Joeky Senders

16 Chapter 1 THESIS OUTLINE Part I: Outcomes and risk factors in neurosurgical oncology This part aims to provide an introduction into the characteristics and outcomes in patients undergoing brain tumor surgery. It comprises three retrospective studies that characterize the incidence and risk factors of postoperative morbidity and mortality. Chapter 2 examines the incidence and predictors of the occurrence of a major complication, extended length of stay, reoperation, readmission, and death within 30 days after surgery. Due to its high incidence and substantial impact on postoperative morbidity, a subsequent in-depth analysis ( Chapter 3 ) was performed in the same cohort to characterize the rates, timing, and predictors of venous thromboembolism and intracranial hemorrhage. This chapter proposes a strategy for optimizing postoperative thromboprophylaxis – continuing prophylactic anticoagulation up to 21 days after surgery – which is evaluated on institutional data in Chapter 4 . Part II: Predictive analytics in neurosurgical oncology This part constitutes a study that evaluates the utility of statistical and machine learning algorithms for outcome prediction in neuro-oncology. Chapter 5 provides a multimodal assessment of a variety of algorithms by evaluating their ability to predict survival in the individual glioblastoma patient based on structured demographic, socio- economic, clinical, and radiographic information. To facilitate the reproducibility and external validation of this chapter, the overall best performing model was deployed as an online survival calculator for glioblastoma patients. Part III: Natural language processing in neurosurgical oncology This part evaluates the utility of natural language processing algorithms for automating clinical chart review in medical research. Chapter 6 presents a natural language processing framework for automating the extraction of clinical information from free- text clinical reports. In this chapter, we analyze a text corpus of radiology reports of glioblastoma patients with a regression-based algorithm. Additionally, we characterize the association between model performance and statistical properties of the variables of interest to provide insight into the methodological boundaries and variables eligible for clinical text mining. In Chapter 7 , we develop several models to classify radiology reports of brain metastases patients into reports that describe solitary versus multiple metastases. This chapter includes an extensive comparison between various statistical, classical machine learning, and deep learning algorithms to provide insight into their