Health worker absence, HIV testing and behavioral
change
Markus Goldstein (World Bank)Joshua Graff Zivin (UCSD)
James Habyarimana (Georgetown)Kiki Pop-Eleches (Columbia)
Harsha Thirumurthy (UNC-Chapel Hill)
Information and behavior
• Information assumed to shape choices and behavior
• This has implications for effects of HIV testing Provides information about future health and
longevity
• Setting for this study Antenatal clinic in Kenya where testing is offered
to pregnant women for prevention of MTCT
• Aim: study take-up of HIV testing and impact of learning HIV status on behavioral outcomes
Two important aspects of HIV testing• Take-up of HIV testing
Supply and demand side factors influence this, but role of each not well understood
• Should testing be an important component of policy response to HIV? Discussion has largely focused on its impact
on sexual behavior (Coates et al. 2000; Thornton 2008)
Impacts on other outcomes largely neglected• Take-up of other valuable health services• forward-looking behavior
Obstacles to scale-up of HIV testing• Supply and demand side factors
Demand side factors (see Thornton 2008 & others)
• This paper focuses on supply side Structural constraints (health worker absence)
• Health worker absence in developing world 35% absence rate among public health
providers (Chaudhury et al. 2005) Limited evidence re: impact on health outcomes
• Also a valid instrument for testing decision
Effects of HIV testing (1): health outcomes• Main reason for testing pregnant women:
provide PMTCT medication & advice In 2005, 11% of HIV+ women in Africa got PMTCT
• Inexpensive & effective meds available for PMTCT Nevirapine ($0.50 per dose) In this setting, ARV therapy also given for PMTCT
• Other possible benefits of PMTCT counseling Healthier mothers and children through safer
delivery and increased take-up of neonatal care
Effects of HIV testing (2): socio-economic behavior• Information about future health &
expected longevity should affect number of inter-temporal investment decisions at household level Fertility – important assumption in macro
models of impact of AIDS epidemic (Young 2005)
Human capital formation Asset accumulation
ANC TEST
Individual
Structural
Inputs
HEALTH OUTCOME
S
BREAST FEED
BIRTH LOCATIO
N
PMTCT
TESTING AND COUNSELLING: HEALTH OUTCOMES
ANC ECONOMIC
OUTCOMES
FERTILITY
SCHOOLING
INVESTMENT
TESTING AND COUNSELLING: ECONOMIC OUTCOMES
TEST
Individual
Structural
Inputs
Summary of results
• Absence rate of 10% of PMTCT nurse Large effect on uptake of testing and counseling
• Large effect on delivery of PMTCT medications Safer delivery Lower likelihood of breastfeeding
• Change in investment behavior among negatives Asset accumulation Schooling
• No effects on fertility preferences
Outline
• Panel survey of pregnant women in Kenya• HIV testing decisions and nurse absence
Survey in Nyanza Province
Kenya adult prevalence 6.7% (1.2 million)
Nyanza Province adult prevalence of 20%
Rural health center provides ANC care, and has HIV clinic that provides ARVs (managed by AMPATH program)
Survey conducted in two waves
Panel survey of pregnant women• Wave 1: In-clinic interview before HIV test (July ‘05 –
Feb ‘06) Only first time visitors for current pregnancy interviewed Short questionnaire, included subjective beliefs about HIV
status 650 women from catchment area enrolled
• HIV testing offered after wave 1 interview
• Wave 2: Household interview (May ‘06 – Feb ‘07) comprehensive socio-economic data collected at home:
• demographics, education, health, employment, sexual behavior, assets, etc
• Interviewed ANC client and spouse• Completed panel on 591 women (9% attrition)• Loss to follow up generally due to relocation out of province
Additional data obtained from clinic• PMTCT logbook
HIV status: continuously updated because pregnant women could have tested on subsequent antenatal visits
Nurse presence/absence: based on # of women tested
• AMPATH records Fraction of HIV+ women who enrolled
(imperfect matching)
PMTCT in western Kenya
• Pregnant women typically get tested at first ANC visit 3 visits recommended
• Most common med for HIV+ women: Nevirapine Given to the mother with onset of labor and
drops given to the baby within 72 hours of birth Reduces the risk of transmission by about 50%
• In our study setting, ART also given for PMTCT
• Breastfeeding generally not recommended
HIV testing in sample of 591 women
61.9%15.2%
22.8%
Negative PositiveNot tested
Source: Chulaimbo ANC Intake, 2006
HIV status of women
Health worker absence & HIV testing
• About 77% of women in panel data tested for HIV 25% of those who tested were HIV+
