Search

Machine learning has immense novel but also disruptive potential for medicine. Numerous applications have already been suggested and evaluated concerning cardiovascular diseases. One important aspect is the detection and management of potentially thrombogenic arrhythmias such as atrial fibrillation. While atrial fibrillation is the most common arrhythmia with a lifetime risk of one in three persons and an increased risk of thromboembolic complications such as stroke, many atrial fibrillation episodes are asymptomatic and a first diagnosis is oftentimes only reached after an embolic event. Therefore, screening for atrial fibrillation represents an important part of clinical practice. Novel technologies such as machine learning have the potential to substantially improve patient care and clinical outcomes. Additionally, machine learning applications may aid cardiologists in the management of patients with already diagnosed atrial fibrillation, for example, by identifying patients at a high risk of recurrence after catheter ablation. We summarize the current state of evidence concerning machine learning and, in particular, artificial neural networks in the detection and management of atrial fibrillation and describe possible future areas of development as well as pitfalls. Graphical abstract Typical data flow in machine learning applications for atrial fibrillation detection.

The BAF (BRG1/BRM-associated factor) chromatin remodelling complex is essential for the regulation of DNA accessibility and gene expression during neuronal differentiation. Mutations of its core subunit SMARCB1 result in a broad spectrum of pathologies, including aggressive rhabdoid tumours or neurodevelopmental disorders. Other mouse models have addressed the influence of a homo- or heterozygous loss of Smarcb1, yet the impact of specific non-truncating mutations remains poorly understood. Here, we have established a new mouse model for the carboxy-terminal Smarcb1 c.1148del point mutation, which leads to the synthesis of elongated SMARCB1 proteins. We have investigated its impact on brain development in mice using magnetic resonance imaging, histology, and single-cell RNA sequencing. During adolescence, Smarcb11148del/1148del mice demonstrated rather slow weight gain and frequently developed hydrocephalus including enlarged lateral ventricles. In embryonic and neonatal stages, mutant brains did not differ anatomically and histologically from wild-type controls. Single-cell RNA sequencing of brains from newborn mutant mice revealed that a complete brain including all cell types of a physiologic mouse brain is formed despite the SMARCB1 mutation. However, neuronal signalling appeared disturbed in newborn mice, since genes of the AP-1 transcription factor family and neurite outgrowth-related transcripts were downregulated. These findings support the important role of SMARCB1 in neurodevelopment and extend the knowledge of different Smarcb1 mutations and their associated phenotypes.

Glaucoma is one of the leading causes of blindness worldwide. Therefore, early detection and diagnosis are key to preserve full vision in patients. As part of the SALUS study, we create a blood vessel segmentation model based on U-Net. We trained U-Net on three different loss functions and used hyperparameter tuning to find their optimal hyperparameters for each loss function. The best models for each of the loss functions achieved an accuracy of over 93%, Dice scores around 83% and Intersection over Union scores over 70%. They each identify large blood vessels reliably and even recognize smaller blood vessels in the retinal fundus images and thus pave the way for improved glaucoma management.

Age-related macular degeneration (AMD) is the leading cause of blindness in the Western world. In this work, the non-invasive imaging technique spectral domain optical coherence tomography (SD-OCT) is used to acquire retinal images, which are then analyzed using deep learning techniques. The authors trained a convolutional neural network (CNN) using 1300 SD-OCT scans annotated by trained experts for the presence of different biomarkers associated with AMD. The CNN was able to accurately segment these biomarkers and the performance was further enhanced through transfer learning with weights from a separate classifier, trained on a large external public OCT dataset to distinguish between different types of AMD. Our model is able to accurately detect and segment AMD biomarkers in OCT scans, which suggests that it could be useful for prioritizing patients and reducing ophthalmologists’ workloads.

A growing number of studies have been researching biomarkers of Parkinson’s disease (PD) using mobile technology. Many have shown high accuracy in PD classification using machine learning (ML) and voice records from the mPower study, a large database of PD patients and healthy controls. Since the dataset has unbalanced class, gender and age distribution, it is important to consider appropriate sampling when assessing classification scores. We analyse biases, such as identity confounding and implicit learning of non-disease-specific characteristics and present a sampling strategy to highlight and prevent these problems.

Deep Learning architectures for time series require a large number of training samples, however traditional sample size estimation for sufficient model performance is not applicable for machine learning, especially in the field of electrocardiograms (ECGs). This paper outlines a sample size estimation strategy for binary classification problems on ECGs using different deep learning architectures and the large publicly available PTB-XL dataset, which includes 21801 ECG samples. This work evaluates binary classification tasks for Myocardial Infarction (MI), Conduction Disturbance (CD), ST/T Change (STTC), and Sex. All estimations are benchmarked across different architectures, including XResNet, Inception-, XceptionTime and a fully convolutional network (FCN). The results indicate trends for required sample sizes for given tasks and architectures, which can be used as orientation for future ECG studies or feasibility aspects.

Even though the interest in machine learning studies is growing significantly, especially in medicine, the imbalance between study results and clinical relevance is more pronounced than ever. The reasons for this include data quality and interoperability issues. Hence, we aimed at examining site- and study-specific differences in publicly available standard electrocardiogram (ECG) datasets, which in theory should be interoperable by consistent 12-lead definition, sampling rate, and measurement duration. The focus lies upon the question of whether even slight study peculiarities can affect the stability of trained machine learning models. To this end, the performances of modern network architectures as well as unsupervised pattern detection algorithms are investigated across different datasets. Overall, this is intended to examine the generalization of machine learning results of single-site ECG studies.

The accurate reconstruction of clonal evolution, including the identification of newly developing, highly aggressive subclones, is essential for the application of precision medicine in cancer treatment. Reconstruction, aiming for correct variant clustering and clonal evolution tree reconstruction, is commonly performed by tedious manual work. While there is a plethora of tools to automatically generate reconstruction, their reliability, especially reasons for unreliability, are not systematically assessed. We developed clevRsim—an approach to simulate clonal evolution data, including single-nucleotide variants as well as (overlapping) copy number variants. From this, we generated 88 data sets and performed a systematic evaluation of the tools for the reconstruction of clonal evolution. The results indicate a major negative influence of a high number of clones on both clustering and tree reconstruction. Low coverage as well as an extreme number of time points usually leads to poor clustering results. An underlying branched independent evolution hampers correct tree reconstruction. A further major decline in performance could be observed for large deletions and duplications overlapping single-nucleotide variants. In summary, to explore the full potential of reconstructing clonal evolution, improved algorithms that can properly handle the identified limitations are greatly needed.

