Pulmonary Embolism Risk Factors in the General Trauma Population: A Nested Case-Control Study Using Nationwide Trauma Registry Data in Japan

Background: Data assessing speci�c risk factors for post-traumatic pulmonary embolism (PE) are scarce. Methods: This was a nested case-control study using the Japan Trauma Data Bank between 2004 and 2017. We enrolled patients aged ≥ 16 years, Injury Severity Score ≥ 9, and length of hospital stay ≥ 2 days, with PE and without PE, using propensity score matching. We conducted logistic regression analyses to examine risk factors for PE. Results: We included 719 patients with PE and 3,595 patients without PE. Of these patients, 1,864 [43.2%] were male, and their median Interquartile Range (IQR) age was 73 [55–84] years. The major mechanism of injury was blunt (4,282 [99.3%]). Median [IQR] Injury Severity Score (ISS) was 10 [9–18]. In the multivariate analysis, the variables spinal injury [odds ratio (OR), 1.40 (1.03–1.89)]; long bone open fracture in upper extremity and lower extremity [OR, 1.51 (1.06–2.15) and OR, 3.69 (2.89–4.71), respectively]; central vein catheter [OR, 2.17 (1.44–3.27)]; and any surgery [OR, 4.48 (3.46–5.81)] were independently associated with PE. Conclusions: Spinal injury, long bone open fracture in extremities, central vein catheter placement, and any surgery were risk factors for post-traumatic PE.


Background
Pulmonary embolism (PE) poses a great risk of mortality and morbidity for patients after trauma 1,2 , among whom prophylaxis for PE is an important consideration.However, patients with lower extremity trauma were unable to receive mechanical prophylaxis.Neither could pharmacologic prophylaxis be applied to patients with high risk of bleeding-especially soon after trauma or surgery.It is thus important to carefully assess risk of PE for each individual patient.
Previous studies evaluated the risk of post-traumatic PE.Aging [3][4][5][6][7][8][9][10] and trauma severity 5-9,11−14 are established risk factors.Other, more controversial risk factors of post-traumatic PE are injury sites, blood transfusion, and timing of de nitive surgery.Previous studies were not designed to assess speci c risk factors for post-traumatic PE, rather investigating the overall outcome of venous thromboembolism (VTE), including deep vein thrombosis (DVT).Few risk assessment studies have focused only on posttraumatic PE.Patients with post-traumatic PE, known as de novo PE, do not often have DVT.The risk of PE and DVT may be different among patients with trauma 15-17.We therefore aimed to investigate speci c unknown risk factors for acute PE such as injury sites and timing of de nitive surgery in trauma patients using a nationwide trauma registry in Japan.

Design and setting
We conducted a nested case-control study between January 2004 and December 2017 utilizing a nationwide trauma registry, the Japan Trauma Data Bank (JTDB) database, established in 2003.In 2017, a total of 264 hospitals, including 95% of all tertiary emergency medical centers in Japan, participated in the database 18 , which includes patient demographics, Abbreviated Injury Scale (AIS) scores, Injury Severity Score (ISS), emergency procedures, in-hospital complications, and clinical outcomes.Data collection was performed as a part of routine clinical patient management.

Participants
We included patients who had blunt or penetrative trauma, were aged ≥ 16 years old, had ISS ≥ 9, and were admitted to the intensive care unit (ICU) or general ward.In addition, we included only patients who survived for more than 2 days after hospital admission in order to exclude the impact of early trauma deaths 14,19 .We excluded patients who were pregnant, underwent chronic dialysis 20 , and received anticoagulation drugs (Fig. 1).We de ned blood transfusion as any blood product within the rst 24 hours-the same method as that used in previous studies [20][21][22] .The database recorded central vein catheter placements only at the emergency department.

Statistical analyses
We included patients from the database without PE as a control, using propensity score (PS) matching to ensure that the PE group and the control group were equally balanced on baseline characteristics and severity of trauma.We used the variables age, gender, vital signs at emergency department (Glasgow Coma Scale, systolic blood pressure, heart rate, and respiratory rate), mechanism of injury, transport type, ISS, and admission ward, to calculate PS.We carefully selected these variables based on clinical relevance and previous research 18, 23 .We employed nearest neighbor matching without replacement in a 1:5 manner.We used a caliper of 0.01 standard deviation of the logit of the PS.We evaluated standardized mean difference of the variables over 0.1 as a meaningful imbalance after PS matching.
We calculated descriptive statistics comparing the PE group and the control group, using the Wilcoxon signed-rank test for continuous variables and the chi-square or Fisher exact test for categorical variables, having rst evaluated the normality of all continuous variables using the Kolmogorov-Smirnov test.No continuous variables had normal distributions.
We used conditional logistic regression models to identify speci c risk factors for post-traumatic PE.In the 1st model, the candidates of risk factors were body regions with AIS ≥ 3.In the 2nd model we included long bone and pelvic fracture instead of AIS ≥ 3 in the upper and lower extremities.In the 3rd model, we categorized long bone and pelvic fracture into open, closed, or unclassi able fractures.In the nal model, we added procedures such as blood transfusion, central vein catheter placement, and any surgery, to the 3rd model.We adjusted all the models for comorbidities (see Supplementary Table S1).We did not adjust all the models for in-hospital complications such as pneumonia and sepsis because the database had no data on whether PE or these in-hospital complications occurred antecedently.For all models we did not use covariates which had 10 or less patients in the PE group or control group.When we conducted conditional logistic regression models, we calculated the variance in ation factor of the covariates to evaluate multicollinearity, evaluating a variance in ation factor > 5 as meaningful multicollinearity.However, we did not con rm any meaningful multicollinearity.All p values were twosided; we considered p values of < 0.05 statistically signi cant.We performed statistical analyses using R software (Version 3.6.2) 24 .

