Background: Criteria for mild cognitive impairment (MCI) predict dementia risk in the clinic. Dementia risk in the population is different and whether there is an optimal MCI-derived threshold for discriminating at-risk from not-at-risk cases in the general older population is not known.
Methods: Data were from the Medical Research Council Cognitive Function and Ageing Study. Two risk thresholds were derived from each of seven different concepts of MCI including: Mayo Clinic-defined amnestic, nonamnestic, multiple, and revised MCI, MCI based on Mini-Mental State Examination derived categories, and the definitions of Cognitive Impairment No Dementia and Age-Related Cognitive Decline (ARCD). Receiver operating characteristic analysis was used to compare the predictive validity of 2-year incident dementia for each risk threshold across the different MCI definitions.
Findings: MCI-derived risk thresholds varied in their ability to predict dementia. MCI thresholds were accurate in identifying individuals not at-risk of dementia progression (false-negative range: 0%–3.4%). No MCI-derived threshold accurately identified an at-risk group with a 2-year progression rate greater than 20%. Criteria for ARCD defined the threshold with the highest sensitivity and specificity for dementia conversion.
Interpretation: MCI-derived thresholds do not reliably identify individuals at-risk of incident dementia at 2 years when applied in the general population. A large subpopulation of individuals not at-risk was more reliably identified. What is considered a sufficient level of accuracy for identification of individuals at increased risk of dementia depends on the motivation for screening and on the safety and efficacy of available interventions.