The Harvard Automated Phone Task: new performance-based activities of daily living tests for early Alzheimer’s disease

Abstract

Background

Impairment in activities of daily living is a major burden for Alzheimer’s disease dementia patients and caregivers. Multiple subjective scales and a few performance-based instruments have been validated and proven to be reliable in measuring instrumental activities of daily living in Alzheimer’s disease dementia but less so in amnestic mild cognitive impairment and preclinical Alzheimer’s disease.

Objective

To validate the Harvard Automated Phone Task, a new performance-based activities of daily living test for early Alzheimer’s disease, which assesses high level tasks that challenge seniors in daily life.

Design

In a cross-sectional study, the Harvard Automated Phone Task was associated with demographics and cognitive measures through univariate and multivariate analyses; ability to discriminate across diagnostic groups was assessed; test-retest reliability with the same and alternate versions was assessed in a subset of participants; and the relationship with regional cortical thickness was assessed in a subset of participants.

Setting

Academic clinical research center.

Participants

One hundred and eighty two participants were recruited from the community (127 clinically normal elderly and 45 young normal participants) and memory disorders clinics at Brigham and Women’s Hospital and Massachusetts General Hospital (10 participants with mild cognitive impairment).

Measurements

As part of the Harvard Automated Phone Task, participants navigated an interactive voice response system to refill a prescription (APT-Script), select a new primary care physician (APT-PCP), and make a bank account transfer and payment (APT-Bank). The 3 tasks were scored based on time, errors, and repetitions from which composite z-scores were derived, as well as a separate report of correct completion of the task.

Results

We found that the Harvard Automated Phone Task discriminated well between diagnostic groups (APT-Script: p=0.002; APT-PCP: p<0.001; APT-Bank: p=0.02), had an incremental level of difficulty, and had excellent test-retest reliability (Cronbach’s α values of 0.81 to 0.87). Within the clinically normal elderly, there were significant associations in multivariate models between performance on the Harvard Automated Phone Task and executive function (APT-PCP: p<0.001), processing speed (APT-Script: p=0.005), and regional cortical atrophy (APT-PCP: p=0.001; no significant association with APT-Script) independent of hearing acuity, motor speed, age, race, education, and premorbid intelligence.

Conclusions

Our initial experience with the Harvard Automated Phone Task, which consists of ecologically valid, easily-administered measures of daily activities, suggests that these tasks could be useful for screening and tracking the earliest functional alterations in preclinical and early prodromal AD.

Introduction

Impairment in activities of daily living (ADL) is a time-intensive, psychological, physical, and financial burden for patients with Alzheimer’s disease (AD) dementia and their caregivers. Traditionally, impairment in basic ADL, which consist of self-care activities, has been associated with moderate to severe dementia, while impairment in instrumental ADL, which consist of activities such as managing the finances, driving, cooking, shopping, and performing household chores, has been associated with mild to moderate dementia. Multiple subjective scales and a few performance-based instruments have been validated and proven to be reliable in measuring instrumental ADL in AD dementia but less so in amnestic mild cognitive impairment (MCI) and preclinical AD [1]. There is a debate in the field about whether or not mild impairment in instrumental ADL should be allowed in the diagnosis of MCI because it can blur the distinction with mild dementia [2, [3, [4]. However, that distinction is often arbitrary and since MCI, which may represent prodromal AD, could progress to AD dementia, they are on a continuum. One step back from MCI may be preclinical AD, in which asymptomatic or minimally symptomatic elderly individuals have biomarker evidence of AD pathology [5]. Like other deficits in AD, subtle difficulties in complex ADL may begin at the transition from preclinical AD to MCI. However, to date few ADL tests have been developed to capture these earliest changes.

Performance-based ADL instruments, in which individuals complete simulated or actual tasks from daily life, are thought to be more objective and ecologically valid than the more widely used subjective ADL questionnaires, in which usually informants and sometimes subjects report about their ability to perform various activities [1, [6, [7]. One of the first performance-based ADL instruments was the Direct Assessment of Functional Status, which targeted a wide range of basic and instrumental ADL in mild-moderate dementia [8, [9]. More recent instruments have focused on MCI [10, [11]. The University of California, San Diego Performance-Based Skills Assessment and the Financial Capacity Instrument (FCI) have both been shown to discriminate well between clinically normal (CN) elderly and MCI [11, [12]. However, both instruments take 30 minutes or longer to complete, require specialized tools and a trained administrator, thus limiting the extent of their use in research and clinical settings.

