Establishment of an early risk prediction model for bloodstream infection and analysis of its predictive value in patients with extremely severe burns
张勤,陈尔真,窦懿,吴蓓雯,杨璐瑜,马珍珠,张寅
DOI: https://doi.org/10.3760/cma.j.cn501120-20210114-00021
2021-01-01
Abstract: To establish a prediction model for bloodstream infection in patients with extremely severe burns based on the screened independent risk factors, and to analyze its predictive value. A retrospective case-control study was conducted. From January 1, 2010 to December 31, 2019, 307 patients with extremely severe burns were admitted to the Department of Burns and Plastic Surgery of Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medcine, including 251 males and 56 females, aged from 33 to 55 years. According to the occurrence of bloodstream infection, the patients were divided into non-bloodstream infection group (221 cases) and bloodstream infection group (86 cases). The gender, age, body mass index, outcome, length of hospital stay of patients were compared between the two groups, and the detection of bacteria in blood microbial culture of patients was analyzed in bloodstream infection group. The included 307 patients were divided into modeling group (219 cases) and validation group (88 cases) according to the random number table with the ratio of about 7∶3. The gender, age, body mass index, total burn area, full-thickness burn area, combination of inhalation injury, implementation of mechanical ventilation, days of mechanical ventilation, days of intensive care unit (ICU) stay, outcome, length of hospital stay, complication of bloodstream infection of patients were compared between the two groups. The patients in modeling group were divided into bloodstream infection subgroup and non-bloodstream infection subgroup. The total burn area, full-thickness burn area, combination of inhalation injury, implementation of mechanical ventilation, days of mechanical ventilation, and days of ICU stay of patients were compared between the two subgroups. The above-mentioned data between two groups were statistically analyzed with independent sample test, chi-square test, and Mann-Whitney test to screen out the factors with statistical significant differences in the subgroup univariate analysis of modeling group. The factors were used as variables, and binary multivariate logistic regression analysis was used to screen out the risk factors of bloodstream infection in patients with extremely severe burns, based on which the prediction model for bloodstream infection in patients with extremely severe burns of modeling group was established. The receiver operating characteristic (ROC) curve of the prediction model predicting the risk of bloodstream infection of patients in modeling group was drawn, and the area under the ROC curve was calculated. The sensitivity, specificity, and the best prediction probability were calculated according to the Youden index. The patients in validation group were divided into bloodstream infection subgroup (21 cases) and non-bloodstream infection subgroup (67 cases). The prediction probability>the best prediction probability of model was used as the judgment standard of bloodstream infection. The prediction model was used to predict the occurrence of bloodstream infection of patients in the two subgroups of validation group, and the incidence, specificity, and sensitivity of bloodstream infection were calculated, In addition, the ROC curve of the prediction model predicting the risk of bloodstream infection of patients in validation group was drawn, and the area under the ROC curve was calculated. Compared with those of non-bloodstream infection group, the mortality of patients in bloodstream infection group was significantly higher (=8.485, <0.01), the length of hospital stay was significantly increased (=-3.003, <0.01), but there was no significant change in gender, age, or body mass index (>0.05). In bloodstream infection group, 110 strains of bacteria were detected in blood microbial culture, among which , and were the top three bacteria, accounting for 35.45% (39/110), 26.36% (29/110), and 13.64% (15/110) respectively. Gender, age, body mass index, total burn area, full-thickness burn area, proportion of combination of inhalation injury, proportion of implementation of mechanical ventilation, days of mechanical ventilation, days of ICU stay, outcome, length of hospital stay, and proportion of complication of bloodstream infection of patients were similar between modeling group and validation group (>0.05). Compared with those of non-bloodstream infection subgroup in modeling group, the total burn area, full-thickness burn area, proportion of combination of inhalation injury, proportion of implementation of mechanical ventilation, days of mechanical ventilation, and days of ICU stay of patients in bloodstream infection subgroup were significantly increased (=-4.429, =-4.045, =7.845, 8.845, =-3.904, -4.134, <0.01). Binary multivariate logistic regression analysis showed that total burn area, days of ICU stay, and combination of inhalation injury were the independent risk factors for bloodstream infection of patients in modeling group (odds ratio=1.031, 1.018, 2.871, 95% confidence interval=1.004-1.059, 1.006-1.030, 1.345-6.128, <0.05 or <0.01). In modeling group, the area under the ROC curve was 0.773 (95% confidence interval=0.708-0.838); the sensitivity was 64.6%, the specificity was 77.9%, and the best prediction probability was 0.335 when the Youden index was 0.425. The bloodstream infection incidence of patients predicted by the prediction model in validation group was 27.27% (24/88), with specificity of 82.09% (55/67) and sensitivity of 57.14% (12/21). The area under the ROC curve in validation group was 0.759 (95% confidence interval=0.637-0.882). The total burn area, days of ICU stay, and combination of inhalation injury are the risk factors of bloodstream infection in patients with extremely severe burns. The prediction model for bloodstream infection risk in patients with extremely severe burns based on these factors has certain predictive value for burn centers with relatively stable treatment methods and bacterial epidemiology.