Search

Contains fulltext : 233871.pdf (Publisher’s version ) (Closed access)

Background: Endovascular treatment (EVT) is effective for stroke patients with a large vessel occlusion (LVO) of the anterior circulation. To further improve personalized stroke care, it is essential to accurately predict outcome after EVT. Machine learning might outperform classical prediction methods as it is capable of addressing complex interactions and non-linear relations between variables. Methods: We included patients from the Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands (MR CLEAN) Registry, an observational cohort of LVO patients treated with EVT. We applied the following machine learning algorithms: Random Forests, Support Vector Machine, Neural Network, and Super Learner and compared their predictive value with classic logistic regression models using various variable selection methodologies. Outcome variables were good reperfusion (post-mTICI ≥ 2b) and functional independence (modified Rankin Scale ≤2) at 3 months using (1) only baseline variables and (2) baseline and treatment variables. Area under the ROC-curves (AUC) and difference of mean AUC between the models were assessed. Results: We included 1,383 EVT patients, with good reperfusion in 531 (38%) and functional independence in 525 (38%) patients. Machine learning and logistic regression models all performed poorly in predicting good reperfusion (range mean AUC: 0.53-0.57), and moderately in predicting 3-months functional independence (range mean AUC: 0.77-0.79) using only baseline variables. All models performed well in predicting 3-months functional independence using both baseline and treatment variables (range mean AUC: 0.88-0.91) with a negligible difference of mean AUC (0.01; 95%CI: 0.00-0.01) between best performing machine learning algorithm (Random Forests) and best performing logistic regression model (based on prior knowledge). Conclusion: In patients with LVO machine learning algorithms did not outperform logistic regression models in predicting reperfusion and 3-months functional independence after endovascular treatment. For all models at time of admission radiological outcome was more difficult to predict than clinical outcome.

1 Introduction Radiotherapy is essential for achieving and maintaining local control in head and neck rhabdomyosarcoma (HNRMS) patients. However, radiotherapy may cause outgrowth disturbances of facial bone and soft tissue, resulting in facial asymmetry. The aim of this study was to develop a method to visualize and measure facial asymmetry in HNRMS survivors using three‐dimensional (3D) imaging techniques. 2 Methods Facial deformity was evaluated in a multidisciplinary clinical assessment of 75 HNRMS survivors, treated with external beam radiotherapy (EBRT, n = 26) or Ablative surgery, MOulage brachytherapy, and REconstruction (AMORE, n = 49). Individual facial asymmetry was measured using 3D photogrammetry and expressed in a raw asymmetry index and a normalized sex–age–ethnicity‐matched asymmetry signature weight. Facial asymmetry was also compared between British and Dutch controls and between survivors and their matched controls. 3 Results Facial asymmetry was more pronounced with increasing age (P < 0.01) in British controls compared with Dutch controls (P = 0.04). Survivors developed more facial asymmetry than matched controls (P < 0.001). The clinical assessment of facial deformity correlated with the raw asymmetry index (r = 0.60, P < 0.001). 4 Discussion 3D imaging can be used for objective measurement of facial asymmetry in HNRMS survivors. The raw asymmetry index correlated with a clinical assessment of facial deformity. Comparisons between treatment groups seemed inappropriate given the differences in facial asymmetry between British and Dutch controls. In future studies, pretreatment images could act as matched controls for posttreatment evaluation.

This thesis presents various secure mechanisms for cross-organisational data sharing during acute stroke care. In such situations, access to the data is urgent. For all the proposals, we assume the patients' EMR are stored in a cloud system to improve accessibility and collection of medical records during the emergency. However, to protect against abuse or internal attacks in the cloud providers, the patient's EMR is stored as ciphertext in the cloud, and the encryption keys are only shared with the involved acute care teams. Additionally, the proposed protocols are used to dynamically grant and revoke access to the patient's EMR for the healthcare teams according to the needs of triage, diagnosis, hospital selection and treatment. This thesis presents mechanisms for secure cross-organisational data sharing during acute stroke care. The proposed mechanisms are applicable and generalisable to other acute and non-acute care cases. These mechanisms focus on improving data availability once the security requirements are fulfilled, so the professionals no longer need break-glass procedures, even in emergencies. Instead, access to data happens with lawful purpose, without compromising patient privacy. Furthermore, cross-organisational data sharing requires trustworthiness among organisations, which needs to be built with regulation compliance and transparency. The proposed mechanisms of this thesis may help future healthcare digital systems comply with the regulations and be more transparent, thus improving security and enabling data sharing across organisations.

