Page 27 - Risk quantification and modification in older patients with colorectal cancer

were not eligible for inclusion in this review. Neither were reviews, editorials and conference abstracts.
Predefined outcomes of interest were any postoperative morbidity (for example, complications, readmission, hospital stay, functional and quality of life outcomes) and postoperative mortality up to 12 months.
All titles and abstracts of the studies retrieved by the search were addressed by two reviewers (ETDS and EB), to determine which studies warranted further examination. Articles in other languages then English, German, French or Dutch were excluded.
The following studies were excluded based on title and abstract: treatment options other than colorectal surgery, no original research, non-human studies, only a subgroup of patients (e.g. only lung metastasis or liver metastasis), the inclusion of postoperative variables in the final model or the outcome of interest was not postoperative morbidity of mortality. All potentially relevant articles were subsequently screened as full text by two authors (ETDS and EB). In the case a model update was published, the updated model was included in the review, but study information of the original study model was used when applicable. Furthermore, references of included publications were cross-referenced to retrieve any additional relevant citations. Finally, only studies that had a score chart or nomogram published or online/offline calculator made available were eligible for data-extraction.
Data extraction and quality assessment
The CHecklist for critical Appraisal and data extraction for Systematic Reviews of prediction Modelling Studies (CHARMS) was used for data extraction.7 For each included study, the following data were independently extracted by two investigators (ETDS, EB): Study date, data ssource (cohort, case-case control, randomised trial or registry data), study population (age, gender, tumour stage and type of surgery), outcome of interest, number of outcome events reported, predictors included in the final model. The final model’s performance was assessed based on its discrimination (AUC of the c-statistic/index, sensitivity and specificity to calculate a Likelihood Ratio) and calibration (accuracy of the predicted risk versus the observed risk, and reported by Hosmer-Lemeshow (H-L) test value,
Risk prediction models for CRC patients
25
2