Passively-Measured Routine Home Computer Activity and Application Use Can Detect Mild Cognitive Impairment and Correlate with Important Cognitive Functions in Older Adulthood

Abstract

INTRODUCTION:

Computer use is a cognitively complex instrumental activity of daily living (IADL) that has been linked to cognitive functioning in older adulthood, yet little work has explored its capacity to detect incident mild cognitive impairment (MCI).

OBJECTIVE:

Examine whether routine home computer use (general computer use as well as use of specific applications) could effectively discriminate between older adults with and without MCI, as well as explore associations between use of common computer applications and cognitive domains known to be important for IADL performance.

METHODS:

A total of 60 community-dwelling older adults (39 cognitively healthy, 21 with MCI) completed a neuropsychological evaluation at study baseline and subsequently had their routine home computer use behaviors passively recorded for three months.

RESULTS:

Compared to those with MCI, cognitively healthy participants spent more time using the computer, had a greater number of computer sessions and had an earlier mean time of first daily computer session. They also spent more time using email and word processing applications, and used email, search and word processing applications on a greater number of days. Better performance in several cognitive domains, but in particular memory and language, was associated with greater frequency of browser, word processing, search and game application use.

CONCLUSION:

Computer and application use are useful in identifying older adults with MCI. Longitudinal studies are needed to determine whether decreases in overall computer use and specific computer application use are predictors of incident cognitive decline.

Introduction

Over 13 million individuals will be living with dementia in the United States by 2050, a figure likely to greatly tax the healthcare system, the economy and care providers [1]. As a result, efforts to identify those early on in the progression of the disease are needed in order to intervene before symptoms reduce older adults’ independence to the point that they require greater care outside the home [1, 2].

Subtle declines in instrumental activities of daily living (IADLs), such as reduced efficiency, accuracy and use of compensatory strategies in managing medications and finances, can signal the transition from normal cognitive function to mild cognitive impairment (MCI), a condition in which noticeable decrements to cognitive functioning emerge and begin to undermine completion of IADLs [3]. While the onset and course of symptoms may vary somewhat between individuals, individuals who develop MCI are at heightened risk for even steeper cognitive and functional decline (i.e., dementia) [3, 4]. Cognitive testing, including brief cognitive screeners and in-depth neuropsychological assessments, has utility in identifying those who have developed MCI. However, these methods are subject to significant clinical and practical limitations, as cognitive screeners have relatively poor sensitivity and specificity to detecting MCI, while neuropsychological evaluations are time and cost-intensive as well as modest in their ability to measure or predict early changes in IADLs [5–7]. Measures designed to directly assess IADLs have also been developed; however, these are frequently prone to report biases, assess marked rather than subtle IADL impairments and often do not capture objective real-world functioning sufficiently [5].

Unobtrusive collection of IADL data in home and community environments represents a relatively new approach to detecting early and subtle changes in cognitively complex IADLs. This method involves using non-invasive consumer devices (e.g., electronic pillboxes, wearable physical activity sensors, automobile sensors, computer monitoring software) to continuously measure older adults’ completion of cognitively demanding tasks on a daily basis. Such technology-aided approaches allow for the assessment of such activities in older adults’ own homes and over long periods of time, thereby improving ecological validity, ability to detect within person change and supplementing standard in-clinic or laboratory appointments. These techniques have proven useful in discriminating healthy older adults from those with MCI based on collection of data in several areas, including mobility [8–10], medication use [11] and sleep habits [10, 12].

Although less frequently assessed in clinical settings than the everyday activities noted above, technology device use represents another cognitively demanding IADL task. Over 90% of individuals over age 50 report using computers, and over 80% of those aged 50 to 64 have smart phones [13]. Despite this, many older adults express limited confidence in their ability to use devices [14]. Multiple factors have been linked to lower frequency and extent of technology use in this population, including greater age, lower income and lower level of education [15, 16]. Cognitive function has also been cited as a predictor of technology use, with results suggesting that those with MCI self-report greater difficulty using technology [17]. Those with MCI also display less efficient use of the computer mouse and longer pauses between mouse movements [18], and those with lower cognitive function employ more unique terms in their online searches [19]. In what is to-date the only study assessing associations between cognitive function and objectively-measured computer use frequency, Kaye and colleagues (2014) found that older adults with MCI used the computer less frequently at home and demonstrated greater day-to-day variability in their use relative to healthy comparisons over a three-year period [20]. Lower amount of daily computer use has also been linked to reduced hippocampal volumes, further suggesting that changes in computer use may be an early sign of a neurodegenerative condition in older adulthood [21].

