To the Editor:
Wunderink et al (November 2003) (1) claim that initial therapy with linezolid was associated with significantly better survival and clinical cure rates compared to that with vancomycin in patients with nosocomial pneumonia (NP) due to methicillin-resistant Staphylococcus aureus (MRSA). However, the authors based their conclusions on subgroup analyses. (2,3)
The use of subgroup analyses to draw conclusions is associated with several difficulties. (4) Past hoc subgroups are not randomized. Nonrandomized data are more likely to reach false-positive conclusions than are randomized data. (5) Randomization controls for unmeasured and unknown factors, as well as for measured factors. While the patient characteristics of the entire population and the MRSA subset appear to be similar in Table 1 of the article, this does not account for other potential unmeasured or unknown factors. The imbalance in treatment groups favoring linezolid therapy in the MRSA subgroup, most notably cardiac disease and diabetes, may influence outcome and survival. Logistic regression may actually, increase bias when adjusting for baseline variables in analyzing nonrandomized data. (6)
When the primary end point in a trial shows similar efficacy for two drugs but a subgroup analysis shows superiority for one of the drugs, it follows that there also is a subgroup in which the other drug must show an advantage. In this article, there is no difference in the mortality or clinical cure rates for all patients with NP or S aureus (SA) NP despite the claimed advantage for linezolid therapy in patients with MRSA NP, This translates into higher survival rates (SS, 1% [67 of 76 patients] vs 76.3% [71 of 93 patients], respectively) and clinical success rates (50% [37 of 74 patients] vs 45% [34 of 75 patients], respectively) for therapty with vancomycin compared to that with linezolid in patients with methicillin-susceptible S aureus (SA) NP. One could question the biological plausibility of this discrepancy, as the in vitro activity of both drugs is similar between MRSA and methicillin-susceptible SA.
It is unclear whether any of these subgroup analyses merely represents chance findings. The authors' subgroup analyses are based on a p value of 0,05 as being statistically significant. A p value of 0.05 means that there is a probability that 1 of 20 comparisons in a clinical trial may represent a false-positive conclusion (ie, an [alpha] or type 1 error). When making multiple comparisons from subgroup analyses, the likelihood of accepting a false-positive conclusion increases. (7) For 10 comparisons, the risk of accepting a false-positive conclusion increases from 5 to 40%. For this reason, it is appropriate to adjust for multiple comparisons using a p value of < 0.05 to indicate statistical significance. (8) This correction is based on the number of comparisons made, not the number of comparisons presented. It is unclear how many comparisons the authors made when evaluating subsets for this article. Regardless, the authors do not present any correction for multiple comparisons.
The results of the trials of linezolid (2,3) support the conclusion that the efficacy of linezolid and vancomycin are similar in patients with NP. Post hoc subgroup analyses can raise hypotheses that require confirmation in other studies before general acceptance. Meanwhile, clinicians should exercise care in drawing conclusions based on subgroup analyses alone. Trials of MRSA NP may have logistical difficulties, but this does not justify accepting conclusions or basing future guidelines for patient care on less than optimal data.
John H. Powers, MD
David B. Ross, MD, PhD
Daphne Lin, PhD
Janice Soreth, MD
US Food and Drug Administration
Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (e-mail: firstname.lastname@example.org).
