RESEARCH ARTICLE Year : 2019  Volume : 10  Issue : 1  Page : 26 Development of a calculated panel reactive antibody web service with local frequencies for platelet transfusion refractoriness risk stratification William J Gordon^{1}, Layne Ainsworth^{2}, Samuel Aronson^{2}, Jane Baronas^{3}, Richard M Kaufman^{4}, Indira Guleria^{5}, Edgar L Milford^{5}, Michael Oates^{2}, Rory Dela Paz^{2}, Melissa Y Yeung^{5}, William J Lane^{4}, ^{1} Department of Medicine, Division of General Internal Medicine and Primary Care, Brigham and Women's Hospital; Department of Medicine, Division of General Internal Medicine, Massachusetts General Hospital; Harvard Medical School, Boston; Partners HealthCare, Somerville, MA, USA ^{2} Partners HealthCare, Somerville, MA, USA ^{3} Department of Pathology, Brigham and Women's Hospital, Boston, MA, USA ^{4} Harvard Medical School; Department of Pathology, Brigham and Women's Hospital, Boston, MA, USA ^{5} Harvard Medical School; Department of Medicine, Division of Renal Medicine, Brigham and Women's Hospital, Boston, MA, USA Correspondence Address: Background: Calculated panel reactive antibody (cPRA) scoring is used to assess whether platelet refractoriness is mediated by human leukocyte antigen (HLA) antibodies in the recipient. cPRA testing uses a national sample of US kidney donors to estimate the population frequency of HLA antigens, which may be different than HLA frequencies within local platelet inventories. We aimed to determine the impact on patient cPRA scores of using HLA frequencies derived from typing local platelet donations rather than national HLA frequencies. Methods: We built an opensource web service to calculate cPRA scores based on national frequencies or customderived frequencies. We calculated cPRA scores for every hematopoietic stem cell transplantation (HSCT) patient at our institution based on the United Network for Organ Sharing (UNOS) frequencies and local frequencies. We compared frequencies and correlations between the calculators, segmented by gender. Finally, we put all scores into three buckets (mild, moderate, and high sensitizations) and looked at intergroup movement. Results: 2531 patients that underwent HSCT at our institution had at least 1 antibody and were included in the analysis. Overall, the difference in medians between each group's UNOS cPRA and local cPRA was statistically significant, but highly correlated (UNOS vs. local total: 0.249 and 0.243, ρ = 0.994; UNOS vs. local female: 0.474 and 0.463, ρ = 0.987, UNOS vs. local male: 0.165 and 0.141, ρ = 0.996;P< 0.001 for all comparisons). The median difference between UNOS and cPRA scores for all patients was low (male: 0.014, interquartile range [IQR]: 0.004–0.029; female: 0.0013, IQR: 0.003–0.028). Placement of patients into three groups revealed little intergroup movement, with 2.96% (75/2531) of patients differentially classified. Conclusions: cPRA scores using local frequencies were modestly but significantly different than those obtained using national HLA frequencies. We released our software as open source, so other groups can calculate cPRA scores from national or customderived frequencies. Further investigation is needed to determine whether a localHLA frequency approach can improve outcomes in patients who are immunerefractory to platelets.
Introduction Platelet transfusions are an essential component of the clinical care for patients undergoing hematopoietic stem cell transplantation (HSCT).[1] Platelet refractoriness, or the insufficient rise in platelet count after repeated transfusions, is a common and challenging clinical problem, and can lead to adverse outcomes, including increased bleeding risk, decreased survival, and longer hospitalizations.[2] The etiology of platelet refractoriness includes nonimmunemediated factors such as medication effects, splenomegaly, or sepsis and immunemediated factors such as alloimmunization to human leukocyte antigen (HLA) or plateletreactive antibodies.[3] Despite leukoreduction, alloimmunization to HLA is still a common complication for patients undergoing HSCT.[4] After the realization that donorspecific HLA antibodies were associated with hyperacute rejection in renal transplant recipients, understanding which patients would be higher risk for HLAmediated rejection has been an important component of candidate prioritization in organ allocation algorithms. Panel reactive antibody (PRA) scoring was developed to determine the degree of HLA sensitization. A PRA value is derived using a panel of normal donors to represent the population frequency of HLA – the percentage of this panel of donors a patient has antibodies to is an individual's PRA score.[5],[6] To improve standardization and take advantage of advances in HLA screening technology, the United Network for Organ Sharing (UNOS) moved toward calculated PRA (cPRA) scores in 2007. cPRA uses a national sample of US kidney donors to estimate the population frequency of unacceptable HLA antigens. Recipient antibodies (and unacceptable antigens) are then compared to this national pool, which results in a cPRA.[6],[7],[8] Prior work has compared various cPRA and PRA assays in the setting of solid organ transplantation.[9],[10] However, platelets, unlike solid organs, express only Class I HLA antigens (HLAA, HLAB, and HLAC, though HLAC is not commonly tested).[11] Because of this, cPRA is often used in clinical practice to estimate the risk of HLAmediated platelet refractoriness. However, there are several potential limitations to using cPRA to predict platelet refractoriness. First, unlike renal transplantation where a donor can donate at most two organs, platelet donors can continuously redonate platelets; thus, the true HLA frequency pool is based not on the number of donors, but the number of units of platelets. Second, the UNOS HLA frequencies are national, but platelet donor pools tend to be regional, potentially leading to important differences in HLA antigen frequency between the two populations. Finally, the UNOS frequencies are renal donors, a different population than platelet donors. Because of the limitations of using cPRA to predict HLAincompatibility for platelet transfusion, we hypothesized that building a custom, local HLA frequency pool, where each platelet unit contributed equally (i.e., plateletdriven frequency as opposed to donordriven frequency), might provide more clinically useful cPRA scores for patients undergoing HSCT. To do this, we constructed a custom HLA frequency data set and compared cPRA scores of patients undergoing HSCT between UNOS and our local, custom calculator. Methods Calculated panel reactive antibody application programming interface UNOS maintains two methods for calculating cPRA. First, they host a manual calculator.