Trauma patients are at increased risk for DVT, which can contribute significantly to morbidity, mortality, and disability as a life-threatening complication following trauma [2,4,7,12]. Despite the innovation in anticoagulant strategy, the morbidity rate of DVT remains very high due to the dramatic increase in the number of patients involved in traumatic events. The estimated annual incidence of VTE has been reported to be 0.1% and there will be approximately 6,000–100,000 deaths of the 1 million VTE patients per year in the United States [21]. The disability-adjusted life years lost as a result of VTE ranks first among all-cause disability globally (DALYs) and VTE has become the third leading cause of cardiovascular death worldwide [22]. Research has showed that early anticoagulant prophylaxis correlates with a 49% and 53% relative risk reduction for the development of symptomatic DVT and asymptomatic DVT, respectively [2,4,5,6,7,9,16]. Therefore, early recognition of patients at high risk of DVT and undertaking prophylaxis strategies to reduce mortality are of great significance. Systematic identification of individual susceptibility to DVT after trauma could provide an opportunity to reduce the risk of thrombotic complications. Furthermore, a simple, practical, quick, and effective prediction method should be recommended for any trauma surgeon in consideration of the specialty of traumatic disease. In our retrospective study, we found that it was feasible to build a risk prediction score based on routinely available clinical data that could be applied by trauma surgeons to identify high risk DVT among trauma patients. The LASSO technique was utilized to screen the best individual risk factors to build the DVT risk score. Our results demonstrated that the discrimination and calibration of the nomogram deriving from the retrospective cohort are reasonably good, and its C-index was calculated to be 0.852 and was confirmed to be 0.832 in the bootstrapping validation. Meanwhile, the multivariable analysis indicated that the predictive capacity will greatly improve, illustrated by the AUC of 0.852, with 85.1% sensitivity and 75.9% specificity.
In the study, eight predictors including age, BMI, ISS, D-dimer levels, FDPs, PT, prealbumin levels, and Hb levels were incorporated into the prediction score panel. All of the clinical data and laboratory results above are available for trauma surgeons within a few hours of admission. Each variables is strongly associated with DVT in trauma patients. It has previously [23] been described that the D-dimer could further improve both the specificity and the positive predictive value based on age-adjusted cutoff values; therefore, it is not surprising that this was incorporated into our predictive model. The risk factor of ISS has been validated as an independent risk factor for DVT diagnosis, and the incidence of DVT significantly increases with increasing ISS trauma score [24]. With the activation of the coagulation cascade in trauma patients, FDPs [25], PT [26] and Hb [27] become stronger markers for patients who are at high risk of DVT. Prealbumin, which is known as a more sensitive nutritional marker that reflects visceral and somatic nutritional status, would add prognostic information to these established risk predictors because of its nutritional role [28]. Obese patients usually present with a procoagulant state, which may be at markedly increased risk of DVT [29]. From the evidence above, features including age ≥46 years, ISS ≥17, BMI ≥22.5 kg/m2, D-dimer levels ≥13.15 µg/mL, FDPs ≥31.84 µg/mL, PT ≥13.8 s, prealbumin levels ≥208 mg/L, and Hb levels ≥121 g/L were identified as independent predictors of DVT in patients with trauma. The DRAS based on the nomogram prediction chart was developed and we further divided the DVT risk score into four groups based on quartiles of the DRAS: low-risk group, medium-risk group, high-risk group, and very high-risk group.
To further evaluate the prediction efficacy of DRAS, a prospective, double blind validation study was designed to predict incident DVT following trauma. Our results showed that a single predictor could be applied for the early identification of risk of DVT, but the discriminatory ability of a single predictor is limited. When these predictors are integrated into the DRAS, the predictive efficiency significantly improved (AUC=0.879 [0.825–0.932]). Higher DRAS was related to higher risk of DVT among trauma patients in the prospective cohort study. At present, the identification of DVT in trauma patients is usually assessed using the Modified Wells criteria score [12,16]. We also performed this analysis to assess the predictive ability of the Modified Wells score in the same study cohort. The clinical application of the Modified Wells score is relatively limited to predicting DVT among trauma patients and the predictive ability was unsatisfactory for the different Wells risk stratification. Our results fall in line with the studies of Beilman et al. [12] and Khorasani et al. [16], they derived the Wells score that identifies hospitalized patients at greater risk of developing DVT after severe trauma; however, the disadvantages of the Wells score is that its risk stratification is not sufficient to rule out DVT. The DVT risk prediction models that have been created specifically and widely used for hospitalized patients in determining DVT risk over the past decade. For example, the Trauma Embolic Scoring System (TESS) in Krasne’s study investigated five risk factors from 19 variables related VTE markers to objectively quantify the risk of VTE in trauma patients; however, the TESS neglected some lab variables associated with a hypercoagulable state and did not take into account population stratification [30-31]. The Risk Assessment Profile (RAP) [32] score used some comprehensive variables to predict an individual’s likelihood to develop DVT, but some of which may not be feasible to measure accurately and evaluate quickly in emergency trauma patients. Caprini’s [33-34] risk assessment model aims to employ a thrombosis rating scale to calculate the total risk factor score by incorporating more than 30 risk factors, producing a valid method for identifying patients at risk of DVT or PE; however, the scale appears to be overly complicated for clinic\l diagnosis. These results indicate that synthetic prediction models play a vital role in the early identification of DVT; however, each of the current prediction models have their own advantages and disadvantages. A reasonable standard for one prediction model may be not applicable in traumatic DVT because of the special status of traumatic disease. However, the DRAS incorporates objective variables including the age, BMI, ISS, D-dimer levels, FDPs, PT, prealbumin levels, and Hb levels in our current study and can streamline the assessment workflow of patients at the highest risk of DVT. The DRAS is more applicable for trauma surgeons and the promotion of earlier management of prophylaxis therapy.
This study has several obvious advantages. First, the LASSO method was more appropriate for the selection of variables and its consolidation of complex clinical data. Moreover, the eight selected predictors can combined into a panel that can be assessed within a few hours after admission. This is a perfect fit for trauma surgeons to identify patients at high risk of DVT. Second, we performed a prospective validation study to verify the clinical feasibility of the DRAS; the results indicated that the predictive capacity of the DRAS was relatively good, demonstrated by the AUC of 0.879(0.825–0.932), with 85.20% sensitivity and 76.80% specificity. The study also has several potential limitations. First, the retrospective study with prospective validation was performed at a single center, and the sample size of 281 patients in the retrospective cohort and 166 patients in the prospective cohort were relatively small since we used strict inclusion criteria to identify the study population. The inconsistent proportions of the different groups of the DRAS in the prospective validation study may have led to an expected relativity bias. Second, the nomogram only underwent internal validation with a bootstrapping cohort, which also may cause a center bias. Although the DRAS was verified using prospective validation, external validation should be taken into consideration, along with its use in diverse populations and different study centers. Additionally, the outcome of early thromboprophylaxis after the identification of high risk DVT was not observed in an independent prospective cohort. A prospective observational study, preferably in a multicenter design, would be desirable to confirm our findings. Finally, we believe that tapping into the association between clinical variables and the risk of DVT is more reasonable and will ultimately help develop an accurate prediction model for the early identification of high risk DVT after trauma.