What Lies Behind the Fall in the HIV Population in India?
Arvind Pandey, D C S Reddy, M Thomas|
Arvind Pandey (firstname.lastname@example.org) and M Thomas (email@example.com) are at the National Institute of Medical Statistics, ICMR, New Delhi. D C S Reddy (firstname.lastname@example.org) is with the World Health Organisation, New Delhi.
Commentary DECEMBER 27, 2008 EPW Economic & Political Weekly
The availability of multiple data sources and new methods of estimation resulted in a more accurate estimate of the HIV population in India in 2006. A critical review of the data and methods used in the past and current estimation processes is offered in this article.
The number of HIV-infected adults and children in India in the year 2006 has been estimated as 2.5 (2.0-3.1) million, down from 5.7 (3.4-9.4) million in 2005 (UNAIDS 2006). The steep downward revision in the estimate is attributed to the availability of multiple data sources and a new method of estimation.
The revision has generated mixed reactions - relief as well as scepticism (Bagla 2007; IANS 2007) and led to a plethora of queries about the process and factors that resulted in the lower estimate. A critical review of data and methods used in the past and current estimation process may provide clarity on these issues.
Method and Data
The size of population and prevalence among each category of risk groups are essential data for estimating the number of infections among adults in the population.
The major high risk behaviour groups (HRG) are female sex workers (FSW), injecting drug users (IDU), clients of FSW and men having sex with men (MSM).
The spouses of the aforementioned groups are categorised in the low risk group (LRG), representing the general population. As a majority of HRG are hidden and inaccessible, their size is generally estimated through specific surveys and/or mapping exercises.
The HIV prevalence among them is determined by assessing their HIV status under targeted intervention approaches. On the other hand, clients and spouses are dispersed in the community and form part of the general population. Their size is determined by subtracting the size of HRG from the national and/or sub-national projected population. Only properly designed population- based surveys can provide reliable estimates of HIV prevalence in this group.
HIV Estimation until 2005 India started estimating the number of HIV infections in 1998 under a broadly consultative procedure (Pandey et al 2007). The consultative group in 1998 suggested using sexually transmitted disease (STD) prevalence as the basis to estimate the size of the high risk behaviour population.
It was further extrapolated to males and females of urban and rural areas, again based on the consensual assumption for the urban-rural differential in STD prevalence. The size of HRG was determined by multiplying the STD prevalence with the projected population of the corresponding year. The size of the general population, male and female in urban and rural areas, was derived after subtracting the said HRG population from the total projected population.
The HIV Sentinel Surveillance (HSS) data had been the only source available for estimating HIV prevalence. It was confined to antenatal clinic (ANC) attendees, STD patients and IDUs. HIV prevalence amongst the first two groups was taken as the proxy for the general population and the high risk sexual behaviour groups respectively.
Since the ANC sites were mainly located in urban areas, consensus assumptions were made to extrapolate the prevalence among men and the rural population.
Infections among children were derived from the expected number of HIV prevalence among pregnant women and probabilities of vertical transmission and survival. The state level estimates were aggregated to arrive at the national estimate and the method was called the worksheet approach. The assumptions were revised in 2003 as more data became available, but they never validated with the community-based population representative surveys.
Over time, the HSS was expanded geographically as well as for other HRGs, FSW and MSM, and state-wise mapping and size estimation of HRG were also undertaken.
With the availability of data, these groups were also incorporated as independent entities in the estimation process. However, the STD patients were not withdrawn under the assumption that they still represented the clients and that the size estimates for HRG were incomplete.
This led to chances of double counting of HIV positives in the estimation process. Based on a study in Guntur district, Dandona et al (2006) observed that there the common practice of referral of HIV-positive/suspected cases to public hospitals and a preferential use of public hospitals by people in the lower socio-economic strata caused overestimation of the HIV burden in India.
HIV Estimation in 2006
The year 2006 presented a landmark when data from multiple sources became available. The third round of the National Family Health Survey (NFHS- 3), which incorporated HIV testing, enriched the data availability in 2006 for the general population.