Subgroup analysis
We conducted a subgroup analysis to investigate the in uence of time to bone xation.We included only patients who had received primary bone xation.We used a non-conditional logistic regression model with covariates time to bone xation and the same covariates as those used in the nal model.We de ned time to bone xation as the time between hospital arrival and surgery.We did not use covariates which had 10 or less patients in the PE group or control group for the model, in the same way as in the main analysis.

Sensitivity analysis
We conducted a sensitivity analysis to exclude the possibility that mechanical and pharmacological prophylaxis could be potential confounding elements.Because prophylaxis is well adhered to in ICU in Japan, we included only patients admitted to the ICU.We used a non-conditional logistic regression model and included the same covariates as those in the nal model.Again, we did not use covariates which had 10 or less patients in the PE group or the control group for the model, using the same method as in the main analysis.

Imputation
Before PS matching, we replaced missing values concerning vital signs at the emergency department with vital signs at the injury site (we replaced no missing records for the Glasgow Coma Scale, 18 missing for systolic blood pressure, 82 missing for heart rate, and 318 missing for respiratory rate among all the 4,314 patients in this study).We de ned missing values for blood transfusion as no blood transfusion (82 missing among the 4,314 patients in this study).We replaced missing values for time of hospital arrival with time when an emergency medical technician had contacted patients at the injury site (67 missing among 1,997 patients who received primary bone xation).

Ethics approval
The Ethics Committee of the Juntendo University approved this study (IRB No. 19 − 010).We con rmed that all methods were performed in accordance with the relevant guidelines.We con rmed that the need for informed consent was waived by The Ethics Committee of the Juntendo University.The JTDB administrators also provided permission for use of the data from their database.

