International AIDS relief stagnated in 2009
BMJ 2010; 341 doi: https://doi.org/10.1136/bmj.c3942 (Published 20 July 2010) Cite this as: BMJ 2010;341:c3942All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
World Health Organisation HIV prevalence estimations in resource-poor
settings have recently been adjusted downwards [1]. This was mainly based
on the introduction of national population-based HIV surveys that were
considered more representative than previous antenatal clinic surveillance
data. However, household surveys pose other problems, such as population
mobility and non-response bias. In this communication an additional
limitation is put forward that seems widely ignored: bias due to
behaviour of the surveyor. Because household surveys are typically
performed on large numbers of respondents by small numbers of surveyors,
this could have disproportional influence on final survey results. It is
unknown which proportion of WHO-interpreted databases suffers from such
surveyor-bias.
In 2007 a population-based representative household survey was
performed in a Sub-Saharan country. The survey randomly assigned a team of
eight trained nurses to perform a medical interview, measure blood
pressure and collect oral fluid samples for an anonymous non-invasive HIV
test on individuals aged 12 years and above. Consent to perform the HIV
test was obtained from (parents of) 2,452 individuals. Samples were
collected with the OraSure collection device (OraSure Technologies Inc.,
Bethlehem, PA) and tested with standard HIV ELISA. The survey followed
practices that are common to DHS+ surveys. Ethical clearance was obtained
from the Ministry of Health and Social Services. The overall HIV
prevalence as determined on the 2,452 individuals was 12.7%.
During the data cleaning it became apparent that HIV prevalence among
respondents sampled by one particular nurse was 3.6 fold higher than
sampled by the other nurses (X2, p<0.0001). The Figure shows weekly
HIV+ percentages stratified by surveyor-nurse. As can be seen, the HIV
estimates of samples obtained by one particular nurse ("H") increased to
above 80% during the second phase of the survey.
No significant systematic differences were found in type of
respondents visited by nurse "H" as compared to the other nurses in terms
of age, sex, education, income, marital status, or household demographics.
Furthermore, half of the geographical area assigned to nurse "H" was
identical to the one assessed by nurse "G" and the remaining area was
assessed jointly with nurses "A" and "B", none of whom demonstrated
similar HIV results. Additional data analysis revealed that hypertension
prevalence rates assessed by this nurse were 2.9 fold higher compared to
the other nurses (X2, p<0.0001). In addition, interview answers
provided by nurse "H" significantly selected for easy answers which do not
require further detailing (X2, p<0,001).
These findings suggest that the suspect nurse had failed to collect data correctly. When all 313
respondents sampled by nurse "H" were excluded from the overall survey
data, this resulted in a drop in the HIV prevalence estimate from 12.7%
(95% CI: 11.4 - 14.0) to 9.6% (95% CI: 7.3 - 11.8). Thus, errors in data collection by a
single interviewer can lead to flawed population HIV prevalence estimates.
The potential magnitude of this limitation of household surveys
depends on the number of surveyors; the actual HIV prevalence in the
country; the percentage of HIV-tests that a surveyor misreports and
whether the incorrect test-results are positive or negative. If 1 nurse in
a team of 10 would produce 50% false positive results, the estimated HIV
prevalence would be biased upwards 1.2-fold, 1.5-fold or 2.6-fold in a
15%, 6% or 2% HIV prevalence setting respectively.
Therefore, in this note a retrospective analysis of representative samples
of pertinent DHS/UNAIDS/WHO data is recommended to avoid that global needs
assessments and funding decisions are based on flawed HIV prevalence rates
(see BMJ 2010;341:c3942) [2].
[1]
http://www.who.int/mediacentre/news/releases/2007/pr61/en/index.html
[2] A longer version of this note is posted on www.pharmaccess.org and www.aiid.org
Competing interests: No competing interests
Re:A cautious note on household survey HIV prevalence estimates in resource-poor settings
Use of Household Surveys for HIV Prevalence Estimates
In their letter to the editor in response to Roehr1, Janssens et al.2
describe a 2007 household survey in an unnamed country in sub-Saharan
Africa in which one out of eight nurses who were collecting samples for
HIV testing apparently began to falsify data after six weeks in the field.
The longer version of their note on the website of the lead author's
institute3, indicates that the suspect nurse apparently started collecting
samples from somewhere other than the assigned households shortly after
another nurse was dismissed from the team for visiting too few households.
The longer version further indicates that the samples for HIV testing were
collected in Windhoek, Namibia, with the OraSure device for oral mucosal
transudates, not blood samples that are commonly used in large-scale
household surveys.
The authors claim that the above survey followed standard practices
common to Demographic and Health Surveys (DHS), including HIV testing.
Nothing could be further from the truth. DHS surveys that include HIV
testing collect dried blood spots on filter paper cards for laboratory
testing, not oral fluid. In DHS surveys, the persons collecting blood are
closely supervised on a daily basis by both the team supervisor and the
field editor on every team, with additional inspections conducted by
visiting monitors from the head office of the implementing agency and
biomarker specialists and survey experts from the Demographic and Health
Surveys programme at ICF Macro. DHS surveys use rigorous hiring, training,
logistics, evaluation, analysis and laboratory testing procedures. These
procedures make it highly unlikely that any member of the field staff
could commit fraud over a long period of time without being detected. In
addition, response rates and survey data in DHS surveys are carefully
scrutinized for potential bias before the results are released.
Most standard DHS surveys conduct HIV tests on at least four times as
many persons as the Windhoek survey, using at least four times as many
nurses/interviewers who collect blood for HIV testing. Nevertheless,
Janssens et al. question WHO's downward adjustment of global HIV
prevalence estimates in 2007 and suggest that HIV prevalence estimates
based on population surveys are likely to be flawed. One case of fraud
found in one local survey that apparently did not have the necessary
protections in place to prevent and detect such practices certainly cannot
be used to conclude that global estimates of HIV prevalence based largely
on HIV tests in high quality, nationally representative household surveys
are suspect.
The country with the largest decrease in the number of persons living
with HIV (from 5.2 million to 2.5 million) according to UNAIDS estimates
is India. The new estimate for India was based largely on HIV tests
conducted on more than 100,000 women and men in the 2005-06 National
Family Health Survey (NFHS-3), which is part of the DHS programme. A panel
of national and international experts closely scrutinized the NFHS data,
as well as information available from the HIV surveillance system and
surveys of high risk groups, before reaching a consensus on the level of
HIV prevalence in India. The blood samples for HIV testing were collected
by more than 500 health investigators attached to the interviewing teams.
Even in the unlikely event that a couple of these investigators had for
some reason provided some fraudulent blood samples that were not detected,
the effect on the overall HIV prevalence estimate for the country would
have been almost negligible.
1 BMJ 341:doi:10.1136/bmj.c3942
2 BMJ 2010; 341:c6323.
3 http://www.aiid.org/docs/AIID_RS_10-08_Cautious_note_HIV_estimates.pdf
Competing interests: No competing interests