Transcript SOCIAL EPIDEMIOLOGY OF HIV: A MEASUREMENT CHALLENGE

SOCIAL EPIDEMIOLOGY OF HIV:
A MEASUREMENT CHALLENGE
FOURTH INTERNATIONAL CONFERENCE ON
“ECOLOGY. RADIATION. HEALTH”,
SEMEY STATE MEDICAL ACADEMY,
KAZAKHSTAN, 28 SEPT. 2007
IRINA CAMPBELL, PhD, MPH
US DEPT. OF STATE FULBRIGHT SCHOLAR IN
GLOBAL HEALTH
[email protected]
www.CampbellHealthAssociates.com
A TRULY GLOBAL PROBLEM
REQUIRING GLOBAL COOPERATION,
AWARENESS, AND ASSISTANCE
The Silk Road of Drugs, Migration, HIV
EVIDENCE-BASED HIV PREVENTION

GREATER ACCURACY & PRECISION IN DESCRIBING ROUTES OF
TRANSMISSION OF HIV AMONG MOST-AT-RISK GROUPS
RATIONALIZES PREVENTION PROGRAMS

THIS PRESENTATION WILL TOUCH ONLY ON A BASIC ISSUE IN ESTIMATING
HIV PREVALENCE IN KAZAKHSTAN

ACCURACY OF ESTIMATES IMPACTS ON DESIGN AND TARGETING OF
EFFECTIVE PROGRAMS

IN 1994, CDC, USA CENTERS FOR DISEASE CONTROL & PREVENTION,
BEGAN RECOMMENDING THAT HIV PREVENTION PLANNING GROUPS
APPLY THE PRINCIPLES OF EPIDEMIOLOGY, EVALUATION & BEHAVIORAL
SCIENCE THEORIES TO DESIGN PREVENTION PROGRAMS IN ORDER TO
GET GRANT FUNDING

SCIENTIFIC METHODOLOGIES WHICH ARE MOST RELEVANT TO
DEFINING & SOLVING THE HIV EPIDEMIC ARE  EPIDEMIOLOGY & SOCIAL RESEARCH METHODS,
 BASIC BEHAVIORAL SCIENCE & CHANGE THEORY,
 EVIDENCE-BASED INTERVENTIONS & EVALUATION
METHODS.
SOCIAL EPIDEMIOLOGY MODELS
BRIEFLY,

EPIDEMIOLOGY IS THE STUDY OF POPULATION HEALTH
THE OCCURRENCE, DISTRIBUTION, NATURAL HISTORY, SOCIAL ETIOLOGY
& CAUSAL PATHWAYS OF DISEASE IN A POPULATION WITH
MICRO + MACRO MODELS

BIOMEDICINE IS THE STUDY OF INDIVIDUAL HEALTH IN THE CLINICAL
CONTEXT WITH
MICRO MODELS

SOCIAL EPIDEMIOLOGY ENCOMPASSES A MULTIDISCIPLINARY,
INTERDISCIPLINARY PARADIGM WHICH OVERLAPS
ENVIRONMENTAL EPIDEMIOLOGY, ECOLOGY, SMALL AREA
ANALYSIS, CHRONIC DISEASE EPIDEMIOLOGY, GEOGRAPHY, &

