Precision Neurocognition: An Emerging Diagnostic Paradigm Leveraging Digital Cognitive Assessment Technology

Abstract

BACKGROUND:

Strategies that may modify Alzheimer’s disease (AD) and other dementia disorders are being developed. To maximize the benefits of these strategies, it is critical that indicators suggesting neurocognitive decline are identified as early as possible. ‘Precision neurocognition’ is a heuristic that seeks to develop methodologies capable of identifying subtle behavior(s) that may flag emerging AD and other dementia related syndromes. Recent research suggests that digital neuropsychological assessment technology may be the platform that can realize the goals of precision neurocognition, i.e., the early detection of neurocognitive difficulties that are prognostic for mild cognitive impairment (MCI) and dementia.

METHODS:

Past research associating 100 percent correct or statistically within normal limits responding using neuropsychological tests with time-based parameters obtained while participants undergo assessment is reviewed.

RESULTS:

Recent research with community dwelling and memory clinic participants examined test scores obtained using commonly available neuropsychological tests. This research extracted a number of discrete latency measures that clearly dissociate between groups, despite final test scores that are either 100 percent correct or statistically within normal limits.

CONCLUSIONS:

Past research using digitally administered neuropsychological tests suggests that the goals of precision neurocognition as related to the early identification of neurodegenerative illness may be realized via an analysis of time derived, process-based behavior using digital assessment technology. Latency or time-based parameters as described in recent research could form the basis of a range of neurocognitive biomarkers for identifying people at risk for developing AD, other dementing disorders, and MCI.

1.). Introduction

1a). Scope of the Problem.

It is commonly understood that early detection of mild cognitive impairment (MCI) and dementia will lead to better and more effective intervention and treatment [1]. Yet, at the current time, in the United States and globally, the care pathways for MCI and dementia is the management of these conditions rather than early identification or mitigation [1,2]. In 2022, 4.0% of adults aged 65 and older in America report they have received a dementia diagnosis [2]. Equally alarming, is that many individuals in the United States who likely meet criteria for a diagnosis for MCI or dementia, have never been properly assessed [1]. Clearly, health systems are struggling to meet these demands, particularly in neuro-specialty care areas [1].

Beta-amyloid plaques (Aβ) and tau neurofibrillary tangles are important biological substrates associated with Alzheimer’s disease (AD). Traditional paper and pencil neuropsychological assessment can do a reasonable job in assessing the brain-behavior relations associated with AD and other dementia illnesses when the illness is more or less present. However, these proteins are thought to accumulate years before the indication for a clinical diagnosis of either MCI or AD [3,4]. This situation constitutes a serious unmet challenge to neuropsychology as it is not clear how well traditional neuropsychological outcome measures operate in the context of subtle, nascent illness.

An additional controversy is the relationship between clinical or neuropsychological indications for suspected AD versus the absence or presence of fluid/ imaging biomarkers for suspected AD. Indeed, past research has shown variability and heterogeneity with respect to the neuropsychological, fluid/ imaging biomarkers, and underlying neuropathology associated with AD and AD-related dementia [5,6]. In this context, the International Work Group [7] has recently updated their diagnostic criteria for AD. These new criteria provide some guidance regarding when and how to differentiate between patients who may be asymptomatic, presymptomatic, or positive for an AD-related dementia. These updated criteria decouple the presence or absence of neuropsychological impairment from the presence or absence of fluid/ imaging evidence for proteins associated with an AD-related dementia. In this context, information obtained using neuropsychological tests, and fluid or imaging biomarkers are accorded equal importance.

All of these circumstances suggest that there is an urgent need for neuropsychological assessment paradigms that can be deployed both routinely and longitudinally so as to detect subtle neurocognitive decline that can predict progression from asymptomatic to presymptomatic to an actual dementia illness.

1b.). Precision Neurocognition.

