This single-center, retrospective observational study was performed at a tertiary teaching hospital in China. The study protocol was reviewed and approved by the Institutional Ethics Committee of People's Hospital of Deyang City (No. 2021-04-019-K01), and the need for informed consent was waived owing to the retrospective nature of the study and anonymous data collection. In this study, all hospitalized patients between January 1, 2012 and July 31, 2022 were screened by searching the HIS database (n = 671,456), as described in our previous work (19). At the initial screening, 20,730 patients with lower extremity compression ultrasonography (CUS) were considered. After that, patients were excluded if they met any of the following criteria: (1) no neutrophil, platelet or lymphocyte data; (2) CUS examination earlier than routine blood test; (3) DVT or PE diagnosed on admission; (4) chronic DVT; (5) upper extremity DVT; (6) arterial thrombosis; (7) superficial venous thrombosis; (8) incomplete medical record; (9) age younger than 18 years. Finally, a total of 16,725 consecutive hospitalized patients were included in this study (Figure 1).

The following patient characteristics were extracted from the electronic medical records: age, sex, body mass index (BMI), smoking and drinking history. According to the Working Group on Obesity in China, obesity was defined as BMI ≥ 28.0 kg/m² (20). Smoking and drinking status were categorized into never, former or current. In accordance with a previous study (21), the International Classification of Diseases-10 (ICD-10) of discharge diagnoses were used to identify comorbidities: hypertension (codes: I10-I13, I15), diabetes mellitus (codes: E11-E14), chronic obstructive pulmonary disease (COPD, codes: J42-J44), atrial fibrillation (code: I48), heart failure (code: I50), stroke (codes: I60-I64), hepatic insufficiency (codes: K72.0-K72.1, K72.9), renal insufficiency (codes: N17-N19) and cancer (codes: C00-C97, D00-D09).

At our institute, blood samples were routinely drawn on the day of admission or the second day after admission. After blood drawing, routine blood tests were performed immediately using automatic hematology analyzers (Sysmex XN2000, Kobe, Japan). In this study, several hematological indicators closely related to DVT were collected, including white blood cell (WBC, reference range: 3.5–9.5 × 10⁹/L) (22), red blood cell (RBC, reference range: 3.8–5.1 × 10⁹/L) (23), and hemoglobin (reference range: 115.0–150.0 g/L) (24).

The SII was calculated as described in our previous study (25), using the following formula: SII = neutrophil count (reference range: 1.8–6.3 × 10⁹/L) × platelet count (reference range: 101.0–320.0 × 10⁹/L)/lymphocyte count (reference range: 1.1–3.2 × 10⁹/L). The unit of SII was expressed as 10⁹/L.

As we described previously (26), lower extremity CUS was performed by an experienced radiologist, and the results were reviewed by another senior radiologist, using a color-coded ultrasound system with a 3–9 MHz linear-array transducer (iU 22, Philips Healthcare, Netherlands). The scanning was routinely performed at iliac vein, common femoral vein, superficial femoral vein, deep femoral vein, popliteal vein, anterior tibial vein, posterior tibial vein, fibular vein and calf muscle vein. DVT was diagnosed according to Robinov group's criteria (27). The primary outcome was the occurrence of all LEDVT events during the hospitalization, which was double confirmed by ultrasound reports and ICD-10 diagnoses of LEDVT (codes: I80.1, I80.2, I80.3, I80.8, I80.9, I82.8 and I82.9). Based on the thrombus location recorded in the ultrasound reports, proximal LEDVT was defined as thrombus occurring in the popliteal vein and/or above, whereas below the popliteal vein as distal LEDVT. Patients with both proximal and distal thrombus were regarded as proximal LEDVT (28). The secondary outcomes were the occurrence of distal and proximal LEDVT.

Prior to analysis, missing values were detected in all variables, with 30.45% missing in obesity (n = 5,093), 4.25% missing in smoking status (n = 710), and 25.72% missing in drinking status (n = 2,629). Given the large number of missing values, missing indicator method was used to handle these missing data, and coded as a separate category (unknown) (29).

