Prediction of outcomes in MCI with (123)I-IMP-CBF SPECT: a multicenter prospective cohort study

The multicenter prospective cohort study (Japan Cooperative SPECT Study on Assessment of Mild Impairment of Cognitive Function: J-COSMIC) aimed to examine the value of (123)I-N-isopropyl-4-iodoamphetamine cerebral blood flow (IMP-CBF) SPECT in regards to early diagnosis of Alzheimer's disease (AD) in patients with mild cognitive impairment (MCI).

Three hundred and nineteen patients with amnestic MCI at 41 participating institutions each underwent clinical and neuropsychological examinations and (123)I-IMP-CBF SPECT at baseline. Subjects were followed up periodically for 3 years, and progression to dementia was evaluated. SPECT images were classified as AD/DLB (dementia with Lewy bodies) pattern and non-AD/DLB pattern by central image interpretation and automated region of interest (ROI) analysis, respectively. Logistic regression analyses were used to assess whether baseline (123)I-IMP-CBF SPECT was predictive of longitudinal clinical outcome.

Ninety-nine of 216 amnestic MCI patients (excluding 3 cases with epilepsy (n = 2) or hydrocephalus (n = 1) and 100 cases with incomplete follow-up) converted to AD within the observation period. Central image interpretation and automated ROI analysis predicted conversion to AD with 56 and 58 % overall diagnostic accuracy (sensitivity, 76 and 81 %; specificity, 39 and 37 %), respectively. Multivariate logistic regression analysis identified SPECT as a predictor, which distinguished AD converters from non-converters. The odds ratio for a positive SPECT to predict conversion to AD with automated ROI analysis was 2.5 and combining SPECT data with gender and mini-mental state examination (MMSE) further improved classification (joint odds ratio 20.08).

(123)I-IMP-CBF SPECT with both automated ROI analysis and central image interpretation was sensitive but relatively nonspecific for prediction of clinical outcome during the 3-year follow-up in individual amnestic MCI patients. A combination of statistically significant predictors, both SPECT with automated ROI analysis and neuropsychological evaluation, may increase predictive utility.