SOCIOLOGICAL CONCEPTS, SUCH AS SOCIAL NETWORKING,
SOCIAL COHESION, SOCIAL CAPITAL, & SOCIAL SUPPORT, TO
ESTIMATE & PREDICT DISEASE PREVALENCE
SOCIAL EPIDEMIOLOGY MODELS
ESTIMATE INCIDENCE, NEW INFECTIONS OF HIV
ESTIMATES PREVALENCE, TOTAL INFECTIONS OF HIV (WHAT)
ESTIMATE DISTRIBUTIONS ACROSS PLACES (WHERE) AND
GROUPS (WHO) - ECOLOGICAL FACTORS
ESTIMATE DISTRIBUTION OF STRUCTURAL (MACRO) &
BEHAVIORAL (MICRO) RISK FACTORS DETERMINING
INCIDENCE & PREVALENCE RATES (WHY)
(see FIGURE 1)
HEALTHY LIFESTYLES MOVEMENT IN PREVENTIVE MEDICINE &
PUBLIC HEALTH IS A RESULT OF THE SCIENTIFIC WORK OF
SOCIAL EPIDEMIOLOGISTS
FIGURE 1: MACRO & MICRO PROPOSITIONS OF GEOGRAPHIC
VARIATION IN HEALTH
Macro proposition:
Micro proposition:
geographic variation due to
Contextual/social causation
hypothesis:
geographic variation due to
Compositional/individual selection
hypothesis:
spacial variation in exposure to
environmental/structural factors:
spacial variation in direct selection:
at-risk people moving/staying in area:
poverty; pollution, traffic, housing;
quality, crime, recreational
resources, sanitation,
access to material or social
resources
poor people living in rundown areas;
downward SES drift/mobility of sick
concentration of sick around
facilities;
concentration of healthy around
parks, or “younger” areas
spacial variation in exposure to
behavioral factors:
spacial variation in indirect selection:
at-risk people with certain traits
moving/staying in area –
drug/alcohol abuse, stress
passive smoking, unsafe driving
community group activities
religious group membership
large, younger, low-income families
blue collar manual workers
older persons w/ low educational
level
ATOMISTIC & ECOLOGICAL FALLACIES VS.
MULTILEVEL MODELS

ATOMISTIC FALLACY – ATTRIBUTING TRAITS OF AN INDIVIDUAL TO A
POPULATION
(HI SES PERSONS LIVING IN SEMEY HAVE HIGHER THAN AVERAGE
LIFE EXPECTANCY & LOW CANCER RATE DOES NOT MEAN SEMEY IS A
WEALTHY HEALTHY CITY - MICRO TO MACRO GENERALIZATION)

ECOLOGICAL FALLACY – ATTRIBUTING TRAITS OF A GROUP/ POPULATION
TO INDIVIDUALS
(HI SES AREA DOES NOT MEAN PERSONS WITHIN AREA ARE WEALTHY
- MACRO TO MICRO GENERALIZATION)

MULTILEVEL MODELS – i.e., SEPARATE ATTRIBUTION OF FACTORS
MEASURED AT SPECIFIC LEVELS, SUCH AS MACRO STRUCTURAL
POPULATION AND MICRO INDIVIDUAL FACTORS, FOR INDIVIDUAL HEALTH
STATUS OUTCOMES
MULTILEVEL MODEL, i.e., can explain simultaneous effect of
both personal SES + place SES on health
Total variance of Yij = sum of between-group vars+ within-group var
Yij = 00 + p0Xpij + 0qZqj + pqZqjXpij + u1jXpij + u0j + eij
where:
p is the number of explanatory variables X at level L1 (individuals),
q is the number of explanatory variables Z at level L2 (urban areas), and
ij is individual level L1 observation i in level L2 (urban areas) j ;
combining terms produces the following general hierarchical linear
equation which separates the fixed and random elements: Yij=
[ 00 + p0Xpij + 0qZqj + pqZqjXpij ]+[ u1jXpij + u0j + eij ]
Fixed part of equation invariate between macro areas
OLS variation at micro level
Random part of equation residual variance between
areas after controlling micro
fixed variables

and where:

Zqj is the cross-level interaction = value of Y-X slope at level L1
(individuals) with Z at level L2 (urban areas);

eij is the between individuals, random residual, mutually independent,
mean=0, homoscedastic, normally distributed, constant across macro units,
random effect = unexplained variability of dependent variable at micro level;

u0j is a between macro unit random residual, mutually independent,
mean=0, homoscedastic, normally distributed, random effect of intercept =
unexplained (by micro level intercept) variability of dependent variable at
macro level;

u1jXpij is the random interaction between macro unit and X; u1j is a
between macro unit and micro unit random residual, independent from the
individual level residuals but correlated to the macro level residuals, random
effect of slopes = unexplained (by micro level slopes) variability of dependent
variable at macro level.