Correspondence to: John H. Powers, MD, HFD-104, 9201 Corporate Blvd, Rockville, MD 20850; e-mail: POWERSJOH@ cder.fda.gov
(1) Wunderink RG, Rello J, Cammarata SK, et al. Linezolid vs vancomycin: analysis of two double-blind studies of patients with methicillin-resistant Staphylococcus aureus nosocomial pneumonia. Chest 2003; 124:1789-1797
(2) Rubinstein E, Cammarata S, Oliphant T, et al. Linezolid (PNU-100766) versus vaneomyein in the treatment of hospitalized patients with nosocomial pneumonia: a randomized, double-blind, multicenter study. Clin Infect Dis 2001; 32: 402-412
(3) Wunderink RG, Cammarata SK, Oliphant TH, et al. Continuation of a randomized, double-blind, multicenter study of linezolid versus vancomycin in the treatment of patients with nosocomial pneumonia. Clin Ther 2003; 25:980-992
(4) Freemantle N. Interpreting the results of secondary end points and subgroup analyses in clinical trials: should we lock the crazy aunt in the attic? BMJ 2001; 322:989-991
(5) Chalmers TC, Celano P, Sacks HS, et al. Bias in treatment assignment in controlled clinical trials. N Engl J Med 1983; 309:1358-1361
(6) Decks JJ, Dinnes J, D'Amico R, Sowden AJ, et al. Evaluating non-randomised intervention studies. Health Technol Assess 2003; 7:iii-x, 1-173
(7) Brookes ST, Whitley E, Peters TJ, et al. Subgroup analyses in randomised controlled trials: quantifying the risks of false-positives and false-negatives, Health Technol Assess 2001; 5:1-56
(8) Bland JM Altman DG. Multiple significance tests: the Bonferroni method. BMJ 1995; 310:170
To the Editor:
Powers and colleagues raise pertinent and well-known concerns regarding the analysis of subgroups in clinical trials. However, we find it surprising to he criticized by members of the United States Food and Drug Administration (FDA) Center for Drug Evaluation and Research for these particular issues.
The use of subgroup analysis is a standard operating procedure for the FDA Center for Drug Evaluation and Research. The actual FDA approval for linezolid is based on subgroup analyses of "clinically evaluable" and "microbiologically evaluable" subgroups of the entire cohort. (1) The FDA approved the use of drotrecogin alfa only for one of the many subgroups analyzed (APACHE [acute physiology and chronic health evaluation] II score, [greater than or equal to] 25), despite the overall benefit in the entire population. (2) No statistical "hit" for multiple comparisons was used by FDA statisticians in making this recommendation. The FDA also gave levofloxacin an indication for penicillin-resistant Streptococcus pneumoniae pneumonia based on 14 patients who were not only not randomized, but were scattered among multiple studies that included both open-label and closed-label, phase III and phase IV, and even observational formats. Therefore, while not ideal, subgroup analysis is used routinely for important decisions, including the approval of investigational agents, in which optimal data are not available.
Powers and colleagues are also incorrect in stating that, if one subgroup shows superiority, then another subgroup must show inferiority if the overall difference is not statistically different. The problematic issue is "statistical" significance. If only a small subgroup has shown a highly significant difference and the remainder of the group demonstrated no difference, the benefit in the small subgroup will be overwhelmed statistically. This was in fact the primary hypothesis of the analysis that we published in CHEST (November 2003). (3) Specifically, the: significant differences in the response to the two agents used in treating methicillin-resistant Staphylococcus aureus (MRSA) pneumonia were being obscured in the overall analysis of the two studies. (1,4) (As an aside, Powers and colleagues miscalculated the mortality rate for MRSA patients treated with vancomycin. The actual numbers are 67 of 86 patients, giving a survival rate of 77.9%, which is not statistically better than the 76.3% achieved with linezolid, even before adjusting for multiple comparisons.)
Powers and colleagues also express concern that clinicians should exercise care in drawing conclusions regarding our subgroup analysis. It is precisely our clinical concern about conclusions drawn from the subgroup analysis performed by the FDA in approving linezolid that prompted this article. Clinicians do not need to know how a drug works for genetic Gram-positive pneumonia. The only cause of Grain-positive pneumonia that is pertinent to a comparison of linezolid and vancomycin is MRSA. Multiple other antibiotics are available and should be preferentially used to treat other Gram-positive infections, including MRSA. Therefore, the rest of the study has little bearing on clinical medicine.
The logical extension of the recommendation of Powers and colleagues is that clinicians should use quinupristin/dalfopristin for treating MRSA pneumonia also, since an FDA-approved randomized clinical trial. (5) in patients with Gram-positive pneumonia showed equivalence to vancomycin. This is despite a clinical success rate for quinupristin/dalfopristin of only 18% in the subgroup with MRSA pneumonia. The pharmaceutical company made a wise decision to not pursue an FDA indication for nosocomial Gram-positive pneumonia, even though it was based on a subgroup analysis with all the flaws described by Powers and colleagues.
