GIS and Infectious Diseases

1
GIS and Infectious Diseases
Lecture 10 GIS and RS in Public HealthEdmund
Seto, Ph.D.School of Public HealthUniversity of
California, Berkeley
2
Previous Lecture

• Ecologies of infectious diseases
• Considering those environmental factors that are
  associated with, or which have changed to promote
  the emergence, transmission and spread, and/or
  virulence of infectious organisms
• Environmental niches that support reservoirs,
  vector habitats, exposure, etc.
• The main focus was on vector-borne diseases

3
Example Endemic Schistosomiasis
4
Potential Problems?

• If we know a priori that certain ecologic
  variables really do limit the geographic range of
  diseases (ie. there is biologic plausibility and
  field evidence), then such GIS analyses can be
  useful.
• However, we should be careful! There is possible
  problem of inferring the geographic range of
  disease transmission based only on location of
  disease occurrence.

5

• We can use GIS tools to compute the temp,
  rainfall, etc. for disease locations, but do we
  really know that there is a causal relationship
  between these ecologic variables and disease
  occurrence?
• For example if we used GIS tools to compute the
  mean temp in endemic areas to be 18 degrees, does
  that mean that disease cannot occur at 20
  degrees?
• Not necessarily. The limiting factor might
  really be rainfall, or some other (unmeasured)
  variable.

6
A Solution

• If we compare locations where disease occurs and
  where disease does not occur, we could construct
  logistic regression models that test the
  hypothesis that ecologic variables really do
  infer something about disease occurrence
• P(occurrence) B0 B1(temp) B2(rain) etc.

7

• This regression approach was used in the Lyme
  Disease case study
• Seroprevalence B0 B1(LC1) B2(LC2) etc.
• Where the model predicts the level of
  seroprevalence based on the predictor variables,
  LC1 and LC2, which are the proportion of
  different landcover classes 1 and 2, respectively
  in a given municipality area

8
Other examples

• Lyme Disease
• Lyme Seroprev B0 B1(LC1) B2(LC2)
• Lyme Seroprev B0 B1(greenness) B2(wetness)
  
• Schistosomiasis
• Odds of mountainous snails vs lower yangtze
  snails
• B0 B1(temp) B2(rain) B3(elevation)

9
Problems with Regression Models

• Although the use of regression models might seem
  quite powerful for modeling disease risk based on
  underlying geographic predictor variables, there
  are potential problems with the traditional
  regression models we learn in our stats classes.
• In particular, there is an assumption that
  observations are independent of one another.
• This is not always true with spatially
  autocorrelated data (recall that with spatial
  autocorrelation, observations close in space will
  have similar values).

10
Spatial Autocorrelation

• You would actually expect there to be strong
  spatial autocorrelation with infectious diseases.
• Your risk is dependent upon whether or not your
  neighbor is infected and sneezing on you!

11
Residual Autocorrelation

• Typically, regular regression models are first
  fit to the data.
• Second, residuals are computed
• Residual Model predicted outcome Actual
  outcome
• Third, test if the residuals are spatially
  autocorrelated. ie. are there clustered areas
  where the model over-predicts the risk? Or
  under-predicts risk? This could be tested by
  computing Morans I.

12

• If there is no spatial autocorrelation in the
  residuals then the predictor variables
  sufficiently model the autocorrelation.
• However, if there is spatial autocorrelation in
  the residuals, then you might try adding
  predictor variables to model the correlation.
  Alternatively, there are classes of Spatial
  Linear Regression models that have extra terms in
  the model that attempt to explain the correlation
  based on neighboring outcome and/or predictor
  variables.
• With each new formulation of your regression
  model, you will want to retest for spatial
  autocorrelation of the residuals until you find a
  model without autocorrelated residuals.

13
(No Transcript)
14
More on Spatial Regression Models

• See Bailey and Gatrells Interactive Spatial
  Data Analysis book.
• Tools for Spatial Regression Modeling
• S-plus spatial stats
• SpaceStat

15
Alternative to Regression Models - Monte Carlo
Methods

• Distribution free
• non-parametric (almost)
• Easier to understand
• Results are as good (or better)
• Can be applied to any geography
• Accounts for spatial autocorrelation (spatial
  confounding)

16
Malaria in a Kenyan village
17
Steps

• Question Is proximity to old tires a risk
  factor for malaria?
• Create buffer around cases, and count the number
  of tires in all the buffers Ncases
• Randomly sample a number of controls, buffer
  around them, count the number of tires in the
  buffers Ncontrols_i, and repeat 1000 times to
  generate an empirical distribution of tires near
  the controls.
• Test whether the the Ncases is statistically
  different from the mean of the 1000 Ncontrols_i
  values.

