Deep vein thrombosis (DVT) is a common complication in patients with lower limb fractures, which can result in fatal pulmonary embolism (PE) (1, 2). The incidence of DVT after surgery for lower limb fractures reported in previous studies ranges from 8.2 to 61.3% (3, 4). As most DVT cases are asymptomatic, early diagnosis is relatively difficult. If not prevented in time, some of the asymptomatic DVT may progress to symptomatic DVT with clinical symptoms such as redness, leg swelling, and tenderness. The consequence of symptomatic DVT often indicates a poor outcome and increases the risk of PE and mortality, and it is therefore highly important to prevent DVT in patients with lower limb fractures during the perioperative period.

DVT diagnosis has traditionally depended on digital subtraction angiograph (DSA). As a gold standard for diagnosis, DSA is a traumatic examination. Hence color Doppler ultrasonography is more frequently used in identifying DVT (5). Considering the actual situation, the objective and timely diagnosis of DVT is frequently delayed in daily practice (6, 7). Ageno et al. found that it required 5 days from symptom onset to the DVT diagnosis for 47.1% of patients and >10 days for 22.6% of patients (6). Hence, there is an urgent need to identify patient at high risk based on the related risk factors or admission laboratory biomarkers. By far, substantial efforts have been made to address this in orthopedics or other fields. The risk factors for DVT include older age, gender, obesity, smoking, cardiovascular disease, a cancer history, fracture location, delay in surgery and so on (8–10). Given so many associated risk factors, early accurate prediction of DVT by utilizing risk factors may be difficult to manage. Although previous studies have demonstrated that many risk factors are associated with DVT after lower limb fractures, they failed to translate the findings into accurate predictive model to identify DVT in elderly patients with lower limb fractures (7, 11–13). Indeed, these identified factors either relied too heavily on the subjective judgment of the assessors, or lacked well-validated biomarkers (14–16). Therefore, it is almost impossible to use these “isolated” risk factors to determine the “intuitive” and “real” possibility that one patient develops a DVT.

Caprini score has been frequently used to evaluate the risk of VTE in surgical patients. Although there is robust evidence to show its validity, it includes more than 30 risk factors in various aspects, which makes it too cumbersome and complex to allow for a rapid clinical diagnosis. Even worse, orthopedic surgeons were disappointed to find that elderly patients with lower limb fractures would always be classified as high-risk group (17–19). The purpose of this study was to develop a simple but valid predictive model to assess the risk of DVT in elderly patients with lower limb fractures, with the hope to provide a more accurate method for identifying and preventing DVT during the perioperative period.

The research protocol for this study was approved by the local ethics committee of Shanghai Pudong New Area People's Hospital. Patients were recruited from the Orthopedics Department from January 2016 to December 2018, and all the patients were diagnosed with new lower limb fractures. All participating patients signed informed consent.

The inclusion criteria were as follows: (1) age ≥60 years; (2) surgical treatment; and (3) internal implants. The exclusion criteria were as follows: (1) pathological fracture; (2) old fracture; (3) open fracture; (4) multiple complicated injuries; (5) anti-coagulation before injury; (6) hematological system disease; and (7) insufficient clinical or follow-up data.

After admission, all patients were treated with low molecular weight heparin (LMWH, 4,000 IU/d) for prophylaxis 12 h after injury. If the patients were hemodynamically unstable, LMWH heparin was not administered until the condition became stable. DVT screening of the lower extremities was performed using color Doppler ultrasonography before and after surgery.

The following data were collected: (1) demographics: gender, age, comorbidities, smoking history, and the time from injury to surgery; (2) location of the fracture: hip, knee, ankle and foot, femoral shaft, tibia and fibula shaft; (3) anesthesia method; (4) time of operation; (5) blood transfusion; (6) implant type; and (7) D-dimer level.

A nomogram is a multi-index graphical calculating device and has been widely used as a prognostic device in cancer research and medicine, and it is a good tool to increase the accuracy of predicting prognosis and makes it more convenient to reach a clinical decision (25–27). To the best of our acknowledge, this study might be the first applied to DVT of lower limb fracture in elderly patients.

Lasso logistic regression is a useful alternative for variable reduction when collinearity is present. With this advantage, we developed and validated this predictive model for DVT risks in elderly patients with lower limb fractures by merely using these six easily available variables. Of all the collective factors, smoking, time from injury to surgery, operation time, blood transfusion, implant type, and D-dimer level were selected by multivariate logistic analysis to form the nomogram to facilitate the DVT individualized prediction of elderly patients with lower limb fractures. Good discrimination and high internal validation of this nomogram were assessed by the C-index. The high C-index value in the interval validation demonstrated that this nomogram might be useful and accurate during clinical practice. For such a high predictive performance of our model, overfitting must be a top concern (28–30). Regrettably, we were not able to find suitable publicly available datasets to verify this issue. The bootstrapping technique was thus used to perform internal validation and our model demonstrated quite stable performance, reflecting its ability to avoid overfitting (11, 31). Given the fact that there are only 6 predictors in our model and they are easy to implement, we are hoping that these results can be further validated by using clinical datasets from other investigators.