Among the growing number of studies collecting everyday activity data using unobtrusive means, a notable limitation concerns their reliance on data collection over the course of many months to several years [22]. To address this, we recently explored whether passively collecting everyday activity data over a comparatively brief time period (i.e., three months) would be sufficient to observe relationships between cognitive function and computer use, sleep, medication adherence and physical activity in a sample of older adults without dementia (Bernstein et al., unpublished results). We found that greater computer use was broadly associated with better performance on several facets of cognitive performance, including memory, language and executive function. In combination with an absence of associations between cognition and both medication adherence and physical activity, results provided preliminary evidence that computer use might be a more sensitive predictor of cognitive function than other everyday activities when assessed over a brief period, potentially attributed to its greater cognitive demands.

Additionally, while patterns of computer use have begun to be described in this population, to our knowledge no studies have used objectively measured data to describe the frequency with which older adults complete specific tasks and activities on the computer, nor have they explored how these behavior patterns may be linked to cognitive function. Indeed, literature in this area is limited to survey studies, which may not fully portray the vast number of computer functions older adults may employ and is limited by self-report methodology [23, 24]. Understanding the underlying cognitive domains that contribute to decreased computer use and computer application use would help inform targeted cognitive rehabilitations as well as computer training efforts in this population, an intervention that has proven beneficial for quality of life and well-being [25]. In addition, it is likely that some computer applications are more cognitively complex and demanding than others, and changes in the use of certain applications could eventually serve as sensitive markers of early cognitive decline.

Importantly, the collection of such sensitive information as home computer use also raises concerns about the ethical use of these types of information in clinical research and practice. With the rapid advancement of technology, a greater variety of tools are available to learn about and describe individuals’ everyday behaviors. While such technologies hold promise in supporting and detecting changes in older adults’ everyday functioning, they must be carefully and thoughtfully employed so as to maximize individuals’ needs and benefits (e.g., physical and mental well-being, independence) while minimizing burdens that may stifle adoption of potentially clinically useful devices (Mahoney et al., 2007). In general, factors such as respect for an older adult’s privacy, preservation of their autonomy, and an appreciation for the vulnerability of many such individuals (particularly those at risk for cognitive decline) to exploitation must all be considered when designing and recruiting for such studies (Diaz-Orueta et al., 2020). This is particularly true for studies examining computer use, an activity that more than ever serves as a window into an older adult’s interests, health, finances, and even comfort with using technology, among a plethora of other sensitive information [24, 26–28]. As such, clinicians and researchers examining older adults’ computer use should collect only the data that is essential to answering their specific questions, while refraining from obtaining even more personal or sensitive information (e.g., monitoring total time spent using the internet, rather than particular web pages visited).

In the present study, we sought to build on both our prior investigation and the extant literature by exploring whether the computer use frequency data described in our previous work (Bernstein et al., unpublished results) could discriminate between cognitively healthy older adults and those with MCI over a brief three-month cross-sectional interval. Given the dearth of literature in this area, this study was considered exploratory in nature. Consistent with our previous findings as well as those of Kaye and colleagues (2014) over a more extended interval [20], we hypothesized that in comparison to healthy older adults, those with MCI would demonstrate lower frequency of computer use and greater day-to-day variability in use. In an effort to further build on the work of Kaye and colleagues, we examined cognitive group differences in time of first and last computer session, with the hypothesis that those with MCI would have a later first session and earlier last session. Additionally, as no studies have utilized unobtrusive methods to describe the tasks or functions that older adults use computers for, we also aimed to address this question through exploration and categorization of application use data. Specifically, we aimed to determine whether use of particular types of applications can discriminate between cognitively healthy and MCI older adults and to examine their cognitive correlates. Given that global cognition (i.e., composite scores that take into account several cognitive domains), memory and executive cognitive functions have been strongly linked to computer use in older adult and MCI samples in the literature [29, 30], we expected to find significant associations between computer applications use and these cognitive domains in our sample. We conclude this paper with a brief discussion of ethical considerations and implications relevant to in-home monitoring of older adults’ computer use.

