Bloody big data : ensuring the health of blood donors and transfused patients with health registers

Abstract

Blood transfusion is widely considered a pillar of modern medicine and is one of the most common medical interventions. A transfusion intimately links two persons and their health through the altruism of blood donors. In turn, this has necessitated the development of rigorous safety measures for both donors and recipients. Hemovigilance is the systematic surveillance of adverse events in the entire blood supply chain, stretching from blood donation to transfusion and follow-up care. Although transfusion safety has dramatically improved in recent decades, especially in regard to transfusion-transmitted infections, there are still evidence gaps for many aspects of blood donation and transfusion safety. This thesis explores several aspects of current controversies in both blood donation and transfusion safety and proposes a ‘big-data’ approach to hemovigilance with the use of data from electronic healthcare records and health registers, and provides a data-driven method for calculating longitudinal transfusion costs.

In Study I, we constructed a new nationwide research database that encompasses all electronically recorded blood donations, transfusions, blood donors, transfused patients, and persons with a blood typing result in Sweden. The database is the Swedish portion of the third iteration of the Scandinavian Donations and Transfusions (SCANDAT3-S) database and contains data spanning from 1968 to 2018 for more than 8 million persons, over 21 million transfusion records, and over 300 million laboratory results. This database is one of the largest and most comprehensive research databases for blood donation and transfusion research and served as the main data source for the other studies in this thesis.

In Study II, we followed up on studies that have observed CD4+ T-cell counts below 200 cells/μl, a level typically indicating AIDS, among frequent platelet donors. This is believed to be caused by frequent leukocyte depletion through the use of widely-used instruments equipped with so-called Leukoreduction System (LRS) chambers. However, it is unknown if this leads to clinically relevant immunosuppression. We conducted a nationwide cohort study of 74,408 platelet and plasmapheresis donors between 1996 and 2017 and observed an increased, donation frequency-dependent hazard ratio (HR) for both immunosuppression-related and common bacterial infections. For a subcohort, we replicated an association between donation frequency and low CD4+ T-cell counts. Together, these findings suggest that frequent platelet donation utilizing the LRS chamber is associated with both T-cell lymphopenia and an increased risk of infections.

In Study III, we aimed to investigate possible transfusion-transmission of cerebral amyloid angiopathy (CAA). CAA is an amyloid-β pathology that co-occurs with Alzheimer’s disease in 80% of cases and is a common cause of multiple spontaneous intracerebral hemorrhages (ICH). Previous studies have demonstrated human-to-human transmission of CAA through peripheral injection of contaminated cadaveric pituitary hormone, but it is unknown if it is transmissible by blood transfusions. Because CAA is rarely diagnosed routinely and would not be frequently registered in health registers, we instead studied possible transfusion-transmission of spontaneous ICH suggestive of CAA. In a binational cohort study of 1,089,370 red-cell recipients between 1970 and 2017, patients that received blood from a donor that subsequently developed multiple spontaneous ICH were associated with an increased hazard ratio for developing spontaneous ICH themselves. This was observed independently in both the Swedish (HR, 2.73; 95% confidence interval [CI], 1.72 to 4.35) and Danish cohort (HR, 2.32; 95% CI, 1.04 to 5.19), after adjusting for known confounders. This association was not observed for ischemic stroke, which is associated with many of the same risk factors as spontaneous ICH but not with CAA. These findings support possible transfusion-transmission of CAA, but other explanations should not be rejected without further mechanistic studies.

In Study IV, we assessed if blood donor sex and parity affected the survival of adult patients transfused with red-cells. Antibodies produced during pregnancy in blood donors have been associated with transfusion-related acute lung injury in recipients and led to the use of predominantly male donor plasma, in some regions also platelets. However, data on red-cell transfusions have been contradictory. Some, but not all, observational studies have found increased mortality among recipients of female donors, parous donors, and sex-mismatched transfusions. We conducted a nationwide study of 368,778 adult patients transfused with red-cells between 2010 and 2018. We demonstrated that donor sex and parity were distributed as-if randomized and constituted a natural experiment. However, female blood donors had lower hemoglobin counts and their units were less efficacious, creating a phenomenon known as treatment-confounder feedback that causes bias in standard regression models. Using inverse probability weighting to mitigate bias due to treatment-confounder feedback, we exploited the natural experiment to emulate a randomized controlled trial. We found no difference in 2-year overall survival comparing transfusions from female to male donor red-cells only (-0.1%; 95% CI, -1.3 to 1.1%) and parous female to male donor red-cells only (0.3%; 95% CI, -0.6 to 1.2%), nor in any subgroup defined by patient sex and age. Comparatively lower hemoglobin counts among female donors is a previously unrecognized source of bias and may help explain the diverging results from previous studies.

In Study V, we calculated longitudinal transfusion costs for patients with myelodysplastic syndromes (MDS) stratified by disease severity defined using the Revised International Prognostic Scoring System (IPSS-R). Transfusion costs included the acquisition cost of blood products, labor costs for nurses, cost for consumables, and costs for laboratory testing. We constructed algorithms to calculate cost based on incident data on blood transfusions and laboratory testing. These algorithms accounted for whether transfusions were administered in an inpatient or outpatient setting, that the first transfusion requires longer administration time than subsequent transfusions, needs for cross-matching, and high patient mortality. In 2018 US dollars, transfusion costs after 4 years ranged from $8805 (95% CI, $6482 to $11,625) in the very low IPSS-R category to $80,106 (95% CI, $61,460 to $95,792) in the very high IPSS-R category.