|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 26, No 11 (April 10), 2008: pp. 1904-1905 © 2008 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.15.9012
In ReplyCancer Research UK Clinical Trials Unit, Division of Cancer Sciences, Faculty of Medicine, University of Glasgow, Glasgow, United Kingdom
Division of Oncology, Washington University School of Medicine, St Louis, MO
UNC Institute for Pharmacogenomics and Individualized Therapy, University of North Carolina, Chapel Hill, NC
Department of Oncology, Imperial College London, Hammersmith Hospital, London, United Kingdom Vach et al raise the issue of the power of our study1 to reliably identify relevant relationships between polymorphisms and toxicity. They are correct to note that by splitting our sample to provide a validation set, we have sacrificed power; however, this was done to ensure the ultimate reliability, and therefore potential clinical utility of any positive findings. Vach et al chose to assess our study against a successive 50% increase in the risk of toxicity going from wild-type homozygous, heterozygous, to homozygous variant (eg, 14%, 21% and 31.5%). This change corresponds approximately to an odds ratio (OR) of 1.7 among successive allele categories. OR is a more convenient form to express these changes and is used for the rest of this response. Our study1 generally does have good power to detect an OR of 3 (eg, 14% v 32.8% v 59.4%) between successive allele categories regardless of the incidence of the toxicity examined, as shown in Table 1.
Although we assessed the power in the validation set in Table 1 for an individual single nucleotide polymorphism (SNP) at the 5% level (two-sided) of statistical significance as in the Vach letter, we should point out that in our study we anticipated validating a prognostic score from a combination of SNPs using a test for trend. The power to validate an association in this scenario is greater than that shown in Table 1. This was in fact done for the combination of CDKN1A 10971C>T and CYP1B1*3 in our article and the power for the trend test in the validation set was 85%. This combination, however, did not approach statistical significance. We believe that the overall statistical properties of our procedure, as described above, are appropriate, and as discussed in the accompanying Editorial to our original article, do meet the current recommendations for the design of these types of study.2 There is, however, debate about what magnitude of difference it is relevant to be able to detect in order to develop a reliable predictive marker that could be adopted for routine use in a clinical setting. We would point out that to our knowledge, no previously published taxane pharmacogenetic studies actually exceed the sample size used in the development set of our study (minimum of 260 patients). Given the concerns raised by Vach et al, we have nevertheless re-examined the toxicity associations in each study arm using our whole data set and employing logistic regression. Table 2 summarizes the power of this procedure (at a 1% level of statistical significance) to detect an OR of 2 or 2.5 across the various toxicity incidences. For the toxicities with low incidence (approximately 13%) we also conducted a logistic regression on ordinal outcome using ungrouped toxicity grades, as suggested by Vach et al, to improve power.
These revised analyses only found one statistically significant association using a 10% false-discovery level. This association was between CYP1B1*3 and grade 3/4 GI toxicity in patients treated with docetaxel (12.9% in homozygous wild type, 23.2% in heterozygous and 31.2% in homozygous variant patients; P = .052 after false-discovery rate correction). This SNP was also identified in the development set in our original article, although it did not survive subsequent validation. This statistically significant association is of modest clinical significance, and moreover will require validation in another independent data set. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST The author(s) indicated no potential conflicts of interest.
ACKNOWLEDGMENTS Supported by Grant No. R21 CA113491 (St Louis, Chapel Hill, Glasgow, London), CRUK programme Grant No. C536/A2662 (Glasgow and London), and the Pharmacogenetics Research Network Grant No. U01 GM63340 (St Louis, Chapel Hill). REFERENCES
1. Marsh S, Paul J, King CR, et al: Pharmacogenetic assessment of toxicity and outcome after platinum plus taxane chemotherapy in ovarian cancer: The Scottish Randomised Trial in Ovarian Cancer. J Clin Oncol 25:4528-4535, 2007 2. Maitland ML, Ratain MJ, Cox NJ: Interpreting P values in pharmacogenetic studies: A call for process and perspective. J Clin Oncol 25:4513-4515, 2007
Related Correspondence
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2008 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
