What is a man’s risk on having event X (symptoms from metastases or prostate cancer death) when he starts following active surveillance? And what are the risks of side effects of a particular treatment (e.g., urinary dysfunction, bowel dysfunction, and erectile dysfunction) and loss of quality of life? The decision to pursue active surveillance is entirely the man’s, but it is the urologist’s responsibility to make sure that the man fully understands all the benefits and risks of active surveillance, as well as the benefits and risks of other treatment options available [11]. In order to assist both clinicians and patients in this critically important treatment-related decision-making process , there are a number of questions which will be addressed:

These unresolved issues in active surveillance for prostate cancer require further study and are identified by the GAP3 clinical experts as the key questions of the global initiative .

What is the most informative data needed to answer these research questions? To achieve a consensus on the design and contents of the global active surveillance database, there was a need for a minimal definition of the GAP3 codebook. A consensus has been reached on the codebook of the GAP3 database among the clinical investigators. The codebook of the global database consists of four sections: inclusion, follow-up, before diagnosis, and end of active surveillance. At time of inclusion, the consortium is interested in recording host characteristics (e.g., age, BMI, race, ethnicity, marital status, educational level, life expectancy, comorbidities/overall health status, etc.) and tumor characteristics (e.g., clinical stage, PSA, biopsy Gleason score , PSA density, number of positive cores, maximum extent cancer per core, etc.). During follow-up, information is gathered on, e.g., PSA, PSA kinetics (PSADT, PSAV), T-stage by DRE, biopsy characteristics, and MRI findings. PSA before diagnosis is also registered. With regard to the end of active surveillance, an inventory is made of, e.g., the reasons for stopping active surveillance, type of metastasis, type of treatment, and cause of death. As a next step, an inventory on available data has been made, including potential markers, imaging, QOL data, and others.

Based on the consensus contents of a global active surveillance database, Philips Research has constructed the GAP3 database. The IT infrastructure of the global database is based on the tranSMART prostate cancer instance of the TraIT IT infrastructure developed by Philips within the Dutch CTMM-TraIT (http://​www.​ctmm-trait.​nl) and CTMM-PCMM (http://​www.​ctmm.​nl/​en/​projecten/​kanker/​pcmm?​set_​language=​en) projects. This infrastructure offers support for collecting and combining the various large, longitudinal datasets from the participating institutes (Fig. 13.3). Transfer of these datasets takes place using the Secure Data Transfer (SDT) tool provided by Philips. The clinical data has been gathered using a global data model, specifically designed to contain all the data items needed by the statisticians to answer the research questions defined by the principal investigators at the start of the project. During this process, several issues arose, such as low data quality and incompatibility of the local data models with the global data model. These issues were solved using quality check software and data mapping software that were developed in-house. The global dataset can be browsed through the web-based tranSMART platform, which is only accessible for a selected group of statisticians. tranSMART supports a number of statistical analyses, such as survival analysis, logistic regression, and correlation analysis, and can incorporate genomic data and imaging metadata as well. The tranSMART instance is connected to RStudio [13] to enable the statisticians to execute their own R [14] scripts on the database, in a secure and reproducible manner.

For quality control of the biopsy Gleason scores used for inclusion of patients for active surveillance and for decision-making during follow-up, a centralized pathology review of approximately 5% of the biopsies by virtual microscopy is currently conducted. It concerns a random selection of 5% of confirmatory biopsies of the active surveillance population of each of the participating centers included in GAP3. The hematoxylin-eosin-stained biopsy slides will be reviewed for Gleason score (primary and secondary GS) and the biopsy core length. In addition, data is collected on the maximum number of cores per cassette and the extent of carcinoma and the total core length. The results will be taken into account in the final statistical analysis. Expected improvements for patient care include the standardization of pathology grading of prostate cancers diagnosed on biopsies, the development of a quality parameter for prostate biopsy core length, and improved selection of men entering an active surveillance program.

Although described broadly as a management option for low-risk prostate cancer patients, there is a semantic heterogeneity in the literature and guidelines on active surveillance. For instance, the specific definitions of the terms active surveillance and watchful waiting (WW) are inconsistent in the published literature and can elicit significant confusion. Further, various definitions of low-risk prostate cancer exist in these guidelines, as specified by different combinations of clinical criteria including clinical and pathological characteristics. Also, definitions of disease reclassification and progression differ among published guidelines, and multiple criteria for the initiation of curative treatment are proposed. Problems resulting from the use of ambiguous language include hindered clinical decision-making, particularly in multidisciplinary collaborations , and limited opportunities for research [15]. Further, it has raised a barrier that hampers exchange of knowledge within and between fundamental domains of research and research groups [16]. An important step toward global consensus was therefore to define some sort of “new (but uniform) Active Surveillance language.” Consensus definitions were derived using a modified Delphi method in which a panel of leading prostate cancer specialists in the field participated [17].

The aim of the GAP3 initiative is to create a global consensus on selecting and monitoring men with low-risk prostate cancer. As a first action, the project leaders started with the development of a guideline consensus on active surveillance based on a review of the current guidelines available around the world. Existing guidelines on active surveillance for clinically insignificant prostate cancer were systematically reviewed to provide a comprehensive view of the recommendations regarding patient selection, monitoring during active surveillance, and disease progression [18].

The consensus-based guidelines will be adapted based on the outcomes of the statistical analysis of the database. A brief description of the statistical analyses plan is provided below.

A conceptual framework was developed to identify which patients are most suitable for active surveillance, relating life expectancy, risk, and treatment choice (Fig. 13.4). If the patient’s life expectancy is relatively short, watchful waiting is the preferable option. If the patient’s life expectancy is relatively long, and the risk of adverse outcomes due to prostate cancer is relatively high, immediate treatment is preferred. Active surveillance is suitable for men with a relatively long life expectancy and low risk of adverse outcomes. The adverse side effects of treatment are avoided, and if disease progression occurs, curative treatment may still be initiated. The key challenge here is to reasonably weigh the harms and benefits of all treatment options and identify the preferred treatment option for a patient.

Two possible end points are relevant for selecting patients that are fit for active surveillance, namely, reclassification (i.e., disease progression after 1 year) and progression (i.e., disease progression after 4 years). To identify patients who leave active surveillance within 1 year of commencing active surveillance, prediction models will be constructed to estimate the probability that a patient moves off the active surveillance protocol within 1 year. A similar approach will be followed when predicting the risk of leaving active surveillance after 4 years, using cox or logistic regression models. Performance measures used to assess the resulting models are discrimination, calibration, and clinical usefulness. Specific attention is needed for dealing with different patient cohorts. Final presentation of models will focus on risk stratification to categorize patients by likelihood of stopping active surveillance, and this will be compared to currently used stratification schemes.