Page 43 - Risk quantification and modification in older patients with colorectal cancer

Discussion
We identified 26 prediction models out of 25 studies for postoperative outcomes of colorectal surgery; ten models studied mortality as an outcome and seven anastomotic leakages. Other outcomes were surgical complications, gastrointestinal problems (including prolonged ileus), perioperative cardiac events, readmissions, and discharge not to home. The average age of included patients ranged from 61 to 76. Two models were exclusively developed for patients of 65 and older. We found no models with quality of life or functional dependency as an outcome. Age, gender and ASA score were common predictors. Twelve studies included intraoperative predictors, such as surgical extent, the distance of the anastomosis, duration of surgery, and intraoperative complications, including both models for older patients, which limits their use for preoperative decision making.29,34
There were methodological concerns relating to sample in size (28%), missing external validation (42%) and not reporting on calibration (28%). Information bias and analysis bias was considered moderate to high in 22 studies (88%). In external validation studies, discrimination and calibration were more likely to be worse compared to the original study. Based on the applicability and methodological concerns, no useful model for older patients was identified that could be used for preoperative shared decision making.
For older patients risks and benefits of treatment must be weighted at an individual level. Identification of high-risk patients enables the initiation of geriatric interventions such as prehabilitation47 that could reduce the risk. In older medical oncology patients, a Geriatric Assessment (GA) has been shown to reveal previously unknown medical issues that are associated with poor outcomes of treatment,48,49 including surgical oncology.50 Potential predictors of surgical outcomes in older patients are comorbidity, functional dependency 13,18,26, falls and cognitive impairments.51 Introduction of such predictors in existing prediction tools may improve a prediction model’s performance for older patients.
Methodological concerns affect clinical applicability and generalizability of prediction models. Especially in small datasets, the effect of included predictors may be overestimated.15,19,23,34 Hence, alternative methods are available for the selection process of candidate predictors to reduce this risk of overestimation. These
Risk prediction models for CRC patients
41
2