Recently, guidelines for the assessment of functional impairment at the preclinical AD stage were suggested [13]. In the current study, we describe a newly developed performance-based ADL instrument, the Harvard Automated Phone Task (APT), targeting individuals with preclinical AD and early prodromal AD. The telephone is still by far the most prevalent technology mode of communication in the elderly, who often need to use an interactive voice response system (IVRS) to complete everyday activities [14]. As part of the Harvard APT, participants navigate an IVRS to refill a prescription (APT-Script), select a new primary care physician (APT-PCP), and make a bank account transfer and payment (APT-Bank). Our objective was to validate the Harvard APT, a set of new performance-based ADL tests for early AD, which are quick and easy to administer and assess high level tasks that challenge seniors in daily life.

Methods

Participants

One hundred and eighty two participants were recruited from the community (consisting of 127 CN elderly) and 45 young normal (YN) participants) and memory disorders clinics at Brigham and Women’s Hospital (BWH) and Massachusetts General Hospital (MGH) (consisting of 10 MCI participants). All participants were in general good health or had stable medical problems and did not have significant psychiatric disorders.

CN participants were ages 60 to 90 years old (inclusive), had a Mini-Mental State Examination (MMSE) [15] score of 26 to 30 (inclusive), and normal memory performance (defined as a Free and Cued Selective Reminding Test (FCSRT) [16] free recall score of >24 and cued recall score of >44). YN participants were ages 18 to 27 years old (inclusive), had an MMSE score of 27 to 30 (inclusive), and normal memory performance (FCSRT free recall score of >24 and cued recall score of >44). MCI participants were ages 61 to 81 (inclusive), had an MMSE score of 25 to 29 (inclusive), and impaired memory performance (FCSRT free recall score of ≤24 and/or cued recall score of ≤44).

The study was approved by the Institutional Review Board (IRB) of Partners Healthcare Inc. Written informed consent was obtained from all participants prior to initiation of any study procedures in accordance with IRB guidelines.

Clinical Assessments

The Harvard Automated Phone Task (APT) was developed at the Center for Alzheimer Research and Treatment at BWH and MGH and the Connected Health Innovation at Partners HealthCare. As part of the Harvard APT, participants perform 3 tasks consisting of navigating an IVRS. The 3 tasks combined can be completed in about 10 minutes using any phone with buttons containing digits and letters. See Appendix for detailed participant instructions, scoring, and schemas of the 3 tasks.

Task 1 (APT-Script) requires participants to call a pharmacy and refill a prescription. Participants are given a mock pill bottle for simvastatin 20 mg orally every day, quantity 30, refills 2, and the pharmacy phone number.

Task 2 (APT-PCP) requires participants to call a health insurance company and select a new primary care physician (Dr. John Smith in Boston, MA). Participants are provided with a card with a member ID number (CDW758421693) and a member services phone number.

Task 3 (APT-Bank) requires participants to make a bank account transfer in order to have enough money to pay their Federal taxes for the year. Participants are provided with the amount of Federal taxes they owe ($4,150), the bank phone number, their checking account number, their savings account number, and a bank statement. Participants are instructed to take notes on a piece of paper in order to complete the task. Once they call in, they are told how much money they have in each account (checking: $2,050; savings: $3,950). They then make a transfer in the appropriate amount (between $2,100 and $3,700; they are told that they need to maintain a minimum of $250 in their savings account in order not to be penalized) from their savings into their checking account in order to have enough money to pay their taxes. They then make the payment.

The 3 tasks are administered in the same sequence (1 through 3) to all participants. Prior to each task, motor speed and hearing acuity are assessed on the phone in order to control for those possible confounds. The 3 tasks are scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion of task. Composite z-scores of time, errors, and repetitions are generated for each task using the mean and standard deviation of the performance of the CN participants as a reference. Individuals who do not complete a task correctly are assigned a total time equivalent to the longest total time among individuals who have correctly completed the task. This adjustment is made in order to avoid the possibility of the appearance of a better performance by individuals who prematurely completed a task incorrectly.

