The results from the HTPN 052 trial reported in August 2011 demonstrated under the controlled conditions of a well-conducted clinical trial that early antiretroviral therapy (ART) can be highly effective in preventing transmission of HIV in stable heterosexual HIV-discordant couples 
. Several experimental studies are currently underway or planned to investigate the effectiveness of HIV treatment as prevention (TasP) in general populations, including in HIV hyperendemic communities in sub-Saharan Africa 
. A few mathematical modeling studies have predicted the cost-effectiveness of TasP, using cost estimates derived from currently existing ART programs 
. Meyer-Rath and Over review prior studies of ART costs, and discuss the cost assumptions used in economic evaluations of HIV treatment 
. They find that economic evaluations of TasP have tended toward a simplified accounting for variation in ART costs across patients and settings, focusing on a limited set of factors such as regimen or disease stage. Meyer-Rath and Over argue that future economic evaluation should account for a range of other factors that may be significant determinants of ART costs, including scale and scope of delivery, health states, ART regimens, health workers' experience, patients' time on treatment, and the distribution of delivery across public and private sectors.
In making this argument, Meyer-Rath and Over distinguish between two categories of ART cost functions 
: “cost accounting identities,” which generate estimates of total costs based on mathematical representations of the production process, and “flexible cost functions,” which generate estimates of total costs based on empirically derived relationships between costs and other factors, while treating the details of the production process as a “black box” (Text S1 of 
). Meyer-Rath and Over find that that “[m]ost existing [ART] cost projections assume a single constant unit cost per patient-year, or per patient-year on a certain regimen,” while a few have allowed for variation of costs by disease stage but not by other factors 
. Concerns with the level of detail in modeling the costs of TasP derive in part from the past focus on predictive, or ex ante
, economic evaluations, which rely heavily on mathematical or statistical models to extrapolate from limited empirical observations (as opposed to ex post
evaluations, which use direct observation of actual costs and benefits) 
It may indeed be ideal to capture the dependence of costs on many factors in economic evaluation of TasP—a task that could theoretically be achieved either by improving our understanding of the production process or through empirical examination of relationships between costs and other factors. However, the necessary data on the ART production process or on the relationship between ART costs and factors such as the scope of delivery or patients' time on treatment are currently largely lacking and may not become widely available for most settings in the near future, despite ongoing studies that will generate such data for a few settings. The absence of empirical data raises the question whether economic evaluation of TasP can be “good enough” without accounting for the dependence of ART costs on many of the factors that Meyer-Rath and Over argue convincingly could be determinants of ART costs.
The answer to this question will depend on both evaluation purpose and perspective. If the purpose is to decide whether or not to implement TasP, less detailed cost functions may be sufficient, because the result will be a yes/no answer indicating whether TasP produces a net benefit to society, or falls below some predetermined cost-effectiveness threshold. Such a result may be relatively robust to imprecision in the specification and estimation of costs. If, on the other hand, the purpose of the evaluation is to establish the most efficient approach to deliver TasP, given that it has been decided that it should be implemented, it will be crucial for the analysis to capture cost variations based on factors such as health worker–to–patient ratio, size and type of health care facility, and the level of integration of TasP programs into the general health care system.
The example Meyer-Rath and Over calculate in their article is a case in point. Based on theoretical considerations of economies of scale and empirical observation of scale effects in most industries, including in the delivery of HIV prevention services 
, they argue that it is unlikely that average costs would remain constant across scale of ART delivery. To demonstrate the potential impact of scale effects on costs, they adjust the estimates for implementing TasP in South Africa produced by Granich et al. 
“for scale and a plausible pattern of distribution of patients into clinics” 
. The result of this adjustment is an increase in total accumulated cost over 40 years from US$75 billion to US$106 billion. While this difference in cost estimates is large, given the dramatic effect of TasP estimated by Granich et al. 
, it may not alter the overall conclusion that TasP is a socially worthwhile intervention. In particular, if TasP could indeed eliminate HIV incidence, as Granich et al. 
assert based on their modeling results, the economic case for TasP would likely be robust to large increases in the cost estimates. At the same time, cost increases of the magnitude calculated by Meyer-Rath and Over would be extremely important for the practical exercises of financial planning and budgeting for TasP.
Whether less detailed cost functions will suffice in a given situation is also affected by the evaluation perspective. For instance, from the perspective of national health policy makers, economic evaluation may not need to account for the relationship between ART costs and the sector of delivery because policy makers may not be concerned about patients who utilize ART in the private sector. Conversely, from the perspective of for-profit companies providing ART in workplace HIV treatment programs, only the private sector costs will be relevant. In such cases, cost functions that account for differences in public versus private sector costs are not required.
Meyer-Rath and Over start the important discussion of which factors to include in cost functions in economic evaluations of TasP. We argue that for particular evaluation purposes (e.g., to establish the social value of TasP) and from particular perspectives (e.g., national health policy makers) “undetailed” cost functions, which do not capture cost dependence on many factors, may be sufficient.