Due to the highly variable clinical phenotype, Klinefelter Syndrome is underdiagnosed.Assessment of supervised machine learning based prediction models for identification of Klinefelter Syndrome among azoospermic patients, and comparison to expert clinical evaluation.Retrospective patient data (karyotype, age, height, weight, testis volume, follicle-stimulating hormone, luteinizing hormone, testosterone, estradiol, prolactin, semen pH and semen volume) collected between January 2005 and June 2019 were retrieved from a patient data bank of a University Centre. Models were trained, validated and benchmarked based on different supervised machine learning algorithms. Models were then tested on an independent, prospectively acquired set of patient data (between July 2019 and July 2020). Benchmarking against physicians was performed in addition.Based on average performance, support vector machines and CatBoost were particularly well-suited models, with 100% sensitivity and93% specificity on the test dataset. Compared to a group of 18 expert clinicians, the machine learning models had significantly better median sensitivity (100% vs. 87.5%, p = 0.0455) and fared comparably with regards to specificity (90% vs. 89.9%, p = 0.4795), thereby possibly improving diagnosis rate. A Klinefelter Syndrome Score Calculator based on the prediction models is available on http://klinefelter-score-calculator.uni-muenster.de.Differentiating Klinefelter Syndrome patients from azoospermic patients with normal karyotype (46,XY) is a problem that can be solved with supervised machine learning techniques, improving patient care.Machine learning could improve the diagnostic rate of Klinefelter Syndrome among azoospermic patients, even more for less-experienced physicians.

Objective Neurological manifestations of autoimmune connective tissue diseases (CTD) are poorly understood and difficult to diagnose. We here aimed to address this shortcoming by studying immune cell compositions in CTD patients with and without neurological manifestation. Methods Using flow cytometry, we retrospectively investigated paired cerebrospinal fluid (CSF) and blood samples of 28 CTD patients without neurological manifestation, 38 CTD patients with neurological manifestation (N-CTD), 38 non-inflammatory controls, and 38 multiple sclerosis (MS) patients, a paradigmatic primary neuroinflammatory disease. Results We detected an expansion of plasma cells in the blood of both N-CTD and CTD compared to non-inflammatory controls and MS. Blood plasma cells alone distinguished the clinically similar entities N-CTD and MS with high discriminatory performance (AUC: 0.81). Classical blood monocytes indicated higher disease activity in systemic lupus erythematosus (SLE) patients. Surprisingly, immune cells in the CSF did not differ significantly between N-CTD and CTD, while CD4+ T cells and the CD4+/CD8+ ratio were elevated in the blood of N-CTD compared to CTD. Several B cell-associated parameters partially overlapped in the CSF in MS and N-CTD. We built a machine learning model that distinguished N-CTD from MS with high discriminatory power using either blood or CSF. Conclusion We here find that blood flow cytometry alone surprisingly suffices to distinguish CTD with neurological manifestations from clinically similar entities, suggesting that a rapid blood test could support clinicians in the differential diagnosis of N-CTD.

Background A thorough analysis of clonal evolution commonly requires integration of diverse sources of data (e.g., karyotyping, next-generation sequencing, and clinical information). Subsequent to actual reconstruction of clonal evolution, detailed analysis and interpretation of the results are essential. Often, however, only few tumor samples per patient are available. Thus, information on clonal development and therapy effect may be incomplete. Furthermore, analysis of biallelic events—considered of high relevance with respect to disease course—can commonly only be realized by time-consuming analysis of the raw results and even raw sequencing data. Results We developed clevRvis, an R/Bioconductor package providing an extensive set of visualization techniques for clonal evolution. In addition to common approaches for visualization, clevRvis offers a unique option for allele-aware representation: plaice plots. Biallelic events may be visualized and inspected at a glance. Analyzing 4 public datasets, we show that plaice plots help to gain new insights into tumor development and investigate hypotheses on disease progression and therapy resistance. In addition to a graphical user interface, automatic phylogeny-aware color coding of the plots, and an approach to explore alternative trees, clevRvis provides 2 algorithms for fully automatic time point interpolation and therapy effect estimation. Analyzing 2 public datasets, we show that both approaches allow for valid approximation of a tumor’s development in between measured time points. Conclusions clevRvis represents a novel option for user-friendly analysis of clonal evolution, contributing to gaining new insights into tumor development.

Predisposing CHEK2 germline variants are associated with various adult-type malignancies, whereas their impact on cancer susceptibility in childhood cancer is unclear. To understand the frequency of germline variants in the CHEK2 gene and their impact on pediatric malignancies, we used whole-exome sequencing to search for CHEK2 variants in the germlines of 418 children diagnosed with cancer in our clinics. Moreover, we performed functional analysis of the pathogenic CHEK2 variants to analyze the effect of the alterations on CHK2 protein function. We detected a CHEK2 germline variant in 32/418 (7.7%) pediatric cancer patients and 46.8% of them had leukemia. Functional analysis of the pathogenic variants revealed that 5 pathogenic variants impaired CHK2 protein function. 6/32 patients carried one of these clearly damaging CHEK2 variants and two of them harbored a matching family history of cancer. In conclusion, we detected germline CHEK2 variants in 7.7% of all pediatric cancer patients, of which a minority but still relevant fraction of approximately 20% had a profound impact on protein expression or its phosphorylation after irradiation-induced DNA damage. Accordingly, we conclude that CHEK2 variants increase the risk for not only adult-onset but also pediatric cancer.

Clinical checklists are the current gold standard to determine whether a child with cancer shows indications for genetic testing. Nevertheless, the efficacy of these tests to reliably detect genetic cancer predisposition in children with cancer is still insufficiently investigated. Here, we assessed the validity of clinically recognizable signs to identify cancer predisposition by correlating a state-of-the-art clinical checklist to the corresponding whole exome sequencing analysis in an unselected single-center cohort of 139 child-parent datasets. We applied a strict testing to only include autosomal dominant or compound heterozygous cancer-related variants.Our study reflects a high consent rate for genetic testing (>90%). In total, 1/3rdof patients had a clinical indication for genetic testing according to current recommendations and 10.8% (n=15/139) of children harbored a proven cancer predisposition based on exome sequencing. Out of these only 73.3% (n=11/15) were identified through the clinical checklist. In addition, >2 clinical findings in the applied checklist increased the likelihood to identifying genetic predisposition from 15% to 50%. While our data revealed a high rate of genetic predisposition (50%, n=5/10) in Myelodysplastic Syndrome (MDS) cases, no cancer predisposition variants were identified in the sarcoma and lymphoma group.In summary, our data showed a low checklist specificity of 68.5%, and missed every 4thchild with genetic predisposition. This highlights the drawbacks of sole clinical evaluation to accurately identify all children at risk and underlines the need for routine germline sequencing of pediatric cancers.

