How Technology is Reshaping Cognitive Assessment: Lessons from the Framingham Heart Study

Abstract

Objective

This article elucidates how the Boston Process Approach (BPA) can amplify the role of neuropsychology in the study of preclinical and clinical dementia, particularly Alzheimer’s disease (AD), and how advancements in technology expand BPA capacity objectively and exponentially.

Method

The BPA is based on a conceptualization of cognition as being comprised of multiple processes, the nature of which could not possibly be captured by a single score on a test or battery of tests. Identification of these processes is only possible with careful observation of an individual during the entire testing process to determine how, when, and why a person fails, which helps to reveal the integrity of the cognitive processes underlying the behavior.

Results

BPA use within the Framingham Heart Study is described, including how digital technology has been incorporated to enhance the sensitivity of BPA to detect insidious onset changes even earlier than had been previously possible. The digital technology movement will dramatically alter the means by which cognitive function is assessed going forward.

Conclusions

Technological advances will catalyze groundbreaking discoveries for effective treatments of neurodegenerative cognitive disorders, such as AD, and inform novel strategies for dementia prevention and sustained lifelong cognitive health.

When neuropsychology began to take root, the zeitgeist of investigations of the brain by clinical psychologists in the United States¹ was dominated by Ward Halstead’s belief that mental activity was a holistic, mass-action event, and the primary goal of assessment was to detect whether there was brain damage; that is, whether atypical behavior was due to “organicity” (Milberg & Hebben, 2006). Although Halstead believed that “biological intelligence” (Halstead, Carmichael, & Bucy, 1946; Halstead, 1951) was comprised of four factors (i.e., abstraction, power, dimension, and memory), he asserted that “collectively they constitute a neural Gestalt which can be selectively impaired or enhanced by certain classes of stress such as anoxia, drugs, hypnotic inhibition, trauma, and disease” (Halstead, 1951). The “Gestalt” he sought to measure would become the Halstead Impairment Index, a single quantitative score that reflected the presence (or absence) of organic brain damage.

Halstead and his student, Ralph Reitan, believed strongly that standardized, replicable, norm-based tests were required for understanding brain-behavior relationships, and that assessment required a “fixed” battery of tests (Reitan, 1994). Reitan actually reported that both he and Halstead believed that observing a patient provided potentially important information, but that “if neuropsychology were to be established as a science rather than an art, it was important to define the standardized testing procedures that would produce quantitative scores separate from the personal and qualitative observations during testing.” In fact, Reitan reported that he could predict whether a lesion was present, what type of lesion it was, and whether it was focal or diffuse by looking at only the test scores (i.e., “blind” to all patient information). Halstead and Reitan were criticized for being too focused on quantitative measurement rather than trying to understand the functional organization of the brain and how it relates to behavior (Luria & Majovski, 1977).

In contrast to Halstead’s emphasis on characterizing the “Gestalt” of mental activity, the Boston Process Approach (BPA; also referred to as the Process Approach) is based on a conceptualization of cognition as being comprised of multiple processes, the nature of which could not possibly be captured by a single score on a test or battery of tests. Identification of these processes is only possible with careful observation of an individual during the entire problem-solving activity to determine how, when, and why a person fails, which helps to reveal the integrity of the various processes involved in the behavior. In order to sort out which processes are affected, the examiner tailors the assessment to the individual and may need to think of novel ways to create conditions that allow for the differentiation of processes. Thus, whereas Halstead and Reitan believed diagnosis could be done “blind” to anything but the quantitative scores, and they believed in a fixed battery of standardized, normed tests, the BPA practitioner is compelled to observe every moment of a patient’s behavior, to target assessment to the specific problems of the patient (i.e., a “flexible” battery), and to allow for alteration of test administration procedures or even create novel tests in the moment.

While the conceptualization of BPA was inspired by the works of many previous (primarily from outside the United States) and contemporary researchers, it was Edith Kaplan, Ph.D. who was “unequivocally regarded as the pioneer and chief architect” (Delis, 2010). Kaplan believed that mental activity was comprised of component processes, and these processes were tied to different brain structures. In order to understand a patient’s brain, and therefore the presenting problems, these component parts had to be identified, and this could only be achieved by careful observation, creative experimentation, and analysis of an individual’s behavior. She noted, “Final global scores are not nearly as informing (and, in some instances, may be misleading) as observing, noting, identifying, and quantifying the behaviors of patients while they are working toward a solution” (Kaplan, 2002).

Much of Kaplan’s early work was based on her detailed observations and experimentation with “case studies,” accompanied by discussions with colleagues. In a field dominated by empiricism, however, the lack of “empirical evidence” supporting BPA drew sharp criticism. Even into the 1980s, the dominant consensus among neuropsychologists was that “mechanical data collection, algorithmic combination, and statistical prediction [were] superior to the clinical collection, synthesis, and diagnostic method” (Poreh, 2006). Although Kaplan and others did start publishing results of studies with more traditional designs, it is really the past 25 years that have witnessed a growing body of both national and international research examining Kaplan’s ideas, standardizing techniques and interpretations to better understand brain-behavior relations, and applying that knowledge for use in research and clinical settings. Ritchie, Odland, Ritchie, and Mittenberg (2012) reported that when they asked neuropsychologists from 75 internship programs about the importance of different types of assessment experiences, 67% identified BPA as “very important” or “essential.”

Kaplan’s work with Heinz Werner, Ph.D. led to the development of the core tenet of the BPA - that cognition and behavior were the outcome of cognitive processes with individual component parts that are associated with neuroanatomical underpinnings. Werner, perhaps the most influential developmental psychologist of the 20^th century, observed that the development of children’s cognition was characterized not by an increase in the quantity of problem-solving capacity, but by a qualitative change in the means by which problems were solved, suggesting that cognition was multifactorial (Werner, 1937). In a seminal paper on education (Werner, 1937), Werner expressed concern about the use of achievement scores to draw conclusions about an individual’s cognitive abilities. As Kaplan (2002) described, Werner’s argument was that the use of a single summary score assumes that cognition is a unitary mechanism, and disregards the reality that the same solution can be derived from different processes, and these processes may be associated with different brain structures. Werner believed that close and careful observation of the behavior an individual displayed during the course of problem solving would provide greater insight into cognitive functioning than looking at a final achievement score. Kaplan would apply these ideas to her work with patients with acquired brain damage. She observed that two people could earn the same score by way of very different strategies as characterized by their errors. Kaplan observed that “the patient who misplaces one of nine blocks on a difficult Block Design item receives the same “zero” score as the patient who throws or eats the blocks” (Bauer & Bowers, 2013). Kaplan’s assertion that the how is as important as how much was revolutionary in American clinical neuropsychology.

