Page 161 - Risk quantification and modification in older patients with colorectal cancer

Oncogeriatric care and Health Related Quality of Life
All scores were compared using two-way Analysis of Variance (ANCOVA). Predefined confounders were age, gender, tumour stage, patients living alone (patients not living together with spouse or partner), open surgery (yes or no) and postoperative treatment with chemotherapy (yes or no).
Linear mixed-effect models (i.e. covariance pattern model with unstructured error covariance matrix and maximum likelihood estimation)32 were used to study the course of HRQoL, functioning and symptom scales over time. This technique uses data efficiently by also including incomplete cases in the analyses. As a result, bias is limited and statistical power is preserved. Dependency status (analysed as a categorical time-invariant predictor: dependent versus independent), time (analysed as categorical predictor with four levels (i.e. four time points)) and confounders as collected from baseline (analysed as time-invariant predictors) were entered in the regression equation as independent variables. The interaction of dependency status and time was tested separately and when these interactions were significantly associated with the individual functioning and symptom scales, stratified analyses were performed per dependency status (i.e. independent and dependent).
The clinical relevance of the differences in HRQoL QLQ-C30 outcomes between functionally independent and dependent patients were estimated using the consensus-based guidelines of Cocks et al.33 Changes over time within a group were separately evaluated on clinical relevance.34 Both guidelines were developed to aid the interpretation of differences in HRQoL scores between groups and the interpretation of change scores over time. Differences in mean scores were categorized into trivial, small, medium or large depending on the scale. Using the global health scale as an example, a difference between two groups of 0-4 would be categorized as trivial, 4-10 as small, 10-15 as medium and >15 as large. For interpreting differences in QLQ-C38 outcomes, Norman’s rule of thumb was used, where a threshold of half an SD was regarded as a clinically relevant difference.35 All analyses were performed using SAS 9.4 (SAS, Cary, NC, USA). A p-value of < 0.05 was considered statistically significant.
159
8