Epidemiology provides information on two distinct but closely related issues:

• How HIV is transmitted
• Who the epidemic affects (in the past, present and future).

In Britain information about individuals who test HIV antibody-positive or who are diagnosed with AIDS is reported in an anonymised form to either the Health Protection Agency (HPA) in London, or the Scottish Centre for Infection and Environmental Health (SCIEH). GPs, NHS hospitals and clinics as well as private hospitals and clinics all take part in this scheme.

Since the beginning of 2000 clinicians in England and Wales and Northern Ireland have been asked to report all first UK diagnoses of HIV infection in those aged 14 and over to the HPA.

Previous to this time, reports were made by the laboratories that actually performed the HIV tests on samples. Clinician reporting has greatly enhanced the quality of information available about the epidemic in the UK. For example, details of ethnicity were given on ninety-nine per cent of clinician reports. This compared to 48% of laboratory reports.

Surveillance programmes have to reach a compromise between completely anonymised reports, which may be inaccurate because there could be numerous duplicates of the same individual if they attend different clinics, and names-based reporting, which has been opposed because it was thought that both doctors and patients would object to the breach of confidentiality.

When AIDS monitoring was first set up in 1984 it was decided that reports of AIDS and (when HIV antibody testing became available in 1985) HIV should be sent to what was then the Public Health Laboratory Service (PHLS) in an anonymised form. The patient’s name and identifying details such as date of birth and address are stripped from the report, leaving the following demographic details:

• Age
• Gender
• Region of Residence
• Probably mode of HIV transmission
• Probably country where HIV was acquired
• Probable mode of transmission and country of acquisition of the infecting partner
• Ethnic origin (since 1995)

Individual diagnosis reports are identified by a code number which is issued by the HIV or GUM clinic or GP practice where the diagnosis is made. The code is a Soundex code, a method of transforming surnames into a simple numerical code that was first used in the US census of 1880.

Some surnames have unique Soundex codes but the majority of codes are shared between different surnames (the point of developing the code in the first place was to be able to trace people who spelt their surname inconsistently). It is based on the pattern of consonants in the surname. For instance the Soundex code for this writer’s name (Cairns) is C652 and is shared by the surnames Carmichael, Carrington, and Cornish, among others.

In theory the HPA could ask the clinic to identify the person behind a particular diagnosis report but in practice this is never done. The point of using the Soundex code as opposed to complete anonymity is to weed out probable duplicate reports. By the end of 1994 over 93% of the 20,407 reports on the HPA database had been Soundex coded, and 22% of the reports of HIV infection were recognised as duplicating earlier reports of infection. It also enabled 70% of AIDS reports to be linked to independent reports of HIV infection.

The collected data allows the HPA to assemble profiles both of the cumulative epidemic, and any changes in its form over time.

The HPA uses a number of exposure categories in order to identify the routes of infection amongst heterosexual men and women and to monitor whether HIV is spreading from particular subgroups amongst heterosexuals. This system is very useful, because it monitors for the spread of infection from groups recognised as being at higher risk of infection (`second generation' transmission).

Estimates of the total number of HIV-infected people in different exposure categories are important for public health planning. It is also useful to know the prevalence of undiagnosed infection in the community, and the future number of people with severe HIV infection.

The ‘direct’ method of estimating the total number of both diagnosed and undiagnosed prevalent HIV infections among adults in the United Kingdom (UK) uses a combination of data from different sources.

The total number of prevalent diagnosed HIV infections in England and Wales was derived by adjusting the number of reported infections from the National Survey of Prevalent Diagnosed Infections (SOPHID) for both under-reporting and failure to access services annually.

The total number of prevalent undiagnosed HIV infections was calculated by applying undiagnosed HIV prevalence estimates, obtained from the Unlinked Anonymous Prevalence Monitoring Programme (UAPMP), to different categories of the population at different levels of risk of HIV infection. These categories included ‘sex between men’, ‘sex between men and women’, and ‘injecting drug users’. Estimates of the total population within these categories were produced by combining the proportion of the population with particular behavioural characteristics, based on the results of the second National Survey of Sexual Attitudes and Lifestyles (Natsal 2000), with mid-year population estimates from the Office for National Statistics (ONS). Each population denominator was then multiplied by the relevant HIV prevalence estimate to derive the number of undiagnosed HIV-infected persons within that category.

The direct method protocol has been developed to include evidence-based adjustments that take account of differences in prevalence between subgroups within the population categories. The behavioural characteristics of the population have been updated using data from Natsal 2000.