The Boston Veteran’s Administration Medical Center (BVAMC) in Jamaica Plain, MA was the incubator of the BPA. In the mid-1970’s, the Halstead-Reitan Neuropsychological Battery (HRB) was well established as the primary method for evaluating brain damage (Milberg & Hebben, 2006), which was an improvement in objective assessment compared to previous efforts to determine “organicity” that had relied on the Rorschach Inkblot Test (Rorschach, 1921) and the Bender-Gestalt (Bender, 1938; Milberg & Hebben, 2006). Halstead’s theory that mental activity was a holistic, mass-action event aligned with the principles of behaviorism, the dominant field of psychology at the time. With her training in developmental psychology, Kaplan did not have the biases endemic to the field of clinical psychology in the United States (i.e., behaviorism and psychoanalysis). At the VA, Kaplan tested her ideas about cognition with veterans suffering a variety of brain disorders (e.g., traumatic brain injuries, tumors, strokes). In 1958, Norman Geschwind, M.D. described as “the father of modern behavioral neurology” (Milberg & Hebben, 2006), joined the team at the Boston VA, bringing his work investigating the connections between neurological mechanisms and behavior. Kaplan, Geschwind, and Harold Goodglass, Ph.D., a clinical psychologist with expertise in aphasia, combined fields of inquiry (e.g., neuropsychology, psychology, psychiatry, speech pathology, cognitive science), which created a vibrant center of neuropsychological and neurobehavioral discovery.

Until 1971 there was no effective or safe way to “look into” the brain (i.e., there were no CT or MRI scans; Beckmann, 2006), and so little was understood about how neuroanatomy was related to cognition in humans. A seminal case described a patient who was able to write fluently and identify objects using his right hand, but could not do the same with his left hand (Geschwind & Kaplan, 1962). It was postulated that the underlying neuroanatomy was likely related to damage to the corpus callosum, which came to be described as the “human disconnection syndrome.” When the patient passed away, autopsy revealed damage to the corpus callosum (Kaplan, 2002), confirming the hypothesized brain-behavior assertions. This case set the foundation for a more widespread adoption of BPA, which then had a significant influence on the development of neuropsychology and behavioral neurology (Libon, Swenson, Ashendorf, Bauer, & Bowers, 2013).

Analysis of errors, which can provide insight into how someone arrives at an incorrect response, is a cornerstone of BPA. As an example, when scoring the Wechsler Adult Intelligence Scale (WAIS) Similarities subtest, Kaplan, Fein, Morris, & Delis (1991) described three error types attributed to executive functions: (1) descriptions of how two items are different, rather than how they are similar (e.g., if the words were “cat” and “mouse,” the response might be, “one is a pet and one is pest”); (2) responses reflecting a characteristic of one of the words, but not both (e.g., for “cat” and “mouse,” the answer could be, “they use a litter box”); and (3) answers that are “stimulus-bound,” such as describing how two things relate to each other concretely (e.g., for “cat” and “mouse,” the answer could be “a cat chases a mouse”). Giovannetti et al. (2001) examined these same error types in a study of Alzheimer’s disease (AD) and vascular dementia (VaD) and further identified semantic error types. A semantic error was one that was a trivial, vague, or concrete detail that reflected degradation of semantic knowledge and inability to identify the most distinctive or salient features of concepts (e.g., cat and dog both eat, or both have tails). In contrast to the executive-based error types which reflect a failure to stay in set (“out-of-set errors”), these semantic errors suggest that the respondent has sufficient executive functioning to maintain set (“in-set errors”), but failure occurred because of deficits in semantic knowledge (Giovannetti et al., 2013). Giovannetti et al. (2001) found that individuals with AD had a higher proportion of in-set (semantic) errors, whereas those with VaD had a higher proportion of out-of-set (executive) errors. Lamar et al. (2010) extended these findings in a study on vascular comorbidities in AD patients. Greater vascular comorbidities (i.e., hypertension, hypercholesterolemia, and diabetes) were related to a higher number of out-of-set errors and a lower number of in-set errors. This in-set and out-of-set error dichotomy has been further validated (Lamar, Swenson, Kaplan, & Libon, 2004), and is consistent with prior research examining set maintenance with other cognitive tests (e.g., Libon, Malamut, Swenson, Sands, & Cloud, 1996).

The analysis of qualitative aspects of test performance, such as types of errors, has been applied to many cognitive tests, such as the Mini-Mental State Examination (MMSE; Jefferson et al., 2002), verbal fluency (Troyer, Moscovitch, & Winocur, 1997), and the Trail Making Test (Ashendorf et al., 2008). Qualitative features such as abnormal serial position effects (Bayley et al., 2000; Kasper et al., 2016; Lafaille-Magnan, Fontaine, & Breitner, 2015), the quality of verbal expression (Mueller et al., 2016; Wetter et al., 2006; Cuentos, Arango-Lasprilla, Uribe, Valencia, & Lopera, 2007; Forbes-McKay & Venneri, 2005), intrusion errors (Libon, Bondi et al., 2011), proactive interference (Loewenstein et al., 2004; Loewenstein et al., 2015), recognition strategies (Bielak, Hultsch, Kadlec, & Strauss, 2007), and non-perseverative errors (Ryu, Lee, Song, Kim, & Lee, 2009) have all been identified as meaningful measures of cognitive impairment in clinical studies.

In addition to examining the processes and error types revealed by performance on standardized tests in general use, a number of cognitive tests have been designed specifically to tease out the component processes underlying cognitive functioning. One of the most widely used tests is the California Verbal Learning Test (CVLT; Delis, Kramer, Kaplan, & Ober, 1987), which evolved into several alternate forms (Delis, et al., 1991; Delis, Kramer, Kaplan, & Ober, 2000; Libon, Mattson et al., 1996; Delis, Kramer, Kaplan, & Ober, 1994). The multiple conditions of the CVLT allows for delineation of where verbal memory performance breaks down (e.g., encoding, storage, retrieval) and provides data reflecting different processes (e.g., learning slope, recall consistency, proactive and retroactive interference, learning strategy, perseveration, intrusions). Other commonly used neuropsychological (NP) tests that were developed around the principles of BPA include the Boston Diagnostic Aphasia Examination (BDAE; Goodglass & Kaplan, 1983; Goodglass, Kaplan, & Barresi, 2000), the Boston Naming Test (BNT/BNT-2; Kaplan, Goodglass, & Weintraub, 1983; Goodglass, Kaplan, & Weintraub, 2001), and the multitude of scoring systems for the Clock Drawing Test (CDT; e.g., Rouleau, Salmon, Butters, Kennedy, & McGuire, 1992; Freedman, et al., 1994; Libon, Malamut et al., 1996; Royall, Cordes, & Polk, 1998; Parsey & Schmitter-Edgecombe, 2011; Nyborn et al., 2013).

While the value of the BPA in clinical settings is more widely accepted, its use in research settings has been more limited because of the presumption that it is subjective, vulnerable to examiner biases, and non-replicable because of the failure to use standardized test administration and norms. The practices of flexibly choosing tests based on the patient’s presenting problem, modifying tasks to tease out what might be the underlying processes attributed to performance, and testing limits beyond standard test discontinue rules are critical features of the BPA, and these have been considered problematic given the rigor of research inquiry that requires consistency in test administration and scoring. The history of BPA use in the studies of cognitive aging and dementia in the Framingham Heart Study (FHS) suggests otherwise.