There is now tremendous interest in the development of digital neuropsychological assessment technology [8,9]. In this paper, the term, ‘precision neurocognition’ is best understood as a heuristic that seeks to develop new methodologies that can operationally-define subtle neurocognitive alterations that might be associated with the emergence of MCI and dementia. ‘Precision neurocognition’ is derived from percepts associated with the Boston Process Approach to neuropsychological assessment as created by Edith Kaplan [10,11]. These percepts suggest that our understanding of the relations between brain and cognition can be enhanced by operationally defining the errors made on neuropsychological tests; and disambiguating the underlying processes used to bring tests to fruition. In this context, the digital administration and scoring of neuropsychological tests are able to reveal subtle, occult, highly nuanced behavior that can operationally define key neurocognitive constructs to identify individuals with MCI and dementia conditions [9]. Given the power of digital assessment technology to reveal subtle alterations in neuropsychological abilities [8,9] it might be possible that indications suggesting nascent neurocognitive decline could be revealed sooner than is currently possible using common paper and pencil neuropsychological tests.

Traditionally, alterations in episodic memory, as assessed with well-known paper and pencil tests, have been viewed as an early cognitive symptom to suggest the onset of neurodegenerative conditions such as AD. However, this notion was developed during earlier generations when many patients presenting for a work-up for suspected dementia were likely exhibiting obvious neurocognitive disabilities. Recent research using digital neuropsychological technology suggests that participants who are well and living in the community, and memory clinic patients diagnosed with MCI can present not only with classic episodic memory difficulty, but with alterations as defined with time-based or latency parameters from a wide array of tests that assess verbal working memory, graphomotor information processing speed, visuoconstructional abilities, and verbal episodic memory recognition test conditions [12–14]. These newly emerging empirical findings constitute a platform suggesting how ‘precision neurocognition’ using percepts derived from the Boston Process Approach [10,11], might be leveraged to identify emergent MCI and dementia.

This paper reviews recent research documenting significant between-group differences with respect to latency in the context of correct or statistically within normal limit responding extracted from digital adaptations of common paper and pencil neuropsychological tests. We define within normal limit responding as performance within the range of one standard deviation from the mean set by normative test data. While many outcome measures from neuropsychological tests are often expressed as simple binary correct or incorrect outcome, the concomitant latency to generate a correct or statistically within normal limits response is a continuous variable.

The thesis to be developed in this paper is that (1) correct or efficient responding should not be viewed as a binary, correct versus incorrect variable; and (2) an analysis of time-based or latency parameters associated with correct responding could result in a set of digital/ neurocognitive biomarkers that might presage the emergence of neurodegenerative illness.

2.). Time-Related Neurocognitive Deficits in Dementia and Mild Cognitive Impairment

2a.). Context Independent versus Context Dependent Neuropsychological Deficits.

Prior empirical research investigating patterns of neuropsychological impairment in insidious onset dementia syndromes, such as AD, and putative vascular dementia (VaD), found that the malignancy of dysexecutive impairment was directly related to the degree or severity of MRI-defined subcortical white matter alterations [WMAs; 15,16]. In the series of investigations, Libon and colleagues [17] observed that the dysexecutive difficulty observed in the context of significant MRI WMAs was context independent, i.e., dysexecutive difficulty was clearly observed across virtually all neurocognitive domains that were assessed including language-related and memory-related tests. By contrast, the dysexecutive difficulty among dementia patients with little WMA was context dependent, i.e., restricted to selected neurocognitive domains.

2b.). The Theory of Executive Attention.

An important observation that resulted from the distinction between context dependent versus context independent dysexecutive difficulty as described by Libon and colleagues [17] was that de-railed performance on executive and other tests was not monotonic but was time dependent. Depending on the task at hand, a low overall summery score appeared to be associated with greater omissions, commissions, and perseverations towards the middle and latter portions of the total response. In other words, performance when assessed in terms of correct responses and the proliferation of errors could be assessed as a function of specific time epochs.

The Theory of Executive Attention rests, in part, upon the construct of the Temporal Organization of Behavior [18], i.e., within the total time necessary to bring a complex test to fruition, it is possible to disambiguate three separate but highly integrated neurocognitive operations involving how well the mental set for the task at hand is established; how well the mental set is maintained; and the emergence and production of errors such as perseverations.

