The rate of new HIV infections has stabilized in recent years (2.7 million infections in 2007 ), and concomitantly, the global number of those infected on antiretroviral therapy (ART) has increased dramatically . Despite this, the rate of new infections in developing countries still outpaces the rate at which individuals are started on treatment , and there is growing concern that this situation is unsustainable [3,4]. Incidence must be further reduced, but disappointingly, few HIV interventions have been shown to be effective in randomized controlled trials in developing countries: behaviour changes following counselling and testing are likely to have a minimal effect or even increase incidence [5,6]; two models of peer education for promoting reductions in risk behaviour have failed [7,8]; risk compensation and low adherence potentially contributed to no effect being found in trials to prevent HIV infection through diaphragm use  and herpes treatment ; and, in the last year, another trial of herpes treatment showed no effect on the rate of HIV transmission from coinfected individuals . These results bring the tally of trials showing no efficacy in reducing HIV incidence to more than 30 . Male circumcision has been shown to reduce the risk of men acquiring infection [13–15], although it is understood that this will not be enough to eliminate HIV, even under the most optimistic conditions [16,17].
In contrast, scale-up of ART has substantially reduced mortality [2,17–19]. As the availability of treatment expanded, Montaner et al. proposed using treatment as an intervention to prevent infection, and Granich et al.  recently used a mathematical model to evaluate that argument. The model suggested that in a high prevalence setting, with an incidence of two per 100 person-years at risk, an intervention that tested everyone annually and initiated treatment immediately if they were infected (‘universal test-and-treat’ intervention) could reduce incidence to below one per 1000 person-years at risk, that is, more than a 95% reduction (described by the authors as elimination). The model predicted that despite high costs during the rollout phase, long-term ART costs would be much lower than the current strategy of treating on the basis of clinical need.
This result has stimulated extensive comment and sparked interest in rolling-out ‘test-and-treat’ type interventions [22–28]. However, it is worth further exploring the findings, as the model provided a limited representation of some aspects of HIV epidemiology and investigated only one type of intervention in one type of epidemiological context. In this article, we investigate the potential of alternate test-and-treat interventions under a range of contexts, and using a different mathematical model that incorporates updated information on the course of HIV infection and transmission rates on treatment along with a more explicit exploration of the potential role of heterogeneity in sexual risk behaviour [29,30].
An HIV transmission model was developed that was defined by a set of partial differential equations incorporating variation in sexual risk behaviour : changes in HIV transmissibility over the course of infection  and observed HIV survival rates from an African setting  (full technical details in online appendix, http://links.lww.com/QAD/A16). In the model, there are two subpopulations, with key parameters being the relative degree of risk behaviour for acquiring/transmitting infection between the two (π), the relative size of the lower risk subpopulation (θ), the value of basic reproductive number in that subpopulation (
) and the degree of sexual contact between individuals in the two subpopulations (ε). It was assumed that the relative size of the two risk groups remained constant over time .
We modelled different ‘test-and-treat’ interventions by altering the time since infection that treatment is started and assuming that individuals on treatment are, on an average, 13-fold less infectious than untreated individuals [30,34]. As patients tend to live longer on ART if they start treatment earlier [18,35], survival on treatment was related to the timing of initiation, with a maximum of 28 years overall survival if treatment is started within 1 year of infection (sensitivity analyses showed that the results could be reproduced assuming no relationship between survival and timing of ART initiation). The timing of treatment initiation was related to the interval between HIV tests and the expected average CD4 cell count at initiation. The trend in CD4 cell count over time since infection was calculated using information from a recent meta-analysis , showing mean time from infection to a level of 200 cells/μl of 7.6 years, and other observational data of the rate of CD4 cell count decline during earlier HIV infection among African populations .
The impact of the intervention was evaluated in different types of populations in which the epidemic was sustained in both groups or only the higher risk group (
), and in which mixing between the two groups was extensive or limited (ε = 0.1 or ε = 0.9). In each case, we assumed that most of the population was at a lower risk and a small minority was at a higher risk, in accord with observational studies . The π parameter was then adjusted, so that the HIV incidence rate was exactly the same in each scenario: 1.5 per 100 person-years at risk before the intervention, which is typical of many countries in Eastern and Southern Africa currently . Thus, the three scenarios were more homogenous distribution of risk (
); heterogeneous risk distribution with random mixing (
); and heterogeneous risk distribution with assortative mixing (i.e. most sex contacts are between individuals in the same subpopulation) (
It was assumed that the sensitivity and specificity of the HIV test was 100%.