History of Cognitive Assessment in FHS

Established in Framingham, Massachusetts in 1948, FHS recruited 5,209 participants (Gen 1 cohort) for a longitudinal study designed to identify common characteristics contributing to cardiovascular disease. In 1971, the biological children of the Gen 1 cohort and their spouses (Gen 2 cohort; n=5124) were recruited for participation (Kannel, Feinleib, McNamara, Garrison, & Castelli, 1979). Most recently, in 2001, a third generation of participants (Gen 3; n=4095), the grandchildren of Gen 1 and children of Gen 2, was recruited for studies of genetic heritability of cardiovascular and cerebrovascular diseases (Splansky et al., 2007). If the spouse of a Gen 2 participant was never enrolled in FHS and if at least two biological children participated in Exam 1 of Gen 3, that spouse was invited to participate in the New Offspring Spouse Exam 1 (NOS; n=103). Pictures 1–3 depict the state of epidemiological research in the pre-analog computer era.

Picture 1.

FHS Recruitment “rolodex”

Picture 3.

Pre-Modern Day Data Analytic Methods

At the time that the Gen 1 cohort was enrolled, only people of Caucasian descent were residents of the town. In 1994, in order to reflect the growing ethnic diversity of the town and its surrounding area, FHS began recruitment of men and women between the ages of 40 and 74 whose ethnicity was Hispanic, non-Hispanic black, Asian, and Native-American, and this group comprises the Omni Generation 1 Cohort (OmniGen 1; n=506). A second generation of Omni participants, consisting of an ethnically diverse group at least 10% of the size of the Gen 3 cohort, was enrolled in 2003 (OmniGen 2). The OmniGen 2 cohort includes both individuals related to the participants of OmniGen 1 and individuals who are unrelated (n=410).

Cognitive aging and dementia research in FHS predated the publication of the National Institute of Neurological and Communicative Disorders and Stroke – Alzheimer’s Disease and Related Disorders Association (NINDS-ADRDA) diagnostic criteria for AD (McKhann et al., 1984). Between 1976–78, the Gen 1 cohort was administered a relatively short battery of cognitive tests, the description of which, along with the normative data, has been described previously (Elias, Elias, D’Agostino, Silbershatz, & Wolf, 1997). This baseline assessment, designed by Drs. Edith Kaplan, Martin Albert, and Harold Goodglass, allowed the initial establishment of a dementia free cohort that could be followed prospectively for incident changes in cognitive function (Farmer et al., 1987).

The integration of BPA into cognitive research at FHS has evolved over the years (Au & Devine, 2013). The proximity of FHS, twenty miles west of Boston, Dr. Kaplan’s home, resulted in the inevitable influence of how to conduct testing. Elements of qualitative scoring date back to the initial administration of cognitive tests in 1976. This 20-minute examination included the Wechsler Memory Scale (WMS) Logical Memory (LM) Story A (Immediate and Delayed Recall); WMS Visual Reproductions (VR: Immediate Recall); WMS Paired Associate Learning (PAL: Immediate Recall); WAIS Digit Span (DS: Forward and Backward); Letter Fluency (FAS), and WAIS Similarities (see Table 1).

Table 1.

FHS Neuropsychological Variables Collected from 1976–78 Test Battery¹

NP variables	Keyed into dataset2
Logical Memory, IR and DR
Total Score	X
Verbatim/paraphrased details
Confabulations (Related to story)
Confabulations (Unrelated to story)
Intrusions (Related to story)
Intrusions (Unrelated to story)
Logical Memory, DR only
Repeated confabulations, related/unrelated
New confabulations, related/unrelated
Visual Reproductions, IR
Total number of points (all designs combined)	X
For each design:
Draws right to left
Contamination from other VR design
Perseveration
Micrographia
Tremors
Confabulation
No attempt to produce drawing
Paired Associates, IR
Total score	X
Correct/Incorrect for each item
If incorrect:
Another word from list
Related to cue word (e.g., school-books)
Unrelated to cue word
No guess
Repetition of incorrect pairing
Digit Span
Scaled Score	X
Longest correct forward span	X
Longest correct backward span	X
Correct/Incorrect for all trials given
Longest incorrect (serial order only) spans
Similarities
Scaled Score	X
For each item:
Score (2, 1, 0, no guess)
Error type: (set loss, concrete, persev.)
Verbal Fluency (FAS)
Total number of words, all conditions	X
For each condition:
Correct words in 15” time intervals
Errors: Wrong first letter (FAS only)
Errors: Broken rules
Errors: Perseverations

¹Baseline neuropsychological testing for Generation 1 (Original Cohort);

²Variables initially entered into database

It is important to note that at the time these data were collected, data entry and cleaning was a time-consuming and costly effort. At the time of this initial NP assessment, 80-column punch cards were not that far in the past. Thus, although BPA data was recorded, the data entered for analyses included only the standard quantitative scores that were in practice at the time. It was not until 25 years later (2001) that technological advances (i.e., personal computers) made entering the BPA variables listed in Table 1 into the FHS electronic database practical.

In order to retain the longitudinal value of the NP test data, the original tests given from 1976–78 remain part of FHS batteries today. The first NIH grant to study prevalence and incidence of dementia within FHS was awarded in 1989, but surveillance and assessment of progression to dementia began informally in the early 1980s. The 1983 battery (Table 2) was administered only to the Gen 1 participants who were thought to be at risk for dementia. Risk was determined by a drop in MMSE performance, referral from examiners in other FHS studies (e.g., stroke, FHS health exam), or self/family report. This test protocol expanded upon the original battery of tests to allow for a more comprehensive evaluation of cognitive function (approximately 1 hour total administration time). Table 2 shows the tests that were included in the 1983 FHS NP test battery. In keeping with the BPA philosophy that was embedded in 1976–78 battery, the additional tests also integrated BPA procedures to allow for a more in-depth assessment of cognitive functioning (Table 2). For example, the CDT was incorporated into the battery, and examiners tracked the participant’s steps in drawing the clock (e.g., sequence and direction of features placed).

Table 2.

FHS Neuropsychological Variables - 1983 Test Battery¹

NP variables
Logical Memory, IR and DR
Verbatim details
Verbatim + paraphrased details
Prompt given?
Visual Reproductions, IR and DR
For each design:
Total points
Draws right to left
Contamination from other VR design
Perseveration
Visual Reproductions (Recognition)
Total number correct
Paired Associates, IR only
Correct/Incorrect/No guess for each item
Easy Pair Total
Hard Pair Total
Total Score (Easy + Hard Pairs)
Digit Span (Forward and Backward)
Correct/Incorrect for all trials given
Longest correct span
Longest incorrect (serial order only) span
Number of re-instructions (Backward)
Similarities
For each item:
Score (2, 1, no guess)
Clock Drawing Test (Command and Copy)
Approximately symmetrical shape
Hour hand present?
Minute hand present?
Numbers present?
Hour hand to correct number?
Minute hand to correct number?
Perseveration of numbers?
Number sequence correct?
Number of hands?
Neglect present?
Multiple attempts?
Time to completion
Verbal Fluency (FAS)
For each condition:
Correct words in 15” time intervals
Errors: Broken rules
Errors: Perseverations
Boston Naming Test (10 items)
For each item:
Correct without cue
Correct with semantic Cue
Correct with phonemic Cue
Errors: Circumlocution
Errors: Perseveration
Errors: Semantic paraphasia
Errors: Phonemic paraphasia
Errors: Perceptual
Time to initial response (seconds)
Time to final response (seconds)
Block Design
For each item:
Time to completion
Correct/incorrect/correct in overtime
Score

¹Neuropsychological battery administered only to participants screened to be at risk for dementia.

