Supplementary Material for: Prognostic Accuracy of Mild Cognitive Impairment Subtypes at Different Cut-Off Levels M.Göthlin M.Eckerström S.Rolstad A.Wallin A.Nordlund 2017 <p><b><i>Background/Aims:</i></b> The prognostic accuracy of mild cognitive impairment (MCI) in clinical settings is debated, variable across criteria, cut-offs, subtypes, and follow-up time. We aimed to estimate the prognostic accuracy of MCI and the MCI subtypes for dementia using three different cut-off levels. <b><i>Methods:</i></b> Memory clinic patients were followed for 2 (<i>n</i> = 317, age 63.7 ± 7.8) and 4-6 (<i>n</i> = 168, age 62.6 ± 7.4) years. We used 2.0, 1.5, and 1.0 standard deviations (SD) below the mean of normal controls (<i>n</i> = 120, age 64.1 ± 6.6) to categorize MCI and the MCI subtypes. Prognostic accuracy for dementia syndrome at follow-up was estimated. <b><i>Results:</i></b> Amnestic multi-domain MCI (aMCI-md) significantly predicted dementia under all conditions, most markedly when speed/attention, language, or executive function was impaired alongside memory. For aMCI-md, sensitivity increased and specificity decreased when the cut-off was lowered from 2.0 to 1.5 and 1.0 SD. Non-subtyped MCI had a high sensitivity and a low specificity. <b><i>Conclusion:</i></b> Our results suggest that aMCI-md is the only viable subtype for predicting dementia for both follow-up times. Lowering the cut-off decreases the positive predictive value and increases the negative predictive value of aMCI-md. The results are important for understanding the clinical prognostic utility of MCI, and MCI as a non-progressive disorder.</p>