Medical Informatics
Master Thesis September 18, 2017Family caregiver opinions on
sensor assisted living situations for
older adults.
Author:
Guido M. Out
Mentor: Prof. Dr. George Demiris Supervisor: Prof. Dr. Monique W.M. Jaspers
University of Washington
Department of Biomedical Informatics and Medical Education
Universiteit van Amsterdam
SRP Details
Student:
BSc. Guido M. Out
Student number: 10774319
Email: g.m.out@amc.uva.nl / outg@uw.edu
Location of Scientific Research Project: University of Washington, Seattle, WA, USA
Department of Biomedical Informatics and Medical Education Period: September 2016 - September 2017
Mentor:
Prof. Dr. George Demiris University of Washington
Department of Biomedical Informatics and Medical Education Email: gdemiris@uw.edu
Tutor:
Prof. Dr. Monique W.M. Jaspers PhD
Academic Medical Center - University of Amsterdam Department of Medical Informatics
Samenvatting
Het is gunstig voor ouderen om ge¨ınformeerd te zijn over hun eigen gezond-heidsstatus om zo gericht maatregelen te kunnen nemen als ze verandering zien. Zogenaamde ’smart home’ sensoren kunnen hierbij helpen door inzicht te geven in de gezondheidsstatus en leefpatronen van hun gebruikers. Ac-ceptatie en gebruiksvriendelijkheid zijn de sleutel voor het succes van deze technologie¨en en met de toenemende mate waarin ouderen last kunnen kri-jgen van cognitieve en functionele beperkingen die een hogere leeftijd met zich mee brengt kan het moeilijker worden om de gezondheidsstatus te mon-itoren. Jongere vrienden of familieleden zijn mogelijk beter bekend met digi-tale interfaces en data visualizaties en kunnen een rol spelen in het managen en interpreteren van gezondheidsdata. Het is nog onbekend of deze jongere groep dezelfde opvattingen heeft over de potentiele voordelen en nadelen (zoals het verlies van privacy) van dit soort sensoren.
In dit onderzoek zijn met behulp van online vragenlijsten de opvattingen van mantelzorgers in kaart gebracht op het gebied van de voor- en nadelen van ’smart home’ sensoren. Deze vragenlijsten zijn verspreid via twee verschil-lende Noord Amerikaanse websites. De onderzoekspopulatie bestond uit 403 Amerikaanse mantelzorgers die zorg verleenden aan een persoon van 65 jaar of ouder. Kwantitatieve data over 12 verschillende sensoren in 3 categorie¨en werd verzameld met behulp van Likert-schaal vragen. Kwalitatieve data over de drie categorie¨en werd verzameld met behulp van vrije-tekst velden. Onze onderzoekspopulatie was overwegend positief in alle categorie¨en, wat lijkt aan te geven dat er interesse is in het gebruik van sensoren in een thuis-zorg situatie. Toekomstige studies zullen moeten onderzoeken of sensoren daadwerkelijk invloed hebben op de gezondheidsstatus van ouderen.
Abstract
Background: It is beneficial for older adults to stay informed about their health so they can take appropriate measures when noticing changes in phys-ical or behavioral parameters. In-home sensors could provide a means for older adults to gain insight into their own living patterns, daily activities, and health status. The integration of such systems with electronic patient records would allow for instant communication of information with all in-volved parties like family members and health care providers. Acceptance and usability are key in the success of self-monitoring, and with increasing levels of cognitive or functional impairment that may come in some cases with aging, self-monitoring can become harder for individuals. Younger family caregivers might be more familiar with digital data forms and data visualizations, and can play a role in the process of monitoring by acting as information managers. However, we do not know if these caregivers have the same attitudes as their older relatives towards the potential benefits and disadvantages (such as a perceived loss of privacy) of such sensors.
Objective: Identify family caregiver attitudes and perceptions on using sensors in their caregiving situation.
Methods: Online survey regarding usefulness, perceived potential breach of privacy and overall desirability of body-worn, object mounted and en-vironmental sensors for family caregivers of individuals 65 years of age or older. Recruitment and collection through Amazon Mechanical Turk and Research Match platforms.
Results: 238 completed surveys from Amazon and 175 completed surveys from Research Match. Perceived positive value outweighs negative percep-tions in all categories (Body-worn: AM 73%/RM 65%—Object mounted: AM 60%/RM 57%—Environmental: AM 57%/RM 49%)
Findings: Large majority of subjects is positive about sensor use in moni-toring older adults. Human centered sensors (sensors that collect data about human behavior) received more positive response than environmental sen-sors (sensen-sors that collect data about characteristics of the living space). Conclusion: Family caregivers in our study were mostly positive about the possibility of using sensors to monitor their loved ones. Future studies should focus on the impact that such monitoring may have on clinical out-comes and the support of successful aging.
Contents
1 Introduction 2 1.1 Background . . . 2 1.2 Research Question . . . 5 2 Methods 6 2.1 Creation of survey . . . 6 2.2 Validation of survey . . . 9 2.3 Deployment . . . 10 2.4 Data analysis . . . 11 3 Results 12 3.1 Demographic data . . . 12 3.2 Amazon results . . . 153.3 Research Match results . . . 17
3.4 Demographic analysis . . . 17
4 Discussion 24 4.1 Findings . . . 24
4.2 In context . . . 26
4.3 About the study . . . 27
4.4 Conclusion . . . 29
5 Appendix 35 5.1 Survey . . . 36
Chapter 1
Introduction
1.1
Background
The number of people with chronic health conditions in western countries is growing [1] and ensuring quality of care for these people is considered a significant challenge for healthcare systems. With current demographic shifts the amount of older adults living with one or more morbidities is projected to grow exponentially, while the amount of people available to provide care to them decreases [1]. Worldwide 70%-80% of impaired older adults are being cared for by a family member [1]. The role of these family members as informal caregivers will become increasingly important for the growing number of older adults trying to age in place.
Smart Home Sensors
One way to allow older adults and their family to stay informed about and in control of their health status involves sensor-based technologies designed to detect and record a person’s activities and environmental characteristics within their living space [2]. The concept of a ”smart home” has been intro-duced to describe residential settings with embedded sensor technologies to facilitate passive monitoring of residents and promote their well-being and safety [3]. The effectiveness of sensor-based technologies has been studied in the past [2, 4, 5] and recommendations for designing and implementing such technologies are available. A large challenge that still remains today in terms of the diffusion of ”smart home” technologies is the acceptance of these technologies by the older adults intended to use them. Acceptance is a crit-ical factor in predicting usage rates and successful implementation of infor-mation technologies. Factors such as age, gender, experience, voluntariness and cultural values have a great impact on potential users’ perceptions [6,7]. Previous studies have investigated acceptance of smart home technologies among older adults. In a study by Courtney et al. [8] older adults were found to rarely base their acceptance of a new technology solely on their privacy.
