Page 104 - The value of total hip and knee arthroplasties for patients

Chapter 6
95% (95% CI:92-98%)).The KL grade in our study was classified as mild in KL 0-2 and severe in KL3-4.
Sociodemographic variables and patient characteristics
Education level was scored on a 8-point scale with answering options representing the education levels in The Netherlands, scores were dichotomized in low level (no education to lower vocational education) versus high level (intermediate vocational education to university). Self-reported weight and height were used to calculate Body Mass Index (BMI)
Analysis
Multivariate linear regression analyses were employed with postoperative pain (KOOS/HOOS pain) and function (KOOS/HOOS function) as dependent variables. Besides the expectation related variables (general outcome expectations, specific outcome expectations and credibility) we selected 7 variables measured preoperatively as candidate predictors of outcome namely preoperative pain, preoperative function, gender, age, education level, BMI, Kellgren and Lawrence score (KL-score), mental health.The selection of these candidate predictors was based on discussions with orthopaedic surgeons about which predictors of outcome they consider in daily practice.
A backwards elimination method was used for these analyses. This procedure started with including all candidate variables in the model, subsequently the least significant variable was removed (the one with the highest p-value).The model was thereafter refitted without this variable, and again the least significant variable was removed.This process was repeated until all predictor variables in the model had a p-value < 0.10.
The models were first ran with the CEQ expectancy subscale as the expectations variable, in case that the CEQ expectancy subscale was included in the final model, this final model was repeated while replacing the CEQ expectancy subscale with the HSS expectation survey subscale corresponding to the outcome of that model (so the HSS expectation function score was used for the models with function as the dependent variable and the HSS expectation pain score was used for the models with pain as the dependent variable). If the CEQ expectancy subscale was not included in the final model, the backwards elimination procedure was completely repeated with the HSS expectation survey score as a candidate predictor instead of
102