Upon completion of all 3 tasks, participants are asked whether these tasks relate to their daily activities on a 5-point Likert-type scale ranging from very little to very much. Then, participants are asked to rank the difficulty of each of the 3 tasks from very easy to very hard.

We also developed alternate versions of the 3 tasks with equivalent psychometric properties (e.g. same character length of physician’s name for Task 2, APT-PCP) for test-retest reliability and in order to avoid practice effects from year to year.

Other assessments used in the current study included the American National Adult Reading Test intelligence quotient (AMNART IQ) [17], an estimate of premorbid intelligence, serving as a proxy of cognitive reserve; the MMSE [15], a measure of global cognition; the FCSRT [16], a measure of episodic memory; Trailmaking Test A (TMT-A) [18], a measure of processing speed; and Trailmaking Test B (TMT-B) [18], a measure of executive function.

Magnetic Resonance Imaging (MRI) Data

MRI scans were conducted with a Siemens Trio 3T scanner (Siemens Medical Systems, Erlangen Germany) at the Charlestown Navy Yard campus of MGH. High-resolution T1-weighted structural images were acquired using a 3D Magnetization Prepared Rapid Acquisition Gradient Echo (MP-RAGE) sequence with the following acquisition parameters: repetition time=2300ms; echo time=2.98ms; inversion time=900ms; flip angle=9°; voxel size=1.0×1.0×1.2mm. Cortical thickness of regions of interest (ROI) was measured using FreeSurfer Version 5.1 (http://surfer.nmr.mgh.harvard.edu/) [19]. Four ROI chosen based on results of prior analyses of neuroimaging correlates of ADL were assessed [20, [21]: 1) inferior temporal cortex 2) posterior cingulate cortex; 3) medial orbitofrontal cortex; and 4) supramarginal cortex.

Statistical Analyses

Analyses were performed using SPSS version 22.0. Most univariate and multivariate analyses focused on the CN diagnostic group, which was the largest group and the group of main interest since the Harvard APT is primarily meant to target individuals with preclinical AD and those transitioning to prodromal AD (MCI).

The Harvard APT distribution was not normal with a negative (left) skew. Therefore, non-parametric tests were employed for univariate analyses. In CN participants, the Harvard APT was related to participant demographics and characteristics using Spearman’s correlations for continuous variables and Mann-Whitney U test for categorical variables (two-tailed p values were reported).

In CN participants, multiple linear regression models with backward elimination (cut-off p<0.05) were performed to assess the association between Harvard APT and cognitive tests, adjusting for demographics that were significant in the univariate analyses. Tasks 1 and 2 (APT-Script and APT-PCP) were the dependent variables in separate models. Predictors included TMT-A, TMT-B, age, education, and AMNART-IQ. These models were repeated in a subset of participants after adding predictors accounting for hearing acuity and motor speed, as well as self-report of difficulty of tasks. Partial regression coefficient estimates (β) with 95% confidence intervals (CI), significance test results (p values), and percent variance accounted for in the dependent variable by the model as a whole (R²) were reported.

Analysis of variance (ANOVA) was used to compare Harvard APT performance (APT-Script and APT-PCP) across diagnostic groups (YN, CN, and MCI). Comparisons of pairs of diagnoses (ex: CN vs. MCI) were also performed for which p values were corrected for multiple comparisons using Tukey post-hoc tests. For APT-Bank, which was completed only in CN and YN participants, a t-test was used to compare performance across diagnostic groups.

In CN participants, reliability of the Harvard APT (APT-Script and APT-PCP) was determined with Cronbach’s α, intraclass correlations. Test-retest was performed with a short interval using the same version, as well as with an alternate version.

In CN and MCI participants combined and then in CN participants alone, multiple linear regression models with backward elimination (cut-off p<0.05) were performed to assess the association between Harvard APT and regional cortical thickness (MRI ROI). Tasks 1 and 2 (APT-Script and APT-PCP) were the dependent variables in separate models. Predictors included the 4 MRI ROI, age, education, and AMNART-IQ. β with 95% CI, p values, R² were reported.

Results

One hundred and eighty two participants (127 CN, 45 YN, and 10 MCI) underwent APT-Script and APT-PCP and 40 of those participants (30 CN and 10 YN) underwent APT-Bank, which was developed later. See Table 1 for participant demographics and characteristics.