For participants who did have privacy concerns, these concerns were not as important as their perception of their need for the technology [8]. Factors influencing this perception were obtained from focus sessions with groups of older adults. Most participants indicated a preference for being able to select only the technology or technologies they perceived they needed, but also consistently described themselves as ’healthy’, ’very healthy’ or ’blessed with good health all my life’. [8] This could cause underestimation of the perceived need for technologies, resulting in a lower acceptance rate. Peek et al. found multiple factors influencing technology acceptance in a pre-implementation stage [9], among which were the ”need for technology”, ”desire to age in place”, ”privacy implications”, ”high costs”, and ”expected increased safety” [9]. Both studies found older adult acceptance to be influ-enced by their perceived need for the proposed technology, making successful implementation highly dependent on personal views. Mortenson et al. found a more philosophical factor influencing acceptance, with participants of their qualitative study reporting that new forms of surveillance would alter the perception of home. This subjective feeling of home can be disrupted by be-ing aware of monitorbe-ing equipment, makbe-ing a monitored individual feel less at ease and perhaps even causing them to change their living patterns [10]. In addition to these personal views, cultural views can also cause signifi-cantly different opinions in areas affecting acceptance, as was found in a study by Jeong et al. investigating preferences influencing the design and operation of smart homes. Opinions were collected using surveys among US and South Korean populations, which resulted in five different factors with statistically significant different preferences between the two cultures [11]. The five factors were 1) environmental connection and control; 2) smart de-vices (appliances) and their control; 3) physical safety and security concerns; 4) comfort and relaxation issues; 5) control restriction issues [11]. ’Environ-mental connection and control’ and ’smart devices (appliances) and their control’ seemed to be the most influential factors for Americans and Kore-ans, respectively. A similar study by Chung et al. comparing the opinions of Koreans and Korean Americans also found cultural differences with Ko-rean older adults regarding the governments role as more important in the adoption and use of new technologies [12]. By focusing on cultural differ-ences like these manufacturers can increase the likelihood of acceptance in multiple geographical markets. All studies described in the section above list perception in one form or another as a factor influencing the acceptance of smart home technologies. This perception can be influenced by culture, experience, self-awareness, expectations, and many other factors, making it difficult to standardize
Sensor use in home care
Home care has been subjected to standardization to ensure a minimum quality of care is provided in an efficient manner. To assess this quality a tool called the Resident Assessment Instrument for Home-Care (RAI-HC) [13] can be used. The basis for this assessment tool is the Minimum Data Set for Home-Care (MDS-HC); a questionnaire assisting the caregivers in assessing a patient’s situation in the areas of physical, mental, cognitive and social health [14]. However, reviewers of this tool were uncertain about the validity of the results, especially over long periods of time, since the questionnaire uses self-reported data from patients [15]. In addition to being prone to over- or underestimation [16] self-reported data can be influenced by both patient and care provider, reducing the validity of the results [15]. To create valid data on the patient’s health status that can assist care providers in validating self-reported data, Nienhold et al. created a concept for a sensor setup that automatically produces valid data on 42% of all questions in the MDS-HC [17]. Their set was based on a minimal set of ambient and non-invasive body sensors, ensuring the feasibility and cost-effectiveness of such a system. Several questions in the MDS-HC requiring more invasive sensors were left unanswered in their sensor setup. The goal of sensor monitoring is only to assist in the assessment process and not to create an all-encompassing health status report, so even with this amount of information a similar setup can prove useful in a large number of situations.
Family Caregiving in the Context of Smart Homes
A family caregiver’s demographics and characteristics may have a signif-icant influence on their physical and mental wellbeing, and indirectly on their perceptions of caregiving. A previous study found some differences be-tween perceived burden of caregiving in the different genders [18]. Women were found to experience a higher burden from providing care, and were also found to use emotion-focused coping and other ineffective coping styles such as fantasy, wishful thinking, denial, escape or avoidance more frequently than men [19,20]. Caregiver wellbeing was also found to be influenced by the degree with which their relative was affected by severe mobility limitations (Performance-Oriented Mobility Assessment), social disability (instrumental activities of daily living), neuropsychiatric disturbances related to cognitive decline (Neuropsychiatric Inventory), and depressive symptoms, all of which increase the likelihood of psychological distress in the caregiver [21]. Using remotely accessible sensors to monitor older adults might reduce the emo-tional burden on family caregivers by allowing less intensive, more frequent checkups on their loved one. In this study, we aimed to determine how much loss of privacy caregivers were willing to place upon their relative to facili-tate the provision of care and obtain their perceptions about the utility of
various distinct sensors systems. Increasing the ease with which family care-givers can monitor their relative might promote their involvement in their family member’s health status. Since many health trends tend to progress gradually over longer periods of time, it can be hard for someone to notice a significant change in their own daily patterns. An outside observer with access to long term data might be able to make relevant observations with-out having to visit every day or rely solely on self-report. Placing sensors in a private space can raise privacy concerns, and especially when the output of the sensors is designed to be viewed remotely it is critical that all parties consent to installing these devices. Caregivers for example, might be more willing to place sensors in an area where they may be perceived as intru-sive or unnecessary to the older adults being monitored. Family caregivers might also like to have more data than older adults themselves are willing to provide, but it is currently still unknown where these two groups show alignment in their perceptions and preferences. Thus, we needed to examine what family caregivers look for in smart home sensor data and how much burden they were willing to place on their family members to obtain infor-mation they consider important. We assessed family caregiver opinions in this subject using an online survey about a number of minimally invasive body-worn, object mounted, and environmental sensors, based on the set of sensors compiled by Nienhold et al.
1.2
Research Question
Main research question
How do family caregivers of older adults value the potential benefits of sonal and environmental sensor data in relation to the care recipients’ per-ceived loss of privacy resulting from the use of these sensors?
Additional research questions
1. What type of sensor do family caregivers think is most useful to have installed in their loved one’s home?
2. What type of sensor do family caregivers think has the largest impact on the privacy of their loved one?
3. What is the general opinion of family caregivers on using sensors to monitor older adults?
Chapter 2
Methods
To answer the main research question, we investigated whether family care-givers perceive specific sensors as useful for their caregiving situation, and whether a perceived breach of privacy caused by these sensors would out-weigh the benefit of the information that is produced. Opinions on these topics were collected using an online survey to reach a large and diverse population of caregivers.
