In This Issue
Technologies for an Aging Population
March 1, 2009 Volume 39 Issue 1

Technology in Support of Successful Aging

Tuesday, March 17, 2009

Author: Misha Pavel, Holly Jimison, Tamara Hayes, and Jeffrey Kaye

Technologies that provide in-time information through unobtrusive, in-home monitoring can improve the daily lives of elders.
 

Introduction: A Personal Scenario
Earlier today I had a terrifying experience when visiting my 80-something year old parents in their little house in Islip, Long Island. It’s not much of a house, but we all love it, and, not surprising, my parents want to live there forever. I was in good spirits as I made the familiar sharp turn into the driveway, reflecting on my last telephone call with my dad just two days ago. He was upbeat—everything was just fine, and both Mom and Dad were looking forward to my visit from Portland, where I am a division head at Oregon Health & Science University (OHSU).

After a seemingly cheerful hug from Dad, I knew instantly something was terribly wrong. Mom did not come to the door, and the house was unusually dark and messy—even for my parents. A pile of dishes in the sink, the stench of old garbage—this was not Mom’s kitchen—a wonderful cook in her day. When I found Mom in her bed, she gave me a big smile, but it was hard for her to move as she attempted to cover up a bruise on her right arm. None of this had come up during my phone call!

Reluctantly, Dad told me that Mom’s bruise was caused by a fall in her bedroom several days ago. The scary part was that he hadn’t discovered her fall for several hours, even though she was wearing an alert pendant and says she was screaming—at the time, Dad was deeply into his favorite TV show. Mom did not press the pendant because she dreaded the havoc of an ambulance—as had happened with a false alarm a few years ago.

Dad also admitted that Mom was sometimes confused, but he did not make much of it because more often than not she was as sharp as ever, and after all, we all have our senior moments. While washing my hands, I discovered a full container of Lasix—medication my Mom swore she was taking religiously once a day. Obviously she was not.

Watching my Dad walk, I could not help noticing a hesitation and shuffle in his gait. It turned out that my parents had stopped taking walks in the nearby park—their main form of exercise—after an unexpected October snowstorm, and they had never resumed them.

I was uneasy. Did Mom have a serious cognitive decline—a mild cognitive impairment? Should I do something? Should I tell their doctor? Should I convince them to move to an elder care facility? I doubt if I could . . . And I live 3,000 miles away . . .

 The parents of Misha Pavel described above belong to the fastest growing, economically dangerous, “epidemic” threat to society—the aging population. Figure 1 shows the problem in terms of a pyramid of aging, the number of people in different age groups. Fifty years ago, it was indeed a regular pyramid with straight sides. Today, however, there is a large bulge in the middle—a “tsunami” wave of baby boomers racing toward retirement age.

Figure 1
FIGURE 1 This pyramid represents the distribution of ages in the U.S. population at three different points in time (1900, 1950, and 2000).
The number of people in each age group is shown on the horizontal axis: males on the left and females on the right. The baby boomers are
represented by the bulge in the pyramid for ages 300‖;â€54 (in 2000).

Economists usually look at this change in terms of the dependency ratio (Figure 2), the ratio of the number of people who need help per working person (i.e., the number of people older than 65 compared to those of working age, 20 to 65). This graph also shows that when the baby-boomer generation crosses the 65 year boundary (the vertical rectangle in Figure 1), the situation will deteriorate even more rapidly. At that point, instead of seven working people per older adult, there will be only three. Advances in health care and changes in lifestyle have extended the life expectancies of the older generation as well as those in succeeding generations. In addition, these relatively affluent elders expect to continue enjoying a high quality of life and want to remain in their own homes as long as possible.


Figure 2

FIGURE 2 The dependency ratio is the number of people older than 64 (presumably retired) compared to the number of individuals in the active workforce (200‖;â€64 years of age), plotted as a function of years. The rapid increase due to the baby-boomer population is indicated by the shaded rectangle. Source: http://www.ssa.gov/OACT/TR/TR06/tr06.pdf.

 

Unless we find a solution quickly, the economic implications of these demographic changes will be devastating. Fortunately, it appears that technology-based solutions may alleviate many of the challenges in economically feasible ways by enabling proactive health care, efficient care giving by remote caregivers, maintenance of elders’ independence, and improvements in their quality of life. In the remainder of this paper, we describe approaches and challenges to technologies for successful aging.

