Smart Home Technology: A New Approach for Performance Measurements of Activities of Daily Living and Prediction of Mild Cognitive Impairment in Older Adults

Abstract

Background:

Functional assessment is of paramount importance when mild cognitive impairment is suspected, but common assessment tools such as questionnaires lack sensitivity. An alternative and innovative approach consists in using sensor technology in smart apartments during scenario-based assessments of instrumental activities of daily living (IADL). However, studies that investigate this approach are scarce and the technology used is not always transposable in healthcare settings.

Objective:

To explore whether simple and wireless technology used in two different smart environments could add value to performance and rater-based measures of IADL when it comes to predicting mild cognitive impairment (MCI) in older adults.

Methods:

Twenty-six (26) cognitively healthy older adults (CH) and 22 older adults with MCI were recruited. Functional performance in a set of five scripted tasks was evaluated with sensor-based observations (motion, contact, and electric sensors) and performance-based measures (rated with videotapes). The five tasks could be performed in any order and were detailed on an instruction sheet given to participants.

Results:

Sensor-based observations showed that participants with MCI spent more time in the kitchen and looking into the fridge and kitchen cabinets than CH participants. Moreover, these measures were negatively associated with memory and executive performances of participants and significantly contributed to the prediction of MCI.

Conclusion:

Simple, wireless, and sensor-based technology holds potential for the detection of MCI in older adults as they perform daily tasks. However, some limits are discussed and we offer recommendations to improve the usefulness of this innovative approach.

Keywords

Activities of daily living memory mild cognitive impairment neurocognitive disorders older adults smart homes technology wireless technology

INTRODUCTION

Timely diagnostic of Alzheimer’s disease (AD) is essential as interventions that delay the onset of symptoms of AD and promote home support are more effective when set up in the early stages of the disease [1-4]. Numerous studies showed that some AD-related biomarkers and symptoms progress slowly and its pre-symptomatic phase extends over a period ranging from 25 to 30 years [5, 6]. Over the last decade, much interest has been given toward the detection of mild cognitive impairment (MCI), a condition that comes with an increased chance of developing dementia: over half of those with MCI will develop dementia within 5 years [7]. MCI is defined as a cognitive decline greater than that expected for an individual’s age and education level which does not interfere notably with function in activities of daily living (ADL) [8]. In contrast, a dementia or a major neurocognitive disorder diagnosis is given when cognitive deficits interfere with functional independence in ADL (Diagnostic and Statistical Manual of Mental Disorders-5). Therefore, functional difficulties are often said to be absent from MCI.

Still, MCI is an evolving concept and a great number of studies now suggest that performance of instrumental activities of daily living (IADL), which refer to activities that are central to a person’s ability to live independently in the community (such as preparing a meal or getting information over the phone), is affected in MCI, and may impact overall quality of life [3, 9-11]. Subtle difficulties in complex IADL could already be present 10 years before AD diagnosis [12]. Moreover, there is growing evidence that difficulties in performance of IADL is an important risk factor for progression from healthy cognition to MCI [9, 13] and from MCI to dementia [14, 15]. Therefore, functional assessment is of paramount importance when a neurocognitive disorder is suspected.

However, it should be noted that common assessment tools for IADL are not very sensitive to determine a person’s real level of functioning [16]. Most researchers and clinicians give questionnaires surveying IADLs to the patient or a caregiver [17-19]. Despite the advantages of questionnaires, such as their rapid administration, simplicity, and low administration costs, it has been shown that older adults with MCI tend to misjudge their ability to perform IADL [20]. In addition, a lack of awareness of functional deficits in MCI is a predictor of progression to AD [21]. This raises concerns about the use of self-report evaluation, but informant report is also susceptible to subjective bias. Indeed, caregivers tend to over- or underestimate functional abilities of care recipients, particularly when they present cognitive disorders [20 , 23]. Performance-based measures are also used to assess performance of IADL [10, 24]. Although these tools may be more sensitive than questionnaires, their administration often requires more time and specially trained personnel.

An alternative and innovative approach to these tools consists in measuring the performances of participants while they carry out IADL (i.e., scenario-based assessments, such as in Schmitter-Edgecombe & Parsey [25]), with the help of new technologies. In fact, new sensor technologies are allowing extensive, non-invasive, and continuous monitoring of performance-based measures of IADL, such as the time spent in each room or the activation of specific sensors attached to key environmental components (e.g., entrance door, microwave). Data gathered through sensor technologies in scenario-based assessments could allow for standardized quantitative and fully automated measurements, thus contributing to functional evaluations made by therapists. Such scenario-based assessments could be administered at different points in time so that longitudinal changes could be objectively quantified. Home sensor technologies have shown promising results in the context of real-life monitoring [26], with data collected in monitoring studies over long periods (e.g., one year). In these studies, participants are only instructed to perform their daily routine naturally. However real-life monitoring is less appropriate for formal evaluation than scenario-based evaluation which is shorter in duration and can be easily administer to several patients by visiting a single apartment equipped with the sensors technology.

Even though few studies have used this new evaluation approach up until now, they have shown encouraging results. In Dawadi et al. [27], participants were invited to perform eight scripted IADL in a smart apartment. The authors were able to observe a significant correlation (r=0.54) between activity quality prediction, performance-based measures, and cognitive assessment. However, while they were able to distinguish participants that had dementia from cognitively healthy adults, MCI participants could not be discriminated from those two groups because of strong correlations with the cognitively healthy group. In Cook et al. [28], machine learning classifiers were able to distinguish healthy older adults from adults with MCI and adults with Parkinson’s disease. Finally, by using sensors installed in a research apartment, Jekel et al. [29] observed that older adults with MCI spent more time performing IADL than healthy older adults. It is worth noting that most of these studies used a large set of sensors (86 in [28], 36 in [27]), which implies an expensive protocol which is hard to replicate in various settings. Also, machine learning algorithms are prone to overfitting [30] especially if experiments are performed in a single apartment. This is why gathering data from different setups is important.