2.1
Creation of survey
Our research question is a general inquiry into the topic of sensor use for monitoring older adults. The market for this type of sensor is growing rapidly, and a multitude of sensors is available for purchase. We based our list of sensors on the set created by Nienhold et al. described in the intro-duction, ensuring a wide variety of applications, high cost-effectiveness, and minimal invasiveness, while also providing data that are considered useful in a caregiving scenario. We included all sensors described by Nienhold et al. and rearranged the list into three separate categories to allow for data collection on a higher level in addition to the individual sensor level.
Body-worn sensors
We placed two sensors in the body-worn sensor category. These sensors are worn on the body, usually the wrist or hip, and measure the wearer’s movement or activities.
1. Body position sensors
Body position sensors is a general term for all sensors that monitor the wearer’s orientation, speed or posture. For the purpose of our study we did not select a specific commercially available sensor but devised a theoretical combination of a gyroscope (producing data on rotation of the body), an accelerometer (producing data on relative speeds), and
a magnetometer (producing data on posture), all three of which were included separately in the list by Nienhold et al. Such a device could produce data on walking speed, sleeping patterns, and fall scenarios. 2. Radio Frequency ID tags
Radio Frequency ID (RFID) tags are small adhesive transmitter chips that can be applied to virtually any object. A user wearing a device designed to interact with these tags records data on interaction dura-tion and frequency with tagged objects. Because all tags have a unique signature, monitoring these interactions can produce a detailed pat-tern of interactions in a daily routine. RFID technology requires the tagged object to be manually operated because the monitoring device is usually worn on the wrist like a watch. For example, if we were monitoring TV watching habits this would mean tagging the remote control and not the TV itself. Other examples of applications for this technology are interactions with light switches, household appliances like fridges and dishwashers, and doors and windows.
Object mounted sensors
Object mounted sensors are sensors that still measure human activities like the body-worn sensors, but in this case the sensors are placed on the objects that are being used to reduce the burden on the monitored subject. Three sensors were chosen for the object mounted sensor category.
1. Reed switch
A Reed switch consist of two parts, one of which is to be mounted on a door or window while the second is to be placed on the respective door- or window frame. The part mounted on the frame is wired to an electrical source and produces a magnetic field that detects the presence of its counterpart. When the door or window is opened, the switch activates and produces a signal, recording the action. Over time a Reed switch can produce data on patterns of going outside, opening the fridge, or opening windows, depending on where the switch is placed.
2. Item presence sensors
An item presence sensor works in a similar fashion to the Reed switch in that a device records the presence of another paired device in its vicinity. The difference with a Reed switch is that an item presence sensor does not have to be wired to an electrical source to function. This allows for the use of such a sensor on more portable items such as medicine bottles or insulin pumps. A downside to this is that the item
problematic when such a sensor is used on a portable item with a user who is prone to forgetting an object’s original position (e.g., a patient with dementia).
3. Object mounted accelerometers
Like the accelerometer in the body-worn category, an object mounted accelerometer measures speed and acceleration. In this case the data are produced by the movement of the object the sensor is applied to. This can be useful in unobtrusively monitoring walking speeds in people using assisting tools like canes or walkers.
Environmental sensors
The set of environmental sensors consisted of devices placed in and around the home that measure some environmental characteristics of the living space, or human influences on this environment. This category was the largest one, containing seven sensors with a large range of applications.
1. Light sensor
Light sensors can be placed in a living space to measure the amount of light that reaches them. Depending on the room such a sensor is installed in, the device can provide information on variables like sleeping patterns (lights off in bedroom), general luminosity of a room during the day (curtains open or closed during the day) or bathroom use during the night.
2. Infrared presence sensor
An infrared presence (IRP) sensor provides information on general activity levels in the targeted area. It does not provide contextual information on the current activity, but movement in general can be used to gain insight into a person’s health status. Especially long-term movement data can show a decline in mobility, while such a gradual decline might go unnoticed without such a reference frame.
3. Microphone sound sensor
A microphone sound sensor, or just microphone, can be used to mon-itor audible cues from a resident. This can be either continuously or situationally. Continuous monitoring could provide information on audible activities like watching TV, showering, or doing the dishes. A combination with other sensors like a body position sensor would al-low for situational monitoring, which can prove useful in fall scenarios where a person is not able to reach a phone or lifeline button but can still call for help. The microphone can confirm a fall in this situation, whereas just the body position sensor can produce false positives when dropped or bumped.
4. Temperature sensor
A temperature sensor, or thermometer, can be used to provide infor-mation on the location of a person within the home in situations where a microphone or IRP sensor would be considered too intrusive, like the bathroom.
5. Electricity consumption sensor
Measuring the amount of electricity that is used by certain devices provides information on the amount of use of these devices. Like the RFID tag this could show information like TV or computer usage. 6. Flow sensor
A flow sensor is placed between the water supply and a faucet, shower, or water consuming device like a dishwasher. This is another relatively unobtrusive method to measure water consuming activities.
7. VOIP usage sensor
Voice Over IP (VOIP) programs are applications that allow users to place voice and video calls to other users over the internet. In older adults, isolation can be challenging may even lead to depression. Mon-itoring VOIP usage can provide details on social connectivity for these older adults, and can be used as an early indicator for isolation. The list of sensors we used for this study is not an exhaustive list, but we feel it contains enough variation within and between the different categories to provide us with a global overview of family caregiver opinions on sensor assisted caregiving.
2.2
Validation of survey
The survey was iteratively developed with multiple rounds of testing by members of the HealthE group at the University of Washington. This group consists of academic researchers in the field of technology and aging, who were expected to be at least reasonably familiar with the topics that were discussed in the survey. The average completion time for the survey within this group was 7 minutes. To account for more time spent on comment fields we expected the participants to spend between 10 and 15 minutes on the survey. The final survey can be found in the appendix of this thesis on page 36. After completion the survey was sent to, and reviewed by the University of Washington’s Internal Review Board (IRB). The IRB deemed the survey exempt from further ethical review, allowing us to start deploying the survey online.
2.3
Deployment
Survey hosting and data collection were performed using the University of Washington’s Catalyst (https://catalyst.uw.edu) survey tool. A link to the survey was created and could be posted externally to refer possible participants to the survey. The survey was administered on two parallel but separate platforms. The first medium that was used for deployment was the Amazon Mechanical Turk (MTurk; http://www.mturk.com) platform. MTurk allows users (also called workers) to complete Human Intelligence Tasks (HITs) for a financial compensation, the amount of which can be set by the task’s requester. HITs offered on this platform consist mostly of short reviews of a variety of goods and services, but with proper precautions the platform lends itself for more academic endeavors as well. MTurk has been successfully used [22–26] and validated [27] in other health-related studies. A downside of online data collection is the possibility for the subjects to provide answers that do not reflect the truth without the researchers knowing. Even though this can also be a problem in paper and pencil surveys, the anonymity of online surveys allows participants to provide information that would be doubted in a face-to-face setting (e.g., age, gender, height, weight). Pauszek et al. compared Mechanical Turk data against laboratory data and found no significant differences between their two populations, despite the lack of control in online testing [28]. Considering our study will only ask qualitative questions with no predefined right or wrong answer we do not see any reason for the participants to answer untruthfully.
