HIV and risk environment for injecting drug users: the past, present, and future

Case 1: Odessa, Ukraine

Ukraine is the most HIV-affected country in Europe and Central Asia, with around 440 000 people aged 15–49 years infected with the virus (HIV frequency of 1.63%).¹¹⁸ In some cities, the frequency of HIV infection in IDUs rose from nearly zero in 1994 to more than 50% in 2 years, and as many as 820 400 people might be infected by 2014.^119,120

Odessa is among the cities bearing the brunt of the HIV epidemic in Ukraine, with the first cases registered in 1987, and outbreaks documented in 1995.¹²¹ As in the rest of country, the epidemic was chiefly attributed to heterosexual transmission. However, by 1997, IDUs accounted for nearly 85% of all cases of HIV infection. Unprotected sex and injecting practices, including drawing of drug solutions from common containers, shared syringe use, and unprotected sex with concurrent casual or sex-trade partners, continued to fuel the epidemic.^95,97 Young IDUs are more likely to engage in risky behaviours; 38% of all people infected with HIV in Odessa are 21–30 years old.¹²² Growing use of stimulants, especially in young people, leads to increased engagement in high-risk sexual and injecting behaviours; and drug-treatment programmes fail to reach stimulant users because they mainly target users of opiates.⁹⁷

High rates of corruption, pervasive poverty, and marginalisation of drug users are epidemic macroenvironmental forces, as explained through the voices of people who inject drugs (panel 2). Since 2004, government spending directed at increased access to antiretroviral therapy, harm reduction, and buprenorphine maintenance has substantially increased,¹¹⁸ but remains suboptimal¹⁰⁹ (table 1). Although 12–13% of the prison population injects drugs, there are no needle and syringe programmes or opioid substitution treatments available to prisoners.

Figure 3 shows the rapid spread of HIV in Odessa in the late 1990s, which has since stabilised at 50% prevalence. Few measures of the early rate of HIV spread and the small sample sizes of available prevalence estimates implied that a wide range of alternative epidemic trajectories were supported by the data, leaving substantial uncertainty in model projections. If we assume there will be no further change in risk in IDUs in Odessa, we predict there will be 6200 (95% credible interval 4500–6500) new HIV infections in IDUs between 2010 and 2015 (figure 3). New HIV infections from injecting drug use are mostly (40–80%) attributed to exposure to contaminated injection equipment, but another substantial route of transmission is unprotected sex between IDUs (15–45%).

Suboptimal access to clinically indicated antiretroviral therapy (CD4-cell count ≤350 cells per μL) causes 40–47% of infections. Many HIV infections in IDUs could be averted through timely and widespread interventions that reduce the frequency with which IDUs use non-sterile equipment. A 60% reduction in the unmet need of opioid substitution services could avert 10% (5–18%) of new infections, and a similar reduction in unmet need ofneedle and syringe programmes could avert 5% (2–8%) (figure 3). However, even in these optimistic scenarios, many infections would still arise because of the residual risk from contaminated injection equipment and unprotected sex. Therefore, a scale-up of these programmes should proceed alongside equitable scale-up of antiretroviral therapy and support for behavioural changes and condom promotion in this setting. 41% (32–47%) of infections could be averted if coverage of opioid substitution therapy and needle-exchange programmes were scaled to reduce unmet need by 60%, and if 60% of IDUs were initiated promptly on antiretroviral therapy when indicated.

Elimination of police beatings in Odessa, Ukraine

Removal of an environmental risk factor can have a substantial effect on incidence of HIV infection. Our reviews^41,51,75 and other studies (see papers 48, 49, 60, 62–64, and 66 in webappendix pp 38–39) suggest that policing practices can directly affect the risk of HIV acquisition by affecting where, when, with whom, and the context within which drugs are injected (eg, IDUs rushing injections or injecting in shooting galleries to avoid arrest). Police can also have an indirect effect on HIV risk, via pressuring or displacing IDUs away from needle and syringe programmes or drug-treatment settings.¹²³ Problematic policing practices include arresting of drug users for carrying of sterile or used syringes, harassments at needle and syringe programmes or drug-treatment clinics, soliciting of bribes to avoid arrest, or in extreme cases, sexual abuse or violence. These actions can increase the chance of IDUs injecting with contaminated equipment. 24% of IDUs in Odessa report ever being beaten by police (table 1 and panel 2),⁹⁹ and these people are more likely to report shared syringe use, use of preloaded syringes, and frontloading or backloading of syringes, among other risk behaviours.¹²⁴ We suggest that elimination of police beatings would remove the excess in risk behaviour, across the population, that stems from some individuals ever having been beaten. The effect of this assumed change in risk behaviour on HIV incidence in Odessa was estimated with the model (see webappendix p 10).