Information Needs and System Requirements
The purpose of the introductory scenario was to illustrate the importance of in-time information in elder care. To understand the need for information, we must first realize that taking care of an elder requires a team. It truly “takes a village”—a team of informal family caregivers, formal caregivers, and health care providers (including geriatricians, neurologists, nurses, and other health professionals), as well as life coaches. Each individual and organization involved in the elder-care team needs different information.

For example, an informal caregiver and a life coach want to know that the elder is engaged and socially and physically active (e.g., how often he or she sees friends). The formal caregiver wants to know whether the elder needs assistance in specific situations and activities, such as hygiene or nutrition. The medical professional wants to be notified when there is a likely reason for medical intervention, such as the onset of a physiological or neurological disorder or an accelerated decline in cognitive or physical abilities. In short, the team needs frequent assessments of activities and health status, including mobility, characteristics of gait and balance, manual dexterity, hygiene, nutrition, weight, blood pressure, medication-taking behavior, and measures of cognitive function, such as attention and memory.

Some of the characteristics of these measurements reflect the medical status of the elderly individual, but others are behavioral markers of physical and cognitive functionality. For example, a decrease in walking speed, in itself, may not necessarily present a problem for the elder. Evidence shows, however, that slower walking or finger tapping may indicate a concomitant or future decline in cognitive function (Camicioli et al., 1998).

A major obstacle in designing reliable monitoring systems is the variability inherent in behavioral assessments. Variability in clinical tests arises from differences among individuals, multiple ailments, unexpected changes and influences, and the inherent variability associated with aging. The traditional approach to assessment based on an in-clinic visit cannot possibly capture the moment-to-moment variability that can be observed informally.

 However, continuous unobtrusive measurements with low-cost sensors have inherent challenges as well as benefits. Sensor data in a natural home environment (as compared to a controlled laboratory setting) are often quite noisy and difficult to interpret on a point-by-point basis. However, by making measurements frequently, it is possible to average out much of the noise. In addition, data from long-term continuous monitoring can be used to assess individual trends and meaningful variability. Thus this approach offers distinct improvements over standard infrequent measurements in a clinic referenced to a population.

Significant individual differences, which occur naturally throughout a lifetime, seem to be exaggerated as people age. Applying population norms—referencing a population distribution—therefore may not be useful for evaluating individuals. Instead, a monitoring system must be adaptable for each individual, and the data must be used as a baseline for the evaluation of further measurements. Finally, because the objective of continuous observation is to assess “normal” behaviors, the sensing process must not interfere with an individual’s life and must require minimal additional effort. In other words, measurements must be unobtrusive and require minimal maintenance. Needless to say, these techniques must also be reliable, self-monitoring, affordable, and scalable.

We note in passing that aging is frequently associated with increasing difficulty in adapting to changes, new environments, and new procedures. The implication of this limitation for the assessment process is that an elder’s behavior must be observed in the normal living environment (e.g., in the home).

In recent years, a number of researchers have developed “Smart Homes,” test laboratories with sophisti-cated sensors, devices, and artificial intelligence algorithms that can monitor and make inferences about the details of human behavior. Although these technologically exciting laboratories (Stefanov et al., 2004) may have some benefits, very few of the proposed technologies are ready for widespread in-home use.

One way to mitigate this problem is to move out of the laboratory setting and into the community. For example, one of our studies at the Oregon Center for Aging and Technology (ORCATECH) is based on a so-called “Living Laboratory,” about 35 volunteers, living in their own residences, who are outfitted with a variety of sensors and communication systems (described in the next section). Another study, which is primarily focused on gathering clinical data, involves about 250 volunteers who are being monitored in their homes over a three-year period. Simultaneously, they are participating in a traditional clinical longitudinal study. We believe this is the largest trial of its kind.

Unobtrusive Sensing at Home
If we had unlimited resources, it would be possible to outfit dwellings with sensors everywhere, including pressure-sensitive carpets, radio-frequency identification devices (RFIDs), accelerometers and gyroscopes in every item and piece of clothing, a multiplicity of video cameras, and so on. However, the cost of the hardware, installation, software, and service would exceed the cost of one or more human caregivers, thus defeating our purpose. We are looking for monitoring methods that not only meet our requirements for unobtrusive monitoring, but are also affordable and manageable on a large scale.

The actual implementation of the monitoring system in our studies uses inexpensive sensors and systems that are easy to install and require minimal maintenance or intervention (Hayes et al., 2008; Hayes et al., in press; Jimison et al., 2004; Kaye and Hayes, 2006).

Figure 3

FIGURE 3 Various sensors and devices in a typical home-monitoring installation including IR sensors for tracking motion, contact switches for tracking outings, load cells under the bed, a medication tracker, a phone-monitoring system, a wrist-worn location-monitoring system, and a home computer for tracking computer use.