Materials and Methods

The present study included participants from two observational cohort studies: (1) the Promote Independent Aging study, which was funded by Veterans Affairs Research & Development and carried out at the Minneapolis VA Health Care System (MVAHCS) in collaboration with the Oregon Center for Aging & Technology (ORCATECH) at Oregon Health & Science University (OHSU) and the Collaborative Aging Research Using Technology initiative (CART; carthome.org); and (2) the Aging Well study, which was funded by the National Institute on Aging and carried out at both the MVAHCS and at OHSU in collaboration with ORCATECH.

Both studies were approved by the MVAHCS and OHSU institutional review boards (IRB numbers 18-00328 and 00019378, respectively) and followed all applicable institutional guidelines. All participants provided both written and oral informed consent before being enrolled in the study, including study staff having participants provide verbal summaries of study procedures to confirm comprehension. Additionally, several efforts were made to ensure participants’ continued consent to remaining enrolled in the study. Staff interacted regularly with participants through phone calls, emails, newsletters and annual study visits, during which participants were asked about any comments or concerns they had about any study procedures, including having their computer use monitored. Participants were informed at their baseline appointments that all data collected would remain confidential, that their data was tied to a unique participant identification number but not to their name, and what data would and would not be collected (see Procedures). Participants were also invited to have someone who knew them well (usually a spouse or adult child) present during the consent process at baseline as well as at subsequent annual study visits. While these individuals were not formally consented, they were provided the opportunity to ask questions about the participant’s involvement in the study and any study procedures. Data collected at all study appointments (see below) were reviewed shortly thereafter as part of a principal investigator clinical consensus meeting, and participants who were suspected of having dementia were not allowed to remain enrolled in the study. In these instances, the participant was notified and a suggestion was made for formal clinical evaluation.

Participants

Participants were 60 community-dwelling older adults recruited from the Minneapolis-Saint Paul, Minnesota and Portland, Oregon metropolitan areas. Of this group, 39 were classified as cognitively healthy and 21 were classified as MCI using established clinical and research measures and consistent with the NIA-AA workgroup criteria for MCI [31]. Inclusion criteria were 65 years of age and older, living within 30 miles of the study site, living independently in their home (living with a companion or spouse was allowed, but not as a caregiver), having a broadband internet connection capability, owning a computer and using it at least once per week and being willing to have their computer use frequency and computer application use frequency passively monitored. Participants also had to be relatively healthy for their age. Individuals with moderate to severe anxiety or depression (i.e., Generalized Anxiety Disorder-7 questionnaire score >5 or Geriatric Depression Scale score-15 ≥7) [32, 33], impaired global cognition (Montreal Cognitive Assessment sex, age, and education adjusted z-scores ≦−2 or global Clinical Dementia Rating Scale score >.05) [34, 35] or a dementia diagnosis were excluded.

Procedures

Participants completed a battery of clinical and neuropsychological measures at study baseline. The standardized battery included an informant-rated functional questionnaire and mental health measures (e.g., Functional Assessment Questionnaire [FAQ], Geriatric Depression Scale [GDS]) [32, 36], as well as a comprehensive battery of neuropsychological tests from the Uniform Data Set (UDS) of the National Alzheimer’s Disease Coordinating Center (NACC) [37] in addition to other validated tests [38]. The neuropsychological examination assessed multiple cognitive domains: attention and processing speed (Number Span Forward, Trail Making Test Part A, Stroop Color Naming, Stroop Word Reading) [39, 40], memory (Craft Story Recall, Consortium to Establish a Registry for Alzheimer’s Disease delayed recall and recognition, and Benson Complex Figure Delayed Recall) [35, 41, 42], language (Category Fluency and Multilingual Naming Test) [35, 43, 44], executive functioning (Number Span Backward; Stroop Color-Word; Trail Making Part B) [35, 39, 40] and visuospatial construction (Benson Complex Figure Copy) [42]. Domain specific and global cognition z-scores were calculated using group mean and standard deviations from the UDS NACC clinically normal cognitive group (n=3550–3991) [37].