Normality of continuous variables (SII, age, WBC, RBC, hemoglobin) were first checked by the Shapiro-Wilk test, showing that none of these variables conformed to the normal distribution. Continuous variables were described as median [interquartile range (IQR)], whereas the other categorical variables were reported as numbers (percentages). The receiver operating characteristic (ROC) curve analysis was used to determine the optimal cut-off value for high and low SII group. Differences between the two groups were compared using Wilcoxon rank-sum test or Pearson's chi-square test. Furthermore, SII data was natural log transformed [ln(SII)], and the differences between patients with and without LEDVT were visualized by box and whisker plots (Tukey method).

Before multivariate analyses, the linearity assumption for continuous variables was checked by the Box-Tidwell test (30), it was found to be violated for age, WBC, RBC and hemoglobin. For this reason, these continuous variables were converted to categorical variables based on the cut-off values by ROC curves (Supplementary Table S1, Figures 2B–E). Multicollinearity was assessed by the variance inflation factor (VIF), with a VIF ≥5 indicating multicollinearity (31). No significant multicollinearity was found between variables (Supplementary Table S2). Subsequently, multivariate logistic regression analyses were performed to evaluate the association between SII and LEDVT risk in different models. Model 1: SII data was entered as a categorical variable (high vs. low SII). Model 2: SII data was entered as a continuous variable [per ln(SII) increase]. Confounders variables that influences both the independent variable (Table 1) and dependent variable (Supplementary Table S3) were included into the multivariate logistic analyses, including age, sex, obesity, diabetes mellitus, COPD, atrial fibrillation, heart failure, stroke, hepatic insufficiency, renal insufficiency, WBC, RBC, and hemoglobin. Adjusted odds ratios (OR) with 95% confidence intervals (CI) were calculated.

To further minimize the effect of potential confounders, a 1:1 propensity score matching (PSM) was performed between patients with high and low SII. Propensity scores for each patient were calculated using a multivariate logistic regression with the following variables: age, sex, obesity, smoking, drinking, hypertension, diabetes mellitus, COPD, atrial fibrillation, heart failure, stroke, hepatic insufficiency, renal insufficiency, cancer, WBC, RBC and hemoglobin. The greedy, nearest-neighbor method without replacement was used for matching, with a caliper width of 0.20 standard deviation (SD) of the logit of the propensity score (32). Standardized mean difference (SMD) was used to evaluate the balance of baseline characteristics before and after matching, with a SMD < 0.1 indicating adequate balance (33). As mentioned above, multivariate logistic regression analyses were also conducted in these matched samples.

Moreover, we examined the association in subgroups stratified by age, sex, obesity, smoking, drinking, diabetes mellitus, COPD, atrial fibrillation, heart failure, stroke, hepatic insufficiency, renal insufficiency, cancer, WBC, RBC and hemoglobin, because these factors were found to be associated with LEDVT (Supplementary Table S3). Considering the sample size of each subgroup, subgroup analyses were limited to the primary outcome only before matching, and adjusted for confounders variables except for the one defining the subgroup. To determine the effect of missing data, sensitivity analyses based on complete cases were also performed.

Finally, we applied restricted cubic spline (RCS) regressions with 5 knots to explore the dose-response relationship between ln(SII) and risk of LEDVT. Nonlinearity was tested with a likelihood ratio test comparing the spline model to a linear model. If the non-linear association was observed, a two-piecewise linear regression model was used to estimate the threshold effect of ln(SII) on LEDVT risk.

All reported P values are two-sided, and P < 0.05 was considered statistically significant. All analyses were conducted using JMP Pro software (version 16.0.0; SAS Institute Inc., Cary, NC, USA) and GraphPad Prism (version 9.1.1; GraphPad Software, San Diego, California, USA).

The baseline characteristics are presented in Table 1. Overall, the median age was 66.0 years, 52.3% were men, 7.1% were obesity, 22.5% were current smoking and 27.5% were current drinking. The most common comorbidities were diabetes mellitus (38.9%) and hypertension (38.4%). The median SII and ln(SII) values were 555.0 (318.6, 1,098.4) × 10⁹/L and 6.3 (5.8, 7.0) × 10⁹/L, respectively. Using the ROC curve analysis (Figure 2A), the optimal cut-off value was 574.2 × 10⁹/L for SII [sensitivity: 68.8%, specificity: 54.1%, area under curve (AUC): 0.647, 95% CI: 0.634–0.660]. The patients were then divided into low SII group (<574.2 × 10⁹/L, n = 8,602, 51.4%) and high SII group (≥574.2 × 10⁹/L, n = 8,123, 48.6%). Significant differences in most variables were observed between the two groups, except for smoking, drinking and cancer (Table 1).