As Mechanical Turk is an online platform that is accessible to anyone with internet access the user base is theoretically not limited to any specific age, gender, or nationality. However, certain age groups and nationalities will be more prominently represented because of the technical nature and the North American origin of the platform. To keep our results representative for the caregiving situation in western countries we will limit our population to English speaking subjects over the age of 18 with an IP address within the United States of America. Basic demographic information (age, gender, and education level) will be collected to compare our population with known metrics about the general caregiving population. Finally, we will require our subjects to be a care provider to someone over the age of 65 to be eligible for participation.
A pilot batch of 20 surveys was created on MTurk offering a $0.75 reward on completion, all of which were completed within 4 hours of deployment. The quality of this batch’s data was deemed acceptable and in response to the rapid rate of participation we decided to lower the reward to $0.50 on completion for the next set of surveys. Another batch of 280 surveys was created to bring the total number of responses up to 300. Completion of this larger set took 8 days. After we reached this threshold of participants a first round of analysis was performed on the data. This first analysis focused on
basic demographics of family caregivers and general opinions of the three sensor categories that were presented in the survey.
The second platform for the survey administration was through recruit-ment of participants using Research Match (www.researchmatch.com). This is another online platform, mediating between researchers and volunteers for studies in the medical domain with a user base of around 110.000 US based volunteers. Filtering on user characteristics is a possibility on Research Match, but being a family caregiver is not one of the possible filters. Thus, an invitation to participate in the survey was sent to multiple subsets of 1500 (the predefined limit for survey recruitment) random users, with the expec-tation to identify some caregivers among them. In total, 23,676 potential subjects were approached, of which 803 agreed to participate.
2.4
Data analysis
Data analysis was performed using the R statistical package [29], utilizing the open source Likert, Wordcloud2 and Correlations packages. Term fre-quency - inverse document frefre-quency word clouds were created using the R statistical package to visually analyze common themes in our subjects’ comments.
Chapter 3
Results
A total of 367 attempted surveys were submitted through Mechanical Turk and collected in our Catalyst database between February 14th and February 28th, 2017. 11 participants did not provide informed consent and did not get forwarded to the webpage containing the survey. 89 participants were not caregivers and 17 participants were caregivers but did not provide care to someone of 65 years of age or older. Because we posted the HITs in three separate batches it was a possibility for participants to enter the study more than once, but because of participant ID records we could exclude any entries beyond the first for each of the 20 subjects where this scenario applied. The final number of eligible participants from the Amazon group was 238.
401 surveys were submitted through the Research Match specific link and after applying the same filters as described above we were left with 175 unique participants for this group. There were no duplicate entries and all participants provided informed consent. Most subjects that were excluded were ineligible because they were not care providers.
3.1
Demographic data
Demographic data for these groups can be found in Table 3.1 (Amazon/AM) and Table 3.2 (Research Match/RM).
Both groups can be considered highly educated with 89.9% and 100% of subjects having obtained at least a college degree or higher in our Amazon (AM) and Research Match (RM) groups respectively. The majority of our participants also either agree or strongly agree to the statement that they are familiar with sensors in a healthcare context (66.4% AM and 77.1% RM). Lastly, there is a comparable division of participants among the different categories of care provision regarding both time spent and intensity of care provided. As can be seen in these tables the two populations are also differ-ent in several areas. The AM group has an average age that is 16 years lower than the RM group, and has a female to male ratio of 1.18 whereas this ratio
Table 3.1: Demographics table (Amazon Mechanical Turk), N = 238
Metric Group N(%)
Gender Female 129 (55.2) Male 109 (44.8) Age (years) Under 30 55 (23.1) 30-49 121 (50.8) 50-69 59 (24.9) 70 or over 3 (1.3) Mean age in years (SD) 38.85 (13.09)
Relative’s age (years) 65-74 111 (46.6) 75-84 65 (27.3) 85 or over 62 (26.1) Mean relative age in years (SD) 76.58 (8.42)
Level of education High school or lower 24 (10.1) Some college or Bachelor 153 (64.3) Masters or higher 61 (25.6) Hours of care providing (hr/week) Under 10 60 (25.2) 10-19 79 (33.2) 20-29 48 (20.2) 30-39 15 (6.3) 40 or over 35 (14.7) Intensity of care provided† Low intensity 92 (38.7) Moderate intensity 108 (45.4) High intensity 38 (16.0) Familiarity with sensors Low familiarity 50 (21.0) Neutral about familiarity 30 (12.6) High familiarity 158 (66.4)
†Example descriptions of three levels of care were provided, and participants were asked which of the three descriptions they indentified the most with. Low intensity was defined as only helping with household tasks like doing dishes and vacuuming. Moderate intensity included eating and moving around in addition to the household tasks described before. High intensity care included all tasks described in both lower levels, in combination with providing aid during dressing, bathing, and other more personal situations.
Age and gender apply to the participant filling out the survey. Care recipient age is listed as relative’s age. Care recipient gender was not recorded.
Table 3.2: Demographics table (Research Match), N = 175
Metric Group N(%)
Gender Female 128 (73.1) Male 47 (26.9) Age (years) Under 30 28 (16.0) 30-49 44 (25.1) 50-69 92 (52.6) 70 or over 11 (6.3) Mean age in years (SD) 54.6 (15.11)
Relatives age (years) 65-74 48 (27.4) 75-84 50 (28.6) 85 or over 77 (44.0) Mean relative age in years (SD) 80.62 (9.06)
Level of education High school or lower 0 (0) Some college or Bachelor 74 (42.3) Masters or higher 101 (57.7) Hours of care providing (hr/week) Under 10 74 (42.3) 10-19 46 (26.3) 20-29 25 (14.3) 30-39 5 (2.9) 40 or over 25 (14.3) Intensity of care provided† Low intensity 100 (57.1) Moderate intensity 38 (21.7) High intensity 37 (21.1) Familiarity with sensors Low familiarity 30 (17.1) Neutral about familiarity 10 (5.7) High familiarity 135 (77.1)
†Example descriptions of three levels of care were provided, and participants were asked which of the three descriptions they identified the most with. Low intensity was defined as only helping with household tasks like doing dishes and vacuuming. Moderate intensity included eating and moving around in addition to the household tasks described before. High intensity care included all tasks described in both lower levels, in combination with providing aid during dressing, bathing, and other more personal situations.