BACKGROUND: Embryonal tumors with multilayered rosettes (ETMR) are pediatric brain tumors bearing a grim prognosis, despite intensive multimodal therapeutic approaches. Insights into cellular heterogeneity and cellular communication of tumor cells with cells of the tumor microenvironment (TME), by applying single-cell (sc) techniques, potentially identify mechanisms of therapy resistance and target-directed treatment approaches. MATERIAL AND METHODS: To explore ETMR cell diversity, we used single-cell RNA sequencing (scRNA-seq) in human (n=2) and murine ETMR (transgenic mode; n=4) samples, spatial transcriptomics, 2D and 3D cultures (including co-cultures with TME cells), multiplex immunohistochemistry and drug screens. RESULTS: ETMR microenvironment is composed of tumor and non-tumor cell types. The ETMR malignant compartment harbour cells representing distinct transcriptional metaprograms, (NSC-like, NProg-like and Neuroblast-like), mirroring embryonic neurogenic cell states and fuelled by neurogenic pathways (WNT, SHH, Hippo). The ETMR TME is composed of oligodendrocyte and neuronal progenitor cells, neuroblasts, microglia, and pericytes. Tumor-specific ligand-receptor interaction analysis showed enrichment of intercellular communication between NProg-like ETMR cells and pericytes (PC). Functional network analyses reveal ETMR-PC interactions related to stem-cell signalling and extracellular matrix (ECM) organization, involving factors of the WNT, BMP, and CxCl12 networks. Results from ETMR-PC co-culture and spatial transcriptomics pointed to a pivotal role of pericytes in keeping ETMR in a germinal neurogenic state, enriched in stem-cell signalling. Drug screening considering cellular heterogeneity and cellular communication suggested novel therapeutic approaches. CONCLUSION: ETMR demonstrated diversity in the microenvironment, with enrichment of cell-cell communications with pericytes, supporting stem-cell signalling and interfering in the organization of the tumor extracellular matrix. Targeting ETMR-PC interactions might bring new opportunities for target-directed therapy.

With increasing numbers of patients recovering from COVID-19, there is increasing evidence for persistent symptoms and the need for follow-up studies. This retrospective study included patients without comorbidities, who recovered from COVID-19 and attended an outpatient clinic at a university hospital for follow-up care and potential convalescent plasma donation. Network analysis was applied to visualize symptom combinations and persistent symptoms. Comprehensive lab-testing was ascertained at each follow-up to analyze differences regarding patients with vs without persistent symptoms. 116 patients were included, age range was 18–69 years (median: 41) with follow-ups ranging from 22 to 102 days. The three most frequent persistent symptoms were Fatigue (54%), Dyspnea (29%) and Anosmia (25%). Lymphopenia was present in 13 of 112 (12%) cases. Five of 35 cases (14%) had Lymphopenia in the later follow-up range of 80–102 days. Serum IgA concentration was the only lab parameter with significant difference between patients with vs without persistent symptoms with reduced serum IgA concentrations in the patient cohort of persistent symptoms (p = 0.0219). Moreover, subgroup analyses showed that patients with lymphopenia experienced more frequently persistent symptoms. In conclusion, lymphopenia persisted in a noticeable percentage of recovered patients. Patients with persistent symptoms had significantly lower serum IgA levels. Furthermore, our data provides evidence that lymphopenia is associated with persistence of COVID-19 symptoms.

OBJECTIVE Various quality indicators are currently under investigation, aiming at measuring the quality of care in neurosurgery; however, the discipline currently lacks practical scoring systems for accurately assessing risk. The aim of this study was to develop three accurate, easy-to-use risk scoring systems for nosocomial infections, reoperations, and adverse events for patients with cerebral and spinal tumors. METHODS The authors developed a semiautomatic registry with administrative and clinical data and included all patients with spinal or cerebral tumors treated between September 2017 and May 2019. Patients were further divided into development and validation cohorts. Multivariable logistic regression models were used to develop risk scores by assigning points based on β coefficients, and internal validation of the scores was performed. RESULTS In total, 1000 patients were included. An unplanned 30-day reoperation was observed in 6.8% of patients. Nosocomial infections were documented in 7.4% of cases and any adverse event in 14.5%. The risk scores comprise variables such as emergency admission, nursing care level, ECOG performance status, and inflammatory markers on admission. Three scoring systems, NoInfECT for predicting the incidence of nosocomial infections (low risk, 1.8%; intermediate risk, 8.1%; and high risk, 26.0% [p < 0.001]), LEUCut for 30-day unplanned reoperations (low risk, 2.2%; intermediate risk, 6.8%; and high risk, 13.5% [p < 0.001]), and LINC for any adverse events (low risk, 7.6%; intermediate risk, 15.7%; and high risk, 49.5% [p < 0.001]), showed satisfactory discrimination between the different outcome groups in receiver operating characteristic curve analysis (AUC ≥ 0.7). CONCLUSIONS The proposed risk scores allow efficient prediction of the likelihood of adverse events, to compare quality of care between different providers, and further provide guidance to surgeons on how to allocate preoperative care.

