Rheumatology 2004; 43: 680-681
Rheumatology Vol. 43 No. 5 (c) British Society for Rheumatology 2004; all rights reserved
Letter to the Editor |
Re: Layton et al. Comparison of the incidence rates of selected gastrointestinal events reported for patients prescribed rofecoxib and meloxicam in general practice in England using prescription-event monitoring data
Boehringer Ingelheim GmbH, Medical Affairs, Ingelheim am Rhein, Germany, 1UZ St Rafael, Biostatistical Centre, Leuven, Belgium, 2Medizinische Hochschule Hannover, Abteilung Rheumatologie, Hannover, Germany
Correspondence to F. Degner. E-mail: degner{at}ing.boehringer-ingelheim.com
SIR, Both papers by Layton et al. [1, 2] address an important aspect of providing evidence on the safety of new drugs under real prescribing conditions. The papers report on the results of three prescription-event monitoring (PEM) studies and compare the incidence rates of selected gastrointestinal (GI) events of meloxicam with those of both rofecoxib and celecoxib. The activities to analyse such data from the actual prescribing setting need to be applauded, in particular due to the inherent difficulties in such an uncontrolled setting to control for multiple potential bias. The authors also address these difficulties in concluding that the results of their studies are only useful if evaluated together with results from other studies.
We feel it is important to highlight some potential biases that have not been addressed in the three studies which refer to data collected for meloxicam between December 1996 and March 1997 while data from cohorts of rofecoxib and celecoxib were collected between 1999 and 2000.
The compilation of the overall findings from both studies seems to show similar crude incidence rate estimates for upper GI complications (0.73–0.90% per yr) obtained for the three drugs monitored. For both the symptomatic upper GI events and the complicated upper GI conditions the Kaplan–Meier estimates showed no difference in both studies in the log rank tests between meloxicam and celecoxib as well as between meloxicam and rofecoxib (P = 0.61, P = 0.30 and P = 0.85, P = 0.87 respectively). Likewise similar crude rate estimates for those aged between 60 and 79 yr (0.6–0.8% per yr) and those aged 80+ yr (2–2.5% per yr) were observed, showing the well-known effect of age as a risk factor for GI complications. The estimates of crude incidence rates for upper GI complications for patients with another well-known risk factor, a history of upper GI problems, are also similar between drugs ranging from 0.9–1.1% per yr (Table 1).
|
However, we have some difficulties regarding the regression models presented. For instance, for at least four other important risk factors, namely concomitant aspirin use, concomitant steroid use, NSAID dose used and disability index, no adjustment was done in the analyses reported. Further, the effects of different types of gastroprotective therapy (e.g. H2-receptor antagonists/proton-pump inhibitors (H2RA/PPI) and misoprostol) were not taken into account. Furthermore, the model selection is not described very well and appears arbitrary as there is no information given on the list of covariates screened for being non-confounding factors as well as on the evaluation of the treatment by covariate interaction. The interpretation of age2 as a confounding factor seems to be lacking. It is known from the statistical literature that when covariate adjustment is not pre-specified the results from a regression approach can vary wildly [3–6]. Hence, it is questionable to us that the results obtained so far are really set in stone.
Another problem is the significant number of missing values, especially regarding age and NSAID history. The authors performed a sensitivity analysis by leaving out more patients (retaining only those patients with no missing value anywhere) and show that the relative risk (RR) does not vary too much. However, this does not tell us anything about the validity of their approach. Indeed, it could well be possible that the authors have performed an unadjusted analysis, found no significant RR and then continued regressing on more covariates (until they found a significant result?). Thus, it seems odd to us to conclude from their exercise that the significant RR was not generated because of missing values. Indeed, the significant RR of 0.56 for celecoxib versus meloxicam can still be the result of a selection bias because the subgroup upon which this significant RR is based is probably not a random sample of the original patient population.
We wonder how the results would look if one were to take the above factors into consideration when performing statistical analyses.
F. Degner is an employee of Boehringer Ingelheim. E. Lesaffre has collaborated with B.I. in various projects. H. Zeidler has received speakers’ honoraria from Boehringer Ingelheim, Merck Sharp & Dohme, Pfizer, Pharmacia and Novartis.
References
- Layton D, Heeley E, Hughes K, Shakir SAW. Comparison of the incidence rates of selected gastrointestinal events reported for patients prescribed rofecoxib and meloxicam in general practice in England using prescription-event monitoring data. Rheumatology 2003;42:622–31.
[Abstract/Free Full Text] - Layton D, Hughes K, Harris S, Shakir SAW. Comparison of the incidence rates of selected gastrointestinal events reported for patients prescribed celecoxib and meloxicam in general practice in England using prescription-event monitoring (PEM) data. Rheumatology 2003;42:1332–41. First published June 16, 2003:10/1093/rheumatology/keg376.
[Abstract/Free Full Text] - Beach ML, Meier P. Choosing covariates in the analysis of clinical trials. Control Clin Trials 1989;10:161S–175S.[Medline]
- Senn S. Testing for baseline balance in clinical trials. Statist Med 1994;13:1715–26.
- Raab GM, Day S, Sales J. How to select covariates to include in the analysis of a clinical trial. Control Clin Trials 2000;21:330–42.[CrossRef][Web of Science][Medline]
- Assmann SF, Pocock SJ, Enos LE, Kasten LE. Subgroup analysis and other (mis)uses of baseline data in clinical trials. Lancet 2000;355:1064–9.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||