Antithrombin use and 28-day in-hospital mortality among severe-burn patients: an observational nationwide study

This study was approved by the Institutional Review Board of the University of Tokyo, which waived the requirement for informed patient consent because of the anonymous nature of the data.

We analysed data from the Japanese Diagnosis Procedure Combination (DPC) database, the details of which have been described previously [14]. Briefly, the DPC database includes administrative claims and discharge abstract data for all inpatients discharged from over 1000 participating hospitals. It covers approximately 92% (244/266) of all tertiary-care emergency hospitals in Japan and 90% (90/100) of institutions certified for training burn specialists by the Japanese Society for Burn Injuries [15]. The database includes the following information for each patient: age; sex; primary diagnosis; comorbidities on admission and post-admission complications coded according to the International Classification of Diseases, 10th revision codes (ICD-10) and written in Japanese; medical procedures, including types of surgery, coded with original Japanese codes; daily records of drug administration and devices used; length of stay; and discharge status. The dates of hospital admission, surgery, bedside procedures, drugs administered, and hospital discharge are recorded using a uniform data-submission format [14‐18]. To optimize the accuracy of the recorded diagnoses, the responsible physicians are obliged to record the diagnoses with reference to medical charts. The diagnostic records are also linked to the payment system, and the attending physicians are required to report objective evidence for the diagnosis of the disease for reimbursement of treatment [14‐18]. Patient follow-up thus begins on the day of admission and ends on the date of discharge, either to home, to another hospital, or as a result of death. We could not follow up patients after discharge from hospital because this information was not available in the database [14‐18].

The database also provides important clinical scores, including burn index and Japan Coma Scale (JCS) scores. Although the percentage of total burn-surface area was not available in the database, the burn index was recorded. The burn index takes into consideration both the surface area and thickness of the burn area: burn index = full thickness of total burn-surface area + 1/2 partial thickness of total burn-surface area [15, 19]. Furthermore, a previous large study suggested that burn index was a better predictor of mortality in burn patients than percentage of total body surface area [20]. The JCS correlates well with the Glasgow Coma Scale, and consciousness scored at 100 points on the JCS is equivalent to a score of 6–9 on the Glasgow Coma Scale. We categorized the JCS scores into four groups: 0 (alert); 1–3 (delirium); 10–30 (somnolence); and 100–300 (coma) [15‐17]. To quantify the extent of the comorbidities, the ICD-10 code for each comorbidity was converted to a score, and the sum was used to calculate the Charlson Comorbidity Index (CCI), as described previously [15, 21, 22]. Briefly, the CCI provides a method of predicting mortality by classifying or weighting comorbidities and has been widely used by health researchers to measure case mixes and disease burdens [21]. We categorized hospital types as academic or non-academic. We defined hospital volume as the number of eligible patients treated in the current study and categorized hospital volume into tertiles (low, medium, and high).

We identified patients with severe burns (burn index ≥ 10) [15] who were recorded in the database from 1 July 2010 to 31 March 2013. We compared patients who were administered antithrombin within 2 days of admission (antithrombin group) and those who were not administered antithrombin (control group). The exclusion criteria for this study were: out-of-hospital cardiac arrest, discharge within 2 days after admission (to avoid immortal time bias), and readmitted patients with burns (to avoid patients with planned operations). All the antithrombin used in the current study was plasma-derived antithrombin. Recombinant antithrombin was not available in Japan during the study period, though recombinant antithrombin use was allowed in a limited number of hospitals from September 2015.

The primary endpoint of the study was all-cause 28-day in-hospital mortality. Ventilator-free days (VFDs) were a secondary endpoint in the propensity score-matched groups [23]. VFDs were defined as the number of days the patient remained alive without mechanical ventilation assistance during the first 28 days after admission; patients who died before day 28 were assigned 0 days [23]. We also evaluated incidence of post-admission complications including intracranial haemorrhage, gastrointestinal bleeding, thrombotic disorders (pulmonary embolism and deep venous thrombosis), and renal failure (acute renal failure and/or requirement of renal replacement therapy).