The hard data are that more patients with MRSA pneumonia who were randomized to therapy with vancomycin died than did patients randomized to therapy with linezolid. The difference is clinically significant, as well as statistically significant. The number needed to treat with linezolid to save one life compared with vancomycin is only six patients. Readers can choose to discount this fact because it is a post hoc analysis or because the diffences do not meet strict statistical criteria. They can hypothesize that the findings are due to differences in baseline characteristics rather than treatment differences, despite there being no statistically significant differences in the baseline characteristics that we analyzed and the retention of treatment group as a significant factor in the multivariate analysis. Readers can hypothesize that other treatment or microbiological differences existed, despite the same pattern seen in patients with bacterermia and in patients who received diagnoses that utilized invasive techniques. But, the fact remains that more patients with MRSA pneumonia who were randomized to therapy with vancomycin (tied than did patients randomized to therapy with linezolid. Therefore, we think that the conclusion that linezolid "is associated" with lower mortality rates and greater clinical success rates than vancomycin in patients with MRSA pneumonia is valid.
We, as the nonemployee authors of these manuscripts, and many other clinicians who have been presented these data have raised the same concerns about subgroup analysis as have Powers and colleagues, Because of these concerns, Pharmacia (now Pfizer Inc) has sponsored a direct comparison of vancomycin and linezolid specifically in patients with MRSA ventilator-associated pneumonia. Until this and possibly other studies are completed, we agree with Powers and colleagues that clinicians should exercise care in drawing conclusions based on subgroup analysis. We differ with them, however, in that we think that the analysis of the MRSA subgroup that we published is more pertinent to the clinical care of patients with MRSA pneumonia than the "clinically evaluable" and "microbiologically evaluable" subgroup analyses of the entire cohort favored by Powers and colleagues.
Richard G. Wunderink, MD, FCCP
Northwestern University Feinberg School of Medicine
Marin Kollef MD, FCCP
Washington University School of Medicine
St. Louis, MO
Jordi Rello, MD
University Rovira I Virgili
All three authors have received research support from Pharmacia and Pfizer Inc, and are consultants to Pfizer Inc.
Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (e-mail email@example.com).
Correspondence to: Richard G. Wunderink MD FCCP Pulmonary and Critical Care Division, Northwestern University Feinberg School of Medicine, 676 St. Clair St, Suite 14-044, Chicago IL 60611; e-mail: firstname.lastname@example.org
(1) Rubinstein E, Cammarata S, Oliphant T, et al. Linezolid (PNU-100766) versus vancomycin in the treatment of hospitalized patients with nosocomial pneumonia: a randomized, double-blind, multicenter study. Clin Infect Dis 2001; 32: 402-412
(2) Bernard GR, Vincent JL, Laterre PF et al. Efficacy and safety of recombinant human activated protein C for severe sepsis. N Engl J Med 2001; 344:699-709
(3) Wunderink RG, Rello J, Cammarata SK, et al, Linezolid vs vancomycin: analysis of two double-blind studies of patients with methicillin-resistant Staphylococcus aureus nosocomial pneumonia. Chest 2003; 124:1789-1797
(4) Wunderink RG, Cammarata SK, Oliphant TH, et al. Continuation of a randomized, double-blind, multicenter study of linezolid versus vancomycin in the treatment of patients With nosocomial pneumonia, Clin Ther 2003; 25:980-992
(5) Fagon J, Patrick H, Haas DW, et al. Treatment of gram-positive nosocomial pneumonia: prospective randomized comparison of quinupristin/dalfopristin versus vancomycin; Nosocomial Pneumonia Group. Am J Respir Crit Care Med 2000; 161:753-762
COPYRIGHT 2004 American College of Chest Physicians
COPYRIGHT 2004 Gale Group