We quantify the impact of the intervention as the reduction in incidence at equilibrium following its introduction compared with the preintervention incidence rate. Treatment load was calculated as the fraction of individuals in the population who are on treatment at equilibrium. Annual costs per capita of the intervention are approximated by summing the product of the number of individuals on treatment and $800, and the product of the annual number of HIV tests and $10 (J.G. Kahn, personal communication), assuming a fixed population size. A cost-efficiency measure was calculated as the reduction in equilibrium incidence divided by annual costs per capita.
Our results broadly confirm the main findings of Montaner et al. and Granich et al.: treatment has the potential to substantially reduce HIV transmission. Figure 1 shows the eventual predicted impact of test-and-treat interventions under three different epidemiological contexts. The contour lines indicate the reduction in incidence for an assumed level of coverage (vertical axis) and treatment start time (horizontal axis). Two other horizontal axes show how this treatment start time corresponds to the average interval between tests required to initiate treatment at that time and the average CD4 cell count (cells/μl) among patients starting treatment. As expected, the impact is greater with higher coverage levels and earlier initiation of treatment (top-left corner of each panel). However, we also find that the expected impact varies depending on the epidemiological context assumed. In a population with a more even risk distribution (‘scenario A’), testing 80% of the population every 2 years and treating immediately is expected to reduce incidence by more than 95% (Fig. 1a). The same intervention in a population with greater variation in risk (‘scenario B’) generates a smaller impact, reducing incidence by approximately 85% and fails to reduce the epidemic to the level termed ‘elimination’  (Fig. 1b). In this scenario, all individuals would need to be treated within 1 month of infection for incidence to be reduced by 95%. If more partnerships are formed between individuals in the same subpopulation (‘scenario C’), incidence is reduced by approximately 60% if individuals start treatment 1 year after infection (Fig. 1c). The reason is that in scenarios B and C, transmission is more dependent on a few individuals who spread infection rapidly once infected.
Our analysis shows that, depending on the epidemiological context, similar reductions in HIV incidence could be generated by less ambitious interventions. For instance, in a population with little variation in risk behaviour and random mixing (‘scenario A’), incidence is still reduced by 95% if 80% of the population is tested only every 3–4 years, corresponding to a mean CD4 cell count at initiation of 400 cells/μl.
In each case, it would take approximately 30 years for these reductions in incidence to be fully realized, and there is the potential for incidence rates to rebound, as the first cohorts starting treatment progress to AIDS (see Figure 2 in the technical appendix, http://links.lww.com/QAD/A16).
We investigated the impact and approximate costs of implementing the test-and-treat intervention, in which the intervals between HIV tests ranged between 1 and 20 years (Fig. 2). Although shorter intervals between tests lead to greater reductions in incidence, the convex shape of the curves indicates a ‘diminishing returns’ relationship (Fig. 2a). The sharp upturn in the impact if individuals are tested more frequently than every 6 months reflects treatment interrupting the period of primary infection when individuals are highly infectious.
The numbers on treatment at equilibrium and the numbers of tests each year for alternative test-and-treat strategies are shown in Fig. 2b and c. For 80% coverage, the treatment load increases as the interval between testing is reduced from 15 to 10 years, as more treatment is provided to those in late-stage disease, without an associated large effect on HIV transmission. Further decreases in the interval between testing from every 10 to 1 years lead to lower treatment loads, as, in this phase, ART is directly reducing the endemic level of HIV and treatment needs. In contrast, if the test-and-treat intervention is scaled-up to only 30 or 50% of the population, more testing gives greater years of treatment per person without attracting large reductions in incidence, so the ART load only increases.
An approximate indication of cost-efficiency is presented in Fig. 3. The highest parts of the curves correspond to the test-and-treat strategy that generates the greatest reduction in incidence per unit cost. The optimal position varies according to the epidemiological context, the level of coverage achieved and the relative costs of treatment and testing. For more fragile epidemics (‘scenario A’, Fig. 3a), the optimal strategy is testing every 4–5 years or initiating treatment at a CD4 cell count of 350–400 cells/μl. Here, the position of the optimum at 80% coverage is determined largely by the frequency of testing that is necessary for HIV elimination.