In this thesis, we performed studies that add new insights into the biology of circular RNAs (circRNAs) in the heart. We characterized the human circRNA landscape in the heart from RNA-sequencing data and identified thousands of circRNAs, with some of them regulated in disease. We found that an unusually high number of circRNAs are produced from the Titin (TTN) gene and show that RNA binding motif 20 (Rbm20)-dependent exon skipping within Ttn's I-band region is associated with circRNA formation. We performed RNA-sequencing on mouse (wildtype and Rbm20 knockout) hearts to investigate Rbm20-regulated circRNA production. We show that the tight correlation between Rbm20-mediated exon skipping and circRNA formation observed in the Ttn gene does not appear a common mechanism for Rbm20 target genes, indicating that circRNA formation is not a general function of Rbm20. In addition, integrated analyses of alternative splicing and circRNA production in human and mouse hearts revealed that alternative splicing only accounts for a small proportion (~10%) of the circRNAs, with the vast majority of the circRNAs arising from constitutive exons indicating that these circRNAs are generated at the expense of their linear counterpart. Finally, we present circRNAprofiler, an R-based computational framework for a comprehensive in silico analysis of identified circRNAs. CircRNAprofiler covers different aspects of circRNA analysis ranging from differential expression analysis, genomic context, evolutionary conservation, biogenesis to functional analysis. The results obtained by running the circRNA analysis workflow implemented by circRNAprofiler can serve as a starting point for further investigation in the role of specific circRNAs in any physiological and pathological condition.

In biomedical research, it has become common to collect data that measures biological functions in different biological levels of an organism. On the biomolecular level, this means that cells and tissues can be described by data gathered from different biomolecular domains, such as by data from the genome, transcriptome or proteome. By collecting such multi modular data, it is hoped that biological processes in cells and tissues can be better modeled and understood. Ultimately, this knowledge can help to better understand health and disease in the whole organism itself. This thesis discusses some of the statistical methods that aim to integrate this type of multi modular biomolecular data.

The development of high throughput technologies has moved biomedical research into the age of omics. By tracking molecules such as DNAs, RNAs, proteins and metabolites, scientists can have better understanding of human diseases. However, it is challenging to translate the large volume of omics data into knowledge. This thesis discussed the use of statistical approaches to decipher omics data. In this thesis, various of statistical approaches were applied to study phenotypes including inflammation, hypercholesterolemia and type 2 diabetes. We argue that statistical analysis is a construction process with a series of component decisions. The components in statistical analyses include data, prior knowledge as well as statistical models. In this thesis, we also discussed other challenges in omics data analysis, such as missing observations, batch effects and the choice of statistical models.

This thesis addresses methodological problems in the evaluation of tests and biomarkers, in quite distinct types of research. The first part addresses systematic reviews and meta-analysis of diagnostic accuracy: discrimination between diseased and non-diseased (cross-sectional design; dichotomous test use). The second part presents primary studies of prognostic factors and clinical prediction rules: the probability of developing an event (continuous test result) in studies that include a time element. Both parts focus on discrimination between patient groups and the best ways to assess and express the discriminating ability of a test or model.

This thesis was part of the The BIOlogy STudy to TAilored Treatment in Chronic Heart Failure (BIOSTAT-CHF) project. BIOSTAT-CHF was a European, multicenter, multinational, prospective, observational study that was especially designed to find biological mechanisms involved with response to ESC guideline recommended pharmacological therapy. The aim and focus of this thesis was twofold. In the first part, we focused on individualized treatment of heart failure. We looked at what heart failure prediction models were published and which factors resulted in more accurate predictions. We tried to identify heart failure patients with low or high risk of mortality and/or heart failure related hospitalization. We also looked at which patients achieved recommended treatment doses for guideline recommended pharmacological therapy and which did not. In addition to this, we tried to estimate if a patient would benefit from receiving recommended treatment doses or not based on prediction models consisting solely on biomarkers. The second part, was of a more methodological nature. We looked at the consistency and robustness of different clustering methods, and we clustered patients based not on clinical, but pathophysiological parameters. We also developed two methods to analyze the enormous amounts of data collected in systems biology. The first method estimates possible existence of splice variants in gene expression data. The second method is a penalized canonical correlation analysis, which enables to analyze multiple high-dimensional datasets in such a way that a set of variables in one dataset correlates maximally with the sets of variables in all other datasets.