Detecting early stages of Alzheimer’s disease using a web-based cognitive battery

Abstract

Portable and efficient cognitive screening measures are needed to address the growing need for effective early detection of Alzheimer’s disease. The Neurotrack Cognitive Battery (NCB) offers an appealing, web-based application that may be sensitive to early cognitive changes associated with Alzheimer’s disease. The NCB contains measures that were conceptually derived from animal lesion studies. The current study sought to investigate the construct and diagnostic validity of the NCB among those with and without mild cognitive impairment (MCI). Participants (n = 47) with and without MCI were administered the NCB and traditional cognitive tests. Three of six NCB measures assessing domains of memory, processing speed, and executive functioning demonstrated moderate to strong associations with well-established cognitive performance tests. In classifying those with and without MCI, sensitivities of these three NCB measures ranged from 0.47 to 0.74, and specificities ranged from 0.78 to 1.00. For traditional cognitive measures, sensitivities ranged from 0.74 to 0.84, and specificities ranged from 0.74 to 0.94. Overall, web-based cognitive test measures pertaining to cognitive domains of memory, processing speed, and executive functioning may serve as highly portable screening tools for detecting the early stages of Alzheimer’s disease. Additionally, these cognitive domains may be valuable in informing back-translational research.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-024-01496-3.

Introduction

As the older adult population continues to grow, the incidence of Alzheimer’s disease and related dementias (ADRD) is increasing [1]. Early detection is valued by patients and family members [2, 3] and is becoming increasingly important within the context of emerging disease-modifying medications [4, 5]. Neurocognitive testing is critical in the diagnosis of ADRD and is sensitive to early disease stages, such as mild cognitive impairment [6]. Further, testing is repeatedly shown to be predictive of those at increased risk of converting to dementia [7–10]. However, traditional cognitive testing is both resource and time-intensive. Computerized cognitive assessment batteries are vital to address growing population needs.

The Neurotrack Cognitive Battery (NCB) is an exceptionally portable computerized cognitive assessment tool that allows for remote evaluation [11]. It consists of six individual tests assessing a wide range of cognitive functions, including attention, processing speed, memory, inhibition, and executive functioning. One of the memory tests implements eye-tracking measurement via webcam, which has demonstrated good concordance with commercial eye-tracking technology [12–14]. This test was conceptually derived from non-human primate lesion research that demonstrated a similar measure of being sensitive to hippocampal damage [15]. Test scores for the NCB are stable, with test–retest correlations ranging from 0.72 to 0.83 [16]. Preliminary studies on individual measures and composite scores are promising [11, 17]. An NCB test measuring processing speed demonstrated convergence with a well-established traditional cognitive symbol-digit coding test in a sample of healthy adults [17]. Further, in a sample of Japanese older adults with and without dementia, several NCB tests demonstrated convergence with well-established cognitive screening tests and accurately differentiated dementia from healthy controls [11]. Remarkably, a composite of NCB tests demonstrated perfect sensitivity and specificity in that sample [11].

The purpose of this study was to evaluate the construct and diagnostic validity of the NCB in a sample of older adults with and without MCI. The sample was specifically selected to exclude non-Alzheimer’s disease explanations for MCI. We hypothesized that individual NCB measures would demonstrate acceptable convergent and discriminant validity with well-established traditional cognitive tests. We also hypothesized that the NCB would adequately classify healthy control and MCI participants. Lastly, we compared the effectiveness of classification between the NCB and traditional cognitive tests.

Method

Participants (N = 47) were recruited from an outpatient neurology clinic, outpatient neuropsychology clinic, and community advertisements. Clinical Dementia Rating (CDR) scale values were assigned by a neurologist or clinical neuropsychologist. Participants were divided into mild cognitive impairment (MCI) and health control (HC) groups based on CDR of 0.5 and 0, respectively. Six HC participants recruited from the community scored between 25 and 18 on the Montreal Cognitive Assessment (MoCA) and were re-assigned to the MCI group. One MCI participant assigned CDR 0.5 from a neurology office visit scored 30 on the MoCA and was re-assigned to the HC group.

