The full content of Annals is available to subscribers

Subscribe/Learn More  >
Articles |

Predicting Deep Venous Thrombosis in Pregnancy: Out in “LEFt” Field?

Wee-Shian Chan, MD, MSc; Agnes Lee, MD, MSc; Frederick A. Spencer, MD; Mark Crowther, MD, MSc; Marc Rodger, MD, MSc; Tim Ramsay, MSc, PhD; and Jeffrey S. Ginsberg, MD
[+] Article and Author Information

From Women's College Hospital, Toronto; Hamilton Health Sciences Corporation, McMaster University Medical Centre, and St. Joseph's Hospital, Hamilton; and The Ottawa Hospital and Ottawa Health Research Institute, Ottawa, Ontario, Canada.

Acknowledgment: The authors acknowledge the contributions of Anne Marie Clement, Pam Stevens, and Myriel Quilacio for assistance in the study.

Grant Support: Partial funding by the Heart and Stroke Foundation of Ontario (grant NA 5048) to recruit half of the patients in the cohort.

Potential Financial Conflicts of Interest: None disclosed.

Reproducible Research Statement:Study protocol: Available from Dr. Chan (e-mail, wee-shian.chan@wchospital.ca). Statistical code and data set: Not available.

Requests for Single Reprints: Wee-Shian Chan, MD, MSc, Department of Medicine, Women's College Hospital, 76 Grenville Street, Toronto, Ontario M5G 1B2, Canada; e-mail, wee-shian.chan@wchospital.ca.

Current Author Addresses: Dr. Chan: Department of Medicine, Women's College Hospital, 76 Grenville Street, Toronto, Ontario M5G 1B2, Canada.

Dr. Lee: Hamilton Health Sciences Corporation, Henderson General Division, 711 Concession Street, Hamilton, Ontario L8V 1C3, Canada.

Drs. Spencer and Ginsberg: McMaster University Medical Centre, 1200 Main Street West, HSC-3X28, Hamilton, Ontario L8N 3Z5, Canada.

Dr. Crowther: St. Joseph's Hospital, Room L 208-4, 50 Charlton Avenue East, Hamilton, Ontario L8N 4A6, Canada.

Dr. Rodger: Division of Hematology, The Ottawa Hospital, General Campus, 501 Smyth Road, Ottawa, Ontario K1H 8L6, Canada.

Dr. Ramsay: Ottawa Health Research Institute, 725 Parkdale Avenue, Ottawa, Ontario K1Y 4E9, Canada.

Author Contributions: Conception and design: W.S. Chan, A. Lee, J.S. Ginsberg.

Analysis and interpretation of the data: W.S. Chan, A. Lee, M. Crowther, M. Rodger, T. Ramsay, J.S. Ginsberg.

Drafting of the article: W.S. Chan, A. Lee, F.A. Spencer, M. Crowther, M. Rodger, J.S. Ginsberg.

Critical revision of the article for important intellectual content: W.S. Chan, A. Lee, F.A. Spencer, M. Crowther, M. Rodger, T. Ramsay, J.S. Ginsberg.

Final approval of the article: W.S. Chan, A. Lee, F.A. Spencer, M. Rodger, T. Ramsay, J.S. Ginsberg.

Provision of study materials or patients: W.S. Chan, M. Crowther, M. Rodger, J.S. Ginsberg.

Statistical expertise: W.S. Chan, M. Rodger, T. Ramsay, J.S. Ginsberg.

Obtaining of funding: W.S. Chan, A. Lee, M. Rodger.

Administrative, technical, or logistic support: W.S. Chan, M. Rodger.

Collection and assembly of data: W.S. Chan.

Ann Intern Med. 2009;151(2):85-92. doi:10.7326/0003-4819-151-2-200907210-00004
Text Size: A A A

Background: Clinicians' assessment of pretest probability, based on subjective criteria or prediction rules, is central to the diagnosis of deep venous thrombosis (DVT). Pretest probability assessment for DVT diagnosis has never been evaluated in pregnant women.

Objective: To evaluate the accuracy of Clinicians' subjective assessment of pretest probability for DVT diagnosis and identify prediction variables that could be used for pretest probability assessment in pregnant women with suspected DVT.

Design: A cross-sectional study conducted over 7 years (March 2000 to April 2007).

Setting: 5 university-affiliated, tertiary care centers in Canada.

Patients: 194 unselected pregnant women with suspected first DVT.

Intervention: Diagnosis of DVT was established with abnormal compression ultrasonography at presentation or on serial imaging. Pretest probability by subjective assessment was recorded by thrombosis experts for each patient before knowledge of results.

Measurements: The sensitivity, specificity, negative predictive value, and likelihood ratios of subjective pretest probability assessment and their corresponding 95% CIs were calculated on the basis of the diagnosis of DVT. Patients were DVT positive if they had diagnostic compression ultrasonography at initial or serial testing or symptomatic venous thromboembolism on follow-up. Patients were DVT negative if they had negative compression ultrasonography at presentation and no venous thromboembolism on follow-up. A prediction rule for assessing DVT was derived, and an internal validation study was done to explore its performance.

