Page 26 - Risk quantification and modification in older patients with colorectal cancer

Chapter 2
Introduction
Older patients make up the majority of new patients with colorectal cancer, and for this heterogeneous population, risks and benefits of treatment must be weighted at an individual level.1-4 Prediction models can be used to facilitate this process and estimate the outcomes of treatment. Morbidity and mortality are important outcomes to discuss when deciding upon cancer treatment, but for older patients with cancer quality of life and retaining functional independence are also important outcomes.5
The aim of this systematic review was to study existing clinical prediction models that were developed to predict postoperative outcomes of colorectal cancer surgery. Quality and accuracy of the prediction models in older patients were studied. Furthermore, their usefulness for preoperative decision making in older patients was evaluated.
Methods
Search strategy and article selection
This systematic review is reported following the recommendations set forth by the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement.6 A literature search was performed on 1 November 2018, in both the electronic databases Pubmed and Embase. The search contained the following key elements: “colorectal”, “surgery” and “prediction” or “risk model” or “nomogram”. No limits in age, language or publication date were included in the search. The full search strategies are shown in Appendix A.
Inclusion criteria for prediction modelling studies were as follows; the study’s main goals included the development of a prediction model for postoperative outcomes of colorectal surgery. The final prediction model included more than one variable, and the model’s performance was reported as an Area Under the Curve (AUC) or C-statistic/index. It was mandatory that pre- or intraoperative predictors were included in the published prediction model. Studies examining the validity of a prediction model outside the development population (the study population on which the prediction model was developed), without calibration or model update,
24