Thus, in the context of the Theory of Executive Attention, working memory is defined as the ability to retrospectively reclaim items from past experience. In this context, working memory can be understood as attention focused on the internal representation of the task at hand. It is here that mental set is initially established. Preparatory set sustains the necessary mental set for an expected response contingent on previous events and information from working memory [18]. The construct of ‘preparatory set’ is comprised of the prospective intention(s) and associated behaviors to respond as requested. In this context working memory and preparatory set work in tandem, i.e., operations related to working memory ‘look back’ or draw upon prior experience to help establish the appropriate context to respond as requested; while operations related to preparatory set ‘look forward’ and marshal the necessary neurocognitive resources to maintain the necessary mental set to complete the task as requested. Finally, inhibitory control is a construct that operates to suppress extraneous internal and external inputs that can derail the task at hand such as the emergence of perseveration and other errors. Although these three constructs are associated with unique, separate underlying neurocognitive abilities, their operations are best understood as a unified gestalt.

Using paper and pencil neuropsychological tests, Lamar and colleagues [19] provided clinical evidence illustrating how constructs from The Theory of Executive Attention can contextualize dysexecutive behavior in patients with dementia. Patients diagnosed with AD and VaD associated with WMAs were studied with the Boston Revision of the Wechsler Memory Scale Mental Control subtest. Performance was divided into three time-based test epochs: the initial, middle, and latter responding. When performance was analyzed as a function of these three test epochs, patients with VaD produced a striking negative slope. That is, performance worsened from test epoch 1, to test epoch 2, to test epoch 3. Also, during the latter two-time epochs there was an accumulation of perseverations and errors of omission/ commission. By contrast, patients with AD declined from test epoch 1 to test epoch 2; however, there was no further decline. Eppig and colleagues [20] documented similar behavior when patients with dysexecutive MCI were compared to other MCI subtypes. With respect to the Theory of Executive Attention, the pattern of performance described above suggest significant difficulty regarding preparatory set, i.e., the ability to maintain the necessary mental set for the task at hand; and inhibitory control, i.e., the emergence of errors and perseverations suggesting a de-railed ability to suppress extraneous internal and external inputs.

3. Latency Parameters from Executive and Verbal Episodic Memory Tests

3a.). Latency and The Backward Digit Span Test.

Recently, Emrani and colleagues [21,22] leveraged constructs from the Theory of Executive Attention to define behavior from discrete time epochs and suggest how these time-based parameters could be related to emergent neurocognitive decline seen in patients diagnosed with MCI. In a series of experiments, Emrani and colleagues used an iPad to administer and score the Backward Digit Span Test (BDST; refs 23,24] and other tests to memory clinic patients. This test was able to measure the latency or reaction time for all emitted responses. Using classification procedures suggested by Jak, Bondi, and colleagues [25,26] memory clinic patients were characterized as presenting with MCI and were compared to a group of patients who did not meet criteria for MCI (i.e., non-MCI). Across both groups, only correct BDST test trials were examined. With respect to latency, several interesting observations were noted. On the 5-span trial block, no between-group differences for average time or latency for the entire trial block were observed. However, in the context of correct responding, differences emerged with respect to how time was distributed across all five emitted responses.

As seen in Figure 1, both groups produced a comparatively lengthy latency before repeating the first digit backward; i.e., approximately 2 seconds. Between-group differences emerged for the latency to repeat the second digit backward. Here, non-MCI patients exhibited a slower latency compared to their MCI counterparts. Between-group differences were also found for the latency to repeat the third digit backward. Now, MCI patients generated a slower latency. Within the context of The Theory of Executive Attention, a slower latency early in the test trial exhibited by non-MCI patients could signal a greater or more efficient capacity to marshal and deploy the necessary neurocognitive resources to establish and maintain the necessary mental set for digit reordering. By contrast, a slower latency toward the middle of the test exhibited by MCI patients could be a marker for subtle impairment for digit re-ordering such that MCI patients may have required greater time to put into place the necessary neurocognitive resources for correct responding.

Figure 1:

Serial Order Correct Trials (95% confidence interval; standard error)

A ratio was calculated comparing latency for the first versus the third digit backwards. Using this metric, MCI patients required approximately 0.75 more seconds than their non-MCI counterparts to generate a correct response. These data suggest that despite a summary score that is 100% correct, the variety of neurocognitive constructs that likely underlie a total summary score might be disambiguated via an analysis of latency or time-based parameters.

