TeleMed and eHealth‘06 Lifestyle Monitoring as a Predictive Tool in Telecare Hanson, J. 1, Osipovič, D. 1, Hine, N. 2, Amaral, T. 2, Curry, R. 3 and Barlow,

TeleMed and eHealth‘06

Lifestyle Monitoring as a Predictive Tool in Telecare

Hanson, J.1, Osipovič, D.1, Hine, N.2, Amaral, T.2, Curry, R.3 and Barlow, J.3

1 The Bartlett School of Graduate Studies, University College London, UK2 School of Computing, University of Dundee, UK

3 Tanaka Business School, Imperial College London, UK

What is telecare?“Care provided remotely by means of information and communication technology (ICT) to people in their own homes.” (Curry et al., 2003).

• Telecare at home can be provided by deploying a wide variety of sensors which monitor:- Security and safety of the domestic environment - Personal safety - Vital health signs - Daily activities or lifestyle

PIR Chair sensor Door sensorFall detector

Three generations of telecare

Three generation of Telecare (Porteus and Brownsell 2000)

• Generations 1 and 2 are in active mode. Real time response to an emergency.

• Third generation is in passive mode. Continuous monitoring of a person’s behaviour within the home.

2nd generation

1st generation

3rdgeneration

Prediction of possibleacute situations

Automatic detection and generation of alert calls

Personal response without system intelligence

• The rationale behind this way of delivering care is an assumption that it will allow older people and people with longstanding health conditions:

- to live independently in their homes for longer

- and at the same time it will save public resources

• £80m Preventative Technology Grant has been made available to local authorities in England for implementing telecare over the next two years.

Aims of Telecare

Why is it needed?

012

3456

789

1998 1999 2000 2001 2002 2003 2004 2005

• Significant demographic, political and economic drivers – ageing, longstanding illness and chronic conditions, care system capacity

• Wider policy agenda provides impetus: focus on capacity, chronic disease, prevention and self care

• Targets. DH, Delivering C21 NHS IT Support (2002, reiterated by ODPM in Nov 2005) and Building Telecare in England (July 2005)

Government and other official reports calling for telecare

Challenges of Telecare• Telecare often presented as an “all win” solution, but

implementation faces challenges:

- technological, not just about re-engineering existing services but moving on from adding to the widely available social alarm technology to an intelligent sensor system

- organisational, diverse and complex service involving a range of stakeholders. Need for clear policy and strategy

- cost-effectiveness, requires tools for evaluation

- ethical, telecare touches on issues of surveillance, empowerment and control. Clear guidance called for when prioritising need, offering the service, obtaining informed consent and activating response protocols

Supporting Independence• Funding from EPSRC (EQUAL Programme)• Interdisciplinary team of academics, charitable housing providers and technology manufacturers• Overall aim is to understand the opportunities for and barriers to ‘mainstreaming’ telecare services in people’s own homes • Study conducted in two locations:

- South Yorkshire – in mainstream housing - Devon – in an extra care housing scheme (the focus in this presentation)

• Use of existing ‘off the shelf’ technology to ‘test’ the LSM concept.

History of LSM• The concept of LSM was developed back in the late 1990s as:

- a non-intrusive, low cost technological solution to enhance care to older people

- reassurance for carers (formal and informal)- proactive rather than reactive service to large numbers of

geographically dispersed clients- cuts the time needed to detect potentially serious problems

• Widespread understanding right across the board that LSM is possible now, and can provide a basis for clinical assessment and intervention.

How LSM works• LSM sensors monitor people’s habitual domestic movements and daily activities such as movement around the house.

– By constant passive monitoring of the domestic environment– integrating into an “intelligent” LSM system, – learning people’s routines – recognising deviations from this norm.

• Some deviations may be interpreted as signs of a forthcoming crisis, in which case upon detecting them an alert is issued to a carer. Therefore the ultimate aim of LSM is to prevent a crisis.

• Our study exposed a number of limitations to this concept of LSM. I will touch upon some of them in this presentation.