German Medical Data Sciences 2022 – Future Medicine: More Precise, More Integrative, More Sustainable!; Amsterdam : IOS Press, Studies in Health Technology and Informatics 296, 33-40 (2022). doi:10.3233/SHTI220801, Published by IOS Press, Amsterdam

Introduction: The long-QT syndrome (LQTS) is the most common ion channelopathy, typically presenting with a prolonged QT interval and clinical symptoms such as syncope or sudden cardiac death. Patients may present with a concealed phenotype making the diagnosis challenging. Correctly diagnosing at-risk patients is pivotal to starting early preventive treatment. Objective: Identification of congenital and often concealed LQTS by utilizing novel deep learning network architectures, which are specifically designed for multichannel time series and therefore particularly suitable for ECG data. Design and Results: A retrospective artificial intelligence (AI)-based analysis was performed using a 12-lead ECG of genetically confirmed LQTS (n = 124), including 41 patients with a concealed LQTS (33%), and validated against a control cohort (n = 161 of patients) without known LQTS or without QT-prolonging drug treatment but any other cardiovascular disease. The performance of a fully convolutional network (FCN) used in prior studies was compared with a different, novel convolutional neural network model (XceptionTime). We found that the XceptionTime model was able to achieve a higher balanced accuracy score (91.8%) than the associated FCN metric (83.6%), indicating improved prediction possibilities of novel AI architectures. The predictive accuracy prevailed independently of age and QTc parameters. Conclusions: In this study, the XceptionTime model outperformed the FCN model for LQTS patients with even better results than in prior studies. Even when a patient cohort with cardiovascular comorbidities is used. AI-based ECG analysis is a promising step for correct LQTS patient identification, especially if common diagnostic measures might be misleading.

The long-QT syndrome (LQTS) is the most common ion channelopathy, typically presenting with a prolonged QT interval and clinical symptoms such as syncope or sudden cardiac death. Patients may present with a concealed phenotype making the diagnosis challenging. Correctly diagnosing at-risk patients is pivotal to starting early preventive treatment.Identification of congenital and often concealed LQTS by utilizing novel deep learning network architectures, which are specifically designed for multichannel time series and therefore particularly suitable for ECG data.A retrospective artificial intelligence (AI)-based analysis was performed using a 12-lead ECG of genetically confirmed LQTS (In this study, the XceptionTime model outperformed the FCN model for LQTS patients with even better results than in prior studies. Even when a patient cohort with cardiovascular comorbidities is used. AI-based ECG analysis is a promising step for correct LQTS patient identification, especially if common diagnostic measures might be misleading.

BackgroundIntensive care unit (ICU) readmissions are associated with mortality and poor outcomes. To improve discharge decisions, machine learning (ML) could help to identify patients at risk of ICU readmission. However, as many models are black boxes, dangerous properties may remain unnoticed. Widely used post hoc explanation methods also have inherent limitations. Few studies are evaluating inherently interpretable ML models for health care and involve clinicians in inspecting the trained model.MethodsAn inherently interpretable model for the prediction of 3 day ICU readmission was developed. We used explainable boosting machines that learn modular risk functions and which have already been shown to be suitable for the health care domain. We created a retrospective cohort of 15,589 ICU stays and 169 variables collected between 2006 and 2019 from the University Hospital Münster. A team of physicians inspected the model, checked the plausibility of each risk function, and removed problematic ones. We collected qualitative feedback during this process and analyzed the reasons for removing risk functions. The performance of the final explainable boosting machine was compared with a validated clinical score and three commonly used ML models. External validation was performed on the widely used Medical Information Mart for Intensive Care version IV database.ResultsThe developed explainable boosting machine used 67 features and showed an area under the precision-recall curve of 0.119 ± 0.020 and an area under the receiver operating characteristic curve of 0.680 ± 0.025. It performed on par with state-of-the-art gradient boosting machines (0.123 ± 0.016, 0.665 ± 0.036) and outperformed the Simplified Acute Physiology Score II (0.084 ± 0.025, 0.607 ± 0.019), logistic regression (0.092 ± 0.026, 0.587 ± 0.016), and recurrent neural networks (0.095 ± 0.008, 0.594 ± 0.027). External validation confirmed that explainable boosting machines (0.221 ± 0.023, 0.760 ± 0.010) performed similarly to gradient boosting machines (0.232 ± 0.029, 0.772 ± 0.018). Evaluation of the model inspection showed that explainable boosting machines can be useful to detect and remove problematic risk functions.ConclusionsWe developed an inherently interpretable ML model for 3 day ICU readmission prediction that reached the state-of-the-art performance of black box models. Our results suggest that for low- to medium-dimensional datasets that are common in health care, it is feasible to develop ML models that allow a high level of human control without sacrificing performance.

Machine learning algorithms become increasingly prevalent in the field of medicine, as they offer the ability to recognize patterns in complex medical data. Especially in this sensitive area, the active usage of a mostly black box is a controversial topic. We aim to highlight how an aggregated and systematic feature analysis of such models can be beneficial in the medical context. For this reason, we introduce a grouped version of the permutation importance analysis for evaluating the influence of entire feature subsets in a machine learning model. In this way, expert-defined subgroups can be evaluated in the decision-making process. Based on these results, new hypotheses can be formulated and examined.

Acute kidney injury (AKI) is a common complication in critically ill patients and is associated with long-term complications and an increased mortality. This work presents preliminary findings from the first freely available European intensive care database released by Amsterdam UMC. A machine learning (ML) model was developed to predict AKI in the intensive care unit 12 hours before the actual event. Main features of the model included medications and hemodynamic parameters. Our models perform with an accuracy of 81.8% on moderate to severe AKI and 79.8% on all AKI patients. Those results can compete with models reported in the literature and introduce an ML model for AKI based on European patient data.

Secondary use, the reuse of medical patient data stored during routine care in the hospital’s electronic medical records (EMR) for research purpose is common, especially for registers and pragmatic trials. Often the medical data items are copied manually from the EMR into the used research database. This process is time consuming and error prone. In the “Safety of the Living Kidney Donor – The German National Register” (SOLKID-GNR), laboratory results gathered during control check-ups of the living donors before and after the transplantation are to be transferred from the EMR into the electronic data capture system REDCap of the register. In this work, we present our approach of realizing an automated transfer of time-dependent laboratory results from the EMR of the University Hospital of Münster to REDCap. A challenge lies in the multi-center structure of SOLKID-GNR. The participating transplant centers are using different EMR systems, which requires a flexible architecture design. In addition, we aimed to support reuse of the implementation for other research settings with other medical data items of interest.