Age and gender apply to the participant filling out the survey. Care recipient age is listed as relative’s age. Care recipient gender was not recorded.
Figure 3.1: Perceived usefulness for bw (body-worn), om (object mounted) and en (environmental) sensors.
is 2.72 in the Research Match group. In addition to these dissimilarities in demographic factors, the two groups are also expected to be inherently dif-ferent due to their motivations for participation (AM paid/RM voluntary). Considering these differences, we decided to keep the two groups separated during analysis. Results for the Amazon and Research Match groups will be presented separately with AM and RM labels respectively.
3.2
Amazon results
In all categories, the majority of our test subjects see positive value in using sensors to monitor their loved ones. Body-worn sensors (BW) received the highest number of positive responses with 75% of participants agreeing to their usefulness. Object mounted sensors (OM) and environmental sensors (EN) received comparable results with 62% and 60% positive responses re-spectively (See Figure 3.1). When considering the possible impact on the monitored individual’s privacy participants were generally still positive in all three categories, with agreement being 3% (BW 72%, OM 59%, EN 57%, Figure 3.2) lower than when just considering usefulness alone. EN sensors showed the largest proportion of participants with concerns about privacy with 39% having some issues with this type of sensor, almost matching the number of participants without privacy concerns which was 43%. The BW and OM categories caused less concern, with respectively 23% and 32% of participants having some concerns (Figure 3.3). Out of all the individu-ally listed sensors the microphone received the highest number of positive responses with 73%, followed by the infrared presence sensor and body-position sensor both at 63%. The sensors with the least amount of positive answers were the electricity and VOIP sensor, at 31% and 32% respectively. Results for all sensors are shown in Figure 3.4.
Figure 3.2: Breach of privacy for bw (body-worn), om (object mounted) and en (environmental) sensors. Answers to the statement: ”I am concerned about privacy violations with the use of [category] sensors”
Figure 3.3: Final perceived value (value outweighs breach of priviacy) for bw (body-worn), om (object mounted) and en (environmental) sensors.
Figure 3.4: Usefulness results for all sensors. irPresence: Infrared presence sensor, rfid: Radio Frequency Identification, VOIP: Voice Over IP
Figure 3.5: Perceived usefulness for bw (body-worn), om (object mounted) and en (environmental) sensors.
3.3
Research Match results
The results obtained from our RM participants were similar to those of the AM group described above, with the most noticeable difference being that the RM group was slightly less positive. General usefulness was rated at 67%/54%/50% (BW/OM/EN), averaging 8.7% lower than the AM group. Privacy concerns were also very similar to the AM group, with the EN sensors producing the greatest number of concerned participants at 35%. OM and BW sensors also received comparable results to the AM group with 28% and 21% concerned participants respectively. When looking at final perceived value in this group, BW was also the best received category with 65% positive response, followed by OM with 57% and EN with 49%. If we take neutral responses out of our analysis, since they are neither positive nor negative, all three categories have a positive final perceived value (See Figure 3.7). When looking at individual sensors for the RM group we see a large overlap with the AM group. The six highest rated sensors are in the same order in both groups and the microphone has the largest amount of positive responses in both groups (See Figure 3.8).
3.4
Demographic analysis
After performing the analysis required to answer our research question, de-mographic information was used to identify other possible factors influenc-ing opinions on sensor use. This second round of analysis produced two apparent trends. When looking at the mean response for privacy concerns across all sensor categories, there are no significant differences between men and women (AM: p=0.288 —— RM: p=0.164), usefulness (AM: p=0.377 —— RM: p=0.262), and perceived value (AM: p=0.250 —— RM: p=0.142). However, when comparing privacy concerns separated by sensor category the
Figure 3.6: Breach of privacy for bw (body-worn), om (object mounted) and en (environmental) sensors. Answers to the statement: ”I am concerned about privacy violations with the use of [category] sensors”
Figure 3.7: Final perceived value (value outweighs breach of privacy) for bw (body-worn), om (object mounted) and en (environmental) sensors.
Figure 3.9: Privacy concerns by gender. F = Female, M = Male, bw = Body-Worn, om = Object Mounted, en = Environmental
in all but the Environmental sensors of the Research Match group (AM: BW p=0.018, OM p=0.106, EN p=0.147 —— RM: BW p=0.018, OM p=0.049, EN p=0.006). Figure 3.9 and 3.10 show these gender differences: note that the O group in Figure 3.10 only contains 3 subjects and was too small to be included in our analyses. Also noteworthy is the fact that the most sig-nificant difference between gender groups in the RM environmental group is in the opposite direction, with men having less concerns. The apparent difference in privacy concerns, regardless of which gender group has more issues, does not cause any significant dissimilarities in final perceived value (BW p=0.155, OM p= 0.104, EN p=0.299).
A second observation made during visual exploration was an apparent correlation between the age of the caregiver’s loved one, and their opinions on privacy concerns when using sensors. Care recipient age and privacy con-cerns are significantly correlated in the Amazon group. All three categories show participants with older relatives having less privacy concerns (AM: Spearman, BW p=0.019, OM p= 0.004, EN p=0.011, Figure 3.11). Our RM participants show a similar, but weaker, correlation with no statisti-cal significance (AM: Spearman, BW p=0.163, OM p= 0.073, EN p=0.074, Figure 3.12). When looking at the usefulness of individual sensors grouped by relative age this trend is also visible, with AM participants showing a defined ”staircase” pattern (Figure 3.13) while the RM participants show a less linear relationship (Figure 3.14). In some cases, especially in the mi-crophone and the body position sensor of the AM group, the trend seems to be reversed with higher averages of privacy concerns as the monitored subject gets older. We did not ask participants about their opinions on the
Figure 3.10: Privacy concerns by gender. F = Female, M = Male, O = Other, bw = Body-Worn, om = Object Mounted, en = Environmental
p = 2.143e-08) but testing correlations between participant age and privacy concerns did not produce any significant results.
Figure 3.11: Privacy concerns by age group of loved one.
