4.5 Discussion

Selecting a portfolio of blood safety interventions is a substantial challenge for blood collection agencies and regulatory bodies worldwide. Epidemiologic variations lead to changes in TTI risk across populations and over time, and the characteristics of available interventions change as economic conditions shift and new technologies become available. Thus, blood safety portfolios must be optimized for local conditions and reassessed periodically. Our framework allows us to identify the optimal portfolio from a set of deferral, testing, or modification interventions for any set of TTIs, and our methods for simplifying the problem make the solution of the model more tractable.

Our evaluation of interventions for WNV and Zika in the U.S. in 2017, 2018, and 2019 found that the optimal portfolio varied considerably by season, year, and geographic region. Our analysis was limited to whole blood donations. A similar analysis could be performed for apheresis platelet, plasma, or red blood cell component collections. Use of plasma pathogen inactivation was never optimal in our analysis, but our analysis considered only two of the many pathogens that could be inactivated, which include emerging TTIs and TTIs for which tests are not available. Inclusion of such benefits could make PI more attractive. Similarly, donor deferral was not optimal for any donor group in our analysis, but inclusion of more TTIs could make donor deferral more attractive. Importantly, our case study was a retrospective analysis of what would have been optimal given perfect information about prevalence. To best inform policymaking, our framework would need to be applied to projections of current and future risks instead of past risks.

Our framework is designed to identify the optimal portfolio of all interventions for all TTIs of concern in a given jurisdiction. However, a comprehensive analysis will require significant effort. Most blood safety cost-effectiveness analyses evaluate the downstream societal costs and QALY losses for a single TTI [9,15–17,37]. Evaluating all interventions will require a similar analysis for each TTI for which risk is impacted. As we demonstrated by comparing performance of state- and zip code-level donor groups, donor segmentation can impact performance of the optimal policy. Donor segmentation becomes more challenging when TTIs with different types of risk factors are considered in the same analysis. For example, geography and season are natural dimensions for segmenting donors based on WNV and Zika risk, but behavioral risk factors are far more relevant for HIV and hepatitis C virus, two major TTIs. Appropriate definition of donor groups, particularly when considering TTIs with different types of risk factors, is an important area for further research.

By identifying the portfolio that minimizes net present monetary costs, our framework is consistent with cost-effectiveness analysis, the most common means of evaluating blood safety interventions. Policymakers may have additional considerations, such as fixed intervention budgets or a limit on the acceptable level of risk. Such considerations could be incorporated into the model through additional constraints or modifications of the cost function. Many assumptions in our model are simplifications of reality. For instance, the model assumes independence across tests and modifications, while in reality tests for the same TTI may be correlated, particularly when they test for the same disease marker (e.g., MP-NAT and ID-NAT both detect viral DNA or RNA). Additionally, we assumed a single test for any given disease marker because multiple tests for the same disease marker are not routinely used in practice. If simultaneous use of such tests is under consideration, one could calculate the expected performance measures for a combination of correlated tests (e.g., ID-NAT + MP-NAT for the same disease) and treat this combination of two tests as a single test in our optimization framework. We also assumed that the risk reduction from multiple modifications is multiplicative, but a similar approach of representing two or more modifications as a single modification could be used in situations where this assumption is not acceptable.

Additionally, our framework assumes that the cost of replacing a donation does not depend on how many donations need to be replaced. However, the donation replacement cost may increase at higher deferral rates due to the difficulty of recruiting new donors once the regular donor pool is exhausted. Our framework could be modified to address such considerations but doing so may change the structural properties in ways that make solution of the model less tractable. Additionally, for MP-NAT testing, we evaluated a generic minipool test with parameters approximating those of the two high-throughput NAT platforms used in the United States. Future analyses could optimize minipool size by evaluating each candidate minipool separately based on the estimated cost, sensitivity, and specificity. Finally, while our framework is designed to maximize expected utility, it could be combined with methods such as robust and stochastic optimization to incorporate different objectives and utility functions.

Our optimization framework could be a valuable tool to support blood safety policymaking. In 2010, the Alliance of Blood Operators convened representatives of blood agencies and regulators from the United States, Canada, Australia, and Europe to establish the blood safety risk-based decision-making (RBDM) framework [13]. Our model could be used within the RBDM framework to elucidate the opportunity cost of all possible portfolios, allowing policymakers to optimize the overall blood safety portfolio instead of looking at each intervention in isolation. As shown in our case study, our model can be used to evaluate the effectiveness of tailoring the set of tests or modifications used for specific donor groups. In the past, the same tests or modifications were applied universally to all collected donations. Recently, however, Canada’s use of an algorithm to target West Nile virus testing and the United States’ policy of only testing first-time donors for Chagas disease demonstrate a growing appetite for tailoring tests and modifications to different donor groups [35,138]. Thus far the RBDM framework has been applied to one-time decisions. In the future, health systems could continually monitor and update their blood safety portfolio by developing a portfolio model of all diseases and interventions, integrating sophisticated disease surveillance models that estimate the risk of diseases across donor groups, and continually updating their model in light of new information. In this way, safety of the blood supply can be ensured now and in the future.