18
Empirical distributions for malaria risks
Cases
Controls
Test for statistical significance
of old tires
19
Trinidad Malaria Study

• Chadee DD, Kitron U Spatial and Temporal
  Patterns of Imported Malaria Cases and Local
  Transmission in Trinidad. American Journal of
  Tropical Medicine and Hygiene 61(1999) 513-517.

20
Background

• Trinidad eradicated malaria in 1965, however
  environmental conditions supporting vector
  habitat still exist on the island.
• But areas surrounding the island have malaria,
  and with the local eradication, the Trinidad is
  susceptible due to lack of immunity in its
  population.
• The challenge is to monitor for malaria cases,
  and to ascertain when malaria is found, if it is
  due to importation or if local transmission is
  occurring.

21
Cluster Analysis

• Calculate Nearest Neighbor (NN) statistic for
  different species of malaria in order to
  determine if there is clustering of cases for
  particular types of malaria
• P. falciparum, P. vivax, P. malariae

22
Spatial Arrangement
Nearest Neighbor Index The Nearest Neighbor
Index measures the degree of spatial dispersion
based on the minimum inter-feature distance and
compares this to what would be expected on the
basis of chance.
23
For each identified cluster

• Follow-up clusters to determine if clusters can
  be traced to overseas travel or local
  transmission
• Results They found clustering of P. malariae and
  P. vivax. The vivax cases were associated with
  an earlier outbreak in 1991. P. malariae was
  traced to local transmission, which relies on a
  particular mosquito vector, A. bellator which
  breeds in bromeliads that grow on the Immortelle
  tree.

24
(No Transcript)
25
Person to Person Diseases

• How can GIS be used in the study of diseases that
  are directly transmissible from person to person?
• For example
• Sexually Transmitted Diseases
• Measles
• Influenza
• Tuberculosis

26
Epidemic Model of Infectious Disease Transmission
Running outof susceptibles
intensity
time
27
Patterns of Transmission

• Contagious Diffusion
• Hierarchical Diffusion
• Network Diffusion

28
Spatial Diffusion
29
Network Diffusion

• Transportation Networks
• Social Networks
• Sexual contact networks - STDs
• Airline travel - Pandemic influenza

30
Spatially-Explicit Metapopulation Models
Traditional models like SIR assume homogenous
populations and homogenous social(spatial)
interaction.
New models are relaxing these assumptions by
modeling subpopulationswith separate
susceptibilities,levels of transmission,
and interactions with othersubpopulations.
Si Ii Rifor each subpopulation i
31
Spatial-Temporal Trends

• How might we visualize these spatial-temporal
  disease patterns with GIS?
• Characterize
• Diffusion pattern?
• How fast is an epidemic spreading?
• Is there a particular point in time where the
  disease spread most rapidly?
• Origin?
• Extent?

32
Map Sequences
Measles cases in Iceland by Month from Nov 1946
to June 1947
33
(No Transcript)
34
(No Transcript)
35
Genetics and GIS

• GIS is playing new role in charting the evolution
  and emergence of pathogens. New genetic
  fingerprinting technology allows researchers to
  track the spatial-temporal spread of particular
  pathogen strains.
• The clearest example of this is studies tracking
  the spread of TB between NY and NJ.
• Biogeographic studies map geography of pathogen
  strains to trace the origin and evolution of
  pathogens.

36
Sexually Transmitted Diseases
37
Baltimore, Maryland Gonorrhea Study

• Becker K, Glass G, Braithwaite W, Zenilman J,
  Geographic Epidemiology of Gonorrhea in
  Baltimore, Maryland, Using a Geographic
  Information System American Journal of
  Epidemiology 147 (1998) 709-716.

38
Gonorrhea

• Caused by Neisseria gonorrhoeae.
• Second only to chlamydial infections in the
  number of cases reported to the Centers for
  Disease Control and Prevention.
• The incidence of gonorrhea is highest in
  high-density urban areas among persons under 24
  year of age who have multiple sex partners and
  engage in unprotected sexual intercourse.
• Recent evidence of antimicrobial resistance.