Research personnel installed Worktime computer use monitoring software on participants’ own desktop or laptop computers (www.worktime.com). This in-the-background, commercially available software collected information regarding the duration and frequency of a participant’s computer sessions, including log-on/log-off times, active/idle times and time spent engaged with each computer application. As part of Worktime implementation, a username/password log-in combination was installed on the computer. Solely the participant (and no other computer users) used this log-in information when initiating a new computer session, thereby ensuring that only the participant’s computer use behaviors were monitored. Broad computer use outcome variables included daily time spent using the computer, number of daily uses, first and last time using the computer each day and whether or not the computer was used at least once during the day. These variables were computed automatically on a daily basis and averaged over the course of the entire study period to yield summary variables for each participant that were included in analyses.

For the computer application use data, all individual computer applications used by any participant at least once during the study period were first categorized by their primary reason for use (e.g., word processing, internet browsing, etc.). Categorization decisions were made by consensus among members of the research team, which included those with clinical and technical backgrounds. A total of 195 distinct applications were used during the study period, each of which were categorized into one of the following 34 categories: Advertising, Anti-malware/Security, App Store, Backup, Browser, Calculator, Consumer Information, Contacts, Email, Exercise, Finance, Game, Geographical Navigation, Graphics Editor, Information Technology Support, Multimedia, Multimedia Editing, News/Audiobooks, Operating System Support, PC-Tablet Interaction, PDF Viewer, Photos, Presentations, Printer, Productivity, Remote Login, Screen Saver, Screenshot, Search, Settings, Spreadsheet, Storage, Teleconferencing, Word Processing. Investigators then selected the application categories that were either most frequently utilized by study participants (in minutes) or of particular a priori clinical interest to investigators given their potential cognitive demands. Application categories selected for inclusion in analyses were email (e.g., Gmail, Microsoft Outlook), games (e.g., chess, solitaire), browser (e.g., Google Chrome, Mozilla Firefox), teleconferencing (e.g., Skype, Zoom), finance (e.g., Quicken), computer search (e.g., Microsoft Search tool) and word processing (e.g., Microsoft Word). For each application category, outcome variables included (1) the total amount of time (in minutes) that each participant used applications in that category over the course of the study period, and (2) the total number of days in which each participant used at least one application within that application category across the study period.

Advanced Encryption Standard encrypted data (FIPS 140-2 compliant) was transmitted to study servers via Transmission Control Protocol connection. Web URLs, names and documents were not monitored. Worktime Corporate does not record passwords, keystrokes, emails, chats, or screen or document content. The computer software was installed on Windows operating system (7, 8, 8.1, and 10) compatible computers.

Statistical Analyses

Descriptive statistics were used to characterize computer and application use data. Differences between cognitive status groups in demographics, cognitive performance and computer and application use measures were assessed using independent-samples t-tests for continuous variables and Pearson’s chi-square tests for categorical variables. Given the sample size and an alpha error probability of 0.05 (two-tailed), a post-hoc power analysis using g*power indicated 44% power to detect a medium effect size (Cohen’s d = 0.5). Pearson’s correlation coefficients were used to assess associations between cognitive performance and continuously measured application use outcome variables in the entire sample. Cognitive domains that were significantly correlated with application use variables were next entered into post-hoc simultaneous multiple regressions to examine which cognitive domains were independently predictive of specific application use variables. Given the exploratory nature of this study, no adjustments for multiple comparisons were made.

Results

Demographic, Computer and Application Use Descriptive Information

Descriptive characteristics of the sample are found in Table 1. In the full sample, participants had a mean age of 73.4 years (SD = 5.3) and a mean education of 15.0 years (SD = 2.8). Participants were 73% male and 92% white. The cognitively healthy and MCI groups did not differ on any demographic variables including age, sex, race, education or depressive or anxiety-related symptoms (all p > .05).

Table 1.