3b.). Latency and Delay Recognition Memory

Verbal serial list learning is a common method to assess the capacity for new learning and the retention of information after a delay. Tests such as the California Verbal Learning Test [CVLT, 27], the Rey Auditory Verbal Learning Test [RAVLT; 28], the Hopkins Verbal Learning Test [HVLT, 29], and the Philadelphia (repeatable) Verbal Learning Test [PrVLT; 30] are commonly used for research and clinical assessment. Common to all of these tests is an assessment of delayed recognition memory whereby all target words are presented along with recognition foils, and participants are asked to respond ‘yes’ for items on the target list or ‘no’ to non-target items.

Behavior not generally analyzed is the latency to generate a recognition yes/ no response. Libon and colleagues [31] assessed memory clinic patients using a brief, digitally administered and scored, neuropsychological protocol. Cluster analysis using digitally obtained outcome measures classified participants into groups suggesting normal cognitive abilities (NL), amnestic mild cognitive impairment (aMCI), dysexecutive mild cognitive impairment (dMCI), and dementia. Memory was assessed with a 6-word version of the P(r)VLT. Behavior from the recognition test was assessed with 6 test trials where each original free recall target word was displayed along with a generic semantic foil (i.e., “wrench, pear”), and a prototypic semantic foil (i.e., “hammer, apple”). Consistent with the research methods of Emrani and colleagues [21], latencies to respond to correct and incorrect recognition choices were obtained. No between-group differences were seen for incorrect recognition responding, suggesting, perhaps, that different neurocognitive constructs could be associated with correct versus incorrect responding.

By contrast, for correct recognition responding, patients with normal cognitive abilities and dMCI did not differ in overall accuracy but responded faster compared to aMCI and dementia patients. Moreover, in subsequent regression analyses Libon and colleagues [31] found that P(r)VLT slower recognition latency for correct responses was highly associated with both episodic memory and executive/ language-related measures, including the production of fewer delay free recall cluster responses; endorsing a greater number of prototypic, but not generic recognition foils; and reduced output on the ‘animal’ fluency test [32].

These data, along with the latency data from Emrani and colleagues [21,22] and Libon and colleagues [12–14] suggest that correct responding on selected neuropsychological test conditions is not necessarily a binary, i.e., correct or incorrect variable. Rather, latency for correct verbal serial list learning recognition test items, along with other serial list learning parameters, could form the basis for novel neurocognitive biomarker(s) potentially signaling an early onset of MCI or a dementia related illness. In other words, the time it takes to correctly respond, recruit the necessary brain regions or strategies for efficient correct responding, may provide rich information regarding the probability of the eventual emergence of serious cognitive decline.

4.). Latency and Visuoconstruction

4a.). Intra-Component Latency and the Digital Clock Drawing Test.

The Clock Drawing Test (CDT) is a widely used neuropsychological test [33–37]. This test consists of two separate, but highly related test conditions; clock drawing to command and clock drawing to copy. Many researchers and clinicians follow Kaplan’s suggestion to ask that the clock hands be set to ‘ten after eleven’. The Digital Clock Drawing Test (DCTclock^™) can now be administered using an iPad [12]. Some of the more interesting behaviors to have emerged using the DCTclock include a variety of time-based or latency parameters. For example, when looking at the total clock drawing time to completion, a number of discrete, underlying higher-order decision making (“think time”) or intra-component latencies can be measured [38].

In an analysis of digital clock drawings from Framingham Heart Study participants, Piers and colleagues [39] noted that these intra-component latencies occur at various inflection points as participants transition from one portion to the next portion of their clock drawing. For example, Post-Clock Face Latency (PCF-L) calculates the time between the final pen stroke used to drawn the clock face and whatever is drawn next; the Pre-First Hand Latency (PFH-L) measures the time between the end of the pen stroke for what was drawn immediately before the first clock hand and the beginning of the first clock hand pen stroke; and, similarly, the Pre-Second Hand Latency (PSH-L) is the time between the end of the last stroke from what was drawn before the second clock hand and the beginning of the pen stroke used to draw the second clock hand.

