Spatial and Computational Models of Risks for Alcohol Users

1
Spatial and Computational Models of Risks for
Alcohol Users

• Edward J. Wegman
• University of Cambridge and George Mason
  University
• Joint work with Yasmin H. Said and William F.
  Wieczorek

2
Agenda

• Spatial Statistics and GIS
• Risk Factors and Social Indicators
• Erie County, New York Risk Factors
• Multivariate Visualization
• Spatial Analysis using CCMaps

3
Spatial Statistics and GIS

• Statistical methods are often used in health
  studies including alcohol studies in order to
  confirm hypotheses about health risks.
• These relatively elementary techniques do not
  exploit the broader newer methods of multivariate
  data visualization and spatial statistics.
• The ability to manipulate multivariate spatial
  data offers the possibility of extracting
  additional meaning and suggests not only the
  possibility of a confirmatory role for
  statistical methods, but also an exploratory
  role.

4
Spatial Statistics and GIS

• Statistical spatial analysis often begins with
  spatial analysis using a geographic information
  systems (GIS).
• Such systems allow the analysis of distance and
  connectivity including
• The measures of distances between points and
  between points and centroids, analysis of
  adjacency, analysis of networks including roads
  and other transportation systems, and analysis of
  buffer areas between otherwise adjacent areas.
• Spatial analysis of this sort can give insight
  into effective distances which may be
  substantially different from apparent Euclidean
  distances.

5
Spatial Analysis and GIS

• Spatial dependencies define the relationships
  among spatially diverse entities, including
  non-random patterns in geographic space,
  clusters, dispersion, and spatial
  autocorrelation.
• Spatial factors are integral to the development
  of alcohol simulation models such as those
  presented in the previous talk by Dr. Said.
• Spatial analysis contributes to hypothesis
  generation, spatial epidemiology,
  multi-level/multi-resolution modeling, spatial
  interaction and travel models, and understanding
  spatial processes in small areas.
• The latter capability allows the development and
  testing of psychosocial models, especially with
  respect to spatial interactions among alcohol and
  drug users.

6
Risk Factors and Social Indicators

• Traditionally, health studies, including alcohol
  studies, collect data by surveys which provide
  data at the individual level.
• It is not always possible to collect data at the
  individual level because of cost, privacy, or
  lack of resources.
• In many situations it is impractical or
  impossible to measure a specific outcome such as
  early drinking, adolescent drug use, or alcohol
  dependence.
• In contrast, information may be easily available
  on factors associated with these phenomena such
  as poverty, immigration status, language
  facility, and alcohol availability.

7
Risk Factors and Social Indicators

• Social indicators are numerical data, usually
  archival in nature, that measure the well-being
  of a population.
• There are frequently issues of data quality
  including reliability and validity.
• Is the indicator a stable measure?
• Is the indicator actually related to the
  phenomenon of interest?
• The advantage of using social indicator data
  include
• The use of substantial amounts of
  administratively available data,
• The ability to make data-driven decisions on
  topics that are impractical to measure directly,
• The fact that specific indicators have conceptual
  and evidential relationship to difficult to
  measure outcomes.

8
Risk Factors and Social Indicators

• Disadvantages of using social indicator data
  include
• The fact that data are collected for purposes
  other than their use as indicator, hence, may not
  have statistical validity,
• That there are few direct indicators
  (relationships of indicator to outcome are
  indirect),
• That are few indicators at local geographic level
  (postal code or census tract, most are at county,
  state, or national levels), and
• That there are a huge number of indicators from
  which to select many of which may be overlapping
  and collinear.

9
Risk Factors and Social Indicators

• Social indicators provide an indirect method of
  needs assessment for public health services.
• They show relative need for services and may be
  used to estimate actual need for services in some
  situations.
• In addition coupled with demographic information,
  social indicator analysis allows for tailoring
  services to population characteristics.

10
Risk Factors and Social Indicators

• Indicators can fall into a number of categories
  including neighborhood indicators, family
  indicators, and individual indicators.
• Neighborhood indicators would include
• Availability of drugs and firearms, community
  attitudes toward laws and social norms, attitudes
  favorable to drug use, firearms and crime, state
  of transition and mobility within the
  neighborhood, levels of neighborhood attachment,
  and community disorganization.
• Family-level indicators include
• Extreme economic privation, family history of
  problem behaviors, family management problems,
  family conflict, and lack of commitment to
  schools.
• Individual indicators include
• Alienation and rebelliousness, early academic
  failure, substance abuse, delinquency, lack of
  parental involvement in problem behaviors, and
  teen pregnancy.