For the most robust epidemic (‘scenario C’, (Fig. 3b), the optimal strategy (at 80% coverage) is testing every 2–3 years, corresponding to treatment at a CD4 cell count above 450 cells/μl. In this scenario, the optimum is determined by a balance between providing ART for longer duration and the greater reductions in incidence, as the testing interval decreases.
For scenario B, the optimum is 1–2 years, the suggested strategy of Granich et al. . Here, equilibrium treatment loads do not change substantially with testing frequency, as the increased duration of treatment is almost exactly offset by concomitant reductions in incidence; so, the position of the optimum is mainly determined by the relative costs of treatment and testing.
It is important that our modelling approach (incorporating recent estimates of transmission in acute infection, transmission rates on treatment and heterogeneity in sexual risk behaviour) has partially confirmed the finding that earlier initiation of ART can lead to substantial reductions in HIV transmission [20,21]. However, our analysis has highlighted three important aspects of ‘test-and-treat’ interventions that should be carefully considered as plans for implementing such an intervention move forward. First, the impact of the intervention depends crucially on the epidemiological context: under some circumstances, we find the effect to be as large as estimated by Granich et al., but in others, the effect is much less. The context is proximally determined by many properties of the sex partner network (such as heterogeneity, concurrency and mixing [38,39]), and it is not easy to determine which of the modelled contexts (A, B or C) is most similar to a particular population. Thus, earlier model assumptions  accurately represent a particular population, even if the prevalence/incidence level appears similar (i.e. the fit of a model to data does not imply that the model is validated). The uncertainty in the specification of the epidemiological context can be reduced by incorporating local behavioural data into more detailed models of HIV transmission , but the remaining variance in projections should be fully reflected in cost-effectiveness calculations. It is likely that the ‘test-and-treat’ approach is much better suited to some populations and poorly suited to others.
The second main finding is that although increasing the frequency of testing does lead to a larger reduction in HIV transmission, there are diminishing returns for increasing testing frequency to the once-per-year levels proposed in the study by Granich et al.. Under some situations, much later initiation can still stall the epidemic. Granich et al. suggest that testing every year and treating immediately was an effective and cost-saving strategy compared with later initiation, but, in our model, the most cost-efficient strategy could be testing everyone 3–5 years, depending on the epidemiological context and the coverage achieved (Fig. 3). However, the position of the optimum is highly sensitive to aspects of the epidemic context, life expectancy on treatment and relative costs of treatment and testing, especially when interventions do not reach universal coverage, making it difficult to formulate firm recommendations without further information and specification.
Our third main finding was that although a high coverage implementation of test and treat could lead to reductions in incidence and ART use, failing to achieve sufficiently high coverage levels or failing to test frequently enough could just lead to a dramatic spiralling of treatment costs. In this scenario, the intervention does not interrupt transmission, so the pool of those developing treatment needs continues to grow. It is essential that this eventuality is avoided, especially in many countries where healthcare systems already struggle to provide care for HIV-infected patients in clinical need. Losses to follow-up, imperfect adherence or the evolution of resistance [41,42] could all contribute to reducing the effective coverage of the programme.
Our analysis was intended to provide qualitative insights rather than precise quantitative predictions about the effect of test-and-treat interventions, and we have not considered the logistical challenges presented by implementing such an intervention. Thus, our estimates of cost and cost-efficiency analysis are simplified, and our consideration of the epidemic is most focused on the equilibrium incidence level. The impact and costs of the rollout phase of interventions is, therefore, not fully captured in all our analyses. Also, our modelled costs are not sensitive to scale, as they can be in practice  (e.g. through clinicians' time being occupied with testing rather than other activities), and we have not explicitly considered the increased chance of adverse events, toxicities and viral evolution and need for second-line therapies associated with long-term use of ART [42,44]. To reinforce our main findings in this analysis, we have not included how the chance of complying with repeat testing and treatment can vary according to how frequently tests are offered, nor how sexual behaviour can change as a result of learning one's serostatus where HIV is negative or positive [45,46]. This could mean that our estimates of costs for testing very frequently are underestimated, and if these were reflected in our analysis, the optimum test-and-treat strategy would be longer intervals between the tests. However, we have also not quantified the extra life-years saved associated with earlier treatment initiation nor how the chance of stopping treatment or developing resistance and moving to second-line therapies could depend on the timing of initiation [18,35], which may favour more frequent testing. All these factors interact with the generalizable issues we have highlighted in this study and will demand attention in further modelling work tailored to specific settings. The model does not explicitly capture the real variation in the risk of transmission between different sexual partnerships due to the frequency of sex, presence of sexually transmitted infections and condom use, and although this detail is not expected to affect the findings presented here, incorporating these factors in further models will improve the specification of the epidemiological context and precision of the projections.