[12] Users select all recipient antibodies and the website calculates a cPRA. Second, as of November 2016, UNOS maintains an application programming interface (API).[13] This API requires registration and does not currently allow for custom antigen frequencies. Because of our requirements, notably, a large number of HLA typed patients and the need to create a custom population frequency, we built a RESTful[14] web service [Figure 1]. This web service supports calculating cPRA using the existing UNOS frequencies, but also allows custom HLA frequencies to be imported and used for calculation. We have released this as an opensource project, so others can calculate cPRA based on their local population frequencies.[15] Our API was built using Spring Boot, an open source, microserviceoriented application generator for java applications.[16] [Figure 1] demonstrates how our API works.{Figure 1} Data sources We collected retrospective Class I HLA antibody data screened at our institution from November 2007 through January 2018. The laboratory uses polymerase chain reactionbased methodologies for HLA typing and Luminex beadbased antiHLA antibody detection assays to process patient samples. Generating local frequencies The current cPRA calculator is rooted in population genetics and based on the HLA and ethnic frequencies of deceased US kidney donors. Specifically, multiple locus haplotype frequencies are estimated using an expectation maximization algorithm, which uses an iterative process to approximate the maximum likelihood estimates of haplotype frequencies in a population of unrelated individuals.[7],[17] The technique is complicated and requires advanced techniques in population genetics, though the Organ Procurement and Transplantation Network has published demonstrations on performing these calculations.[8] To generate HLA frequencies based on our local apheresis platelet donor pool, we aggregated class I HLA antigen data for all typed platelet units donated at our institution from September 2016 through June 2019. We picked September 2016 as our start date as this was when our platelet donor center began proactively identifying women who selfreported a current or past pregnancy in line with AABB standard 5.4.1.3.1.[18] Frequencies were calculated by adding every unique combination of A and B loci antigens and dividing by the total number of platelet units. [Figure 2] demonstrates this methodology in more detail. Source code to generate the frequency file has also been made available online.[15]{Figure 2} Statistical analysis For each patient in our data set with at least one HLA antibody, we calculated a cPRA score using the UNOS frequencies and a cPRA score for our local, platelet unitdriven frequencies. We created histograms to examine the distribution of cPRA scores for our population, segmented by gender. To compare differences in the median cPRA scores between groups, we used the Wilcoxon SignedRank test. To compare interpopulation median cPRA gender differences, we applied the Mann–Whitney test. We used the Spearman's rho correlation to examine the relationship between scores from the two calculators. To look at the patientlevel magnitude of the difference between calculators, we created histograms, segmented by gender, for the median difference between calculators. Finally, we then placed each cPRA score into one of three buckets: <20% (low cPRA), 20%–90% (medium cPRA), and >90% (high cPRA). We used these buckets based on our clinical experience interpreting cPRA for patients undergoing workups for platelet refractoriness. We looked at intercategory movement of patients between the three cPRA score buckets (low, medium, and high) to assess the clinical impact of using the platelet unitdriven calculator. All analyses were conducted using R (2016. Foundation for Statistical Computing, Vienna, Austria. URL: https://www.Rproject.org). The Partners HealthCare Institutional Review Board reviewed and approved our study (protocol number, 2017P000637). Results We calculated cPRA scores for 4,282 samples and 3,124 unique patients undergoing HSCT at our institution using our customdeveloped web service [Figure 1]. Of those, there were 2,531 samples with at least one antibody, which we used for the cPRA calculations. The distribution of cPRA scores between the UNOS calculator and our local calculator is shown in [Figure 3], segmented by gender, along with median and sample size for each group. There was a statistically significant difference between the median cPRA scores between the UNOS population (median = 0.249, interquartile range [IQR] of 0.067– 0.697) and our local population (median = 0.243, IQR of 0.069–0.681) (P < 0.001). We also found a statistically significant difference when we segmented by gender. For men, the UNOS median was 0.165 (IQR of 0.029–0.418). The median for our local population was 0.141 (IQR of 0.004–0.408) (P < 0.001). For women, the UNOS median was 0.474 (IQR of 0.111–0.872) and the local population median was 0.463 (IQR of 0.112–0.856) (P < 0.001), all shown in [Table 1].