Blood products are essential components of any healthcare system, and their safety, in terms of being free of transfusion-transmittable infections, is crucial. While the Food and Drug Administration (FDA) in the United States requires all blood donations to be tested for certain infection types, it does not dictate which particular tests should be used by blood centers. Multiple FDA-licensed blood screening tests are available for each infection type, and screening tests are imperfectly reliable and have different costs. In addition, infection prevalence rates within the donor population are uncertain for both emerging and established infection types. In this setting, the budget-constrained blood center’s objective is to devise a “robust” postdonation bloodscreening scheme that minimizes the risk of an infectious donation being released into the blood supply. Toward this goal, we study the minimization of the transfusion-transmittable infection risk considering regret- and expectation-based objectives, and we characterize structural properties of their optimal solutions. This allows us to gain insight, derive the price of robustness, and develop efficient algorithms. The proposed robust solution lowers the expected infection risk over various FDA-compliant testing schemes as well as the expectation-based scheme under forecast error. These findings have important public policy implications. The e-companion is available at https://doi.org/10.1287/opre.2017.1658 .

European law now requires AI to be explainable in the context of adverse decisions affecting the European Union (EU) citizens. At the same time, we expect increasing instances of AI failure as it operates on imperfect data. This paper puts forward a neurally inspired theoretical framework called “decision stacks” that can provide a way forward in research to develop Explainable Artificial Intelligence (X-AI). By leveraging findings from the finest memory systems in biological brains, the decision stack framework operationalizes the definition of explainability. It then proposes a test that can potentially reveal how a given AI decision was made.

We study the design of large-scale group testing schemes under a heterogeneous population (i.e., subjects with potentially different risk) and with the availability of multiple tests. The objective is to classify the population as positive or negative for a given binary characteristic (e.g., the presence of an infectious disease) as efficiently and accurately as possible. Our approach examines components often neglected in the literature, such as the dependence of testing cost on the group size and the possibility of no testing, which are especially relevant within a heterogeneous setting. By developing key structural properties of the resulting optimization problem, we are able to reduce it to a network flow problem under a specific, yet not too restrictive, objective function. We then provide results that facilitate the construction of the resulting graph and finally provide a polynomial time algorithm. Our case study, on the screening of HIV in the United States, demonstrates the substantial benefits of the proposed approach over conventional screening methods. Summary of Contribution: This paper studies the problem of testing heterogeneous populations in groups in order to reduce costs and hence allow for the use of more efficient tests for high-risk groups. The resulting problem is a difficult combinatorial optimization problem that is NP-complete under a general objective. Using structural properties specific to our objective function, we show that the problem can be cast as a network flow problem and provide a polynomial time algorithm.

Blood products are vital to healthcare, and their safety, in terms of being free of transfusion-transmittable infections (TTIs), is crucial. Unlike serological assays that screen for antibodies and antigens, Nucleic Acid Testing (NAT) technology screens for genetic material, enabling an earlier detection of the infection. However, because NAT is more expensive than serological testing, most blood centers administer NAT to pools of blood samples from multiple donors. This reduces the testing cost, but also decreases the test's sensitivity due to a dilution effect. Further, first-time donors typically have higher prevalence rates than repeat donors, and therefore, a flexible, non-universal strategy which allows first-time donors to undergo more extensive screening may reduce the TTI risk in transfusion recipients. Since some donor characteristics are uncertain, we develop a chance-constrained model that determines the optimal NAT pool sizes for the various TTIs, considering both non-universal and universal (i.e., common testing for all donors) strategies, so as to minimize the TTI risk while remaining within the testing budget with a high probability. Our case study of the United States quantifies the risk reduction benefit of non-universal NAT testing schemes over current practice.