In addition to the collection of HIV and AIDS diagnoses, the HPA also compile two other ongoing sets of data.

Unlinked Anonymous Prevalence Monitoring Programme (UAPMP)

This provides similar information without the biases involved in voluntary, `elective' testing. In anonymised seroprevalence studies, samples of blood taken for other medical purposes at hospitals, clinics, casualty departments and maternity units are tested for HIV. Any information which could identify the individual is removed, leaving only the basic demographic information.

At present the HPA has three anonymised seroprevalence studies in its Unlinked Anonymous Prevalence Monitoring Programme (UAPMP). These are:

• Attendees at GUM clinics: samples provided for syphilis serology are tested.
• Injecting drug users: unlinked anonymous saliva samples are collected from drug users in contact with over 50 specialist agencies in England, Wales and Northern Ireland.
• Pregnant women and their newborn babies: samples collected from women attending antenatal clinics and from newborn infants.

Survey of Prevalent HIV Infections Diagnosed (SOPHID)

This provides data on those patients diagnosed with HIV and attending HIV-related care in clinics throughout the UK. As well as the data collected in initial diagnosis reports, information on the stage of disease reached and the level of antiretroviral therapy used is collected too.

After all the information has been compiled and analysed, the statistics are released in the form of monthly updated bulletins from the HPA. The Department of Health also issues these as press releases. Standard monthly statistical updates include only AIDS statistics; quarterly tables also include HIV figures. More detailed breakdowns are made widely available to such people as senior medical figures and health authority managers. More complex surveys such as the anonymised seroprevalence surveys are published in journals such as HPA's Communicable Disease Report Weekly.

Limitations to the available statistics

Firstly, the interpretation of the data related to diagnosed infections must take account of who comes forward for testing. By definition, reported positive HIV antibody test results include only those individuals who have taken the test. Those who choose to get tested may not be representative of the total population of HIV–infected people.

Secondly, people who visit sexually transmitted infection clinics are likely to be among the more sexually active members of the population. HIV infection rates among them may therefore be higher than among the population as a whole. However they are also more likely to test for HIV.

In 2003 the HPA estimated that based on figures from the UAPMP, the proportion of gay men attending GUM clinics but did not know it was 25%. However in 2005 they revised their estimate of the total proportion of gay men who were undiagnosed upwards to 34%, after anonymised surveys in London, Manchester and Brighton suggested a higher figure (see the Dodds survey under Will we observe increases in HIV infection amongst gay men? below).

They also estimated that 28% of women in general with HIV and 39% of men (including gay men) were undiagnosed. It is clear that only a proportion of those who are HIV–positive come forward for testing, and this proportion may well be different in the various risk groups and in different parts of the country. Furthermore, reporting of identified infections and AIDS cases may be incomplete or delayed.

Conversely, despite the weeding out of duplicate reports by Soundex code, it is clear that a proportion of patients still get ‘double counted’ because they have presented for care at several clinics. HIV diagnoses may also overestimate the overall prevalence on HIV in the country (once the likely proportion of undiagnosed people has been estimated) because many people with HIV come from highly mobile populations and may leave the UK after diagnosis.

The same applies to intravenous drug users. A recent study in which saliva samples were collected by street agencies in touch with drug users that did not use traditional drug agencies¹ found that HIV prevalence was higher among these largely younger users, especially in London. 

Having said this, the UK probably has one of the best HIV surveillance programmes in the world. In the USA, for instance, each state has its own surveillance programme for HIV diagnoses. Some use a system of codes based on names, as in the UK; others first collect epidemiological data by name before coding them and sending them to the federal Centers for Disease Control; others keep the patients’ names linked to their data and rely on agreements to maintain patient confidentiality. As a result, the CDC is still unable to issue reliable figures for HIV infection that cover all 50 US states because it is not confident that each state collates statistics to the same standard of accuracy. It has recently announced that federal funding will from 2006 be withheld from states that do not adopt confidential names-based reporting, which they regard as the only safe method of weeding out duplicate records.

AIDS diagnoses, on the other hand, have by federal law to be reported on a name-by-name basis so, whatever concerns there may be about patient confidentiality, at least state-by-state comparisons are possible and the CDC can issue reliable national figures.

In Europe surveillance systems range from the completely anonymous, as in France and Germany, to Iceland, which has a fully named-patient system. Other countries add a measure of traceability to patient records by retaining initials and/or part of the patient’s social security number. No other country uses the Soundex code. Some countries like Italy and Ireland do not have a mandatory national surveillance system but rely on extrapolations from a sample of lab reports. France only recently instituted a national system in 2003, though because this was done so late, its national patient database is now probably the most sophisticated in the world.

