GRADUATE SCHOOL OF PUBLIC HEALTH PILOT GRANT INITIATIVE COMPUTATION AND SYSTEM MODELS IN PUBLIC HEAL

1
GRADUATE SCHOOL OF PUBLIC HEALTHPILOT GRANT
INITIATIVECOMPUTATION AND SYSTEM MODELS IN
PUBLIC HEALTHSocial Networks and Tobacco Use
in First Year College Students

• PI Stephanie R. Land, Ph.D.
• Department of Biostatistics

2
Research Team

• Brian Primack, MD, MEd, MS, co-I
• Deborah Moss, MD, Co-I
• Ju-Sung Lee, PhD, Consultant
• Melissa Jones, MPH, CHES, Study Coordinator

3
Tobacco use during the first year of college

• 21.2 of Pitt undergraduates smoke cigarettes
• Overall, cigarette smoking declines during
  college, although there is some initiation (11
  of non-smokers in one study)

• Hookah (waterpipe) smoking uptake seems to double
  (15 to 30 ever use) during the first year of
  college

4
How does tobacco use relate to social networks
among first year college students?
Are smoking students in the center of social life
or marginalized? How is a students tobacco use
predicted by friends attitudes? Are the answers
different for cigarettes versus hookah?
5

• Smokers in a cluster of the Framingham social
  network, 2001-2002
• (Christakis, NEJM, 2008)
• Smokers became socially marginalized from 1971 to
  2003

6
Specific Aims To examine

• Feasibility of dormitory social network study
• Longitudinal trends in social network and tobacco
  outcomes over the course of the year
• Associations between quantitative features of the
  network and tobacco outcomes in college students.
  Additionally, we will estimate the extent to
  which students select friends with similar
  behavior, rather than influencing one another to
  adopt similar behavior.

7
Research Plan

• Questionnaire survey to be completed in residence
  hall common room.
• Participants students of 10 selected first-year
  floors (n280)
• Resident assistant helps to recruit and
  administer survey
• Surveys in March 2009 (pilot), August 2009,
  January 2010.
• Incentives toys, pizza, raffle iPod Shuffle

8
Questionnaire

• Items regarding participant (ego)
• Demographics
• Smoking status
• Cigarette attitudes, beliefs
• Hookah attitudes, beliefs
• Items regarding friends (alters)
• Strength of all pairwise associations (including
  those between alters)
• Smoking behavior, approval

9
Mapping the social network

• Software Statnet Ver. 2.1 (R)
• Measures
• Mean ego-network density ratio of the number of
  ties among that egos contacts, divided by the
  number of possible ties.
• Centrality number of ties reported by an ego
• Cluster size number of smokers connected
• Analyses will assume undirected associations.

10
Aim 1 (Feasibility) Analysis

• Participation rates (expect n28/floor)
• Compare demographics of sample to University
  first year students (?2 tests)
• Mean ego-network density (0.4 to 0.8 is typical)
• Adequate variability in centrality

11
Aim 2 Analysis

• Ever-use of cigarettes at baseline versus hookah
  (McNemars test)
• Initiation of cigarettes versus hookah (McNemars
  test)
• Tabulate conversion between smoking behaviors
• Density and centrality baseline mid-year

12
Aim 3 Analysis

• Repeated measures logistic regression of
  participants mid-year cigarette smoking status,
  with explanatory variables
• participants centrality at baseline mid-year,
• participants smoking behavior at baseline,
• smoking behavior at baseline and mid-year of the
  students social contacts,
• perceived smoking approval of the contacts at
  baseline mid-year,
• interactions (e.g., between the students
  contacts baseline smoking)
• sex, race, educational achievement and
  socioeconomic status of the student.
• To include alters whose relationship either
  arises or dissolves between fall and winter, we
  will classify alters as not a contact,
  non-smoking, or smoking at each time point.

13
Aim 3 Analysis (contd)

• Selection versus influence?
• We will use the exponential random graph
  approach, which models the probability of tie
  creation/deletion from baseline to mid-year as a
  function of the behavioral characteristics of the
  alters.

14
Aim 3 Analysis (contd)

• Logistic regression analysis will be repeated for
  hookah, and for smoking attitudes and intentions
• Graphically compare smoker cluster sizes between
  cigarette and waterpipe, and between time points

15
Continuation study

• Contingent on additional funding
• Survey same students in April, 2010
• Additional analyses
• How associations between network features and
  tobacco variables differ based on demographics
  and friend directionality
• Permutation test for the existence of clusters of
  smokers or students with pro-tobacco attitudes,
  beliefs and intentions. Clustering in the
  observed network is compared to that in simulated
  networks.
• Similarly estimate the influence of one students
  tobacco outcomes on another students, as a
  function of their social distance (degrees of
  separation).

16
Future directions

• Larger college study
• Socially-based interventions for tobacco use
  prevention
• Facebook to obtain network and tobacco use
  information, and transmit smoking prevention
  messages.
• Adult populations, e.g. occupational setting,
• Other health behaviors, including obesity and
  fitness.

17
Limitations/Strategies

• Not including all alters as participants. E.g.
  cant measure eigenvector centrality, which
  counts students contacts, weighting each contact
  by the number of other contacts he or she has.
• Low participation at baseline? Consider hosting a
  second event or performing assessments via an
  email survey conducted by each RA.
• Participation at winter survey low? Biased
  self-selection at the second time point?
  Targeted survey of the missing baseline
  participants.
• Sparse social network of isolated individuals?
  Conduct a secondary recruitment of alters.
• Can construct network based on student major and
  dormitory floor residence, although past research
  regarding tobacco use suggests that neighbors are
  not as influential as friends and coworkers.

18
xij 1 if there is an edge between node i and
node j c(?) is a normalizing constant. ?t is
transpose of ?.
David Hunter, http//www.stat.psu.edu/dhunter/tal
ks/ergm.pdf
19
Timeline