The included participants were aged 60 to 90 and had no contraindications for completing brain MRI. Participants were excluded for CDR > 0.5 or likely non-Alzheimer’s disease conditions such as Parkinson’s disease or Parkinson-plus syndrome, prion disease, frontotemporal disease, imaging evidence of hemorrhage, cortical infarcts, hydrocephalus, or brain tumor. Participants were also excluded for medical conditions that could interfere with study procedures or confound cognitive testing results such as congestive heart failure, unstable angina, moderate to severe renal or liver impairment, uncontrolled hypertension (systolic/diastolic blood pressure measurements of 165/100 three times or more), epileptic seizures within the last 10 years, cancer diagnosis (excluding non-melanoma skin cancer) within the last 5 years, or unstable depression or anxiety. Two participants were excluded from the MCI group. One participant scored 13 on the MoCA, which estimated CDR > 0.5. The second participant was excluded for suspected Parkinson-plus syndrome based on the development of tremors and bradykinesia as assessed by a neurologist. With exclusions, the total number of participants included 23 HC and 22 MCI participants. Demographics for the sample are presented in Table 1.

Table 1.

Sample demographics

	Healthy control (N = 23)	Mild cognitive impairment (N = 22)	t or χ2	p
Age M (SD)	67.35 (5.67)	70.18 (5.59)	− 1.68	0.099
Education M (SD)	17.30 (2.51)	15.77 (2.96)	1.87	0.068
Gender			0.02	0.879
Men	12	10
Women	11	12
Race/ethnicity			0.002	0.963
White	21	20
Other	2	2

Measures

Neurotrack cognitive battery (NCB)

The NCB comprises six individual computerized subtests measuring cognitive domains of memory, attention, processing speed, and executive functioning (Table 2). Both memory and executive functioning are measured by two subtests. Below is a brief description of each measure. More detailed descriptions have been previously reported [11].

Table 2.

Convergent partial Spearman associations between Neurotrack and traditional cognitive tests controlling for age and education

Neurotrack measure	Convergent measures	N	rs	p
Image pairs	WMS-IV VR-I	41	0.34	0.032
	RAVLT trials 1–5	41	0.26	0.108
	WMS-IV VR-II	41	0.07	0.684
	RAVLT delayed recall	41	0.29	0.067
Item price	WMS-IV VR-I	40	0.33	0.045
	RAVLT trials 1–5	40	0.55	< 0.001
	WMS-IV VR-II	40	0.44	0.006
	RAVLT delayed recall	40	0.50	0.001
Symbol match	SDMT written	43	0.39	0.013
	SDMT oral	43	0.52	< 0.001
Light reaction	Stroop color-word	39	0.37	0.026
	TMT-B	39	− 0.28	0.100
Arrow match	Stroop word	39	0.14	0.395
	Stroop color	39	0.02	0.912
	TMT-A	39	0.22	0.192
Path points	Stroop color-word	42	0.48	0.002
	TMT-B	42	− 0.47	0.002

r_s Spearman correlation, WMS-IV Weschler memory scale fourth edition, VR-I/II visual reproductions immediate and delayed recall, RAVLT Rey auditory verbal learning test, SDMT symbol digit modalities test, TMT trail making test

Image Pairs is a measure of memory that displays 10 pairs of images. There are four phases to this subtest: (1) learning trial of the image pairs, (2) eye-tracking measurement of gaze duration towards a novel image, (3) second learning trial of initial pairs and novel images, and (4) recognition test of target and foil images, with targets including initial images or novel images presented in Phases 2 or 3. Scores on this measure are a combination of gaze duration toward the novel image in Phase 2 and correct responses in Phase 4.

Item Price is a measure of memory that displays a series of common supermarket items paired with prices for 4 s, and examinees then immediately respond to a recognition trial of target and foil time-price pairs. They are provided 4 s to respond. Scores on this measure are the number of correct responses, with non-responses counted as incorrect.