Baseline Demographic, Functional and Cognitive Sample Characteristics

Variable	Total (N = 60)	Healthy (N = 39)	MCI (N = 21)
	M (SD) or N (%)	M (SD) or N (%)	M (SD) or N (%)
Age (years)	73.4 (5.3)	72.6 (4.7)	75.0 (6.2)
Sex (male)	44 (73.3%)	28 (71.8%)	16 (76.2%)
Race (white)	55 (91.7%)	35 (89.7%)	20 (95.2%)
Education (years)	15.0 (2.8)	15.1 (2.5)	14.8 (3.3)
FAQ*	0.6 (1.6)	0.2 (0.7)	1.4 (2.3)
MoCA**	25.0 (2.7)	26.2 (2.1)	22.7 (2.3)
GDS	1.1 (1.3)	1.1 (1.2)	1.1 (1.6)
GAD	1.5 (1.5)	1.3 (1.5)	1.8 (1.5)
Global Cognition**	−.25 (.64)	.03 (.48)	−.77 (.57)
Memory**	−.29 (.67)	−.02 (.41)	−.79 (.66)
Language*	−.17 (.49)	−.02 (.41)	−.44 (.52)
Attention and Processing Speed**	.17 (.53)	.30 (.53)	−.09 (.46)
Executive Functioning**	−.11 (.63)	.15 (.54)	−.61 (.45)
Visuospatial Perception/Construction**	−.44 (.70)	−.22 (.63)	−.84 (.65)

Note: FAQ = Functional Assessment Questionnaire; MoCA = Montreal Cognitive Assessment; GDS = Geriatric Depression Scale; GAD = 7-item Geriatric Anxiety Scale;

^*p < .05;

^**p < .01

Computer and application use summary statistics in the full sample and within the healthy and MCI groups are in Table 2. In the full sample, participants spent an average of 91 minutes per day using their computer (SD = 77), with significant day-to-day variability (i.e., 62 minutes). Participants averaged 10 computer use sessions per day (SD = 6), with an average first daily use time of 10:42am and an average last daily use time of 4:47pm. Participants used the computer on 66% of the days in the study period. With regard to application use, participants spent the greatest amount of total minutes using web browsers (M = 4,224), followed by games (600) and search tools (339). Participants spent the greatest number of total days with at least one use of web browsers (56), search tools (32) and email (16).

Table 2.

Computer and Application Use Summary Metrics Over Three-Month Interval (N = 60)

Variable	Total (N = 60)	Healthy (N = 39)	MCI (N = 21)	Effect Size
	M (SD)	M (SD)	M (SD)	Cohen’s d
Computer Use Variables
Computer Use Time (minutes)*	91.0 (77.1)	105.9 (88.9)	63.4 (36.4)	0.57
Computer Use Time Variability (minutes)	61.3 (38.8)	67.8 (43.0)	49.2 (26.5)	0.49
Number of Sessions*	10.1 (6.8)	11.6 (7.6)	7.3 (3.8)	0.65
Time of First Session*	10:42am (2:43:38)	10:04am (2:37:11)	11:53am (2:34:52)	0.70
Time of Last Session	4:47pm (2:35:39)	5:04pm (2:41:15)	4:14pm (2:22:43)	0.32
% Days with at least One Session	65.1% (34.2)	70.4% (33.1)	55.3% (34.7)	0.45
Application Use Variables
Email Use Time, minutes*	322.0 (750.4)	398.4 (841.0)	46.8 (114.4)	0.53
Game Use Time, minutes	600.4 (1839.6)	504.7 (1404.3)	424.4 (1748.0)	0.07
Browser Use Time, minutes*	4224.5 (4607.9)	5078.2 (5212.6)	2602.9 (2757.6)	0.60
Teleconferencing Use Time, minutes	2.5 (12.6)	3.2 (15.2)	0.0 (0.0)	0.26
Finance Use Time, minutes	32.6 (109.9)	30.4 (111.6)	42.1 (116.4)	0.08
Search Use Time, minutes	338.6 (1070.9)	436.2 (1324.2)	160.4 (288.0)	0.27
Word Processing Use Time, minutes*	215.3 (465.5)	299.9 (542.3)	60.0 (242.1)	0.54
Email Use, Days*	16.9 (29.9)	20.7 (34.3)	6.5 (13.8)	0.52
Game Use, Days	9.5 (23.1)	11.4 (25.7)	4.0 (12.9)	0.33
Browser Use, Days	56.4 (30.9)	61.7 (30.6)	48.4 (30.4)	0.51
Teleconferencing Use, Days	0.3 (1.6)	0.4 (2.0)	0.0 (0.0)	0.26
Finance Use, Days	2.1 (8.4)	2.6 (10.3)	1.5 (3.3)	0.14
Search Use, Days*	32.2 (25.3)	36.5 (27.5)	23.6 (18.4)	0.59
Word Processing Use, Days**	10.8 (19.0)	15.5 (22.3)	1.8 (3.7)	0.77

