Non-contact sensor based falls detection in residential aged care - - PowerPoint PPT Presentation

Non-contact sensor based falls detection in residential aged care facilities: Developing a real-life picture

Cecily Gilbert Health and Biomedical Informatics Centre, University of Melbourne

1

Falls facts in Australia

More frequent for those aged 65+:

• 30% of people living at home will have a

fall each year.

• much higher rate in aged care facilities.
• leading reason for admission to

hospital: 38% compared to 13% for transport related injuries.

2

Study Aim and method

Objective: to test the feasibility and acceptability of a ambient non-wearable sensor technology with older participants in a residential care facility. Mixed method approach comprising: a) Empiric study implemented at a residential care facility using purposive sampling b) Evaluation and post-study interviews c) Analysis and review of results

3

Study setting and criteria

4

Purpose-built aged care facility:

• 170 places, 200 staff
• Less than 8 years old
• Bed-exit and pressure mat alarms wired

to nurse-call system

• IT support outsourced

Involvement in study:

• Management & senior staff supportive
• Agreed to screen occupants and approach

eligible residents (or authorised family members) for consent to participate

Participant selection criteria:

• Aged 65+
• Previous falls history (e.g. two or more in

past 6 months)

• Able to walk either independently or with

staff assistance (+/- gait aid)

Exclusion criteria:

• Bed bound, or require hoist for transfers
• Residents currently receiving palliative

care

• Residents who have had no falls in

previous 12 months

4 male residents in the pilot study

• average age 87 years
• complex chronic diagnoses

Assessment Resident 1 Resident 2 Resident 3 Resident 4

Fall risk assessment

HIGH HIGH HIGH HIGH

Previous falls? No history available 5 falls in prior 6 months 13 falls in prior 6 months 1 fall in prior 6 months Uses gait aid? 4-wheel walker 4-wheel walker Wheelchair beyond room 4-wheel walker Mobility assistance? Assist X 1 Assist X 1 Transfer to wheelchair assist X 2 Supervision X 1 Level of care High Care High Care High Care High Care

5

In each participant’s suite:

• one sensor in bedroom
• another sensor in en-suite

bathroom.

Sensor installation

6

Rolled out sequentially:

• set up, test with healthy volunteer,

live-test, then moved to next suite.

• 8 sensors installed in total.

Prototype sensor adapted by industry partner

• Privacy preserving
• Optical, non-contact
• On-board cognitive processing
• Skeletal pose tracking
• Suited to indoor environment
• Designed for 24/7 operation

Depth images

Results

ITEM DATA Sensors functioning 8 installed, only 7 operated reliably Total days of sensor operation 122 Monitoring duration Range 5 – 22 days per participant Data generated 18 GB – 25 GB per day per room, saved as high compression files onto secure dedicated server Fall events One known fall occurred, but was not captured because sensor cable was faulty at the time.

8

Challenges and unexpected events

Gaps in network connectivity e.g. not all room data points were cabled to the core network. Wireless workaround devised, not

• ptimal.

Radio-frequency interference from wall-mounted TV screens in bedroom disabled sensor wireless network: Sensors were re-positioned, but this caused tracking performance to decline, increased ‘noise’. As a result, the event detection threshold setting was raised to detect only medium or high fall-like events. Range and quality of the commercial pose-tracking component in the sensors was more limited in practice than shown in the lab testing: i.e. pose-tracking healthy volunteers was not adequate to determine sensor effectiveness with

• lder person.

9

Acceptability

Post-study interviews with staff indicated strong acceptance:

10

Interview Question Response Did the sensors change the amount or type of contact between residents and staff? No, it really didn’t have any negative impacts

• n anybody.

Do you have any concerns with the display of visual images from the sensors? It’s not a facial picture – just stick figures

• really. So that’s good for privacy and

confidentiality. How is the sensor data useful for residents with cognitive disabilities? I think it would be useful in [the RAC] overall, for people who can’t articulate how a fall happened. Has being part of the sensor trial changed your view about sensors for fall detection or prediction? It’s exciting where the research is taking us…ultimately we want to keep our people safe.

Conclusions – Operational lessons learned

Unexpected technical difficulties delayed full implementation of sensors in participants’ rooms

• End-to-end live testing of hardware and networks is essential before launching.

Optimal placement of sensors is not straightforward

• Suggests need to involve clinicians in realistic calibration for sensors to balance sensitivity and

specificity.

Staff and carer attitudes to the sensors were positive overall

i.e. envisaged a range of benefits if proven to work: knowing the events that lead to a fall, earlier detection of falls if sensors are linked to alarm system etc.

11

Acknowledgements and team members

Thanks to:

• Study participants and their carers
• Staff at the residential facility
• Study Advisory Board members.

Funding gratefully received from the Melbourne Networked Society Institute.

12

Study team:

• A/Prof Ann Borda1
• Dr Cathy Said 2
• Mr Frank Smolenaers 3
• Mr Michael McGrath 4
• A/Prof Kathleen Gray 1
• Ms Cecily Gilbert 1

(1) Health and Biomedical Informatics Centre, University of Melbourne (2) Physiotherapy Directorate, Western Health (3) Australian Centre for Health Innovation, Alfred Health (4) Semantrix Pty Ltd

Thank you

cecilyg@unimelb.edu.au