Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
PACT: Private Automated Contact Tracing
Presented at the IEEE EMBS 2020 Annual Conference
Marc Zissman, PhDMIT Lincoln Laboratory
24 July 2020
Sponsors:
DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited. This material is based upon work supported by the Defense Advanced Research Projects Agency under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Defense Advanced Research Projects Agency. © 2020 Massachusetts Institute of Technology. Delivered to the U.S. Government with Unlimited
Rights, as defined in DFARS Part 252.227-7013 or 7014 (Feb 2014). Notwithstanding any copyright notice, U.S. Government rights in this work are defined by DFARS 252.227-7013 or DFARS 252.227-7014 as detailed above. Use of this work other than as specifically authorized by the U.S. Government may violate any copyrights that exist in this work.
Page 2MAZ 07/24/2020
COVID-19 Infection Progression
Exposure
10 days 20 days5 days 15 days 25 days0
InfectedContact
Symptom Onset
IndexExposure
IndexCase Infectious
Recovery
Recovery
Infectious
Incubation
Page 3MAZ 07/24/2020
COVID-19 Infection Progression
Exposure
10 days 20 days5 days 15 days 25 days0
InfectedContact
Symptom Onset
IndexCase
https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-guidance-management-patients.htmlThe Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Applicationhttps://wwwnc.cdc.gov/eid/article/26/7/20-0282_article https://www.harvardmagazine.com/2020/05/r-noughthttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC7054855/ ; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7121626/
Infectious
Recovery
Recovery
Goal: Find this person before they might infect others• Identify “contacts” that could have infected this person (reverse)• Identify “contacts” that this person could infect (forward)
IndexExposure
Incubation
Infectious
https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-guidance-management-patients.htmlhttps://www.acpjournals.org/doi/abs/10.7326/M20-0504https://wwwnc.cdc.gov/eid/article/26/7/20-0282_articlehttps://www.harvardmagazine.com/2020/05/r-noughthttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC7054855/https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7121626/
Page 4MAZ 07/24/2020
• Traditional uses:– Tuberculosis (2.9 cases / 100k people,
> 900 cases / 100k for COVID-19 Ϯ )– Smallpox– Sexually transmitted diseases
Contact Tracing
Contact tracing is an epidemiological technique used to identify people who have had “contact” with an infected person
Contact tracing can help inform public health interventions to slow virus transmission
Infected
Uninfected
Time
Transmission
Ϯ https://www.cdc.gov/tb/statistics/indicators/2018/incidence.htm for Massachusettshttps://www.cdc.gov/mmwr/volumes/69/wr/mm6915e4.htm
https://www.cdc.gov/tb/statistics/indicators/2018/incidence.htmhttps://www.cdc.gov/mmwr/volumes/69/wr/mm6915e4.htm
Page 5MAZ 07/24/2020
• Prior to COVID-19, contact tracing was primarily a manual process• Primarily used for diseases with longer temporal characteristics
Contact Tracing Tools
Public Health Contact Tracing ToolsChallenges• Index case has to remember who they
were in contact with, where they were
• Labor intensive and time consuming
• Increased risk of data errors
• Difficult to apply analytics
• Does not scale to need
• Need to know identifying information for contacts
Advanced contact tracing tools are urgently needed to handle COVID-19
Page 6MAZ 07/24/2020
Animation Video
Animation Video: https://www.youtube.com/watch?v=yuXzAh4slNw
https://www.youtube.com/watch?v=yuXzAh4slNw
Page 7MAZ 07/24/2020
Likely Late Summer Scenarioin a U.S. State
Approx 6 feet
ChirpLog
Database ofAnonymized Contacts
BluetoothChirp
RecordsInfectionCertification
ChirpLog
BluetoothChirp
Contact Distance
IndexCase
Contact IndexCase
Contacts
ManualContacts Records
Contact Detection
1 COVID-19 Test Confirmed Positive
2 Exposure Query & Notification
3 Contact Actions
4
State Dept of HealthVirtual Community Tracing Center
ManualNotification
AutomaticQuery &
Notification
Test
Symptom check
SelfQuarantine
Contacts
iPhone orAndroid
• Updated Apple & Google OS w/BT contact tracing built-in
• State-acquired and A|G-approved app for user interface
1
• State-acquiredsynchronized database
2
• State-acquired appwill provide tailoredinstruction to contacts
3
Page 8MAZ 07/24/2020
“Too Close for Too Long” – TC4TL
Range
Expo
sure
Tim
e
0m
10min
Hug
30min
1hr
8hrs
1m 2m 3m 4m
1min
SameClassroom
Subway,One Stop,Same Car
HandshakeNo Alert
• Public health authorities define alert region• Engineers try to implement it accuratelyALERT!
Side-by-Side@ Coffee Shop
SameTable @School
NotionalAlert Region
Page 9MAZ 07/24/2020
Assessing TC4TL Systems
Probability of False Positive(Leads to applying effort & treatment unnecessarily to a healthy person)
Prob
abili
ty o
f Fal
se N
egat
ive
(a.k
.a. M
iss)
(Lea
ds to
with
hold
ing
effo
rt &
trea
tmen
t fro
m a
sic
k pe
rson
)
0.01% 99%0.1% 1% 10% 50%
“Flip a Coin”
0.01
%99
%0.