Dion and colleagues [40] observed slower command and copy PCF-L and PFH-L among their sample of MCI patients; moreover, these slower intra-component latencies were negatively associated with performance on tests measuring working memory (WAIS-III Letter Number, Sequencing, WAIS-III Digits Backwards); graphomotor information processing speed (WAIS-III Digit Symbol, Stroop Color Word- reading test condition); and language (Boston Naming Test, ‘animal fluency’). Dion and colleagues [41] also analyzed clock drawing intra-component latencies obtained from patients with PD with greater dysexecutive versus amnestic neuropsychological profiles. PD-dysexecutive patients continued to present with slower PCF-L.

4b.). Digital Clock Drawing and AD Polygenic Risk.

In an analysis of DCTclock drawing data from over 2,000 participants in the Framingham Heart Study, Thompson and colleagues [42] examined associations between DCTclock index scores and AD polygenic risk scores. They found that an information processing speed composite score, capturing latencies, pauses, and relative time spent ‘thinking’ (not drawing) versus ‘inking’, (actively drawing), was uniquely associated with greater AD genetic risk even when the sample was restricted to only cognitively healthy older adults. These findings lend support for the prospect of using processing-speed-related digital markers to capture some of the earliest AD-related neurocognitive changes affecting aging adults as well as identifying points of neurocognitive variation that exist on the basis of genetic risk factors for AD throughout adulthood.

4c.). DCTclock Latency and MCI.

An example, where measures associated with clock drawing latency could suggest emergent neurocognitive difficulty is displayed in Figure 2. This memory clinic patient was ultimately diagnosed with subtle cognitive impairment [43] using standard paper and pencil neuropsychological testing. As displayed in Figure 2, both command and copy drawings appear to be relatively intact with respect to the production of pertinent clock components, i.e., round clock faces were drawn, numbers inside the clock faces are more or less correctly positioned, and the representation of the requested time is relatively intact.

Figure 2:

72-year-old patient; MMSE= 29; diagnosis- subtle cognitive Impairment (Edmunds et al., 2015)

Command – Percent Ink Time: The percentage of the test time spent actively drawing with the pencil on the iPad. MoCA 3/3, correct; Libon et al., (1996) 1 error

Copy: average speed: The average speed of the pencil for all pencil strokes used during the drawing of the clock face. MoCA 3/3, correct; Libon et al (1996) 2 errors

However, indices from the DCTclock found subtle motor-related difficulty. Thus, in the command test condition the Simple/ Complex Motor Index was at the 10^th percentile due to slow time to completion. In the accompanying copy test condition, the Simple/ Complex Motor Index score continued to be below the first percentile, also suggesting slow time to completion. By contrast, The Spatial Reasoning Index quantifies how well the internal clock components are placed. As displayed in Figure 2, these indices were intact (command Spatial Reasoning; 58^th percentile; copy Spatial Reasoning; 24^th percentile). The motor-related parameters described above constitute another set of potentially interesting neurocognitive biomarkers that could flag emergent illness. These data are consistent with the results reported by Emrani and colleagues [19] and clearly show that an overall summary score is not monotonic, but comprised of a number of discrete, neurocognitive operations.

4d.). Relations Between P(r)VLT Recognition Memory and Clock Drawing Latencies.

Subsequent research analyses [31] were undertaken to assess possible relations between P(r)VLT recognition latency for correct responses and DCTclock latency and constructional measures. A series of regression analyses were performed whereby P(r)VLT recognition latency for correct responses was the dependent variable.

DCTclock independent variables included (1) the mean latency for pauses between the production of pen strokes, (2) and the average variability latency for pauses also between the production of pen strokes. Both of these metrics assess ‘think’ or non-drawing time, as described above. Additional DCTclock measures included (3) the average speed with which pen strokes were drawn (i.e., drawing or ‘ink’ time), and (4) the total number of pen strokes to produce the clock. Separate analyses were obtained for command versus copy test behavior and were controlled for age, sex, and education. These analyses revealed faster P(r)VLT recognition latencies in relation to shorter or less DCTclock non-drawing time, but only in the copy test condition. A similar analysis using both the ‘ink’ or drawing time, and the constructional metrics described above, i.e., total pen strokes were not significant for either the command or copy test conditions.

