Reply: The impact of dementia prevalence on the utility of the AD8

Sir, Dr Christensen brings up an important point when considering the utility of dementia screening tests, namely the prevalence of disease. Psychometric properties of screening tests should also be closely examined when deciding how one can best utilize a test and interpret its meaningfulness. The most common properties, namely sensitivity/specificity and predictive values differ greatly in this regard. Sensitivity and specificity refer to the likelihood of a positive or negative test result when the disease is present or absent. Because these properties are inherent to the test, they do not vary with disease prevalence. Predictive values, on the other hand, refer to the likelihood of disease based on the test result, and are therefore directly related to prevalence of the disease within the study population.

Dr Christensen correctly points out that the world-wide prevalence of dementia ranges from 2.4% to 19.5% in a meta-analysis of Asia, Europe and the Americas, while the prevalence rates for the studies we have published on the AD8 are significantly higher, being drawn from a longitudinal study and a memory disorders clinic. While the difference in prevalence rates would not impact the sensitivity and specificity of the AD8 with both properties averaging 85% in the multiple studies, the positive predictive value of the AD8 would be significantly lower when applied to a broad population. However, four important points are lost in the discussion.

First, population-based studies often underestimate the frequency of Alzheimer’s dementia because their diagnostic methods tend not to capture individuals with milder stages of dementia (Krysio et al., 2004). This may be due to differences in the diagnostic criteria applied to the sample, at which level the diagnostic instrument is operating (mild versus moderate cases of disease), and ascertainment of information (review of records versus in-person interview).

Second, dementia screening as a secondary prevention measure is proposed as an early detection of symptoms in individuals in whom the disease process has already begun and who may be experiencing very mild clinical symptoms not yet detected in the usual course of medical care (Galvin, 2011). Broad dementia screening in the general population makes little sense. The prevalence rate of Alzheimer’s disease and related disorders in individuals is too low to yield sufficient results. However, since the risk of dementia increases substantially with age, approaching 42% in individuals over the age of 85, targeted screening of at-risk individuals makes perfect sense. Dr Christensen points this out in her letter.

Third, the inherent challenges of broad dementia screening of a population with a low prevalence would be an obstacle with all dementia screening tools. Without a targeted approach, every instrument—both informant- and performance based—would suffer from decrements in the positive predictive value. Dr Chistensen cites an excellent paper by Holsinger et al. (2007): ‘AD8 (Galvin et al., 2005) is a promising instrument but has only been evaluated in one study’. At the time of the Holsinger et al. (2007) paper, that was true. However, we have since published four additional papers on independent samples showing similar psychometric properties (Galvin et al., 2006, 2007a, b, 2010) and in the last paper, demonstrating the biological validity of the AD8. Malmstrom et al. (2009) used the AD8 in a population-based sample of African-American older adults drawn and reported good sensitivity and specificity with an area under the curve of 84.7% discriminating cognitively normal older adults from those with very mild dementia. Additionally, international studies in South Korea (Ryu et al., 2009) and Taiwan (Yang et al., 2011) report identical properties of the AD8 with ongoing studies in Brazil, Argentina, Ecuador, Spain, Norway, Canada and the Philippines.

Last, predictive values may not be the most effective method of evaluating screening tests. The likelihood ratios (McGee, 2002) of any screening test is the probability that a positive test is found in persons with disease divided by the probability of the same finding in persons without disease (McGee, 2002). Likelihood ratios range from 0 to infinity with larger numbers providing more convincing evidence of disease; smaller numbers argue the disease is less likely. When the positive likelihood ratio is >5 or the negative likelihood ratio is <0.2, the pre-test probability of a patient having the disease tested can be used to estimate a post-test probability of the existing disease state. This same standard is applied by Holsinger et al. (2007). In the case of the AD8, we report the likelihood ratio of a positive test is 5.8 (95% confidence interval: 5.4–6.3), which is similar to the ratio of 5.6 reported by Holsinger et al. (2007).

As we state in the article, dementia screening requires a consideration of the population at risk and the sensitivity and specificity of the instruments used (Holsinger et al., 2007). A large number of false positive individuals might expend limited health-care dollars; a large number of false negative individuals would be denied treatment and miss opportunities to participate in clinical research. Thus, a staged dementia screening approach would make the most sense clinically and economically. The value of the AD8 is that it is not only brief but corresponds to more detailed evaluations, neuropsychological testing and Alzheimer’s disease biomarkers (Galvin, 2010), an argument that many other screening tests have not yet made. A more important point we make is that informant-based methods, whether the brief AD8 or the more detailed Clinical Dementia Rating, are at least as good as any objective measures. Thus, if simple screening is the goal, because the AD8 is comparable to the Clinical Dementia Rating both in direct comparison and in its relationship to biomarkers, it could be recommended for this simple goal on the basis of utility and brevity.

Funding

This study was supported by grants (P01 AG03991, P01 AG026276 and P50 AG05681) from the National Institute on Aging, National Institutes of Health.