Results
A total of 155,525 patients among the patients who were registered in the JTDB database were eligible for this study after inclusion and exclusion criteria.A total of 772 (0.5%) patients developed posttraumatic PE.After PS matching, the PE group comprised 719 patients and the control group comprised 3,595 patients (Fig. 1).We detected no meaningful imbalance of the variables for the PS matching between the two groups (Table 1).With regard to baseline characteristics, the PE group had head injury with AIS ≥ 3 less frequently and lower extremities and pelvic injury with AIS ≥ 3 more frequently than those of the control group (16.6% vs. 29.2%,66.6% vs. 47.0%,respectively) (Table 1).Speci cally, the PE group had long bone and open fractures in the upper and lower extremities more frequently than the control group.We found no consistent pattern of comorbidities between the two groups (e-Table 1).However, we noted several obvious concomitant complications that occurred more frequently in the PE group than in the control group (see Supplementary  S3).
The overall in-hospital mortality was higher in the PE group than in the control group (5.8% vs. 3.4%, P = 0.003) (Table 2).In terms of survivor dispositions, the PE group were more likely to be transferred to other facilities than the control group (68.9% vs. 58.4%,P < 0.001 All the models were also controlled for comorbidities.The 3rd model and the nal model were adjusted for unclassi able fracture in lower extremity and pelvis.a Covariate which was not used for the model because of less than 10 patients in PE group De nition of abbreviations: OR = odds ratio; CI = con dence interval; AIS = Abbreviated Injury Scale score In the subgroup analysis, we included 1,997 patients treated with primary bone xation.Our nonconditional logistic regression model showed that time to primary bone xation of 24-120 h-compared to that within 24 h-was associated with PE (Fig. 2).
Sensitivity analysis showed that the following covariates were independently associated with PE, which is consistent with the nal model: spine AIS ≥ Long bone open fractures in upper and lower extremities emerged as high risk trauma for PE.Lower extremity fracture is an established risk factor for PE; however, in previous studies, upper extremity fracture has not been risk factor for PE 3,8,10,21 .Whether open or closed fracture was a risk factor for PE was also unknown.In this study, long bone open fracture in the lower extremities was the most relevant trauma with PE, and long bone open fracture in the upper extremities was the second most relevant trauma.The possible underlying pathophysiology may be that open fracture has greater soft tissue damage involving the venous endothelium and has more bleeding from open wounds, resulting in more severe coagulopathy, than closed fracture [25][26][27] .Venous stasis caused by immobility could also play a role in patients with lower extremity fracture 28 .Open fracture in the extremities could be related to occurrence of PE.
We con rmed that spinal injury 9,12,21,29−31 , central vein catheter placement 6,32,33 , and surgery 3,10,12,21 were risk factors for PE, which is consistent with previous studies.Spinal injury is related to venous stasis because of long-term immobility.Vascular-related procedures such as central vein catheter placement and surgery could be related to endothelial injury and hypercoagulability.
Head, thoracic, and abdominal trauma were not risk factors for PE in this study, although whether these injury sites are risk factors for PE has been controversial in previous studies 3,4,7,9,11 .Compared with this study, previous studies were limited by small sample size and uncontrolled potential confounders such as comorbidities, severity, and surgery.Regarding head trauma, recent studies focusing on the timing of post-traumatic PE showed that head trauma was associated with late onset of PE 16,34−36 .Our cohort might have captured only early PE.
Early de nitive surgery for bone fracture could reduce risk for PE.Previous studies showed that delayed de nitive surgery for bone fracture (after 24 to 48 h) led to pulmonary complications such as pneumonia and acute respiratory distress syndrome because post-traumatic and post-surgical in ammation act as "two hit model" 22,37−39 .However, few studies have showed an association between timing of de nitive surgery and post-traumatic PE because it is a relatively infrequent complication 40 .A subgroup analysis in this study showed that bone xation during 24-120 h after injury was associated with a higher risk for PE.Our study suggests that early total care is better than damage control orthopedics regarding PE among patients with bone fracture.
In this study, patients with spinal injury, open fracture in the extremities, central vein catheter placement, and any surgery were at high risk of post-traumatic PE.Patients with these characteristics should be probably treated with pharmacological prophylaxis as soon as bleeding risk is adequately controlled.
The current study has several important limitations that warrant discussion.First, neither mechanical nor pharmacological prophylaxis was recorded in this database and could be potential confounding factors.
Therefore, we performed a sensitivity analysis among ICU patients because prophylaxis is well adhered to in ICU in Japan (see Supplementary Table S4).The sensitivity analysis showed similar results, four of ve risk factors for PE in the main analysis were independently associated with PE.Second, because they were risk factors for PE in previous studies, pneumonia and sepsis also could be potential confounding factors 13,32,41 .However, we did not include these complications to logistic regression analyses because we did not have onset-time data of complications in our database.Third, we did not include patients who did not survive < 48 hours to exclude the impact of early trauma deaths.Some patients with PE might have been lost in our study.However, because VTE prophylaxis such as anticoagulants could not have been used in early phase of trauma care, these patients were presumably small number and their deaths were unpreventable.Fourth, there might be an information bias.The diagnosis of PE was based on the reports of the physician in charge, and it might have been an underdiagnosis.However, the incidence of PE among patients with eligible criteria was 0.5% (772/155,525) in this study, which is consistent with previous studies (0.1-2.6%) 3,[5][6][7][9][10][11]40 . Diagnosis was likely to have been correct. Moreoer, physicians in Japan, compared with those in other countries, have relatively easier access to the use of computed tomography, because there are many computed tomography scanners in Japan 42 .

Conclusions
Among general trauma population, long bone open fracture in extremities, spinal injury, central vein catheter placement, and any surgery could be risk factors for post-traumatic PE.Among patients with bone fracture, delayed bone xation also could be a risk factor for PE.

c
Other surgeries include facial, neck, and skin surgeries.De nition of abbreviations: PE = pulmonary embolism; SMD = standardized mean difference; IQR = interquartile range; ED = emergency department; GCS = Glasgow Coma Scale; SBP = systolic blood pressure; HR = heart rate; RR = respiratory rate; ICU = intensive care unit; ISS = Injury Severity Score; AIS = abbreviated injury scale score Missing data: None

Figures Figure 1
Figures The primary outcome was occurrence of PE during hospital stay.We categorized AIS codes representing fractures into open, closed, or unclassi able, using AIS 90 Update 98 and AIS 2005 Update 2008.We categorized AIS codes such as 752604.3(humerus fracture), 752804.3(radius fracture), 753204.3(ulna fracture), 853422.3(tibia fracture), and 852604.3(pelvic fracture) in AIS 90 Update 98 as open fractures in this study because almost all were open, although these codes include several other kinds of fractures.

Table 1
Baseline characteristics of patients with and without pulmonary embolism

Table S2
both groups, bone xation was the most frequent surgery (72.5% and 41.4%).Among patients who received bone xation, a lower proportion of the PE group than that of the control group received bone xation within 24 hours (22.4% vs. 27.5%)(see Supplementary Table
).The PE group had longer hospital stays than the control group (32 [interquartile range (IQR): 19-60] days vs. 23 [IQR: 13-36] days, P < 0.001).De nition of abbreviations: PE = pulmonary embolism; IQR = interquartile range.Missing data: In-hospital mortality = 32; Place after discharge = 203 This nested case-control study of general trauma population revealed that long bone open fracture in extremities, spinal injury, central vein catheter placement, and any surgery were risk factors for PE.Bone xation was the most frequent surgery, and delayed bone xation also could be a risk factor for PE. Discussion