The analysis provides three main results. First, PrEP used prior to ART initiation can prevent infections in HIV-1 serodiscordant couples and, although the initial costs are high, they are substantially offset by reduced future ART costs among HIV-1–uninfected partners who remain uninfected. In some circumstances (e.g., with effectiveness of 80% and used in couples that remain at high-risk), PrEP could be cost-saving overall. Second, PrEP in serodiscordant couples could be as cost-effective as earlier initiation of ART (compared to existing practice) if PrEP has a sufficiently high effectiveness (>70%) and low cost of delivery (<40% annual cost of ART). If used in couples that remain at high risk, PrEP could be as cost-effective as earlier ART even if PrEP had effectiveness of ~40%. Third, in lower risk couples, earlier ART at CD4<500 may be the most cost-effective strategy, but, in couples that remain at high risk, PrEP and ART could be used together (PrEP in the uninfected individual prior to ART initiation for their HIV-1–infected partner) to deliver maximal benefit and best cost-effectiveness. We hope this might inform the choices that will be available for HIV prevention in couples. We note, however, that it is important that many other considerations besides cost-effectiveness should inform decision-making for HIV treatment initiation and provision of PrEP in couples, including equitable access and the preferences of the couples themselves.
The principal determinants of the eventual use of PrEP among stable serodiscordant couples will be PrEP effectiveness, relative costs of PrEP and ART delivery, and couples' sexual behaviour. Using the model, we have defined a “target product profile,” the cost and effectiveness level for PrEP at which its use in a couple would to be at least as cost-effective as starting ART earlier in couples with different patterns of behaviour. The model shows that, if couples risk behaviour is reduced through risk reduction counselling, and becomes more like the behaviours reported by the “partners in prevention” clinical trial couples, then earlier initiation of ART is probably a more cost-effective way to manage infection and prevent HIV-1 infection (i.e., keeping couples “alive and HIV free”), unless PrEP in “real world” settings is at least as effective as indicated in recent trials among couples 
. However, the model also shows that, in couples with risks similar to those recorded in observational studies (“more typical” behaviour assumptions 
), with a PrEP effectiveness similar to that observed in recent trials 
, and at a cost of delivery consistent with optimistic forecasts 
, PrEP use among the uninfected partner could be as cost-effective as earlier treatment, and even a cost-saving intervention in its own right. This outcome highlights how the behavioural profile of couples influences the potential utility of PrEP and illustrates the importance of maximizing efficiency by prioritizing interventions for highest risk couples. It also shows the need for further research into the behaviours of those in long-term serodiscordant couples, their responses to the counselling, and their preferences for these different forms of intervention, in order to develop responsive and appropriate programs.
We note that although the feasibility of delivering ART is proven, the feasibility of PrEP is unknown and currently being investigated, so the information available about each option is not equal. Nonetheless, this analysis does support that PrEP could become one reasonable option that couples in this situation can be offered. And greater choices in HIV prevention should be welcomed as this can lead to increased uptake of services and better protection overall.
These calculations should also inform decision making about investment in new technologies—for instance, by setting a limit on the cost for potential future longer-lasting PrEP formulations that may be more effective. All these considerations are, of course, influenced by the estimated cost of ART, which, through renegotiated drug supply contracts and task-shifting in clinics, might be expected to fall considerably in the coming years 
, which would tend to make earlier ART more cost-effective.
We have explored these trade-offs using a detailed mathematical model that is parameterized and calibrated with data from stable serodiscordant heterosexual couples in South Africa, which included information on the sources of infection for those acquiring HIV-1 (i.e., whether infected by their stable partner or another partner). However, these couples may have lower risks of infection than HIV-1 serodiscordant couples in the general population due to study eligibility criteria and their participation in intensive HIV-1 prevention counselling during a clinical trial. Nonetheless, PrEP delivery programs would require initial HIV testing, and ideally will promote and provide couples HIV counselling and testing, so knowledge of serostatus and condom use will likely increase as has been reported among HIV serodiscordant couples in other studies 
. Sensitivity analyses were used to explore how differences in sexual behaviour affected the results. The data available from the Partners in Prevention trial 
cannot fully specify the long-term behaviour of couples in the model because couples were only followed for a 2-y period, whereas the model tracks individuals over their adult lifetimes. The use of the extended time-horizon of the simulation enabled the analysis to reflect the cumulative risk of transmission/death and total costs, whereas a short-term approach would not indicate whether infections in couples are averted or just delayed and would not capture the full cost implications of different strategies (e.g., because life-years saved and ART costs may follow many years after initial PrEP costs). The choice of outcome measure depends upon the relative value placed on preventing death and preventing HIV infection. The QALY approach emphasizes reduced deaths whereas the “alive and HIV free” metric gives more weight to HIV infection, which would often be survived with treatment. Giving more weight to averted infections also helps to implicitly reflect reduced risk of onward HIV transmission.
