Jordy van Sambeeck

Chapter 2 34 Observers Four observers with different medical experience performed all of the measurements mentioned above on radiograph, CT and MRI in 2 cycles with a minimum of 4 weeks apart. The four observers included an orthopedic surgeon, a radiologist, an orthopedic resident and a medical student. IRB approval was received from the review board of Canisius Wilhelmina Hospital, Nijmegen, The Netherlands, ID number: 039-2017. Statistical analysis For each measurement (IS, BP, CD and MIS on CR, CT and MRI) the intra- and interrater reliability was determined by calculating the intra-class correlation coefficient (ICC). To compare results of the IS, BP, CD and MIS measurements between CR, CT and MRI (inter-method reliability) an ICC was also calculated. The ICC estimates the average correlation among pairs of data and gives a value between 0 and 1. Scores were interpreted as follows: a score of 0 to 0.50 indicating poor reliability, 0.50 to 0.75 indicating moderate reliability, a score of 0.75 to 0.90 indicating good reliability, and a score higher than 0.90 indicating excellent reliability [29]. The ICC is a qualitative measure of reproducibility. In order to further quantify the reliability of the measurements the Bland Altman analysis was used to assess agreement [2]. It evaluates the mean difference in measurements and a range of agreement within which 95% of the differences between one measurement and the other are included. A Bland Altman analysis was performed for measurements with an ICC ≥ 0.70. A mean difference < 0.20 on the different measurement methods was deemed acceptable for clinical use. SPSS software (version 24.0, SPSS Inc., Chicago, IL, USA) was used to calculate the intra-class correlation coefficients and Microsoft Office Excel (version 14.0, Microsoft Corp., Redmond, WA, USA) was used for all Bland Altman analyses calculating mean differences and limits of agreement. We included all patients who had all three imaging modalities available in the selected time frame. In a review article, Bujang et al. [9] provide a guide to determine the minimum sample size required for estimating the desired effect size of ICC. According to this guide, the minimum sample size requirement for our study is 25 subjects when alpha is pre-specified to be 0.05, power to be 0.90, an acceptable ICC of 0.70 and an expected ICC of 0.9.