Open-chest versus closed-chest cardiopulmonary resuscitation in blunt trauma: analysis of a nationwide trauma registry

The Japan Trauma Data Bank (JTDB) is a nationwide trauma registry established in 2003, and all trauma patients with an abbreviated injury scale (AIS) score ≥3 in any anatomical region are required to be registered. During the study period, the JTDB received records from 256 hospitals, of which 95% were government-approved tertiary emergency medical centers. The database includes information on injury mechanism, pre-hospital time course, patient baseline characteristics including vital signs at the scene of injury and on arriving at the ED, procedures performed, and status at hospital discharge (deceased or alive). Procedures performed in the ED are stipulated separately from other procedures performed in the operating theater after admission to the ward.

This was a retrospective cohort study evaluating the effectiveness of OCCPR in patients with blunt trauma, using registry data available in the JTDB. We collected data on patients with blunt injury who underwent cardiopulmonary resuscitation (CPR) in the ED. The patients were divided into two groups according to whether or not OCCPR was performed in the ED. We compared the outcomes of the two groups, adjusting for clinical background using propensity score matching analysis. We also evaluated the potential benefits and drawbacks of OCCPR compared with CCCPR among subgroups. The ethics committee of Tokyo Medical and Dental University approved this study (#2192).

We included patients with blunt trauma who received CPR in an ED from January 2004 to December 2015. We excluded patients with an AIS score of 6 because, by definition, these injuries are fatal (i.e. anatomically unsalvageable injury). Patients with a missing AIS score in any anatomical region were excluded to avoid analyzing the patients who potentially had an anatomically unsalvageable injury. Patients transferred from another hospital were also excluded.

We collected the following patient information from the JTDB: age, sex, year of injury, pre-hospital vital signs (systolic blood pressure, heart rate, and respiratory rate), time from emergency medical service (EMS) dispatch to ED arrival, pre-hospital treatment by paramedics (chest compression and defibrillation), vital signs (systolic blood pressure, heart rate, and respiratory rate) and Glasgow coma scale (GCS) score on ED arrival, AIS score for each region, injury severity score (ISS), resuscitation method (OCCPR or CCCPR), status at hospital discharge (survival or death), and time to death or hospital discharge. In addition, we collected hospital information, including the annual number of registered patients, OCCPR cases, and unexpected survivors defined as patients who had survived with a probability of survival based on the trauma and injury severity score (TRISS) <0.5 [16].

Cardiac arrest was defined as a registered systolic blood pressure = 0 mmHg based on the JTDB data registration instructions, which say that systolic blood pressure in a patient whose pulse is not palpable should be registered as 0 mmHg even though actual blood pressure could not be 0 mmHg (i.e. cardiac arrest). The primary study outcome was survival to hospital discharge and the secondary study outcome was survival over 24 hours following arrival at the ED.

Statistical analysis was performed using R 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria). Missing mechanism of data in the naïve dataset were assumed clinically as missing at random (Additional file 1), so that missing data on the collected variables were complemented by the method of multivariate imputation by chained equations with 15 iterations using the package “mice” [17] and 51 datasets were produced. Descriptive statistics were used to display categorical variables as counts and percentages, and numeric or ordered variables as medians and 25^th–75^th percentiles, after pooling all the imputed datasets into one dataset. Predictive statistics were used to display the estimators as point estimation and 95% confidence intervals (CI) integrated across the imputed datasets, based on Rubin’s rule [18].

We used propensity score matching to compare the outcomes between the OCCPR group and the CCCPR group. The propensity score for predicting OCCPR was calculated by logistic regression analysis using variables pertaining to the year of injury, patient factors (age and sex), time from EMS dispatch to ED arrival, mean number of unexpected survivors per year in the treating hospital that the patient was transferred to, vital signs at the scene of injury (systolic blood pressure, heart rate, and respiratory rate), whether or not cardiac arrest at the scene of injury was observed, vital signs (systolic blood pressure, heart rate, respiratory rate, and body temperature) and GCS on ED arrival, whether or not cardiac arrest was observed on ED arrival, AIS score of each region, and ISS. Propensity score matching extracted 1:1 matched pairs of subjects from the OCCPR group and the CCCPR group using the values of the logit-transformed propensity score calculated for each imputed dataset and averaged across the datasets. Match balance between the two groups was assessed by the absolute standardized mean difference of all the variables, and values lower than 0.1 were regarded as acceptable. To achieve this match balance, the caliper width of matching was set at 0.01. Intergroup comparison of the outcomes with propensity-score-matched subjects was performed using the chi-square test. Survival curves were constructed using Kaplan-Meier estimates for the propensity-score-matched subjects and were compared using a log-rank test.