1%1%
10%
50%
System A
System BEverybodyStay Home
Do Nothing
Better SystemPerformance
*A. Martin, A., G. Doddington, T. Kamm, M. Ordowski, and M. Przybocki. "The DET Curve in Assessment of Detection Task Performance",Proc. Eurospeech '97, Rhodes, Greece, September 1997, Vol. 4, pp. 1895-1898.
IdealPerformance
• Want to give public health authorities freedom to pick from a range of operating points
• If we can estimate Pr(pos), Pr(neg), Cost(FN), Cost(FP), we can find the lowest-cost operating point on the DET* curve
Decision Error Tradeoff (DET*) Curve
Page 10MAZ 07/24/2020
Initial Bluetooth Data Collection
Generate “Near” pdfs Generate “far” pdfs
• Empirically measure RSSI vs range for many conditions
– Phone orientation– Location of phone– Phone model – Multipath environment
Probability Density Function of RSSI vs RangeRotation through 3600
RSS
I (dB
m)
Distance (ft)
Prob
abili
ty
RSSI Variability
Phone conditions lead to 20dB (100x) variation in RSSI at fixed distance
Page 11MAZ 07/24/2020
Preliminary TC4TL Performance**
Decision Error Tradeoff (DET) Curve* • 1 of 6, Agnostic: 30 min duration,phone wakes up 6 times, one attenuation per wake up period, no info re phone carriage
• M of 24, Agnostic: 30 min duration,phone wakes up 6 times, 4 samplesper wake up period, no info re phone carriage
• 1 of 6, Cognitive: Same as 1 of 6 above except threshold specific to perfect phone carriage information
• M of 24, Cognitive: Same as M of 24 aboveexcept threshold specific to perfect phone carriage information
• Data: “Range-Angle” data set. Collected at distances of 3 to 15 feet at every 45-degree angle for a variety of conditions (e.g. phone in pocket, hand, purse, bag, etc.)
Better signal processing with additional metadata sent with chirp(no additional power, storage) can reduce EER from ~40% to ~15% EER
*A. Martin, A., G. Doddington, T. Kamm, M. Ordowski, and M. Przybocki. "The DET Curve in Assessment of Detection Task Performance",Proc. Eurospeech '97, Rhodes, Greece, September 1997, Vol. 4, pp. 1895-1898.
** G. F. Hatke et al. “Using BLE Signal Strength Estimation to Facilitate Contact Tracing for COVID-19” https://arxiv.org/abs/2006.15711
Equal ErrorRate (EER) Line(PFA = Pmiss)
https://arxiv.org/abs/2006.15711
Page 12MAZ 07/24/2020
Advanced Proximity Testing
Mannequins on
TurtleBots
Robotic Testing in the ASDF MIT LL @ Waypoint Robotics
Bluetooth
Scanning Apps
• LL custom• 3rd party
ASDF
ASDF = MIT LL Autonomous System Development Facility (Group 76)
Animation Video: "Robot Dance" https://youtu.be/1F6TEvpy5g0
https://youtu.be/1F6TEvpy5g0
Page 13MAZ 07/24/2020
PACT Public Data Repositoryhttps://mitll.github.io/PACT/
https://mitll.github.io/PACT/
Page 14MAZ 07/24/2020
Evaluations: NIST TC4TL Challengehttps://www.nist.gov/itl/iad/mig/nist-tc4tl-challenge
• Explore promising new ideas in TC4TL detection using BLE signal• Support the development of advanced technologies incorporating
these ideas• Measure and calibrate the performance of the state-of-the-art
TC4TL detectors
Page 15MAZ 07/24/2020
User Interfaces
Epidemiology DashboardsMobile Phone App
• For individual citizens
• Alerts user of potential exposure
• Turns exposure tracing on and off
• For public health professionals
• Helps contact tracers manage cases
• Provides summary epidemiological metrics
Page 16MAZ 07/24/2020
PACT Accomplishments
ImPACT 2020 Conference
Exposure notification system for iOS and Android
• Welcome remarks by MA Gov Baker
• Technical exchange of ideas
• 500+ global participants
• A|G ENS consistent with the PACT protocol
• Providing technical advice to Apple and Google– “Too close for too long” detector design– Laboratory RF measurements and analysis
• Weekly meetings with the A|G technical teams
• Trusted technical advisor to US CDC
• Trusted technical advisor to Massachusetts and Pennsylvania
Page 17MAZ 07/24/2020
Advanced Concepts
Improved Inter-Device Ranging “Wearables” and Tokens
Apple | Google Subsystem Compatibility Trustless Servers and Data Stores
UltrasoundUltra wideband(G39’s LLDART) Smart watches
RF tokens & bands
Custom Hardware
Key Exchange
Bluetooth Chirps
LLDART = Low-cost Localization using Distributed Adaptable Response Transponders
Page 18MAZ 07/24/2020
• Contact tracing combined with public health action, such as testing and quarantining, will help reduce virus spread
• Automated contact tracing can supplement manual efforts– Automation of contact analysis– Automatic detection of high-risk exposure events
• PACT seeks to advance the state-of-the-art in private automated contact tracing solutions
• Significant opportunities for future technical innovation exist
Summary
PACT serves as trusted technical advisor to federal, state and local public health authorities (PHAs) and as a convening center for collecting and sharing data and best
practices for private automated contact tracing