Page 199 - Risk quantification and modification in older patients with colorectal cancer

Selecting the appropriate cancer treatment for older patients with cancer remains challenging.
Personalised treatment recommendations require the evaluation of patient-, disease- and treatment characteristics in combination with individual patient’s needs, values, and preferences to weigh gain and burden of treatment and disease. Ideally, more personalised outcome information regarding the risk of postoperative complications and mortality, but also regarding postoperative physical functioning and quality of life is available to support treatment advice.
Part I of this thesis addresses methods to quantify the risk of postoperative complications for older patients with non-metastatic CRC cancer. We have incorporated our findings into a new prediction model for severe complications of surgery. In Part II of this thesis, we have studied interventions designed to modify the risk for poor surgical outcomes in this patient group.
In this chapter, implications for future research (prognostic research, body composition research and prehabilitation research) and clinical practice (pre- and postoperative care) are discussed, and an adapted care pathway for older non- metastatic CRC patients is proposed.
Implications for future research
Prognostic Research
Prognostic research can provide tools for personalised outcome information. However, the implementation of these tools in clinical practice, requires critical evaluation. To this purpose, future prediction model studies should systematically use the TRIPOD guidelines to allow critical assessment of a model’s applicability, bias performance. For performance assessment, discrimination as well as calibration measures need to be reported, and external validation should be available before considering implementation into clinical practice.1
Using these guidelines, we concluded that most prediction models are not useful for older patients with CRC (Chapter 2). Good discrimination does not always mean there is proper calibration. As shown in Chapter 2, many prediction models
General discussion and future perspectives
197
11