0
Correspondence |

Assessing the Performance of the HAS-BLED Score: Is the C Statistic Sufficient? FREE TO VIEW

Ulisse Corbanese, MD
Author and Funding Information

From the Department of Anesthesia and Intensive Care, Ospedale S. Maria dei Battuti.

Correspondence to: Ulisse Corbanese, MD, Department of Anesthesia and Intensive Care, Ospedale S. Maria dei Battuti, Via B. Bisagno, 31015 Conegliano, Italy; e-mail: ucorbanese@hotmail.com


Financial/nonfinancial disclosures: The author has reported to CHEST that no potential conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article.

Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (http://www.chestpubs.org/site/misc/reprints.xhtml).


© 2011 American College of Chest Physicians


Chest. 2011;139(5):1247-1248. doi:10.1378/chest.10-2995
Text Size: A A A
Published online

To the Editor:

I read with interest the article by Pisters et al1 in CHEST (November 2010) that addresses the problem of major bleeding in patients with atrial fibrillation during treatment with oral anticoagulant drugs. The authors suggest a new user-friendly score in order to assess the risk of major bleeding and, possibly, to support clinical decision making about antithrombotic therapy in patients with atrial fibrillation. They evaluated the predictive accuracy of the model by using the C statistic. Although the C statistic should offer a simple and intuitive measure of the accuracy of predictions using a single test, several readers might not be so familiar with it. Conventionally, the evaluation of a new scoring system is performed using calibration, discrimination and, to a lesser extent, classification measures (mainly sensitivity and specificity), and likelihood ratios.2-4

The calibration is generally assessed with the Hosmer-Lemeshow goodness-of-fit test,5 which is a summary measure of the model’s ability to predict outcome for groups of patients having different levels of risk. Patients are rank-ordered according to outcome probability; they then are divided into deciles of risk. Expected and observed outcomes are compared within each decile of risk. The results of comparisons for each cell of the contingency table are summed, and that result is compared with the χ2 distribution: P values larger than .05 demonstrate adequate model calibration across the entire range of risks.

On the other hand, discrimination is commonly evaluated using the area under the receiver operating characteristic curve.2-4 The area under the receiver operating characteristic curve summarizes in a single number the overall discrimination across the range of risks, independently of disease prevalence and without loss of information due to the choice of a particular decision criterion, as happens for classification measures.2,3 The area can range from 0.5 to 1.0; 0.7 is considered the minimal value acceptable in the validation of a model. Finally, although the role of classification measures in the assessment of the performance of scoring systems has been questioned,2,3 most readers are familiar with these measures, particularly with sensitivity and specificity. In my opinion, such an interesting article could have benefited if the authors had reported in a table at least some of the above-mentioned time-honored statistics, making easier the interpretation of the performance results of the model, as well as its comparison with other scoring systems.

Pisters R, Lane DA, Nieuwlaat R, de Vos CB, Crijns HJGM, Lip GYH. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation: the Euro Heart Survey. Chest. 2010;1385:1093-1100. [CrossRef] [PubMed]
 
Lemeshow S, Le Gall JR. Modeling the severity of illness of ICU patients. A systems update. JAMA. 1994;27213:1049-1055. [CrossRef] [PubMed]
 
Lett RR, Hanley JA, Smith JS. The comparison of injury severity instrument performance using likelihood ratio and ROC curve analyses. J Trauma. 1995;381:142-148. [CrossRef] [PubMed]
 
Sackett DL, Haynes RB, Tugwell P. Clinical Epidemiology: A Basic Science for Clinical Medicine. 1985;1st ed Boston Little, Brown & Co:59-138
 
Hosmer DW, Lemeshow S. Applied Logistic Regression. 1989; New York, NY John Wiley & Sons Inc
 

Figures

Tables

References

Pisters R, Lane DA, Nieuwlaat R, de Vos CB, Crijns HJGM, Lip GYH. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation: the Euro Heart Survey. Chest. 2010;1385:1093-1100. [CrossRef] [PubMed]
 
Lemeshow S, Le Gall JR. Modeling the severity of illness of ICU patients. A systems update. JAMA. 1994;27213:1049-1055. [CrossRef] [PubMed]
 
Lett RR, Hanley JA, Smith JS. The comparison of injury severity instrument performance using likelihood ratio and ROC curve analyses. J Trauma. 1995;381:142-148. [CrossRef] [PubMed]
 
Sackett DL, Haynes RB, Tugwell P. Clinical Epidemiology: A Basic Science for Clinical Medicine. 1985;1st ed Boston Little, Brown & Co:59-138
 
Hosmer DW, Lemeshow S. Applied Logistic Regression. 1989; New York, NY John Wiley & Sons Inc
 
NOTE:
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).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543