Lower limb arthroplasty involving surgical hip or knee joint replacement is a common treatment for advanced hip or knee joint diseases. The increasing prevalence of osteoarthritis due to aging and obesity as well as considerable improvements in postoperative quality of life are key reasons for the rapidly increasing implementation of this procedure (1). Arthroplasty is anticipated to become the most common elective surgery in the United States (2). Similarly, the volume of arthroplasty procedures in China is growing at a rate of 20% annually (3).

Despite its benefits, arthroplasty involves a substantial risk of deep vein thrombosis (DVT). In the absence of pharmacological prophylaxis, the perioperative incidence of all DVTs was 16.8–46.8% in several studies based on an Asian population (4–6). Many DVTs are isolated distal DVTs, and about 10–20% of isolated distal DVTs may extend to proximal DVTs (7), with 4–15% causing symptoms or pulmonary embolism (8).

The incidence of DVT during the perioperative period following major orthopedic surgery is highest within the first 24 h (9), with a peak occurrence of symptomatic thrombosis observed at one week postoperatively (10, 11). To balance the risk between postoperative bleeding and DVT, mechanical prophylaxis is typically used alone during the initial 12–24 h postoperatively, leaving a gap for pharmacologic prophylaxis (9). The average length of hospitalization after arthroplasty is three days (12). Relying solely on early signs or symptoms of DVT is an unreliable strategy for preventing clinically significant thromboembolic events (13). Early and accurate prediction of the risk of perioperative DVT is crucial for preventing severe complications.

Recently, the association between immune-inflammatory and venous thrombosis has been recognized (14). Furthermore, an interaction between inflammatory and hypercoagulable states has been recognized (15, 16). Thromboelastography (TEG), a method for assessing coagulation dynamics using whole blood, facilitates dynamic monitoring of the entire clotting reaction and exhibits relatively high sensitivity in detecting hypercoagulable states (17). Novel inflammatory markers derived from hematological reports, such as the monocyte-to-lymphocyte ratio (MLR), neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio, platelet-to-neutrophil ratio, systemic immune-inflammation index (SII), systemic immune response index (SIRI), and aggregate index of systemic inflammation (AISI), serve as a cost-effective and readily accessible indicator of overall inflammation levels in the body (18). These peripheral blood-derived indices have been increasingly employed in the study of complications in large-scale orthopedic surgeries, demonstrating promising predictive values (19–23).

Despite extensive research on the risk factors for DVT, the focus has primarily been on symptomatic DVT. To our knowledge, reliable predictive models for asymptomatic DVT during the perioperative periods are currently lacking. Therefore, in this study, we aimed to integrate the perioperative immune–inflammatory status and hypercoagulable state before anticoagulation intervention to identify risk factors for perioperative DVT in patients undergoing arthroplasty. We sought to create and validate a nomogram prediction model to estimate the risk of early postoperative DVT in a simple and accurate manner.

The decrease in the length of stay for arthroplasty has increased the need for enhanced perioperative monitoring (32). We hope that preemptive risk alerts can guide personalized anticoagulation before discharge. To our knowledge, this was the first study to develop a clinical prediction model for early postoperative DVT by combining the perioperative immune–inflammatory status with hypercoagulable states. In this retrospective study, a clinical prediction model for DVT during the perioperative period of arthroplasty was developed using LASSO regression and visualized by a nomogram. The nomogram involved seven predictive variables including age, chronic heart failure, stroke, tourniquet use, post-MLR, post-α, and post-D-dimer. Dividing the dataset based on the chronological order of surgery dates, as opposed to random sampling, more closely simulates true prospective prediction (33). The Brier score (0.1762 < 0.25), Hosmer–Lemeshow goodness-of-fit (p = 0.503 > 0.05), and AUC values (training set, 0.737; validation set, 0.683) demonstrated the model nomogram’s reasonable and robust predictive capability (34, 35). However, the calibration curve overestimated the risk in the higher range of postoperative DVT probability (>70%), potentially owing to the limited number of complex cases available for modeling. Given the strong correlation between immune-inflammatory and coagulation markers, one of these features tend to be selected in LASSO regression to manage multicollinearity and reduce model variance (36). Consequently, LASSO regression was chosen instead of the traditional stepwise regression to screen for risk factors in this study.

Postoperative DVT is a serious but preventable complication during and after hospitalization (37), closely associated with baseline characteristics of patients, such as advanced age and cardiovascular diseases (9, 13). Studies on these factors in patients undergoing surgery, under trauma, or with cancer have confirmed their significance in the development of DVT (38, 39). With advancing age, there is a gradual decline in vascular function and blood circulation (40). A history of stroke or chronic heart failure may result in reduced blood flow, elevated blood viscosity, impaired endothelial function, and damaged vascular walls, consequently increasing the risk of DVT (41–43). In addition, our study did not establish a correlation between BMI and DVT occurrence, possibly attributable to the relatively limited distribution of extreme BMIs in the Chinese population (44), or the encouragement for patients to lose weight preoperatively.

