Timo Soeterik

174 CHAPTER 10 lymph node metastasis at primary staging, use of a nomogram to predict the risk of LNI will not impact change of management. However, since sensitivity of PSMA PET-CT at present remains moderate (40%), more than half of patients with pelvic LNI remain node-negative on PSMA PET-CT. 27–29 For these patients, nomograms can still be of value to decide whether ePLND should be performed. Furthermore, recent data of Meijer and colleagues showed that combining mpMRI information with PSMA PET-CT outcomes can improve the negative predictive value. 30 In case the number of patients undergoing PSMA PET-CT increases, sufficient data will become available to develop novel nomograms including PSMA PET-CT information. Combined with mpMRI data, such as the PI-RADS score and the mpMRI T-stage, this can potentially lead to promising new prediction tools further improving the staging of primary prostate cancer. Artificial intelligence and predictive modelling The traditional risk prediction tools are mostly based using on multivariable logistic regression analysis. Advanced statistical and mathematical techniques, such as machine learning and deep learning algorithms, can further enhance risk prediction. The main limitation of a logistic regression based prediction tool is that the outcome is rather static. 31 For instance, one can calculate the risk of pelvic LNI in newly diagnosed patients based on clinical stage, PSA, relative number of positive cores and the highest ISUP grade on biopsy. However, if information derived from other staging modalities, not included in the model becomes available, the nomogram-calculated risk should be interpreted differently. For example, if the nomogram-calculated risk of LNI in a patient is 10%, but subsequent PSMA PET-CT shows high suspicion for LNI, this risk of LNI has drastically increased. The advantage of machine learning tools is that they can be updated faster and more efficient by delivering novel data; compared to traditional logistic regression models. For developing a novel model; specific steps should be undertaken every time which are time-consuming. 32 A machine learning tool enabling immediate model updating and validation based on recent data collected during routine clinical care, can substantially speed up the development of novel highly accurate prediction tools. Value-Based Healthcare The Santeon Value-Based Healthcare Initiative (VBHC) for prostate cancer formed an excellent basis for the research presented in this thesis. This quality improvement initiative, started in 2012, had resulted in interesting insights as well as accumulated data proven to be very useful for scientific research. Although it remains difficult to define the added value of such quality improvement initiatives, the collaboration itself has proven to be a successful starting point for several research initiatives. In general, the adoption of VBHC as a concept is growing on a national and international level. Within Santeon, number of staff members involved in VBHC teams are growing. During the study period