Data source
We used the Japanese Diagnosis Procedure Combination database, which has been used for clinical studies on various topics, including burns [13–17]. The database contains discharge abstracts and administrative billing data from more than 1,000 voluntary participating hospitals in Japan. Until 2010, annual data were collected over a six-month period, from July 1 to December 31. Since 2011, data have been collected on a year-round basis. Larger hospitals were more likely to contribute their data to the database, with all 82 hospitals being academic; more than 90% of tertiary emergency hospitals were in the database, and 90% (90/100) of institutions were board-certified by the Japanese Society for Burn Injuries to impart training to burn specialists [13].
The dataset includes hospital identification number, hospital academic status, patient demographic characteristics, admission and discharge dates, main and comorbid diagnoses at admission, in-hospital complications (which were coded according to the International Classification of Diseases [ICD-10]), discharge status, procedure or operation dates and codes (in Japanese), and prescribed medications. Physicians attended to patient records to establish diagnoses at discharge. In order to optimise the accuracy of the recorded diagnosis, the attending physician was obligated to record the diagnosis with reference to the medical record.
Moreover, the database includes burn indices, which are calculated using both burn surface area and thickness: burn index = full thickness of total burn surface area + 1/2 partial thickness of total burn surface area [13, 18]. Previous reports suggested that the burn index was a good predictor of patient mortality [13, 19].
Study Participants
We included patients aged 15 years or older, with emergency hospitalisation for burns (based on ICD-10 codes T20-T30) between July 2010 and March 2018, who started mechanical ventilation within 3 days of admission, and were administered at least one of the following sedative agents: dexmedetomidine, midazolam, or propofol [4, 16, 20]. In addition, we restricted the study participants to patients with severe burns, defined as burn index ≥ 10 [3, 14, 15]. We excluded patients with missing records of burn index, and those who died on the day of mechanical ventilation initiation to avoid immortal time bias.
Exposure
The exposure of interest was the use of dexmedetomidine. The dexmedetomidine group was made of patients who were administered dexmedetomidine on the day of mechanical ventilation initiation, whereas the control group was made of patients who were administered either propofol or midazolam on the day of mechanical ventilation initiation. A small number of patients receiving both dexmedetomidine and propofol or midazolam were included in the dexmedetomidine group since our interest was the effect of dexmedetomidine.
Outcomes
The primary outcome was all-cause 30-day in-hospital mortality. Secondary outcomes included length of hospital stay and duration of mechanical ventilation in all patients and survivors.
Covariates
We considered the following characteristics as potential confounding factors: age; sex; body mass index; Charlson comorbidity index; burn index; the presence of inhalation injury and head/neck burns; Japan Coma Scale score at admission; hospital characteristics, including academic/non-academic status, the presence or absence of certification by the Japanese Society for Burn Injuries, and hospital volume; ICU admission; transportation from another hospital; and mechanical ventilation initiation day (i.e., day 0, 1, 2, or 3). Furthermore, we considered the following treatment regimens on the day of mechanical ventilation initiation: the use of other sedatives (propofol and midazolam), opioids, vasoactive agents (dopamine, dobutamine, adrenaline, noradrenaline, and vasopressin), albumin, intravenous antibiotics, transfusion (red blood cells, platelets, and fresh frozen plasma), continuous renal replacement therapy, enteral tube feeding, and surgery.
The body mass index, calculated using the formula: body mass index = mass (in kg)/height squared (in m2), was categorized into four groups: <18.5, 18.5 to < 25, 25 to < 30, and ≥ 30 kg/m2 [21]. The Charlson comorbidity index reflects the underlying disease or condition [22], and is defined based on ICD-10 codes recorded in the database [23]. In this study, Charlson comorbidity index was categorised into three groups as follows: low, 0; medium, 1; and high, ≥ 2. The Japan Coma Scale score is used to assess the level of consciousness, and correlates with the Glasgow Coma Scale score. Neurologic dysfunction scored at 100 points on the Japan Coma Scale is equivalent to scores of 6–9 on the Glasgow Coma Scale [24]. We categorised the Japan Coma Scale scores into four groups: 0, alert; 1–3, delirium; 10–30, somnolence; and 100–300, coma. Hospital volume was defined as the number of patients with severe burns admitted per year during the study period and was categorised into low, medium, and high hospital volume. Hospital types were categorised based on the accreditation status issued by the Japanese Society for Burn Injuries.
Statistical analysis
We evaluated baseline characteristics and crude 30-day in-hospital mortality in the dexmedetomidine and control groups. Continuous and ordinal variables were expressed as medians and interquartile ranges. Categorical variables were presented as numbers and percentages. Comparisons were performed using the t-test for continuous variables and chi-squared test for categorical variables.
We conducted a one-to-one propensity score matched analysis to determine the conditional probability or the likelihood of administering dexmedetomidine using measured pre-treatment factors. The propensity score was estimated using a multivariable logistic regression model that adjusted for the aforementioned covariates (i.e., patient characteristics including age, squared terms of age, sex, comorbidities, degree of burn, consciousness; hospital characteristics and treatment regimen on the day of mechanical ventilation initiation). Each patient in the dexmedetomidine group was matched to a patient in the control group using nearest-neighbour matching, a calliper width equal to 0.2 of the standard deviation of the propensity score, and without replacement. The C-statistic was calculated to evaluate the discriminative ability of the propensity score estimation. We used the standardized difference to measure covariate balance, whereby an absolute standardized difference > 10% represented meaningful imbalance [25].
In the propensity score-matched patients, we performed logistic regression analyses fitted with generalised estimating equations and paired t-tests to examine the association between dexmedetomidine use and 30-day mortality, as well as the length of hospital stay and the duration of mechanical ventilation in all eligible patients and survivors (to account for competing risk of death) [25].
We also conducted several sensitivity analyses. In the first sensitivity analysis, we used the inverse probability of treatment weighting (IPTW) method. First, we estimated the propensity score of receiving dexmedetomidine using the aforementioned covariates. Thereafter, we calculated the inverse probability of treatment. Finally, we estimated the effect of dexmedetomidine use on 30-day in-hospital mortality using a generalised linear model with a logit link function for a binomial outcome weighted by IPTW [26, 27]. In the second sensitivity analysis, the outcome definition was changed to all cause in-hospital mortality.
In the third sensitivity analysis, we judged the exposure status (the use or non-use of dexmedetomidine) on the day of or the day following mechanical ventilation initiation. For this analysis, we excluded patients who died on the day of or day following mechanical ventilation initiation to avoid immortal time bias.
All tests of significance were two-tailed, and p < .05 was considered significant. Variables were analysed using Stata version 15 (Stata-Corp, College Station, TX, USA).