Abstract
Objectives
The aim of this research was to compare the accuracy of the modified Wells, the Wells, the Kahn, the St. André, and the Constans score for the diagnosis of deep vein thrombosis of the lower limb in unselected population of outpatients and inpatients.
Method
The pretest of probability score was employed in consecutive 500 outpatients and inpatients with suspicion of deep vein thrombosis. All patients were examined with compression ultrasonography.
Results
Deep vein thrombosis was confirmed in 26.4%. In the unselected population of outpatients and inpatients, the accuracy of the modified Wells score and the Constans score was higher than other scores. Both scores were more accurate for the outpatients. There was no accurate score for the inpatient subgroup.
Conclusions
The modified Wells and the Constans score appear to be useful in the unselected population of outpatients and inpatients and particularly in the outpatient subgroup.
Introduction
Due to the fatal complication of acute deep vein thrombosis (DVT) from pulmonary embolism, early detection for DVT and immediate anticoagulant therapy should be performed. 1 Empirical treatment with anticoagulant in all suspected DVT cases have a risk of bleeding. Correct diagnosis and treatment are necessary. Generally, individual clinical signs and symptoms of DVT are nonspecific for the diagnosis of DVT. 1 Several clinical prediction rules have been developed to improve the accuracy of diagnosis of patients with suspected DVT.2–6
Clinical pretest probability scores for deep vein thrombosis (DVT).
Risk category: low risk ≤ 0 points; intermediate risk = 1 or 2 points; high risk ≥ 3 points.
Risk category: Unlikely DVT (low risk) < 2 points; likely DVT (high risk) ≥ 2 points.
The aim of this prospective cohort study was to compare the accuracy of these scores in predicting DVT in the unselected population of outpatients and inpatients with clinical suspicion of acute DVT and to identify the independent clinical predictors of DVT in the unselected population of outpatients and inpatients.
Method
Consecutive 500 outpatients and inpatients aged above 18 years with clinically suspected lower limb DVT addressed to the thrombosis consultation of the division of Vascular Surgery of a university hospital from March 2010 to September 2011 were included in this study. The approval of the institutional review board of ethic committee was obtained. All patients also gave informed consent.
The patient’s demographic characteristics and clinical predicting scores of DVT including the Wells, the modified Wells, the Kahn, the St. André, and the Constans scores were recorded. In the Wells, the Kahn, the St André, and the ambulatory Constans score, a score of 0 or less was defined as low probability, a score between 1 and 2 was defined as moderate probability, and a score of 3 or more was defined as high probability. In the modified Wells score, a score of 2 or more meant likely DVT (high probability) while a score less than 2 meant unlikely DVT (low probability). After recording the scores, the patients were examined with compression ultrasound (CUS) of the lower extremities by experienced vascular surgeons.
CUS was performed on GE LOGIC 9 (GE Healthcare, USA), following a standardized protocol using real-time B mode compression ultrasonography with a linear probe of 5–10 MHz. 8
For each score, a receiver operating characteristic (ROC) curve was drawn. The area under the ROC curve with corresponding 95% confidence interval was calculated. 9 The DeLong’s test for two correlated ROC curves was used to assess the difference in the areas under the ROC curves. 10 Subgroup analyses of the ROC curves for outpatients and inpatients were performed.
We also identified the clinical predicting factors associated with DVT by logistic regression analysis for univariate analysis. The evaluated factors were male gender, cancer, lower limb paralysis or immobilization, confinement to bed more than three days, orthopedic surgery within six months, unilateral lower limb pain, local warmth, localized tenderness, whole limb enlargement, calf enlargement of more than 3 cm compared to the other sites, unilateral pitting edema, superficial venous dilatation, other diagnosis at least as plausible as DVT, previous venous thromboembolism, use of oral contraceptive pills, use of corticosteroid, and pregnancy.
All variables with p < 0.20 in univariate analysis were included in multivariate analysis. A backward stepwise procedure was then applied to remove variables that were not associated with deep venous thrombosis (p < 0.05). Hosmer-Lemeshow test was used to evaluate the goodness-of-fit of the logistic regression model.
Data were prepared and analyzed using PASW statistics 18.0 and R program version 3.0.1 (http://www.r-project.org). All statistical tests were two-tailed with p < 0.05 considered significant.
Results
Baseline characteristics of patients with or without deep vein thrombosis (DVT).