These findings raise several interesting considerations for interpretation. For example, if P(r)VLT recognition latency for correct responding was associated with all of the DCTclock metrics described above, it might be reasonable to conclude that all of these parameters are related to a single, gross underlying neurocognitive construct. However, the observation that faster P(r)VLT recognition latency for correct responding appears to be exclusively related to shorter DCTclock non-drawing time, only in the copy test condition suggests a more nuanced explanation.

Prior clock drawing research has found that among selected groups of dementia patients, improvement from the command to copy test conditions was associated with relatively less dysexecutive impairment revolving around a better capacity to alter the necessary mental set as one transitioned from the command to the copy test condition [37,44–45]. Also, as reported above, Libon and colleagues [31] found that faster P(r)VLT recognition latency for correct responding was associated with better executive test scores. Thus, better executive abilities might underlie the relationship described above between recognition latency and selected DCTclock latency measures. Further research is necessary to ensure the validity of these findings. Still, the relationships among latency metrics drawn from these tests could provide an additional neurocognitive biomarker to flag emergent neurodegenerative illness.

5.). Graphomotor Information Processing Speed and Functional Abilities

Evidence to suggest the presence of a decline in everyday, functional abilities or instrumental activities of daily living [IADL; 46], such as taking medication properly, understanding financial matters, shopping independently, and driving an automobile, needs to be present before a dementia illness such as AD can be diagnosed. Recent research suggests that problems with IADL activities can be particularly associated with dysexecutive difficulty as previously assessed with paper and pencil tests [47–48]. Indeed, in recent research with memory clinic patients diagnosed with MCI, Libon and colleagues [49] also found that IADL abilities as measured with the IADL-C, an informant-based IADL questionnaire was highly associated with dysexecutive difficulty. Among the executive tests examined by Libon and colleagues [50] was the classic paper and pencil Trail Making Test- Part B (TMT- B; refs 51,52) where total time to completion was longer in relation to greater informant-based IADL difficulty.

In subsequent research, Libon and colleagues [13] collected data from a sample of community-dwelling participants along with their study partners. These research participants were assessed with a digital version of the Trail Making Test-Part B (dTMT-B). Study partners were queried regarding IADL activities with the Functional Activities Questionnaire (FAQ). Similar to prior research with the DCTclock, dTMT-B total time to completion was fractionalized into a number of time-based outcome variables. Several of these comparatively new dTMT-B outcome variables have been described by Fellows and colleagues [53]. Libon and colleagues were particularly interested in putative relations between subtle to mild IADL difficulty and these new dTMT-B outcome measures. As such, three FAQ groups were constructed. An FAQ group with a score of 0 suggests unimpaired informant-based IADL abilities; an FAQ score between 1–4 could suggest only subtle IADL decline; and an FAQ score between 5–8 might suggest greater, but nonetheless, mild IADL decline.

Compared to the FAQ-unimpaired group, other groups required greater time or pauses inside target circles and completed their tests with more total pen strokes. The FAQ-subtle group needed more time to complete the entire test; and the velocity of their lines drawn connecting successive target circles was slower compared to the FAQ-unimpaired group. Lines connecting successive circle targets were less straight among FAQ-mild, compared to FAQ-unimpaired participants. A stepwise nominal regression analysis (reference group= FAQ-unimpaired) found that pause duration inside target circles was able to correctly classify participants into their respective groups.

6.). Clinical Decision Making

Additional considerations concern how digitally obtained latency outcome measures are used for clinical decision making. For example, in a primary care setting, where the patient is, say, only 60 years old, and is well known to the attending physician, slow responding could be caused by several problems such as psychological upset, poor medication management, or some kind of medical/ metabolic disorder. A nascent dementia may not be the first rule out. On the other hand, if the patient is seen in memory clinic where the family has expressed concern about declining neurocognitive abilities, and, say, the MRI study of the brain has disclosed some positive findings, the same pattern of performance might trigger a different care path, perhaps, involving serum and imaging analysis of dementia related proteins.