Arrow Match is a measure of attention that has the examinee indicate whether an arrow in the center of a series of arrows is pointing in the same or opposite direction of adjacent arrows. The score is a composite of response accuracy and speed.

Symbol Match is a measure of processing speed. Examinees are presented with a key of symbols and numbers, with two symbols matching the number “7.” Symbols are presented, and examinees are instructed to respond if the symbol matches “7” or not. Scores on this measure are the number of incorrect responses subtracted from correct responses.

Light Reaction measures the inhibition aspects of executive functioning and is a version of a “go/no-go” test [18]. Examinees are show trials of “red” and “green” lights. They are instructed to respond when they see a “red” light and not respond when they see a “green” light. Scores on this measure are the reaction time in milliseconds on “go” trials.

Path Points is a measure of executive functioning and is a version of a trail-making test. Examinees used a computer mouse to alternate letter and number sequencing. The score on this measure of total completion time.

Traditional cognitive tests (TCT)

The TCT battery was selected based on well-established tests measuring domains of memory, attention, processing speed, and executive functioning. Measures of language and visuoperception were also included for descriptive purposes and were not used to evaluate the validity of the NCB. Memory was measured using the Rey Auditory Verbal Learning Test (RAVLT) and Weschler Memory Scale, Fourth Edition Visual Reproductions subtest (WMS-IV VR) immediate and delayed recall trials. Published regression-based norms were used for the RAVLT [19], and manual norms were used for the WMS-IV VR. Attention was measured using the color and word trials of Golden’s version of the Stroop Color and Word Test (SCWT) and the Trail Making Test part A (TMT-A). Manual norms were used for the SCWT. Heaton norms were used for TMT-A [20]. Processing speed was measured using written and oral trials of the Symbol Digit Modalities Test (SDMT). Published regression-based norms were used. Executive functioning was measured using the Trail Making Test part B (TMT-B) and the color/word interference trial of the SCWT. Heaton norms were used for TMT-B [20] and manual norms were used for the SCWT. Additional cognitive performance tests were Judgement of Line Orientation, Boston Naming Test, Controlled Oral Word Test, and Animal Naming Test.

Emotional functioning

The Geriatric Depression Scale and Geriatric Anxiety Scale were administered to investigate the possible effects of emotional distress on cognitive test performances.

Procedure

Data were drawn from a larger study that collected neuroimaging data to assess multimodal measurement of cognitive decline in Alzheimer’s disease. Study procedures were approved by the University of Oklahoma Health Sciences Center Institutional Review Board. Participants provided written informed consent. Participants for this study also included a sample of patients diagnosed with dementia, which were not included in this study. All participants provided informed consent and study procedures were approved by an institutional review board. NCB and TCT measure administration was counterbalanced. NCB tests were administered in a private room on a desktop computer. Instructions for NCB measures were automated and included text instructions and sample trials. Sample test stimuli are displayed in Supplementary Fig. 1. Measures were administered by a trained research psychometrist supervised by a board-certified clinical neuropsychologist.

Data analysis

Correlations between NCB measures and TCT measures assessed convergent and discriminant validity. Sporadic missing values were present on the NCB measures, and pairwise deletion was employed. Due to several significant associations between demographics and NCB measures (presented in Supplementary Table 1), convergent and discriminant associations were assessed with partial correlations controlling for age and education. Convergent associations are presented in Table 2, and discriminant associations are presented in Table 3. Discriminant associations were only assessed for NCB measures that demonstrated evidence of convergence with TCT measures. Spearman correlations were calculated to minimize the effects of outliers [21].

Table 3.