In univariate analysis, we found that the variable predictors for DVT were cancer, confinement to bed more than three days, unilateral lower limb pain, local warmth, whole limb enlargement, calf enlargement of more than 3 cm compared to other sites, and previous venous thromboembolism with p < 0.05 (Table 2).
Independent predictors of deep vein thrombosis (DVT) in 500 patients.
There was no statistically significant difference between outpatient and inpatient baseline characteristics except lower limb immobilization which was higher in inpatients (22.9% vs. 15.3%; p = 0.04) and superficial venous dilatation which was higher in outpatients (23.5% vs. 13.9%; p < 0.05).
Patient risk classification and prevalence of deep vein thrombosis (DVT) in outpatients and inpatients.
Risk category: low risk ≤ 0 points; intermediate risk = 1 or 2 points; high risk ≥3 points.
Risk category: unlikely DVT (low risk) <2 points; likely DVT (high risk) ≥2 points.
The ROC curves of each score in unselected outpatients and inpatients, outpatient subgroup, and inpatient subgroup were plotted (Figures 1 to 3).
Receiver operating characteristic curve of prediction scores in the unselected population of outpatients and inpatients. Receiver operating characteristic curve of prediction scores in outpatients. Receiver operating characteristic (ROC) curve of prediction scores in inpatients.
Areas under the receiver operating characteristic (ROC) curves.
Comparison of the areas under the receiver operating characteristic curves for clinical predictor score of deep vein thrombosis.
In total 500 cases and the outpatient subgroup, the modified Wells, the Constans, and the Wells score had higher area under the ROC curve than the Kahn and the St André score. There was no statistically significant difference between the modified Wells and the Constans scores. The modified Wells score had higher area under the ROC curve than the Wells score. However, the area under the ROC curve of each score in the outpatient subgroup was higher than in the unselected population of outpatients and inpatients.
The area under the ROC curve of each score in the inpatient subgroup was lower than in the outpatient subgroup and in the unselected outpatients and inpatients.
Discussion
Our results demonstrate that the modified Wells and the Constans scores were the most accurate prediction scores in the unselected population of outpatients and inpatients with clinical suspicion of lower limb DVT. The modified Wells score had high sensitivity but low specificity while the Constans score had high specificity but low sensitivity.
In the unselected population of outpatients and inpatients, the prevalence of DVT in the high-risk category of the modified Wells score and the Constans score was about 33% and 40%, respectively. Both scores performed better in the outpatient subgroup than in the unselected population of outpatients and inpatients.
In the outpatient subgroup, the prevalence of DVT in the high-risk category of the modified Wells score and the Constans score was about 39% and 53%, respectively. The prevalence of DVT in the high-risk category of the original studies of the modified Wells score and the Constans score was 27.8% and 58%, respectively.1,7 So, the predictive ability of the modified Wells and the Constans scores in this cohort was consistent with the original papers.
However, both scores performed rather poorly in the inpatient subgroup (area under the ROC curve: 0.59–0.60). Our results show that the modified Wells scores and the Constans scores do not accurately predict the presence of DVT in hospitalized patients. The modified Wells score and the Constans scores were derived from the outpatients with clinical suspected DVT, potentially explaining the lower diagnostic performance among the inpatients and the unselected population of outpatients and inpatients. In the hospitalized patients, the differential diagnosis of limb swelling may be broader than in the outpatients, leading to a decrease in specificity of the scores.
The modified Wells score and the Constans score could be useful in prediction of DVT in the unselected population of outpatients and inpatients. However, the accuracy of the scores in the inpatients was lower. Due to the lower accuracy of the prediction scores in the unselected population of outpatients and inpatients, developing the new scoring system for the unselected population of outpatients and inpatients with our four independent variables from multivariate analysis including unilateral lower limb pain, confinement to bed more than three days, calf enlargement of more than 3 cm compared to the other sites, and previous venous thromboembolism should be further investigated.
Footnotes
Contributorship
NS and TP researched literature and conceived the study. NS, TP and PS involved in protocol development, gaining ethical approval, patient recruitment and data analysis. NS and PS wrote the first draft of the manuscript. All authors reviewed and edited the manuscript and approved the final version of the manuscript.
Ethic Approval
The ethic committee of the Siriraj Institutional Review Board approved this study (027/2554 EC1).
Acknowledgements
The authors would like to acknowledge the following contributors to the work presented in this paper – Siriraj Vascular Surgeon team and Siriraj Vascular OPD-nurse team.
Conflict of interest
None declared.
Funding
This study was supported by Faculty of Medicine, Siriraj Hospital, Mahidol University.