• Controlling for selection into testing 10% absence rate for PMTCT nurse (relatively
small)
• Useful as an instrumental variable to deal with selection Effect of absence on testing is first stage Also control for day of the week and prior beliefs
Table 2a: Nurse absence and testing
Table 2b (1) (2)Age in years 0.001 0.001
(0.002) (0.002)Completed primary school -0.020 -0.017
(0.026) (0.026)Married 0.021 0.027
(0.039) (0.038)# of church attendances (past 4 wks) 0.011 0.012
(0.004)* (0.004)**Number of sexual partners (past 6 mths) 0.007 0.010
(0.035) (0.034)Boils drinking water 0.035 0.038
(0.026) (0.025)HIV subjective beliefs Moderate chance 0.012 0.010
(0.039) (0.040) Small chance -0.029 -0.032
(0.037) (0.038) No chance at all -0.071 -0.078
(0.049) (0.049)Livestock ownership 0.003 0.003
(0.002) (0.002)*House has non-grass roof -0.033 -0.032
(0.028) (0.028)Household resides in clinic catchment -0.001 -0.001
(0.028) (0.028)Day of week = Tuesday -0.013
(0.029)Day of week = Wednesday -0.028
(0.030)Day of week = Thursday -0.085
(0.033)**Day of week = Friday -0.193
(0.046)**Constant 0.865 0.909
(0.081)** (0.080)**Observations 577 574R-squared 0.03 0.08
Nurse present at time of woman's first visit
HIV testing and behavior change• Instrument for testing offers opportunity
to examine whether behavior changes after learning HIV status We estimate separate effects for HIV- and
HIV+ women
• Comparison group? We compare to women who do not get tested
but have similar pre-test beliefs about own status
IV strategy for estimating impact of testing (by test
result)• Ideally:
Do not know status for non-testers
• Therefore, we assume that
Control for pre-test subjective beliefs Our assumption (non-testers’ behavior shaped
by beliefs)
Subjective beliefs about HIV status• First, are pre-test subjective beliefs good
proxy for HIV status? i.e. do pre-test beliefs predict actual test
result?
• Second, do beliefs change after learning HIV status? Perhaps a prerequisite for behavioral change
to occur We examine beliefs about own status and
partner’s status
Variation in pre-test subjective beliefs
12.1%
18.5%
51.8%
17.7%
Great ModerateSmall None at all
Source: Chulaimbo ANC Intake, 2006
Distribution of priors
Pre-test beliefs do have predictive power
10.7%
18.9%
52.6%
17.8% 21.1%
25.6%44.4%
8.9%
9.7%
12.7%
54.5%
23.1%12.1%
18.5%
51.8%
17.7%
Negative Positive
Not tested Total
Great Moderate
Small None at all
Source: Chulaimbo ANC Intake, 2006
Distribution of priors
Pre-test beliefs & actual test result
Tested positive
Tested positive
Tested positive
(1) (2) (3)
Chance of having HIV- great 0.272 0.267 0.226(0.097)** (0.098)** (0.098)*
Chance of having HIV- moderate 0.171 0.168 0.126(0.081)* (0.082)* (0.080)
Chance of having HIV- small 0.077 0.079 0.055(0.059) (0.059) (0.059)
Day of week controls? N Y YOther controls N N YMean of dep. variable 0.197 0.197 0.197Sample Size 453 452 452
Testing and changes in subjective beliefs
HIV testing and behavior change• Evidence of changes in subjective beliefs
about one’s status provide motivation for other behavioral responses
• We first estimate average effect across all women who learn they are HIV+ and HIV- (not interacted with pre-test beliefs)
Why pre-test beliefs may matter (Boozer & Philipson 2000)
Prior belief (subj. belief)
Costs and Benefits of testing
High-risk
Low-risk
Cost of testing
Benefit of testing
Literature has focused largely on the effect of HIV testing on sexual behavior• Policy rationale
Those who test HIV- may have incentives to avoid infection
Those who test HIV+ can be encouraged to adopt safe sex practices
• Effects are theoretically ambiguous In both cases, the opposite response possible
• Existing studies of testing and sexual behavior Coates el al. 2000: VCT, Kenya and Tanzania Thornton 2005: community-based VCT, Malawi
Impacts on other outcomes also important for evaluation of HIV testing• PMTCT take-up, health outcomes• Socio-economic behavior, particularly
forward-looking decisions
Effects on PMTCT and health outcomes
Socio-economic behavior
Heterogeneous response by level of surprise and updating• As in Boozer and Philipson (JHR 2000)• Do women who learn more from the HIV
test have larger changes in behavior? Not much evidence that this matters Actual test result is more salient than how
surprised one is by the test result
Summary of results
• Absence rates are moderate but have large effects on PMTCT outcomes
• Pre-test beliefs do predict HIV status, and these beliefs evolve over time
• Increases in investment behavior for women who test HIV-negative
• No effect on fertility