We believe that smart home technology could help to discriminate between normal aging and MCI in formal evaluations conducted in clinical settings. The goal of the present study was to determine whether simple and affordable wireless technology can help to identify difficulties in the performance of IADLs in a population with MCI through scenario-based evaluations in a laboratory testbed. Patients could be invited to perform a set of tasks in clinical settings. To do so, we have recruited cognitively healthy (CH) older adults as well as older adults with MCI. Neuropsychological tests were administered and participants were invited to perform a set of five daily tasks in an apartment equipped with only eleven sensors. Performance-based measures collected during these tasks were rated through videotape by two judges. Our goal was to examine how strongly the sensor-based data was related to the MCI clinical diagnosis, cognitive performance, and performance-based measures. We expected that sensor-based data would be significantly associated with all these variables. We also wanted to examine the predictive value of sensor-based data for MCI diagnosis and compare it with performance-based measures and cognitive performance. We hypothesized that sensor-based data would significantly improve the predictive value of performance-based measures, therefore suggesting that both measurement approaches could be used in synergy to improve diagnostic procedures in patients with MCI.

METHODS

Participants

Participants without cognitive impairment were recruited from research pools of volunteers in Montreal and Sherbrooke (Quebec, Canada). Inclusion criteria were: 1) 65 years of age or older; and 2) Montreal Cognitive Assessment (MoCA) score considered clinically normal (over Z= -1.5), based on normative data for middle-aged and elderly French Quebeckers [31]. Participants with MCI were recruited via memory clinics of the CIUSSS-Institut universitaire de gériatrie de Montréal (CIUSSS-IUGM) and at the CSSS-Institut universitaire de gériatrie de Sherbrooke (CSSS-IUGS), via lists of participants. Inclusion criteria were: 1) 65 years of age or older; and 2) clinical diagnosis of MCI established according to consensus criteria [8, 32]. Exclusion criteria for both groups were related to health conditions that could cause cognitive impairment: 1) history of cerebral disease (e.g., head trauma, stroke, encephalopathy); 2) uncontrolled diabetes or hypertension; 3) presence of psychiatric disorders: (e.g., schizophrenia, bipolar disorder, anxiety, or depression); 4) delirium in the last six months; 5) intracranial surgery; 6) vitamin B₁₂ deficiency or ethylism; 7) use of medication that can influence cognition and alertness (e.g., hypnotics, neuroleptics, or anticonvulsants); and 8) physical impairments limiting the ability to move alone and safely in the intelligent apartment. This research was approved by the IUGM and the CSSS-IUGS ethical review boards.

In total, forty-eight (48) participants were recruited and classified in two groups: 1) CH participants (n = 26; 16 in Montreal and 10 in Sherbrooke), and 2) participants with MCI (n = 22; 12 in Montreal and 10 in Sherbrooke). Participants were met individually once or twice (about four hours in total) at the laboratory apartment of the École de Réadaptation de l’Université de Montréal (Montreal) or at the DOMUS laboratory (Sherbrooke). Three experimenters (2 occupational therapists and one graduate student in occupational therapy) were responsible for the administration of screening tests and for supervising participants in the laboratory apartments. They were not blind to the diagnosis.

Five Tasks Procedures

Participants were instructed to carry out five activities detailed on an instruction sheet. Four of the selected tasks had been validated in previous studies [33] and one (obtain information by phone) was based on the IADL Profile [34, 35]. Participants were asked to complete the five tasks (Table 1) in any order but within 45 minutes, based on the procedure of the Six Elements Test [36]. However, this time, pressure was not enforced as all participants were allowed to finish the task if more time was required. The instruction sheet was available at all time so that remembering instructions was not mandatory. The only exception was task 5 (folding clothes), whose instructions were given only once on the phone, and never repeated.

Table 1

Title	Living areas	Objectives*
Task 1	Entrance	1- Place belongings in the wardrobe
Task 2	Kitchen and Dining	2- Add one cooked egg in the plate
		3- Add two toasts in the plate
		4- Add jam on the toasts
		5- Make a drinkable hot beverage
		6- Put milk in the beverage
		7- Put sugar in the beverage
		8- Set dining table
		9- Clean dishes after
		10- Put away the dishes
Task 3	Bathroom	11- Flush the toilet
		12- Wash the mirror
		13- Wash the sink
		14- Use and put away the cleaning product
Task 4	Living room	15- Get bus schedule between Montreal and Toronto for the day
		16- Give it to the experimenter
Task 5	Bedroom	17- Answer the phone when ringing
		18- **Fold the three clothes on the bed
		19- **Put clothes away in the right drawer

Note. *Each objective is worth one point after completion for a total of 19 for the five task combined; **These instructions were not provided on the instruction sheet but given on the phone.

While performing the tasks, experimenters remained in specific areas of the apartment that were blind spots for sensors. They were also instructed to intervene as little as possible, but could do so 1) when participants wandered or got stuck on a task for about two minutes, and 2) when a dangerous behavior occurred. If they had to intervene, experimenters provided incremental assistance [37]: action priming, cueing, explicit advice. Action priming consisted in inviting the participant to plan their actions (e.g., “What would you do at home?”, “Have you looked at the instruction sheet?”), without giving any indication on how to perform the tasks. Cueing consisted in providing the participant with extra information in order to help him progress (e.g., “Is everything cleaned/stored?”, “What do we put in the coffeemaker”). Finally, explicit advice included direct instructions (e.g., “You could look in the phonebook”, “The coffee filters are in this cupboard”) [37]. All five scripted tasks were performed, either in Montreal or Sherbrooke, in a smart apartment. Since it is known that CH adults and adults with MCI present a greater level of compensation and more strategies than those with dementia [38], the differentiation of these subgroups is more complex if the tasks are performed in highly familiar environments.