In the present study, 68 patients out of 576 were examined as asymptomatic proximal DVT by color Doppler ultrasonography above the popliteal vein, and the overall incidence of DVT was 11.81%. During the perioperative period of patients with lower limb fractures, many demographic, traumatic, and treatment features might be associated with the development of DVT. Our nomogram demonstrated that patients who had smoking history, delayed surgery, prolonged operation time, blood transfusion during operation, arthroplasty, and D-dimer level ≥5 µg/ml might be more likely to develop DVT (32).

As shown by previous studies (33, 34), smoking was associated with the occurrence of DVT, but whether it was an independent risk factor was also controversial (35, 36). Our study suggested that patients who had smoking habit in our series had a higher incidence of DVT, and the logistic regression analysis found smoking to be an independent predictor in this nomogram. Many previous studies indicated that delayed hip fracture operation on the elderly could significantly increase the incidence of DVT (2, 37–39). However, clinical practice shows that elderly patients often have other concomitant diseases, especially those over 80 years old, and eliminating the contraindications before operation often delays the operation. In our nomogram, a six-day or longer waiting before operation is an important factor contributing to the higher risk of DVT. Similar to the finding of previous studies (33, 40), longer operation time increased the risk of developing DVT, and more than two hours' operation time significantly increased the rate of DVT after lower limb fracture operation. In addition, intra-operative blood transfusion may also increase the risk of DVT, especially when the transfusion amount is more than 800 ml (41–44). The implant type also has impact on the risk of DVT. The implants used in this study included the plate and screw, intramedullary nail and artificial joint, of which the artificial joint was half hip or total hip replacement for femoral neck fracture. The incidence of DVT in hip replacement was higher than that in the internal fixation system, which was consistent with the previous reports (45) Previous studies demonstrated (46, 47) that the elevated level of D-dimer showed a significant association with DVT in patients with lower extremity fractures. In clinical practice, D-dimer is usually used as a reliable screening tool to predict DVT. But the precise level of D-dimer is still not clear (48, 49). It was found in our study that the rate of DVT significantly increased when the level of D-dimer exceeded 5 μg/ml.

The pathogenesis of DVT includes slow blood flow, a hypercoagulable state of blood and injury of the venous wall (50). The factors which can influent these pathological pathways may have impact on the risk of DVT. In traumatic orthopedics, fracture-associated peripheral capillary injury is often the initial factor of DVT formation. It is acknowledged that the incidence of DVT in lower limb fractures is much higher than that in other parts of the body, which is mainly related to the injury mechanism and the long-term bed confinement required for previous treatment (51). The incidence of DVT in elderly patients is significantly higher than that in the young patients due to their decreased motor function and cardiovascular function.

Several scoring systems have been built to help to predict the risk of DVT for general patients, such as Padua score and Caprini score (17–19). Unfortunately, elderly patients with lower limb fractures were all at the highest level according to above mentioned scoring systems (17–19). Based on this criterion alone, it is difficult for surgeons to make an effective and balanced decision and to know which patient should be focused on. That is why we not only made an incidence and risk factor analysis of DVT but also developed this nomogram of prediction model. For instance, the total points of a simple hip replacement patient were 98 and the incidence of DVT was much smaller than 10%, but if the patient had a smoking history and a high D-dimer level (>5 mg/ml), the total points would be 260(98 + 62 + 100) and the incidence of DVT would be almost 50%. Our model makes it possible to select patients who need early evaluation and intervention. We hope our work will provide an effective approach to improve accuracy and efficiency of predicting DVT so as to improve the prognosis of elderly patients with lower limb fracture.

There are several limitations in this study. First, our data were collected between January 2016 and December 2018, which might cause some selective bias. Second, we only collected the DVT consequence in the injury limbs and ignored the asymptomatic DVT in the uninjured leg. Third, we only performed color Doppler ultrasonography up to 7 days, and did not study long-term DVT results. Finally, although we have used bootstrap testing to examine internal validation of our nomogram, we are not sure whether this nomogram can be used universally in other hospitals or centers because of lack of external validation. Further studies are required to evaluate the long-term efficacy of this nomogram in larger-sample studies.

This study has developed a relatively accurate nomogram to assess the risk of DVT in elderly patients with lower limb fractures. In clinical practice, surgeons can use this nomogram to estimate the risk of DVT of individual patients, and perform relevant prophylaxis to prevent the development of DVT. Nevertheless, external validation is needed to determine whether individual interventions based on this nomogram can reduce the DVT risk.