11
Erie County Risk Indicators

• Wieczorek and Delmerico (2005) assembled a
  database of risk indicators for Erie County, NY
  using several sources.
• Erie County includes the city of Buffalo, New
  York. This database provides a data-rich snapshot
  of a relatively small county-level geographic
  area.
• The sources include U.S. Census 2000, New York
  State Education Department, New York state
  Department of Criminal Justice Services.
• At the local level, sources include the Center
  for Health and social Research, City of Buffalo
  Police Department, Erie County Board of
  elections, Erie County Department of Health, Erie
  County Department of Mental Health, and the
  Roswell Park Cancer Institute.

12
Erie County Risk Indicators

• Because all indicators are essentially ratios of
  the form cases/population (expressed as percent
  or per 10,000), it is important to avoid
  unreliable indicator values due to small
  populations.
• For this reason an arbitrary threshold of
  population greater than 100 was set. Records for
  zip codes and tracts with populations below 100
  have been removed from the database.
• Sometimes the source data for calculation of the
  indicators were available at a spatial level
  other than census tract or zip code area.
• In these cases risk indicators were first
  calculated at the available level, and then
  imputed to the zip level. The imputation was
  performed using population-based weighting method.

13
Multivariate Analysis

• Alcohol use and abuse can be thought of in terms
  of both a cause and an effect.
• Alcohol use and abuse is a cause insofar as it
  leads to acute outcomes such as DWI/DUI, DWI with
  fatal crashes, assault, domestic violence, child
  abuse, sexual assault, murder, suicide as well as
  chronic outcomes such as cirrhosis of the liver
  and other alcohol induced diseases.

14
Multivariate Analysis

• Some social indicators for these outcomes from
  the Erie County Risk Indicators Database include
  
• crm.dwi (DWI crime),
• de.traffic (fatal crash deaths),
• crm.viol (violent crime),
• de.trauma (trauma deaths),
• jar.viol (juvenile crime),
• crm.drug (drug-related crimes),
• de.suicide (suicide deaths), and
• de.cirrohsis (cirrhosis deaths).

15
Multivariate Analysis

• Conversely, alcohol use and abuse can be thought
  of as being caused by
• poverty,
• marital unhappiness,
• poor education,
• drug and alcohol availability,
• neighborhood factors,
• parental alcoholism, and
• ethnicity issues.

16
Multivariate Analysis

• Some social indicators in the Erie County Risk
  Indicators Database include
• fam.pov (family poverty),
• med.income (median income),
• unem (unemployment),
• divorce (divorce rates),
• nv.married (never married),
• edu.g8 (education below 8th grade level),
• edu.col.d (educated beyond college),
• dropout (dropout rates),
• alc.all (all alcohol outlets),
• alc.off (off license outlets),
• tobacco (tobacco outlets),
• vacant (neighborhood vacancies),
• vote.gen (general voting registrations) and
• poor.eng (poor household English usage rates).

17
Multivariate Analysis

• An indicator of overall alcohol problems for Erie
  County is the rate of admissions to treatment for
  alcoholism and substance abuse.
• The appropriate indicator is oasas.18ov, which is
  the rate per 10,000 by zip code for individuals
  over the age of 18.

18
Multivariate Analysis
19
Spatial Analysis Using CCMaps
20
Spatial Analysis Using CCMaps
21
Spatial Analysis Using CCMaps
22
Spatial Analysis Using CCMaps
23
Spatial Analysis Using CCMaps
24
Acknowledgements

• The work of Dr. Wegman is supported in part by
  the U.S. Army Research Office under contract
  W911NF-04-1-0447.
• The work of Dr. Said is supported in part by
  grant number F32AA015876 from the National
  Institute on Alcohol Abuse and Alcoholism.
• The work of Dr. Wieczorek is supported in part by
  grant number R01AA016161 from the National
  Institute on Alcohol Abuse and Alcoholism and by
  a contract from Western New York United Against
  Alcohol and Drug Abuse/Erie County Department of
  Mental Health.
• The content is solely the responsibility of the
  authors and does not necessarily represent the
  official views of the National Institute on
  Alcohol Abuse and Alcoholism or the National
  Institutes of Health.
• Drs. Wegman and Said were Visiting Fellows at the
  Isaac Newton Institute for Mathematical Sciences
  at the University of Cambridge in Cambridge,
  England. We are indebted for the support
  provided by the Newton Institute, which has made
  the successful completion of this work possible.

25
Contact Information

• Edward J. Wegman
• ewegman_at_gmail.com
• [phone]
• Yasmin H. Said
• ysaid99_at_hotmail.com
• [phone]
• William F. Wieczorek
• wieczowf_at_buffalostate.edu
• [phone]