The survey provided HIV prevalence in the general population and female- male and urban-rural ratios for each of the five high prevalence states (Andhra Pradesh, Karnataka, Maharashtra, Manipur, and Tamil Nadu) individually and for the remaining states together.
These data were used to calibrate HIV prevalence rate among ANC attendees and to validate the assumptions. Data generated through the Integrated Biological and Behavioural Assessment (IBBA) survey among HRG and clients in the six high prevalence states was used to validate the HSS results for HRG.
Further, the World Health Organisation (WHO)/United Nations AIDS (UNAIDS) workbook (Walker et al 2004) formed the worksheet anchor. Given the same inputs, both the approaches were found to generate the same results.
In 2006, the general population and HRG (including FSW, MSM, IDU) and long distance truckers were included for estimation. The STD population was dropped, while inclusion of HRG and truckers was considered important to account for missing (mobile/hidden)
Population in community-based surveys.
The HIV prevalence rates among ANC attendees in HSS were adjusted for intra and inter-state variations by applying mixed-effects logistic regression models, using SAS version 9.1.3 (SAS Institute, Cary, North Carolina). The adjusted HIV prevalence estimates were then calibrated against the same in NFHS-3 before entering into the workbook.
In order to obtain the trend estimates with the new method, point estimates were computed for five years starting from 2002. The trend estimates of HIV prevalence among ANC attendees for previous years had also been adjusted for inter-and intra-state variation and then calibrated to the NFHS-3 results.
The projection of adult HIV prevalence for the period 1985-2010 was generated by fitting a logistic curve to the five-point estimates.
Numeric results of the curve were then entered into the "Spectrum" (Stover et al 2006) to derive the epidemic curve for all ages.
For this, additional data such as population distribution, fertility rates, migration as well as uptake of antiretroviral treatment and prophylaxis for prevention of mother to child transmission were inputs into the model.
The number of infections in all ages (adults and children) in 2006 was estimated to be 2.5 million.
What Caused the Reduction?
The HSS was initiated with the objective of monitoring trends. These data, though, comparable over time, cannot be generalised even for all women. This limits their use in estimations. For example, over four fifths of antenatal clinic attendees are in the age range of 20-29 years, sexually more active and have had unprotected sex. The HIV prevalence observed among them is likely to be high and not representative of HIV prevalence among all adult women.
Further, low utilisation of antenatal services, particularly in the public sector facilities where the HSS are mostly located, also contribute to poor representation of antenatal clinic data as clearly brought out by Dandona et al (2006) in their study in Guntur district.
For this very reason, in 15 out of 20 countries of Africa where demographic and health surveys (DHS) were undertaken, the HIV prevalence in the DHS survey was lower than that was estimated among ANC attendees in HSS (Gouws 2006). The use of such exaggerated HIV prevalence to all women and men, in turn, inflates the estimate.
A community-based study in Cambodia also observed that though HIV prevalence in ANC data can be used for estimations, it suffers from the limitation of overestimating the infection in younger age groups (Saphonn et al 2002).
Despite recognising this limitation, the use of HIV prevalence among ANC attendees in HSS continued without correction as evidenced by the results of some community-based studies in Tamil Nadu (Thomas et al 2002; Kang et al 2005) which matched with the results of HSS in the state.
In retrospect, it was realised that these studies had low power and were conducted either by cluster sampling or by camp approach, which probably led to exaggerated HIV prevalence.
Secondly, continued use of STD population as a risk group even after inclusion of FSW and MSM in the estimation process also pushed the estimates upwards in the past. The assumption that they stood proxy for clients is not tenable because, as mentioned earlier, the HIV prevalence among clients and their spouses is encompassed in general
Inclusion of this group, therefore, led to double counting. Further, the HIV prevalence rates documented in STD sites of HSS were also exaggerated because a large proportion of the STD sites were located in tertiary hospitals, which mostly receive referred and chronic patients.