{Figure 3}{Table 1} While the median cPRAs were statistically different between populations, we were unsure of the clinical significance of these findings given the similar values. To further evaluate the implications of these findings, we performed several analyses. First, we examined the magnitude of the cPRA score difference between the two calculators, segmented by gender, by calculating the absolute difference between the UNOScalculated cPRA and the localcalculated cPRA scores for every patient in our data set with at least one antibody (e.g., a sidebyside comparison of the two calculators for every patient in our analysis set). As shown in [Figure 4], the median difference between the UNOScalculated cPRA and the localcalculated cPRA was below 0.25 for every patient, with the majority even lower (male median difference 0.014, IQR of 0.004–0.029, female median difference 0.013, IQR of 0.003–0.028). Second, we looked at the correlation between the three comparison groups (total, female, and male). Each group was highly correlated (rho of 0.994, 0.996, and 0.987, P < 0.001 for all three comparisons) [Table 2].{Figure 4}{Table 2} Finally, we looked at whether the calculators would place patients into different cPRA groups. Based on our clinical experience, we picked <20%, 20%–90%, and >90% as three clinically significant cPRA scoring groups (mildly, moderately, and highly sensitized, respectively). Overall, the number of patients in each group was similar, with most patients falling in the low or moderately sensitized groups in each population [Figure 5]. At the patient level, there was minimal but important movement between groups. For example, 25 patients were differentially categorized from the UNOS calculator as compared to the local calculator and 50 patients were differentially categorized from the local calculator to the UNOS calculator, for a total of 2.96% of patients differentially classified (75/2531) [Table 3].{Figure 5}{Table 3} Discussion Platelet transfusions are a critical supportive therapy for patients undergoing HSCT, though many patients will develop antibodies to HLA making therapeutic platelet transfusion challenging. cPRA scoring is used by clinicians as a populationbased method of determining the likelihood of HLAmediated platelet refractoriness. Because cPRA is currently based on a population frequency of deceased US kidney donors, we hypothesized that using a local frequency that considers the number of platelet units expressing HLA (as opposed to the number of individual donors expressing HLA) might result in different cPRA scores for patients undergoing HSCT at our institution. While we found statistically significant differences between the two calculators, the clinical significance of the results is less clear. The groups were highly correlated, and when we put patients into cPRA “bins” (mildly, moderately, or highly sensitized), there was some, but not great movement between the groups. These methods and findings are important for several reasons. First, in order to do these calculations at scale, we had to build a cPRA calculator. We have open sourced the computer code for this web service and posted it online.[15] In addition, we have posted our custom frequency files as well as the code to generate these files, so that others can use our web service to not only calculate cPRAs based on UNOS frequencies for their own population, but also build local, custom frequencies. Second, we calculated cPRA scores based on a platelet unitdriven local frequency. Unlike the US kidney donor pool where there is at most a 1:2 ratio between donors and kidneys, many platelet donors give numerous units of platelets over time.[19] Thus, the true HLA “pool” is a product of platelet donors and platelet units donated. Third, we used a simpler, countingbased method of determining HLA frequencies, as opposed to the more complicated expectationmaximization/maximum likelihood method used by OPTN. The correlation suggests that frequency counting may be sufficient for determining HLA frequencies, though further validation is needed across different populations. Finally, we note that our population is US based (as is UNOS), and it is possible that different countries will have different results based on their own local frequencies. We believe that our methodology, a mechanism to scale calculation based on populationspecific frequencies of genetic information, provides a way to enable more locally precise driven clinical decision support. We believe that this model has value beyond cPRA and HLA. An additional important finding is the difference in cPRA scores by sex, with females having a statistically and clinically significant higher median cPRA score than males. Given that pregnancy is a known risk factor for HLA sensitization,[20],[21] this is not altogether surprising, and our findings support a lower threshold for contemplating HLA matching for female patients that appear refractory to random platelet transfusions. There are important limitations to our work. First, this is a singlecenter study, and our custom frequencies were also derived from a singlecenter donation site. Second, we did not look at clinical outcomes, for example, how cPRA predicts transfusion response, which we leave to a future study. Third, not all of our platelet donors have undergone HLA typing, given the cost and logistics of performing typing on all donors. While we did not analyze this formally, we suspect this would bias towards frequent donors, as they are more likely to get typed over time. Future work could look at how cPRA predicts platelet count response, as well as how cPRA scores differ across multiple sites, each with their own locally derived frequencies. Conclusions In conclusion, a customderived, platelet unitdriven HLA frequency cPRA calculator statistically differed from the UNOS calculator, with unclear clinical implications. Importantly, we have made the tooling available online via an opensource web service, so similar analyses can be done for different population sets. Thrombocytopenia is a dangerous complication for patients undergoing HSCT, and cPRA is one component of managing these patients. Continued efforts to better understand how we can minimize thrombocytopenia will be important for maximizing limited resources such as platelet transfusion products and taking optimal care of these patients. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest. References