Variables

Cognitive performances

The cognitive evaluation mainly focused on memory and executive functions which are known to predict functional difficulties in IADL [39, 40]. To prevent multiple testing problems, and especially minimize task impurity associated with testing of executive functions [41], we calculated composite scores based on z-scores of performances on executive and memory assessments included in a cognitive assessment battery. For the executive function composite score, we added; 1) the ratio between the time required to complete the switching condition (trail B) and the baseline condition (trail A) in the Trail Making Test (TMT) [42], 2) the ratio between the time required to complete the color naming in the discordant condition (3^rd condition) and the concordant condition (2^nd condition) in the DKEFS Color Stroop Test, 3) the total score of the DKEFS Towers [43], and 4) the visuo-executive sub-score of the MoCA [44]. For the memory composite score, we added; 1) the total number of words recalled over the five trials of the Rey Auditory Verbal Learning Test (RAVLT) (maximum of 75) [45] and 2) the recall subscore of the MoCA [44].

IADLs sensor-based observations

Smart apartments were equipped with twelve sensors connected to a server. Wireless Z-wave infrared motion detectors were installed in each of the 5 living areas (i.e., bedroom, living room, dining room, kitchen, and bathroom). Since participants were allowed to switch between task and perform them in any order, each task had to be performed in a different area. The total amount of time used to perform a task was then determined by adding together the time spent by the participant in the different steps of this task, in the specific room dedicated to the task. This information was abstracted from motion sensors. The refresh time of motion sensors varied between 45 and 60 s. This means that once the sensor detected a movement in a specific room, it emitted a first signal; the next signal would be triggered only about 50 s later, if the person was still in the same room (though another motion sensor could trigger a signal elsewhere). Therefore, motion sensors could inform us of activity in a given room which, in the current protocol, constitutes a proxy of what the participants are doing. However, the number of triggers from motion sensors could not be used as proxy for activity level, movement or speed. Also, the coffeemaker and toaster were connected to electric Z-wave sensors detecting the electric consumption, in order to measure usage time (i.e., when the consumption was above 5 watts). Finally, Z-wave electromagnetic contact sensors were installed on a bedroom drawer, on the fridge, and on two cupboards in the kitchen to indicate whether they were open or closed. The total duration for which each object remained open was measured (i.e., the time interval during which the two parts of the magnetic sensor were apart from each other) and was used as a proxy of the total time spent searching.

IADLs performance-based measures

While performing the five tasks procedures, participants were videotaped with a mobile camera in the Montreal laboratory, and a set of seven fixed cameras in the Sherbrooke laboratory. To assess qualitative IADL performance-based measures, videos were analysed and scored by two independent raters, one of which was blinded to the participants group IADL. Three different variables were rated: objectives achieved, assistance interventions, and errors. Participants received a score based on the number of objectives achieved by the end of the evaluation. The maximum score for the five tasks combined was 19 and participants received one point for each successfully met objective (see Table 1). Raters also counted the number of times the participant asked for assistance, as well as the number of times the experimenters intervened without being asked to. Finally, raters counted the total number of errors committed by participants. Errors were divided in five categories inspired by Giovanetti et al. [46]: omissions, substitutions, perseverations searching, and dropping out. "Omission" referred to a missing and unattemped step (e.g., forgetting to put milk in coffee; not using soap to clean the dishes). "Substitution" referred to an attempt to perform a step with the wrong object or target (e.g., put a coat on the couch instead of in the closet; putting peanut butter on toast instead of jam). "Perseveration" referred to repetitive or unrequired behavior (e.g., cooking two eggs; cleaning the bath). A behavior was tagged as “searching” if participants spent more than two minutes to find an item or to determine how a device worked. Finally, a behavior was tagged as “dropping out” when participants voluntarily decided not to comply with an instruction or showed signs of giving up (e.g., “I don’t like milk in my coffee so I won’t”, “I haven’t taken the bus in a while, do I really need to?”). It is important to note that participant could notice their omissions or substitution errors, either spontaneously or through revision of the instruction sheet. If they corrected their errors by themselves, errors would still be included in the total, but the completion score would not be affected.

The level of agreement between the raters was satisfactory (standardized Cronbach’s alpha were 0.98, 0.096, 0.95 for completion score, amount of assistance interventions, and errors respectively). Also, for all three measures, in-between raters discrepancies of 3 points/units or more were discussed and revised in order to reach a consensus. If the discrepancies were smaller than 3 points/units, their scores were averaged.

Data analysis

Data analysis was conducted with IBM SPSS 21.0. First, demographic data and cognitive performance of participants were analysed to better describe both groups. Composite scores were calculated from memory and executive function z-score performance to streamline later analyses. They were calculated by combining subjects’ scores in tests measuring executive functions and memory. Second, ANOVAs were used to compare the two smart apartments with regard to sensor-based observations. This was done to confirm whether data obtained through sensors in the two laboratories were comparable. Third, we compared cognitively healthy and MCI groups for sensor-based and IADL performance-based measures with ANOVAs to determine which measures significantly differed among groups. Afterward, composite scores were also calculated with sensor-based and performance-based measures that were sensitive to MCI transformed in z-score. These observations were transformed in z-scores and averaged for all participants. Simple regressions were then performed to determine which IADL observations were associated with cognitive composite scores. Finally, the added value of sensor-based observation in relation to performance-based measures for the prediction of MCI diagnosis and cognitive composite scores was examined. Stepwise regressions were performed to observe the predictive relationship between IADL performance-based composite scores, cognitive performance composite scores, and MCI clinical diagnosis.

RESULTS

Participants

Demographic and cognitive measures as well as ANOVAs are presented in Table 2. There was a larger proportion of men in the MCI group, which was slightly older than the CH group. Total years of education were comparable. Compared to the CH group, MCI group performances were significantly lower in all cognitive measures as well as in the executive composite score, F(1, 46) = 27.95, p<0.01, and memory composite score, F(1, 46) = 47,76, p<0.01.