The basic difference between the ordinary least squares regression model
(OLS) and the hierarchical linear model is the complex random residual term,
[ u1jXpij + u0j + eij ]. The contextual effects or unexplained variance of the
outcome due to macro units as estimated by the random residuals, u0j and
u1j , are assumed to be independent between macro units but correlated
within macro units; independent of the micro level residuals; with population
mean = 0, a multivariate normal distribution, and constant covariance
WHAT RELEVANCE DOES THE MULTILEVEL
EPIDEMIOLOGY MODEL HAVE FOR HIV
EPIDEMIOLOGY?
INCLUDE STRUCTURAL FACTORS (i.e.,
SOCIAL NETWORKS, PLACE) AS
PREDICTORS + INDIVIDUAL RISK
FACTORS (IDU, MSM, CSW)
GLOBAL STAGING OF HIV ACROSS
CENTRAL ASIAN REGIONS

WORLD BANK MODELS OF STAGING HIV EPIDEMIC
1-UNKNOWN
2-NASCENT Epidemic Stage 1: 1987- Dominant
- Sexual
transmission
3-CONCENTRATED Epidemic Stage 2: 1991- Concentrated
Dominant transmission – IntraVenous Drug Use
4-GENERALIZED Epidemic Stage 3: 2005 - Generalized
Dominant transmission: >Sex+IVDU

H0: STAGE 5 - GENERATIONAL Epidemic Stage 4: 2006 Generational Dominant transmission: Adolescents & Children
– parental-father-mother to child transmission
–
Young People lifestyle behaviors
STAGING OF HIV IN ECA REGION
PREVALENCE OF HIV/ OBLAST, KAZAKHSTAN, 2006
national average = 11.4/ 100,000 persons
IDU & HIV in Kazakhstan
MAJOR TRANSMISSION ROUTES

IDU MAJOR ROUTE OF TRANSMISSION OF HIV - MOSTAT-RISK AND MOST-HARD-TO-FIND GROUPS

THUS DETERMINING SIZE/ LOCATION/ DEMOGRAPHIC
COMPOSITION OF IDU POPULATION FOCUSES
PREVENTION INTERVENTIONS AT THE POINT OF
GREATEST TRANSMISSION TO CONTAIN EPIDEMIC

NEED > ACCURATE METHODS TO ESTIMATE & LOCATE
THIS MOST-AT-RISK GROUP
KAZAKHSTAN HIV EPIDEMIC TRANSITIONING FROM
3-CONCENTRATED Dominant transmission – IDU
AND
4-GENERALIZED Dominant transmission: >Sex + IDU
TO
GENERATIONAL Increasing transmission:
– parent-father-mother to child transmission
– Young People lifestyle behaviors
HIV PREVALENCE IN PREGNANT WOMEN, SCREENING RESULTS, KAZAKHSTAN,
2006 (% HIV among screened)
HIV PREVALENCE AMONG IDU, HEALTH SCREENING RESULTS, KAZAKHSTAN,
2006 (% HIV among screened)
HIV PREVALENCE AMONG PRISON POPULATION, HEALTH SCREENING
RESULTS, KAZAKHSTAN, 2006 (% HIV among screened)
MOST-AT-RISK GROUPS FOR HIV ALSO
MOST-HARD-TO-FIND, ESTIMATES VARY BY
METHOD