Discriminant partial Spearman associations between Neurotrack and traditional cognitive tests controlling for age and education

Neurotrack measure	Discriminant measures	N	rs	p
Item price	Stroop word	41	0.23	0.157
	Stroop color	41	− 0.01	0.924
	TMT-A	41	− 0.14	0.380
	Stroop color-word	41	0.21	0.202
	TMT-B	41	− 0.06	0.739
Symbol match	WMS-IV VR-I	42	0.41	0.009
	RAVLT trials 1–5	42	0.22	0.171
	WMS-IV VR-II	42	0.27	0.100
	RAVLT delayed recall	42	0.21	0.186
Path points	WMS-IV VR-I	41	0.52	< 0.001
	RAVLT trials 1–5	41	0.41	0.010
	WMS-IV VR-II	41	0.57	< 0.001
	RAVLT delayed recall	41	0.39	0.013

r_s Spearman correlation, WMS-IV Weschler memory scale fourth edition, VR-I/II visual reproductions immediate and delayed recall, RAVLT Rey auditory verbal learning test, SDMT symbol digit modalities test, TMT trail making test

Group differences in NCB and TCT scores are summarized in Tables 4 and 5. Standard t-tests, as well as 20% trimmed bootstrapped t-tests, were calculated simultaneously to evaluate mean differences between the groups. Diagnostic validity was assessed using logistic regression and ROC analysis. Candidate NCB and TCT measures were selected from those that demonstrated statistically significant differences between the HC and MCI groups. Logistic regression analyses were conducted to evaluate the ability of each candidate measure to classify participants into the HC and MCI groups. Composite scores were calculated both for the NCB and TCT candidate measures to compare the overall ability of the NCB and TCT tests to classify the groups. Composite scores were the average of the candidate measure z-transformed raw scores. ROC analysis was conducted based on each logistic regression. Significance was set at < 0.05 for all analyses.

Table 4.

Descriptive statistics for Neurotrack tests

	Healthy control		Mild cognitive impairment		t	p	t (20% trim)	p
	M (SD)	N	M (SD)	N
Image pairs	0.40 (0.14)	23	0.33 (0.19)	19	1.22	0.233	1.27	0.065
Item price	0.66 (0.13)	22	0.56 (0.16)	19	2.35	0.025	1.86	0.032
Symbol match	24.22 (5.59)	23	18.40 (6.65)	20	3.08	0.004	3.04	0.002
Light reaction	− 556.68 (84.70)	22	− 634.76 (130.31)	17	2.15	0.041	1.81	0.095
Arrow match	1.14 (0.32)	21	0.96 (0.31)	18	1.78	0.083	1.87	0.072
Path points	− 16.18 (3.13)	23	− 24.84 (13.85)	19	2.67	0.015	2.67	0.012

t (20% trim) = bootstrapped (n = 499) t-test with 20% trim

Table 5.