Table 2

Participants’ demographic and cognitive measures

	Cognitively healthy	Mild cognitive impairment	ANOVA
Demographics
	n=26	n=22	F(1, 46)
Sex - women	18 (69%)	8 (36%)	5.57, p <0.01
Age (y)	72.0 (6.4)	75.8 (6.5)	4.22, p <0.05
Education (y)	16.8 (4.0)	14.7 (4.1)	p >0.05
Cognition
MoCA	28.4 (1.7)	23.3 (3.1)	52.17, p <0.001
Executive function
MoCA Visuo-executive	4.6 (0.7)	3.6 (1.1)	14.58, p <0.001
TMT switching cost	1.2 (0.8)	2.1 (1.3)	8.53, p <0.01
London tower achievement score	16.4 (3.7)	13.5 (4.1)	6.74, p <0.05
Stroop inhibition ratio	1.5 (0.6)	2.3 (1.3)	7.62, p <0.005
Executive function composite score	0.37 (0.31)	-0.44 (0.68)	27.96, p <0.001
Memory
MoCA Recall	4.3 (1.0)	1.9 (1.9)	35.54, p <0.001
RAVLT total correct recall	52.1 (8.3)	32.6 (12.7)	35.67, p <0.001
Memory composite score	0.61 (0.47)	-0.72 (0.85)	47.76, p <0.001

Sensor-based observation of IADL performances

Inter-location validation of sensor data was examined: both locations were comparable in terms of time spent in each living area as well as for the use of domestic appliances and storage. The only significant difference observed was that participants in Montreal spent more time in the living room doing the “finding bus schedule” task, F(1, 46) = 15.48, p<0.001. This difference did not significantly interact with the group (CH or MCI).

In regard to comparisons between participants with MCI or CH, scores and ANOVAs for sensor-based observation are presented in Table 3. The MCI group spent significantly more time in the kitchen than the CH group. Both groups spent a comparable amount of time in all the other living areas. Also, MCI group spent more time looking into the fridge and into the kitchen cabinets than the CH group. No difference was observed for the bathroom cabinet or for the electric appliance in the kitchen. Therefore, a sensor-based composite score was calculated based on the features sensitive to MCI: time spent in the kitchen, usage of the fridge, usage of the kitchen cabinets.

Table 3

Sensor-based observation of IADL performance measures

	CH n=26	MCI n=22	ANOVA F(1, 46)
Time spent in minutes (standard deviation)
Kitchen	22.7 (7.5)	28.7 (6.3)	8,78, p <0.005
Dining room	4.7 (4.1)	4.8 (3.5)	p = 0.905
Living room	10.3 (7.3)	12.2 (7.4)	p = 0.378
Bathroom	4.2 (1.6)	4.3 (2.7)	p = 0.840
Bedroom	1.3 (0.6)	1.3 (0.8)	p = 0.698
Time of usage in minutes (standard deviation)
Fridge	1.0 (0.4)	1.3 (0.5)	6.06,p <0.05
Kitchen cabinets	3.2 (2.0)	4.5 (2.2)	4.50, p <0.05
Toaster	2.7 (0.9)	4.0 (1.3)	p = 0.390
Coffeemaker	5.9 (6.2)	5.6 (6.1)	p = 0.864
Bedroom drawer	0.3 (0.3)	0.3 (0.3)	p = 0.676

Finally, the association between sensor-based observations and cognitive composite scores were evaluated: for both groups, the total time spent in the kitchen was associated with the executive composite score, F(1, 46) = 7.67, p<0.01, _adjustR² = 0.12, β = -0.38, and the memory composite score, F(1, 46) = 8.17, p <0.005, _adjustR² = 0.13, β = -0.39. Also, only the memory composite score was related to the amount of time spent looking into the fridge, F(1, 46) = 4.62, p <0.01, _adjustR² = 0.07, β = -0.30, and looking into the cabinet, F(1, 46) = 15.07, p <0.001, _adjustR² = 0.23, β = -0.50. For the same analysis performed by group, only the association between cabinet searching and memory composite score in MCI participants remained significant, F(1, 20) = 7.23, p <0.01, _adjustR² = 0.23, β = -0.52.

Performance-based measures of IADL

Performance-based measures of IADL in MCI and CH groups are presented in Table 4. In regard to the completion score, CH participants achieved more objectives than MCI participants in the five scripted tasks. The MCI group did not ask for more assistance from experimenters than the CH group. However, experimenters decided to intervene more often with MCI participants without their explicit request, compared to CH participants. Globally, the MCI group made more errors than the CH group. All error types were observed in the MCI group, but omission and substitution were the most frequent errors.

Table 4

Performance-based measures of IADLs

	No-MCI n=26	MCI n=22	F(1, 46)
Completion score (standard deviation)
Objectives completed (/19)	17.2 (1.6)	13.9 (2.0)	40.26, p <0.001
Number of assistances (standard deviation)
Total assistance given	2.3 (2.9)	7.1 (6.3)	11.31, p <0.001
Total solicited assistances	1.3 (1.7)	2.8 (3.9)	3.39, p = 0.07
Total unsolicited assistances	1.12 (1.70)	4.23 (3.19)	18.53 p <0.001
Number of errors (standard deviation)
Total errors	3.17 (3.09)	9.50 (4.18)	36.21, p <0.001
Inefficient search	0.77 (0.99)	2.52 (2.49)	10.88, p <0.01
Omission	1.33 (1.40)	3.41 (1.50)	24.55, p <0.001
Substitution	0.50 (0.66)	1.89 (1.11)	28.50, p <0.001
Perseveration	0.12 (0.26)	0.66 (1.22)	4.93, p <0.05
Not complying	0.33 (0.63)	0.96 (1.13)	5.85, p <0.05

In a subset of analyses, we looked if there were differences in performances-based measures of IADL between tasks. MCI participants made more errors in tasks 2 (cooking) F(1, 46) = 28.85, p <0.001, 3 (cleaning bathroom) F(1, 46) = 10.48, p <0.01, 4 (bus phone) F(1, 46) = 7.83, p <0.01, and 5 (cloth folding) F(1, 46) =5.35, p <0.05. Also, MCI participants required more assistance and had lower scores in tasks 2, 3, and 4 (respectively F(1, 46) = 8.04, 3.47, 5.31, ps <0.05. No difference was detected in Task 1 (put away coat). Also, interestingly, men and women performances were comparable, though the sample might be too small to reach a conclusion.