A typical sensing system in the “Living Laboratory” is shown in Figure 3. The movement of people in their homes is monitored by passive infrared (IR) motion detectors with pyroelectric sensors (MS16A, X10.com). The IR sensors, typically used in home-security systems, sense changes in heat energy from moving thermal sources, in this case, human bodies. The sensitivity of these sensors is sufficient, but their spatial resolution is limited. Thus when an IR motion detector senses a movement that exceeds a fixed threshold, the sensor transmits a message identifying the type of event. Upon receipt, the message is time-stamped and stored on a computer.

To estimate an elder’s gait velocity, we installed several motion sensors with restricted field of view to about 8 degrees to ensure precise measurements. These sensors have been used successfully to measure walking speed in Parkinson’s patients (Pavel et al., 2007). In a similar way, various actions, such as opening drawers, doors, and so on, are sensed by magnetic-contact switches (DS10A, X10.com). Sleep patterns and weight are observed by load cells under the bed (Adami et al., in press). Medication adherence is tracked by a wireless “MedTracker” (Hayes et al., 2006; Leen et al., 2007; Hayes et al., in press).

In houses with more than one resident, the participants wear active wireless devices (e.g., HomeFree™ http://www.homefreesys.com/) to help us identify and localize the individual who triggers a particular sensor (Hayes et al., 2007). These wearable devices include accelerometers, gyroscopes, and sensors to indicate that they are properly worn.

For individuals who use computers, the system records their interactions with their computers during normal usage, as well as while playing specially designed computer games. Developed in collaboration with Spry Learning Company, these games record every move (Jimison et al., 2004). The records are analyzed and interpreted in terms of computational models of the players, in which the parameters of the model represent the key cognitive characteristic. In one game, for example, the system estimates an individual’s working memory capacity.

The data from various sensors are collected by a dedicated computer in the home and transmitted securely to OHSU. The raw data are then checked for integrity and stored in a database on secure servers at the university. The data and the functionality of the systems at the client sites are monitored using a comprehensive web-based console tool that provides not only remote monitoring, but also management of the subject cohorts and technologies. Figure 4 shows a view from the console that includes a summary of sensor activities and use of a computer mouse.

Figure 4

 

FIGURE 4 This screen shot shows the console used by technicians and clinical research assistants to monitor the subject cohorts, the status of monitoring systems, and the integrity of the collected data. The data shown represent the total activity of the sensors and the number of computer logins.

In summary, the raw data from remote systems comprise time-stamped events associated with the detection of physical movements, the movement of doors and drawers, and various specialized devices similar to those shown in Figure 3. In the next section we describe how the data are interpreted.

Signal Processing, Pattern Recognition, and Inference
The initial goal of our analysis is to estimate an elder’s activity level and the characteristics of certain behaviors, such as gait velocity, medication adherence, and the number and types of outings. This unobtrusive, low-cost measurement system yields data that are only indirectly related to the more variable characteristics we want to measure. For example, instead of a direct measurement of gait velocity, the system records only motion detection.

To compensate for this imprecision, we must develop mathematical models of the relationship between the raw, observed data and the desired characteristics (Pavel et al., 2006a,b). To assess an elder’s gait velocity, for instance, we must first select motion-detector data events that are likely to include the desired information and then use them in a statistically efficient way.

The selection of the relevant data from the vast number of observed events requires a computation of the probability that the motion detectors are triggered by the targeted individual, rather than the spouse, visitors, caregivers, or large animals. In a multi-person dwelling, identification of the individual cannot be taken for granted, even if the individual has been asked to wear an identification device. Using such devices for localization is typically based on the strength of the received signal as a function of the distance from a fixed base station. This inference is difficult because the strength of the signal in a real dwelling is highly variable, and contrary to the theory based on open space, the signal strength does not decrease monotonically with distance from the transmitter. This is because of the complex environment of a building with irregular walls, furniture, appliances, and so on.

In one study we used a simple Markov model in conjunction with a triangulation procedure based on the strength of the signal transmitted by devices worn by an elder and spouse and received by several base stations, as shown in Figure 5. The model was used to estimate the location of each individual and then infer the probability that the correct individual had triggered particular motion detectors. In this particular study of elders with Parkinson’s disease, we also measured subjects’ walking speed in the clinic. The home monitoring estimates and the observed speeds in the clinic were very similar, indicating the reliability of the model (Pavel et al., 2007).