Participants’ profiles at the start of the research (October 2005)

Pseudonym of Participant

Gender Age Household Composition

Health Conditions

Mrs Adams Female 80 one person household

vision impairment, hypertension, diabetes

Mrs Bishop Female 68 married couple household

vision impairment, hypertension

Miss Evans Female 84 one person household

vision impairment, hearing impairment, coronary heart disease, epilepsy

Mr Gibson Male 82 one person household

vision impairment, hypertension, diabetes, cancer

Mr Heaton Male 47 one person household

vision impairment, cerebral palsy, hypertension

Mrs Jenkins Female 84 one person household

diabetes, coronary heart disease

Data and Methodological Approach

• Mixed research methodology :

- Monitoring began in January 2006 and lasted for 10 months.

- Detailed floor plans prepared, taking account of the furniture.

- On average 14.8 sensors were installed in each of the six flats.

- Four rounds of in-depth interviews have been conducted.

- All six participants received regular blood pressure monitoring, and two participants received blood sugar monitoring.

- Half way through the project a systematic review of sensor output data was conducted. This allowed us to create a number of “vignettes”.

- Vignettes represent case studies of sensor activity around the time of a specific, known event in the life of our participants, together with an attempt to interpret this pattern of activity with the benefit of available contextual information.

‘Busyness’• Busyness is a measure of movement within a dwelling, and a count of interactions

with sensored objects. These interactions ‘time tag’ aspects of routine and result in regular patterns that can be detected within periods of each day, within days of each week, monthly, annually and so forth.

• Busyness is measured by normalising the overall sensor activity graphs by the number of sensors active in the particular period of time, applying weightings to each sensor depending on number of firings.

• Busyness should not be attributed directly to people, but to the firing patterns of the sensors in their home

• Flagging up changes in the busyness of individual sensors or sensor arrays that indicate deterioration of health or well-being.

Example of a Vignette Emergency Hospitalisation

• Miss Evans is 84 years old and lives alone in a bedsit flat that she rents from the charitable housing provider. She has both vision and hearing impairment, epilepsy and a history of coronary heart disease. Her personality could be described as “a bit of a worrier”.

• She has a fairly regular daily routine. For example round about lunch time she normally takes an afternoon nap. As she explains herself:

• “I go to dinner and after dinner I have a little nap on the bed, not a sleep so much as to get my back flat because the doctor said I must lie flat for an hour or so during the day.”

little oven

D

B

C

Ep

E

D

ECBD

D

DE

BEmai01

EPmai01

PImai01

CHmai01

PImai02

DRmai01

PIbat01

DRbat01

DRhal01

PIhal01

PIkit01

ELkit01

ELkit02

DRkit01

ELmai01

DRkit02

Sensors installed in Miss Evans’ flat

The Event Itself• Until the day of the crisis all of Miss Evans’ blood pressure readings were within a range considered “normal” for this individual.

• On the 3rd of March 2006 quite ‘out of the blue’ Miss Evans felt very ill, her blood pressure was very high and she was taken to hospital. She was discharged from hospital on the 17th of March 2006.

• The hospitalisation of Miss Evans is precisely the type of crisis event that lifestyle monitoring system aims to predict and prevent. In this case, analysis of lifestyle monitoring devices was done retrospectively.

