The method of data collection to estimate HIV prevalence and burden in India has improved considerably. But more needs to be done, including setting up more high-risk group sites and increasing data collection from the general population, says M Prasanna Kumar
‘Surveillance’ is the collection of data for action. ‘Sentinel surveillance’ for HIV is done to provide information on the trend and level of the HIV epidemic in specified populations. This can serve as an early warning of a situation that needs action. This article discusses the findings of the 2007 round of sentinel surveillance in India, an exercise that has been conducted nationwide annually since 1998.
What surveillance consists of
Sentinel surveillance for HIV is conducted on specified high risk groups such as sex workers (SWs), men who have sex with men (MSM) and injecting drug users (IDUs). It also looks at the low risk group of pregnant women seeking antenatal care (ANC) which represents the general male and female adult population. Surveillance ‘sites’, from where blood samples are obtained, are chosen for each group. Information on pregnant women is obtained from antenatal clinics in government hospitals. Government hospitals are also the sites for taking samples from people with sexually transmitted diseases. In the case of high risk groups, the sites are the drop-in centres of non-government organisations providing services for these groups.
Some 250 blood samples are supposed to be taken from each high risk group site and 400 samples are taken from each site for pregnant women. The number of sites is: 248 STD, 52 IDU, 40 MSM, 137 SW, 3 migrant, 1 transgender and 7 truckers. In order to maintain consistency, surveillance sites do not change from year to year. Samples are collected for any three consecutive months every year, throughout India.
In theory, surveillance is done on samples collected for other purposes. In practice, a special programme is run for surveillance during which samples are collected and tested for syphilis (the results of which are given to the participant), and an unused portion of the sample is used for HIV testing. Personal data such as age, sex, educational status, occupation and migration status are collected separately. One portion of the sample is sent to the lab to test for HIV after removing all identifying labels so that it is impossible to link the test result to the person from whom the sample was taken, to maintain his/her anonymity.
(The use of annual sentinel surveillance and the results of previous surveillance rounds are described in more detail on http://hivaidsonline.in/index.php/Debates/20-million-or-2-million.html )
The 2006 surveillance
The 2006 sentinel surveillance was a landmark for a number of reasons. The number of sentinel sites was expanded from 703 to 1,122, thus making possible representation of almost all districts of the country. The HIV estimation process also underwent a shift in that the WHO/UNAIDS Workbook was used for deriving HIV estimates. This made international comparisons possible. In addition, other sources of data such as the results of the third National Family Health Survey (NFHS-3) and the second round of the National Behavioural Surveillance Survey were used to calibrate and refine HIV estimates. The result was a halving of the previous national estimate of adult HIV prevalence (from 0.91% to 0.36%) and of the total number of people with HIV (from 5.2 million to 2.47 million). This meant that for many years the country had been overestimating the prevalence of HIV (percentage of infected people in the population) as well as the burden of the disease (the total number of people infected).
The same methodology was also used to re-estimate HIV prevalence for each risk group every year, going back to 2002, using HIV prevalence data from those sites that had adequate sample sizes in the previous five years. The newly derived HIV prevalences and the 2006 data were used to create an epidemic curve that projected the adult HIV prevalence from 1985 to 2010.
Spectrum software, which is used globally for projecting the impact of the AIDS epidemic, was then utilised with additional information on the numbers of people on antiretroviral treatment, the fertility rate and other parameters, to estimate and project the number of those living with HIV at all ages and the national HIV burden from 1985 to 2010. (1)
One should keep in mind that all these are working estimates. They are needed by policy makers to monitor the trends of the epidemic and to plan and implement control measures. The basis for all these estimations and projections are samples from heterogenous population groups and not actual head counts. Hence the results certainly will have a margin of error. Further, estimates using different methods are not comparable. Future refinements of the methodology used today may even lead to changes in the present figures.
Hence, although the 2006 estimates did lead to a downward revision of the HIV prevalence and the burden, it did not connote a decline in the epidemic; it only gave a truer picture of the epidemic.