Our model projects that a substantial number of new HIV infections could be prevented in Odessa by 2015 through elimination of police beatings (figure 4). In two other Ukrainian cities, Makeevka and Kiev, where beatings are less associated with increased risk behaviour (Booth R, unpublished data), we estimate 2–9% of infections could be averted. The influence of police beatings and other practices on HIV incidence probably varies by location and operates through several direct and indirect causal pathways, making precise quantitative estimates impossible to generate without additional data. Additional studies with quantitative and qualitative methods are needed to elucidate the magnitude and direction of these associations, and intervening variables and the interplay between them. We will then be able to assess the effect of elimination of these human rights violations on HIV incidence in IDU populations.

Case 2: Karachi, Pakistan

Pakistan has had a thriving heroin consumption market for decades, and is neighbour to Afghanistan, the world’s largest producer of opium. In the 1980s and 1990s, national surveys¹⁰⁷ suggested that the number of drug users increased yearly by 12%, with most people who used drugs taking heroin by inhalation or smoking. In the past 10 years, however, changes in drug trafficking and distribution prompted transitions from smoking of heroin to injection of heroin, liquid opiates, and other pharmaceuticals, these being readily available over-the-counter at chemist shops. By 2000, there were an estimated 500 000 people who used drugs nationwide,¹⁰⁸ of whom 15–30% were IDUs.¹⁰⁷ Although non-therapeutic drug use, prostitution, adultery, and homosexuality are illegal in Pakistan, all are widespread.¹²⁵

HIV prevalence in Pakistan’s population of 91 million aged 15–49 years is low at 0.1% (~96 000 people). However, HIV outbreaks in IDUs occurred in Larkana in mid-2003,¹⁰⁷ and in Sargodha, where HIV prevalence in IDUs rose from 9% in 2005–06, to 51% in 2007.⁵⁰ HIV prevalence was 23% in Karachi, Pakistan’s most populous city in 2003,¹⁰¹ where there were 9000 IDUs in 2006,¹⁰⁷ and was 19.6% in nearby Sukkur and 18.3% in Hyderabad. As in other cities, most IDUs in Karachi are male; 2–3%¹⁰⁷ of female sex workers inject drugs, and 20–25% of these workers report having sex with IDUs.¹⁰⁷ About half of male IDUs in Karachi are married;¹⁰¹ 47% reported having sex with female sex workers,¹⁰⁵ and 9–14%^101,105 reported having sex with a male sex worker or hijra (transsexual) in the past month—two-thirds never use condoms.¹⁰⁵

Challenges to prevention include the tendency for IDUs to use professional injectors (hit doctors), inject together in groups with common syringes or ampoules (79.5% of all injection drug users),¹⁰⁵ and mix their blood with drug preparations before passing it on in an effort to potentiate the high. Furthermore, in the previous year, 16% of IDUs in Karachi reported visiting other cities in Pakistan,¹⁰¹ and 28% reported selling their blood.¹⁰⁶

Needle and syringe programmes are present in Karachi and in many major cities in Pakistan,¹⁰⁷ some of which provide mobile services, but coverage is grossly insufficient.¹⁰⁹ Detoxification and residential rehabilitation are the predominant strategies for drug treatment. Although buprenorphine maintenance therapy has been piloted in a few cities,¹⁰⁹ opioid substitution remains unavailable. In 2009, less than 1% of IDUs with HIV infection were receiving antiretroviral therapy.¹⁰⁹ Although Pakistan’s epidemic is concentrated, it could be increasingly difficult to contain if reported rates of highrisk behaviours continue, or if high rates of transition to injection happen as a result of macroenvironmental changes in drug markets (ie, reduced availability or purity and increased price of heroin), in the absence of effective prevention scale-up.

In Karachi, the HIV epidemic spread in the late 1990s, although the earliest prevalence measurement in 1995 suggested little evidence of HIV penetration in IDUs. However, by 2003, estimated HIV prevalence in this group was 23% (figure 5), which rose to 30.1% in 2006, and decreased to 23.1% in 2008.¹⁰¹ We estimate that HIV incidence will stabilise, with 2400 (95% credible interval 1300–4400) new HIV infections in IDUs in 2010–15.

