The present report describes the utility of an entirely automated, very brief assessment of common dementia risk factors and memory performance in a sample of >3,000 older adults, and provides preliminary results from a validation study of persons seeking evaluation for possible dementia. We found that age, sex, education, complaints of severe memory problems, and reported histories of cerebrovascular disease (hypertension, stroke and transient ischemic attack), Parkinson's disease (PD), and brain tumor all differentiated normal from low memory performance. Although our predictive model accounts for only 14.5% of the total variance in memory score, this is considered a medium effect and is comparable to that obtained with in-person administration of traditional cognitive tests 
There have been several previous attempts to develop algorithms for the prediction of dementia risk. Kivipelto and colleagues 
tested regression models based on data collected in mid-life for dementia risk 20 years later. They found that age, education, apolipoprotein E (ApoE) status, systolic blood pressure, body mass index, total cholesterol, and level of physical activity could be combined to produce a total score that predicts up to a 16-fold increased risk of dementia. Barnes and colleagues 
developed a late-life dementia risk index that stratifies older adults into low, moderate, or high risk of developing dementia within 6 years. Age, cognitive test performance, body mass index, ApoE status, findings of white matter disease or ventricular enlargement on brain MRI scans, carotid artery thickening on ultrasound, history of bypass surgery, slowed physical performance, and total abstinence from alcohol all contributed to the index score. Four percent of older adults with low scores on this index developed dementia within 6 years, compared with 23% of those with moderate scores, and 56% of those with high scores. Note that both of these models require DNA analysis and other biomedical assessments.
The results of our clinical validation study suggest that the DRA holds promise in the identification of cognitive disorders diagnosed by the ‘gold standard’: formal evaluation by a geriatric neuropsychiatrist, including supporting laboratory and imaging studies. DRA recognition memory test performance was highly correlated with both in-person cognitive performance (MMSE) and caregiver ratings of cognitive decline (IQCODE). The criterion validity of our memory test is established by its clear differentiation of persons seeking clinical evaluations who are found to meet criteria for AD or another dementia syndrome from those who do not (including persons with MCI and neuropsychiatric conditions). In our small dementia clinic sample, where the base-rate of cognitive disorder was high, a ‘positive’ test (i.e., scoring below .25) had 68% predictive validity for a dementia diagnosis.
The entirely automated and highly accessible format of the Dementia Risk Assessment may make it useful in large-scale screening programs, as might be required to identify at-risk elderly for inclusion in prevention trials. It also provides a way for those who are unable or unwilling to visit a dementia specialist to learn about their risk factors, and encourages those at high risk to seek clinical evaluation. The DRA may also provide reassurance to those concerned about developing dementia but whose empirical risk is low. It should be emphasized, however, that the DRA does not attempt to diagnose dementia, and this is clearly stated by the program.
Several limitations of the DRA instrument and of this study must be acknowledged. First, in an effort to keep the DRA brief (and thereby encourage its completion), several important risk-factors were excluded. Future versions might include additional health and lifestyle variables, such as body mass index, alcohol and tobacco use, physical exercise, and mentally stimulating activity. Second, although our Anonymous Internet Sample was open to persons anywhere in the world who are competent in English and who had access to the Internet, the representativeness of our two research samples and their comparability to future users is unknown. Third, there are limitations inherent in the online administration. Although Internet-based assessments are becoming increasingly prevalent and results demonstrate good validity 
, the conditions of administration are not controlled and there is no way to ensure legitimate results. It is certainly feasible that some participants did not tell the truth on risk factor questions or violated instructions to obtain higher scores on the memory test. Alternatively, poor performances may have been due to environmental distracters or waning motivation, rather than genuine memory impairment. Other individuals may have completed the DRA on more than one occasion. Given that participants are self-selected, it may be argued that the vast majority of respondents who are sufficiently motivated to engage in online dementia screening would complete the instrument honestly.
Probably the most useful data on dementia risk are those provided by self- and proxy-reports about the same person, and comparing self-report ratings with informant reports will be important for future studies. The number of such pairs in our existing Internet sample is extremely small, probably due in part our desire to keep the data anonymous. Overcoming this limitation will be challenging, but the continued evolution of technology may allow us to address this issue in future versions of the DRA. Finally, the size in our Clinical Validation Sample was small and did not allow us to test robust predictive models. A larger-scale clinical validation study is currently underway.