Descriptive statistics for traditional cognitive tests

	Healthy control M (SD)	Mild cognitive impairment M (SD)	|t|	p	|t| (20% trim)	p
WRAT-IV read (stnd. score)A	121.18 (15.78)	113.64 (14.38)	1.66	0.105	1.86	0.068
MoCA (raw score)	25.96 (2.64)	22.64 (2.11)	4.65	< 0.001	3.77	0.002
Processing speed/attention
TMT-A (t-score)ASE	47.96 (9.88)	46.23 (9.53)	0.59	0.553	0.54	0.568
Stroop word (t-score)AE	38.83 (14.51)	39.09 (12.41)	0.07	0.948	0.49	0.624
Stroop color (t-score)AE	39.61 (12.71)	37.59 (12.65)	0.53	0.596	0.17	0.883
SDMT written (T-score)ASE	53.22 (11.10)	42.34 (12.76)	3.04	0.004	3.32	0.003
SDMT oral (t-score)ASE	49.04 (11.82)	39.26 (10.17)	2.98	0.005	2.59	0.003
Executive functioning
Stroop color-word (t-score)AE	44.48 (8.44)	42.73 (10.86)	0.60	0.551	0.56	0.539
TMT-B (t-score)ASE	50.87 (10.14)	49.80 (15.62)	0.35	0.731	0.26	0.813
Memory
WMS-IV VR-I (scaled score)A	11.74 (2.47)	9.05 (3.24)	3.12	0.003	3.45	< 0.001
WMS-IV VR-II (scaled score)A	12.48 (2.41)	8.50 (3.88)	4.11	< 0.001	4.24	< 0.001
RAVLT 1–5 (t-score)ASE	50.70 (9.20)	41.14 (15.29)	2.52	0.016	2.27	0.025
RAVLT delay (t-score)ASE	53.65 (8.60)	42.82 (13.19)	3.24	0.003	2.26	0.013
Language
BNT-2 (t-score)ASE	53.77 (12.00)	53.36 (11.38)	0.11	0.908	0.30	0.770
COWAT (t-score)ASE	49.74 (12.77)	46.41 (9.16)	1.01	0.319	1.14	0.229
ANT (t-score)ASE	48.57 (11.39)	41.91 (9.18)	2.16	0.036	1.48	0.101
Visuoperception
JoLO (scaled score)A	12.48 (2.68)	11.50 (3.17)	1.11	0.271	0.92	0.321
Emotional functioning
GDS (raw score)	6.48 (5.91)	9.09 (6.50)	1.41	0.165	1.64	0.114
GAS (raw score)	10.39 (7.90)	14.23 (10.24)	1.40	0.168	1.29	0.180

t (20% trim) = bootstrapped (n = 499) t-test with 20% trim

MoCA Montreal cognitive assessment, WMS-IV Weschler memory scale fourth edition, VR-I/II visual reproductions immediate and delayed recall, RAVLT Rey auditory verbal learning test, SDMT symbol digit modalities test, TMT trail making test, WRAT-4 wide-range achievement test, BNT-2 Boston naming test, second edition, COWAT controlled oral word association test, ANT animal naming test, JoLO judgement of line orientation, GDS geriatric depression scale, GAS geriatric anxiety scale

^AAge-corrected

^SSex-corrected

^EEducation-corrected

Results

Convergent and discriminant validity

Item Price, Symbol Match, and Path Points demonstrated consistent moderate to large convergent associations with TCT measures (Table 2). Image Pairs demonstrated a moderate association with the WMS-IV Visual Reproductions immediate recall trial. However, no other convergent associations were significant. Light Reaction demonstrated a moderately significant association with Stroop Color-Word, but the association with TMT-B was non-significant. Arrow Match did not demonstrate significant convergent associations with TCT measures. Overall, Item Price, Symbol Match, and Path Points demonstrated good evidence of convergence. Image Pairs and Light Reaction demonstrated partial evidence of convergence, and Arrow Match demonstrated poor evidence of convergence.

Discriminant validity was evaluated for Item Price, Symbol Match, and Path Points (Table 3) as they demonstrated good evidence of convergence. Item Price demonstrated no significant discriminant associations. Symbol Match demonstrated a moderate to strong association with the WMS-IV Visual Reproductions immediate recall trial but otherwise did not demonstrate significant discriminant associations. Path Points demonstrated moderate to strong associations with all TCT discriminant measures. Overall, Item Price and Symbol Match demonstrated adequate discrimination based on consistently higher convergent than discriminant associations. Path Points demonstrated good convergence and poor discrimination.

Image Pairs and Symbol Match demonstrated the strongest evidence for construct validity, based on patterns of convergent and discriminant associations with traditional cognitive tests. Path Points demonstrated good evidence of convergence, but poor evidence of discrimination due to moderate to strong associations with memory measures. There was overall poor evidence for construct validity for Image Pairs, Light Reaction, and Arrow Match.