The association between cognitive composite scores and performance-based measures was assessed. The memory composite score predicted the objective completion score, F(1, 46) = 21.32, p <0.001, _adjustR² = 0.30, β = 0.56. While the executive composite score was associated with the amount of unsolicited assistance received, F(1, 46) = 22.14, p <0.001, _adjustR² = 0.31, β = -0.57), neither memory nor executive composite scores were associated with the assistance sought. Finally, the total number of errors was predicted by both the memory composite score, F(1, 46) = 32.42, p <0.001, _adjustR² = 0.40, β = -0.64, and the executive composite score, F(1, 46) = 15.19, p <0.001, _adjustR² = 0.23, β = -0.50. When the same analysis was performed by group, only the association between the number of errors and the memory composite score in MCI participants remained significant, F(1, 20) = 5.59, p <0.05, _adjustR² = 0.18, β = -0.46. Therefore, an IADL performance-based composite score was calculated by combining the completion score, the amount of assistance interventions, and the number of errors.

Predictive relationship between IADL

performance-based composite scores, cognitive performance composite scores, and MCI clinical diagnosis.

First, simple regressions were calculated with each composite score. All composite scores significantly predicted MCI clinical diagnosis (F(1, 46) = 12.18, p <0.001, _adjustR² = 0.19, β = 0.30 for sensor-based observations; F(1, 46) = 35.50, p <0.001, _adjustR² = 0.32, β = 0.36 for performance-based measures; F(1, 46) = 27.96, p <0.001, _adjustR² = 0.37, β = -0.37 for executive function, F(1, 46) = 47.76, p 0<.001, _adjustR² = 0.50, β = -0.24 for memory). When all composite scores were entered in a stepwise regression to predict MCI diagnosis, only sensor-based observations were not kept in the model (see model 1 on Table 5). However, when cognitive composite scores were not entered, both direct and sensor-based observations predicted MCI diagnosis (model 2). Moreover, both direct and sensor-based observations predicted memory and executive function composite scores (models 3 and 4).

Table 5

Summary of stepwise regression analyses in relation to MCI diagnostic and composite scores

Predicted variable	Entered Variables	B	SE	R²	F	Sig
1. MCI diagnosis	Memory	-0.14	0.04	0.62	26.62	<0.001
	Executive	-0.17	0.06
	Direct IADL	-0.15	0.07
	Sensor-based IADL	-	-
2. MCI diagnosis	Direct IADL	-0.32	0.06	0.47	22.01	<0.001
	Sensor-based IADL	-0.17	0.08
3. Memory	Direct IADL	-0.81	0.19	0.47	21.80	<0.001
	Sensor-based IADL	-0.68	0.22
4. Executive	Direct IADL	-0.36	0.12	0.29	10.57	<0.001
	Sensor-based IADL	-0.33	0.15

DISCUSSION

The main objective of this study was to determine whether simple and affordable wireless sensor-based technology, used in two different smart research environments, could add value to performance-based measures of IADL when it comes to predicting MCI in older adults. Participants were invited to perform five scripted IADL tasks in a laboratory testbed. The MCI group had significantly lower performances than the CH group on the scripted IADL task, according to both sensor-based and performance-based measures of IADL. With sensors, it was observed that participants with MCI spent more time than CH participants in the kitchen and looking into the fridge or the kitchen cabinets. For performance-based measures, older adults with MCI completed fewer objectives, required more unsolicited assistance and made more errors than CH participants. Moreover, sensor-based observations were associated with memory and executive performances, and significantly contributed to the prediction of MCI when no cognitive performance was included in the model. The findings from this study suggest that smart environment devices can help to detect significant differences between healthy individuals and individuals with MCI performing IADLs.

Concerning performance-based measures of IADLs, our findings correspond to those of previous studies. In Jekel et al. [29], participants with MCI showed more searching and task-irrelevant behavior than CH participants, as in the present study. Also, we observed that MCI received more unsolicited assistance than CH individuals, similar to Seelye et al. [47] who observed greater need for prompting assistance in the multi-domain amnestic MCI. Experimenters in our study were not blinded to group assignment, which is a potential limitation. Even though instructions were given to deliver unsolicited assistance objectively, according to subjects’ difficulties, possible bias in interventions cannot be completely excluded in the present setting. Interestingly, we observed that the amount of unsolicited assistance was associated with poorer executive function scores in participants, regardless of diagnosis. Therefore, it might be that individuals with weaker executive functioning are not seeking the help they need in situations where they experience difficulties, potentially because they fail to disengage from their initial plan or because they are more impulsive. However, it is also possible that examiners trained in occupational therapy tend to detect errors and difficulties and to spontaneously support hesitant individuals while they are performing IADL. Thus, future studies should attempt to keep blinded all those who interact with participants in order to reduce potential biases.

To determine whether sensor-based data were comparable across different sites, participants were tested in two different smarthome laboratories. Sensor data collected were quite comparable in the two research laboratories, which indicates that similar protocols can be used in different settings. Such smart apartments could be installed in various clinical settings and be used as part of a larger formal evaluation to detect MCI. The only noticeable exception was that Montreal participants spent more time in the living room. In the Montreal apartment, participants had to press one additional digit before making an external call, while this was not required in Sherbrooke and some participants struggled with that additional action.

