© 2007 American Thoracic Society
On the Correct Statistical Analysis of Exacerbation Rates in Clinical TrialsTo the Editor:Kardos and colleagues, in their randomized trial assessing the effect of salmeterol/fluticasone propionate (SFC) versus salmeterol (SAL) on exacerbations in the treatment of severe chronic obstructive pulmonary disease (COPD), report a rate ratio of exacerbation of 0.65 (95% confidence interval, 0.57–0.76; P < 0.0001) with SFC relative to SAL (1). Although the rate ratio is accurate, the P value and confidence interval are not. The authors correctly recognize that the statistical analysis has to account for both "the (duration of) treatment exposure of each patient and for between-patient variation in exacerbation rates," in referring to my previous article on the subject (2). Yet, the authors appear to use the standard method of Poisson regression, which only accounts for the variable duration of treatment exposure, to produce the correct weighted estimate of the rate ratio, as suggested by their statement that their "study accounted a priori for differential dropout using Poisson regression with time in study as an offset variable." This offset does not per se account for between-patient variation in exacerbation rates, which can be considerable in COPD. To account for this "extra-Poisson" variability between patients, an overdispersion parameter must also be included along with the Poisson regression analysis (2). If not, the P values and confidence intervals are seriously underestimated since the method will then assume that the exacerbation rates of all patients in the SFC group are identical and all equal to the reported 0.92 per year, while those of the SAL group are all equal to 1.4, with no variation between patients. This leads to unusually low P values such as those of less than 0.0001 reported by the authors (2). The correct approach, which uses the overdispersion parameter, has been used in several trials in COPD (3–5). It should be applied to these data to provide the correct values that may be necessary in future meta-analyses.
McGill University Health Centre, Montreal, Canada FOOTNOTES Conflict of Interest Statement: S.S. has been reimbursed for attending advisory board meetings and conferences, and participating as a speaker in scientific meetings financed by various pharmaceutical companies (AstraZeneca, Boehringer Ingelheim, GlaxoSmithKline, Pfizer, and Sepracor); he has received funding for research grants from AstraZeneca and GlaxoSmithKline. REFERENCES
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
