Separation of cognitive domains to improve prediction of progression from mild cognitive impairment to Alzheimer's disease

Addressing causes of heterogeneity in cognitive outcomes is becoming more critical as Alzheimer's disease (AD) research focuses on earlier disease. One of the causes of this heterogeneity may be that individuals with deficiencies in different cognitive domains may perform similarly on a neuropsychological (NP) test for very different reasons. Tatsuoka and colleagues have applied a Bayesian model in order to integrate knowledge about cognitive domains relevant to each NP test with the observed outcomes from the Alzheimer's Disease Neuroimaging Initiative (ADNI) mild cognitive impairment data. This approach resulted in better prediction of AD diagnosis than more traditional approaches.


Introduction
Alzheimer's disease (AD) research is beginning to focus on earlier stages of disease, prior to formal diagnosis of AD dementia. Th is focus on early symptomatic disease and populations with mild cognitive impairment (MCI) introduces issues of heterogeneity since patients at these earlier stages have not yet met the more standardizing criteria of an AD diagnosis. Understanding this heterogeneity is the fi rst step in eff ectively planning studies for these early populations.

Review
One of the barriers to understanding the heterogeneity in these early populations is the diffi culty in isolating cognitive domains for measurement. Most neuropsychological (NP) tests require a number of cognitive processes and, therefore, measures of multiple domains. In the previous issue of Alzheimer's Research & Th erapy, Tatsuoka and colleagues [1] get around this problem by identifying the multiplicity of cognitive domains measured by each NP test and also the level of ability in that domain that is required for maximum performance on the test. Th ey use this information in a Bayesian statistical model that is designed to separate out the eff ects of diff erent cognitive functions on the NP outcomes. Based on NP data collected in MCI subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) study, the approach yielded several potential combinations of cutoff s for baseline NP test scores and apolipoprotein E (APOE) status values which predict imminent risk of progression to AD over the course of a 24-month period.
Importantly, the authors also illustrate that more commonly applied methods, such as factor analysis and logistic regression, are not able to achieve the same level of success in predicting a diagnosis of AD, presumably because the specifi c domains are not able to be separated with those models. Separation of specifi c domains also allows grouping of the specifi c combinations of defi cits into diagnostic subgroups (for example, amnestic singledomain MCI, amnestic multidomain MCI, and nonamnestic multidomain MCI) which then can be analyzed in terms of biomarkers and disease outcomes.

Conclusions
Th e ultimate validation of the method awaits replication in larger, more diverse MCI populations, as the authors acknowledge, although the method appears promising. Many diff erent statistical approaches are currently being applied to clinical outcomes in order to identify patients who will progress to AD or patients who will have diff erent decline rates over time. Th e results from these analyses can be used to help us identify a particular population and then a sensitive clinical outcome for study in a clinical trial that will result in improved power for detecting a diff erence. Th ese results may also be Abstract Addressing causes of heterogeneity in cognitive outcomes is becoming more critical as Alzheimer's disease (AD) research focuses on earlier disease. One of the causes of this heterogeneity may be that individuals with defi ciencies in diff erent cognitive domains may perform similarly on a neuropsychological (NP) test for very diff erent reasons. Tatsuoka and colleagues have applied a Bayesian model in order to integrate knowledge about cognitive domains relevant to each NP test with the observed outcomes from the Alzheimer's Disease Neuroimaging Initiative (ADNI) mild cognitive impairment data. This approach resulted in better prediction of AD diagnosis than more traditional approaches.
helpful for identifying important stratifi cation variables to include in a statistical model for analyzing the data from a clinical trial. Ultimately, these diff erent statistical approaches will be evaluated by comparing their success in predicting conversion or maximizing decline rates in new populations.