Measuring a Human Contact Network for Epidemiology Research Maria - - PowerPoint PPT Presentation

▶

Jan 11, 2024 153 likes •470 views

Measuring a Human Contact Network for Epidemiology Research Maria Kazandjieva, Jung Woo Lee, Marcel Salath Marcus Feldman, James Jones, Philip Levis Stanford University HotEmNets 2010 Contact Networks Specify physical proximity interactions

SLIDE 1

Measuring a Human Contact Network for Epidemiology Research

Maria Kazandjieva, Jung Woo Lee, Marcel Salathé Marcus Feldman, James Jones, Philip Levis Stanford University

HotEmNets 2010

SLIDE 2

Contact Networks

Specify physical proximity interactions over time Used by National Institute of Health Center for Disease Control Key in determining how to limit disease spread e.g. vaccinations

SLIDE 3

Epidemiology’s Problem

Not enough data! Pen-and-paper and survey methods are inaccurate have limited coverage

SLIDE 4

The Alternative

Use technology to get more fine grained and accurate data Previous work has used cell phones and RFID on small populations

SLIDE 5

The Deployment

8-hour day at a U.S. high school

closed network environment

850 participants

students, teachers, staff

5+ school buildings 170 location nodes

SLIDE 6

the motes

SLIDE 7

The Code

Mote type Number Beacon interval TX Power participant 850 20 sec

16.9 dBm

location 170 20 sec

11 dBm

Over 1000 motes broadcasting periodic beacons

SLIDE 8

Data Format

source bcn_seqno rssi local_seqno

SLIDE 9

Preparation

Program 1000+ motes Place 170 location motes the night before Prepare batches of 10 to 15 motes for teachers to distribute to students Brief the school

SLIDE 10

D-Day

In at 6:30 am, out by 9 pm Distribute motes, pouches, and assent forms Participants write down the time and start the mote by inserting the second battery 4 pm - motes start to come back

SLIDE 11

SLIDE 12

SLIDE 13

The Rest of This Talk

Pre-deployment considerations working with epidemiologists working with human subjects Post-deployment woes node resets and disconnections

SLIDE 14

Epidemiology Considerations

Trade-off between time resolution of data and coverage

20 seconds is good enough

SLIDE 15

Opting in and out

user button vs. batteries

Indicate state of the mote

red vs. yellow LED

Human Subject Considerations

SLIDE 16

The Deployment

8-hour day at a U.S. high school 850 participants 5+ school buildings

SLIDE 17

D-Day Outcome

792 traces from participants Total of 3 million contact entries

SLIDE 18

But During the Deployment …

Batteries fell out Batteries were taken out Students played with the reset button Motes were accidentally bumped against desks Students banged motes against hard surfaces Students swung motes by the lanyards Motes were rubbed together … to see what happens

SLIDE 19

Node reboots

272 nodes with uninterrupted data traces a.k.a. ‘safe motes’ 520 nodes with a total of over 1500 reboots

SLIDE 20

Hardware Resets

source bcn_seqno rssi local_seqno

450 300 220 370 789 414 223 370 450 301 221 371 450 302 220 0

SLIDE 21

Extended Disconnection

80 60 local 200 120 60 safe node 150 330 180 40

SLIDE 22

Extended Disconnection, case 2

200 60 local 200 60 safe node 330 300 40 8 am 30

SLIDE 23

Global Time

No time synchronization code on the motes Only need relative time to place all subject interactions in context Tactic: choose a popular mote!

SLIDE 24

Global Time, p2

Everyone likes food… 93.6% of participants received a beacon from one mote in the dining area it became the global reference clock

SLIDE 25

Global Time mote 101 example

source bcn_seqno local_seqno 450 10 0 ….. ….. 10055 1750 50 global 1700 1750

SLIDE 26

Global Time, result

A lookup table

Node ID Global time at start 101 1700 102 1800 … … 145 ???

SLIDE 27

You did all this work, so what?

SLIDE 28

SLIDE 29

Takeaways

Sensor networks provide an opportunity improve the understanding of disease spread. Large-scale deployments can benefit from standard tools for mass programming.

SLIDE 30

Measuring a Human Contact Network for Epidemiology Research

Maria Kazandjieva, Jung Woo Lee, Marcel Salathé Marcus Feldman, James Jones, Philip Levis Stanford University

Contact Networks

Specify physical proximity interactions over time Used by National Institute of Health Center for Disease Control Key in determining how to limit disease spread e.g. vaccinations

Epidemiology’s Problem

Not enough data! Pen-and-paper and survey methods are inaccurate have limited coverage

The Alternative

Use technology to get more fine grained and accurate data Previous work has used cell phones and RFID on small populations

The Deployment

8-hour day at a U.S. high school

850 participants

5+ school buildings 170 location nodes

the motes

The Code

Mote type Number Beacon interval TX Power participant 850 20 sec

location 170 20 sec

Over 1000 motes broadcasting periodic beacons

Data Format

source bcn_seqno rssi local_seqno

Preparation

Program 1000+ motes Place 170 location motes the night before Prepare batches of 10 to 15 motes for teachers to distribute to students Brief the school

D-Day

In at 6:30 am, out by 9 pm Distribute motes, pouches, and assent forms Participants write down the time and start the mote by inserting the second battery 4 pm - motes start to come back

The Rest of This Talk

Pre-deployment considerations working with epidemiologists working with human subjects Post-deployment woes node resets and disconnections

Epidemiology Considerations

Trade-off between time resolution of data and coverage

20 seconds is good enough

Opting in and out

user button vs. batteries

Indicate state of the mote

red vs. yellow LED

Human Subject Considerations

The Deployment

8-hour day at a U.S. high school 850 participants 5+ school buildings

D-Day Outcome

792 traces from participants Total of 3 million contact entries

But During the Deployment …

Batteries fell out Batteries were taken out Students played with the reset button Motes were accidentally bumped against desks Students banged motes against hard surfaces Students swung motes by the lanyards Motes were rubbed together … to see what happens

Node reboots

272 nodes with uninterrupted data traces a.k.a. ‘safe motes’ 520 nodes with a total of over 1500 reboots

Hardware Resets

source bcn_seqno rssi local_seqno

450 300 220 370 789 414 223 370 450 301 221 371 450 302 220 0

Extended Disconnection

Extended Disconnection, case 2

Global Time

No time synchronization code on the motes Only need relative time to place all subject interactions in context Tactic: choose a popular mote!

Global Time, p2

Everyone likes food… 93.6% of participants received a beacon from one mote in the dining area it became the global reference clock

Global Time mote 101 example

source bcn_seqno local_seqno 450 10 0 ….. ….. 10055 1750 50 global 1700 1750

Global Time, result

A lookup table

Node ID Global time at start 101 1700 102 1800 … … 145 ???

You did all this work, so what?

Takeaways

Sensor networks provide an opportunity improve the understanding of disease spread. Large-scale deployments can benefit from standard tools for mass programming.

Discussion