At FHS, the MMSE has long served as a screening tool to track changes in cognitive status. It was first added to the Gen 1 health exam in 1981 (i.e., Exam 18). A drop in MMSE score of 3 or more points from the immediately prior exam, or 5 or more points from any previous exam, triggers a referral for a more in-depth assessment for possible dementia. While the NP test battery that was used for dementia assessment was considered comprehensive for an epidemiologic study in the 1980s, administration was limited to those who were first evaluated by a neurologist. It was the neurologist who then decided whether cognitive testing was warranted. Further, only if the neurology examination yielded a preliminary diagnosis of dementia at moderate or greater severity level was the case brought to consensus review for confirmation of diagnosis. The certainty of AD diagnosis was far less so in these early years and FHS used the moderate severity guideline to ensure higher diagnostic accuracy. The first report of incident dementia from FHS relied on this conservative method for case identification (Bachman et al., 1992; Bachman et al., 1993), which resulted in FHS prevalence and incidence rates being lower than those published by later studies (Hebert et al., 1995; Wilson, Beckett, Bennett, Albert, & Evans, 1999; Seshadri et al., 2011; Satizabal et al., 2016).

Two factors colluded that led to a change in the FHS dementia case ascertainment protocol. First the Gen 1 cohort as a whole was becoming an elderly cohort. By Exam 22 (1990–1994), the average age of survivors was 80 years. Second, it had become increasingly clear during the dementia diagnostic consensus review meetings that NP testing was revealing cognitive impairments indicative of dementia earlier than the standard neurological exam and that relying on the neurologic exam outcomes to determine whether a NP assessment was warranted was creating a bottleneck for incident dementia identification. Thus, in 1992, FHS uncoupled the NP exam from the neurology exam, leading to an eventual drastic increase in the number of cognitive assessments and a significant reconfiguring of the role NP testing had in FHS dementia case ascertainment. Figure 1 illustrates the number of NP versus neurology exams that have been conducted at FHS from 1982 to 2016.

Figure 1.

FHS Neurology and Neuropsychology Examinations Administered Each Year

The impact of untethering NP testing from neurology evaluations had a profound impact on the FHS dementia diagnosis consensus process. In 1990, the data collection form used for the diagnostic meeting was still a carryover from the previous decade during which diagnostic criteria for dementia subtypes largely did not exist, and there was no specificity in the documentation of the progression of disease. With the addition of repeated detailed NP examinations, decline in domain specific cognitive functioning produced data on disease progression with far greater granularity than had previously been possible. This NP data also created significant pressure to alter the conservative FHS practice of using moderate severity as the determinant for dementia case identification. It was becoming increasingly apparent that the additional NP data was altering the threshold for detection of dementia, particularly AD. The bar for diagnosis was lowered in 1999 to be consistent with mild severity, but this internal shift was not being captured in the dementia consensus diagnostic data collection form. After a nearly 2-year iteration process, the dementia diagnostic protocol was re-configured and included the added delineation of disease progression from normal cognition to, where applicable, documentation of the dates of cognitive impairment onset and progressively more severe stages of the disease. Contemporary dementia subtype diagnostic categories were also added and led to a multi-year effort in which these new criteria were applied to the 995 cases that had been reviewed previously.

With documentation of cognitive function taking on a dominant role in incident dementia surveillance, the impact of the embedded BPA in the NP testing was also having a significant influence on case identification, particularly as the threshold was shifted from moderate disease severity to mild. It is general practice in a clinical work up to seek consistent patterns of performance within domain specific tests in order to properly diagnose intact versus impaired cognitive skills. Applying the BPA to a clinical exam strengthens the mapping of cognitive profiles both within a test and across tests. For example, documenting perseverative responses produced during a verbal memory test and poor performance on the Wisconsin Card Sorting Test (WCST) would lend stronger support to a diagnosis of dysexecutive function than just results from the WCST alone. And perseverative responses documented on a verbal memory test, visual memory test, naming test, visuospatial test, and the WCST would be even stronger evidence of a persistent deficit in executive function. It is likely that repetition of an error type across different tests would signal a more definitive impairment compared to a single measure. In the case of executive function, which comprises multiple components (e.g., abstracting, planning, cognitive flexibility, initiating, inhibiting), it is possible for the quantitative score on a specific test to be within expected limits, but errors made during the course of task completion reveal executive deficits related to other underlying processes. Or conversely, poor test performance might be related to the specific characteristics of the test itself, and not any significant domain specific impairment (e.g., low scores on abstract reasoning tests are also associated with low education and cultural differences). Thus, using BPA can enhance both within test and across test reliability in assessment of domain specific skills. Lastly, the added insight provided by the BPA measures informs diagnostic impression and dementia diagnoses, particularly related to AD. In the FHS diagnostic meetings, once cases are reviewed to determine if criteria for dementia are fulfilled, qualitative findings from NP test performance are often used in tandem with traditional quantitative scores to designate date of onset (earliest symptoms), date of diagnosis (earliest date when diagnostic criteria are met), dates of transition in disease severity (mild to moderate to severe), and dementia subtype. This system has enabled FHS researchers to systematically track cognitive status before and after the diagnosis of dementia, permitting fairly accurate ascertainment of the temporal evolution of cognitive impairment.

Preclinical AD and Expanded Use of BPA

The insidious nature of AD makes demarcating where normal cognition ends and clinical disease begins difficult to determine. When Petersen et al. (1999) initially defined mild cognitive impairment (MCI), it was a noteworthy attempt to differentiate a symptomatic period that did not meet current diagnostic criteria for dementia. Neuropathological studies of MCI cases, however, revealed that hallmark AD tau and amyloid pathology were already present in medial temporal and cortical regions of the brain (Forsberg et al., 2008; Jack et al., 2008; Jack et al., 2009; Koivunen et al., 2011; Okello et al., 2009; Mufson et al., 2012). Further, the continued failure of clinical trial studies reflects, in part, treatment interventions that are too late to reverse course of the disease (Sperling, Karlawish, & Johnson, 2013; Salloway et al., 2014; Doody et al., 2014; Belluck, 2016; Taylor, 2017). A recent revision of diagnostic criteria for MCI (Albert et al., 2011) and separate research criteria for preclinical disease that precedes MCI (Sperling et al., 2011) reflect increasing recognition of AD as a life course disease.