Of these patients, 1,962 (11.7%) were diagnosed with LEDVT and confirmed by CUS examinations. According to the thrombus location, 1,440 (73.4%) were distal LEDVT and 522 (26.6%) were proximal LEDVT. As shown in Figure 3A, patients with LEDVT had significant higher ln(SII) levels than those without (6.8 [6.2, 7.5] vs. 6.3 [5.7, 6.9] × 10⁹/L, P < 0.001). This trend was also seen in distal and proximal LEDVT (P < 0.001).

After adjusting for confounding factors (Table 2, Model 1), patients in the high SII group showed a 1.740-fold risk of LEDVT (95% CI: 1.546–1.959, P < 0.001). When SII was entered into the multivariate analysis as a continuous variable (Model 2), elevated ln(SII) still remained a significant risk factor (OR = 1.361, 95% CI: 1.278–1.449, P < 0.001). Also, this positive relationship existed in both distal and proximal LEDVT (Figure 4A). At the same time, age, sex, obesity, diabetes mellitus, atrial fibrillation, heart failure, stroke, hepatic insufficiency, WBC, RBC and hemoglobin were independently associated with a higher risk of LEDVT (Table 2, all P < 0.05).

After PSM, 7,814 patients were identified (3,907 in each group), and the baseline characteristics were well balanced between the two groups, with SMDs less than 0.1 for all variables (Table 1, Supplementary Figure S1). Among these patients, 905 (11.6%) had LEDVT, including 670 (74.0%) with distal LEDVT and 235 (26.0%) with proximal LEDVT. Similarly to the results before matching, patients with LEDVT exhibited higher ln(SII) levels than those without, and this difference remained significant in patients with distal and proximal LEDVT (Figure 3B). The results of multivariate analyses indicated that both high SII group (OR = 1.422, 95% CI: 1.232–1.641, P < 0.001; Table 3, Model 1) and elevated ln(SII) (OR = 1.326, 95% CI: 1.200–1.465, P < 0.001; Table 3, Model 2) were significantly associated with an increased risk of LEDVT. Moreover, the forest plot showed that high SII was an independent risk factor for both distal and proximal LEDVT (Figure 4B, P < 0.05).

To further verify the robustness of this association, a series of subgroup analyses were conducted, and the results are graphically presented using a forest plot (Figure 5). Consistent with the main findings, patients with elevated ln(SII) had significantly higher adjusted ORs for LEDVT in nearly all subgroups (OR range: 1.257–1.483, all P < 0.05), except for patients with heart failure (P = 0.908), and renal insufficiency (P = 0.216).

After excluding 7,040 patients with any missing data, a total of 9,685 patients (1,015 with distal LEDVT, 309 with proximal LEDVT) before matching, and 4,642 patients (475 with distal LEDVT, 138 with proximal LEDVT) after matching were available for further analyses. All sensitivity analyses yielded similar findings to the main analyses (Table 4).

As shown in Figure 6, significant non-linear relationships were observed between ln(SII) and risk of all, distal and proximal LEDVT events (P < 0.001). As ln(SII) increased, the risk of LEDVT first decreased, reached a minimum and then increased rapidly. Using two-piecewise linear regressions, the threshold values of ln(SII) were 5.6 × 10⁹/L for all LEDVT (SII = 270.4 × 10⁹/L), 5.5 × 10⁹/L for distal LEDVT (SII = 244.7 × 10⁹/L), and 5.7 × 10⁹/L for proximal LEDVT (SII = 298.9 × 10⁹/L). Below the threshold, increased ln(SII) was not associated with risk of LEDVT (Table 5, all P > 0.05). Nevertheless, above the threshold, each unit increase in ln(SII) had a 1.369-fold, 1.338-fold and 1.390-fold increased risk of all, distal and proximal LEDVT, respectively (P < 0.001).