Evaluation and Analysis of Incidence and Risk Factors of Lower Extremity Venous Thrombosis After Urologic Surgeries: A Prospective Two-Center Study Using LASSO-Logistics Regression.


 Purpose: To analyze the incidence and risk factors of lower extremity venous thrombosis after urologic surgeries. Methods: A prospective two-center study was conducted from August 2019 to January 2020. 1122 consecutive patients who underwent urologic procedures were enrolled. The study primary end point was the detection of asymptomatic or symptomatic deep vein thrombosis (DVT) of lower extremity within 7 days after the surgeries. Univariate and LASSO-logistics regression analysis were performed.Results: We excluded 111 patients who met exclusion criteria. Totally, 56 (5.54%) out of 1011 patients had developed DVT. In the univariate analysis, Barthel Index ≤40, D-dimer levels ≥0.5mg/L and age≥ 60 yrs (p <0.001) were the most significant risk factors. The LASSO-logistics regression model identified 9 factors including age, history of DVT, lymph nodes dissection, perioperative steroid use, Caprini Score, Barthel Index, D-dimer levels, cystectomy and prostatectomy.Conclusion: Our study used the LASSO-logistic regression model to provide reliable data on risk factors of DVT after comprehensive urologic surgeries. It might facilitate individualized anticoagulant management of patients undergoing urologic procedures.Trial registration: ChiCTR1900024784


Introduction
Deep vein thrombosis (DVT) is known as one of the most frequent complications of surgery. It may cause mortality or long-term unfavorable health outcomes, which requires extensive treatment and follow up (1).
The risks of DVT after surgery include the type of disease, age, body mass index (BMI), personal or family history of DVT and the approach of surgery, etc. (2,3). Until now, a large proportion of urologic surgeries are endoscopic and the risks of DVT are lower compared with conventional open surgeries. The overall DVT incidence of laparoscopic surgeries for urologic cancers is between 0.7% and 10.3% (2,3). A systematic review pointed out that pharmacological anticoagulation prophylaxis decreases the risk of DVT in surgical patients by approximately 50% (4). However anticoagulation has side effects, especially the increased bleeding tendency (5). It is necessary to nd out the patients who actually need the perioperative pharmacological anticoagulation. A recent retrospective study demonstrates that age greater than 60 years, condition of disseminated cancer, congestive heart failure (CHF) and anesthesia time greater than 120 minutes are independently associated with DVT formation after common urologic procedures. (6) Another Australian cohort study assessed risk factors among patients undergoing major pelvic urologic surgery and revealed that the risk factors for DVT included long operative time of greater than 4 hours, lymph node dissection (LND) and blood transfusion (7).
In 2017, European Association of Urology (EAU) published its rst guideline for DVT prophylaxis (8).
However, the risk classi cations were too simple, which only included age, BMI and the history of DVT.
This guideline needs more convincing prospective evidence-based data.
Least absolute shrinkage and selection operator method (LASSO) is a popular and robust method for regression with high dimensional predictors (9)(10)(11). In this prospective two-center study, we estimated the incidence of DVT and constructed a LASSO-logistics regression model to evaluate risk factors for the development of DVT among patients undergoing urologic surgeries.

Patients
We prospectively collected the data of patients who received urologic procedures between August 2019 and January 2020 in Xiangya Hospital and Hunan Cancer Hospital. The exclusion criteria were as follows: (1) age < 18 years old; (2) using anticoagulant or antiplatelet therapy before or after surgery; (3) patients underwent day-time surgeries (ureteroscopy, ureteral stent placed or removed, cystoscopy, transperineal prostate biopsy, etc.) (4) patients found with DVT before surgery.

DVT risk factors
We assessed the following parameters to screen the possible risk factors on DVT development: (1) basic and demographic characteristics: age, gender, BMI, history of smoking, history of DVT, history of any diseases or surgeries; (2) information about surgical procedures: methods of surgery, operative time, intraoperative blood loss; (3) postoperative information: complications (infection, hemorrhage and etc.), D-dimer levels (on the rst postoperative day), immediate postoperative Caprini Score(a predictive risk assessment tool), Barthel Index (BI) (to evaluate patients' functional status) (12,13), absolute bed rest time, platelet count and mean corpusular hemoglobin concentration (MCHC) (on the rst postoperative day) (11,12).

