恶性血液病伴碳青霉烯类耐药菌感染病人死亡风险建模

目的 探寻影响恶性血液病伴碳青霉烯类耐药菌感染病人死亡风险因素、建立死亡风险预测模型，并对模型进行评估和验证。方法 收集2018年11月至2021年2月郑州大学第一附属医院158例碳青霉烯类耐药菌感染的恶性血液病病人资料（建模组121例，验证组37例），建模组，根据其离院时存活与否，分为58例的存活亚组、63例的死亡亚组。纳入病人基本资料、耐碳青霉烯类抗菌药物细菌感染前各种抗生素应用情况、血液病种、血常规及生化、侵入性操作或者手术、造血干细胞移植术（HSCT）、入住重症监护室（ICU）以及基础疾病或合并症等因素，进行单因素分析。logistic回归分析方程中纳入那些P<0.20的指标，以其作为自变量，应用方程所输出的风险因子以及因子的相关系数，建立死亡风险回归方程。进而纳入验证组37例病人资料对模型进行验证。结果 建模组存活亚组和死亡亚组在耐药菌感染前住院时间[（13.40±10.02）d比（23.35±15.52）d]、抗生素使用时间[（11.03±9.33）d比（19.56±15.43）d]、抗生素使用种类[（2.81±1.87）种比（4.00±1.86）种]、最多几种抗生素...

Modeling the mortality risk in patients with malignant hematological disorders accompanied by carbapenem-resistant bacterial infections

Objective To explore the risk factors affecting the mortality of patients with malignant hematologic diseases with carbape-nem-resistant bacterial infections to establish a mortality risk prediction model, and to evaluate and validate the model.Methods Data from 158 malignant hematologic patients with carbapenem-resistant bacterial infections in the First Affiliated Hospital of ZhengzhouUniversity were collected from November 2018 to February 2021 (121 in the modeling group and 37 in the validation group), and theywere divided into a survival subgroup of 58 cases and a death subgroup of 63 cases according to whether they were alive or not at thetime of their discharge from the hospital. Basic information, various antibiotic applications before carbapenem-resistant antimicrobial bacterial infections, types of hematological diseases, routine blood and biochemistry, invasive operations or surgeries, hematopoieticstem cell transplantation (HSCT), admission to the intensive care unit (ICU), and underlying diseases or comorbidities of the patientswere included in the one-way analysis. Logistic regression analysis equations incorporated those indicators with P < 0.20 as indepen-dent variables and applied the risk factors output from the equations as well as the correlation coefficients of the factors to create a re-gression equation for the risk of death. The model was further validated by including the data of 37 patients in the validation group.Re- sults The survival and death subgroups of the modeling group were characterized by the duration of hospitalization before infectionwith resistant bacteria [(13.40±10.02) d vs. (23.35±15.52) d], the duration of antibiotic use [(11.03±9.33) d vs. (19.56±15.43) d], the types of antibiotics used [(2.81±1.87) species vs. (4.00±1.86) species], the most types of antibiotics in combination [(1.67±0.925) spe- cies vs. (2.10±0.59) species], use of carbapenem antimicrobials or not (40/18 vs. 58/5), use of the remaining special-use class of antibiot- ics or not (24/34 vs. 45/18), neutrophil counts [0.55(0.05, 3.64) ×109/L vs. 0.06(0.03, 0.22) ×109/L], duration of granulopathy [3.0 (0, 13.0) d vs. 12.0 (8.0, 19.0) d], whether there was a history of cardiovascular disease (10/48 vs. 21/42), and the spectrum of hematologic disorders (22/13/6/5/12 vs. 34/6/10/10/3) were significantly different (P < 0.05). Whether or not they had received chemotherapy (51/7 vs. 61/2) and whether or not they had received HSCT (8/50 vs. 4/59) were 0.05 < P < 0.20 in the results of univariate analyses. Univari- ate analysis variables with P<0.20 were included in the binary logistic regression equation for modeling, and a total of 4 risk factorswere output to obtain the modeling equation: logistic (P) = 0.061 × a (days of hospitalization before infection) + 1.868 × b (applicationof the rest of the special-use class antibiotics before infection or not) - 0.412 × c (preinfection neutrophil count) + 1.345 × d (cardiovas- cular disease or not) -1.309. The predictive model was validated and evaluated: the likelihood ratio chi-square test showed = 42.26, P < 0.001; the H-L test yielded chi-square = 2.985, P = 0.935. The ROC curves were plotted with the data of the modeling group. TheAUC (i.e., area under the curve) was 0.82, the optimal cutoff value was 0.50, the sensitivity was calculated to be 79.4%, the specificitywas 75.9%, and the Jordon index was 0.56. The ROC curves were plotted with the data of the validation group, the AUC was 0.85, andthe optimal cutoff value was 0.48. The sensitivity was 81.0%, the specificity was 75.0%, and the Jordon index was 0.56.Conclusion The constructed prediction model has good calibration and differentiation and has good efficacy in predicting the prognosis of patientswith malignant hematological disorders accompanied by carbapenem-resistant bacterial infections.