Table 1.

Demographics and characteristics of all participants.

	YN	CN	MCI	Correlations with Tasks in CN*
				Task 1: APT-Script		Task 2: APT-PCP
n	45	127	10	rs	p-value	rs	p-value
Age (years)	21.4±2.3	73.8±6.8	72.1±6.5	−0.22	0.02	−0.20	0.02
Sex (% Female)	71.1	72.4	40.0		0.34		0.96
Race (% White)	73.3†	75.8‡	100		0.57		0.21
Education (years)	14.4±1.6	16.4±2.6	17.0±1.6	−0.06	0.48	0.18	0.04
AMNART IQ	122.0±4.5	122.3±7.7	123.0±6.0	0.07	0.47	0.32	<0.001
MMSE	29.7±0.7	29.1±1.1	27.4±1.7	−0.12	0.16	0.16	0.07
FCSRT free recall	38.6±3.5	33.3±4.6	19.9±4.2	0.06	0.55	0.11	0.25
FCSRT cued recall	47.9±0.3	47.9±0.4	40.4±8.1	−0.11	0.21	−0.09	0.34
TMT-A (sec)	20.7±5.9	39.7±15.2	35.0±20.9	−0.32	<0.001	−0.35	<0.001
TMT-B (sec)	45.7±10.9	91.8±45.5	98.0±58.9	−0.24	0.007	−0.38	<0.001

AMNART IQ (American National Adult Reading Test intelligence quotient), APT (Automated Phone Task), CN (clinically normal elderly), MCI (mild cognitive impairment), MMSE (Mini-Mental State Exam), FCSRT (Free and Cued Selective Reminding Test), TMT (Trailmaking Test), YN (young normal).

Values represent mean ± standard deviation (except for n, Sex, and Race).

^*Mann-Whitney U test used instead of Spearman correlation for Sex and Race.

^†YN minorities: 4.4% African American, 17.8% Asian, and 4.4% Hispanic.

^‡CN minorities: 20.0% African American, 2.5% Asian, and 1.7% Hispanic.

Table 2 provides information about performance on Tasks 1 and 2 (APT-Script and APT-PCP) across diagnostic groups. Correct task completion across groups was high (98% for APT-Script and 91% for APT-PCP). The average time to complete APT-Script was about 1 minute and APT-PCP about 3 ½ minutes. The average time to complete APT-Bank was about 5 minutes (CN: 314.5±73.6 seconds; YN: 254.2±97.7 seconds). APT-Script and APT-PCP were significantly associated across groups (r_s=0.30, p<0.001). APT-Bank was significantly associated with APT-PCP (r_s=0.41, p=0.008) but not with APT-Script (r_s=0.09, p=0.57).

Table 2.

Performance on Harvard APT Tasks 1 and 2.

Phone Tasks		YN	CN	MCI
n		45	127	10
Task 1: APT-Script	Total Time (sec)	59.1±14.5	71.0±27.4	88.7±44.1
	Errors	0.3±0.6	0.4±0.8	0.5±1.1
	Step repetition	0.1±0.3	0.3±0.6	0.9±0.6
	Completed (%)	100	98.4	90.0
Task 2: APT-PCP	Total Time (sec)	154.1±56.7	211.8±107.9	276.8±120.1
	Errors	0.9±1.5	1.6±2.2	3.7±2.2
	Step repetition	0.4±0.5	0.6±0.7	0.6±0.5
	Completed (%)	97.8	89.8	77.8

APT (Automated Phone Task), CN (clinically normal elderly), MCI (mild cognitive impairment), YN (young normal).

Values represent mean ± standard deviation (except for n and Completed).

Analysis of variance (ANOVA) revealed performance differences between groups on composite z-scores (Task 1: p=0.002, Task 2: p<0.001) with MCI performing worse than CN and CN worse than YN participants.

