Dementia, a syndrome characterized by a progressive cognitive decline that interferes with independent functioning, affects a large and growing number of older adults. With increases in life expectancy in the population, the individual, family, and societal burdens associated with dementia are expected to accelerate rapidly, making the disease one of the major public health challenges of this century [1]. It is increasingly recognized that the identification of preclinical markers that are predictive of the future transition from healthy cognition to dementia is of paramount importance [2,3]. Such markers could allow earlier intervention and even modest delays in disease onset or progression, which would have a substantial impact on incidence rates and health care costs.

Despite existing efforts to discover biomarkers of cognitive decline and dementia [4,5], there is a need to develop and investigate nonbiological, behavioral markers that can be turned to when using biomarkers is impractical [6]. Accumulating research has shown that subtle alterations in daily functioning, for example, telephone or computer use, managing finances, and completing forms, are among the earliest and strongest signals that predict cognitive decline and dementia [4,5,7-9]. Reductions in the efficiency, speed, and consistency of performing these and other cognitively demanding activities can be observed up to a decade before diagnosis [3,7]. Accordingly, recommendations by the US National Institute on Aging and Alzheimer’s Association highlight the need to identify and validate new neurobehavioral measures, including assessments of very early deficits in functional performance, that could detect preclinical manifestations of the disease [2,3].

We propose to develop indicators of functioning that can be gleaned from the way older people complete survey assessments, that is, questionnaires and computer-administered assessments. Surveys are ubiquitous in research and clinical practice, and many large-scale national and international longitudinal panel surveys on aging are presently in the field. A survey is a small slice of everyday functioning; nevertheless, completing a survey is a complex and cognitively demanding task that requires attention, working memory, executive functioning, and short- and long-term memory. Research on the psychology of survey methods confirms that answering survey questions requires considerable cognitive effort: respondents are expected to interpret the meaning of each question, retrieve relevant information from memory, integrate it into a summary judgment, and map their judgment onto the provided response alternatives [10]. As a result, examining older people’s survey response behaviors, which focus on how respondents complete surveys irrespective of the content being sought by the questions, may represent a valuable but often neglected resource that can be leveraged to develop behavior-based early markers of cognitive decline and dementia that are cost-effective, unobtrusive, and scalable for use in large population samples.

The purpose of this paper is to describe the study protocol of the “Testing Early Markers of Cognitive Decline and Dementia Derived from Survey Response Behaviors” study. The study has the following 3 specific aims:

As mentioned in the previous section, we will have identified 15 longitudinal ageing studies as eligible data sources for creating questionnaire answer pattern indices by October 2022 (Table 1). We have identified a total of 20 questionnaire answer pattern indices (Table 2) and 20 paradata indices (Table 3). We have conducted a preliminary investigation to test the utility of the questionnaire answer pattern indices for the prediction of cognitive decline and dementia. The tested indices included the proportion of skipped items [18], the multivariate outlier rate [22], the normed Guttman errors [26], the inter-item standard deviation [28], and the person-total correlation [23] derived from the 2006/2008 waves in the Health and Retirement Study (N=12,554). Survival analyses (using the Fine and Gray competing-risk model [38]) were used for each index to examine whether it predicted dementia onset in the subsequent 8 years, while statistically controlling for age and education and while taking death into account as a competing event. Dementia status was ascertained using the criteria by Langa and Weir, which categorize respondents as cognitively normal, cognitively impaired (not dementia), or having dementia. The classification is based on cognitive test scores with cutoffs calibrated against in-person clinical assessments in the Aging, Demographics, and Memory Study [13]. It is not a clinical diagnosis but has proven useful for research purposes. As shown in Table 4, the hazard rates of dementia onset are between 1.52 and 2.25 times as high in respondents with questionnaire answer patterns that suggested cognitive problems (highest tertile) as for those in the lowest tertile (P<.001 for each index), and the middle tertile fell between the lowest and the highest. Details on the results of these analyses have been published elsewhere [39].

To assess the predictive power of the para-data indices, we derived participants’ response times to survey items in 15 UAS surveys that were fielded from mid-2014 to 2015. We examined whether they were predictive of subsequent cognitive functioning measured in 2016 and of cognitive function change from 2016 to 2018. The response times to UAS survey items were log-transformed to reduce skewness. Cognitive functioning was evaluated by using the verbal analogies test, an executive functioning and analogical reasoning task with good specificity and sensitivity in discriminating normal controls from probable dementia patients. The analysis included 3829 UAS panelists 50 years and older and examined both zero-order associations and associations controlling for age and education. As shown in Figure 2, longer response times on 14 of the 15 surveys were significantly correlated at the 0.05 level with lower subsequent cognitive test scores in 2016 as well as with worsening cognitive functioning from 2016 to 2018. The significance and direction of these associations did not change after controlling for age and education.

Notably, these early results are based on only a subset of questionnaire response patterns and para-data indices but are suggestive of the findings that we anticipate will emerge from the planned analyses of multiple behavioral indices derived from many diverse studies.

This paper describes the protocol of a multiyear research project that aims to develop early markers of cognitive decline and dementia from survey response behaviors. The project plans to develop 2 types of indices, known as questionnaire answer pattern indices and paradata indices, summarizing different aspects of older adults’ survey response behaviors. The project will examine the psychometric properties of the created indices and create predictive models of cognitive decline and dementia using the indices. Preliminary results are suggestive of the predictive power of survey response behavior indices for forecasting cognitive decline and dementia. To the best of our knowledge, this study is the first multiyear funded project aspiring to develop nonbiological early markers of cognitive decline and dementia from large-scale survey response behavior data at older ages.

If successful, the study may open novel opportunities for advancing research on health care for cognitive decline and dementia. Indices derived from survey response behaviors that could serve as preclinical markers of cognitive decline could greatly enhance the usefulness of questionnaires collected in large-scale surveys for epidemiological research on the risk and protective factors of dementia. Questionnaires are also routinely collected in medical care settings, and it is possible that survey response behavior indices extracted in these settings could potentially supplement information from standardized cognitive tests. In clinical trial research, the Food and Drug Administration recommendations for early-stage Alzheimer disease trials encourage the development of novel approaches for the evaluation of preclinical functional deficits [40], and response behavior indices extracted from questionnaires administered in clinical trials might potentially supplement core clinical measures that serve as trial endpoints. Furthermore, the survey response behavior indices may be useful as a proxy for cognitive impairment in aging survey research where standardized cognitive functioning measures are not assessed.

Survey response behaviors are a relatively inexpensive data source, but they are seldom used directly for epidemiological research on cognitive impairment at older ages. This innovative yet unconventional approach may complement existing approaches aimed at the early detection of cognitive decline and dementia.