Results: The prevalence of DVT was 8.8%. Clinicians' subjective assessment of pretest probability categorized patients into 2 groups: low pretest probability (two thirds of patients) with a low prevalence of DVT (1.5% [95% CI, 0.4% to 5.4%]) and a negative predictive value of 98.5% (CI, 94.6% to 99.6%), and nonlow pretest probability with a higher prevalence of DVT (24.6% [CI, 15.5% to 36.7%]). Three variables (symptoms in the left leg [L], calf circumference difference ≥2 cm [E], and first trimester presentation [Ft]) were highly predictive of DVT in pregnant patients.

Limitations: Few outcomes occurred. Altogether, 17 events were diagnosed during the study. The prediction rule derived should be validated on an independent sample before applying it to clinical practice.

Conclusion: Subjective assessment of pretest probability seems to exclude DVT when the pretest probability is low. Moreover, 3 objective variables (“LEFt”) may improve the accuracy of the diagnosis of DVT in pregnancy. Prospective validation studies are needed.

Primary Funding Source: Heart and Stroke Foundation of Ontario.





Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).


Submit a Comment
Developing prediction rules should be based on methodological guidelines
Posted on July 31, 2009
Kristel JM Janssen
University of Utrecht
Conflict of Interest: None Declared

With interest we read the article by Chan et al, in the Annals of Internal Medicine.1 This article describes the development of a prediction rule that could be used for pretest probability assessment in pregnant women with suspected DVT. Prediction rules are valuable tools in daily clinical practice, as they provide absolute risks for individual patients that can be used to decide on treatment choices, or further diagnostic work-up. However, before a prediction rule can be used safely in daily clinical practice, the methodological soundness of the development steps needs to be assured. The authors made some methodological choices that we would have advised against. We will point out these methodological issues one by one.

First, the authors included predictors in the prediction rule based on the c-statistic. Predictors were included if they increased the c-statistic by a least 0.03. However, this approach has been severely criticised, as it is less sensitive than measures based on the likelihood or other global measures of model fit.2-4 We would have advised the authors to use the likelihood to include predictors in the model, if predictor selection is to be used at all (not advised).

Second, the authors did a univariate analysis, and the predictors that were significantly associated with presence of DVT at a P value of 0.05 or less were entered into a multivariate model. Selecting predictors at such a low P value has been strongly advised against, as it leads to poor model performance when evaluated on new patients (external validation), especially in small datasets.5;6 This is particularly important in this dataset, as the authors explored too many predictors considering the low number of events: 11 predictors on 17 events. The authors acknowledged that they needed at least 5 to 10 events per predictor when they developed the rule. They state that by using an initial model with 6 predictors, this would imply that they needed between 30 and 60 events. However, the authors did not take the univariate step into account in this calculation. They should have accounted for this analysis as well, implying that for the initial 11 predictors that have been analysed, 55 to 110 events would have been needed.

Third, the authors did not use any imputation method to handle the missing values in their dataset. This means that a complete case analysis was performed, and that only the data of the patients with complete records were analysed. One of the predictors in the model (difference in calf circumference), was missing in 46 of the 194 patients. This implies that the information of only 75% of the patients was analysed. Besides a loss of power, it is widely acknowledged that ignoring the missing values in a dataset by conducting a complete case analyses may lead to biased study results.7;8 Prediction rules are valuable tools for daily clinical practice. Not surprisingly, the number of prediction rules presented in the medical literature has increased enormously (it has more than doubled between 1995 and 2005).9 The majority of these articles concern the development of prediction rules. Considering this rapid increase, we would like to emphasize the necessity to develop these prediction rules according to methodological guidelines. We would therefore like to recommend researchers that aim to develop prediction rules to use these guidelines.4 -6;10-13 A selection of these recommended references can also be found in the Information for Authors of the Annals of Internal Medicine. Reference List

(1) Chan WS, Lee A, Spencer FA, Crowther M, Rodger M, Ramsay T et al. Predicting deep venous thrombosis in pregnancy: out in "LEFt" field? Ann Intern Med 2009; 151(2):85-92.

(2) Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 2007; 115(7):928-935.

(3) Cook NR. Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin Chem 2008; 54(1):17-23.

(4) Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15(4):361-387.

(5) Steyerberg EW, Eijkemans MJ, Habbema JD. Stepwise selection in small data sets: a simulation study of bias in logistic regression analysis. J Clin Epidemiol 1999; 52(10):935-942.

(6) Steyerberg EW, Eijkemans MJ, Harrell FE, Jr., Habbema JD. Prognostic modelling with logistic regression analysis: a comparison of selection and estimation methods in small data sets. Stat Med 2000; 19(8):1059-1079.

(7) Little RJ, Rubin DB. Statistical analysis with missing data. Hoboken, New Jersey: John Wiley & Sons; 1987.

