Background and Aims Acute gastrointestinal bleeding is the most common gastrointestinal cause for hospitalization. For high risk patients requiring intensive care unit stay, predicting transfusion needs through dynamic risk assessment may help personalize resuscitation. Methods A patient cohort presenting with acute gastrointestinal bleeding (N = 2,524) was identified from the Medical Information Mart for Intensive Care III (MIMIC-III) critical care database, separated into training (N = 2,032) and validation (N = 492) sets. 74 demographic, clinical, and laboratory test features were extracted and consolidated into 4-hour time intervals over the first 24 hours from admission. The outcome measure was the transfusion of packed red blood cells during each 4-hour time interval. A long short-term memory (LSTM) model, a type of Recurrent Neural Network (RNN) that processes and learns the best information storage plan, was compared to the Glasgow-Blatchford Score (GBS) for the first 24 hours from admission. Results The LSTM model performed better than GBS in predicting packed red blood cell transfusion overall (AUROC 0.81 vs 0.63;P<0.001) and at each 4-hour interval (P<0.01). Sensitivity analysis also demonstrated superiority over GBS for patients directly admitted from the ED to ICU (0.82 vs 0.63;P<0.001), with upper GIB (0.84 vs 0.68;P<0.001), with lower GIB (0.77 vs 0.58;P<0.001), and with unspecified GIB (0.85 vs 0.64;P<0.001). Conclusions A LSTM model can be used to predict the need for transfusion of packed red blood cells over the first 24 hours from admission to help personalize the care of high-risk patients with acute gastrointestinal bleeding.

中文翻译：

---

神经网络预测进入重症监护病房的急性胃肠道出血患者需要输血

背景和目的急性胃肠道出血是住院的最常见胃肠道原因。对于需要重症监护病房住院的高风险患者，通过动态风险评估预测输血需求可能有助于个性化复苏。方法从重症监护医学信息中心（MIMIC-III）的重症监护数据库中识别出出现急性胃肠道出血的患者队列（N = 2,524），将其分为训练（N = 2,032）和验证（N = 492）组。在入院后的最初24小时内，提取了74个人口统计学，临床和实验室测试特征并将其合并为4小时时间间隔。结果测量是在每个4小时的时间间隔内输注堆积的红细胞。长短期记忆（LSTM）模型，在入学后的头24小时内，将一种处理和学习最佳信息存储计划的递归神经网络（RNN）与格拉斯哥-布拉奇福德分数（GBS）进行了比较。结果LSTM模型在预测整体红细胞整体灌注（AUROC 0.81 vs 0.63； P <0.001）以及每4小时间隔（P <0.01）方面表现优于GBS。敏感性分析还显示，直接从ED入ICU的患者优于GBS（0.82 vs 0.63; P <0.001），GIB较高（0.84 vs 0.68; P <0.001），GIB较低（0.77 vs 0.58; P <0.001） ）和未指定的GIB（0.85 vs 0.64; P <0.001）。