Nonetheless HIV statistics are usually more reliable than those for other diseases, since most developed countries at least instituted some kind of systematised reporting of HIV and AIDS. Compare, for instance, the situation with hepatitis C. The UAPMP figures suggest that about one-third of the people with HIV in the UK are undiagnosed. The HPA can therefore make an estimate of the number of new infections (incidence) and the number living with HIV (prevalence) which has a statistical uncertainly of about 18% either way: that is, UK prevalence is in the order of 45,000 to 65,000.

In contrast there could be anything between 250,000 and 500,000 people living with hepatitis C in the UK, an uncertainty factor of 50% either way. Because reporting is not mandatory, there were only just over 8,000 new hepatitis C diagnoses notified to the HPA last year, scarcely more than the 7,500 or so HIV diagnoses.

How statistics are presented

The month-on-month developments in the HIV and AIDS statistics provide an indication of the evolving dynamics of the epidemic. However, unless viewed with caution and an awareness of their limitations they can be more confusing than enlightening.

The fact that HIV infection usually remains asymptomatic for years means that new positive HIV antibody tests in a given month or year do not necessarily reflect transmission patterns as they are taking place at that time. A person may choose to take the HIV antibody test one month after becoming infected, only many years later or when they already have symptoms suggestive of AIDS. Thus HIV statistics can only reveal the cumulative shape of the epidemic, rather than provide a snapshot of trends at any given moment.

This problem is all the more significant with AIDS statistics. Although it is likely that AIDS statistics are more fully documented than positive HIV antibody test results, simply because a person who develops AIDS is likely to seek medical attention, the long average period between the contracting of HIV infection and the development of symptomatic disease means that AIDS statistics are unlikely to be representative of current patterns of HIV transmission.

Incidence and prevalence

Epidemiologists use two different measures of the severity and burden of diseases and it is important to understand the distinction between the two.

Incidence is the number of new cases of the disease diagnosed within a specific time period. It measures the spread of the disease. It is usually expressed as the number of cases per patient-years. For instance, once all late reports come in, about 7500 people will have been diagnosed with HIV in the UK in 2004. This represents an annual incidence of 0.0125% or 1.25 cases per 10,000 patient/years. Incidence in vulnerable groups may be very much higher: for instance gay men in the UK are probably between 50 and 100 times as likely to get HIV as a member of the general population.

Incidence depends critically on case reports and tests and can’t be retrospectively determined from anonymous prevalence surveys. Therefore, as we said above, the rate of new diagnoses of HIV is not indicative of current incidence.

Given the ‘time lag’ in diagnoses, how can true incidence be estimated? There are a couple of clever ways to estimate incidence within given populations who do not all come forward for testing. At present these are only used as research tools but some, for instance, a detuned assay (see HIV testing) which detects which HIV diagnoses are of recent infections (usually ones contracted in the last six months) can determine the proportion of people testing HIV-positive who have recent infections and therefore the overall recent incidence in the population.

It was by using anonymous HIV testing coupled to a detuned assay that, for instance, a survey in Baltimore in the USA this year² found an annual HIV incidence of 8% among gay men, four times larger than the proportion of men reporting HIV diagnoses.

Conversely the incidence in San Francisco was only half of the new diagnosis rate, and previously estimated incidence, of 2.2%, indicating a possible move among gay men in the city towards behaviours less likely to result in IV infection.

Prevalence is the total number of people who have the disease at any one time. For instance in the UK about 60,000 people or one in 1,000 of the general population are currently living with HIV. Prevalence measures the overall burden of disease in a society. It depends crucially not only on the incidence but also on mortality (from the disease or from other causes) or on spontaneous recovery or cure rates if the disease is not lifelong like HIV.

Only by mapping both incidence and prevalence is it possible to predict whether an epidemic is getting more severe or waning.

While in general the fewer people suffer from a serious infection like HIV the better, there are situations in which increasing prevalence may be a form of good news; and in which declining prevalence may be bad news.

For instance: there are currently about eight times as many people living with HIV in the UK as catch it every year. But because the death rate among people with HIV is now only 6% of the annual incidence, it can easily be seen that this figure will continue to grow for the foreseeable future. The increasing prevalence of HIV in the developed world is primarily a sign of success in reducing the death toll from AIDS rather than failure to contain HIV.