Diagnosis of DVT
All patients underwent a physical examination and venous ultrasound of the lower extremity at the day of admission. The patients were performed with ultrasound when they were suspected of clinical DVT. All patients received a second venous ultrasound 7 days after surgery to detect those asymptomatic DVT. A standardized complete compression ultrasound protocol was utilized by experienced sonographers.

Ethics
Written informed consent was obtained preoperatively from all patients. All surgery procedures were performed by experienced surgeons. The study was approved by Hospitals Ethic Committee (IRB number: 201906142) and has been registered in the Chinese clinical trial registry (ChiCTR1900024784).

Statistics
We compared two groups using the t-test for continuous variables and χ² test for categorical variables by using SPSS (IBM Corp. Released 2017. IBM SPSS Statistics for Macintosh, Version 25.0. Armonk, NY: IBM Corp.). We used the LASSO-logistics regression model to select the most useful prognostic risk factors of lower extremity DVT after urology surgery. All types of operations were identi ed with a dummy variable. We used the R software version 3.6.1 and the "glmnet" package (R Foundation for Statistical Computing, Vienna, Austria) to do the LASSO-logistics regression analysis. All statistic evaluations were performed by the Department of Statistics, School of Mathematics and Statistics, Beijing Technology and Business University, China.

Results
A total of 1122 patients underwent regular urologic procedures between August 2019 and July 2020 in two hospitals. One hundred and eleven patients were excluded since they met exclusion criteria. The nal number of eligible participants included in the analysis was 1011 (Fig. 1). Among these patients, DVT developed in 56 (5.54%) and PE developed in 2(0.2%) within 7 days after surgery. The in-hospital mortality rate was zero. Table 1 summarized the incidence of DVT by surgical types. Patients performed with radical prostatectomy had the highest incidence (23.3%) of DVT after the surgery, whereas transurethral resection of bladder tumor (TURBT) had the lowest risk of DVT (0.0%).  The LASSO-logistics model went through generalized cross-validation. With the change of the log (λ) value of harmonic parameter, the area under the curve (AUC) value of the ordinate model changed as well. Table 3 showed all the assignments of variables associated with DVT. The numbers of corresponding variables screened out by the model were listed in Fig. 2A. We used a LASSO-logistics regression model to build a risk factors classi er (Fig. 2B). After LASSO-logistic analysis, 9 risk factors were selected including age, history of DVT, LND, perioperative steroid use, Caprini score, BI, D-dimer levels, cystectomy, prostatectomy (Table 4). Discussion DVT events might be a serious complication of common urologic procedures (14). Understanding the risk factors for DVT is of importance in detecting DVT high risk group and decreasing morbidity and mortality after urologic surgeries. This study collected the most comprehensive risk factors of DVT after urologic surgeries in the prospective two-center cohort. With great help of LASSO-logistics regression analysis, 9 risk factors were demonstrated as the most meaningful factors to select high-risk group of DVT. The overall incidence of DVT after surgeries was 5.54% in this study. It could be seen that the incidence of DVT was not low after urologic surgery. This was in agreement with other studies describing the incidence of DVT after urologic surgery (15,16).
Previous studies have identi ed multiple risk factors related to DVT (17)(18)(19). In particular, radical cystectomy, personal history of DVT and LND have been shown to be associated with the occurrence and development of DVT. However, these studies have been limited by a small number of risk factors screened, small sample sizes, retrospective nature and the inappropriate statistical methods. The LASSOlogistics regression allowed us to integrate multiple possible risk factors into one tool, which has signi cantly greater prognostic accuracy than that of the single-factor and multi-factor logistic regression analysis.
This study rstly included the Caprini Score, BI, absolute bed time, platelet count, MCHC and non-cancer urologic surgeries as potential risk factors. The Caprini Score is one widely accepted model with an established history and utilization as a reliable predictive DVT risk assessment tool. Based on the scores, it can be divided into low, moderate and high-risk subtypes. This rating scale can be used in numerous surgical elds, but not for urologic surgeries speci cally (20)(21)(22). In our univariate analysis, high-risk subtype categorized by Caprini score represented a risk factor (HR 3.710, 95%CI 2.112-6.514, p < 0.001).
LASSO-logistics regression model had also selected this rating scale. Thus, it is necessary to include the Caprini Score in the routine preoperative evaluation.