The 2007 surveillance round
In the 2007 round, the same methodology was followed as in 2006. A total of 358,797 blood samples were taken from 1,134 sites in the country; 646 from pregnant women at ANC clinics, and 488 from high risk group sites (for SW, MSM, IDU, male migrants and truckers). (2)
The 2007 round found an HIV prevalence of 7.2% among IDUs, 7.4% among MSM, 5.1% among SWs and 0.48% among pregnant women. When compared to the previous year’s figures (6.9% among IDUs, 6.41% among MSM and 4.9% among SWs and 0.60% among pregnant women) (3), the 2007 results suggest that the trend in the epidemic continues; it is concentrated in the high risk groups with relatively little spill-over into the general population.
In the 2007 round, Andhra Pradesh is the only state with an HIV prevalence of more than one per cent among pregnant women. However, 87 districts in the country (out of 476 districts with sites for pregnant women) were found to have more than one per cent prevalence among pregnant women. Some districts recorded a prevalence of more than one per cent for the first time. Several of these districts are in states previously believed to be low prevalence states. (2)
These results suggest that at the all-India level, while the HIV prevalence among SWs and pregnant women is clearly declining, the prevalence is stable to rising among IDU and rising in MSM.
Levels of the epidemic
The HIV epidemic is described as low level if less than 5% of any high risk group and less than 1% of the general population is infected.
If more than five out of every 100 people in any high risk group is infected, but less than one in every 100 people in the general population is infected, the epidemic is described as a concentrated epidemic: infections are mostly confined to people with high risk behaviour with the general population largely unaffected. Based on the prevalence estimates from the 2006 and 2007 surveillance, India’s HIV epidemic is a concentrated epidemic.
If more than 1% of the general population is infected, it is a more serious matter and known as a generalised epidemic. Depending on the extent of linkages between the high risk population and the general population – through unprotected sex or shared injecting drug use – a concentrated epidemic can evolve into a generalised epidemic, slowly over time or rapidly.
Concentrated epidemics are for the most part confined to high risk groups whose numbers are limited. This situation offers a window of opportunity for controlling the epidemic, if effective and prompt interventions are implemented. Generalised epidemics are harder to control since the risk behaviour of the entire population has to be addressed for arresting the epidemic.
What do these results mean?
In a large country such as India, it is best to consider the HIV epidemic as a series of localised epidemics among different populations such as SWs and their clients, MSMs and IDUs, with some spill-over into the general population. In various regions, particular sub epidemics predominate. Thus, we have a fairly intensive – or concentrated -- epidemic in south India which is mainly sexually driven and may be related to easy availability of commercial sex, even in small towns, and the traditional forms of sex work such as devadasi, that is prevalent in some states. The other concentrated epidemic is in the northeast (due to injecting drug use). Much of north India shows low levels of infection. Hence it is quite possible to have an overall nationwide reduction of prevalence in the epidemic even while a few localised areas show an increase in HIV prevalence. In the low prevalence states, too, we have pockets of much higher prevalence. (2)
HIV prevalence among all high risk groups has declined in most high prevalence states – Andhra Pradesh, Karnataka, Maharashtra, Tamil Nadu (now no longer a high-prevalence state), Manipur and Nagaland. This could be due to effective interventions among these groups. However, more research is needed before we can arrive at a conclusion on this subject.
While HIV prevalence among IDUs in various parts of the country is generally stable, it is increasing in some sites. It was formerly believed that IDUs were present only in the north east of the country and in major metropolises. However, pockets of IDUs are now being identified in most states. In view of the fact that explosive HIV epidemics can occur among IDUs in a matter of months, far more intensive efforts at mapping them and starting interventions among them are called for.
Injecting drug use is a criminal activity and is carried out surreptitiously. To understand the true extent of injecting drug use, mapping of the IDU sites is done using ex-users and/or currently active users. They can identify other users and through a snowballing technique unravel the true extent of injecting drug use in an area, including the numbers of users, sites of drug use, etc. Unless such mapping is done policy makers will not be aware of the extent of drug use in their community.