We performed one-to-one matching analysis between the antithrombin and control groups, based on estimated propensity scores for each patient [24, 25]. We assessed the propensity score by fitting a logistic regression model for antithrombin use as a function of the patients’ demographic and clinical characteristics, and hospital factors, including the following, which were previously reported to have the potential to affect mortality and the extent of haemostatic changes in patients with severe burns: age; sex; burn index; CCI, level of consciousness on admission; hospital type and hospital volume; inhalation injury with/without requirement for mechanical ventilation; use of catecholamines (dobutamine, norepinephrine, and/or dopamine); treatment with albumin, haptoglobin, and other blood products; use of drugs for disseminated intravascular coagulation (heparin, thrombomodulin, gabexate mesilate, nafamostat mesilate, or ulinastatin); and requirement for escharotomy and debridement [1, 4, 5, 13, 15, 17, 19, 26‐31]. The C-statistic for evaluating the goodness-of-fit was calculated. One-to-one matched analysis using nearest-neighbour matching was performed based on the estimated propensity scores of the patients; a match was accepted when a patient in the antithrombin group had an estimated score within 0.2 standard deviations of a patient in the control group [24]. We examined the balance in baseline variables using standardized differences, where an absolute value <10% was regarded as balanced [24]. Data are expressed as number (%) or mean (standard deviation). Continuous variables were compared between groups using t tests, and categorical variables were compared using χ² test or Fisher’s exact tests. A value of P less than 0.05 was considered statistically significant. Cox regression analysis was used to assess differences in 28-day in-hospital mortality between propensity score-matched patients with and without antithrombin treatment [24]. We performed logistic regression analysis fitted with generalized estimating equations to examine the association between antithrombin use and 28-day survival accounting for the paired nature of the propensity score-matched patients [24]. Because of its retrospective nature, we did not perform sample-size estimation for the current study. All statistical analyses were performed using IBM SPSS version 22 (IBM Corp., Armonk, NY, USA).

A total of 3223 patients treated at 618 hospitals during the 33-month study period were identified as eligible. Patients were divided into an antithrombin group (n = 152) and a control group (n = 3071), from which 103 propensity score-matched pairs were generated (Fig. 1). The C-statistic indicated a goodness-of-fit of 0.95 for the propensity score model.

Table 1 shows the baseline characteristics of the unmatched and propensity score-matched groups. Patients were more likely to receive antithrombin if they had severe burns, a higher burn index, and more requirements for mechanical ventilation, catecholamines, and other treatments, according to comparisons between unmatched groups. After propensity score matching, most of the baseline patient characteristics were well balanced between the groups. There was no significant difference in associated trauma lesions between the two groups (Additional file 1: Table S1). The median dose of antithrombin administered in the antithrombin group was 1500 U/day (minimum 500 U/day, maximum 3000 U/day; 90 percentiles 1500–3000 U/day) for 3 days (minimum 1 days, maximum 48 days; 90 percentiles 2–17 days). The median length of hospital stay among eligible patients was 54 days.

Overall 28-day mortality was 14.7% (475/3223) in this cohort. Twenty-eight-day mortality was higher in the antithrombin group compared with the control group in unmatched analysis (control vs. antithrombin, 13.5 vs. 39.5%; difference −26.0%; 95% confidence interval [CI] −31.7 to −20.2), but 28-day mortality was lower in the antithrombin compared with the control group in propensity-matched analysis (control vs. antithrombin, 47.6 vs. 33.0%; difference 14.6%; 95% CI 1.2–28.0) (Table 2). Cox regression analysis showed a significant difference in 28-day in-hospital mortality between the control and antithrombin propensity-matched groups (hazard ratio 0.58; 95% CI 0.37–0.90) (Fig. 2). Logistic regression analyses using generalized estimating equations accounting for the paired nature of the propensity score-matched patients showed a significant association between antithrombin use and 28-day mortality in the propensity-matched groups (odds ratio 0.54; 95% CI 0.31–0.95).

There were significantly more VFDs in the antithrombin group compared with the propensity score-matched control group (control vs. antithrombin, 12.6 vs. 16.4 days; difference −3.7; 95% CI −7.2 to −0.12) (Table 3). There was no significant difference in the incidence of post-admission complications between patients with and without antithrombin in the matched groups.