BI is a 10-item measure of activities of daily living originally described in 1965 (13). BI has been used in clinical practice to assess baseline abilities and quantify functional changes. In our study, we rstly assessed the relationship between BI and DVT. Our data suggested that patients whose BI was less than 40 may need more thromboprophylaxis measures after surgeries.
In the univariate analysis, absolute bed rest time ≥ 48 h was associated with a 2.484-fold increase in risk for DVT and platelet count ≥ 300*10 9 /L was associated with a 3.49-fold increase. This result indicated that when absolute bed rest time and platelet count exceeded a certain threshold, the DVT incidence would increase. The pathophysiological mechanism of absolute bed rest time may be the blood stasis in the lower extremity venous (23).
The kind of operations was found an important aspect the of our research. LASSO-logistics regression model also selected these two operation as important risk factors of DVT. Prostatectomy was associated with the highest rate of DVT (23.3%) and cystectomy had the second-highest rate (16.1%). Prostatectomy and cystectomy are the most complicated regular pelvic surgeries in urology. Except the common factors, patients who undergoing these two surgeries should be alert to any sign of DVT. It should be noted that TURP and percutaneous nephrolithotomy had 14.6% had 8.8% rate to develop DVT, respectively. Our data suggested that these two procedures should deserve attention, although the nal model did not select.
Data of these two surgeries should be further validated by large sample prospective study.
In addition to the factors screened above, the LASSO model and univariate analysis founded that the infusion of red blood cells (≥ 2U) and perioperative steroid use increased the risk of thrombosis. This result supports the ndings of Beyer et al. (19), which proves that there is a statistical signi cance between blood transfusion greater than 2U and thrombosis. Many studies also considered blood transfusions as a risk factor for DVT (24)(25)(26). Steroid use was rst considered as meaningful in a retrospective analysis by Tyson et al.(6) Using prospective two-center data, we demonstrated perioperative steroid use is a signi cant risk factor of DVT.
D-dimer levels were increased in almost all cases of DVT in our study. Any process that increases brin production or breakdown will increase D-dimer levels. Shi et al. con rmed that elevated D-dimer early after operation is an independent predictor of DVT in patients undergoing urologic tumor surgery (27).
According to the LASSO model, D-dimer levels could also apply to the non-cancer urologic surgeries. However, a single D-dimer assay should not be used to diagnose DVT. Our result showed that among the negative patients, 379 (39.7%) patients were found with D-dimer levels higher than 0.5 mg/L. Therefore, patients suspected to have DVT should be performed with ultrasound to con rm.
It is additionally notable that history of DVT had no signi cance with DVT on univariate analysis, but the LASSO-logistic regression model screened it out as a risk factor. The result further illustrated the LASSOlogistic regression model is superior to multivariate regression analysis with high dimensional data. To obtain relatively important variables as much as possible, the model chooses the ideal value of λ to maximize the AUC and obtain the optimal number of compression variables.
There are some limitations of this study. Ultrasound was performed within 7 days after the operations, and no further screening was carried out beyond that duration. Therefore, some patients might develop DVT after discharge would be overlooked. Another limitation was the lack of consistency amongst sonographers performing ultrasound examinations due to nature of two-center study. In addition, exclusion of anticoagulation patients after surgery may also affect the incidence rate. Thus, the nding of DVT may be affected by detection bias.

Conclusions
It is the rst prospective study in analyzing risk factors of DVT development in urologic surgeries for cancer or non-cancer. The results suggested that in addition to the traditional risk factors (age, history  The study was approved by Hospitals Ethic Committee (IRB number: 201906142) and has been registered in the Chinese clinical trial registry (ChiCTR1900024784).

Consent to participate
Informed consent was obtained from all individual participants included in the study.

Consent for publication
The authors all made a signi cant contribution to this manuscript and agreed to publish.

Availability of data and material
Contacting the corresponding author by email to get the data and meterial.

Code availability
Contacting the corresponding author by email to get the code.

Con icts of interest
The authors declare that they have no con icts of interest. Patients' cohort.