2.2 Methods

2.2.1 Model structure

Prior evaluations of blood safety policies have taken a cohort approach in which recipients with a transfusion-transmitted infection (TTI) are assumed to have the same general characteristics as all blood recipients [8,8,15,64]. However, the number and types of components to which a recipient is exposed influence risk for TTI. Those with high component exposure may differ from the “average” recipient in age, sex, or expected survival, which can influence the health economic consequences of transfusion transmission. Recognizing this, we developed a recipient-level microsimulation model.

Our model aggregated the blood center costs, adverse events, and lifetime health economic consequences of adverse events due to ZIKV transfusion transmission during 1 year under each policy. Lifetime societal costs due to TTZ, including productivity loss due to illness or death, and lifetime utility loss of quality-adjusted life-years (QALYs) were estimated and discounted in continuous time at 3% per year. True-positive and false-positive donations removed from the blood supply were also totaled for the year, along with an associated per-donation cost. We took a societal perspective and converted all costs to 2016 U.S. dollars. Parameters estimated from the literature are listed in Table 2.1 <[9,34,61,65–79]>.

Our model randomly sampled the estimated number of transfusion recipients in a year (5 million for the 50 states and 31,000 for Puerto Rico) from a representative database and determined outcomes for each recipient individually. Because a suitable data set was not available for U.S. recipients, we sampled characteristics—including age, sex, and number of each component received (red blood cells, platelets, and plasma)—from a large database of Swedish and Danish transfusion recipients [80] with similar characteristics to U.S. recipients [81]. Recipients’ expected survival was sampled from a distribution based on observed posttransfusion survival of U.S. donors [81] adjusted to reflect recipient age and component exposure (see supplemental methods). The National Blood Collection and Utilization Survey was used to determine the number of components prepared per donation collected after accounting for wastage [34]. We estimated the probability that a ZIKV-infectious component was transfused on the basis of test sensitivity and specificity and the rate of donations testing positive for ZIKV RNA measured in the donor population. The probability that a transfusion resulted in infection was based on component-specific estimates of transfusion transmissibility (Table 2.1).

Health consequences of transfusion transmission included asymptomatic infection, mild febrile illness, and Guillain−Barré syndrome. Guillain−Barré syndrome resulted in temporary symptoms, permanent disability, or premature death. Male recipients could transmit sexually. The likelihood of sexual transmission was calculated based on the age-dependent probability of penetrative sex and an assumed probability of sexual transmission. Infected sexual partners could have the same adverse events as recipients. Female recipients and sexual partners were assigned an age-dependent probability of being pregnant. The likelihood that maternal infection lead to congenital Zika syndrome was based on a prior analysis (Figure 7.1) [70].

The medical costs, productivity costs, and QALYs lost due to all adverse events were totaled. Uncertainty was estimated by probabilistic sensitivity analysis, during which the model was run 10,000 times with different input parameters sampled from distributions (Table 2.1). Linear regression meta-modeling was used to assess sensitivity to specific variables (see supplemental methods) [82].

2.2.2 Rate of ZIKV-infectious donations

Donations confirmed positive for ZIKV RNA were considered infectious whether they tested positive or negative for IgM. For Puerto Rico, the rate of ZIKV-positive donations was based on ID-NAT results from 3 April 2016 to 1 April 2017. To enable evaluation of seasonally targeted strategies, the 52-week period was divided into a 26-week season of high mosquito activity (April through September) and a 26-week season of low mosquito activity (October through March).

In the 50 states, implementation began in May 2016 and was completed in December 2016, after the epidemic had waned. Blood centers with a higher perceived risk for ZIKV-infectious donations implemented ID-NAT earlier than low-risk centers [51]. To derive a rate of ZIKV-infectious donations that included early testing but also reflected the quantity of donations typically screened in a year, we extended our period of analysis from 23 May 2016 to 4 November 2017.

