Page 117 - Risk quantification and modification in older patients with colorectal cancer

GerCRC risk prediction model for severe complications
Imputation, LASSO shrinkage and bootstrap validation were analysed with R (Version 3.5.2) using “mice”, “rms”, “glmnet” packages. All other analyses were performed using SPSS version 23.0 (SPSS, Inc., Chicago, IL).
Results
Participants
The total cohort consisted of 1366 older patients who underwent colorectal resection between January 2014 and December 2017 (Figure 1). From one hospital, data was only available from January 2014 until December 2015 because of a change in EHR registration. There were no missing demographic data (Table 1). The number of complete cases was 977 (89.8%), 87 cases (8%) had one missing candidate predictor, 24 (2%) had 2 or more missing candidate predictors. Mean age was 77.7 (SD 5.2), there were 498 (46%) females, 270 (25%) patients with rectal cancer and 354 (33%) had an ASA score of III or IV.
Model development
There were 171 patients (16%) with one or more severe complications recorded; 51 patients were admitted to the ICU for more than two days, 26 of whom had a reintervention. A total of 121 patients (including 29 ICU patients) had a hospital stay of > 14 days; 30-day mortality was 1.7% (n=19). The distribution of severe complications is available in Appendix A.
Unadjusted associations between each candidate predictor and severe complications are shown in Table 1 and Appendix B. For the demographic model development, with only demographic candidate predictors, the final predictors were age, gender, COPD/Asthma/Emphysema, previous PE or DVT, ASA score and tumour location. The AUC of the demographic model was 0.65 (95% CI 0.62- 0.70), which was corrected to AUC 0.62 after internal validation.
The discriminatory performance of the preoperative model improved to 0.69 (95% CI 64-0.73) when the geriatric predictors delirium, cognitive impairments, ADL assistance, the use of mobility aid, and polypharmacy were included. The optimism corrected AUC was 0.65. Table 2 shows the regression coefficients of the demographic and geriatric models.
115
6