Modelling provides most support for rapidly growing epidemics beginning after 2000, because of the high degree of shared use of injection equipment.^105,106 Although a quarter of male IDUs have sex with female sex workers,¹⁰¹ the low HIV prevalence in female sex workers in Karachi and the relative frequency and probability of transmission through sex and injection behaviours suggest that little transmission is attributable to this route. Furthermore, only about 4.6% of HIV infections are attributed to unprotected sex with other IDUs (figure 5). However, the proportion of HIV infections attributable to inadequate antiretroviral therapy coverage is substantial (19–40% if antiretroviral therapy is started at CD4-cell count ≤350 cells per μL), but is marginally lower than that reported in Odessa because of the higher proportion of cases of incident HIV caused by injection behaviour in Karachi.

Non-injecting drug users might increasingly transition to injecting in the coming years, and, because of the large numbers of these people in Karachi,⁵⁰ the spread of the epidemic could accelerate. If 10% of non-injecting drug users started injecting in 2010, the number of new HIV infections could increase by 82% by 2015 (~4000 extra infections) compared with the most conservative scenario that assumes no transitions (figure 5). If 12% of non-injecting drug users started injecting, the increase in new infections in IDUs in Karachi would be close to 98% (4400 extra infections). Because of the virtual absence of opioid substitution in Pakistan, implementation and scale-up of this therapy could have a substantial effect on reduction of shared syringe use by lessening of the frequency of injections or the number of active IDUs. If 60% of IDUs in Karachi were to be given opioid substitution, the number of new HIV infections could decrease by up to 28% in 2010–15. Thus, actions to limit the quantity of drug users who transition to injecting and a substantial scale-up of opioid substitution therapies are two priority interventions for Pakistan.

Case 3: Nairobi, Kenya

Kenya is in eastern Africa, borders the Indian Ocean between Somalia and Tanzania, and has a population of 39 million, 42% of whom are younger than 15 years of age. 1.5–2 million adults and children had HIV in 2007, of whom 1.4–1.8 million were aged 15 years and older.¹²⁶ Only a quarter of Kenyans with HIV infection know their serostatus.¹¹⁷ Urbanisation is happening at an estimated 4% annual rate of change (2005–10 estimate), and the capital city, Nairobi, has an estimated population of 3.04 million. Recreational drugs taken in Kenya include bhangi (herbal cannabis), heroin, khat (a plant with amphetamine-like properties), flunitrazepam, diazepam, glue, and alcohol.¹²⁷ Estimates of the number of people who take heroin vary from 12 201,¹¹⁷ to 18 000,¹²⁷ up to 24 500.¹¹⁶ Overall, less than 10% of heroin users are thought to have injected.¹¹⁷ However, one study¹²⁸ reported that 44.9% of Nairobi heroin users had injected and 52.5% of 101 present injectors were HIV positive, compared with 13.5% of 181 non-injecting heroin users.¹²⁸ Hepatitis C prevalence was 61.4% in the presently injecting group compared with 3.8% in those not injecting. In 2005, although only 0.3% of the adult male population were injecting drugs, an estimated 4.8% of new HIV infections happened through this route.¹¹⁶ Although 73% of Nairobi injectors knew about the risks of HIV transmission in 2005, 28% had used a needle after someone else, and 43% had passed their own needle to another person in the past 6 months.¹²⁹ By 2008, injection-drug use accounted for 3.8% of incident HIV infections in Kenya (5.8% in Nairobi).¹¹⁷

Kenya’s national AIDS strategic plan for 2010–13 referred to development of policies and legislation for the provision of needle and syringe programmes and opioid substitution therapy for the first time. Needle and syringe programmes are not permitted by law in Kenya,¹⁰⁹ but some non-governmental organisations in Nairobi and Mombasa reportedly provide education about harm reduction. There are very few opioid substitution services in private clinics in Kenya, and substitution therapy is not permitted in public clinics because of rigid interpretation of a law regulating opiate pain medications. Treatment that is not reliant on non-opioid substitution for drug dependence is available in government hospitals and is provided by non-governmental organisations.¹²⁹ The number of people receiving antiretroviral therapy in Kenya rose from 29 000 in 2004,¹²⁶ to 242 881 in 2008,¹³⁰ but less than 1% of IDUs who were HIV positive were receiving therapy in 2008.¹⁰⁹