Present research on preclinical AD centers on identifying biomarkers predictive of cognitive impairment and decline within individuals who are largely asymptomatic (Sperling et al., 2011). The presumption is that these biomarkers may signal increased risk for disease well before cognitive function is affected. Yet as early as 1995, before the publication of Petersen’s 1999 seminal paper defining MCI, Linn et al. (1995) reported that FHS participants who were 65 years and older and were subsequently diagnosed with AD performed worse on measures of memory, verbal learning, abstract reasoning, and executive function as much as 13 years earlier compared to those who remained cognitively intact. The average number of years between baseline NP testing and date of diagnosis was 7 years.

With continued surveillance of the FHS dementia cohort, Elias et al. (2000) re-examined the length of the pre-clinical phase of AD up to 22 years after the baseline NP examination. Results aligned with those of the Linn et al. study. Because FHS had still been operating under its pre-NINDS-ADRDA legacy, Ellis, et al. recognized that both the earlier findings from Linn, et al. and the initial results from their own study could be biased by inclusion of mild AD cases in what was identified as a dementia-free cohort. Thus, additional analyses were conducted censoring any incident cases within 5 and 10 years of baseline NP. They found that even among persons who were dementia-free for 10 years and who were subsequently diagnosed with AD, performance on verbal memory, verbal learning, and abstracting reasoning tests was significantly lower compared to those who remained cognitively intact. The average length of time between baseline NP and dementia diagnosis within this group was 13 years, and extended up to 20 years.

The results of Elias et al. (2000) suggest that pre-clinical indicators of disease may have been evident as long as two decades before disease onset. Linn et al. (1995) and Elias et al. (2000) suffered from important methodological limitations that did not meet contemporary standards, such as the exclusion of AD cases of mild severity and people with low education. However, more recent research also suggests that the earliest indicators of AD may predate clinical onset by as much as two decades (Reiman et al., 2012), which would mean that these markers are not restricted to the oldest decades of life. This raises the issue that if a detectable difference in cognition is evident decades before clinical disease is diagnosed, and extends to those of middle age, the concept of what is “pre-clinical” may need to be readdressed and methods for early detection may need to be modified. To date, much research on MCI continues to rely on the same measurement tools as those used to determine presence of clinical disease. Further, the diagnostic criteria for MCI is based on a modification of those used for AD. If research is to explore the very earliest potential markers of disease, it seems unlikely that current test methodologies will be adequate to do so. It is within this context that FHS turned to expansion of the BPA approach as providing a potential solution.

One of the great strengths of FHS is that it comprises nearly 7 decades of prospective study of a multi-generational cohort. Being tied to the past, however, also presents challenges. As science and knowledge advance, so too do the research methods and instrumentation. Applied to FHS, the challenge is in preserving the longitudinal integrity of the data while adopting new approaches that are aligned with what is current state-of-the-art. It is with this concern that FHS greatly expanded its use of BPA as a means of potentially increasing NP test sensitivity to detect more subtle changes in cognitive functions while also protecting the capacity to document longitudinal changes from NP tests that were given as far back as 1976.

The diagnostic criteria for MCI was first developed and widely adopted by clinical researchers within Alzheimer’s Disease Research Centers (ADRC), and thus applied to biased study samples. In parallel, from 1999–2005, FHS began a baseline cognitive assessment of the Gen 2 and OmniGen 1 cohorts. This NP evaluation was administered to the Gen 2 cohort when the average age was near that of the Gen 1 cohort when it underwent initial NP testing. In fact, the exact same versions of tests administered in 1976–78 were carried over into the NP test battery. Since NP testing itself had evolved during the ensuing two decades, new tests were also added, essentially doubling the time of testing. These new tests were also added to the NP evaluation of those being followed for progression of dementia, thus further increasing sensitivity to pick up cognitive changes related to neurodegeneration. In addition to following an at-risk subsample, FHS was also testing a community-based cohort, and so it was in a unique position to see the range of performance on cognitive testing that many ADRCs were not. Between 1999–2001, FHS tested 3,062 participants, and, during this 3-year period, it was increasingly clear that BPA responses that had been previously attributed to those with suspected dementia were also evident in those who were clearly not demented. Further, because NP scoring was still largely restricted to documenting defined correct responses only, it was evident that the NP data collection methods were not capturing the richness of cognitive performance. This clinical observation led to a complete overhaul of the FHS NP scoring system. Details on the rationale and full-scale implementation of BPA across all NP tests are provided in Au & Devine (2013).

While the need for an expanded BPA scoring system was strongly apparent by 2001, it was not fully implemented until 2005 when the Gen 2 and OmniGen 1 cohorts were invited back for a second NP assessment to document longitudinal differences relative to their baseline. The 2005 battery (Table 3) included additional tests and expanded on the number of qualitative variables collected (e.g., see Logical Memory, Visual Reproductions, and Clock Drawing Test).

Table 3.

New Qualitative Neuropsychological Variables Recorded and Coded - 2005 Battery¹

NP variables
Logical Memory, IR and DR
Verbatim/paraphrased points (by detail)
Confabulations (Related to story)
Confabulations (Unrelated to story)
Intrusions (Related to story)
Intrusions (Unrelated to story)
Logical Memory, DR only
Repeated confabulations/intrusions, related and unrelated
New confabulations/intrusions, related and unrelated
Logical Memory, Recognition
Response chosen for each question
Visual Reproductions, IR and DR
For each design:
Micrographia
Tremors
Confabulation
Starts to draw before told to begin
No attempt to produce drawing
Indicates “I do not remember”
Indicates of awareness of 4 drawings
Visual Reproductions (Delay only)
Incorrect IR production repeated exactly
Incorrect IR production repeated partially
Visual Reproductions (Recognition)
Response chosen for each design
Choice same as DR production?
Paired Associates, IR and DR
If incorrect:
Another word from list
Related to cue word (e.g., school-books)
Unrelated to cue word
Repetition of incorrect pairing
Paired Associates (Recognition)
Response chosen
Digit Span (Forward and Backward)
Correct/Incorrect for all trials given
Similarities
For each item:
Error type: (set loss, concrete, persev.)
Clock Drawing Test (Command, Copy, Pre-drawn)
Attempt to self-correct any error
Outline:
Present (y/n/not circular)
Formed by continuous line
Direction of motion during drawing
Longest diameter
Perpendicular diameter
Direction of longest diameter
Outline overdrawn (>2 times around)
Outline perseverated (multiple circles)
Numeral Placement:
Rotated paper while placing numbers
Anchor numerals or substitutes placed before any others (all that apply)
Numerals are absent, written, Arabic, Roman, combined
Measurement of most medially deviated
Anchor displacement
Non-anchor displacement
Numbers/substitutes placed on or outside the outline
Dots, words, or symbols substituted for one or more numerals
Two 12’s are present
Omission of number(s)
Numerals beyond 12 are present
Sequencing errors
Time Setting
Needed reminder of time for hand-setting (Command only)
Hand points to “11” OR correct position (not both)
Hand points to “2” OR correct position (not both)
One line drawn to two numbers
One hand incorrectly points to the 10
Length of hour versus minute hands
Center of hands on horizontal axis
Center of hands on vertical axis
Extraneous marks in clock
Tester’s clinical assessment of clock drawing (normal, mild, mod, sev)
Verbal Fluency (FAS, Animals)
For each condition:
Correct words in 15” time intervals
Boston Naming Test (36 items)
For each item:
Errors: Circumlocution
Errors: Perseveration
Errors: Semantic paraphasia
Errors: Phonemic paraphasia
Errors: Perceptual
Trailmaking A and B
Time to completion (sec)
Number of self-corrected errors
Number of examiner-corrected errors
Obvious Tremor
Number of Pen Lifts
Early start for sample
Early start for test
If discontinued, number of circles completed
Wide Range Achievement Test – Reading
Correct/Incorrect for each word
Finger Tapping
Hooper Visual Organization Test
For each item:
Score (0, 0.5, 1, no guess)
Errors (Isolate, perceptual, other)
WAIS Block Design
For each item:
Broken configuration
Incorrect due to only one block in error
WAIS Information
Correct/incorrect for each item
Cookie Theft
Type of writing (print/cursive/combined)
Syntactic complexity (sentences are incomplete/simple/complex)
Inclusion of major events
Unusual spatial clumping of words?
Incorrect use of capital letters?
Incorrect use of punctuation
Rotated paper
Total time for testing beginning to end
Factors affecting testing
Behavioral Observations

