Bundeswehr Institute of Radiobiology affiliated to the University Ulm, Neuherbergstraße 11, 80937, Munich, Germany.
Department of Radiation Sciences (DRS), Institute of Radiation Biology, Helmholtz Zentrum München, Oberschleißheim, Germany.
Health Phys. 2021 Apr 1;120(4):400-409. doi: 10.1097/HP.0000000000001353.
A suite of software tools has been developed for dose estimation (BAT, WinFRAT) and prediction of acute health effects (WinFRAT, H-Module) using clinical symptoms and/or changes in blood cell counts. We constructed a database of 191 ARS cases using the METREPOL (n = 167) and the SEARCH-database (n = 24). The cases ranged from unexposed (RC0), to mild (RC1), moderate (RC2), severe (RC3), and lethal ARS (RC4). From 2015-2019, radiobiology students and participants of two NATO meetings predicted clinical outcomes (RC, H-ARS, and hospitalization) based on clinical symptoms. We evaluated the prediction outcomes using the same input datasets with a total of 32 teams and 94 participants. We found that: (1) unexposed (RC0) and mildly exposed individuals (RC1) could not be discriminated; (2) the severity of RC2 and RC3 were systematically overestimated, but almost all lethal cases (RC4) were correctly predicted; (3) introducing a prior education component for non-physicians significantly increased the correct predictions of RC, ARS, and hospitalization by around 10% (p<0.005) with a threefold reduction in variance and a halving of the evaluation time per case; (4) correct outcome prediction was independent of the software tools used; and (5) comparing the dose estimates generated by the teams with H-ARS severity reflected known limitations of dose alone as a surrogate for H-ARS severity. We found inexperienced personnel can use software tools to make accurate diagnostic and treatment recommendations with up to 98% accuracy. Educational training improved the quality of decision making and enabled participants lacking a medical background to perform comparably to experts.
已开发出一套软件工具，用于使用临床症状和/或血细胞计数变化进行剂量估算（BAT、WinFRAT）和急性健康影响预测（WinFRAT、H-Module）。我们使用 METREPOL（n = 167）和 SEARCH 数据库（n = 24）构建了一个 191 例 ARS 病例数据库。这些病例从未暴露（RC0）到轻度（RC1）、中度（RC2）、重度（RC3）和致命性 ARS（RC4）不等。2015 年至 2019 年，放射生物学学生和两名北约会议的参与者根据临床症状预测临床结局（RC、H-ARS 和住院）。我们使用相同的输入数据集评估了来自 32 个团队和 94 名参与者的预测结果。我们发现：（1）无法区分未暴露（RC0）和轻度暴露个体（RC1）；（2）RC2 和 RC3 的严重程度被系统高估，但几乎所有致命病例（RC4）均被正确预测；（3）为非医生引入预先教育部分可将 RC、ARS 和住院的正确预测率提高约 10%（p<0.005），方差降低三倍，每个病例的评估时间减半；（4）正确的结果预测与所使用的软件工具无关；（5）比较团队生成的剂量估算值与 H-ARS 严重程度反映了仅剂量作为 H-ARS 严重程度替代物的已知局限性。我们发现经验不足的人员可以使用软件工具做出准确的诊断和治疗建议，准确率高达 98%。教育培训提高了决策质量，使缺乏医学背景的参与者能够与专家相媲美。