The HIV epidemic in IDUs in Nairobi will grow substantially; 4000 (95% CI credible interval 2600–5400) new HIV infections are expected from 2010 to 2015, 95% as a direct result of exposure to contaminated injecting equipment and 5% from sexual transmission (figure 6). In the model, the much higher rate of non-sterile equipment use in Nairobi than Odessa or Karachi leads to a lower proportion of risk of HIV infection for injection drug use via unprotected sex there, despite the much higher prevalence of HIV infection in sexual partners. The high rate of non-sterile equipment use could also undermine the effect of needle and syringe and opioid substitution programmes (figure 6). Scale-up of needle and syringe programmes to reach 40% of the population might reduce HIV prevalence by as little as 4%, assuming that such a programme halves the frequency of injections with nonsterile equipment.

However, a combination of interventions at high coverage would have a far greater effect; 80% coverage of both opioid substitution and needle and syringe programmes could avert around 1100 (26%) infections by 2015. Elimination of laws prohibiting opioid substitution and scaling up services to 80% of IDUs could reduce the number of incident HIV infections by 14%. If increased coverage of antiretroviral therapy is added to opioid substitution and needle and syringe programmes, with rapid initiation when CD4-cell counts are lower than 350 cells per μL, about 1800 (45%) infections could be averted. However, the high rate of reported risk behaviour in the population implies that, even with all these interventions at full coverage, there could still be more than 2000 new infections in IDUs during 2010–15. Thus, to further reduce the effect of this epidemic, sustaining of high-coverage, high-effectiveness needle and syringe programmes and opioid substitution interventions will be crucial. With an effectiveness rate of 70% (rather than 50% assumed elsewhere, see webappendix p 11), and with 80% of the population accessing needle and syringe and opioid substitution programmes, our model suggests that almost 2500 (~67%) new infections could be prevented.

Comment

Our systematic review of the published work from the past decade, and modelling scenarios of local HIV epidemics in three diverse cities emphasises the important role of the risk environment in vulnerability to HIV infection in people who inject drugs. Although the published work suggested that physical, social, economic, and political environments can be microenvironmental or macroenvironmental, and are independently associated with HIV infection in injectors, only a quarter of studies were designed to assess environmental influences. Even fewer tried to measure macroenvironmental influences, or economic or political determinants. Fewer than a third of studies were done in countries with the greatest burden of HIV infection in IDUs, suggesting that research has not kept pace with the rapid evolution of HIV epidemics in people who inject drugs.

Research of the HIV risk environment

Heuristics of the HIV risk environment (figure 1 and panel 1) provide a loose theoretical framework for empirical efforts to improve the evidence-base on the social relations of HIV transmission.^11,13 Undertaking of epidemiological studies that explicitly delineate social-determinant causal pathways for HIV infection to inform social and structural approaches to HIV prevention is daunting.¹³¹ Although our review shows that epidemiological studies rarely explicitly embrace investigations of social determinants, they nonetheless identify environmental risk factors as important. We contend that future epidemiological studies of HIV in IDUs should make use of the methods and findings of qualitative and social science research in their efforts to systematically delineate how microenvironmental and macroenvironmental influences combine to increase or reduce HIV risk. We advocate for fully integrated mixed-method approaches that triangulate findings from qualitative and quantitative methodologies.

Findings from our review present challenges for future research. Investigations should shift from binary epidemiological models of straightforward cause and effect to multifaceted models that emphasise HIV infection as an outcome of many contributing factors interacting at once.¹¹ Social determinants are often complex and indirect in their effects with intervening factors (panel 2), and are a substantial challenge to delineation of causal relationships. For example, policing practices can affect the extent to which IDUs engage in protective behaviours in many ways, from reducing their ability to avoid use of contaminated injection equipment or access to needle and syringe programmes, to increasing the likelihood of injection in locations with a high prevalence of HIV (eg, shooting galleries or prison).

Theoretical models of HIV risk environments need to shift from heuristics that list factors (panel 1) to those that model interactions, processes, and pathways (figure 1). Furthermore, an emphasis on protective factors such as resilience, social cohesion, and solidarity is needed for individuals, communities, and populations. Because present epidemiological studies mainly focus on determinants of risk behaviour, rather than HIV infection, there is a need to understand the links between downstream or proximate factors and upstream or more distal influences on behaviour. Interventions on the basis of observational studies that are too narrow or superficial in scope tend not to address social contexts that shape behaviour, and therefore might be unable to create enabling environments that preserve and promote the protective resilience that can prevent the emergence of HIV epidemics.