¹This battery was administered to Gen 1 (Original Cohort), Gen 2 (Offspring Cohort), and Omni (multi-ethnic cohort) regardless of whether they were deemed at risk for dementia.

Fully embracing the BPA methods for cognitive assessment required accurate capture of the full range of spoken responses. The decision to significantly increase BPA scoring proved inadvertently critical for setting the foundation for innovation. Because it is not physically possible to write down everything a participant says for any given test, audio recording of test sessions became necessary. This is in keeping with longstanding FHS quality control practices to maximize reliability in data collection and data cleaning. The importance of generating high quality data is central to all research and results in documentation of protocols and procedures to provide clear transparency. Within FHS, there is an amplified appreciation for scrupulous documentation because FHS data will be available and potentially of continued value to the scientific community decades later. The current scientific community benefits from FHS data that has been collected for nearly 70 years, and any contemporary work may be of interest to researchers 70 years from now. Thus, it was with this rationale that the decision was made to record these sessions digitally. Doing so provided a streamlined method for storing these voice recordings. Further, these digital recordings were used not just to ensure that participant responses were being meticulously documented, but they also provided a method for ensuring test administration and scoring was adhered to longitudinally. Since testing a cohort requires a team of trained examiners, and there is turnover in these testers, each month a recording of one test session for each examiner is randomly selected and is listened to to check that both the test administration and scoring protocols are being followed, preventing a “testing/scoring drift” that can often happen over prolonged periods of repeated testing.

While NP tests are typically identified as domain specific, known to any clinician or researcher engaged in cognitive assessment is the fact that all NP tests in fact tap into multiple domains. Clinically, experienced neuropsychologists do not make diagnostic determinations on the basis of any one test but rather on the pattern of performance across tests. As championed by Dr. Kaplan, using BPA enhances the understanding of the brain-behavior relationships that a single score does not provide. In essence, BPA considers the multiple domain dimension of any NP test. With the expanded BPA scoring system, FHS began to quantify, in much greater detail, the multi-faceted dimensionality of NP tests, something that has not been widely considered nor applied in a research context. In fact, most studies of preclinical AD continue to depend on the same types of cognitive assessment tools that were initially developed to differentiate between those with frank clinical symptoms of disease and those who had not yet met threshold for clear cognitive dysfunction.

The widespread practice of using NP tests as a measure of a specific cognitive domain in research has had an arguably profound impact on the efficacy of finding effective treatment for AD. As reflected in the widely cited Jack et al. (2010) theoretical model for disease progression, significant cognitive symptoms are deemed not to emerge until much later in its course. The compendium of AD clinical trial studies presumes the temporal relationship reflected in the Jack et al. model, leading to a focus on imaging and fluid biomarkers (Lim et al., 2014; Lim et al., 2012; Lim et al., 2013a; Lim et al., 2013b; Small et al., 2012; Ellis et al., 2013; Morris et al., 2009) as indices of preclinical disease, despite the fact that the prevalence of these biomarkers in a normal population remains unknown. What has been documented in neuropathological studies is that there is a subset of the population in which AD pathologies are evident that meet the threshold of neuropathological diagnosis, but the disease was not clinically expressed. Furthermore, there are non-AD pathologies that are incorrectly diagnosed as AD clinically. Through its own brain donation program, FHS has conducted independent ante-mortem and post-mortem diagnoses on 200 autopsy cases; 53 clinically diagnosed as AD did not meet neuropathological criteria for AD. These findings suggest that the clinical expression of AD symptoms is related to variability in underlying pathology and may not solely be related to presence of amyloid or tau.

Cognition as a surrogate marker of preclinical neurodegenerative disorders has only been recently considered. Studies have reported that using cognition instead of fluid, imaging, or genetic biomarkers may be more efficacious in predicting progression from preclinical to disease state (Twamley, Ropacki, & Bondi, 2006). In FHS, clinical diagnosis of early symptoms is dependent on cognitive assessment, and the FHS dementia review process has always been aided by the availability of the BPA data embedded in the NP test administration and scoring protocols.

From 2005–2011, FHS documented baseline BPA performance across an entire NP test battery in a community based cohort, and, to our knowledge, this is the first epidemiologic study to have done so. An expanded BPA scoring protocol was implemented while preserving standardized test administration and scoring procedures. This was an important consideration since one of the primary reasons BPA has not been widely adapted in research is because of the alterations in standard NP testing that are generally presumed warranted. Most published studies that report BPA related findings rely on administering modified versions of standardized tests or using altogether newly designed tests. Often lost are the well-studied, well understood measures against which to compare new findings to existing. For large-scale epidemiologic studies such as FHS, any changes to study protocol entail the risk of losing what is the greatest value, which is its ever-extending longitudinal nature. The fact that FHS can look at cognitive changes across decades using the same metrics of comparison is what sets FHS apart from more recently initiated longitudinal studies.

Because of the novelty of the BPA metrics, a number of FHS studies have tried to determine their potential clinical utility. In the first published study of BPA normative data on a community based study, not surprisingly errors were found to be more prevalent in older compared with younger individuals and less prevalent in individuals who completed at least some college (Hankee et al., 2013). Normative values for the Clock Drawing Test (CDT) features were also published (Nyborn et al., 2013) and preliminary findings suggest that incorrect placement of the minute hand by cognitively intact individuals may be pathognomonic of preclinical dementia (Piers, Ning, et al., 2016).

