Establishment of Mortality Risk Prediction Model for Bloodstream Infections in Patients with Hematological Diseases
Jinjin Wang,Hui Zhou,Mengyao Wang,Ailin Zhao,Ting Niu
DOI: https://doi.org/10.1182/blood-2022-168082
IF: 20.3
2022-01-01
Blood
Abstract:Backgrounds: In recent years, with the continuous development of treatment for hematological diseases, the remission rate and survival rate of patients have been significantly improved. However, due to the progression of the hematological disease and immunosuppression, the incidence of bloodstream infection was greatly increased in patients with hematological disease. The occurrence of bloodstream infection had many negative effects on patients with hematological diseases, including the increase in mortality, the adding of medical bills, the prolongation of hospitalization period, and so on. West China Hospital is a national center for diagnosing and treating difficult and critical diseases in Western China. In our center, we found that the incidence of bloodstream infection in hematological patients was not low, and it was often one of the common causes of death. So, our purpose was to explore the risk factors for the 30-day death and establish a mortality risk prediction model in patients with bloodstream infections in hematological diseases, to provide evidence for the prevention and treatment of bloodstream infection in hematological diseases. Methods: The clinical data of 412 strains of pathogens of bloodstream infection from patients with hematological diseases were retrospectively collected from June 2015 to June 2020 in the Department of Hematology, West China Hospital, Sichuan University. Statistical analysis was conducted with IBM SPSS Statistics (version 26.0). 70% of the patients were randomly selected as the development group and 30% as the validation group. Results: A total of 338 patients with hematological diseases with positive blood culture were collected and 412 strains of pathogens of bloodstream infection were isolated. The most common primary hematological disease was acute myeloid leukemia, accounting for 44.2%. Among the included cases, the 7-day and 30-day mortality rates of all patients were 5.6% and 32.0%, respectively. All cases were randomly divided into the development group (n = 289) and validation group (n = 123) by 7:3. Univariate and multivariate logistic regression analysis revealed that septic shock [P = 0.001, OR = 4.126], albumin < 30g/L [P < 0.001, OR = 3.622], platelets < 30×109/L before infection [P = 0.002, OR = 4.202], bleeding [P = 0.029, OR = 2.308], placement of urinary ducts [P = 0.008, OR = 3.769], and inappropriate empiric antibiotic treatment [P = 0.007, OR = 2.523] were independent risk factors for 30-day death after bloodstream infection in patients with hematological diseases. In the light of the results of multivariate analysis, the equation of risk prediction model of 30-day death for bloodstream infections in patients with hematological diseases was established as follows: P = ea/ (1 + ea), a=-4.172 + (1.417 × septic shock) + (1.287 × albumin < 30g/L) + (1.436 × platelets < 30×109/L before infection) + (0.836 × bleeding) + (1.327 × placement of urinary ducts) + (0.925 × inappropriate empiric antibiotic treatment). The Hosmer-Lemeshow Chi-square test showed P = 0.582, and the area under ROC curve was 0.863±0.022 [95%CI (0.820-0.906)], indicating that the model had a good prediction coincidence and discriminant validity. The model was verified using the data of the validation group, and the results exhibited that the sensitivity, specificity, positive predictive value, negative predictive value, and coincidence rate were 97.3%, 86.0%, 75.0%, 98.7%, and 89.4%, respectively. Discussion and Conclusions: This study showed that about 32.0% of patients with bloodstream infections died within 30 days, which was consistent with other data reported at home and abroad. In the mortality risk prediction model established in this study, the four risk factors for death included septic shock, albumin < 30g/L, placement of urinary ducts, and inappropriate empirical antibiotic treatment, which have also been reported in other studies. A notable finding in this study was the identification of previously unnoted risk factors for death, including platelets < 30×109/L before infection and bleeding. The prediction model in our study was verified that the model had a good prediction coincidence, discriminant validity, and clinical predictive value, which can provide evidence for clinicians to evaluate the death risk and make individualized treatment plans for patients.