Diagnostic validity

Diagnostic validity was assessed for Item Price, Symbol Match, and Path Points as they demonstrated significantly lower scores in the MCI group (Table 4). TCT measures of memory, as well as the SDMT, were selected for comparison as the MCI group also had significantly lower scores on these tests (Table 5). Due to sporadic missing values on NCB measures, three MCI cases were excluded. HC cases were then age and education matched to these MCI cases, resulting in N = 19 for both groups. Age and education were not significantly different between these groups (p > 0.09). Composite NCB and TCT measures were computed by taking the mean of sample-referenced z-scores. Cronbach’s alpha was 0.77 for the NCB composite and 0.87 for the TCT composite. TCT Memory (α = 0.83) and Processing Speed (α = 0.90) composite measures were also calculated and analyzed.

Group classification statistics from logistic regression and receiver operating characteristic (ROC) analyses are presented in Table 6. Sensitivity and specificity were calculated at each measure’s optimal Youden index, which indicates the highest combined sensitivity and specificity. All measures adequately classified MCI and HC participants. Odds ratios and areas under the curve (AUC) of the overall and Memory TCT composites surpassed all NCB measures. All TCT measures demonstrated a good balance of sensitivity and specificity. Symbol Match and Path Points demonstrated a good balance of sensitivity and specificity. The NCB composite and Item Price demonstrated strong specificity (1.00 and 0.95, respectively) but poor sensitivity (0.56 and 0.47, respectively).

Table 6.

Classification statistics for traditional cognitive and Neurotrack tests

	Odds ratio*	p	AUC (95%CI)	Sensitivity	Specificity	Youden index
Neuropsych. composite	24.74	0.002	0.90 (0.80, 0.99)	0.84	0.83	0.84
Memory	8.80	0.002	0.88 (0.76, 0.99)	0.74	0.94	0.84
Processing speed	4.67	0.007	0.78 (0.64, 0.94)	0.74	0.74	0.74
Neurotrack composite	6.33	0.014	0.77 (0.62, 0.92)	0.56	1.00	0.78
Item price	2.18	0.041	0.70 (0.54, 0.87)	0.47	0.95	0.71
Path points	7.66	0.019	0.78 (0.63, 0.93)	0.74	0.79	0.77
Symbol match	3.35	0.017	0.78 (0.63, 0.93)	0.74	0.78	0.76

^*Scores were inverted so that higher scores were associated with greater odds of MCI group membership. Sensitivity and specificity were calculated at each measure’s Youden index

Memory: composite of WMS-IV VR I and II, RAVLT trials 1–5, and RAVLT delayed recall

Processing speed: composite of SDMT oral and written

Neuropsych composite: composite of Memory and Processing Speed measures

Neurotrack composite: composite of Item Price, Path Points, and Symbol Match

NCB measures adequately classified MCI and HC participants, with Path Points and Symbol Match demonstrating a better balance of sensitivity and specificity. NCB measures performed similarly to the TCT Processing Speed composite but were outperformed by the Memory and overall TCT composites.

Associations with demographics and anxiety and depression

Associations between demographics and emotional functioning with NCB measures are presented in Supplementary Table 1, and associations between demographics and emotional functioning with TCT measures are presented in Supplementary Table 2. Significant associations were observed with age and education, but not gender, depression, or anxiety. NCB measures did not show a stronger pattern of associations with demographic factors or emotional distress than TCT measures.

Discussion

Results showed that a test of memory, a test of processing speed, and a test of executive functioning from a novel web-based computerized cognitive battery demonstrated adequate convergent and diagnostic validity. However, another test of memory, a test of inhibition, and a test of attention did not. While none of these tests outperformed well-established cognitive performance tests in classifying those with MCI from HC, NCB measures of processing speed and executive functioning were comparable to their traditional test counterparts. Some NCB tests hold the promise of operating as effective and highly portable screening measures for the early stages of Alzheimer’s disease.