Additional published papers have documented commission of error responses on memory and executive function tests, with the presumption that they may serve as additional potential markers of early cognitive change (e.g., Gupta et al., 2015; Nishtala et al., 2014). Recent studies have sought to validate error metrics against neuroimaging biomarkers. While temporal horn volume (THV, a surrogate measure for hippocampal volume) was associated with poorer performance on a verbal memory test (WMS Logical Memory), white matter hyperintensities (WMH) were associated with a greater number of related errors (Libon et al., 2015). On the WMS Visual Reproductions-delayed free recall subtest, an error score that combined intrusions, perseverations, and repeated incorrect responses was computed, and results suggest that among participants with high quantitative scores (i.e., good recall of story details), those who also had high error scores had larger THV (which is interpreted as lower hippocampal volume) compared to those who made few errors (Libon, Preis, et al., 2014).

BPA in the Current Decade (2010–2020)

FHS recently completed a repeat administration of its BPA enhanced NP test protocol to the Gen 2 and OmniGen 1 cohorts (2011–2016), and is now beginning a third round of BPA NP data collection. Of significance is that the average age of the Gen 2 cohort is currently 75.4 years, with 90.9% aged 65 years or older, which means that as a cohort they are now at the age of highest risk for dementia/AD. The youngest generation cohort (Gen 3/OmniGen 2) is currently undergoing a second round of BPA enhanced NP testing as well, providing FHS with the opportunity to document incident changes in both traditional and BPA cognitive measures that span the entire adult lifespan. Anticipated during the remainder of this decade will be the opportunity to validate the clinical utility of the BPA approach in providing preclinical cognitive markers of AD.

As noted earlier, to facilitate the BPA scoring strategy, FHS digitally audio recorded NP testing beginning in 2005. To date FHS has accumulated 7500+ recordings of spoken responses to NP test questions, which includes those from 2230+ participants who have two or more recordings. Among those with multiple recordings, 130+ subjects, who were considered cognitively intact at the baseline recording, were subsequently diagnosed with incident cognitive impairment or dementia. With the rise in voice recognition and voice analysis software, these digital voice recordings have evolved into digital voice data from which potential cognitive biomarkers of incident disease can be determined. Some of the voice related measures being derived in collaboration with Evidation Health (a Silicon Valley based company) and Jim Glass, Ph.D. and his research team at Massachusetts Institute of Technology (MIT) include changes in word use/choice, increased number and latencies of pauses between spoken responses, shifts in tonal quality, etc. The value of having digitally stored data is that as automated algorithms for extracting new features are created, they can be easily applied to all available recordings, virtually instantly, creating additional longitudinal metrics. In collaboration with the researchers from MIT, analysis of a subset of recordings has been conducted in which 256 prosodic and speech-to-text features were extracted and used to predict those with incident cognitive impairment (Alhanai, Au, & Glass, under review). Some of the features measured included number of words, vocabulary, pitch, speaker turn taking, hesitations, and speaking rate. We anticipate that as more participants are diagnosed with incident dementia during the next 3–5 years, we will be able to further confirm the predictive utility of these digital voice markers.

While FHS digital voice data was serendipitous, extending NP data collection methods to include digital capture of written responses was deliberate. Starting in 2011, using the academic version of proprietary software developed and patented by researchers at MIT and Lahey Hospital and Medical Center (U.S. patent number US 2008/9243933A1), FHS substituted use of a regular ballpoint pen with a digital pen for the Clock Drawing Test. The MIT/Lahey-owned software transforms the data captured by the digital pen into more than 1000 analyzable features that have been combined to reveal decision-making latencies and graphomotor characteristics that may serve as measures of underlying cognitive processes (Piers et al., under review; Souillard-Mandar et al., 2016; Cohen et al., 2014; Libon, Penney, et al., 2014), taking the BPA approach to a level of precision that hand scored BPA efforts cannot. For example, preliminary exploration of metrics obtained using the digital Clock Drawing Test (dCDT) suggests that information processing speed is not a single, all-encompassing construct (Piers, Devlin, et al., 2016). Rather, it may be differentiated into at least three components: (1) motor behavior, (2) non-motor behavior, and (3) higher-order decision latencies. Motor behavior can be assessed by variables such as the total number of strokes needed to complete the drawing. Non-motor behavior can be assessed by activity such as the percent time participants devote to thinking (‘think time’) and the percent time participants devote to drawing (‘ink time’) (Davis, Libon, Au, Pitman, & Penney, 2014; Cohen et al., 2014). Finally, higher-order decision latencies are reflected in the time intervals as participants transition from one portion of their drawing to the next (Libon, Penney et al., 2011; Libon, Penney et al., 2014; Penney et al., 2014). Clinically relevant higher-order decision latencies include, but are not limited to: (1) the latency after drawing the clock face, (2) the latency before drawing the first hand, (3) the latency before drawing the second hand, and (4) the latency between drawing the numbers. The dCDT provides a rigorous, psychometrically objective assessment of the constructs underlying clock drawing behavior (Piers et al., under review).

In 2013, FHS extended the use of the digital pen to all participant drawn tests and is developing software to extract derived metrics. For example, the Trail Making Test evaluates processing speed and executive functioning. In Trails A, the participant must draw a continuous line connecting sequentially labeled targets, 1 to 25. In the second trial, Trails B, the targets alternate between a sequence of numbers and letters, and the line must be drawn from 1 to A, A to 2, 2 to B, and so on. Scoring of the traditional pen and paper version includes the temporal variable total time to completion. However, capturing the performance digitally allows for assessment of many additional features, such as latencies between targets, existence of tremor, and additional metrics of pen speed, acceleration, and path.

Cognitive decline is one of the most concerning behavioral symptoms associated with AD. While cognitive complaints serve as a surrogate preclinical measure of decline (Buckley et al., 2015; Snitz et al., 2015; Perrotin et al., 2017; La Joie et al., 2016), they are subject to the inherent bias of self-report (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003; van de Mortel, 2008). Integrating digital capture of performance on NP tests allows precision measurement of behaviors, which is, in essence, providing a window into the brain as it constructs responses to the NP task at hand. We consider that the granularity of performance that can now be measured through digital capture can be described as BPA on steroids. Importantly, FHS continues to give its NP tests with the same test administration procedures that date back to 1976. The preservation of the traditional paper-pencil methods is what distinguishes the FHS approach from that of current computerized NP tests such as CogState, CAMCI (The Computer Assessment of Memory and Cognitive Impairment), and CANTAB (Cambridge Cognition) that also produce novel and precise performance metrics. We contend that, especially for our oldest FHS participants, asking them to do NP testing using a mouse or keyboard is not a comparable test situation to traditional paper-pencil testing. The FHS approach of preserving the past, while finding ways to incorporate new technological advancements, is a particular advantage over other large-scale studies that have long been collecting longitudinal NP data. FHS provides a means to leveraging the past while also enabling the future.

To take one step further, beginning in 2014, in collaboration with researchers from Tsinghua University and Peking Union Medical College Hospital, FHS initiated creation of a platform that will capture all NP testing digitally. This tablet-based platform is customizable to any existing NP test and utilizes a different digital pen than is currently used. The pen uses blue tooth technology to wirelessly connect to the tablet, and gathers a participant’s response drawn on a regular piece of paper and transfers the time-stamped latency and graphomotor data to a secure server (via the tablet) in real-time. Importantly, as has always been the case within FHS, participants will still use pen and paper when performing tasks.

