Interobserver variability of pediatric AVM grading: a disagreement index for training and calibration
摘要
To quantify interobserver variability in Spetzler–Martin (SM) and Spetzler–Ponce (SP) grading of pediatric arteriovenous malformations (AVMs) and to develop a Disagreement Index (DI) capturing case-level grading instability and rater-dependent reclassification.
MethodsIn this single-centre retrospective study, 45 consecutive pediatric AVMs were graded on pretreatment imaging by three raters. SM components/grade, SP class, and compact versus diffuse nidus morphology were recorded. Agreement was assessed using intraclass correlation coefficients (ICC) and κ statistics. A case-level DI (0–1) aggregated between-rater spread in SP class, dispersion of SM scores, and mismatches in SM components.
ResultsSM score agreement was moderate (ICC = 0.72), whereas categorical agreement on SM grades was poor (Fleiss’ κ = 0.04). Collapsing into SP classes improved agreement (κ = 0.49), yet 17/45 AVMs (38%) crossed at least one SP boundary across raters. Eloquence (κ = 0.29) and nidus morphology (κ=−0.15) were least reproducible; nidus size and deep venous drainage showed higher agreement (κ = 0.41 and 0.58). DI ranged from 0.15 to 1.00 (median 0.46); high-DI cases were driven mainly by discordant eloquence and threshold crossings (SM II/III; SP A/B).
ConclusionsIn pediatric AVMs, grading variability clusters around subjective descriptors and clinically meaningful thresholds, allowing the same lesion to be assigned to different risk strata. DI may support trainee feedback, data-quality checks, and prioritisation of cases for multidisciplinary review.