Page 124 - Risk quantification and modification in older patients with colorectal cancer

Chapter 6
performance improved (data not shown). When our GerCRC model will be externally validated, more focus will be on the calibration of the different risk groups to judge the performance and clinical usefulness of this model.37
A head-to-head comparison with the POSSUM,14 CR-BHOM18 and ACS-NSQIP original and recently published universal model10,19 is with caution because differences in the definition of severe complications, the use of perioperative predictors and the lack of external validation. The GerCRC model is the only model that uses a prolonged length of hospital stay in the definition of a severe complication, accounting for a possible negative impact of a prolonged hospital stay on physical functioning and quality of life. The use of perioperative predictors in the other models limits preoperative decision making.
External validation for all models (including the GerCRC model) has not been performed or was shown to be somewhat disappointing for older patients. The POSSUM was shown to overpredict complication and mortality risk. A recent evaluation of the performance of the POSSUM in 1380 UK patients (with surgery between 2008-2013) confirmed its poor discriminatory performance for severe complications (AUC 0.51).38 The discriminatory performance for prediction morbidity in 204 Portuguese octogenarians, was 0.65 for the POSSUM and 0.66 for the CR-BHOM model with poor calibration.39 The original ACS-NSQIP surgical risk model was not specifically developed for colorectal cancer surgery and also the accuracy of the universal ACS-NSQIP model for severe complications or its performance for outcomes in older CRC patients has not been published. The accuracy of the universal ACS-SNQIP model for severe complications in 200 older gynecologic oncology patients undergoing laparatomie (2009-2013), was only 0.62also with poor calibration.40 To account for possible heterogeneity between cohorts30 external validation of the proposed prediction models is required; also changes in the healthcare setting and geographic differences are reasons for periodic updating and recalibration.41 This applies to the ACS-NSQIP model that had not been validated outside the USA, as well as for the GerCRC model. More detailed comparison of the preoperative GerCRC, CR-BHOM and ACS-NSQIP models are shown in Appendix C.
122