The analysis of correct responding in relation to latency has implications for other psychiatric and neurologic illnesses such as major depression, multiple sclerosis (MS), or epilepsy. Again, if the patient is well known to the attending physician and a baseline regarding neurocognitive test performance has been established, a disparity between correct responding and the pattern of accompanied latencies could be the first signal that a treatable medical problem is emerging.

7.). Summary and Conclusions

‘Precision neurocognition’ is a heuristic that seeks to deploy new methods that can operationally-define subtle neurocognitive alterations that might predict the emergence of MCI and dementia. The heuristic of ‘precision neurocognition’ is based on percepts from the Boston Process Approach to neuropsychological assessment [Kaplan; 10,11]. The latency parameters described above suggests that 100 percent responding or responding that is statistically within normal limits on neuropsychological tests, does not necessarily rule out the presence of neurocognitive difficulty. The data reviewed above look beyond the level of our typical binarized system for neuropsychological test interpretation, and suggest that a mismatch between correct summary scores, and the latencies to generate a response could constitute a portfolio of neurocognitive biomarkers for emergent illness.

We acknowledge that a number of factors must be considered in future research regarding the validity of discrete time-based outcome measures as a neurocognitive outcome measure for MCI and dementia. For example, insights in lifespan cognition research have shown that time-based variables are highly related to effects of normal age-related changes [54]. As digitally derived, time-based outcomes are developed, the effects of normal aging need to be considered. Also, the data described above clearly suggest that many neuropsychological outcome measures are not monotonic. Item Response Theory (IRT) describes a variety of statistical modeling techniques that seek to understand the relationship between a latent trait, i.e., a presumed neurocognitive ability, and how this trait or neurocognitive ability can be measured from test outcome variables. The application of IRT modeling to the latency data described above could provide greater insight regarding when and which time-based parameters are most sensitive to early, emergent dementia.

The current research has put forth the notion of ‘precision neurocognition’ as a heuristic that seeks to develop new methods for the detection of emergent MCI and dementia. As such, we feel that the heuristic of Precision Neurocognition is essentially agnostic with respect to methodology. In this context, we readily acknowledge that the goals of precision neurocognition could be realized using different methodologies and research strategies. For example, intra-individual variability (IIV) describes within person, trial-to-trial fluctuations or variations in reaction time [55]. Increasing IIV has been shown to be present along with significant MRI WMAs [56] and in patients with AD [57]. These findings are akin to some of the data reported above. Extracting IIV outcome measures with digital assessment could be another powerful tool to flag emergent dementia illness.

Additional research methods that could be used to achieve the goals of precision neurocognition include the analysis of speech/ prosody. This research has demonstrated that lower scores on neuropsychological tests that assess episodic memory and selected executive functions are associated with alterations in prosodic features including reduced speech duration, fundamental frequency (pitch), and variations in signal frequency (jitter) and signal amplitude (shimmer), as well as other acoustic measures [58,59]. Additionally, these and other voice metrics appear to be able to differentiate between cognitively normal and MCI subtypes [60,61].

Giovannetti and colleagues [62] have pioneered research that seeks to develop a theoretical and empirical framework to understand how MCI and dementia can be assessed in the context of declining everyday functioning. In a series of studies, these researchers have demonstrated how neurocognitive abilities can be revealed by analyzing the behavior of patients with MCI and dementia as they use common, everyday objects to carry out routine tasks. More recently, Hackett and Giovannetti [63] have suggested that early, emergent, suspected dementia may be flagged with an analysis of the variability by which everyday tasks are carried out.

The validity of the time-based empirical and theoretical suppositions described above require additional research. Longitudinal research is clearly needed to truly assess how well digitally obtained, time-based parameters can operationally-define patterns of neurocognitive behavior that might meaningfully predict the eventual emergence of a dementia illness. Despite these challenges, digital assessment technology clearly deserves a seat at the table along with other existing technologies to help mitigate the malignant impact of AD and other dementia illnesses.