Parkinson’s disease (PD) is a common neurodegenerative disorder that severely impacts quality of life as the condition progresses. Early diagnosis and treatment is important to reduce burden and costs. Here, we evaluate the diagnostic potential of the Non-Motor symptoms (NMS) questionnaire by the International Parkinson and Movement Disorder Society based on patient-completed answers from a large single-center prospective study. In this study data from 489 study participants consisting of a PD group, a healthy control (HC) group and patients with differential diagnosis (DD) have been recorded with a smartphone-based system. Evaluation of the study data has shown a significant difference in NMS between the representative groups. Cross-validation of Machine Learning based classification achieves balanced accuracy scores of 88.7% in PD vs. HC, 72.1% in PD vs. DD and 82.6% when discriminating between all movement disorders (PD + DD) and the HC group. The results indicate potentially high feature importance of a simple self-administered questionnaire that could support early diagnosis.

Secondary use, the reuse of medical patient data stored during routine care in the hospital's electronic medical records (EMR) for research purpose is common, especially for registers and pragmatic trials. Often the medical data items are copied manually from the EMR into the used research database. This process is time consuming and error prone. In the "Safety of the Living Kidney Donor - The German National Register" (SOLKID-GNR), laboratory results gathered during control check-ups of the living donors before and after the transplantation are to be transferred from the EMR into the electronic data capture system REDCap of the register. In this work, we present our approach of realizing an automated transfer of time-dependent laboratory results from the EMR of the University Hospital of Münster to REDCap. A challenge lies in the multi-center structure of SOLKID-GNR. The participating transplant centers are using different EMR systems, which requires a flexible architecture design. In addition, we aimed to support reuse of the implementation for other research settings with other medical data items of interest.

Machine learning algorithms become increasingly prevalent in the field of medicine, as they offer the ability to recognize patterns in complex medical data. Especially in this sensitive area, the active usage of a mostly black box is a controversial topic. We aim to highlight how an aggregated and systematic feature analysis of such models can be beneficial in the medical context. For this reason, we introduce a grouped version of the permutation importance analysis for evaluating the influence of entire feature subsets in a machine learning model. In this way, expert-defined subgroups can be evaluated in the decision-making process. Based on these results, new hypotheses can be formulated and examined.

Acute kidney injury (AKI) is a common complication in critically ill patients and is associated with long-term complications and an increased mortality. This work presents preliminary findings from the first freely available European intensive care database released by Amsterdam UMC. A machine learning (ML) model was developed to predict AKI in the intensive care unit 12 hours before the actual event. Main features of the model included medications and hemodynamic parameters. Our models perform with an accuracy of 81.8% on moderate to severe AKI and 79.8% on all AKI patients. Those results can compete with models reported in the literature and introduce an ML model for AKI based on European patient data.

This systematic review assesses the reporting quality and risk of bias in studies evaluating the diagnostic test accuracy (DTA) of clinical decision support systems (CDSS).The Cochrane Library, PubMed/MEDLINE, Scopus, and Web of Science were searched for studies, published between January 1, 2016 and May 31, 2021, evaluating the DTA of CDSS for human patients. Articles using a patient's self-diagnosis, assessing disease severity, focusing on treatment/follow-up, or comparing pre-post CDSS implementation periods were excluded. All eligible studies were assessed for reporting quality using STARD 2015 and for risk of bias using QUADAS-2. Item ratings were presented using heat maps. This study was reported as per PRISMA-DTA.In total, 158 of 2,820 screened articles were included in the analysis. The studies were heterogeneous in terms of study characteristics, reporting quality, risk of biases, and applicability concerns with few highly rated studies. Mostly the overall quality was deficient for items addressing the domains 'methodology,' 'results,' and 'other information'.Our analysis revealed shortcomings in critical domains of reporting quality and risk of bias, indicating the need for additional guidance and training in an interdisciplinary scientific field with mixed biostatistical expertise.

Digital acquisition of patients' self-reports on individual risk factors and symptom severity represents a promising, cost-efficient, and increasingly prevalent approach for standardized data collection in psychiatric clinical routine. Yet, studies investigating digital data collection in patients with a schizophrenia spectrum disorder (PSSDs) are scarce. The objective of this study was to explore the feasibility of digitally acquired self-report assessments of risk and symptom profiles at the time of admission into inpatient treatment in an age-representative sample of hospitalized PSSDs. We investigated the required support, the data entry pace, and the subjective user experience. Findings were compared with those of patients with an affective disorder (PADs). Of 82 PSSDs who were eligible for inclusion, 59.8% (n=49) agreed to participate in the study, of whom 54.2% (n=26) could enter data without any assistance. Inclusion rates, drop-out rates, and subjective experience ratings did not differ between PSSDs and PADs. Patients reported high satisfaction with the assessment. PSSDs required more support and time for the data entry than PADs. Our results indicate that digital data collection is a feasible and well-received method in PSSDs. Future clinical and research efforts on digitized assessments in psychiatry should include PSSDs and offer support to reduce digital exclusion.

Patients with myeloid neoplasia are classified by the WHO classification systems. Besides clinical and hematological criteria, cytogenetic and molecular genetic alterations highly impact treatment stratification. In routine diagnostics, a combination of methods is used to decipher different types of genetic variants. Eight patients were comprehensively analyzed using karyotyping, fluorescence in situ hybridization, array-CGH and a custom NGS panel. Clonal evolution was reconstructed manually, integrating all mutational information on single nucleotide variants (SNVs), insertions and deletions (indels), structural variants and copy number variants (CNVs). To allow a correct integration, we differentiate between three scenarios: 1) CNV occurring prior to the SNV/indel, but in the same cells. 2) SNV/indel occurring prior to the CNV, but in the same cells. 3) SNV/indel and CNV existing in parallel, independent of each other. Applying this bioinformatics approach, we reconstructed clonal evolution for all patients. This generalizable approach offers the possibility to integrate various data to analyze identification of driver and passenger mutations as well as possible targets for personalized medicine approaches. Furthermore, this model can be used to identify markers to assess the minimal residual disease.