Figure 5


FIGURE 5 Speed of walking in m/sec as measured at home and in the clinic. The three black lines represent the 20, 50, and 80 percentiles, from bottom to top. The values indicated by the square markers are those measured in the clinic.

A more sophisticated approach to estimating an elder’s location in the home is to use Bayesian recursive estimation, a method that recursively estimates, step by step, the probability distributions of the elder’s possible locations (Paul and Wan, 2008). The advantage of this approach is that the estimated location is given in considerably more detail and can be used to assess the elder’s activity. We anticipate that this approach would also enable us to detect falls.

With accurate estimates of gait velocity, it should be possible to detect changes that would warrant an intervention. The algorithms developed by our group to detect such changes as early as possible estimate the capabilities of each individual by combining the individual’s measurements with those obtained from the larger population. The parameters of these mixed-effect models that characterize the participant’s relative performance, such as the slope of the changing speed of walking, are then used to perform the detection (Lu et al., 2009).

Intervention
The ultimate goal of continuous monitoring is to provide information to support decisions to intervene by various caregivers and clinicians. For example, early detection and diagnosis of declines in some functions will enable a health care provider to administer drugs or design an effective mitigation program that may include changes in diet, physical exercise, balance training, or cognitive exercises. In some situations, technology may be used to mitigate problems associated with normal or accelerated decline. For example, the detection of non-adherence to taking medications may trigger the use of a sophisticated, context-aware alerting system (Lundell et al., 2007).

A recent ongoing study involves the development of technology for remote coaching of exercises to maintain cognitive function (Jimison and Pavel, 2008). In this study, volunteers used computer technology to communicate with a health coach, set health goals, and monitor progress toward meeting those goals. We emphasize cognitive-health coaching and include interventions in the form of adaptive cognitive computer games, novelty exercises, physical exercise, and advice on sleep behaviors.

With the approach to health and wellness management described above, we can provide continuity of care with fewer resources. Prompting for coaching messages is automated, and special interfaces with the data are available to family members and clinicians. This type of technology offers a scalable approach to integrating lower cost personnel, as well as motivated patients and family members, into the care team.

Discussion
Technology-based care for elders is a rapidly emerging field at the intersection of engineering, medicine, biology, care-giving, and family life. The management of the multidisciplinary problem of caring for elders requires organizations that bring together experts and practitioners. ORCATECH is one such organization, spearheading one of the largest ongoing longitudinal studies on clinical evidence to investigate unobtrusive monitoring as compared to traditional, clinical assessment techniques.

Although this article is focused on our work at ORCATECH, several other organizations are also addressing these issues. The most comprehensive list of these can be found on the website of the Center for Aging Services and Technologies (http://www.agingtech.org/index.aspx). In addition, several well funded projects to address the problems of elder care have been initiated in the European Union. In this short paper, we could not address a multitude of related issues associated with ethics, privacy, and security. However, a good deal of research is under way to address these issues (Alwan et al., 2006; Demiris et al., 2008; Mahoney and Tarlow, 2006; Wild et al., 2008).

In summary, our studies combining unobtrusive monitoring with sophisticated computational algorithms represent the first steps in the development of technology-based care for elders. If these can be successfully integrated into clinical practice and caregivers’ workflow, we might avoid the scenario described in the introduction to this article.

Acknowledgments
The work described in this article is supported by National Institutes of Health PHS grants P30-AG008017, P30-AG024978, R01-AG024059, Intel Corporation, Kinetics Foundation, Alzheimer’s Association, and a NIST ATP grant.

References
Adami, A.M., M. Pavel, T.L. Hayes, and C.M. Singer. In press. Detection of movement in bed using unobtrusive load cell sensors. IEEE Transactions on Information Technology in Biomedicine.

Alwan, M., S. Dalal, D. Mack, S. Kell, B. Turner, J. Leachten-auer, and R. Felder. 2006. Impact of monitoring technology in assisted living: outcome pilot. IEEE Transactions on Information Technology in Biomedicine 10(1): 192–198.

Camicioli, R., D. Howieson, B. Oken, G. Sexton, and J. Kaye. 1998. Motor slowing precedes cognitive impairment in the oldest old. Neurology 50(5): 1496–1498.

Demiris, G., B.K. Hensel, M. Skubic, and M. Rantz. 2008. Senior residents’ perceived need of and preferences for “Smart Home” sensor technologies. International Journal of Technology Assessment in Health Care 24(1): 120–124.

Hayes, T.L., J.M. Hunt, A. Adami, and J.A. Kaye. 2006. An electronic pillbox for continuous monitoring of medication adherence. Pp. 6400–6403 in EMBS ’06, Proceedings of the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, August 30–September 3, 2006. New York, NY: IEEE.