STD – SEXUALLY TRANSMITTED DISEASE CASES

IDU – INJECTION DRUG USERS

CSW – COMMERCIAL SEX WORKERS

MSM – MEN HAVING SEX WITH MEN

HOMELESS YOUTH – ORPHANS, RUNAWAYS,
ABANDONED

YOUNG PEOPLE – POPULATION AGE 10-24 YRS (WHO)
HIV PREVALENCE /100,000 POP, KAZAKHSTAN, 1987 – 2006,
KAZAKHSTAN REPUBLICAN CENTER FOR THE PREVENTION OF HIV
NUMBER OF PERSONS WITH HIV+ (lt. blue), AIDS+ (dark blue), AND DEATHS
(red), KAZAKHSTAN, 2004 – 2006, REPUBLICAN CENTER FOR THE PREVENTION
OF AIDS
HIV PREVALENCE IN SYPHILIS + (dark blue) & SYPHILIS – (lt. blue) PERSONS
UNDER SURVEILLANCE, KAZAKHSTAN, 2006 (IDU, CSW, PRISONERS, STD+,
PREGNANT WOMEN, from left to right)
ESTIMATES OF % IDU AMONG HEPATITIS C SURVEILLANCE GROUPS (CSW n=2105,
PRISONERS n=4487, STD n=4836) BY HEPATITIS C PREVALENCE (blue), IDU AMONG
HEPATITIS C (orange), IDENTIFIED SELF AS IDU IN SURVEY (green), KAZAKHSTAN, 2006.
HIV PREVALENCE AMONG COMMERCIAL SEX WORKERS (CSW), KAZAKHSTAN,
2006 (% of CSW in Oblast/ Region, National Average = 2.5%)
PREVALENCE OF SYPHILIS BY OBLAST/ REGION, KAZAKHSTAN, 2006 (% of
Syphilis in Oblast/ Region, National Average = 26%)
HIV PREVALENCE AMONG CSW WITH SYPHILIS + AND/ OR HEPATITIS C+,
KAZAKHSTAN, 2006 (HPT C+/Syphilis+; HPT C+/Syphilis-; HPT C-/Syphilis+; HPT
C-/Syphilis-; from left to right)
SYPHILIS PREVALENCE AMONG IDU, KAZAKHSTAN, 2006 (n=4553, National
Average=11%)
HIV PREVALENCE AMONG IDU, KAZAKHSTAN, 2006 (n=4553, National
Average=3.4%)
Number IDU REPORTING CASUAL & CSW SEXUAL CONTACT BY OBLAST/
REGION, DURING PAST 6 MONTHS, KAZAKHSTAN, 2006 (total n=4553, National
Average=47%)
Number IDU IDENTIFIED WITH VOLUNTARY HIV TESTING BY OBLASST/ REGION,
KAZAKHSTAN, 2006 (total n=4553, National Average=47%)
NUMBER OF PERSONS SURVEYED FOR HIV, KAZAKHSTAN, 2004-2006
INCIDENCE OF HIV, KAZAKHSTAN, 2004-2006
CHANGES IN HIV EPIDEMIOLOGY DUE TO INCREASED SCREENING OF POPULATION FOR
HIV OR CHANGES IN EPIDEMIOLOGICAL FACTORS, KAZAKHSTAN 2004-2006 (orange=n
cases based on changing factors; teal=n cases due to increased screening, 0 cases
screened 2004 vs. 311 cases screened 2006)
ANNUAL REGISTRATION OF NEW HIV CASES, KAZAKHSTAN, 1987-2006

> N CASES HIV BETWEEN 2004 - 2006 DUE TO > EPIDEMIC, NOT TO
BETTER SCREENING OR TESTING

HIV AMONG IDU INCREASED FROM
2003
3,8% TO
2006
5.8%
UNEVEN DISTRIBUTION AMONG OBLASTS

MOST REPUBLIC OF KAZAHSTAN AIDS PREVENTION CENTER DATA
DERIVED FROM CDC SPONSORED SNOWBALL SAMPLING VARIANT,
RESPONDENT DEVELOPED SAMPLE (RDS)