The microenvironment of cancer cells is receiving increasing attention as an important factor influencing the progression and prognosis of tumor diseases. In multiple myeloma (MM), a hematological cancer of plasma cells, mesenchymal stem cells (MSCs) represent an integral part of the bone marrow niche and tumor microenvironment. It has been described that MM cells alter MSCs in a way that MM-associated MSCs promote the proliferation and survival of MM cells. Yet, our understanding of the molecular mechanisms governing the interaction between MM cells and MSCs and whether this can be targeted for therapeutic interventions is limited. To identify potential molecular targets, we examined MSCs by RNA sequencing and Western blot analysis. We report that MSCs from MM patients with active disease (MM-Act-MSCs) show a distinct gene expression profile as compared with MSCs from patients with other (non-) malignant diseases (CTR-MSCs). Of note, we detected a significant enrichment of the PI3K-AKT-mTOR hallmark gene set in MM-Act-MSCs and further confirmed the increased levels of related proteins in these MSCs. Pictilisib, a pan-PI3K inhibitor, selectively reduced the proliferation of MM-Act-MSCs as compared with CTR-MSCs. Furthermore, pictilisib treatment impaired the MM-promoting function of MM-Act-MSCs. Our data thus provide a deeper insight into the molecular signature and function of MSCs associated with MM and show that targeting PI3K-AKT-mTOR signaling in MSCs may represent an additional therapeutic pathway in the treatment of MM patients.

Journal of medical internet research : JMIR 24(1), e25440 (2022). doi:10.2196/25440, Published by Healthcare World, Richmond, Va.

Growth Factor Independence 1 (GFI1) is a transcription factor with an important role in the regulation of development of myeloid and lymphoid cell lineages and was implicated in the development of myelodysplastic syndrome (MDS) and acute myeloid leukaemia (AML). Reduced expression of GFI1 or presence of the GFI1-36N (serine replaced with asparagine) variant leads to epigenetic changes in human and murine AML blasts and accelerated the development of leukaemia in a murine model of human MDS and AML. We and other groups previously showed that the GFI1-36N allele or reduced expression of GFI1 in human AML blasts is associated with an inferior prognosis. Using GFI1-36S, -36N -KD, NUP98-HOXD13-tg mice and curcumin (a natural histone acetyltransferase inhibitor (HATi)), we now demonstrate that expansion of GFI1-36N or –KD, NUP98-HODXD13 leukaemic cells can be delayed. Curcumin treatment significantly reduced AML progression in GFI1-36N or -KD mice and prolonged AML-free survival. Of note, curcumin treatment had no effect in GFI1-36S, NUP98-HODXD13 expressing mice. On a molecular level, curcumin treatment negatively affected open chromatin structure in the GFI1-36N or -KD haematopoietic cells but not GFI1-36S cells. Taken together, our study thus identified a therapeutic role for curcumin treatment in the treatment of AML patients (homo or heterozygous for GFI1-36N or reduced GFI1 expression) and possibly improved therapy outcome.

Screening for eligible patients continues to pose a great challenge for many clinical trials. This has led to a rapidly growing interest in standardizing computable representations of eligibility criteria (EC) in order to develop tools that leverage data from electronic health record (EHR) systems. Although laboratory procedures (LP) represent a common entity of EC that is readily available and retrievable from EHR systems, there is a lack of interoperable data models for this entity of EC. A public, specialized data model that utilizes international, widely-adopted terminology for LP, e.g. Logical Observation Identifiers Names and Codes (LOINC®), is much needed to support automated screening tools.The aim of this study is to establish a core dataset for LP most frequently requested to recruit patients for clinical trials using LOINC terminology. Employing such a core dataset could enhance the interface between study feasibility platforms and EHR systems and significantly improve automatic patient recruitment.We used a semi-automated approach to analyze 10,516 screening forms from the Medical Data Models (MDM) portal's data repository that are pre-annotated with Unified Medical Language System (UMLS). An automated semantic analysis based on concept frequency is followed by an extensive manual expert review performed by physicians to analyze complex recruitment-relevant concepts not amenable to automatic approach.Based on analysis of 138,225 EC from 10,516 screening forms, 55 laboratory procedures represented 77.87% of all UMLS laboratory concept occurrences identified in the selected EC forms. We identified 26,413 unique UMLS concepts from 118 UMLS semantic types and covered the vast majority of Medical Subject Headings (MeSH) disease domains.Only a small set of common LP covers the majority of laboratory concepts in screening EC forms which supports the feasibility of establishing a focused core dataset for LP. We present ELaPro, a novel, LOINC-mapped, core dataset for the most frequent 55 LP requested in screening for clinical trials. ELaPro is available in multiple machine-readable data formats like CSV, ODM and HL7 FHIR. The extensive manual curation of this large number of free-text EC as well as the combining of UMLS and LOINC terminologies distinguishes this specialized dataset from previous relevant datasets in the literature.

Intensive care unit readmissions are associated with mortality and bad outcomes. Machine learning could help to identify patients at risk to improve discharge decisions. However, many models are black boxes, so that dangerous properties might remain unnoticed. In this study, an inherently interpretable model for 3-day ICU readmission prediction was developed. We used a retrospective cohort of 15,589 ICU stays and 169 variables collected between 2006 and 2019. A team of doctors inspected the model, checked the plausibility of each component, and removed problematic parts. Qualitative feedback revealed several challenges for interpretable machine learning in healthcare. The resulting model used 67 features and showed an area under the precision-recall curve of 0.119±0.020 and an area under the receiver operating characteristic curve of 0.680±0.025. This is on par with state-of-the-art gradient boosting machines and outperforms the Simplified Acute Physiology Score II. External validation with the Medical Information Mart for Intensive Care database version IV confirmed our findings. Hence, a machine learning model for readmission prediction with a high level of human control is feasible without sacrificing performance.

Background/Aim: In the field of cancer research, reconstructing clonal evolution is of major interest. The technique provides new insights for analysis and prediction of tumor development. However, reconstruction based on mutational data is characterized by several challenges. Materials and Methods: By performing extensive literature research, we identified 51 currently available tools for reconstructing clonal evolution. By analyzing two cancer data sets (n=21), we investigated the applicability and performance of each tool. Results: Seventeen out of 51 tools could be applied to our data. Correct clustering of variants can be observed for 4 patients in the presence of ≤3 clusters and ≥5 time points. Correct phylogenetic trees are determined for 10 patients. Accurate visualization is possible, by applying adjustments to the original algorithms. Conclusion: Despite bearing considerable potential, automatic reconstruction of clonal evolution remains challenging. To replace tedious manual reconstruction, further research including systematic error analyses using simulation tools needs to be conducted.