If further data become available about the added clinical benefits to patients of ART initiation at higher CD4 cell counts rather than indicated in current national and international guidelines, then these should be used to update the model and revise this analysis. We also note that in the analysis the wider benefits of the intervention (or the cost of nonintervention), such as increased labour availability and economic growth, are not included in the calculations. Issues regarding the trade-offs between PrEP and ART for immediate clinical need, including the attendant ethical considerations, are important in the wider debate about resource allocation in HIV programs, but were not relevant here because we only investigated use of PrEP in couples after universal access to ART (at current national and international guidelines) has been achieved. Many countries aim to achieve this by 2015 
, but we recognize that realistically this may not be achieved until many years later 
Many simplifying assumptions were made in the model, including not representing any change in risk behaviours during ART or PrEP use (i.e., potential “risk compensation” from feeling less at risk due to PrEP or ART use), the long-term interaction between PrEP and ART effectiveness through selection of resistant strains of virus 
, or potential effects of sexually transmitted infections on the efficacy of ART or PrEP 
. The model, and the chosen outcome measures, also do not capture the external sexual partner network so that, for instance, it does not account for the possibility that an averted infection terminates a chain of further infections 
, including averted infections among children. This factor may be expected to influence the estimated impact of ART and PrEP similarly (although further work is required to examine this) because, while ART reduces transmission to the infected individual's other partners, PrEP reduces the chance of infection and the subsequent risk of onward transmission to their partners, including during the initial highly infectious phase 
. The model also does not reflect that the HIV-1 prevalence and infectiousness among external partners will be influenced by patterns of PrEP and ART use in the wider population 
. The impact of ART on the incidence of other diseases, particularly tuberculosis, was not explicitly captured and this could lead to an underestimation of the benefit of ART, although the CD4-level–specific mortality rates in untreated individuals and utility-weights in the QALY analysis should implicitly reflect the deterioration in health that is associated with advanced HIV-1 infection 
. Interpretation of the results is further complicated by key uncertainties in the estimates of the cost of PrEP, which is inevitable given that PrEP delivery programs have not yet been implemented. However, the analyses presented here reflect these uncertainties and it is reassuring that our assumptions for the annual cost of PrEP (and the ratio of PrEP to ART costs) are similar to those independently derived by Pretorius et al. 
. Although these results suggest that the use of PrEP in HIV-1 serodiscordant couples could be cost-effective and have a significant impact on HIV incidence for that group, there are still significant logistical challenges that are not captured in the model. The identification and retention of discordant couples in services varies from setting to setting, and has been shown to be particularly difficult in South Africa. In such settings the feasibility and cost of targeting discordant couples (and, in particular, couples in which the woman might be pregnant/trying to conceive) could make an intervention utilizing PrEP much more expensive. Lastly, although we hope that this model will assist in policy-making decisions, we recognize that other factors beyond effectiveness and cost will also influence the introduction of PrEP for certain groups.
This analyses focuses on heterosexual HIV-1 serodiscordant couples in sub-Saharan Africa but similar questions could be asked for other groups such as men who have sex with men (MSM) in Africa and elsewhere. Different behavioural, biological, and program parameter values would be required for analyses in these different high risk groups reflecting for example the much higher risk of transmission per unprotected sex act 
in MSM and the higher cost of treatment for MSM in developed countries. However, the same general principles would apply: the lower the cost and the higher the effectiveness of PrEP, the more likely it is that PrEP will be a cost-effective way to support serodiscordant couples.
In summary, PrEP might become a valuable addition to combination approaches for HIV-1 prevention among stable serodiscordant couples in sub-Saharan Africa, in conjunction with ART. If PrEP is used by individuals that remain at high risk of infection prior to a partner's ART initiation, the additional cost per infection averted might be smaller than previously anticipated or the intervention could even be cost-saving, and the use of PrEP could be as cost-effective as earlier ART initiation. However, this outcome completely relies on the delivery cost of PrEP meeting current forecasts, and the “real-world” effectiveness of PrEP in couples being comparable to that found in the clinical trial 
: if adherence to PrEP outside of trials is lower, or if PrEP is more expensive to deliver than expected, PrEP could be much less cost-effective. It is vital to understand these trade-offs as soon as possible so that programmatic decision making and implementation can quickly proceed.