Rapid Prediction of Hematologic Acute Radiation Syndrome in Radiation Injury Patients Using Peripheral Blood Cell Counts.

Rapid clinical triage of radiation injury patients is essential for determining appropriate diagnostic and therapeutic interventions. We examined the utility of blood cell counts (BCCs) in the first three days postirradiation to predict clinical outcome, specifically for hematologic acute radiation syndrome (HARS). We analyzed BCC test samples from radiation accident victims (n = 135) along with their clinical outcome HARS severity scores (H1-4) using the System for Evaluation and Archiving of Radiation Accidents based on Case Histories (SEARCH) database. Data from nonirradiated individuals (H0, n = 132) were collected from an outpatient facility. We created binary categories for severity scores, i.e., 1 (H0 vs. H1-4), 2 (H0-1 vs. H2-4) and 3 (H0-2 vs. H3-4), to assess the discrimination ability of BCCs using unconditional logistic regression analysis. The test sample contained 454 BCCs from 267 individuals. We validated the discrimination ability on a second independent group comprised of 275 BCCs from 252 individuals originating from SEARCH (HARS 1-4), an outpatient facility (H0) and hospitals (e.g., leukemia patients, H4). Individuals with a score of H0 were easily separated from exposed individuals based on developing lymphopenia and granulocytosis. The separation of H0 and H1-4 became more prominent with increasing hematologic severity scores and time. On day 1, lymphocyte counts were most predictive for discriminating binary categories, followed by granulocytes and thrombocytes. For days 2 and 3, an almost complete separation was achieved when BCCs from different days were combined, supporting the measurement of sequential BCC. We found an almost complete discrimination of H0 vs. irradiated individuals during model validation (negative predictive value, NPV > 94%) for all three days, while the correct prediction of exposed individuals increased from day 1 (positive predictive value, PPV 78-89%) to day 3 (PPV > 90%). The models were unable to provide predictions for 10.9% of the test samples, because the PPVs or NPVs did not reach a 95% likelihood defined as the lower limit for a prediction. We developed a prediction model spreadsheet to provide early and prompt diagnostic predictions and therapeutic recommendations including identification of the worried well, requirement of hospitalization or development of severe hematopoietic syndrome. These results improve the provisional classification of HARS. For the final diagnosis, further procedures (sequential diagnosis, retrospective dosimetry, clinical follow-up, etc.) must be taken into account. Clinical outcome of radiation injury patients can be rapidly predicted within the first three days postirradiation using peripheral BCC.