Note: Computer use variables are expressed as quantities per day. Application use variables are expressed as total quantities over the three-month period;

^*p < .05;

^**p < .01

Associations Between Computer Use and Cognitive Status

Independent samples t-tests indicated that in comparison to participants with MCI, healthy participants spent more time using the computer daily (t(57) = 2.47, p < .05, d = 0.63), had a larger number of computer sessions daily (t(57) = 2.23, p < .05, d = 0.72) and had an earlier first time of computer use daily (t(57) = −2.37, p < .05, d = 0.70). A statistical trend was also noted whereby healthy participants had greater variability in their amount of use time per day (t(57) = 1.91, p = .06, d = 0.52).

Associations Between Application Use and Cognitive Performance

With regard to application use, independent samples t-tests indicated that in comparison to participants with MCI, healthy participants spent relatively more total minutes using email (t(57) = 2.60, p < .05, d = 0.59), web browsers (t(57) = 2.41, p < .05, d = 0.59) and word processing (t(57) = 2.37, p < .05, d = 0.57) during the three-month period. Healthy participants also spent relatively more total days using email (t(57) = 2.27, p < .05, d = 0.54), search tools (t(57) = 2.16, p < .05, d = 0.55) and word processing (t(57) = 3.80, p < .01, d = 0.86) than those with MCI.

Correlations between cognitive performance (global and domain-specific z-scores) and computer application use are found in Table 3. Pearson’s correlations indicated that more total minutes using web browsers was associated with better performance on language (r = .38, p < .01), executive function (r = .27, p < .05) and global cognition (r = .27, p < .05). More total minutes spent using word processing was associated with better performance on attention (r = .37, p < .01), executive function (r = .36, p < .01) and global cognition (r = .26, p < .05). Greater number of days with a game application use was associated with better memory scores (r = .29, p < .05). Greater number of days with browser use was associated with better memory (r = .32, p < .05), language (r = .26, p < .05) and global cognition scores (r = .30 , p < .05). Greater number of days with search tool use was associated with better memory (r = .28, p < .05), language (r = .30, p < .05) and global cognition scores (r = .33, p < .05). Greater number of days with word processing use was associated with better memory (r = .41, p < .01), language (r = .35, p < .01), attention (r = .35, p < .01), executive function (r = .46, p < .01) and global cognition scores (r = .46, p < .01).

Table 3.

Associations Between Three-Month Application Use Metrics and Baseline Cognitive Performance (Global and domain-specific z-scores) (N = 60)

	Global	Memory	Language	Attention	Executive Function	Visuospatial
Email Use Time	.19	.17	.27	.23	.20	.03
Game Use Time	.06	.21	−.03	−.05	−.08	.00
Browser Use Time	.27*	.18	.38**	.13	.27*	.10
Teleconferencing Use Time	.11	.04	.09	.14	.16	.15
Finance Use Time	−.12	.01	.06	−.03	−.18	.10
Search Use Time	.04	.01	.06	.10	.00	.12
Word Processing Use Time	.26*	.22	.18	.37**	.36**	.08
Email Use Days	.07	.06	.18	.13	.06	.01
Game Use Days	.16	.29*	.07	.09	.01	.02
Browser Use Days	.30*	.32*	.26*	.15	.25	.03
Teleconferencing Use Days	.11	.04	.09	.14	.16	.15
Finance Use Days	−.11	.00	−.02	−.08	−.10	.17
Search Use Days	.33*	.28*	.30*	.15	.23	.18
Word Processing Use Days	.46**	.41**	.35**	.35**	.46**	.16

^*p < .05;

^**p < .01

For application use variables that were associated with multiple cognitive domains, post-hoc simultaneous multiple linear regression analyses were next conducted to determine whether individual cognitive domains were independently associated with application use. Only memory and language were significantly associated with application use. Better memory performance was associated with more days with word processing use (F(2, 58) = 5.22, p < .05, f² = 0.38), and there was a non-significant trend for greater memory to be associated with greater number of days with at least one use of browser (F(2, 58) = 4.05, p = .06, f² = 0.14). Better language performance was associated with more minutes of browser use (F(2,58) = 4.88, p < .05, f² = 0.17). No other cognitive domains were predictive of use of any application use variables (all p > .05).