Conversely where the mortality from a disease is particularly high, prevalence may fall even though incidence remains stable. This was the case in Uganda in the 1990s. HIV prevalence among people in the Rakai district fell from 17.6 to 11.4 per cent from 1990 to 2002, even though the annual incidence remained unchanged. This was largely because more people were dying from AIDS at the time than were catching HIV; the wave of deaths was reflecting the extremely high incidence of the previous decade.

The fact that this declining prevalence was seen as evidence for the success of the HIV prevention programmes of the preceding few years shows how the significance of prevalence and incidence figures can get very confused in people’s minds. The undoubted success of the Ugandan ‘zero grazing’ campaigns of the late 80s-early 90s are reflected not only in the declining incidence at the time but also in the fact that mortality has just started to decrease since 2000, as in the graph below:

(Source: Wikipedia)

The epidemiologist Roy Anderson sums up the situation with regard to HIV thus:

“The prevalence of HIV in a particular population will not grow indefinitely, it will saturate at some level. Following the initial spread of HIV there is likely to be a fall in the incidence of infection followed in turn by a resultant reduction in prevalence. A fall in incidence is likely to precede the fall in prevalence, as the time scale over which the epidemic saturates is likely to be more rapid than the time scale on which HIV associated mortality increases.”

Thus, even in a situation where there is no HIV treatment and mortality runs unchecked, there may be a considerable time lag between a fall in incidence and one in prevalence. This is likely to be prolonged by decades in a country that has universal HIV treatment, and if incidence remains unchecked, prevalence will not decline until the number of ‘at risk’ people in the population is saturated (see chart below). (Source: WHO)

This means that prevention campaigns have to be extremely successful to have a short-term effect on prevalence. This does appear to have happened in Thailand, where the institution of near-universal condom use among men visiting sex workers in the early 1990s brought the HIV incidence in young men down from 2.5 per cent in 1991 to 0.5 per cent in 1993. This rapid fall in HIV incidence was followed equally rapidly by a fall in prevalence (see below).

HIV-1 prevalence in Thai military conscripts

(Source: WHO)

This may be one of the few occasions in which an HIV prevention campaign has had such an immediate effect, apart from very early in the history of the epidemic, where rapid behavioural changes accompanied by high mortality achieved permanent declines in prevalence rates in groups like intravenous drug users in the UK.

Incidence and prevalence figures will tend to run ‘neck and neck’ in an acute infection such as influenza, and indeed prevalence is rather a meaningless term in a situation where prevalence during one week is a ‘snapshot’ of recent infection with virtually the same meaning as incidence, and by the end of the year may be zero.

In non-infectious conditions like cancer, annual prevalence is the most useful tool to map both factors giving rise to the disease and the effect of treatment. If incidence changes quickly, an unusual influence like an environmental agent may be suspected.

In a chronic, slowly-accumulating infectious disease like HIV, incidence therefore is the best measure of whether behavioural change or prevention interventions are having an effect on the epidemic, and can also be used to make predictions as to future prevalence. Prevalence (current and projected) is the best indicator of the burden of the epidemic and is the figure to use in planning services.

Difficulties in using statistics to plan services

At a local level, primary care trusts often experience problems in obtaining up–to–date and comprehensive data on the numbers of people testing positive in their district (and demographic breakdowns of potential service users) due to a lack of provision for the exchange of data between services, lack of resources to collate and analyse such data, and insistences that all data must remain confidential.

Even when data have been obtained from clinics, for instance, it may still not relate to actual service needs in particular districts because people may take HIV antibody tests at clinics outside their district of residence.

There is even a phrase – ‘perverse incentives’ – that refers to the fact that area-based health services like PCTs have a disincentive to improve open-access services like GUM clinics, as they will attract more patients and increase running costs.

The lack of data sharing between districts may perpetuate the phenomenon of local residents going outside the district for HIV/AIDS services. If health authorities and local authorities perceive an apparently low demand for services within the district, they will not consider it necessary to provide them. For instance, districts in south London may not be aware of the extent of their HIV–positive populations because people are using hospitals north of the river, or Welsh PCTs because of patients using English clinics.

Despite the need for reliable data for use as a basis for planning, many hospitals and local authorities have not had the resources to devote time to data collection, analysis or publication. The quality of epidemiological data is strongly related to the financial incentives available to district health authorities. Thus, if funds are allocated on the basis of HIV reporting, districts will have an incentive to invest in improved reporting and analysis, which in turn will yield more cost effective targeting of services.

More detail on how UK statistics are compiled can be obtained from the Health Protection Agency’s website at http://www.hpa.org.uk/. Links to this site and other useful sources of statistical information on HIV can be found at aidsmap.com.