Tourniquet application is recognized as a significant surgical factor influencing postoperative DVT. It not only limits postoperative mobility and exacerbates pain but also markedly stimulates tissue factor release through ischemia–reperfusion injury, thereby disrupting the coagulation balance (45). Prolonged tourniquet compression may lead to vascular endothelial damage, further promoting tissue factor release and platelet aggregation, thus heightening the risk of thrombosis. In a prospective study by Huang et al., the incidence of DVT in TKA was significantly lower in the early tourniquet release group (4.6% [9/196]) compared to the late release group (12% [24/200]; OR 0.35, 95% CI 0.16–0.78, p = 0.008) (46). This may also contribute to the observed differences in DVT incidence between hip and knee arthroplasty.

The study shows that on the first postoperative day, MLR exhibits higher predictive potential than NLR, SII, AISI, and SIRI, consistent with the findings of Zhu et al. (20). The difference in predictive ability between NLR and SIRI may be related to the temporal changes in inflammatory dynamics, which are influenced by the primary onset and subacute or acute inflammatory states (19, 23). After surgery, the trend for NLR and SII usually shows an initial rise followed by a decline, peaking within 48 h postoperatively (47). The reduced predictive value of AISI and SII may be attributed to increased platelet activation and consumption caused by surgical bleeding and wound healing, leading to a reduction in the postoperative platelet count that requires time for recovery or reactive increase (21, 48). Although controversy remains regarding the peripheral blood-derived index with the highest predictive value (22), a dynamic and combined monitoring approach can improve diagnostic efficacy.

This study confirmed the predictive role of TEG in identifying DVT, underscoring its superiority over conventional coagulation function screening methods (49). The α angle reflects the rate of blood clot formation and is determined by fibrinogen and platelet interaction, but mainly reflects fibrinogen function in the early stages of coagulation. The larger α angle indicates hyperfunction of fibrinogen, which will directly enhance the fibrin crosslinking speed, leading to the acceleration of the coagulation cascade and the rapid formation of the fibrin network, which can jointly build a stable blood clot with platelets and promote thrombosis (50). Our results demonstrate that postoperative TEG holds predictive value for postoperative DVT, while preoperative TEG does not, aligning with previous findings (51). Brill et al. conducted a study involving 983 patients with trauma and found that the risk of DVT increased two-fold in patients with high coagulation status detected by TEG (52). In addition, some studies (53, 54), have suggested that maximum amplitude (MA) exhibits high sensitivity in diagnosing elevated coagulation status. MA reflects the maximum strength of clots formed through the interaction between fibrinogen and platelets via the GPIIb/IIIa receptor and is positively correlated with platelet count and fibrinogen levels (49). Consequently, the lack of significant sensitivity on postoperative day one may be attributed to surgical bleeding and fluid dilution.

D-dimer serves as a valuable but controversial biomarker for detecting DVT. Its utility is influenced by key variables, such as age and etiology (55, 56), leading to significant fluctuations among patients with surgical trauma, with two distinct peaks observed on postoperative days one and seven, often exceeding the conventional institution-set limit of 0.5 FEU μg/mL set (57). In this study, the AUC for D-dimer in diagnosing DVT was 0.6244, with an optimal cutoff value of 4.52 FEU μg/mL, indicating the stability of D-dimer levels on postoperative day one in predicting DVT, however, dynamic adjustments of the cutoff values are necessary based on temporal variations (58, 59).

This study collected various perioperative medical data before anticoagulation intervention to develop a nomogram prediction model that simply and accurately distinguishes high-risk patients. However, there are some limitations. First, as this is a single-center study, future multicenter research and ongoing data gathering are required to optimize the prediction model and increase the results’ generalizability. To further reduce the population’s heterogeneity, a more detailed subgroup study is required. Second, the validation dataset may be influenced by selection and time biases, hence we are constantly gathering data for prospective validation. Third, because of the brief hospital stay, we are intensifying follow-up to consider the impact of the time course of inflammatory markers on DVT risk and to account for DVT risk over longer timeframe. Finally, we established an association between the variables and DVT rather than a causal relationship, thus, requiring cautious interpretation of our results.

In conclusion, age, chronic heart failure, stroke, tourniquet usage, post-MLR, post-α, and post-D-dimer are independent factors associated with predicting perioperative DVT in hip and knee arthroplasty. Based on these seven indicators, a predictive model was developed, which demonstrated good efficacy in predicting DVT risk. This model holds the potential to increase the clinical benefits for patients by identifying high-risk individuals early in the postoperative period.