Chapter 4
Discussion
4.1
Findings
The majority of our population of family caregivers sees value in the use of sensors to aid their caregiving. Not all sensors were perceived as equally valuable but the majority of our participants responded positively to the idea of implementing one or more sensors into their daily routines. The sensors that provide human oriented information like the microphone and movement sensors were seen as more useful than the sensors that only described general environmental metrics like the electricity and water flow sensors. This seems to show that our subjects prefer direct information about personal activities over information about the home environment that could lead to inferences about behavior. Comments by participants added to this by pressing the fact that they did not feel comfortable ”spying” on their loved one, and were only interested in providing the best possible care. However, this does not mean the environmental sensors are not useful in any situation. There was still a sizeable portion of participants that was positive about this category of sensors. Care intensity does not seem to affect the acceptance of potential users. This indicates that there are no specific ”high intensity sensors” that are more useful for people receiving high intensity care and less useful for low intensity care recipients.
Privacy concerns play a large role in the acceptance rate of sensors and our participants did not seem to have any impactful concerns. The majority reported that they value the produced information of sensors higher than the breach of privacy these sensors cause. Because of the high education levels in our two populations we cannot confidently draw any conclusions about the opinions of the general public. Especially when the topic of interest is tech-nical in nature, digital literacy plays an important role in acceptance. We expect that a less educated population might have reported more privacy concerns. This was a limitation of the platforms we used to recruit par-ticipants, and future studies should try to focus on a more representative
population.
The trend we found when looking at privacy concerns in relation to the age of our participants’ loved ones does not seem to be connected to the participant’s age itself. While our data are not granular enough to pro-vide an explanation for this trend, we can still hypothesize about a possible cause. So-called ”ageism” is a phenomenon previously documented in set-tings such as nursing homes, where older residents are treated differently than younger residents based on negative stereotypes [30]. Nursing staff switches to infantilized communication methods in a large number of cases, most notable of which is the collective pronoun substitution (e.g. calling people ”hon”/”honey” or ”sweetheart”/”sweetie”) [30]. The notion that older adults are less capable or cognitively impaired by default is incorrect but widespread. This belief is apparent, even among family caregivers who deal with this demographic group frequently [30]. Caregivers might compare older adults to small children, in the sense that they need constant help and cannot function on their own. This might be an explanation for why the age of a loved one is this strongly correlated with the perceived reduced need for privacy. Alternatively, caregivers may hypothesize that aging leads to an increased need for intensive and continuous monitoring and to loved ones becoming frailer and less independent.
An empirical study of technology researchers’ perceptions of ethical is-sues in developing smart-home health technologies by Birchley et al. yielded interesting insights about privacy in a smart-home setting [31]. Researchers considered physical privacy to a lesser extent, claiming the choice of an end-user to use a technology was a solution to ethical dilemmas [31]. Putting all responsibility on the user like this may be irresponsible in certain situations, especially when considering other types of privacy like privacy regarding personal digital information. We can expect an average older adult to confi-dently decide about whether they would like to allow their movement within the home to be monitored because this is a very tangible subject. However, when we are asking this same user about the storing and sharing of their health information using online services like the cloud it is largely depen-dent on the user’s experience with such systems if they can make a fully informed decision. Because of this, it will be important to educate possible users about the implications of information technology systems, regardless of digital literacy levels.
Even though our quantitative results seem unaffected by the issue of privacy, there were still several comments expressing a lack of trust in our proposed sensors. Of course, these comments are personal opinions from a minority of our subjects, but they represent the opinions of a sizeable part of the target population and these issues should be addressed thoroughly to aid the general public in forming opinions on sensor use. Some comments
”Any sensor that can monitor basic activity would be great. Most of these sensors seem to be excessive and unnecessary, and concern me greatly over someones ability to hack into these de-vices.”
”I think it is important for monitors to monitor falls and allow the individual to seek help but after that I believe that there might be a lot of privacy invasion issues and needs to be discussed and looked at before using. I would also worry would people hack into this to monitor when to rob or do harm to my loved one.” ”overall, the benefits outweigh any loss of privacy because the safety of the older adult is paramount at this point.”
”I think that the most important thing is that the older adult knows 100% that they are being monitored and in what ways.” ”I think they are a good idea but I would have many privacy concerns”
Due to the relatively low number of comments, and the large variance between these comments the inverse term frequency analysis did not provide any significant results. The created word clouds did confirm the quantita-tive results showing a high rate of acceptance with the number of posiquantita-tive comments far outweighing the number of negative ones.
4.2
In context
We found a relatively high rate of acceptance in our pre-implementation study. Whether a new technology will be adopted depends on many fac-tors. A systematic review about technology acceptance among older adults aging in place by Peek et al. found multiple factors influencing technol-ogy acceptance in a pre-implementation stage [9]. The factors with the largest number of associated studies were the ”need for technology”, ”de-sire to age in place”, ”privacy implications”, ”high costs”, and ”expected increased safety” [9]. Costs were not discussed in our survey but the other influencing factors all had to be considered by our participants. Even when considering some of the most concerning of these factors (privacy, need for technology, safety), our results still show a high rate of acceptance in a pre-implementation stage. This is a positive indicator for actual acceptance when the technology is deployed.
Even more important than initial acceptance is continued motivation to use the products after implementation. Knowledge about acceptance influencing factors in a post-implementation setting is scarce, which is a problem because using sensors will only be useful when the data they pro-duce can be seen in context. Repeated measurements allow for comparisons
with periods of known health or sickness. A study by Van Kasteren et al. found that smart home sensor data accurately reflected daily routines of older adults [32]. This monitoring of daily patterns is important to ”iden-tify changes in routine that signal illness, recovery from bereavement, and gradual deterioration of sleep quality and daily movement” [32]. Therefore, post-implementation acceptance rates are just as, if not more important than the initial acceptance rates we collected in this study. Users will likely not see any noticeable change within a short period after deploying sensors, so it is paramount that they stay motivated and keep up their sensor related routines consistently in the long term. Possible solutions to encourage con-tinued use could be an online community to discuss results and problems with other users. This community could be implemented in the form of an accompanying app, also allowing it to notify users when they failed to check their results for a period of time. Such an app could also be used to set more short term personal goals and offer a platform for friendly competition with friends and family.
4.3
About the study
Conversion rate
The percentage of participants that was filtered out after trying to complete a survey through Research Match was 56%. This was 18% higher than the number of excluded participants in the Amazon group where only 38% was filtered out. We expect this to have been caused by the different method of recruitment. The AM population was shown a page with detailed infor-mation about the study before participating, whereas the Research Match group had to decide whether to participate based on a short recruitment text that was sent to them through email. This recruitment text was less detailed because of the imposed word limit, which probably caused the larger number of misunderstandings about the purpose of the study. This will not have in-fluenced the quality of our data since only completed surveys were included in our analyses. Once participants were deemed eligible they reached the survey page which was always the same, regardless of the source of recruit-ment. The resulting number of participants with completed questionnaires was still large enough to answer our research question with confidence.