The impact of many interventions, including test and treat, can be amplified by targeting those that are most at risk of acquiring and transmitting infections . Therefore, in the generalized epidemics in southern Africa, testing women in beer halls  and truck drivers  most often might improve impact and cost-efficiency. The extent of that amplification also depends on the epidemiological context (especially extent of contact between higher and lower risk individuals ), so that the advantage of targeting would be modest in some situations (e.g. scenario A) but great in others (e.g. scenario C). Combining interventions can also lead to synergies, so that applying two interventions together leads to increases in effectiveness . Therefore, the opportunity to counsel those testing HIV-negative, promoting behaviour change to reduce the risk of infection [5,46], should not be missed.
Leveraging the infrastructure and capacity that have so rapidly grown-up to support the expansion of ART programmes in Africa to also reduce HIV transmission is a promising strategy. However, by failing to capture some important features of HIV epidemiology, overoptimistic projections can be generated. It is also essential to recognize that testing every year is not necessarily the most cost-efficient strategy, and that failing to fully implement the test-and-treat strategy could perversely lead to overall increased long-term ART costs.
For funding support, T.B.H. thanks The Wellcome Trust and G.P.G. thanks MRC and UNAIDS for funding support.
We declare that no conflicts of interest exist.
3. Piot P, Bartos M, Larson H, Zewdie D, Mane P. Coming to terms with complexity: a call to action for HIV prevention. Lancet 2008; 372:845–859.
4. Potts M, Halperin DT, Kirby D, Swidler A, Marseille E, Klausner JD, et al
. Public health. Reassessing HIV prevention. Science 2008; 320:749–750.
5. Sherr L, Lopman B, Kakowa M, Dube S, Chawira G, Nyamukapa C, et al
. Voluntary counselling and testing: uptake, impact on sexual behaviour, and HIV incidence in a rural Zimbabwean cohort. AIDS 2007; 21:851–860.
6. Corbett EL, Makamure B, Cheung YB, Dauya E, Matambo R, Bandason T, et al
. HIV incidence during a cluster-randomized trial of two strategies providing voluntary counselling and testing at the workplace, Zimbabwe. AIDS 2007; 21:483–489.
7. Gregson S, Adamson S, Papaya S, Mundondo J, Nyamukapa CA, Mason PR, et al
. Impact and process evaluation of integrated community and clinic-based HIV-1 control: a cluster-randomised trial in eastern Zimbabwe. PLoS Med 2007; 4:e102.
8. Pequegnat W; The NIMH Collaborative HIV/STD Prevention Trial Group. Results of the RCT to test the community popular opinion leader (C-POL) intervention in five countries [LBPE1166]
. In: XVII International AIDS Conference
; 3–8 August 2008; Mexico City, Mexico.
9. Padian NS, van der Straten A, Ramjee G, Chipato T, de Bruyn G, Blanchard K, et al
. Diaphragm and lubricant gel for prevention of HIV acquisition in southern African women: a randomised controlled trial. Lancet 2007; 370:251–261.
10. Watson-Jones D, Weiss HA, Rusizoka M, Changalucha J, Baisley K, Mugeye K, et al
. Effect of herpes simplex suppression on incidence of HIV among women in Tanzania. N Engl J Med 2008; 358:1560–1571.
11. University of Washington. Herpes medication does not reduce the risk of HIV transmission from individuals with HIV and genital herpes but demonstrates modest reduction in HIV disease progression and leads to new important insights about HIV transmission, UW-led international study finds.