BackgroundDigital acquisition of risk factors and symptoms based on patients’ self-reports represents a promising, cost-efficient and increasingly prevalent approach for standardized data collection in psychiatric clinical routine. While the feasibility of digital data collection has been demonstrated across a range of psychiatric disorders, studies investigating digital data collection in schizophrenia spectrum disorder patients are scarce. Hence, up to now our knowledge about the acceptability and feasibility of digital data collection in patients with a schizophrenia spectrum disorder remains critically limited.ObjectiveThe objective of this study was to explore the acceptance towards and performance with digitally acquired assessments of risk and symptom profiles in patients with a schizophrenia spectrum disorder in comparison with patients with an affective disorder.MethodsWe investigated the acceptance, the required support and the data entry pace of patients during a longitudinal digital data collection system of risk and symptom profiles using self-reports on tablet computers throughout inpatient treatment in patients with a schizophrenia spectrum disorder. As a benchmark comparison, findings in patients with schizophrenia spectrum disorder were evaluated in direct comparison with findings in affective disorder patients. The influence of sociodemographic data and clinical characteristics on the assessment was explored. The study was performed at the Department of Psychiatry at the University of Münster between February 2020 and February 2021.ResultsOf 82 patients diagnosed with a schizophrenia spectrum disorder who were eligible for inclusion 59.8% (n=49) agreed to participate in the study of whom 54.2% (n=26) could enter data without any assistance. Inclusion rates, drop-out rates and subjective experience ratings did not differ between patients with a schizophrenia spectrum disorder and patients with an affective disorder. Out of all participating patients, 98% reported high satisfaction with the digital assessment. Patients with a schizophrenia spectrum disorder needed more support and more time for the assessment compared to patients with an affective disorder. The extent of support of patients with a schizophrenia spectrum disorder was predicted by age, whereas the feeling of self-efficacy predicted data entry pace.ConclusionOur results indicate that, although patients with a schizophrenia spectrum disorder need more support and more time for data entry than patients with an affective disorder, digital data collection using patients’ self-reports is a feasible and well-received method. Future clinical and research efforts on digitized assessments in psychiatry should include patients with a schizophrenia spectrum disorder and offer adequate support to reduce digital exclusion of these patients.

Background Medical information management for stroke patients is currently a very time-consuming endeavor. There are clear guidelines and procedures to treat patients having acute stroke, but it is not known how well these established practices are reflected in patient documentation. Objective This study compares a variety of documentation processes regarding stroke. The main objective of this work is to provide an overview of the most commonly occurring medical concepts in stroke documentation and identify overlaps between different documentation contexts to allow for the definition of a core data set that could be used in potential data interfaces. Methods Medical source documentation forms from different documentation contexts, including hospitals, clinical trials, registries, and international standards, regarding stroke treatment followed by rehabilitation were digitized in the operational data model. Each source data element was semantically annotated using the Unified Medical Language System. The concept codes were analyzed for semantic overlaps. A concept was considered common if it appeared in at least two documentation contexts. The resulting common concepts were extended with implementation details, including data types and permissible values based on frequent patterns of source data elements, using an established expert-based and semiautomatic approach. Results In total, 3287 data elements were identified, and 1051 of these emerged as unique medical concepts. The 100 most frequent medical concepts cover 9.51% (100/1051) of all concept occurrences in stroke documentation, and the 50 most frequent concepts cover 4.75% (50/1051). A list of common data elements was implemented in different standardized machine-readable formats on a public metadata repository for interoperable reuse. Conclusions Standardization of medical documentation is a prerequisite for data exchange as well as the transferability and reuse of data. In the long run, standardization would save time and money and extend the capabilities for which such data could be used. In the context of this work, a lack of standardization was observed regarding current information management. Free-form text fields and intricate questions complicate automated data access and transfer between institutions. This work also revealed the potential of a unified documentation process as a core data set of the 50 most frequent common data elements, accounting for 34% of the documentation in medical information management. Such a data set offers a starting point for standardized and interoperable data collection in routine care, quality management, and clinical research.

The current draft of the German Hospital Structure Law requires remuneration to incorporate quality indicators. For neurosurgery, several quality indicators have been discussed, such as 30-day readmission, reoperation, or mortality rates; the rates of infections; or the length of stay. When comparing neurosurgical departments regarding these indicators, very heterogeneous patient spectrums complicate benchmarking due to the lack of risk adjustment. In this study, we performed an analysis of quality indicators and possible risk adjustment, based only on administrative data. All adult patients that were treated as inpatients for a brain or spinal tumour at our neurosurgical department between 2013 and 2017 were assessed for the abovementioned quality indicators. DRG-related data such as relative weight, PCCL (patient clinical complexity level), ICD-10 major diagnosis category, secondary diagnoses, age and sex were obtained. The age-adjusted Charlson Comorbidity Index (CCI) was calculated. Logistic regression analyses were performed in order to correlate quality indicators with administrative data. Overall, 2623 cases were enrolled into the study. Most patients were treated for glioma (n = 1055, 40.2%). The CCI did not correlate with the quality indicators, whereas PCCL showed a positive correlation with 30-day readmission and reoperation, SSI and nosocomial infection rates. All previously discussed quality indicators are easily derived from administrative data. Administrative data alone might not be sufficient for adequate risk adjustment as they do not reflect the endogenous risk of the patient and are influenced by certain complications during inpatient stay. Appropriate concepts for risk adjustment should be compiled on the basis of prospectively designed registry studies.

Lower extremity artery disease (LEAD) affects millions of elderly patients and is associated with elevated cardiovascular morbidity and mortality. Risk factor modification, including the therapy of dyslipidaemia, is mandatory to reduce cardiovascular event rates and to improve survival rates. However, only a minority achieve the recommended low-density lipoprotein cholesterol (LDL-C) target level < 55 mg/dL, according to the current ESC/EAS guidelines on the treatment of dyslipidaemia. This study elucidated the implementation of the lipid-lowering guideline recommendations of 400 LEAD patients with LDL-C > 100 mg/dL and their adherence to treatment adjustment during follow-up. Despite a sustained statin prescription in 93% of the patients, including 77% with high-intensity statins at follow-up, only 18% achieved the target level. Ezetimibe appeared in 21% and LDL-C goals were reached significantly more often with combination therapy. Recurrent revascularization appeared more often (28%) than coronary artery or cerebrovascular disease progression (14%) and 7% died. Despite the frequent use of high-intensity statins and expandable rates of ezetimibe, the progression of cardiovascular events remained inevitable. Only 18% of the patients had received recommendations on lifestyle modification, including dietary adaptations, which is key for a holistic approach to risk factor control. Thus, efforts for both pharmacological and behavioral strategies are needed to improve clinical outcomes and survival rates.