Association with demographics and cognitive measures in CN participants

Composite z-scores for each task accounting for time, errors, and repetitions were created. As illustrated in Table 2, in unadjusted analyses, greater age was significantly associated with worse APT-Script performance (p=0.02) and worse APT-PCP performance (p=0.02), lower education and lower AMNART IQ were significantly associated with worse APT-PCP performance (education: p=0.04; IQ: p<0.001), worse processing speed (TMT-A) was significantly associated with worse APT-Script performance (p<0.001) and worse APT-PCP performance (p<0.001), and worse executive function (TMT-B) was significantly associated with worse APT-Script performance (p=0.007) and worse APT-PCP performance (p<0.001). APT-Script was not significantly associated with education or AMNART IQ. APT-Script and APT-PCP were not significantly associated with sex, race, MMSE, or FCSRT. Worse APT-Bank performance was significantly associated with lower education (p=0.003) and worse executive function (p=0.05) but not with any other demographic or cognitive variables.

The multiple linear regression model revealed a significant association between worse APT-Script performance and worse processing speed (β=−0.02, 95% CI for β=−0.03, −0.01, p<0.001). No other predictors were retained in the model (R²=0.11, p<0.001 for overall model). A separate multiple linear regression model revealed a significant association between worse APT-PCP performance and worse executive function (β=−0.01, 95% CI for β=−0.01, −0.005, p<0.001). No other predictors were retained in the model (R²=0.19, p<0.001 for overall model).

Association with cognitive measures independent of hearing acuity and motor speed

In order to correct for hearing acuity and motor speed, prior to each task participants were asked to enter 8 numbers separately and time and errors were recorded, see Appendix for details. Data from 55 participants (34 CN, 20 YN, and 1 MCI) completing this pre-task revealed an average time of 15.9±2.3 and nearly no errors (0.0±0.1). Across all participants, slower performance on pre-task was significantly associated with worse APT-Script performance (r_s=−0.37, p=0.005) and worse APT-PCP performance (r_s=−0.27, p=0.05); however, there was no significant association in CN participants alone. When adding average pre-task time to the regression models in CN participants above with APT-Script and APT-PCP as dependent variables, the associations between APT-Script and processing speed (p=0.009) and APT-PCP and executive function (p<0.001) remained significant.

Association with cognitive measures independent of self-report of difficulty of tasks

Upon completion of the tasks, 46 of the participants (36 CN and 10 YN) were asked whether these tasks relate to their daily activities on a 5-point Likert-type scale ranging from very little to very much. Then, participants were asked to rank the difficulty of each task from very easy to very hard, see Appendix for details. On average, participants agreed (2.1±1.1) that the tasks relate to their daily activities. Across all participants, those who performed better on the tasks indicated that they were easier (APT-Script: r_s=−0.28, p=0.06; APT-PCP: r_s=−0.37, p=0.01). When adding self-report of difficulty of tasks to the regression models in CN participants above with APT-Script and APT-PCP as dependent variables, the associations between APT-Script and processing speed (p=0.005) and APT-PCP and executive function (p<0.001) remained significant.

Discrimination between diagnostic groups

ANOVA revealed significant performance differences between groups (APT-Script: p=0.002, APT-PCP: p<0.001) with MCI performing worse than CN participants (Tukey post-hoc tests for APT-Script: p=0.05, APT-PCP: p=0.12) and CN performing worse than YN participants (Tukey post-hoc tests for APT-Script: p=0.05, APT-PCP: p=0.004), see Table 2 and Figure 1. Only CN and YN participants underwent APT-Bank, and a t-test revealed significant performance difference between the groups with CN performing worse than YN participants (p=0.02).

Figure 1.

Bar graphs with error bars of composite Z-scores for Task 1 (APT-Script) (LEFT) and Task 2 (APT-PCP) (RIGHT) in YN, CN, and MCI participants. MCI performed worse than CN and CN performed worse than YN participants on both tasks. P values are corrected for multiple comparisons using Tukey post-hoc tests. APT (Automated Phone Task), CN (clinically normal elderly), MCI (mild cognitive impairment), YN (young normal).

Test-retest reliability

Data was obtained for APT-Script and APT-PCP in CN participants using the same and alternate versions over short intervals. See Appendix for details of task versions A and B. Thirty one participants underwent the same version after 9.0±3.1 days, yielding a Cronbach’s α value of 0.81. Eleven participants underwent an alternate version after 8.3±4.2 days, yielding a Cronbach’s α value of 0.87.