Our analyses present some limits. Sensor data were mainly used to calculate duration. In preliminary analysis, we also looked at the number of times each sensor was triggered, but this approach proved to be limited. First, regarding motion sensors, the number of times a motion sensor triggered could not be used as a proxy for activity level, movement or speed, but could solely indicate presence in a specific room, given the 45-60 s refresh rate of motion sensors. Second, regarding contact sensors (e.g., fridge, cabinet), the total amount of time spent searching for objects was a more sensitive indicator in participants with MCI than the number of door openings. This could either be due to MCI participants spending more time finding a target among distractors or because the time spent searching is a continuous variable instead of a discrete variable. Another sensor-based analysis approach would have been to analyze if the sequence in which sensors were activated was significantly related to a specific group. However, the protocol used in the present study was designed to be unstructured in some aspects. This is because, as stated by Lezak [48]: “the more open-ended and unstructured the task, the more likely will impairments in programming become evident (p. 290)”. We anticipated that this unstructured approach would facilitate differentiation between MCI and CH older adults. However, the presence of unstructured aspects may also involve many optimal or possible orders for task performance, and creates a variety of sequences that cannot be statistically appraised. Moreover, difficulties that arise in unstructured contexts can be especially apparent in the formulation of goals, during planning and verification of goal attainment [49]. The problem in our study was that these difficulties can be more adequately assessed by a trained observer, while sensor data only represent the task execution segment. It should be pointed out that, by focusing mainly on time-based results, having an observant managing several timers might have brought similar results to those brought here by the sensors. However, we believe this is but a first step and that sensor analyses will have more potential as we gain a better understanding of how to use them. Another limit is that the nature of the tasks selected may have impacted the efficacy of sensor-based analyses. Only sensor-based measures of Task 2 (cooking) were significantly related to MCI, while significant differences for Tasks 3, 4, and 5 were observed with performance-based measures. This is coherent with the literature [50]: Task 2 had the most instructions and therefore, the most objectives to achieve. It also involved several interactions with everyday technology, which creates difficulties in people with MCI [51]. On the one hand, this could suggest that sensor-based measures lack sensitivity when it comes to simple tasks. On the other hand, sensor data from the present study were highly driven by performance duration, while behavioral data were more related to the quality of the performance. While the two are often related, as demonstrated in the present study, an inaccurate behavior can sometimes take the same or even less amount of time (e.g., omission and impulsivity). Moreover, Seligman et al. [52] suggest that subtle errors are more present in MCI than overt errors. This raises a great challenge when it comes to smart home binary sensors, as subtle errors, especially when not self-corrected, do not always slow down the performance.

In consideration of those limits, we offer recommendations for future studies on scenario-based evaluation. First, a greater number of sensors might not necessarily help to detect a wider variety of subtle errors. Besides the increase in costs for observants, this kind of errors may need other measures than those provided by binary sensors. Instead of using more sensors for the detection of MCI in smart environments, we recommend a more structured approach in IADL task execution, with a clear optimal path for the completion of all tasks. This structure should still require careful planning and be coherent with everyday IADL routine. It should also allow for binary sensors to easily detect when a participant is diverging from the optimal answer (e.g., first do A, when everything is done do B, after A do C, while waiting the end of B, please do D, when the timer beeps, do E, etc.). Second, daily tasks performed should be fairly complex and involve more interactions with simple technology (e.g., setting an alarm clock in the bedroom, washing a load of laundry in the bathroom, put 37.85 in an envelope and add the address for a specific bank that is in the telephone directory, etc.). Third, tasks should be designed in a way that allows the evaluator to simply visit each room and evaluate the quality of the performance based on final results, without being required to watch the whole execution (e.g., is the amount of money in the envelope correct?). The absence of observers would 1) encourage independent problem-solving, 2) prevent observer interventions and bias, and 3) be more resource-efficient. Fourth, as sensor-based data are more suitable for timing task execution, uncontrollable waiting time (e.g., holding the line) should be avoided, as well as tasks that could be performed in many different ways (e.g., having the possibility to either boil or fry the eggs). We believe that, in the context of predicting MCI with sensors, a more structured execution of IADL and fewer unwanted variations would help to clarify what is the optimal performance from the standpoint of sensor-based analysis. Methods based on machine learning for the automatic identification of behavioral patterns could also be more effective in this context and improve the predictive value of this new technological approach [53, 54]. Still, machine-learning usually requires large datasets to achieve generalizations and accuracy, so this approach would be more relevant with large samples of participants [55].

In conclusion, the aim of this study was to explore if a minimal, affordable, non-intrusive kit of wireless sensors could improve detection of MCI through scenario-based formal evaluation. Using smart apartment technologies in two laboratories, we were able to detect significant differences between a group of individuals with MCI and a healthy control group in relation to the performance of scenario-based evaluations of IADL. Moreover, sensor-based observations were not only associated with clinical diagnosis of MCI but were also indicative of cognitive performance. While the relation is modest, it is encouraging given the small sample size. Sensor-based environments could thus be used in combination with more traditional clinical evaluations (e.g., scenario-based assessments) to improve MCI detection.

Footnotes

ACKNOWLEDGMENTS

This study was supported by the Quebec Network of research in Aging and the INTER network (Interactive technologies of engineering in rehabilitation). ML was supported by a postdoctoral award from Fonds de recherche du Québec – Santé (FRQS). NB, CH, and MG received salary support from the FRQS.

Authors’ disclosures available online ().

References

Barnett

, Lewis

, Blackwell

, Taylor

(2014) Early intervention in Alzheimer’s disease: a health economic study of the effects of diagnostic timing. BMC Neurol 14, 101.

Robinson

, Tang

, Taylor

J-P

(2015) Dementia: timely diagnosis and early intervention. BMJ 350, h3029.

Rodakowski

, Saghafi

, Butters

, Skidmore

(2015) Non-pharmacological interventions for adults with mild cognitive impairment and early stage dementia: An updated scoping review. Mol Aspects Med 43, 38–53.

Van Vliet

, de Vugt

, Bakker

, Koopmans

, Verhey

(2010) Impact of early onset dementia on caregivers: a review. Int J Geriatr Psychiatry 25, 1091–1100.

Bateman

, Xiong

, Benzinger

, Fagan

, Goate

, Fox

, Marcus

, Cairns

, Xie

, Blazey

(2012) Clinical and biomarker changes in dominantly inherited Alzheimer’s disease. N Engl J Med 2012, 795–804.