The primary outcome was compared in the propensity-score-matched cohort across subgroups stratified according to sex, age (<50 or ≥50 years), time from EMS dispatch to ED arrival (<30 or ≥30 min), systolic blood pressure on ED arrival (<60 or ≥60 mmHg), whether or not cardiac arrest was observed at the scene of injury, whether or not cardiac arrest was observed on ED arrival, ISS (<30 or ≥30), and AIS score of head, chest, abdomen, and pelvis and lower extremities (0–2 or 3–5). For each subgroup, logistic regression analysis was used to estimate the odds ratio (OR) and interaction for survival to hospital discharge in the OCCPR group and the CCCPR group.

We performed instrumental variable analysis, which is an established technique used to control unmeasured confounding in nonrandomized data [19], as the sensitivity analysis for the propensity score matching. This approach was conducted using a two-stage least-squares regression analysis adjusted by following variables: year of injury, sex, TRISS, systolic blood pressure at the scene of injury, and AIS of head, chest, abdomen, and pelvis and lower extremities, on multiply imputed and not propensity score-matched dataset. The mean number of registered patients per year in the hospital was also incorporated into the model as a measure of quality of trauma care provided by the hospital [20]. Issues with variable multicollinearity were assessed by variance inflation factor (VIF) analysis and the tolerance value was set at <2. We selected the variable of “the mean number of OCCPR cases per year in the hospital” as the instrumental variable. The null hypothesis was that there was no association between the mean number of OCCPR cases per year in the hospital and the actual implementation of OCCPR. A partial F test was conducted to assess an issue of weak instruments, and a value of F-statistic >10 was regarded as acceptable. The level of significance was defined as p < 0.05 for all statistical analyses.

The flow diagram of the patient selection process is shown in Fig. 1. A total of 6510 patients who received CPR in the ED were identified and divided into the OCCPR group (n = 2192) and the CCCPR group (n = 4318), from which 1804 propensity-score-matched pairs were generated.

The baseline characteristics of the multiply imputed dataset before and after propensity score matching is shown in Table 1. The distribution of naïve data and the proportions of missing values are shown in Additional file 2.

The in-hospital and 24-hour survival rates in the OCCPR group were 1.8% (40/2192) and 5.6% (123/2192), and those in the CCCPR group were 3.6% (156/4318) and 9.6% (416/4318), respectively.

The standardized mean difference in the variables according to the estimated propensity score indicated a well-matched balance (Table 1). Of the propensity-score-matched subjects, the proportion of patients surviving to hospital discharge in the OCCPR and CCCPR groups were 1.2% (22/1804) and 3.3% (60/1804), respectively. The proportion of patients in each group surviving over 24 hours after ED arrival were 4.9% (89/1804) and 8.1% (147/1804), respectively. The OCCPR group had significantly lower odds of survival to hospital discharge than the CCCPR group (OR (95% CI) = 0.41 (0.25–0.68)), and the OCCPR group had significantly lower odds of survival over 24 hours after ED arrival than the CCCPR group (OR (95% CI) = 0.59 (0.45–0.79); Table 2).

Survival curve analysis of 30-day mortality was performed (Fig. 2). The log-rank test results revealed significant superiority of CCCPR compared to OCCPR (p < 0.001).

Results of the subgroup analysis are shown in Fig. 3. There were no statistically significant differences between dichotomized subgroups in any variables except for AIS score for pelvic and lower extremity injuries. In patients with severe pelvic and lower extremity injury (AIS score ≥3), OCCPR was significantly associated with a poor outcome compared to CCCPR (p for interaction = 0.038).

The two-way least-squares analysis with this instrumental variable did not alter the original analysis by propensity score matching (mean difference (95% CI) = –5.0% (−9.2, –0.8) for survival to hospital discharge and mean difference (95% CI) = –5.6% (−12.3, 1.1) for survival over 24 hours after ED arrival, respectively; Table 3).