Association with regional cortical atrophy

Structural MRI data was available for 19 participants (15 CN and 4 MCI) who underwent APT-Script and APT-PCP. Multiple linear regression models examining 4 ROI revealed a significant association between reduced inferior temporal cortical thickness and worse APT-PCP performance in all participants (β=3.01, 95% CI for β=0.45, 5.57, p=0.02; R²=0.52, p=0.04 for overall model) and in CN participants alone (β=8.07, 95% CI for β=4.27, 11.87, p=0.001; R²=0.68, p=0.01 for overall model), see Figure 2. There were no significant associations with APT-Script performance.

Figure 2.

Regression models showing the association between Task 2 (APT-PCP) and inferior temporal cortical thickness in all participants (LEFT) and in CN participants alone (RIGHT), adjusted for age, education, and AMNART IQ. AMNART IQ (American National Adult Reading Test intelligence quotient), APT (Automated Phone Task), CN (clinically normal elderly).

Discussion

This initial experience with the Harvard APT, a new performance-based ADL instrument, which consists of real-life, practical, quick (about 10 minutes altogether), and easy to administer tasks, is very encouraging. We demonstrated differential performance across diagnostic groups (CN, MCI, and YN), associations with processing speed, executive function, and regional cortical thinning within CN elderly independent of hearing acuity and motor speed, as well as incremental complexity for the three tasks and excellent test-retest reliability. The more complex tasks were influenced by level of education as expected but not by race. Our results suggest that the Harvard APT could be of useful for screening and tracking the earliest functional alterations in preclinical and early prodromal AD.

Recently, a leading expert in the field published recommendations for an effective ADL test for preclinical AD [13]. These guidelines stipulated that the test should: “1) assess cognitively complex functional abilities relevant to independent living; 2) use an interval scaled, direct performance measure that evaluates performance variables in a highly detailed and granular manner; 3) include time limitations for performance items in order to enhance item difficulty; and 4) include task completion time variables in order to capture subtle processing speed changes”. One of the few examples of ADL tests that fulfil these criteria is a new short-form version of the FCI, with which Marson and colleagues have shown subtle financial skill decline in individuals with preclinical AD participating in the Mayo Clinic Study of Aging, thus demonstrating the utility and feasibility of a performance-based ADL instrument in preclinical AD [22]. Our data suggest that the newly developed Harvard APT similarly fulfils these criteria and therefore has the potential to be an effective ADL test in preclinical AD.

Among CN elderly, the more complex Harvard APT tasks (APT-PCP and APT-Bank) were associated primarily with executive function, while the simpler task (APT-Script) was associated primarily with processing speed. These findings are in agreement with prior studies in early AD, which have shown that among various cognitive domains, instrumental ADL are associated most closely with executive function [23, [24]. There was no association between Harvard APT and global cognition or episodic memory. However, this could be partly due to the narrow range of MMSE and FCSRT scores in the CN sample of the current study. As expected, there was an association between performance on the more complex Harvard APT tasks and level of education and premorbid intelligence. Therefore, those tasks may require an education or IQ adjustment when used in the future. Similarly, there was an association with age and Harvard APT task performance. Such associations are commonly found with cognitive tests as well. That said, after adjusting for those variables, there was still a significant association between Harvard APT and executive function and processing speed in CN elderly. On the other hand, there was no association between Harvard APT performance and race in a sample consisting of about one quarter minorities. Furthermore, adjusting for hearing acuity and motor speed did not affect the association with executive function and processing speed. Finally, on average, participants agreed that the Harvard APT relates to their daily activities, which has been confirmed in a recent study that indicated wide use of phones, and in particular IVRS, among elderly individuals [14]. These findings bode well for the generalizability of the Harvard APT and its ability to overcome common confounds in the assessment of elderly individuals.

We were also able to demonstrate that the Harvard APT can discriminate well between diagnostic groups (CN, MCI, and YN); it has a wide range of performance in CN elderly, especially for the more complex tasks, which can help avoid floor or ceiling effects; it is comprised of tasks with incremental complexity; and it has excellent test-retest reliability using the same version or an alternate version. These are important properties for a new test to possess.

A small subset of CN and MCI participants in the current study underwent structural MRI. Among those, there was an association between worse performance on the Harvard APT and greater inferior temporal atrophy, even within CN elderly alone. Inferior temporal atrophy appears early in the course of AD [25] and has been associated with impairment in instrumental ADL, measured by a conventional subjective scale, in the early AD spectrum [20]. Instrumental ADL impairment in early AD has also been associated with frontal and parietal atrophy and hypometabolism [21, [26, [27, [28].