Villemagne

, Burnham

, Bourgeat

, Brown

, Ellis

, Salvado

, Szoeke

, Macaulay

, Martins

, Maruff

(2013) Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer’s disease: a prospective cohort study. Lancet Neurol 12, 357–367.

Gauthier

, Reisberg

, Zaudig

, Petersen

, Ritchie

, Broich

, Belleville

, Brodaty

, Bennett

, Chertkow

(2006) Mild cognitive impairment. Lancet 367, 1262–1270.

Petersen

(2004) Mild cognitive impairment as a diagnostic entity. J Intern Med 256, 183–194.

Chen

, Denny

, Harvey

, Farias

, Mungas

, DeCarli

, Beckett

(2017) Progression from normal cognition to mild cognitive impairment in a diverse clinic-based and community-based elderly cohort. Alzheimers Dement 13, 399–405.

10.

Jekel

, Damian

, Wattmo

, Hausner

, Bullock

, Connelly

, Dubois

, Eriksdotter

, Ewers

, Graessel

(2015) Mild cognitive impairment and deficits in instrumental activities of daily living: a systematic review. Alzheimers Res The 7, 17.

11.

Farias

, Mungas

, Reed

, Harvey

, Cahn-Weiner

, DeCarli

(2006) MCI is associated with deficits in everyday functioning. Alzheimer Dis Assoc Disor 20, 217.

12.

Pérès

, Helmer

, Amieva

, Orgogozo

J-M

, Rouch

, Dartigues

J-F

, Barberger-Gateau

(2008) Natural history of decline in instrumental activities of daily living performance over the 10 years preceding the clinical diagnosis of dementia: a prospective population-based study. J Am Geriatr Soc 56, 37–44.

13.

Farias

, Lau

, Harvey

, Denny

, Barba

, Mefford

(2017) Early functional limitations in cognitively normal older adults predict diagnostic conversion to mild cognitive impairment. J Am Geriatr Soc 65, 1152–1158.

14.

Devanand

, Liu

, Brown

(2017) Impact of functional deficits in instrumental activities of daily living in mild cognitive impairment: a clinical algorithm to predict progression to dementia. Alzheimer Dis Assoc Disord 31, 55–61.

15.

Di Carlo

, Baldereschi

, Lamassa

, Bovis

, Inzitari

, Solfrizzi

, Panza

, Galluzzo

, Scafato

, Inzitari

(2016) Daily function as predictor of dementia in cognitive impairment, no dementia (CIND) and mild cognitive impairment (MCI): an 8-year follow-up in the ILSA study. J Alzheimers Dis 53, 505–515.

16.

Bottari

, Dassa

, Rainville

, Dutil

(2009) The criterion-related validity of the IADL Profile with measures of executive functions, indices of trauma severity and sociodemographic characteristics. Brain Inj 23, 322–335.

17.

Bell-McGinty

, Podell

, Franzen

, Baird

, Williams

(2002) Standard measures of executive function in predicting instrumental activities of daily living in older adults. Int J Geriatr Psychiatry 17, 828–834.

18.

Mioshi

, Kipps

, Dawson

, Mitchell

, Graham

, Hodges

(2007) Activities of daily living in frontotemporal dementia and Alzheimer disease. Neurology 68, 2077–2084.

19.

Barberger-Gateau

, Fabrigoule

, Helmer

, Rouch

, Dartigues

(1999) Functional impairment in instrumental activities of daily living: an early clinical sign of dementia? J Am Geriatr Soc 47, 456–462.

20.

Okonkwo

, Wadley

, Griffith

, Belue

, Lanza

, Zamrini

, Harrell

, Brockington

, Clark

, Raman

(2008) Awareness of deficits in financial abilities in patients with mild cognitive impairment: going beyond self-informant discrepancy. Am J Geriatr Psychiatry 16, 650–659.

21.

Tabert

, Albert

, Borukhova-Milov

, Camacho

, Pelton

, Liu

, Stern

, Devanand

(2002) Functional deficits in patients with mild cognitive impairment prediction of AD. Neurology 58, 758–764.

22.

Debettignies

, Mahurin

, Pirozzolo

(1990) Insight for impairment in independent living skills in Alzheimer’s disease and multi-infarct dementia. J Clin Exp Neuropsychol 12, 355–363.

23.

Doble

, Fisk

, Rockwood

(1999) Assessing the ADL functioning of persons with Alzheimer’s disease: comparison of family informants’ ratings and performance-based assessment findings. Int Psychogeriatr 11, 399–409.

24.

Belchior

, Korner-Bitensky

, Holmes

, Robert

(2015) Identification and assessment of functional performance in mild cognitive impairment: a survey of occupational therapy practices. Aust Occup Ther J 62, 187–196.

25.

Schmitter-Edgecombe

, Parsey

(2014) Assessment of functional change and cognitive correlates in the progression from healthy cognitive aging to dementia. Neuropsychology 28, 881.

26.

Lussier

, Lavoie

, Giroux

, Consel

, Guay

, Macoir

, Hudon

, Lorrain

, Talbot

, Langlois

(2018) Early detection of mild cognitive impairment with in-home monitoring technologies using functional measures: A systematic review. IEEE J Biomed Health Inform, doi: 10.1109/JBHI.2018.2834317.

27.

Dawadi

, Cook

, Schmitter-Edgecombe

, Parsey

(2013) Automated assessment of cognitive health using smart home technologies. Technol Health Care 21, 323–343.

28.

Cook

, Schmitter-Edgecombe

, Dawadi

(2015) Analyzing activity behavior and movement in a naturalistic environment using smart home techniques. IEEE J Biomed Health Inform 19, 1882–1892.

29.

Jekel

, Damian

, Storf

, Hausner

, Frölich

(2016) Development of a proxy-free objective assessment tool of instrumental activities of daily living in mild cognitive impairment using smart home technologies. J Alzheimers Dis 52, 509–517.

30.