SNOWBALL SAMPLING = NONRANDOM SELECTION,
NONREPRESENTATIVE, SAMPLE OF CONVENIENCE
– NEED SAMPLING AMONG RISK GROUPS TO > EFFICIENCY BUT
PROBLEMS WITH GENERALIZATION FROM
NONREPRESENTATIVE SAMPLE, THEREFOR RDS SAMPLING
RESPONDENT DRIVEN SAMPLING (RDS), NULL WAVE, IDU CASE #1 &
IDU CASE #2, EACH ASKED FOR 3 REFERRALS
RESPONDENT DRIVEN SAMPLING (RDS)
WAVE 2 CASES IDU #3 - #8
RESPONDENT DRIVEN SAMPLING (RDS) WAVE 3, IDU CASES # 9-16;
WAVE 4, IDU CASES #17-30; WAVE 5, IDU CASES #31-45
NETWORK OF RECRUITED IDU CASES FROM IDU CASE #1, YANGIUL,
2004
NETWORK OF 400 IDU CASES RECRUITED IN YANGIUL, 2004
COMPARATIVE METHODOLOGICAL ASSESSMENT
OF DRUG USE IN KAZAKHSTAN
RESEARCH STUDY BY MINISTRY OF HEALTH,
REPUBLIC OF KAZAKHSTAN APPLIED RESEARCH CENTER
FOR MEDICOSOCIAL PROBLEMS IN NARCOTICS,
NATIONAL CENTER FOR THE PREVENTION OF HEALTHY
LIFESTYLE DEVELOPMENT (NCPHLD), REPUBLIC OF
KAZAKHSTAN CENTER FOR PSYCHIATRY,
REPUBLIC OF KAZAKHSTAN CENTER FOR PREVENTION OF
AIDS, 2004
METHODOLOGICAL INNOVATIONS TO LOCATE
MOST-AT-RISK GROUPS

COLLABORATIVE STUDY WITH ASSISTANCE FROM MINISTRY OF
INTERIOR, JUSTICE, POLICE DEPT., CDC, UNICEF, UNAIDS

RESEARCH FRAMEWORK – ALL OBLASTS OF KAZAKHSTAN

COMPARED TO EXISTING SOCIOLOGICAL STUDIES OF HIV
PREVALENCE

INVESTIGATION LOCATED 201,045 DRUG USERS IN KAZAKHSTAN
IN 2004

METHOD USED = UN EXPRESS-EVALUATION/ MONITORING FOR
DRUG ABUSERS, ADAPTED TO KAZAKHSTAN BY RK CENTER FOR
PREVENTION OF AIDS
– 4 PARTS TO METHOD –
 1 – BASED ON EXISTING OFFICIAL STATISTICS
 2 – METHOD OF MULTIPLICATION
 3 – METHOD OF NOMINATION
 4 – METHOD OF TRADITIONAL SOCIOLOGICAL INVESTIGATIONS IN
MEDICINE
METHOD 1 TO LOCATE MOST-AT-RISK GROUPS

STUDY FRAME = 14 OBLASTS + ASTANA CITY, ALMATY CITY,
ARKALYK, BALKHASH, ZHEZKAZGAN, SEMIPALATINSK, TEMIRTAU,
EKIBASTUZ

DATA COLLECTION INSTRUMENT = SURVEY QUESTIONNAIRE

SAMPLING FRAME =
LIST 1 - DRUG USERS REGISTERED IN NARCOLOGICAL CLINICS
LIST 2 - DRUG USERS REGISTERED
BY POLICE
N DRUG USERS
LIST 1
LIST 2
GROUP A
DRUG CLINIC
REGISTRY
+
POLICE
REGISTRY
+
GROUP B
__
+
GROUP C
+
__
need to find
GROUP X, not
screened by list
1 or list 2
__
__
METHOD 1 TO ESTIMATE MOST-AT-RISK FOR HIV
LIST 1 + LIST 2 +
LIST 1 -- LIST 2 +
GROUP a
GROUP b
LIST 1 + LIST 2 –
LIST 1 - LIST 2 –
GROUP c
GROUP x
ax = bc
x = bc/a
X = UNKNOWN POTENTIAL HIV / IDU CASES
TOTAL IDU N(1) = a + b + c + x
METHOD 2 (p) TO ESTIMATE MOST-AT-RISK
GROUPS

