Contact diaries

Report
Network Measurement: past and future
Ken Eames
Centre for the Mathematical Modelling of Infectious Diseases
London School of Hygiene and Tropical Medicine
Outline
Measuring networks:
• Direct measurement
• Proxy measures
Complications:
• Dynamics
• Influence of infection
• What we really want
Ongoing/future studies:
• Linking contact behaviour to infection
• Improved inference
• Low-tech and high-tech solutions
Networks in epidemiology
As far as we’re concerned, a network consists of a population
and its interactions.
Conlan et al, Proc. R. Soc. B, 2011
Node
Link
Node attributes: age, gender, location, vaccination status...
Link attributes: duration, frequency, stability, proximity,
timing…
Example: sexual networks in Canada
•
•
Contact tracing, linked to medical records.
Reported sexual partnerships, and infection status.
Wylie & Jolly, Sex. Transm. Dis., 2001
Direct measurement
•
•
•
•
Observation: visual, video.
Contact tracing: with or without infection information.
Contact diaries: paper-based, interview-led, or electronic.
Electronic tagging: radio, Bluetooth.
Please cut off this
column before
returning the form.
This questionnaire is for the pers
can complete it on their behalf but
Measuring social networks: contact diaries
Today’s date ___________
Age
Gender
Was
(or age (Male or there
range) Female) skin to
skin
contact
(Yes or
No)?
Over th
you wit
Under 5 mins
Name (or description) of
contact
Edmunds et al, Proc. R. Soc. B, 1997; Mossong et al, PLoS Med, 2008
Measuring social networks: contact diaries
Who?
Physical
contact?
Duration
Location
Frequency
Contact diary
This questionnaire is for the person who has been given antiviral medication . If that person is a child then an adult
can complete it on their behalf but should respond from the point -of-view of the child.
Today’s date ____________________
•
•
•
Never met
before
Less than
monthly
Once or twice
monthly
Once or twice
weekly
Daily or
almost daily
Other
Leisure
activity
Travel
How often?
How often do you normally meet
this person (tick one)?
Work/ School/
College
Where?
Where did you meet this person
(tick all that apply)?
Over 4 hours
(Yes or
No)?
How long?
Over the day, for how long were
you with this person (tick one)?
1-4 hours
Was
(or age (Male or there
range) Female) skin to
skin
contact
10mins-1 hour
Gender
5-10 mins
Age
Under 5 mins
Name (or description) of
contact
Day of the week____________________
Home
Please cut off this
column before
returning the form.
List of all contacts over the course of day(s).
Collect information about participant, the people they meet,
and the encounter.
Low-tech, but lots of information.
Please turn over if you need more space
Contact diaries: Warwick contact network
• If names are recorded accurately we can use contact diaries
to reconstruct a social network.
• Only likely to be possible in a limited setting.
• 8661 encounters, 3528
different individuals.
• 49 participants.
• 310 shared contacts.
• 3218 one-off contacts.
Read, Eames, & Edmunds J. R. Soc. Interface, 2008
Contact patterns: who mixes with whom?
• If names are not recorded, we can still get useful data.
• Studies show high levels of like-with-like mixing by age.
• Largest amount of mixing is between school children.
• It’s useful and interesting, but it’s not a network.
Data: POLYMOD; Image: Rohani & King
Networks of attribute types
•
•
•
Might not need a network (i.e. named contacts) study at all.
Depends on how much people know about those they
interact with.
Depends on what we want to do with the data.
Town A
young
old
young
old
Town B
Contact diaries and ego networks
•
•
Familiar contact diary, plus clustering question.
Aim to determine which contacts know each other.
Contact diaries and ego networks
Home
Work
Respondent
Social
•
•
Information about clustering within egonet.
Measure of clustering by age, social setting, etc.
Measuring social networks: electronic tags
Measuring social networks: electronic tags
• Bluetooth has the advantage that it’s commonly accessible.
• The disadvantage that it’s fairly long-range (& goes through
walls).
• Custom-designed RFID tags can be tuned to record at a
required distance, but are relatively expensive.
• Both approaches allow finer temporal resolution data to be
collected.