The current study had several limitations. First, the study sample was highly educated and intelligent. However, this is typical of clinical trial and imaging study samples. Therefore, participants of such studies are likely to perform on the Harvard APT similarly to the participants of the current study. On the other hand, unlike many of those type of studies, our sample consisted of about one quarter minorities, potentially increasing the generalizability of the Harvard APT. Further validation in a population-based sample will help clarify this important issue as we hope to be able to utilize this test in the clinical setting. Second, the current study had a small sample of impaired participants with MCI used to compare to the diagnostic group of interest, CN elderly. Future studies of the Harvard APT with larger samples of impaired individuals, as well as those with subjective cognitive decline, will help further characterize this test in the early AD spectrum. Third, we did not have the opportunity to compare the Harvard APT to established ADL scales, such as conventional subjective scales, which could contribute to convergent validity. However, we did have access to sensitive memory and executive function tests, the latter of which were associated with performance on the Harvard APT. Fourth, APT-Bank was not associated with APT-Script; however, APT-Bank was associated with APT-PCP and APT-PCP was associated with APT-Script. There is an incremental complexity in these tasks (APT-Script is the least complex, APT-PCP is intermediate, and APT-Bank is the most complex), which may be the reason for this differential in associations. As such, the less complex tasks may prove more useful in symptomatic individuals with MCI or mild dementia, while the more complex tasks may prove more useful in asymptomatic CN elderly or in minimally symptomatic individuals with subjective cognitive decline. Finally, only a small subset of participants in the current study had imaging data though even with this small group we found significant associations with the Harvard APT in CN elderly.

In conclusion, the Harvard APT, a novel performance-based ADL instrument developed to target CN elderly and individuals with MCI at risk for early AD, is a promising, ecologically valid, quick, and easy to administer assessment. ADL are the practical extension of cognitive function that is of the utmost importance to patients and caregivers. Accordingly, the Food and Drug Administration (FDA) has required co-primary outcome measures of cognition and ADL for clinical trials in mild to moderate AD dementia for many years now. However, recently, the FDA issued new guidance for clinical trials in early AD allowing a single cognitive outcome measure for prevention trials in preclinical AD due to the thought that there are no meaningful changes in ADL at that stage of AD [29]. That said, the guidance also stipulated the requirement to eventually demonstrate a clinically relevant benefit from the intervention, which is where a sensitive ADL assessment fits in. As such, the Harvard APT provides a unique opportunity to fill this important gap in early functional assessment both in the research and clinical setting.

Appendix

Harvard APT Subject Instructions, Scoring, and Schemas (Version A)

Initial Dial in

Note: This portion is not counted toward the subject’s performance.

In order to test for deficits in hearing acuity and motor speed, the subject will be first asked to enter 8 numbers separately, and the system will record whether the correct number was entered and how much time it took to enter the number.

The subject ID will be 5 digits long; the first 2 digits will determine the task (18 = Task 1; 26 = Task 2; 49 = Task 3); the last 3 digits will identify the subject.

Subjects will dial in to the same number 3 separate times in order to complete the 3 tasks.

Task 1: Refilling a Prescription (APT-Script)

Instructions for subject

The purpose of this task is to have you use a pharmacy automated phone menu to refill a prescription.

You will be given a pill bottle with a label that includes a pharmacy phone number (9-1-917-525-4971), a prescription number (487216), and a medication name (simvastatin), dose (20 mg orally), frequency (every day), quantity of pills (30), and number of refills (2). You will use this information to complete the task.

To start the task, please use our phone to call the pharmacy phone number that appears on the pill bottle.

Your subject ID is: 18XXX

* Items in parentheses above appear on the pill bottle and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Task 2: Selecting a New Primary Care Physician (APT-PCP)

Instructions for subject

The purpose of this task is to have you use a health insurance company automated phone menu to select a new primary care physician.

You will be given a health insurance member card which includes a member ID number (CDW758421693) and a member services phone number (9-1-917-525-4971). You will use this information to complete the task.

To start the task, please use our phone to call the member services phone number that appears on the card.

Your subject ID is: 26XXX

The name and address of the primary care physician you will select is: Dr. John Smith in Boston, MA.