SURVEYS OF RISK GROUPS ESTIMATED % OF IDU LOCATED BY
SURVEY WHO ARE REGISTERED – CLINICS

CALCULATE MULTIPLICATIVE FACTOR p OF IDU NOT
REGISTERED IN CLINICS

MULTIPLY EXISTING OFFICIAL LIST 1 OF CLINIC REGISTRY BY p

TOTAL IDU N(2) = N p
METHOD 3 (k) – SOCIAL NETWORK THEORY TO
ESTIMATE MOST-AT-RISK GROUPS

DURING SURVEY - RESPONDENTS ASKED TO LIST
FRIENDS WHO ARE IDU

CALCULATE NOMINATIVE FACTOR k OF IDU NOT
LISTED IN CLINIC REGISTRY

TOTAL IDU N(3) = N k
AVERAGE ESTIMATE m OF IDU
COEFFICIENT
m = ∑ k , p / 2 = IDU
METHOD 4 – TRADITIONAL SURVEY RESEARCH
TO ESTIMATE MOST-AT-RISK GROUPS

2004 QUESTIONNAIRES, DESIGNED BY UNICEF/ WHO,
FOCUSED ON KNOWN RISK GROUPS – IDU, CSW, MSM,
YOUTH – TOTAL N SURVEYED = 15,863

YOUTH SAMPLING FRAME – PROBABILITY
NONREPEATING SELECTION OF 10 SCHOOLS, AGED 1114 / 15-17 YRS (200 OF EACH GENDER), TOTAL N = 7200

DESIGN OF SAMPLING FRAME OF OTHER RISK GROUPS
WAS NOT EXPLICITY DESCRIBED IN THIS
METHODOLOGICAL REPORT
PREVALENCE DERIVED FROM OFFICIAL DATA &
ESTIMATES

1.01.2004 TOTAL N IDU OFFICIALLY REGISTERED IN KZ = 46,940
316/100,000 POP, DIAGNOSED WITH DRUG ABUSE
2004 TOTAL IDU BY 4 METHODS

OFFICIAL CLINIC REGISTRY
N = 46,340

METHOD 1 (N=a+b+c+x)
N = 175,024

COEFFICIENT M
N = 227,066

AVERAGE OF OFFICIAL CLINIC REGISTRY DATA + METHOD 1 +
COEFFICIENT M
N=201,045
NARCOTICS USE AMONG KAZAKHSTAN YOUTH

2004 STUDY FOUND THAT YOUTH 11-17 YRS OLD IN 9 LARGE KZ
CITIES N=13,158 HAD USED NARCOTICS RECREATIONALLY AT
LEAST ONCE (WHICH CAN QUICKLY CHANGE TO ADDICTION)

THIS AMOUNT IS MANY TIMES LARGER IN JUVENILE DETENTION
HOMES & ORPHANAGES (10% - 24%) THAN IN THE GENERAL POP
OF YOUTH (2.2% - 4.6%)

NARCOTICS ARE MAJOR CAUSE FOR INITIATION INTO
ADOLESCENT SEXUAL ACTIVITY
OFFICIAL REGISTRY DATA FOR YOUTH ARE INACCURATE
UNDERESTIMATES, AS FOLLOWS :
DRUG REGISTRY
N CHILDREN = 53; N ADOLESCENTS = 823
EPISODIC USE REGISTRY N CHILDREN = 312; N ADOLESCENTS = 1342
2004 METHODS STUDY
TOTAL N = 13,158
HIV PREVENTION THROUGH
Strategic information, including
monitoring & evaluation,
surveillance &
management information systems
http://www.globalhivevaluation.org/toolbox.aspx
ADDITIONAL RESOURCES
Title: Monitoring & Evaluation Capacity Building for
Program Improvement - Training Presentations
Agency: Centers for Disease Control and
Prevention/Global AIDS Program (CDC/GAP)