(8) Rubin DB. Multiple Imputation for Nonresponse in Surveys. Hoboken, New Jersey: John Wiley & Sons; 1987.

(9) Toll DB, Janssen KJ, Vergouwe Y, Moons KG. Validation, updating and impact of clinical prediction rules: A review. J Clin Epidemiol 2008; 61(11):1085-1094.

(10) http://biostat.mc.vanderbilt.edu/ManuscriptChecklist.

(11) http://symptomresearch.nih.gov/chapter_8.

(12) Harrell FE, Jr. Regression modelling strategies. Springer-Verlag, New York; 2001.

(13) Steyerberg EW. Clinical Prediction Models. A Practical Approach to Development, Validation, and Updating. Springer; 2009.

Response Predicting Dvt In Pregnancy: Out In "left" Field?
Posted on September 16, 2009
Wee-Shian Chan
University of Toronto
Conflict of Interest: None Declared

We appreciate Dr Janssen's comments on our article "Predicting deep venous thrombosis in pregnancy: out in "LEFt" field?" published in Ann Intern Med 2009; 151(2):85-92 [1]. Dr Janssen has raised some important and valid concerns which we will address below.

As pointed out by Dr Janssen, prediction rules in clinical practice are important to guide clinical practice in many instances as they provide absolute risks for individual patients that can be used to decide on treatment choices, or further diagnostic work-up. We would agree that prior to adopting any rule, it must be properly validated (which we have not done so in this initial study) and the methodology on which it is derived must be sound.

The first comment that the c-statistic is not the most sensitive measure of how much an additional variable contributes to estimates of individual risk may be true. However, the reader should note that it is unlikely that with an overall c-statistic for the final model with three variables of 0.943 (corresponding to the area under the receiver operator curve), that any other significant predictors were missed using this method.

We agree with the points raised in the second comments. Our low event rate in such a study reflects the reality seen in the few diagnostic studies of DVT in pregnant women. The prevalence of DVT in this particular cohort of patients is low (<10%). Despite our best efforts over seven years, we have been limited to 17 events (out of 194 patients). The rule of that one should have 5 to 10 events per variable in a multivariable logistic model is based on the fact that fewer events leads to unstable parameter estimates. Thus, even though it would have been unreasonable to attempt to fit a single model with 11 independent variables to our data, this in no way implies that there is a problem with fitting 11 single- variable models.

We acknowledge the last comment by Dr Janssen as valid. We would like to draw your readers to the two objectives of the paper: a) to determine how clinicians determine the presence of DVT in pregnant patients by subjective means and b) if there are "objective" predictors which can assist clinicians to do this. We believe that we have achieved both objectives in this paper. We share all the concerns raised by Dr Janssen and we share his concerns regarding the development of prediction rules. As we have repeatedly emphasized throughout this paper that this rule should not yet be applied in daily practice until it has been properly validated.

Among pregnant women, symptoms mimicking DVT are common (e.g. leg swelling and pain). At the very least, such a study will raise awareness that when certain "symptoms" (e.g. left leg presentation, asymmetry etc) are present, one should be more vigilant for the presence of DVT and arrange for appropriate testing.


(1) Chan WS, Lee A, Spencer FA, Crowther M, Rodger M, Ramsay T et al. Predicting deep venous thrombosis in pregnancy: out in "LEFt" field? Ann Intern Med 2009; 151(2):85-92.

Conflict of Interest:

None declared

Submit a Comment

Summary for Patients

Predicting Deep Venous Thrombosis in Pregnancy

The summary below is from the full report titled “Predicting Deep Venous Thrombosis in Pregnancy: Out in “LEFt” Field?” It is in the 21 July 2009 issue of Annals of Internal Medicine (volume 151, pages 85-92). The authors are W.S. Chan, A. Lee, F.A. Spencer, M. Crowther, M. Rodger, T. Ramsay, and J.S. Ginsberg.


Clinical Slide Sets

Terms of Use

The In the Clinic® slide sets are owned and copyrighted by the American College of Physicians (ACP). All text, graphics, trademarks, and other intellectual property incorporated into the slide sets remain the sole and exclusive property of the ACP. The slide sets may be used only by the person who downloads or purchases them and only for the purpose of presenting them during not-for-profit educational activities. Users may incorporate the entire slide set or selected individual slides into their own teaching presentations but may not alter the content of the slides in any way or remove the ACP copyright notice. Users may make print copies for use as hand-outs for the audience the user is personally addressing but may not otherwise reproduce or distribute the slides by any means or media, including but not limited to sending them as e-mail attachments, posting them on Internet or Intranet sites, publishing them in meeting proceedings, or making them available for sale or distribution in any unauthorized form, without the express written permission of the ACP. Unauthorized use of the In the Clinic slide sets will constitute copyright infringement.


Buy Now

to gain full access to the content and tools.

Want to Subscribe?

Learn more about subscription options

Related Articles
Related Point of Care
Topic Collections
Forgot your password?
Enter your username and email address. We'll send you a reminder to the email address on record.