Domingos

(2012) A few useful things to know about machine learning. Commun ACM 55, 78–87.

31.

Larouche

, Tremblay

M-P

, Potvin

, Laforest

, Bergeron

, Laforce

, Monetta

, Boucher

, Tremblay

, Belleville

(2016) Normative data for the Montreal Cognitive Assessment in middle-aged and elderly Quebec-French people. Arch Clin Neuropsychol 31, 819–826.

32.

Winblad

, Palmer

, Kivipelto

, Jelic

, Fratiglioni

, Wahlund

, Nordberg

, Bäckman

, Albert

, Almkvist

(2004) Mild cognitive impairment–beyond controversies, towards a consensus: report of the International Working Group on Mild Cognitive Impairment. J Intern Med 256, 240–246.

33.

Chikhaoui

, Wang

, Pigot

(2011) A frequent pattern mining approach for ADLs recognition in smart environments. Advanced Information Networking and Applications (AINA), 2011 IEEE International Conference on IEEE, pp. 248–255.

34.

Bier

, Belchior

PdC

, Paquette

, Beauchemin

, Lacasse-Champagne

, Messier

, Pellerin

M-L

, Petit

, Mioshi

, Bottari

(2016) The instrumental activity of daily living profile in aging: A feasibility study. J Alzheimers Dis 52, 1361–1371.

35.

Bottari

, Dassa

, Rainville

, Dutil

(2009) The factorial validity and internal consistency of the Instrumental Activities of Daily Living Profile in individuals with a traumatic brain injury. Neuropsychol Rehabil 19, 177–207.

36.

Shallice

, Burgess

(1991) Deficits in strategy application following frontal lobe damage in man. Brain 114, 727–741.

37.

Le Dorze

, Villeneuve

, Zumbansen

, Masson-Trottier

, Bottari

(2014) Verbal assistance within the context of an IADL evaluation. Open J Ther Rehabil 2, 182.

38.

Tomaszewski Farias

, Schmitter-Edgecombe

, Weakley

, Harvey

, Denny

, Barba

, Gravano

, Giovannetti

, Willis

(2018) Compensation strategies in older adults: association with cognition and everyday function. Am J Alzheimers Dis Other Demen 33, 184–191.

39.

Marshall

, Rentz

, Frey

, Locascio

, Johnson

, Sperling

(2011) Executive function and instrumental activities of daily living in mild cognitive impairment and Alzheimer’s disease. Alzheimers Dement 7, 300–308.

40.

Overdorp

, Kessels

, Claassen

, Oosterman

(2016) The combined effect of neuropsychological and neuropathological deficits on instrumental activities of daily living in older adults: a systematic review. Neuropsychol Rev 26, 92–106.

41.

Miyake

, Friedman

(2012) The nature and organization of individual differences in executive functions: four general conclusions. Curr Dir Psychol Sci 21, 8–14.

42.

Tombaugh

(2004) Trail Making Test A and B: normative data stratified by age and education. Arch Clin Neuropsychol 19, 203–214.

43.

Delis

, Kramer

, Kaplan

, Holdnack

(2004) Reliability and validity of the Delis-Kaplan Executive Function System: an update. J Int Neuropsychol Soc 10, 301–303.

44.

Nasreddine

, Phillips

, Bédirian

, Charbonneau

, Whitehead

, Collin

, Cummings

, Chertkow

(2005) The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc 53, 695–699.

45.

Lavoie

, Bherer

, Joubert

, Gagnon

J-F

, Blanchet

, Rouleau

, Macoir

, Hudon

(2018) Normative data for the Rey Auditory Verbal Learning Test in the older French-Quebec population. Clin Neuropsychol 32(sup1), 15–28.

46.

Giovannetti

, Libon

, Buxbaum

, Schwartz

(2002) Naturalistic action impairments in dementia. Neuropsychologia 40, 1220–1232.

47.

Seelye

, Schmitter-Edgecombe

, Cook

, Crandall

(2013) Naturalistic assessment of everyday activities and prompting technologies in mild cognitive impairment. J Int Neuropsychol Soc 19, 442–452.

48.

Lezak

(1982) The problem of assessing executive functions. Int J Psychol 17, 281–297.

49.

Bottari

, Dassa

, Rainville

, Dutil

(2009) , The criterion-related validity of the IADL Profile with measures of executive functions, indices of trauma severity and sociodemographic characteristics. Brain Inj 23, 322–335.

50.

Goldberg

, Koppel

, Keehlisen

, Christen

, Dreses-Werringloer

, Conejero-Goldberg

, Gordon

, Davies

(2010) Performance-based measures of everyday function in mild cognitive impairment. Am J Psychiatry 167, 845–853.

51.

Nygård

, Starkhammar

(2007) The use of everyday technology by people with dementia living alone: Mapping out the difficulties. Aging Ment Health 11, 144–155.

52.

Seligman

, Giovannetti

, Sestito

, Libon

(2014) A new approach to the characterization of subtle errors in everyday action: implications for mild cognitive impairment. Clin Neuropsychol 28, 97–115.

53.

Chikhaoui

, Bier

, Lussier

, Gagnon

, Giroux

, Pigot

(2018) Automatic identification of behavior patterns in mild cognitive impairments and Alzheimer’s disease based on activities of daily living, Mokhtari

, Abdulrazak

, Aloulou

, eds. In Smart Homes and Health Telematics, Designing a Better Future: Urban Assisted Living. Springer International Publishing, https://doi.org/10.1007/978-3-319-94523-1_6

54.

Caroux

, Consel

, Dupuy

, Sauzéon

(2014) Verification of daily activities of older adults: a simple, non-intrusive, low-cost approach. In Proceedings of the 16th International ACM SIGACCESS Conference on Computers & Accessibility. ACM, pp. 43–50.

55.

Alsheikh

, Lin

, Niyato

, Tan

H-P

(2014) Machine learning in wireless sensor networks: Algorithms, strategies, and applications. IEEE Commun Surv Tutor 16, 1996–2018.