Daily firings of bed occupancy sensor

000005

010015

020025

030035

040

Jan-

26Ja

n-28

Jan-

30F

eb-0

1F

eb-0

3F

eb-0

5F

eb-0

7F

eb-0

9F

eb-1

1F

eb-1

3F

eb-1

5F

eb-1

7F

eb-1

9F

eb-2

1F

eb-2

3F

eb-2

5F

eb-2

7M

ar-0

1M

ar-0

3M

ar-0

5M

ar-0

7M

ar-0

9M

ar-1

1M

ar-1

3M

ar-1

5M

ar-1

7M

ar-1

9M

ar-2

1M

ar-2

3M

ar-2

5M

ar-2

7M

ar-2

9M

ar-3

1A

pr-0

2A

pr-0

4A

pr-0

6A

pr-0

8A

pr-1

0A

pr-1

2A

pr-1

4A

pr-1

6A

pr-1

8A

pr-2

0A

pr-2

2A

pr-2

4A

pr-2

6A

pr-2

8A

pr-3

0M

ay-0

2

BEmai01

Series2

Bed occupancy sensor

Hourly firings of bed occupancy sensor, 37 days before hospitalisation

000

010

020

030

040

050

060

070

080

090

0 1 2 3 4 5 6 7 8 9 10

11

12

13

14

15

16

17

18

19

20

21

22

23

BEmai01

Hourly firings of bed occupancy sensor, 37 days after hospitalisation

000

010

020

030

040

050

060

070

080

090

0 1 2 3 4 5 6 7 8 9 10

11

12

13

14

15

16

17

18

19

20

21

22

23

BEmai01

Chair occupancy sensor

000010020030040050060070080090

Jan

-25

Jan

-27

Jan

-29

Jan

-31

Fe

b-0

2F

eb

-04

Fe

b-0

6F

eb

-08

Fe

b-1

0F

eb

-12

Fe

b-1

4F

eb

-16

Fe

b-1

8F

eb

-20

Fe

b-2

2F

eb

-24

Fe

b-2

6F

eb

-28

Ma

r-0

2M

ar-

04

Ma

r-0

6M

ar-

08

Ma

r-1

0M

ar-

12

Ma

r-1

4M

ar-

16

Ma

r-1

8M

ar-

20

Ma

r-2

2M

ar-

24

Ma

r-2

6M

ar-

28

Ma

r-3

0A

pr-

01

Ap

r-0

3A

pr-

05

Ap

r-0

7A

pr-

09

Ap

r-11

Ap

r-1

3A

pr-

15

Ap

r-1

7A

pr-

19

Ap

r-2

1A

pr-

23

Ap

r-2

5A

pr-

27

Ap

r-2

9M

ay-

01

Ma

y-0

3

CHmai01

Daily firings of chair occupancy sensor

000020040060080100120140160180200

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

CHmai01

000020040060080100120140160180200

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

CHmai01

Hourly firings of chair occupancy sensor, 37 days before hospitalisation

Hourly firings of chair occupancy sensor, 37 days after hospitalisation

Validation of the vignette• Although this interpretation seems to fit in well with the presented picture of sensor activity, without some form of validation it remains speculation.

• We therefore re-interviewed Miss Evans some weeks after she had returned from hospital, to see how she felt about the episode.

• Factors emerged which suggest that it would be premature to attribute the change in her activity pattern to a change for the better in her health or well being.

- Miss Evans did not make a rapid recovery but rather took a long time to get better.

- a medication change altered her routine

- daily visits from care workers during the recovery period also altered the participant’s routine

From a retrospective to a predictive model

• The construction of vignettes is a good way of becoming familiar with sensor output data and therefore it is an appropriate technique to be used at the exploratory and preliminary data analysis stages.

• However, even when used retrospectively, the vignette must be used cautiously, avoiding over-interpretation of sensor data.

• Even it is possible to make a retrospective link between busyness and the build up to a crisis, we cannot assume that this will automatically translate into a predictive approach. The build up to the next crisis could be different.

Making sense of sensors

• In order to “make sense of sensors” one needs rich contextual information about the events and everyday activities of participants to set alongside the sensor data

• This is crucial both at the stage of choosing the right LSM sensors and when interpreting their output

• Obtaining such information depends on establishing continuous effective communication - a dialogue of care - which requires a rapport between patients and their carers.

Lessons for mainstream telecare• What are the challenges if we want to move towards predictive telecare?

• Technological. Gathering and analysing data on LSM is currently time consuming and laborious. We need more sophisticated sensors, data mining and visualisation techniques. Intelligent decision making tools.

• Organisational. Decisions will need to be made about which activities are monitored, who monitors them and which pattern of sensor firings represents a “norm” in the case of any particular individual.

• Resource. Who is going to pay for the service? Can we measure the costs and benefits?

• Ethical. Explaining the concept, capturing user wishes, and obtaining informed consent.

LSM ‘Lite’• There are potential benefits to LSM for both active / reactive and passive / predictive telecare.

• In active mode the signals are usually clear. There is either an alert or there is not.

• In passive mode the signals are both less clear and more open to interpretation. The real challenge is knowing when a weak signal indicated by a change in busyness is showing that something is going wrong.

• In the long run, LSM will probably have to become much more sophisticated but in the short term we may need to make it simpler - LSM ‘Lite’ - selecting one or a few key sensors to monitor, based on individual case histories.