Multicenter development and validation of a probability-based model to diagnose Lewy body disease using ¹²³I-meta-iodobenzylguanidine

Purpose: We previously proposed that a probability-based sympathetic 123I-meta-iodobenzylguanidine (mIBG) index (SMILe) could distinguish the presence or absence of Lewy body disease (LBD) based on findings at a single center. However, whether the model would be useful in the real world remained uncertain. Therefore, we updated and evaluated its performance at five Japanese and three European institutions. Methods: We compared data from 967 patients with suspected LBD with 62 controls from a normal database (NDB). Of 815 patients with guideline-based diagnoses, 483 had LBD (Parkinson disease [PD] or dementia with Lewy bodies [DLB]) and 332 did not have LBD. Heart-to-mediastinum (H/M) ratios were standardized using a phantom-based method. Logistic regression models included early and late H/M ratios, age, sex, and comorbidities. We assessed diagnostic performance using ROC analysis and cross-validation. Results: The updated model discriminated LBD from other diseases (AUC for early and late H/M, 0.880 0.899, respectively). Age correction of H/M ratios based on the NDB did not improve accuracy. Median early H/M ratios [SMILe probability] were 3.09 [12.8%] for NDB, 2.57 [37.5%] for Alzheimer disease, 1.76 [84.7%] for PD, and 1.62 [89.0%] for DLB, with significantly lower H/M ratios and higher probabilities in PD and DLB compared with controls (p < 0.0001). Late-phase imaging added value mainly in intermediate borderline (30%–70%) situations. Coronary artery disease attenuated the diagnostic performance of SMILe. Conclusion: The probability-based 123I-mIBG model reliably differentiated LBD from other diseases. Standardization among sites supports global applicability and reflects real-world clinical practice.