To enable evaluation of targeted strategies, data on donations collected during local transmission in south Florida (August 2016 to June 2017) and the Texas Gulf Coast (December 2016 to August 2017) were separated. The remaining donations in the 50 states’ supply were divided between those collected from donors with a history of travel to an area of local transmission in the past year and those from donors without such a history. The number of donors with a travel history was estimated from the 1995 American Travel Survey [66], the only comprehensive survey of both domestic and international travel destinations for U.S. residents. The proportion of ZIKV-positive donations due to travel acquisition was estimated from surveillance data on the sources of Zika infections in the 50 states compiled by the Centers for Disease Control and Prevention [61].

2.2.3 Policies considered

Assessed policies differed in how they applied screening tests across donor subpopulations. Screening tests include ID-NAT and MP-NAT, where NAT is done on a sample pooled from 6 to 16 donations. We considered the following 5 policies for both Puerto Rico and the 50 states: no Zika NAT screening, universal ID-NAT, universal MP-NAT, a separate-inventory policy wherein components from donations intended to be transfused to women of childbearing age were screened with ID-NAT and others were not screened, and a similar separate-inventory policy wherein others were screened with MP-NAT.

For Puerto Rico, we also considered the following 3 seasonal strategies: MP-NAT in high mosquito season and no screening in low season, ID-NAT in high mosquito season and no screening in low season, and ID-NAT in high mosquito season and MP-NAT in low season.

For the 50 states, we also considered the following 4 strategies that targeted donors who resided in or had recently traveled to areas of known local transmission: ID-NAT in areas with a known local transmission only (location-adaptive), location-adaptive MP-NAT, ID-NAT for donors who traveled to or resided in an area with a known local transmission only (travel-adaptive), and travel-adaptive MP-NAT.

Areas with a known local transmission were defined geographically and temporally (see supplemental methods).

2.2.4 Techniques to improve model efficiency

Our Monte Carlo simulation model used 2 efficiency improvement techniques to generate more precise outcome estimation with less computation. First, common random numbers were used across all policies. Second, a conditional Monte Carlo approach was used for the primary analysis (see supplemental methods). In a secondary analysis, we simulated whether patients had TTZ using traditional Monte Carlo methods and compared the component exposure, age at transfusion, and baseline survival of recipients with TTZ to those of other recipients.

2.2.5 Determining cost-effectiveness

Routine disease-marker tests of donated blood have been shown to exceed the willingness-to-pay thresholds of $50 000 to $150 000 that are commonly used for health care interventions in the United States [7–9,83]. Because of this precedent, the history of transfusion-transmitted HIV and hepatitis C virus, and the normative idea that decision makers might be willing to make larger investments to prevent health system–acquired illness than illness acquired elsewhere, we chose an alternate cost-effectiveness threshold of $1 million per QALY gained, as proposed previously (53).

We include the cost-effectiveness ratio (CER) of universal ID-NAT versus no screening as a key result because the FDA mandated ID-NAT in 2016. We also include the incremental CER (ICER) of each intervention, which describes the cost per benefit compared with the nondominated alternative we considered that was next most effective. An intervention can be cost-effective compared with no screening (CER below willingness-to-pay threshold) while not being cost-effective compared with alternatives (ICER above willingness-to-pay threshold).

2.2.6 Two-year follow-up

Two years after the original pubication, I developed a follow-up analysis in which I estimated (1) the relationship between the rate of Zika-infectious donations and the rate of adverse outcomes due to TT-Zika in the 50 states without screening and (2) the 2018 cost-effectiveness of universal screening. For that analysis, I simulated 13 Zika-infectious donation rates between 0.01 and 10,000 per million donations, resampling parameters were resampled 10,000 times from distributions reflecting uncertainty. I estimated rates of mild febrile illness, congenital Zika syndrome, and Guillain-Barré syndrome cases in transfusion recipients and their sexual partners, linearly interpolating between the mean, 1st percentile, and 99th percentile of the rate of outcomes to generate a function mapping the rate of Zika-infectious donations and the rate of adverse outcomes. I also estimated the cost-effectiveness of universal MP-NAT and ID-NAT as compared to no screening in the 50 states based on the rate of Zika-infectious donations observed in 2018 [61].

The analytic code is available on request; all analysis was done using Python 2.7.5 (Python Software Foundation) or R, version 3.4.3 (R Foundation for Statistical Computing).