Purpose: To evaluate the normality of internal quality control (IQC) data used for measurement uncertainty (MU) estimation in a nuclear medicine laboratory and to implement a structured framework for adaptive and conservative MU model selection. Metho...
Purpose: To evaluate the normality of internal quality control (IQC) data used for measurement uncertainty (MU) estimation in a nuclear medicine laboratory and to implement a structured framework for adaptive and conservative MU model selection. Methods: IQC datasets from 45 assays performed between May and July 2025 at Asan Medical Center were segmented according to reagent kit lot continuity, yielding 129 pooled datasets. Normality was assessed using the Shapiro–Wilk test and visualized with Q–Q plots. Segments were classified as normal (p ≥ 0.05), mildly non-normal (0.01 ≤ p < 0.05), non-normal (p < 0.01), or non-testable (n < 15). Measurement uncertainty was calculated using four models: CV-based, Algorithm A, robust (MAD/IQR), and rectangular (Type B), selected by an integrated decision algorithm incorporating sample size, normality, and Δ-based stability indices. Results: Among 129 pooled datasets, 123 (95.3%) had n ≥ 15 and underwent normality testing; 59 (48.0%) were normal, 12 (9.8%) mildly non-normal, and 52 (42.3%) non-normal. Dataset size was the only significant predictor of normality (p = 0.003). Of the 127 analyzable segments (n ≥ 2), 100 (78.7%) required non-Gaussian or conservative MU models. MU from the proposed framework strongly correlated with CV-based MU (Spearman r = 0.97, p < 0.001) without systematic bias (Wilcoxon p = 0.503). Relative differences in MU was near zero across normality groups, but model-specific differed: Algorithm A showed minimal deviation (−0.03%), robust MAD and IQR produced higher MU (+22.3% and +12.0%), and the rectangular model showed wide variability. Overall, 62% of datasets yielded higher MU under the proposed framework. Conclusion: More than half of IQC datasets with sufficient sample size deviated from Gaussian assumptions, indicating that routine CV-based MU estimation may be inappropriate for many assays. The proposed framework maintains consistency with conventional estimation under normal conditions while adaptively providing more conservative uncertainty estimates when distributional assumptions are violated. These results support its applicability as a reproducible approach consistent with current international guidelines for MU evaluation in nuclear medicine laboratories.