This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Bonetti, M. Articles by Gelber, R. D. Search for Related Content PubMed PubMed Citation Articles by Bonetti, M. Articles by Gelber, R. D. Social Bookmarking                            What's this?

Another STEPP in the Right Direction

Department of Decision Sciences, Bocconi University, Milan, Italy

Bernard F. Cole

Department of Mathematics and Statistics, University of Vermont, Burlington, VT

Richard D. Gelber

Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Boston, MA

To the Editor:

We welcomed the editorial by Royston and Sauerbrei,¹ which highlighted the importance of evaluating the magnitude of treatment effect differences as a function of continuously measured covariates in clinical research. Our work in this area includes the development of the subpopulation treatment effect pattern plot (STEPP),^2,3 a statistical method that assesses treatment effects for overlapping subgroups of patients defined by a covariate of interest such as age. STEPP relies on estimates obtained from standard survival curves, such as those produced with the Kaplan-Meier methodology.³ The approach recommended by Royston and Sauerbrei is based on models of multivariable fractional polynomial interaction (MFPI) and searches over a set of possible functional forms for the relationship between the covariate and the survival end point.^4,5

Royston and Sauerbrei faulted STEPP because, in contrast to MFPI, STEPP does not provide a single numerical measure of an interaction. We consider this a strength, however, because STEPP avoids distilling potentially complicated patterns of effect modification into a single number. Instead, graphical methods are used to display these patterns, showing the clinically relevant absolute treatment effect as the covariate ranges from low to high values. These graphs are readily interpretable by clinicians, as illustrated in the recent analysis of Viale et al,⁶ which evaluated the predictive value of estrogen and progesterone receptors in breast cancer. This is similar to the use of Kaplan-Meier survival curves for comparing treatments. There is no single summary measure, yet the usefulness of displaying estimated survival curves is well recognized.

Though we agree that STEPP is useful as an exploratory tool, we disagree with the suggestion that its primary role is to check MFPI results. This is analogous to preferring a parametric model (eg, exponential) to estimate survival distributions and using Kaplan-Meier estimates to check the model fit. Clearly, there are advantages in directly estimating effects using nonparametric methods. Of course, we support the continued development of both parametric (or semiparametric) and nonparametric methods and point out that another useful analytic approach has been developed by the National Surgical Adjuvant Breast and Bowel Project statistical group.⁷

Finally, we note that the study of treatment-covariate interactions typically requires large sample sizes to produce reliable results. Even in a large-scale clinical trial, subtle levels of effect modification may be difficult to detect with STEPP or MFPI. We are currently extending STEPP to evaluate the magnitude of treatment effect differences in meta-analyses.

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The author(s) indicated no potential conflicts of interest.

REFERENCES

1. Royston P, Sauerbrei W: Interactions between treatment and continuous covariates: A step toward individualizing therapy. J Clin Oncol 26:1397-1399, 2008[Free Full Text]

2. Bonetti M, Gelber RD: A graphical method to assess treatment-covariate interactions using the Cox model on subsets of the data. Stat Med 19:2595-2609, 2000[CrossRef][Medline]

3. Bonetti M, Gelber RD: Patterns of treatment effects in subsets of patients in clinical trials. Biostatistics 5:465-481, 2004[Abstract]

4. Royston P, Sauerbrei W: A new approach to modeling interactions between treatment and continuous covariates in clinical trials by using fractional polynomials. Stat Med 23:2509-2525, 2004[CrossRef][Medline]

5. Sauerbrei W, Royston P, Sapient K: Detecting an interaction between treatment and a continuous covariate: A comparison of two approaches. Comp Statistic Data Anal 51:4054-4063, 2007[CrossRef]

6. Viale G, Regan MM, Maiorano E, et al: Chemoendocrine versus endocrine adjuvant therapies for node-negative breast cancer: Predictive value of centrally reviewed expression of estrogen and progesterone receptors—International Breast Cancer Study Group. J Clin Oncol 26:1404-1410, 2008[Abstract/Free Full Text]

7. Jeong JH, Costantino JP: Application of smoothing methods to evaluate treatment-prognostic factor interactions in breast cancer data. Cancer Invest 24:288-293, 2006[CrossRef][Medline]

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Bonetti, M. Articles by Gelber, R. D. Search for Related Content PubMed PubMed Citation Articles by Bonetti, M. Articles by Gelber, R. D. Social Bookmarking                            What's this?