异基因造血干细胞移植后出血患者预后分析和预测模型构建

Prognostic analysis and predictive model construction of bleeding events in allogeneic hematopoietic stem cell transplant patients

目的研究发生异基因造血干细胞移植(allo-HSCT)相关出血患者的预后并构建出血预测模型.方法回顾性分析苏州大学附属第一医院2004年5月1日至2012年4月1日期间接受allo-HSCT的555例恶性血液病患者的临床资料并构建出血事件预测模型.结果在全部555例患者中302例(54.0%)发生出血事件,其中Ⅰ度出血151例(27.0%),Ⅱ度出血63例(11.0%),Ⅲ度出血48例(9.0%),Ⅳ度出血40例(7.0%).多因素分析显示,出血等级较高(Ⅲ、Ⅳ度)患者的总死亡率(HR=12.53,95%CI7.91~19.87,P<0.001)及非复发死亡率(HR=23.79,95%CI 12.23~46.26,P<0.001)均较高于低出血等级组.另外,供者合并心脑血管疾病、移植物抗宿主病(GVHD)分度、血小板重建不良以及血小板无效输注与出血风险呈现独立相关.通过以上变量构建的出血模型展现了较好的准确度(C-index=0.934),其效能比既往出血模型具有明显优势.结论allo-HSCT患者发生高等级(Ⅲ/Ⅳ度)出血事件后死亡风险增加.经交叉验证的出血风险预测模型对其出血事件的预测和提前干预具有一定价值.

Objective To study hematopoietic stem cell transplantation-related bleeding prognosis and construct a bleeding prediction model.Methods The clinical data of 555 patients with malignant hematologic diseases who underwent allogeneic hematopoietic stem cell transplantation between May 1st 2004,and April 1st 2012 was analyzed retrospectively,and a prediction model was constructed.Results Of the 555 patients,a total of 302(54.0%)patients exhibited bleeding events of varying degrees,including 151(27.0%)with gradeⅠbleeding,63(11.0%)with gradeⅡbleeding,48(9.0%)with gradeⅢbleeding,and 40(7.0%)with gradeⅣbleeding.Multifactorial analysis showed that the overall mortality(HR=12.53,95%CI 7.91-19.87,P<0.001)and non-recurrence mortality(HR=23.79,95%CI 12.23-46.26,P<0.001)were higher in patients with higher bleeding grades(ⅢandⅣbleeding)compared to those with lower bleeding grades.Additionally,the donor’s underlying disease,graft-versus-host disease(GVHD)score,poor platelet reconstitution,and ineffective platelet transfusion were independently associated with bleeding risk.The bleeding model constructed using the above variables showed good accuracy(C-Index=0.934),and its efficacy was significantly higher than previous bleeding models.Conclusion Hematopoietic stem cell transplant patients are at increased risk of death after a bleeding event.The cross-validated bleeding risk prediction model is valuable for early intervention.