Joeky Senders

172 Chapter 10 major complications, and intracranial hemorrhage as the most common reason for reoperation. These results indicate blood coagulation as a primary challenge in the perioperative management of glioblastoma patient with a careful balance, often deviating in both directions. In a subsequent in-depth analysis ( Chapter 3 ) , intracranial hemorrhages occurred predominantly within the first days of surgery, whereas the risk of thrombogenic complications, and pulmonary embolisms in particular, extended beyond the period of hospitalization. The hemorrhagic and thrombogenic risk patterns, which diverge over time, suggest caution with regards to starting anticoagulation shortly after surgery, as well as a potential role for continuing it beyond the period of hospitalization. In a retrospective cohort study investigating this prophylactic strategy ( Chapter 4 ), the rate of venous thromboembolism remained nevertheless similar in patients receiving short (i.e., up to discharge) versus prolonged (i.e., 21 days after surgery) thromboprophylaxis. A higher rate of intracranial hemorrhages was even observed in the latter group. Based on these findings, we do not recommend the routine use of prolonged thromboprophylaxis in patients undergoing craniotomy for high-grade glioma. Part I characterized risk factors of postoperative complications, as well as the safety and efficacy of thromboprophylaxis, in patients undergoing craniotomy for a primary malignant brain tumor. However, the interpretable coefficients to quantify these effects remain group-level estimates and do not necessarily apply to each individual patient to the same extent. After all, the risk of venous thromboembolism in the individual patient can be very different from the cohort’s average. Although routine use of prolonged thromboprophylaxis did not significantly reduce the rate of venous thromboembolism at the group-level, this does not preclude selected individual patients to benefit from this strategy. Predictive analytics could help in personalizing clinical decision-making to the characteristics and needs of the individual patient. Part II: Predictive analytics in neurosurgical oncology In Chapter 5 , we developed a model to predict survival in the individual glioblastoma patient. We trained several statistical and machine learning algorithms based on structured demographic, socio-economic, clinical, and radiographic information. The accelerated failure time model demonstrated superior performance in terms of discrimination, calibration, interpretability, predictive applicability, and computational efficiency compared to Cox proportional hazards regression and other machine learning algorithms.