Seattle, Washington: University of Washington; 2009. http://depts.washington.edi/hsnews
12. Weiss HA, Wasserheit JN, Barnabas RV, Hayes RJ, Abu-Raddad LJ. Persisting with prevention: the importance of adherence for HIV prevention. Emerg Themes Epidemiol 2008; 5:8.
13. Auvert B, Taljaard D, Lagarde E, Sobngwi-Tambekou J, Sitta R, Puren A. Randomized, controlled intervention trial of male circumcision for reduction of HIV infection risk: the ANRS 1265 trial. PLoS Med 2005; 2:e298.
14. Bailey RC, Moses S, Parker CB, Agot K, Maclean I, Krieger JN, et al
. Male circumcision for HIV prevention in young men in Kisumu, Kenya: a randomised controlled trial. Lancet 2007; 369:643–656.
15. Gray RH, Kigozi G, Serwadda D, Makumbi F, Watya S, Nalugoda F, et al
. Male circumcision for HIV prevention in men in Rakai, Uganda: a randomised trial. Lancet 2007; 369:657–666.
16. Hallett TB, Singh K, Smith JA, White RG, Abu-Raddad LJ, Garnett GP. Understanding the impact of male circumcision interventions on the spread of HIV in southern Africa. PLoS One 2008; 3:e2212.
17. UNAIDS/WHO/SACEMA Expert Group on Modelling the Impact Cost of Male Circumcision for HIV Prevention. Male circumcision for HIV prevention in high HIV prevalence settings: what can mathematical modelling contribute to informed decision making?PLoS Med
18. Braitstein P, Brinkhof MW, Dabis F, Schechter M, Boulle A, Miotti P, et al
. Mortality of HIV-1-infected patients in the first year of antiretroviral therapy: comparison between low-income and high-income countries. Lancet 2006; 367:817–824.
19. Jahn A, Floyd S, Crampin AC, Mwaungulu F, Mvula H, Munthali F, et al
. Population-level effect of HIV on adult mortality and early evidence of reversal after introduction of antiretroviral therapy in Malawi. Lancet 2008; 371:1603–1611.
20. Montaner JS, Hogg R, Wood E, Kerr T, Tyndall M, Levy AR, Harrigan PR. The case for expanding access to highly active antiretroviral therapy to curb the growth of the HIV epidemic. Lancet 2006; 368:531–536.
21. Granich RM, Gilks CF, Dye C, De Cock KM, Williams BG. Universal voluntary HIV testing with immediate antiretroviral therapy as a strategy for elimination of HIV transmission: a mathematical model. Lancet 2009; 373:48–57.
22. Garnett GP, Baggaley RF. Treating our way out of the HIV pandemic: could we, would we, should we? Lancet 2009; 373:9–11.
23. Cohen MS, Mastro TD, Cates W Jr. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1077, author reply 1080–1081.
24. Ruark A, Shelton JD, Halperin DT, Wawer MJ, Gray RH. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1078, author reply 1080–1081.
25. Epstein H. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1078–1079, author reply 1080–1081.
26. Jurgens R, Cohen J, Tarantola D, Heywood M, Carr R. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1079, author reply 1080–1081.
27. Hsieh YH, de Arazoza H. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1079–1080, author reply 1080–1081.
28. Assefa Y, Lera M. Universal voluntary HIV testing and immediate antiretroviral therapy. Lancet 2009; 373:1080, author reply 1080–1081.
29. Hollingsworth TD, Anderson RM, Fraser C. HIV-1 transmission, by stage of infection. J Infect Dis 2008; 198:687–693.
30. Attia S, Egger M, Muller M, Zwahlen M, Low N. Sexual transmission of HIV according to viral load and antiretroviral therapy: systematic review and meta-analysis. AIDS 2009; 23:1397–1404.
31. Gregson S, Nyamukapa CA, Garnett GP, Mason PR, Zhuwau T, Carael M, et al
. Sexual mixing patterns and sex-differentials in teenage exposure to HIV infection in rural Zimbabwe. Lancet 2002; 359:1896–1903.
32. Todd J, Glynn JR, Marston M, Lutalo T, Biraro S, Mwita W, et al
. Time from HIV seroconversion to death: a collaborative analysis of eight studies in six low and middle-income countries before highly active antiretroviral therapy. AIDS 2007; 21:S55–S63.