Hayes, T.L., M. Pavel, N. Larimer, I.A. Tsay, J. Nutt, and A.G. Adami. 2007. Distributed healthcare: simultaneous assessment of multiple individuals. IEEE Pervasive Computing 6(1): 36–43.

Hayes, T.L., F. Abendroth, A. Adami, M. Pavel, T.A. Zitzelberger, and J.A. Kaye. 2008. Unobtrusive assessment of activity patterns associated with mild cognitive impairment. Alzheimer’s and Dementia 4(6): 395–405.

Hayes, T.L., N. Larimer, A. Adami, and J. Kaye. In press. Medication adherence in healthy elders: small cognitive changes make a big difference. Aging and Health.

Jimison, H., and M. Pavel. 2008. Integrating Computer-Based Health Coaching into Elder Home Care. Pp. 122–129 in Technology and Aging, vol. 21, edited by A. Mihailidis, J. Boger, H. Kautz, and L. Normie. Fairfax, Va.: IOS Press.

Jimison, H., M. Pavel, J. McKanna, and J. Pavel. 2004. Unobtrusive monitoring of computer interactions to detect cognitive status in elders. IEEE Transactions on Information Technology in Biomedicine 8(3): 248–252.

Kaye, J.A., and T.L. Hayes. 2006. Home health monitoring: a system to assess motor and cognitive function. Generations 30(2): 61–63.

Lu, Z., T.K. Leen, and J. Kaye. In press. Hierarchical Fisher Kernel for Longitudinal Data. Presented at Advances in Neural Information Processing Systems Conference (NIPS 22), December 8, 2008, Vancouver, British Columbia. La Jolla, Calif.: Neural Information Processing Systems Foundation.

Lundell, J., T.L. Hayes, S. Vurgun, U. Ozertem, J. Kimel, J. Kaye, F. Guilak, and M. Pavel. 2007. Continuous Activity Monitoring and Intelligent Contextual Prompting to Improve Medication Adherence. Pp. 6286–6289 in Proceedings of the 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Lyon, France. Piscataway, N.J.: IEEE.

Mahoney, D.F., and B. Tarlow. 2006. Workplace response to virtual caregiver support and remote home monitoring of elders: the WIN project. Studies in Health Technology and Informatics 122: 676–680.

Paul, A.S., and E.A. Wan. 2008. Wi-Fi Based Indoor Localization and Tracking Using Sigma-Point Kalman Filtering Methods. Pp. 646–659 in Proceedings of 2008 IEEE/ION Position, Location, and Navigation Symposium. Piscataway, N.J.: IEEE.

Pavel, M., A. Adami, M. Morris, J. Lundell, T.L. Hayes, H. Jimison, and J.A. Kaye. 2006a. Mobility Assessment Using Event-Related Responses. Pp. 71–74 in Proceedings of the 2006 Transdisciplinary Conference on Distributed Diagnosis and Home Healthcare. Piscataway, N.J.: IEEE.

Pavel, M., T.L. Hayes, A. Adami, H. Jimison, and J. Kaye. 2006b. Unobtrusive Assessment of Mobility. Pp. 6277–6280 in Proceedings of the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Piscataway, N.J.: IEEE.

Pavel, M., T. Hayes, I. Tsay, D. Erdogmus, A. Paul, N. Larimer, H. Jimison, and J. Nutt. 2007. Continuous assessment of gait velocity in Parkinson’s disease from unobtrusive measurements. Pp. 700–703 in Proceedings of the 3rd International IEEE/EMBS Conference on Neural Engineering, CNE ’07. Piscataway, N.J.: IEEE.

Stefanov, D.H., Z. Bien, and W.C. Bang. 2004. The smart house for older persons and persons with physical disabilities: structure, technology arrangements, and perspectives. IEEE Transactions on Neural Systems Rehabilitation Engineering 12(2): 228–250.

Wild, K., L. Boise, J. Lundell, and A. Foucek. 2008. Unobtrusive in-home monitoring of cognitive and physical health: reactions and perceptions of older adults. Journal of Applied Gerontology 27(2): 181–200.

 

About the Author:Misha Pavel is professor and division head of biomedical engineering. Holly Jimison is associate professor of medical informatics and clinical epidemiology. Tamara Hayes is an assistant professor in the Department of Biomedical Engineering. Jeffrey Kayene is professor of neurology and biomedical engineering and diretor of the Layton Aging and Alzheimer's Disease Center. All are at the Oregon Health & Science University in Portland.