An added operational benefit of the platform is the automation of much of the data collection, scoring, and entry, which will significantly reduce the labor and costs of current digital data acquisition and analysis. For example, in addition to recognizing words, the tablet-embedded voice recorder can also extract key metrics, such as the latencies for spoken words and pauses between each word, and it can differentiate between fragmented versus fluid responses. The pen-based responses can also yield information about features such as the latencies for different graphomotor characteristics, behaviors when the pen is off the paper (e.g., “hovering” between test items), and fragmented versus fluid writing. Anticipated machine learning modeling methods are expected to detect cognitive changes earlier and increase diagnostic accuracy. The training algorithms used to collect and analyze performance data is self-correcting and will increase in predictive accuracy over time.

The Future of Cognitive Assessment: Vision for 2025

Cognition is at the heart of what differentiates humans from all other forms of life. By optimizing brain health all else is enabled, because everything we do, we do through our brains. While neuroscience has laid a foundation of understanding of brain-behavior relationships, the current methods of NP testing are an artificial means by which to measure cognitive capabilities. Technology is changing what we can do and how we can do it. E-health and the internet-of-things (IoT) have translated into a plethora of wearable sensors and smart home devices flooding the market, with more devices in the pipeline. Together they offer a solution for monitoring and detecting changes at the earliest stages, enabling feedback and intervention strategies.

While still in its nascent stages, digital health is being pursued earnestly by academia, industry, and government. Computational methods create derived measures from digital signals that are validated against defined gold standards. However, since these gold standards were created using data collection methods that were intermittent, adhering to these methods of validation leads to incremental results. The prior precedent of the medical model, with its focus on the treatment of disease after symptoms emerge, is being challenged with the insinuation of the technology sector into the healthcare space. In the U.S., major corporations such as Apple, Google, General Electric, Amazon, and IBM have all moved rapidly into the arena of health and wellness. Worldwide, mature and startup companies are upending the traditional convention of waiting until key health metrics reach thresholds that trigger action by the medical community, a practice in the U.S. that has led to the intractable problem of high healthcare costs and low health quality. A whole new sector of the health industry is growing from the development of new technologies and applications to prevent disease and promote healthy aging and wellness. The economic impact is as significant as the social one as there is nothing that will reduce healthcare costs faster than removing the need for it.

This technology movement will also dramatically alter the ways in which cognitive function will be assessed. As the myriad of smart sensors pick up continuous measures of behavior, it should soon be possible to infer cognitive capabilities through the integration and interpretation of multi-sensor signals. These algorithms are already being built in artificial intelligence (AI) laboratories, with astonishing advances routinely showcased in public events, including IBM Watson’s defeat of the world’s bests in chess and Jeopardy, and Google DeepMind value network solution’s mastery of the game of Go (Silver et al., 2016; Cyranoski, 2017; Obermeyer & Emanuel, 2016). The leading medical journals have not only highlighted how AI is revolutionizing healthcare practice, they have also created new journals to provide a venue for publishing technology-medicine advancements (Lee, Campion, Morrissey, & Drazen, 2015; BMJ Innovations, 2014). This is also true in brain related research, such as the Alzheimer’s Association recent releases of their new journals, Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring (Snyder, 2015) and Alzheimer’s & Dementia: Translational Research and Clinical Intervention (Alzheimer’s Association Update, 2015).

FHS is positioned to be part of this future vision. At its foundation is the decades of existing health related data to validate new digitally acquired cognitive indices. Our current work seeks to determine whether there are a common set of digital cognitive biomarkers that are agnostic to test. Since responses to all NP tests tap into multiple domains, the precision of behavior captured through digital acquisition should include overlapping features regardless of test. Removing test specific variability will potentially revolutionize how cognition is assessed in both research and clinical settings. For example, the impact could radically change the conduct of clinical trials in which cognitive impact is of interest, including drug discovery for neurodegenerative disorders as well as drug side effects. Rather than reliance on a battery of domain specific tests which produce discrete scores that are subject to confounding factors that might mask true cognitive status, studies can be optimized for both time and costs based on a much smaller set of validated objective metrics that in combination may reflect a more accurate cognitive profile that accounts for fluctuations in performance. Performance noise and cost are the most significant barriers to scaling cognitive assessment in any large-scale study.

Another intriguing concept being explored at FHS is whether machine learning technology offers the possibility of analyzing raw digital signals with little to no clinical guidance. There is evidence suggesting that these approaches can produce highly accurate clinical diagnoses (Kolachalama et al., under review). Catalyzing groundbreaking discoveries for cognitive impairment prevention by identifying determinants of progressive dementing disorders, such as AD, will inform new strategies for sustained lifelong cognitive health. While using data on hand provides a practical blueprint on how to leverage existing resources, empowering what does not exist is what will most enable the future.

The home provides an opportunity to collect rich data more passively and potentially in a less burdensome fashion to the person. As poorer health outcomes increase, mobility decreases, leading to higher subject attrition for any study of a progressive condition or disorder. Embedded technologies in the home, done in a way that is not personally intrusive, may be able to capture much of the important data needed for those who are at higher risk for AD and other neurodegenerative disorders. With this in mind, the FHS cognitive aging research program is collaborating with other FHS investigators who have launched the e-FHS study in which health related measures are being collected using smartphone and wearable devices (Fox et al., 2016). Pushing data collection from the artificial setting of a clinic and into a person’s natural environment is the first step toward the exciting prospect of continuous monitoring of cognitive capabilities. Rather than waiting for overt symptoms, it will now be possible to detect the subtle, yet persistent, indications of a negative trajectory of change. Interventional strategies at these stages may not simply attenuate a progressive decline, they may well alter the trajectory altogether.

Cognitive impairment and decline does not have to be an inevitable part of aging. The biological mechanisms underlying non-genetically dominant neurodegenerative disorders remain largely unknown. As identification and treatment for these disorders increasingly center on preclinical detection, the heterogeneous cognitive profiles of at-risk individuals may explain the high clinical trial failure rates. Technological advancements are disrupting known best practices, and in this context the role of cognition should not be underestimated. IBM Chief Executive Ginni Rometty has repeatedly noted in her public comments that cognitive computing is going to lead to a new era in personalized and precision medicine. While she is focused on how computers and AI will metamorphosize healthcare, people still remain at the forefront of what machines are trying to mimic. It is a pursuit that has no end because it is the power of the human brain that generates the next set of challenges that hardware and software are trying to emulate. Cognition in its raw human form is something that will always be the driving force for human understanding, and the vision for 2025 is that neuropsychologists should embrace the opportunity to lead this cognitively-driven future.

Picture 2.

Blood Pressure Measurement in the Early Years

Public Significance Statement.

This article discusses how the Boston Process Approach enhances the ability to detect changes associated with dementia, particularly Alzheimer’s disease (AD). Technology is changing what we do and how we do it, which is expected to catalyze groundbreaking discoveries for effective treatments and prevention of neurodegenerative cognitive disorders, such as AD.