Anonymity
The AM group participants were only identified by their Amazon worker number and no names or addresses were recorded. The RM group partici-pants were approached through a personal email address but after clicking
for our study. A danger of anonymous surveys is the lack of certainty about the truthfulness of the participant’s answers. The RM participants were all volunteers and without any personal gain for them, we can assume they would have no reason to misrepresent their opinions. The AM group was fi-nancially compensated for participation which may introduce a bias for some participants. We think this did not affect us in a noticeable manner since the compensation was minimal at $0.50. The reward was very comparable to many other shorter and easier tasks available on MTurk. The strength of our survey being anonymous comes from the nature of the subject. Home care can be a very personal matter and our survey allowed participants to speak their mind, even if they felt any questions or topics may have been sensitive.
Analysis
Because this study was set up as a first look into the area of sensor assisted caregiving from the perspective of a family caregiver we set out to identify the opinions of potential users. Our research question was answered without any subgroup analysis but with all the demographic information we collected other possible interesting discoveries could be made. To give as much validity as possible to our results we tried to prevent unnecessary repeated statistical testing by visually identifying regions of interest first, and only testing their significance later. By allowing the researcher to define what was interesting we introduced a certain degree of observer bias. With this method of analysis we did not alter the validity of our answers to the original research question and only provided areas for future studies to focus on.
Population
This study specifically looked at caregiver opinions about smart home sen-sors while previous studies have limited their focus mostly on older adults who would reside in a smart home setting. Those previous studies have shown that there certainly is some interest among a select group of older adults but due to the voluntary nature of those studies a selection bias is present. Only older adults who are interested in, or at least not opposed to, sensor technology will participate skewing the results towards the positive. The older adults who do not participate in such studies might still benefit from sensors being deployed in their homes. In such cases where the older adult is not interested but would still benefit, it is up to a family member or loved one to convince the older adult.
According to the Family Caregiver Alliance 75% or more of all care-givers are female, and spend 50% more time providing care than males do [33]. The number of women in our Amazon population is lower than this expected number at 55%, while the Research Match population comes
very close at 73%. The Amazon group does however reflect the expected gender demographics of the Amazon mTurk platform which varies between 53% and 58% being female according to the demographics section of the mturk tracker website [34]. Time spent providing care is not significantly different between male and female caregivers in either of the two groups. Another notable difference for our participants is in the average education level. The large majority of subjects have obtained a bachelor’s degree or higher. The technical (online) nature of both platforms in addition to our highly educated populations is a probable explanation for the results we see in the self-reported familiarity with sensors. Previous studies have shown a relation between education levels and digital literacy [35], which in turn has been shown to increase acceptance of new technologies [35]. Consider-ing this, our two groups might not be perfect representations of the general caregiving population. However, with their intrinsic interest in technology and innovation our subjects are in fact the target demographic for the use of sensors such as the ones we have shown here. Thus, even though one may not be able to extrapolate the results to the general public, they are still valuable as a starting point for future in-depth studies in this area.
Sensor selection
The list of sensors we used was informed by a previous study [17], and even though this list was varied it was not all-encompassing. Recent advances in technology have increased the popularity of personal health gadgets like smart watches and health trackers. These devices are capable of tracking heart rates and movement speed, among other things, and could be viable instruments in a home care situation. Since this type of sensor was not included in the set our sensor selection was based on we did not include it in this study. However, because of the popularity of these devices it is very likely we will see more of them appear in a home care setting in the future.
4.4
Conclusion
Our study shows that a considerable portion of family caregivers is interested in sensor-assisted caregiving. With this knowledge in mind we can start looking for specific situations where certain sensors are most useful. The monitored older adult still has to be convinced of the value of a sensor but the largest burden, caused by the constant analysis of their own data, is taken off their shoulders. Future research should focus on the effects of sensor use on the caregiving process, and the resulting changes in health status for the older adults being monitored. Having a large variety of tested sensors available on the market will ensure the right sensor is available for
With this number of sensors, it is no surprise that many different opinions arise. Home care is a very personal subject and every situation is different. The main conclusion we can draw from this study is not which one of the described sensors is best for everyone, but the fact that many people are open to the idea of using sensors in general. The design and implementation of sensor systems need to address end users’ needs and preferences including older adults and their family members. With this approach, sensor use may allow families to stay informed when separated by significant geographic distance, reduce the burden on the caregiver, and improve the quality of life for older adults trying to age in place.
Acknowledgements
I would like to thank George Demiris, Monique Jaspers, and everyone in the University of Washington’s Health-E group for their academic contributions to my thesis. Discussing my study with you gave me many new insights and this report wouldn’t have been nearly as comprehensive without you. Special thanks are in order for Yong Choi, who took me under his wing and spent the time to discuss my progress in great detail even after my study plans changed and my work was no longer a part of his project.
I would also like to thank Shawn Banta for helping me with the process of registering with the UW and moving to the US. Without her all the paperwork would probably have gotten the better of me and I wouldn’t have had a home to live in when I first arrived. Another personal thank you goes out to Christian and his girlfriend Lea who invited me to everything American that happened in their home, and gave me a true taste of American life.
Lastly I’d like to thank my friends and family in The Netherlands, who kept me motivated when I started to feel homesick from the 5000 miles that separated us. The most special person among this group is of course my girlfriend Myrthe who motivated me to seek out this adventure in the first place, and supported me in doing so even though it meant us having to spend over half a year apart. Thank you.
Bibliography
[1] K. T. Washington, S. E. Meadows, S. G. Elliott, and R. J. Koopman. Information needs of informal caregivers of older adults with chronic health conditions. Patient Educ Couns, 83(1):37–44, Apr 2011.
[2] T. Le, N. C. Chi, S. Chaudhuri, H. J. Thompson, and G. Demiris. Understanding Older Adult Use of Data Visualizations as a Resource for Maintaining Health and Wellness. J Appl Gerontol, Jul 2016. [3] G. Demiris and B. K. Hensel. Technologies for an aging society: a
systematic review of ”smart home” applications. Yearb Med Inform, pages 33–40, 2008.
[4] T. Le, B. Reeder, D. Yoo, R. Aziz, H. J. Thompson, and G. Demiris. An evaluation of wellness assessment visualizations for older adults. Telemed J E Health, 21(1):9–15, Jan 2015.
[5] T. Le, B. Reeder, J. Chung, H. Thompson, and G. Demiris. Design of smart home sensor visualizations for older adults. Technol Health Care, 22(4):657–666, 2014.