A few studies were exceptions to the rule, and can serve as models for future studies. In an ecological study of 96 metropolitan areas in the USA, Friedman and colleagues⁷⁵ reported that three measures of legal repressiveness (arrests for possession of cocaine or heroin, police staff per capita, and costs for corrections) were independently associated with population HIV prevalence. Innovative use of administrative data allowed the investigators to change the unit of analysis from the individual to the community, allowing simultaneous study of the effect of macrosocial and macroeconomic influences on HIV prevalence. Wood and colleagues³⁰ prospectively examined the effects of reduction of plasma HIV-1 viral load in a community on HIV incidence in IDUs. In this study, undetectable rates of HIV-1 viral load in the community were a marker of antiretroviral therapy coverage, which permitted causal inferences to be made about the effect of policies of universal access to antiretroviral therapy on risk of HIV acquisition for the individual.

Review limitations

Other eligible reports could have been selected with different keywords, and we might not have included all studies that explored the effect of risk environments on high-risk behaviours. Because we explicitly focused on the epidemiology of HIV risk environments, our review did not include empirical or theoretical social science studies on the social, economic, and structural determinants of HIV and health inequality. A well-established body of published work from social science exists on the political economy of HIV and drug-related health.¹¹

The environmental risk determinants summarised in panel 1 are inevitably restricted in scope to those factors that epidemiological research has previously operationalised and found significant. Epidemiological research and subsequent reviews should build on social science to hypothesise and test other social, political, and economic influences that indirectly affect the risk of HIV acquisition through many, presently not delineated, pathways.

Another limitation is that models of risk environment draw distinctions between microenvironmental and macroenvironmental factors, and forms of risk environment (physical, social, economic, political). Analytically, these definitions are often somewhat overlapping of each other in practice. Police practices can be classified as a microenvironmental effect in some studies, affecting social environment, or a macroenvironmental effect of law enforcement, policy, and other cultural practices.

Modelling of the HIV risk environment

Our modelling scenarios confirmed the importance of structural change on the projected HIV epidemic trajectories in Odessa, Karachi, and Nairobi in the next 5 years. Since coverage of antiretroviral therapy for IDUs in these cities is very low (<1%), the benefits of providing optimal therapy for those with CD4-cell counts lower than 350 cells per μL are compelling, and could avert 40–50% of incident HIV infections in Odessa and Karachi. Implementation of widespread coverage of antiretroviral therapy to IDUs when clinically indicated to avert immediate HIV-related deaths is a clinical and moral imperative. Our results and other findings^30,132 suggest that antiretroviral therapy can contribute to subsequent reductions in HIV transmission. However, the resources needed to meet such targets, including voluntary HIV testing, efficient linkage and retention in care, CD4 diagnostics, adherence counselling, and resistance monitoring are substantial.

Our results support the well-established role of sterile syringe provision and opioid substitution as cornerstones of HIV prevention in IDUs. However, modelling results suggest that epidemic context alters their potential effect. Our findings from Odessa, Karachi, and Nairobi strongly advocate simultaneous combination of opioid substitution, needle and syringe programmes, and antiretroviral therapy scale up wherever possible, since their effects are synergistic.¹³² An important outcome from the Nairobi models is that, in an epidemic in which the force of infection is great (eg, an outbreak), the approximate 50% effectiveness of needle and syringe programmes and opioid substitution (see webappendix p 11) restricts effect as much as does coverage. This finding implies that investment in strategies to optimise intervention effectiveness is needed as coverage increases. Straightforward incremental scale-up will not achieve intended direct and indirect effects if interventions are not tailored, intensified, and otherwise adapted to address local epidemic dynamics.