* Items in parentheses above appear on the health insurance member card and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Task 3: Bank Account Transfer and Payment (APT-Bank)

Instructions for subject

The purpose of this task is to have you use a bank automated phone menu to make a bank account transfer in order to have enough money to pay your Federal taxes for the year.

You will be given a bank account summary that contains your checking account number (0095834881), your savings account number (0095834899), and the bank phone number (9-1-917-525-4971). You will use this information to complete the task. You will be provided with your account balances once you call in.

Amounts of money are in dollars only (not in cents).

You should take notes on a piece of paper in order to complete the task.

To start the task, please use our phone to call the bank phone number that appears on the summary.

Your subject ID is: 49XXX

The amount you owe in Federal taxes is $4,150. You will need to make the payment from your checking account.

* Items in parentheses above appear on the bank account summary and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Harvard APT Subject Instructions, Scoring, and Schemas (Version B)

Initial Dial in

Note: This portion is not counted toward the subject’s performance.

In order to test for deficits in hearing acuity and motor speed, the subject will be first asked to enter 8 numbers separately, and the system will record whether the correct number was entered and how much time it took to enter the number.

The subject ID will be 5 digits long; the first 2 digits will determine the task (19 = Task 1; 27 = Task 2; 50 = Task 3); the last 3 digits will identify the subject.

Subjects will dial in to the same number 3 separate times in order to complete the 3 tasks.

Task 1: Refilling a Prescription (APT-Script)

Instructions for subject

The purpose of this task is to have you use a pharmacy automated phone menu to refill a prescription.

You will be given a pill bottle with a label that includes a pharmacy phone number (9-1-917-525-4971), a prescription number (394172), and a medication name (lisinopril), dose (5 mg orally), frequency (every day), quantity of pills (30), and number of refills (3). You will use this information to complete the task.

To start the task, please use our phone to call the pharmacy phone number that appears on the pill bottle.

Your subject ID is: 19XXX

* Items in parentheses above appear on the pill bottle and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Task 2: Selecting a New Primary Care Physician (APT-PCP)

Instructions for subject

The purpose of this task is to have you use a health insurance company automated phone menu to select a new primary care physician.

You will be given a health insurance member card which includes a member ID number (CDW946382715) and a member services phone number (9-1-917-525-4971). You will use this information to complete the task.

To start the task, please use our phone to call the member services phone number that appears on the card.

Your subject ID is: 27XXX

The name and address of the primary care physician you will select is: Dr. Paul Jones in Denver, CO.

* Items in parentheses above appear on the health insurance member card and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Task 3: Bank Account Transfer and Payment (APT-Bank)

Instructions for subject

The purpose of this task is to have you use a bank automated phone menu to make a bank account transfer in order to have enough money to pay your Federal taxes for the year.

You will be given a bank account summary that contains your checking account number (0084925772), your savings account number (0084925788), and the bank phone number (9-1-917-525-4971). You will use this information to complete the task. You will be provided with your account balances once you call in.

Amounts of money are in dollars only (not in cents).

You should take notes on a piece of paper in order to complete the task.

To start the task, please use our phone to call the bank phone number that appears on the summary.

Your subject ID is: 50XXX

The amount you owe in Federal taxes is $5,350. You will need to make the payment from your checking account.

* Items in parentheses above appear on the bank account summary and not on the written instructions.

Scoring

The task will be scored based on total time (until disconnected), number of errors, number of repetition of steps, and correct completion.

Self-rating of Harvard APT

After completing the 3 tasks, please answer the following question:

These tasks relate to your daily activities?

1. Strongly agree

2. Agree

3. Neither agree nor disagree

4. Disagree

5. Strongly disagree

Also, please circle one of the following options for each task.

Task 1: Pharmacy task (APT-Script)

The task you just completed was:

1. Very easy

2. Easy

3. Neither easy nor hard

4. Hard

5. Very hard

Task 2: Health Insurance Company task (APT-PCP)

The task you just completed was:

1. Very easy

2. Easy

3. Neither easy nor hard

4. Hard

5. Very hard

Task 3: Bank task (APT-Bank)

The task you just completed was:

1. Very easy

2. Easy

3. Neither easy nor hard

4. Hard

5. Very hard