This study is a post-hoc analysis of the database of the Japan Septic Disseminated Intravascular Coagulation (JSEPTIC DIC) study (University Hospital Medical Information Network Individual Clinical Trials Registry (UMIN-CTR000012543, http://www.umin.ac.jp/icdr/index-j.html). This study followed the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of each participating hospital (Additional file 1: Table S1). Because of the anonymous and retrospective nature of this study, the board of each hospital waived the need for informed consent.
The JSEPTIC DIC study was conducted using data from 42 intensive care units (ICUs) in 40 institutions throughout Japan [4]. We reviewed all patients admitted to ICUs between January 2011 and December 2013 for the treatment of sepsis (formerly defined as severe sepsis by the International Sepsis Definitions Conference criteria, 2003 [5]). Patients younger than 16 years old and patients who developed sepsis after their ICU admission were excluded.
The following data were collected: ICU characteristics (number of beds, ICU model, preference for disseminated intravascular coagulation (DIC) therapy), age, gender, weight, admission route to the ICU, Acute Physiology and Chronic Health Evaluation (APACHE) II score, pre-existing organ dysfunction (using chronic health evaluation score in APACHE II), pre-existing hemostatic disorders, Sequential Organ Failure Assessment (SOFA) score (days 1, 3, and 7), systemic inflammatory response syndrome (SIRS) score (days 1, 3, and 7), primary infection site, blood culture results, microorganisms responsible for sepsis, daily results from laboratory tests during the first week after ICU admission, serum lactate levels (days 1, 3, and 7), administration of adjunctive medications (including anti-DIC drugs, other anticoagulants, IVIgG, and low-dose steroids) during the first week after ICU admission, transfusion volume (red blood cell (RBC) concentration, fresh frozen plasma (FFP), platelet concentrate) and bleeding complications during the first week after ICU admission, therapeutic interventions including surgical interventions at the infection site, renal replacement therapy, renal replacement therapy for non-renal indications, polymyxin B direct hemoperfusion, plasma exchange, extracorporeal membrane oxygenation (ECMO), and intra-aortic balloon pump use during the first week after ICU admission, duration of mechanical ventilation, vasoactive drugs and renal replacement therapy use up to 28 days after ICU admission, and ICU mortality and in-hospital mortality.
Statistical analysis
Data are expressed as number (%), or median (interquartile range (IQR)), as appropriate. Patients who received IVIgG were compared with patients who did not receive IVIgG. To estimate the association between IVIgG therapy and mortality rates (ICU mortality and in-hospital mortality), multivariable logistic regression modeling and propensity score matching were used. We performed 1:1 nearest neighbor matching without replacement between the IVIgG and no-IVIgG groups based on estimated propensity scores for each patient. For propensity score matching, a caliper was set at 20% of the standard deviation of the logit of the propensity score. To calculate a propensity score, we fitted a logistic regression model for IVIgG administration adjusted for the following factors: ICU characteristics, age, gender, weight, admission route to the ICU, pre-existing organ dysfunction, pre-existing hemostatic disorders, APACHE II score, SOFA score of each organ on day 1, SIRS score on day 1, primary infection site, blood culture results (positive, negative, or not taken), causative microorganisms, surgical interventions to the infection source, and laboratory test results (white blood cell count, platelet count, hemoglobin level) on day 1. Other laboratory data collected (including fibrinogen, fibrin/fibrinogen degradation products, d-dimer, anti-thrombin, and lactate) were not used to estimate the propensity score since the proportion of missing data was >10%. Other therapeutic interventions were not included for the estimation of the propensity score because timing data of those interventions were not recorded in the database. Standardized difference was used to evaluate covariate balance, and an absolute standardized difference of >10% represents meaningful imbalance. To make the results more robust, we used generalized estimating equations fitted with logistic regression models in the matched groups to assess the association between IVIgG and mortality adjusting for clustering within hospitals and other therapeutic interventions which were not used to estimate the propensity score (anti-thrombin, recombinant human thrombomodulin, heparinoid, protease inhibitor, low-dose steroid, renal replacement therapy, renal replacement therapy for non-renal indications, polymyxin B direct hemoperfusion, plasma exchange, veno-arterial ECMO, veno-venous ECMO, intra-aortic balloon pump, and the volume of transfusion (RBC, FFP, platelet concentrate)).
The database does not include the exact timing of administration of IVIgG within the first week. The timing of administration in some patients might be better correlated with severity on day 2 or later. To adjust the severity within the first week, we added a supplemental analysis, using generalized estimating equations fitted with logistic regression models adjusting for clustering within hospitals, other therapeutic interventions, and SOFA score (each organ score) on days 3 and 7.
The survival curve was generated by the Kaplan-Meier method and hazard ratios for administration of IVIgG were estimated using the multivariable Cox regression model. Univariate differences between groups were assessed using the Mann-Whitney U test for continuous variables and chi-square test or Fisher’s exact test for categorical variables.
Interaction between high (29 and over, highest interquartile range) and low (less than 29) APACHE II score groups was tested using the Breslow-Day statistic in matched groups created by propensity score. Interactions between immunodeficiency and effects of IVIgG were evaluated by subgroups with and without immunodeficiency in the same way. A p value of 0.05 was considered statistically significant. All analyses were performed using IBM SPSS Statistics version 22 (IBM Corp., Armonk, NY, USA).