P14.109 Standard error of measurement (SEM) and intraclass coefficient (ICC) in volumetric assessment of postoperative residual tumor volume in glioblastoma

Abstract

Abstract BACKGROUND Residual contrast-enhancing tumor volume (RTV) is an important parameter in glioblastoma (GBM) treatment. MR imaging within 48hrs after surgery is the preferred imaging modality but is prone to variability because of different interpretation by different raters. Variability can be statistically assessed using agreement and reliability parameters. The former are underreported in medical literature and are often mixed up with the latter. But agreement parameters are to be preferred in assessment of the level of agreement between repeated measurements. A typical agreement parameter is the standard error of measurement (SEM), which is expressed in the same unit as the original variable. Reliability parameters on the other hand relate the measurement error to the variability between study objects. A typical and well-known reliability parameter is the intraclass coefficient (ICC), expressed on a scale from 0 to 1. We investigated both SEM and ICC between and within raters concerning RTV in GBM. MATERIAL AND METHODS Ten cases were randomly selected from a glioblastoma database. Three observers (a neuroradiologist, NR; a neurosurgeon, NS; a junior neurosurgical trainee, JNT) performed two sessions of volumetric assessment of RTV, blinded for each others results and independently from each other. At least 2 months passed in between sessions. Volumetric assessment was performed using Magnetisation Prepared-Rapid Echo Gradient (MPRAGE) 0.9mm images in standard neuronavigation software (Medtronic Inc, Louisville, CO, USA), in comparison with other images when deemed necessary by the individual rater. Semi-automated segmentation was applied, meaning that the rater used automated segmentation technique on each slice, but adjusted according to his of her own judgement (segmentational growing around the region of interest). Volumes were expressed in mL. Statistical analysis was performed using R software (version 3.4.3 (30-11-2017); Attached packages: psych foreign pander date knitr) to calculate ICCs. SEMs were manually calculated using the formulae by De Vet et al. RESULTS Intrarater ICC agreement was 0.96 (0.87–0.99), 1 (0.98–1) and 0.77 (0.34–0.94) for NS, NR and JNT resp. SEM agreement was 0.87mL, 0.25mL and 2.67mL resp. Interrater variability was as follows: NS-NR 0.89 (0.65–0.97) ICC and 1.42mL SEM; NS-JNT 0.93 (0.77–0.98) ICC and 1.52mL SEM; NR-JNT 0.74 (0.28–0.93) ICC and 2.81mL SEM. CONCLUSION Agreement on postoperative residual contrast-enchancing tumor volume in GBM was high between neurosurgeon and neuroradiologist, with SEM 1.42mL. SEM probably reflects the differences in interpretation, especially of areas with blood products on postoperative MR imaging. Both intrarater and interrater agreement was highest for more experienced raters.