33. Walker PT, Hallett TB, White PJ, Garnett GP. Interpreting declines in HIV prevalence: impact of spatial aggregation and migration on expected declines in prevalence. Sex Transm Infect 2008; 84(Suppl 2):ii42–ii48.
34. Wilson DP. Data are lacking for quantifying HIV transmission risk in the presence of effective antiretroviral therapy. AIDS 2009; 23:1431–1433.
35. Sterne JA, May M, Costagliola D, de Wolf F, Phillips AN, Harris R, et al
. Timing of initiation of antiretroviral therapy in AIDS-free HIV-1-infected patients: a collaborative analysis of 18 HIV cohort studies. Lancet 2009; 373:1352–1363.
36. Eligibility for ART in lower income countries collaboration (eART-linc) collaboration. Duration from seroconversion to eligibility for antiretroviral therapy and from ART eligibility to death in adult HIV-infected patients from low and middle-income countries: collaborative analysis of prospective studies
. Sex Transm Infect
37. Laurent C, Bourgeois A, Faye MA, Mougnutou R, Seydi M, Gueye M, et al
. No difference in clinical progression between patients infected with the predominant human immunodeficiency virus type 1 circulating recombinant form (CRF) 02_AG strain and patients not infected with CRF02_AG, in Western and West-Central Africa: a four-year prospective multicenter study. J Infect Dis 2002; 186:486–492.
38. Morris M, Kretzschmar M. Concurrent partnerships and the spread of HIV. AIDS 1997; 11:641–648.
39. Ghani AC, Garnett GP. Risks of acquiring and transmitting sexually transmitted diseases in sexual partner networks. Sex Transm Dis 2000; 27:579–587.
40. Hallett TB, Gregson S, Gonese E, Mugurungi O, Garnett GP. Assessing evidence for behaviour change affecting the course of HIV epidemics: a new mathematical modelling approach and application to data from Zimbabwe. Epidemics 2009; 1:108–117.
41. Ndiaye B, Ould-Kaci K, Salleron J, Bataille P, Bonnevie F, Cochonat K, et al
. Characteristics of and outcomes in HIV-infected patients who return to care after loss to follow-up. AIDS 2009; 23:1786–1789.
42. Baggaley RF, Garnett GP, Ferguson NM. Modelling the impact of antiretroviral use in resource-poor settings. PLoS Med 2006; 3:e124.
43. Marseille E, Dandona L, Marshall N, Gaist P, Bautista-Arredondo S, Rollins B, et al
. HIV prevention costs and program scale: data from the PANCEA project in five low and middle-income countries. BMC Health Serv Res 2007; 7:108.
44. Montessori V, Press N, Harris M, Akagi L, Montaner JS. Adverse effects of antiretroviral therapy for HIV infection. CMAJ 2004; 170:229–238.
45. Cremin I, Nyamukapa C, Sherr L, Hallett TB, Chawira G, Cauchemez S, et al.Patterns of self-reported behaviour change associated with receiving voluntary counselling and testing in a longitudinal study from Manicaland, Zimbabwe
. AIDS Behav
2009. [Epub ahead of print]
46. Hallett TB, Dube S, Cremin I, Lopman B, Mahomva A, Ncube G, et al
. The role of testing and counselling for HIV prevention and care in the era of scaling-up antiretroviral therapy. Epidemics 2009; 1:77–82.
47. Garnett GP, Anderson RM. Strategies for limiting the spread of HIV in developing countries: conclusions based on studies of the transmission dynamics of the virus. J Acquir Immune Defic Syndr Hum Retrovirol 1995; 9:500–513.
48. Lewis JJ, Garnett GP, Mhlanga S, Nyamukapa CA, Donnelly CA, Gregson S. Beer halls as a focus for HIV prevention activities in rural Zimbabwe. Sex Transm Dis 2005; 32:364–369.
49. Ramjee G, Gouws E. Prevalence of HIV among truck drivers visiting sex workers in KwaZulu-Natal, South Africa. Sex Transm Dis 2002; 29:44–49.
50. Hallett T, Garnett G, Mupamberiyi Z, Gregson S. Measuring effectiveness in community randomized trials of HIV prevention. Int J Epidemiol 2008; 37:77–87.