NCB measures of Item Price, Symbol Match, and Path Points demonstrated the strongest evidence of construct and diagnostic validity. Each measure showed generally consistent moderate to strong associations with well-established cognitive performance measures assessing respective domains of memory, processing speed, and executive functioning. Item Price and Symbol Match also demonstrated adequate to good evidence of discriminant validity based on smaller associations with non-domain traditional cognitive performance measures. Curiously, Path Points, a measure of executive functioning, demonstrated moderate to strong associations with all traditional memory tests. This pattern may be explained by both Path Points and memory scores being lower in the MCI group, thus inflating the relationship between these measures. It is possible that the failure of Path Points to discriminate is reflective of the pattern of cognitive impairments in early Alzheimer’s disease involving both memory and executive functioning [10]. However, processing speed is also implicated in early Alzheimer’s disease [10, 22], and Symbol Match demonstrated better discriminant validity than Path Points.

Item Price, Symbol Match, and Path Points all demonstrated adequate diagnostic validity based on logistic regression and ROC analyses. Sensitivities ranged from 0.47 to 0.74 and specificities ranged from 0.78 to 0.95. Symbol Match and Path Points demonstrated balanced sensitivity and specificity, while Item Price demonstrated very high specificity (0.95) and low sensitivity (0.47). A similar pattern of higher specificity than sensitivity was observed for traditional memory measures, but the discrepancy was not as large (sensitivity = 0.74 and specificity = 0.94). Overall, NCB measures did not outperform TCT measures in classifying those with MCI. This is not surprising as there were several more TCT measures included in composite scores, amplifying the measurement reliability of their respective cognitive domains. Notably, however, the overall classification accuracy of a brief TCT measure of processing speed was comparable to that of a composite measure of Item Price, Symbol Match, and Path Points. This suggests that this brief TCT measure of processing speed may perform similarly to well-performing NCB measures in screening for MCI. However, the NCB measures offer greater portability and clinical efficiency as they are web-based and do not require a specialty-trained examiner.

Three NCB measures demonstrated poor construct and diagnostic validity in this sample. Previous studies have shown that these measures (Image Pairs, Arrow Match, and Light Reaction) were effective in discriminating healthy controls from those with dementia in a sample of older adults in Japan [11]. It is possible that these three underperforming measures in the current sample may be effective in identifying dementia, even though they were not able to effectively classify MCI. Curiously, Image Pairs, a memory measure with an eye-tracking component, was not an effective classifier of MCI or measure of memory function in this sample. In previous samples, other versions of Image Pairs have effectively predicted conversion from MCI to dementia, shown good concordance with commercial eye-tracking technology, and can differentiate between healthy controls and those with MCI [12–14, 23, 24]. The current results may have been related to potential eye movement measurement difficulties via a webcam.

Results from this study should be interpreted within the context of some notable limitations. Healthy control selection, and in some cases CDR estimation, was based on MoCA scores. This could have led to some initial inaccurate group classifications, which then influenced diagnostic validity results for both the TCT and NCB measures. The sample, which was also relatively small, consisted of predominantly White individuals with advanced levels of educational attainment. These results may not generalize to more diverse samples. The current results would be strengthened by verification from replication in another sample. Although initial data suggest consistency in test scores of the NCB over repeated administrations [16], specific analysis is needed to address possible improvements in test scores following repeated administrations.

Overall, NCB measures of Item Price, Symbol Match, and Path Points may operate as effective screening measures for detecting MCI due to Alzheimer’s disease. Additional studies are needed to make direct comparisons between NCB measures and well-established screening measures, ideally with neuropsychologically-defined MCI diagnosis. Further, appropriate normative samples are needed before clinical interpretation from NCB test scores can be confidently made. Additionally, longitudinal research is needed to ascertain the ability of the NCB in predicting conversion to MCI or dementia.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This study was funded by a grant from the Alzheimer’s Drug Discovery Foundation (ORA00025797). CP receives grant support from the US Department of Veterans Affairs (Merit award CX000340).

Data Availability

Data from this study are available upon request for data verification purposes.

Declarations

Conflict of interest

JH, BR, CC, and JS declare no conflicts of interest. CP serves as an associate editor for GeroScience. JG is an employee of Neurotrack Technologies, Inc.; he receives a salary and holds equity in the company.