[6] F. J. Infurna, M. A. Okun, and K. J. Grimm. Volunteering Is Associated with Lower Risk of Cognitive Impairment. J Am Geriatr Soc, Oct 2016. [7] H. Y. Chiu, F. C. Lai, P. Y. Chen, and P. S. Tsai. Differences Between Men and Women Aged 65 and Older in the Relationship Between Self-Reported Sleep and Cognitive Impairment: A Nationwide Survey in Taiwan. J Am Geriatr Soc, Sep 2016.
[8] K. L. Courtney, G. Demiris, M. Rantz, and M. Skubic. Needing smart home technologies: the perspectives of older adults in continuing care retirement communities. Inform Prim Care, 16(3):195–201, 2008. [9] S. T. Peek, E. J. Wouters, J. van Hoof, K. G. Luijkx, H. R. Boeije, and
H. J. Vrijhoef. Factors influencing acceptance of technology for aging in place: a systematic review. Int J Med Inform, 83(4):235–248, Apr
[10] W. B. Mortenson, A. Sixsmith, and R. Beringer. No Place Like Home? Surveillance and What Home Means in Old Age. Can J Aging, 35(1):103–114, Mar 2016.
[11] K. A. Jeong, G. Salvendy, and R. W. Proctor. Smart home design and operation preferences of Americans and Koreans. Ergonomics, 53(5):636–660, May 2010.
[12] J. Chung, H. J. Thompson, J. Joe, A. Hall, and G. Demiris. Examining Korean and Korean American older adults’ perceived acceptability of home-based monitoring technologies in the context of culture. Inform Health Soc Care, 42(1):61–76, Jan 2017.
[13] J. N. Morris, B. E. Fries, K. Steel, N. Ikegami, R. Bernabei, G. I. Carpenter, R. Gilgen, J. P. Hirdes, and E. Topinkova. Comprehensive clinical assessment in community setting: applicability of the MDS-HC. J Am Geriatr Soc, 45(8):1017–1024, Aug 1997.
[14] F. Landi, E. Tua, G. Onder, B. Carrara, A. Sgadari, C. Rinaldi, G. Gambassi, F. Lattanzio, and R. Bernabei. Minimum data set for home care: a valid instrument to assess frail older people living in the community. Med Care, 38(12):1184–1190, Dec 2000.
[15] J. E. Byles. A thorough going over: evidence for health assessments for older persons. Aust N Z J Public Health, 24(2):117–123, Apr 2000. [16] P. L. Enright. The six-minute walk test. Respir Care, 48(8):783–785,
Aug 2003.
[17] D. Nienhold, R. Dornberger, and S. Korkut. Healthcare Assessment Questions Automatically Answered Using Non-Invasive, Ambient Sen-sors. 2016 IEEE International Conference on Health Informatics. [18] N. Sharma, S. Chakrabarti, and S. Grover. Gender differences in
care-giving among family - caregivers of people with mental illnesses. World J Psychiatry, 6(1):7–17, Mar 2016.
[19] S. M. Lutzky and B. G. Knight. Explaining gender differences in care-giver distress: the roles of emotional attentiveness and coping styles. Psychol Aging, 9(4):513–519, Dec 1994.
[20] S. H. Parks and M. Pilisuk. Caregiver burden: gender and the psycho-logical costs of caregiving. Am J Orthopsychiatry, 61(4):501–509, Oct 1991.
[21] S. Rote, J. L. Angel, and K. Markides. Health of elderly Mexican American adults and family caregiver distress. Res Aging, 37(3):306– 331, Apr 2015.
[22] J. Kristan and B. Suffoletto. Using online crowdsourcing to under-stand young adult attitudes toward expert-authored messages aimed at reducing hazardous alcohol consumption and to collect peer-authored messages. Transl Behav Med, 5(1):45–52, Mar 2015.
[23] A. M. Turner, M. Bergman, M. Brownstein, K. Cole, and K. Kirchhoff. A comparison of human and machine translation of health promotion materials for public health practice: time, costs, and quality. J Public Health Manag Pract, 20(5):523–529, 2014.
[24] B. Yu, M. Willis, P. Sun, and J. Wang. Crowdsourcing participatory evaluation of medical pictograms using Amazon Mechanical Turk. J. Med. Internet Res., 15(6):e108, Jun 2013.
[25] Z. C. Merz, R. Van Patten, and J. Lace. Current Public Knowledge Per-taining to Traumatic Brain Injury: Influence of Demographic Factors, Social Trends, and Sport Concussion Experience on the Understanding of Traumatic Brain Injury Sequelae. Arch Clin Neuropsychol, Oct 2016. [26] A. M. Paine, L. A. Allen, J. S. Thompson, C. K. McIlvennan, A. Jenk-ins, A. Hammes, M. Kroehl, and D. D. Matlock. Anchoring in Destination-Therapy Left Ventricular Assist Device Decision Making: A Mechanical Turk Survey. J. Card. Fail., 22(11):908–912, Nov 2016. [27] J. Chandler and D. Shapiro. Conducting Clinical Research Using
Crowdsourced Convenience Samples. Annu Rev Clin Psychol, 12:53–81, 2016.
[28] Joseph R. Pauszek, Pedro Sztybel, and Bradley S. Gibson. Evaluating amazon’s mechanical turk for psychological research on the symbolic control of attention. Behavior Research Methods, pages 1–15, 2017. [29] R Core Team. R: A Language and Environment for Statistical
Comput-ing. R Foundation for Statistical Computing, Vienna, Austria, 2016. [30] J. M. S. Sao Jose, C. A. F. Amado, S. Ilinca, S. G. Buttigieg, and
A. Taghizadeh Larsson. Ageism in Health Care: A systematic review of operational definitions and inductive conceptualizations. Gerontologist, May 2017.
[31] G. Birchley, R. Huxtable, M. Murtagh, R. Ter Meulen, P. Flach, and R. Gooberman-Hill. Smart homes, private homes? An empirical study of technology researchers’ perceptions of ethical issues in developing smart-home health technologies. BMC Med Ethics, 18(1):23, Apr 2017.
Everyday Household Routines: A Mixed Method Case Study. JMIR Mhealth Uhealth, 5(6):e52, Jun 2017.
[33] Family Caregiver Alliance caregiver statistics demographics. https://www.caregiver.org/caregiver-statistics-demographics. Ac-cessed: 2017-06-14.
[34] MTurk Tracker mechanical turk demographics. http://demographics.mturk-tracker.com/gender/US. Accessed: 2017-06-14.
[35] M. Mackert, A. Mabry-Flynn, S. Champlin, E. E. Donovan, and K. Pounders. Health Literacy and Health Information Technology Adoption: The Potential for a New Digital Divide. J. Med. Internet Res., 18(10):e264, Oct 2016.