39
Symptoms

• Symptoms usually appear within two to 10 days
  after sexual contact with an infected partner.
• Women
• bleeding with vaginal intercourse painful
  urination and/or vaginal discharge that is
  yellow or bloody. More advanced symptoms, which
  indicate development of PID, include cramps and
  pain, bleeding between menstrual periods,
  vomiting, or fever.
• Men
• pus from the penis and pain, or a burning
  sensation during urination

40

• Studies of STDs have focused on clustering within
  risk groups (social network diffusion), known as
  core groups.
• Associated with these core groups are core areas,
  geographic areas of increased incidence and
  unusually high transmission.
• Baltimore study geocode residential addresses of
  persons diagnosed with gonorrhea, and assign to
  census tract.

41
(No Transcript)
42
Counts vs Rates?

• Which one better represents the core area?
• But, almost 85 of the people with gonorrhea
  lived outside the core neighborhoods.
• And over 90 of the core area residents did not
  have gonorrhea.
• Issues of unfairly stigmatizing areas.

43
(No Transcript)
44
Immunization
45
Vaccinations

• Background
• Typically its a school requirement now that
  children receive their immunizations
• However, just because they say its so, doesnt
  mean its so in 1998 only 79 of children 19 to
  35 months old had complete immunizations.
• As adults, are we current on our immunizations?

46
Targeting Immunization Programs

• Although there is a big need to identify
  population groups that are under-immunized,
  typically there are no surprises
• Low-income and minority children
• Inner city
• The so-called pockets of need
• The problem each incremental improvement in
  immunization rate is harder than the last! More
  resources are needed, as is better targeting.

47
Florida Study

• J Devine et al. Identifying Predicted
  Immunization Pockets of Need, Hillsborough
  County, Florida, 1996-1997. Journal of Public
  Health Management and Practice 5, no. 2 (1999)
  15-16.
• Create underimmunization risk map

48
Data

• 3 years of data from annual survey of 2 year
  olds immunization levels
• Birth Certificate
• Census Block Data

49
Analysis

• Geocoded each survey response to the appropriate
  census block
• Extracted underlying census block data so that
  each individuals community-level factors were
  ascertained
• Logistic regression model predicting probability
  of complete immunization by age 2 yrs based on
  individual and community level risk factors.
• Map these probabilities for census blocks in the
  county

50
PAGE 106
51
Utah Study of Hepatitis A

• TL Schlenker et al Incidence Rates of Hepatitis
  A by ZIP Code Area, Salt Lake County, Utah
  1992-1996. Journal of Public Health Management
  and Practice 5, no. 2 (1999) 17-18.
• Vaccine-preventable diseases are reportable
• In this Salt Lake County, Utah study, simply
  mapping the case rates for diseases like
  Hepatitis A helps to target immunization programs.

52

• Created a thematic map of hepatitis A incidence
  by zipcode for a five-year period from 1992-1996.
• The incidence data suggested that preschool aged
  children were at highest risk of infection.
• Core areas of infection incidence were identified.

53
PAGE 107
54
Evaluating Intervention

• The previous map can be used to target pockets of
  need for immunization.
• By continuing to monitor rates of
  vaccine-preventable diseases and creating these
  maps after intervention programs, it is possible
  to see the effectiveness of control.
• This is unlike cancer, where the effects take a
  while to manifest themselves.

55
Treatment
56
Tuberculosis

• Background
• TB is a major worldwide disease
• 19-43 of the worlds population is infected
• Estimated 3 million people die each year from TB.
• Within the US, estimated 15 million infected

57
Treatment problem

• Treatment of active TB requires several medicines
  taken for a minimum of 6 months.
• Failure to complete the treatment will result in
  disease recurrence with added risk of drug
  resistance.
• WHO promotes a program called Directly Observed
  Treatment (DOT)

58
South African Study

• Tanser F, Wilkinson D. Spatial implications of
  the Tuberculosis DOTS strategy in rural South
  Africa A novel application of geographic
  information system and global positioning system
  technologies. Tropical medicine and
  international health 4 (1999) 634-638.

59

• TB in Kwazulu-Natal, South Africa greatly
  increased with HIV epidemic.
• Between 1993 and 1997, HIV prevalence among
  adults with TB increased from 36 to 67.
• Active cases are treated 2 wks in hospital
  followed by community-based DOT.
• GIS was used to manage the DOT program

60
(No Transcript)
61

• GPS and aerial photos to map homesteads in the
  rural area
• Similarly mapped DOT sites and potential DOT
  sites such as clinics, community health worker
  homes, churches, shops, and local hospitals.
• Mapped DOT utilization and distance to nearest
  DOT over time 1991 to 1996.

62
Findings

• Despite increase in caseload, hospital treatment
  decreased with the increase in community DOT
  sites
• Average distance to DOT sites decreased in half
  over the study period