Page 121 - Risk quantification and modification in older patients with colorectal cancer

GerCRC risk prediction model for severe complications
Based on the weight of the regression coefficients, all predictors were given 2 points except for tumour location (1 point) and the (self-reported) need for ADL assistance (1 point). In the simplified model, a total of 14 points can be obtained (Table 3A). The number of patients with a score of 0 or 1 was 171 (16%) and 46 (4%), respectively. The maximum score obtained by patients in our study was 11 (n=3), of whom two (67%) had a severe complication.
After grouping patients with a score of 0-1 and 7 or higher, Table 3B shows the corresponding predicted proportion of complications with corresponding sensitivity and specificity. At a score of 5, the difference between predicted risk and observed risk was 6% (19% versus 13%), at a score of 7 this was 14% (31% versus 45%, respectively).
Table 2 Model development and multivariable regression coefficients after shrinkage
Demographic Model
Geriatric Model
Betaa
-2.64 0.04 0.32 -
-
- 0.34 0.35 0.02 0.12 -
-
- 0.33 0.09 0.16 0.43 0.35
0.687 0.650
Predictors
Cohort Model Estimates
Intercept
Age (for every 10 years)
Male gender
BMI, kg/m2
History of abdominal surgery Cardiac comorbidity COPD/ASTMA/Emfysema Previous PE or DVTb
ASA score
Rectal tumour
Tumour stage
Reported falls
Risk for malnutrition
Previous delirium
Self-reported cognitive impairment Self-reported need for ADL assistance Mobility aid
Polypharmacy (≥5)
Mode performance (AUC)
Model after bootstrapping
Optimism corrected model
a Regression coefficient after shrinkage using LASSO method - candidate predictor was not selected after shrinkage
* candidate predictor was not used in model development
b PE, Pulmonary Embolism; DVT, Deep Venous Thrombosis
Betaa
-6.64 0.14 0.26 -
-
- 0.27 0.37 0.2 0.03 *
*
*
*
*
*
*
*
0.648 0.623
119
6