Catutto et al, PLoS One, 2010
Measuring social networks: proxy measures
• Phone call/social network data: alterative network.
• Travel patterns: mass movements for spatial models (e.g.
air-travel data); individual movements for networks (e.g. cattle
movements, commuting data).
• Location & time-use data: inferring networks from location
information.
Place
Person
1
A
Place
Person
3
2
3
B
4
C
5
Time 1
1
1
A
2
B
Network
4
C
5
Time 2
Þ
2
5
4
3
Issues with network data collection
•
•
•
•
•
•
•
•
•
Fatigue: reporting lots of contacts gets very boring.
Recall: unless reporting is instant, contacts are forgotten.
Accuracy: about the person: how old?
about the encounter: how long did we meet for?
Matching contacts: how many John Smiths are there?
Partial samples: if we only sample a small fraction of the
population, is the network still meaningful?
Boundaries: where should the study stop?
Definition: how do we define a meaningful contact?
Many of these are issues for any social contact data study,
not just network studies.
Automated data collection solves some problems, but only
really suitable (currently) for small, specific, populations – a
school, or a hospital.
Other things that matter: dynamics
Networks constantly change:
Natural dynamics – birth, death, ageing.
Seasonal changes – school holidays, weekends.
Day-to-day variation; predictable and chance encounters.
• The more dynamic a
network is, the harder it is to
measure.
• Population mixing becomes
less “network” and more
“mass action”.
• Can return to betweengroup rather than betweenindividual models (maybe).
Number of contacts
•
•
•
•
Participant
Other things that matter: illness
•
Illness can have a large effect on social behaviour: time off
work/school means fewer contacts, and a different agedistribution of contacts.
Network likely to be reduced – for people who are heavily
symptomatic – to “strong” home contacts.
100
5.5
●
80
60
40
●
2.61
●
1.39
●
●
20
●
●
0
Home
Average No. of Contacts
Duration of Contacts, %
•
<5 min
5−10 min
10 min−1h
1−4 h
>4h
NA
0.73
●
●
●
Leisure Other
0.04
Work
Location of Contacts
• Impact of illness on the healthy: social distancing, from real
or perceived risk.
What do we really want?
•
•
•
•
•
•
We want to know “how people interact”, but this is very woolly.
What we really want to know is whom each person would
infect if they were infectious with pathogenTopical.
There are lots of factors we don’t understand, so instead try to
determine whom they’ll meet when they’re ill (&, ideally, give
each encounter a transmission probability).
We’re not going to be able to measure encounters when
people are ill… so instead just measure normal mixing.
How do we define “normal”? Let’s just measure mixing today.
What do we mean by a contact? Let’s just measure what we
can.
Linking contacts and infection
•
•
Want to link contact behaviour (ideally, network position) to
risk of infection.
Combine social contact studies and swabbing/symptom
surveillance.
•
Which properties of an individual, or their network, matter?
• Do we really need to measure everything?
• Until we have measured “everything”, we can’t be sure.
•
•
Is it going to be the same for every infection? Obviously not.
Is it going to be the same in different populations or at
different times? Probably not.
Can we find robust proxies? Maybe.
•
Linking contacts and infection
Gardy et al, NEJM, 2011
Linking contacts and infection
Figure removed –
unpublished data
from Kucharski et al
• 762 survey participants in
Hong Kong.
• Social contact data and
paired sera pre- and post2009 H1N1 epidemic.
Kucharski, Cowling, Lessler, Read, Cummings, Riley, …
Statistical inference framework
Welch et al, Epidemics, 2011
To do
•
•
•
•
•
Fill in the low-tech gaps. Simple behavioural things we
know very little about. Illness, seasonality, stability.
Find out what matters, of the stuff we’re missing –
modelling to exploit fully current data and guide future data
collection.
Match the model (& the data) to the intervention.
Link to outbreak/illness data.
Innovate (not forgetting to handle privacy and anonymity
sensitively).
“It is essential to begin joint studies that involve epidemiologists,
biostatisticians, and modellers. Maybe such an approach would
make everyone hold hands, smile and produce something novel.”
Sally Blower, in Models for Infectious Human Diseases; their
structure and relation to data

similar documents