Page 123 - Risk quantification and modification in older patients with colorectal cancer

GerCRC risk prediction model for severe complications
and tumour related predictors improved the model’s prognostic accuracy for older patients. With a AUC of 0.65 after optimism correction stronger predictions are needed for better discrimination.
Gender, COPD/Asthma/emphysema, previous PE or DVT, rectal cancer, previous delirium, self-reported need for ADL assistance and polypharmacy were selected as predictors to develop the GerCRC clinical prediction model. Gender, rectal cancer and severe comorbidity are well-known predictors for poor outcomes of colorectal surgery, also in older patients.25 We recently showed strong associations between ADL and postoperative complications21 in line with other studies in older CRC and non-CRC patients.31-33 A recent geriatric pilot of the ACS- NSQIP among orthopedic and vascular surgery patients, also identified physical functioning, the use of a mobility aid preoperatively, and cognitive functioning as important predictors for 20 of the 25 outcomes measured.20 For polypharmacy and postoperative outcomes, results have been conflicting.34
In contrast to other prediction models for mortality, anastomotic leakage or surgical site infections,9,11,15,35,36 in our study age and ASA score had no additional predictive value. This is in accordance with a study among older patients with CRC referred for GA.33 Several explanations can be put forward. First, because our study population was limited to older patients, the age distribution is smaller and therefore less likely to be discriminative. Possibly, in our model, calendar age (and possibly ASA score) were replaced by measures of age related problems such as cognitive functioning, functional performance and comorbidity. Second, in the Netherlands, national guidelines recommend geriatric screening of older patients planned for CRC surgery to identify high-risk surgical patients and guide interventions or adapt treatment plans. This means our study population could be somewhat selected, as we have no information on the non-surgically treated older patients in our cohort.
After interval validation, the expected discrimination of our model was 0.65. Because we aimed to develop a model that can be used in preoperative decision- making, we did not include predictors such as the surgical technique (laparoscopic surgery or not) or perioperative complications, were not included. Also, high- risk patients such as a patient with metastatic disease or acute surgery were not included4 When these predictors and patients were added, the GerCRC model
121
6