Discussion

Although literature using self-report data has suggested a relationship between older adults’ routine home computer use and cognitive status, this has not been extensively investigated using objectively measured computer use data. The entire cohort in the present study used the computer for an average of 1.5 hours per day over an average of ten daily logins, which is consistent with prior objective monitoring research demonstrating an average use of 1.5 hours/day among both MCI and cognitively intact cohorts over a one-month period [20]. While there was notable day-to-day variability in computer use (i.e., 62 minutes), participants used the computer in approximately 66% of the days in the study period.

The vast majority of older adults own and use a computer [13]. Despite this, the current study is the first to objectively monitor how older adults spend their time using computer applications. In this study, the vast majority of computer application time and greatest number of days (approximately 2/3 of total monitoring days) were spent using web browser applications. This is not surprising given that 67% of seniors endorse going online regularly as well as considering the vast array of activities older adults can complete through their web browsers, including social interaction and communication, entertainment and learning, facilitation of routine tasks (e.g., banking and shopping), accessing health services and mental stimulation and challenge [23–25]. After Browser Use Time, older adults spent the most amount of time on Game Use, followed by Search Use, Email Use and Word Processing Use. A prior study of self-reported computer application use among a 50+ aged cohort found that word-processing, Internet and email were the most frequently used applications, which may reflect the evolution in older adults’ technology and computer knowledge [45](Goodman et al., 2003). Interestingly, while the second most in terms of time spent, Game Use Time was the fifth most in terms of total days with at least one use, potentially attributed to lower popularity among the entire cohort but greater engagement by a select set of participants.

Prior objective in-home computer monitoring research demonstrated that routine computer use measured over three years discriminated between healthy cognition and MCI older adults, whereas there were no significant group differences in computer use in a shorter one-month baseline period [20]. The present study demonstrated significant, medium-to-large group differences between MCI versus healthy older adults in similar aspects of routine home computer over a three-month baseline period. Similar to Kaye et al (2014) [20], we found that individuals with MCI had fewer daily computer sessions and used the computer for less time daily as compared to their healthy counterparts. In addition, we examined time of first computer use in our study and found that MCI participants’ time of first computer use was approximately two hours later in the day than healthy participants, which is consistent with prior work suggesting that later computer start time of day is associated with MCI and poorer memory and visuospatial performance [46]. These results provide additional evidence to support the relationship between MCI and reductions in engagement in routine computer use.

Percent days with at least one computer session did not differ between groups, which is surprising given the relationship of days of computer use with cognition in other research [20] and may be related to the fact that cognition was dichotomized in the current study (i.e., MCI vs. healthy) rather than treated as a continuous variable. The present study also identified a trend toward greater variability in daily time of computer use in the healthy participants as compared to the MCI group, a finding which conflicts with Kaye and colleagues (2014) [20], who found that MCI participants exhibited greater day-to-day variability in computer use over a three-year period. The difference in monitoring periods may partially explain this contrast in findings. Shorter periods of monitoring may capture the fact that healthy individuals are flexible in their timing and use of more complex applications whereas individuals with MCI require more daytime routine and structure to engage in the task.

In addition to exploring associations between cognitive status and broad computer use, the present study is the first to examine relationships between cognitive status and use of specific computer applications. As compared to individuals with MCI, healthy participants spent more time using email and word processing applications and used email, search, and word processing applications on a greater number of days, with largely medium effects observed. We then explored cognitive correlates of use of specific computer applications in the entire sample. Better global cognition was associated with greater browser use time and both word processing use time and days of use. Prior self-report research in this area has focused exclusively on Internet use and found a similar positive relationship between cognition and Internet use both cross-sectionally and longitudinally [14, 15]. This study is the first to demonstrate that objectively assessed computer application use, including word processing, email and search use, are significantly associated with cognitive status and global cognition.

Number of days with word processing use was the application use variable most frequently and strongly correlated with several aspects of cognition. Compared to healthy individuals, participants with MCI also used word processing less frequently on a daily basis and on fewer numbers of days. Such findings may be suggestive of the broad cognitive faculties required to successfully use word processing, including language (e.g., sentence formation, object naming, syntactical knowledge), memory (e.g., working memory to keep track of information and decide the next step, episodic memory to remember content, procedural memory to initiate routines necessary to use the program), attention (e.g., engagement in the task) and executive functioning (e.g., concept formation, initiation and execution of commands). The relationship between cognitive functioning and computer application use is complex, such that current cognitive functioning likely results in greater application use while greater computer application use also potentially results in benefits to cognition over time [47].