The Portal of Medical Data Models has been developed since 2011 by the University of Münster. Its main goals are transparency, standardization and secondary use of medical metadata. Via two online surveys feedback from stakeholders of German health research was collected regarding the portal's contents. The surveys confirmed great interest in secondary use of medical forms.

Smartwatches provide technology-based assessments in Parkinson’s disease (PD). It is necessary to evaluate their reliability and accuracy in order to include those devices in an assessment. We present unique results for sensor validation and disease classification via machine learning (ML). A comparison setup was designed with two different series of Apple smartwatches, one Nanometrics seismometer and a high-precision shaker to measure tremor-like amplitudes and frequencies. Clinical smartwatch measurements were acquired from a prospective study including 450 participants with PD, differential diagnoses (DD) and healthy participants. All participants wore two smartwatches throughout a 15-min examination. Symptoms and medical history were captured on the paired smartphone. The amplitude error of both smartwatches reaches up to 0.005 g, and for the measured frequencies, up to 0.01 Hz. A broad range of different ML classifiers were cross-validated. The most advanced task of distinguishing PD vs. DD was evaluated with 74.1% balanced accuracy, 86.5% precision and 90.5% recall by Multilayer Perceptrons. Deep-learning architectures significantly underperformed in all classification tasks. Smartwatches are capable of capturing subtle tremor signs with low noise. Amplitude and frequency differences between smartwatches and the seismometer were under the level of clinical significance. This study provided the largest PD sample size of two-hand smartwatch measurements and our preliminary ML-evaluation shows that such a system provides powerful means for diagnosis classification and new digital biomarkers, but it remains challenging for distinguishing similar disorders.

Smartwatches provide technology-based assessments in Parkinson’s disease (PD). We present results for sensor validation and disease classification via Machine Learning (ML). A comparison setup was designed with two different series of Apple smartwatches, one Nanometrics seismometer and a high-precision shaker to measure tremor-like amplitudes and frequencies. Clinical smartwatch measurements were acquired from a prospective study including 450 participants with PD, differential diagnoses (DD) and healthy participants. All participants wore two smartwatches and within a 15-min examination. Symptoms and medical history were captured on the paired smartphone. A broad range of different ML classifiers were cross-validated. Amplitude and frequency differences between smartwatches and the seismometer were under the level of clinical significance. The most advanced task of distinguishing PD vs DD was evaluated with 74,1% balanced accuracy, 86,5% precision and 90,5% recall by Multilayer Perceptrons. Deep Learning architectures significantly underperformed in all classification tasks. Smartwatches are capable of capturing subtle-tremor signs with low noise. This study provided the largest PD sample size of two-hand smartwatch measurements and our preliminary ML-evaluation shows that such a system provides powerful means for diagnosis classification and new digital biomarkers but it remains challenging for distinguishing similar disorders.

BACKGROUND Medical information management for stroke patients is currently a very time-consuming endeavor. There are clear guidelines and procedures to treat patients having acute stroke, but it is not known how well these established practices are reflected in patient documentation. OBJECTIVE This study compares a variety of documentation processes regarding stroke. The main objective of this work is to provide an overview of the most commonly occurring medical concepts in stroke documentation and identify overlaps between different documentation contexts to allow for the definition of a core data set that could be used in potential data interfaces. METHODS Medical source documentation forms from different documentation contexts, including hospitals, clinical trials, registries, and international standards, regarding stroke treatment followed by rehabilitation were digitized in the operational data model. Each source data element was semantically annotated using the Unified Medical Language System. The concept codes were analyzed for semantic overlaps. A concept was considered common if it appeared in at least two documentation contexts. The resulting common concepts were extended with implementation details, including data types and permissible values based on frequent patterns of source data elements, using an established expert-based and semiautomatic approach. RESULTS In total, 3287 data elements were identified, and 1051 of these emerged as unique medical concepts. The 100 most frequent medical concepts cover 9.51% (100/1051) of all concept occurrences in stroke documentation, and the 50 most frequent concepts cover 4.75% (50/1051). A list of common data elements was implemented in different standardized machine-readable formats on a public metadata repository for interoperable reuse. CONCLUSIONS Standardization of medical documentation is a prerequisite for data exchange as well as the transferability and reuse of data. In the long run, standardization would save time and money and extend the capabilities for which such data could be used. In the context of this work, a lack of standardization was observed regarding current information management. Free-form text fields and intricate questions complicate automated data access and transfer between institutions. This work also revealed the potential of a unified documentation process as a core data set of the 50 most frequent common data elements, accounting for 34% of the documentation in medical information management. Such a data set offers a starting point for standardized and interoperable data collection in routine care, quality management, and clinical research.

INTRODUCTION: Molecular subgrouping of Medulloblastoma (MB) has expanded our understanding of its biology and the impact on clinical parameters. However, detailed analysis of inter- and intratumoral heterogeneity on a metabolic level is currently lacking. Within this study, we aimed at improving our understanding of metabolic heterogeneity between the MB subgroups, between samples within these subgroups and how these differences affect prognosis. METHODS: We analyzed metabolic characteristics of four MB cohorts covering 1,804 samples in total. In 911 samples (ICGC and MAGIC cohort), we explored metabolic programs on RNA level. In two cohorts (ICGC and G3/G4 samples from the HIT cohort; n=1,035) we examined genetic alterations on DNA level. Furthermore, single-cell RNA-sequencing data of six samples were used to explore intratumoral metabolic heterogeneity. Inter- and intratumoral heterogeneity were correlated to clinical data. RESULTS: Using publicly available gene signatures, we discovered significant differences in metabolic gene expression comparing established MB subgroups. Three metabolically distinct clusters of G3/G4 samples could be defined by unsupervised analyses in two independent cohorts. We were able to confirm our finding of intertumoral metabolic differences on single-cell RNA level. Additionally, our analysis revealed the possibility of sample-specific metabolic features. On DNA level, we identified regulatory genes with known role in MB development to be predominantly associated with lipid metabolic processes. After all, lipid metabolism and metabolism of nucleotides in MB have prognostic value and correlate with the outcome of patients. CONCLUSION: Our data highlight the importance of metabolic properties in MB. We show the distinct metabolic signatures are clinically relevant and, thus, might provide opportunities for novel target-directed therapeutic options in the future.