The role of men having sex with men (MSM) in the Indian epidemic had not been appreciated to any great extent in the past. Now it is realised that their role has been greatly underestimated. An expert committee (4) has estimated that 5% of sexually active men engage in same sex activities and 20% of these men have had more than five partners in the previous month.
MSM sites in high prevalence states have shown a high HIV prevalence. Among the 40 districts that had MSM sites, 21 had more than 5% HIV prevalence. In nine sites the HIV prevalence exceeded 15%. MSM are a hidden population and need specially designed programmes to be effective. A major problem in designing and implementing effective programmes is that communities refuse to accept that same sex behaviour is commonplace and deny the need for interventions. Here, too, mapping of MSM by using their peers is necessary to discover the extent of the sexual networks and run effective programmes for them. There is considerable commercial MSM activity as well, which is one of the factors leading to the high HIV prevalence seen among this group.
Need for more data on high risk groups
While the 2007 surveillance round provides more information than before, there are a number of limitations resulting from the surveillance process. Most importantly, there are insufficient surveillance sites among high risk groups. Only 176 districts out of the 622 districts in the country have high risk group sites. Even these are for the most part in high prevalence states. Maharashtra, Karnataka and Tamil Nadu among the high prevalence states have high risk group sites in only less than a third of the districts. The existing high risk group sites are mainly SW sites with relatively fewer MSM and IDU sites. Since 47 out of 129 districts with SW sites showed a prevalence of more than 5%, the number of SW sites itself may be increased. Uttar Pradesh has only nine sentinel surveillance sites among high risk groups. Rajasthan has no sentinel surveillance site for MSM so we have no information on the prevalence of HIV among this group in that state. Far more high risk group sites need to be set up in low prevalence states. (5)
Since 1998 STD clinics have been considered as high risk group sites. In the 2007 sentinel surveillance round, there were 248 sites in STD clinics. But there are major limitations in the choice of STD clinic attendees as representative of high risk groups. The HIV prevalence among STD clinic attendees certainly will offer indications regarding the state of the epidemic in the region which the clinic serves. But practically all STD clinics are in referral hospitals and attract a select group of individuals who have symptoms of STDs following unsafe sex and clearly have not taken any protective measures. So they will include a higher proportion of HIV infected individuals. Hence sites in STD clinics need to be gradually eliminated. Since 2006, STD clinic data is no longer used for estimating the HIV burden though it is used to monitor the trend in a high risk population.
For the general population, too, more data are needed. State-specific data are available under NFHS-3 only for the high prevalence states and UP. More extensive community level surveys may be needed to get a better picture at the state level in other states.
For many parameters that are used, only estimates are available. Some of the critical parameters are the sizes of populations at higher risk.
HIV prevalence among pregnant women
Prevalence among pregnant women has been showing a considerable decline over some years. An analysis of sentinel surveillance data of HIV prevalence among young pregnant women (15-24 years) has shown that new HIV infections have decreased by as much as 54% in the southern states between 2000 and 2007. (6) This may well be because effective prevention activities were in place earlier and for a longer time in the south and the impact is now becoming evident.
But although the decline in prevalence is very evident in the southern as well as north-eastern states, there is an actual increase in the northern states. Most of the increase is contributed by Gujarat, Rajasthan, Orissa and West Bengal.
Estimates of HIV burden
The HIV prevalence among various groups is also used to estimate the HIV burden of the country. Until 2005, an algorithm was employed which used a large number of assumptions to estimate the HIV burden. Some of the major assumptions were the ratio of HIV infected men to women, the rural/urban distribution of HIV infection, and the STD prevalence in urban and rural areas.
Since 2006 the WHO/UNAIDS Workbook method, modified for India, is used to estimate the HIV burden. In this method, the population is divided into several compartments with higher risk such as SWs, MSM, IDUs and truckers, and the general adult population, both male and female, in rural as well as urban areas, which is at lower risk. For every state, the HIV prevalence in each of the above compartments is multiplied by the number of individuals in the compartment to give the estimate of the HIV burden for that group. These estimates are added up to arrive at the state’s total HIV burden. Several statistical adjustments are made for greater accuracy.
