys Anti-SARS-CoV-2 S-immunoassay, Roche) at the earliest four weeks after the second vaccination. To assess the response, we compared five patient cohorts: Patients who received (i) B cell depleting antibodies, (ii) checkpoint inhibitors (ICI), (iii) chemotherapy, or (iv) tyrosin kinase inhibitors (TKIs), and (v) healthy controls. The patients treated with ICI or TKI showed a comparable vaccination response to the healthy patients, while patients receiving Rituximab/Obinutuzumab showed no significant humoral vaccination response at all. The more severe disease course of patients infected by the SARS-CoV-2 delta variant compared to the initial waves of infections strongly underline the importance of vaccination in cancer patients.

Providing a suitable rehabilitation after an acute episode or a chronic disease helps people to live independently and enhance their quality of life. However, the continuity of care is often interrupted in the transition from hospital to home. Virtual coaches (VCs) could help these patients to engage in personalized home rehabilitation programs. These coaching systems need also to be fed with procedural precepts in order to work as intended. This, in turn, relates both to properly represent the clinical knowledge (as the VC somehow replaces the formal caregivers that cannot be fully present) as well guide the patient correctly (in order to follow the medically desired procedures given the need for personalisation according to individual needs). Therefore, we outline our technical approach to deal with this. In particular, clinical pathways in terms of semi-formal procedure models in combination with machine learning components processing and powerful user interfaces providing these pathway information and feeding the VC are presented. The system is currently under testing in a participatory design phase called Living Lab. Thus, initial user feedback for further improvements is about to come.

Clinical Practice Guidelines (CPGs) contain expert knowledge on the diagnosis and treatment of diseases. They can be regarded as state of the art and standardized procedures that have been established by consensus of the clinical expert community. In this work, we show how CPGs can be formalized by activities of the Unified Modeling Language (UML), and can subsequently be translated into PROforma models. UML activities allow for a comprehensible representation of the underlying process, whereas PROforma models can be directly executed in a dialog system and support the practitioner during the diagnosis or treatment process. In this work, we expand our approach from [1] to include more complex diseases like Primary Myelofribrosis (PMF) and Immune Thrombocytopenia (ITP) and show the applicability for exemplary patients.

With the rise of personalized medicine, the number of individualized treatment options and related decisions is increasing tremendously. Thereby, the acquisition, incorporation and representation of the patient’s individual preferences in upcoming, modern, AI-based medical decision support systems play a decisive role. E.g., for patients with advanced breast cancer, there are various therapeutic options associated with different outcomes to choose from. In our contribution we show a first approach to model Preference Elicitation (PE) via card sorting using a utility function. Based on this, we present further ideas for extending and improving the approach.