Like word processing, greater browser use time and number of days with browser use were associated with memory, language and executive functions, with MCI participants using browser applications for less total time than healthy participants. Prior imaging research has found greater brain activation in frontal regions during Internet search tasks among older adults with computer use experience [48], which is broadly consistent with positive relationships observed between browser use and executive performance in the present study. Greater search use time was also linked to better memory and total cognition. While previous studies have explored associations between age, cognition and use of internet search tools, to our knowledge this is the first study to establish that use of the computer search tool may similarly be linked to cognitive function.

Of the cognitive domains assessed, language and memory were most consistently linked to application use. The importance of memory functioning for technology use is highlighted by prior imaging studies, documenting reduced computer engagement among those with decreased hippocampal and mesial temporal lobe volume, brain regions critical for memory functioning [21]. Most computer applications are language-mediated and depend on free retrieval and application of knowledge. Thus, computer engagement may be limited when those domains are impaired, especially for those applications that require independent generation of information (e.g., word processing applications) and less context or structure (e.g., browser applications). Given that decline in memory and language are two sensitive indicators of early Alzheimer’s disease [49, 50], computer application use monitoring holds promise as a potential real-world marker of cognitive decline in these areas.

Given the promise of computer and application use monitoring in the detection of cognitive impairment, studies with larger samples and longer periods of monitoring (i.e., years) are warranted. Monitoring over such an extensive time period in combination with serial administration of cognitive test batteries would allow for more nuanced appreciation of the temporal relationship between declines in cognition and alterations in computer and application use. Consistent with the present study’s approach, from a logistical and ethical perspective, such work should promote frequent communication between participant and study staff to ensure the former has adequate understanding about benefits to the individual or older adults more broadly as well as both the extent and limits of data collected via computer monitoring. Studies should take significant efforts to maintain older adults’ privacy (e.g., excluding collection of website or document contents), as well as respect their autonomy throughout study involvement (e.g., acknowledge participants’ right to decline or limit engagement, encourage questions, recognize that research norms may not always align with participants’ preferences – see Robillard and Feng (2017) for review) [51]. If the present study’s findings hold in larger-scale projects, significant decreases in use or dramatically later uses throughout study participation may indicate the need for staff follow-up to determine whether additional, formal cognitive assessment or intervention may be indicated.

Limitations

The current study is not without limitations. This relatively small pilot study recruited a predominantly white and male sample, with all participants living independently and with few health and psychiatric comorbidities, which limits generalizability to broader older adult populations as well as our statistical power to detect group differences. Additionally, inclusion criteria required that all participants have broadband internet connection and own and use a computer at least once per week, which limits the translational value to more diverse samples with less computer knowledge. For this reason, consideration of other moderating variables that may impact activity tracking, such as technology usability and digital readiness, is important given their impact on adherence [52]. Conversely, while the computer use monitoring software was installed on the computer the participant reported spending the majority of their use time on, an increasing number of older adults use more than one device at home (e.g., second computer, tablet, smartphone). As the computer monitoring software was solely able to capture use data for PCs, and not Mac products or other devices, the study’s capacity to comprehensively measure all facets of participants’ technology use is limited in this regard [53]. In comparison to other studies assessing objectively measured computer use in older adults, the present study also collected data over a comparatively short time interval (three months).

As noted previously, since the present study was exploratory in nature, multiple comparisons were not controlled for in order to reduce the risk of type 2 errors. Similarly, because cognition was assessed solely at baseline, this study was unable to determine whether changes in cognition may have been linked to changes in computer or application use, thereby limiting capacity to comment on the direction of relationships. Future studies will examine associations between cognitive functioning and computer and application use in larger samples and over more extended time periods.

Conclusion

Our study adds to an emerging literature suggesting that computer use behaviors may be helpful in the detection of cognitive impairment in older adulthood. Additionally, this study is the first to show that use of particular computer applications may aid in identification of such individuals. Future work should examine longitudinal changes in computer and application use and concurrent changes in cognitive and functional abilities to better understand these relationships.