The 2007 HIV burden estimate is not just based on the sentinel surveillance prevalence data. The NFHS-3 data are also used to calibrate the HIV prevalence obtained through sentinel surveillance. This is done because it was realised that HIV-positive individuals were overrepresented among sentinel surveillance site attendees. Sentinel surveillance sites are mostly in urban areas. So in order to factor in the rural population who form 75% of the entire population, it is necessary to employ a correction factor which is derived from the rural-urban differential in HIV prevalence obtained from NFHS-3 the previous year. Importantly, men from the general population are not sampled in sentinel surveillance; the number of infected men is estimated indirectly by multiplying the HIV prevalence in pregnant women by the ratio of HIV-infected men to women that was obtained from NFHS-3 in 2006. At the all-India level, the ratio of infected men to women is 1.6:1 as per the NFHS-3 data.
However, NFHS-3 data have some limitations. For one, samples were taken from only those who gave their consent. It is possible that those who believed that they might be HIV-positive were more likely to refuse to give their consent. This would have introduced a bias in the sample. High risk groups, who may constitute up to 15% of the total HIV burden, are underrepresented in NFHS-3, which was a household based study and thus will exclude people on the move such as migrants and others with higher risk with no regular living place, such as SWs.
Such limitations introduce some degree of uncertainty in the final estimate. But for all that, the estimation process follows a very thorough methodology.
Despite their limitations, the annual sentinel surveillance rounds carried out since 1998, give a good picture of the trends of the HIV epidemic in the country. This was possible only because the sentinel surveillance programme was planned and executed meticulously over the years with the help of several national institutions and the involvement of teams of experts in every state.
The data sources have increased markedly, but there are still many uncertainties. More sentinel sites are required to uncover newly emerging localised pockets of infection. HIV incidence studies, if instituted, will give clues to the direction the epidemic is taking. Incidence studies estimate the number of new infections over a given period, usually a year. Prevalence studies estimate the proportion of infections in the population at a given point of time.
The Tamil Nadu experience
Tamil Nadu has been a trailblazer in meeting the AIDS challenge. It can justly claim success in epidemiological reversal. The HIV prevalence in this state of 70 million people has come down to 0.36% among pregnant women in 2006 from 1.59% in 2000, from 63.8% among IDUs in 2003 to 16.8% and the STI prevalence in the general community is down from 15% in 1998 to 10.6% in 2004. This drop was a result of the state governemnt’s aggressive multi-sectoral, multidimensional response which was launched early in the epidemic. HIV prevalence among pregnant women in the state has been consistently well below one per cent for more than five years. It is no longer a high prevalence state. It has shown - for the first time in this country - that control measures do work.
Tamil Nadu pioneered the AIDS Control Society model for efficient programme management at the state level, a model which was soon followed by the rest of the country. The Tamil Nadu AIDS Control Society and AIDS Prevention and Control Project of Voluntary Health Services provided a strong system under which all stakeholders and organisations working in the area of HIV prevention in the state were able to come together and work in a synergistic fashion. Many innovative collaborative efforts such as a network of 55 NGOs implementing targeted interventions for high risk groups, annual rounds of behavioural surveillance, numerous exemplary research studies and workplace interventions have been the outcome. (7)
(Dr M Prasanna Kumar is former deputy director of the Kerala State AIDS Control Society)
4. Report of the Expert group on Size Estimation for NACP-3 planning, 2004.
5. Technical consultation to review HIV surveillance in India. New Delhi: WHO SEARO, NACO MOHFW; 2008 Apr 23-5. Available from: http://www.searo.who.int/LinkFiles/Publications_HIV_Tech_Consul.pdf
6. Arora P, Kumar R, Bhattacharya M, Nagelkerke NJ, Jha P. Trends in HIV incidence in India from 2000 to 2007. Lancet 2008 Jul 26;372(9635):289-90.
7. http://www.pibchennai.tn.nic.in/karuvoolam/Releases per cent202007/November per cent202007/30112007/HIV-AIDS per cent20feature.htm
InfoChange News & Features, December 2008