A comparison of HIV prevalence rates between the STD sites located in medical colleges and those in district hospitals has demonstrated this point.
NFHS-3 covered a sample of over 1,02,000, for an assumed prevalence rate of 0.9%. Now that they have a much smaller prevalence rate, about one- third of the assumed value, many people question the validity of these results. It is common knowledge that an estimate lower than the one assumed for sample size calculations increases the error bounds rather than invalidates the results. These errors are accounted for in the range provided around the estimate.
On the other hand, the proportion of the sample from low prevalence states is considerably small. As a result, one calibration factor had to be developed for all the low prevalence states together. This is expected to mask the magnitude of difference in the estimate of HIV prevalence between the states and the range of the estimate will be
much wider in these states.
In order to facilitate comparison, estimates were derived for five years starting from 2002 and it is found that the epidemic is stable at the national level, although at the state level some high prevalence states showed a decline and some in the low prevalence areas showed an increase in the epidemic.
However, the decline was significant only in Tamil Nadu. Further, in several districts of high and low prevalence states, HIV prevalence among ANC women was more than 1%.
The new emerging areas with high HIV transmission have been identified. The HIV prevalence among IDUs remains stable. This abundantly makes it clear that the lowered estimate does not indicate a decline in the epidemic but a correction for some incongruities in the data and in the previous method of estimation.
Bagla, Pallava (2007): "Don't Be Misled on AIDS", Times of India, New
Delhi, 10 July.
Dandona, L, V Lakshmi, T Sudha, G A Kumar and R Dandona (2006): "A
Population-Based Study of Human Immunodeficiency Virus in South India
Reveals Major Differences from Sentinel Surveillance-Based
Estimates", BMC Medicine,
Gouws, Eleanor (2006): "Comparison of Country Level ANC Prevalence in
Household Surveys and ANC in South India", paper presented in the
meeting of WHO/UNAIDS Reference Group on Estimates, Modelling,
Projections, held in Prague, Czech Republic, 29 November-1 December.
IANS (2007): "Experts Challenge New India HIV Estimate", Indo-Asian
New Service, Yahoo! India News.htm, 7 July.
Kang, G, R Samuel, T S Vijayakumar et al (2005): "Community
Prevalence of Antibodies to Human Immunodeficiency Virus in Rural and
Urban Vellore, Tamil Nadu", National Medical Journal of India, 18(1): 15-17.
Pandey, Arvind, M Thomas, D C S Reddy, Kant Shashi and M Bhattacharya
(2007): Indian Journal of Public Health, January-March.
Saphonn, V, L B Hor, S P Ly, S Chhuon, T Saidel, R Detels
(2002): "How Well Do Antenatal Clinic (ANC) Attendees Represent the
General Population? A Comparison of HIV Prevalence from ANC Sentinel
Surveillance Sites with a Population-Based Survey of Women Aged 15-49 in Cambodia", International Journal of Epidemiology, April, 31(2): 449-55.
Stover, J, N Walker, N C Grassly and M Marston (2006): "Projecting
the Demographic Impact of AIDS and the Number of People in Need of
Treatment: Updates to the Spectrum Projection Package", Sexually Transmitted Infections, June, 82 (Supplement 3), iii45-50.
Thomas, K, S P Thyagarajan, L Jeyaseelan et al (2002): "Community
Prevalence of Sexually Transmitted Diseases and Human Immunodeficiency Virus Infection in Tamil Nadu, India: A Probability Proportional to Size Cluster Survey", National Medical Journal of India, 15(3):135-40.
UNAIDS (2006): Report on the Global AIDS Epidemic, UNAIDS/06.13E,
Walker, N, J Stover, K Stanecki, A E Zaniewski, N C Grassly, J M
Garcia-Calleja, P D Ghys (2004): "The Workbook Approach to Making
Estimates and Projecting Future Scenarios of HIV/AIDS in Countries
with Low Level and Concentrated Epidemics", Sexually Transmitted Infections, August, 80 (Supplement 1) i10-13.