Data from our review of the published work^{19,27,56,57,65,80,82} and our modelling suggest that opioid substitution scale-up should be an intervention priority. Methadone and buprenorphine maintenance have been added to WHO’s essential drug list, but have yet to be implemented widely in Kenya, Ukraine, and Pakistan, with potentially disastrous results. Access to therapeutic opioid medications is regulated by specific provisions in UN conventions, and these treaties recognise them as indispensable for medical care.¹³⁴ Despite this benefit, treatment for dependency with opiates is not available in Kenya, where the amount of morphine available for palliative care was 0.14 mg per head in 2004—less than 2.5% of the global mean of 5.9 mg per head.¹³⁵ One hospice physician stated “Physicians are afraid of morphine...Doctors [in Kenya] are so used to patients dying in pain...they think that this is how you must die. They are suspicious if you don’t die this way—[and feel] that you died prematurely.”¹³⁶ These reports suggest a substantial disconnect between drug-control authorities and the health-care system.¹³⁷ Similar to many other lowincome and middle-income countries, national drug control authorities in Kenya have concentrated more on control of illicit use than on facilitation of legitimate medical use.¹³⁸ WHO’s guidelines for assessment of national opioid control policies adopted in 2000 are anchored in a four-pillar approach aimed at achievement of a balance so drug control does not interfere with medical availability of opioid analgesics for pain relief and opioid substitution.¹³⁹ Our model suggests that legislative changes to permit opioid substitution in IDUs in Nairobi could prevent up to 14% of new HIV infections in the next 5 years, assuming 80% coverage.

Apart from the need for enhanced HIV and drug addiction prevention and treatment programmes and policies in Karachi, HIV epidemic projections advocate prevention of the transition from use of non-injection drugs to injection drugs. In Pakistan, where the number of people who smoke or inhale heroin is substantial,¹⁰⁷ macroenvironmental changes in drug trafficking routes that change the availability, purity, or price of heroin can lead to increased transitions from non-injecting to injecting drug use, as was reported after the terrorist attacks of Sept 11, 2001.¹⁴⁰

Our modelling scenarios suggest that an 8–12% increase in transition from non-injection drug use to injection drug use, even in the short-term, could lead to around a 65–98% rise in HIV incidence in Karachi in the next 5 years. Such effects are potentiated in the absence of opioid substitution, which could otherwise have a substantial benefit via reduction of the frequency and duration of injections. These findings suggest the need to invest in interventions to prevent transitions to injection and monitoring of perceived changes in local drug markets that might serve as harbingers of changes in injection behaviours.

Modelling limitations

Our modelling analysis allowed a direct quantification of the uncertainty in projections, leveraging substantial information from a wide range of sources. We showed patterns and trends in common with previous work,^112–114 and the explosive HIV epidemics occurring in Vancouver, Canada and Kathmandu, Nepal despite the existence of needle and syringe programmes.^141,142 Estimates of HIV prevalence in IDUs are a good source of epidemiological information, allowing model projections for Odessa and Karachi to be based on direct observational data.

However, our modelling results are limited by the need to make detailed assumptions about forms of behaviour that are highly stigmatised in a hard-to-reach population. Some of the most important model variables, such as IDU population size and the frequency with which IDUs were exposed to potentially contaminated syringes were imprecisely estimated. Reporting of highly stigmatised behaviours can be unreliable. If individuals under-report injecting behaviour, investigators can overestimate the rate of spread of infection or degree of epidemic saturation. If injecting behaviour is under-reported compared with sexual behaviour, the model could overestimate the contribution of sex in epidemics. There tend to be few estimates of HIV prevalence at the start of the epidemic, so the time and speed of initial spread cannot be known precisely. This uncertainty leads to substantial doubt about model projections for future courses of local HIV epidemics, especially because the link between risk behaviours and HIV acquisition is complex, and warrants caution in interpretation of our findings.¹¹⁴

Our models used a straightforward representation of opioid substitution and needle and syringe programmes; the effects of opioid substitution on non-HIV mortality and individuals moving in and out of programmes were not captured here, although they are addressed elsewhere in this Series.¹³³ The deterministic framework used in our models constrained characterisation of the network of contacts between IDUs, and might have influenced our results. If a more complex network topology were represented, for example with distinction of isolated pockets of high-risk behaviour within the IDU population, then the modelled spread of infection might have been more gradual in Karachi, compared with the explosive spread suggested by our model, and therefore the predicted effect of interventions would be reduced. The Nairobi model assumed a homogeneous IDU population that, although not unreasonable on the basis of available reports from outreach groups, could have led to overestimation of the potential for epidemic growth in this population.

Survey data capturing intersecting types of risk in IDU populations and links between IDUs and other at-risk groups, and improved measurement of HIV prevalence, will be most valuable for development of our understanding of epidemics, and modelling and assessment of the effects of interventions.