Rapid Responses to:

EDITORIALS: Peter C Gøtzsche Why we need a broad perspective on meta-analysis BMJ 2000; 321: 585-586 [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

We need a broader perspective on data interpretation.

David W Noble   (7 November 2000)

We need a broader perspective on data interpretation. 7 November 2000
David W Noble Send response to journal: Re: We need a broader perspective on data interpretation.	Editor - Although Gotzsche's editorial appears to call for a broad perspective, in reality it does the opposite, and is an injustice to the complexities of critical appraisal and interpretation of data.1 Ideally, we would all like to view the truth from the Universal Frame of Reference. However even from that frame, the risks and benefits for a given treatment are likely to be different for different patient populations and different indications, be that for albumin, penicillin, tPA or any other treatment. In the absence of very good, reliable and robust evidence the totality of information from animal data, personal experience, audit, observational studies and non-randomized clinical trials must be weighed and integrated with all other critically appraised evidence as guidance as to what is probably best for an individual patient. In this context the albumin example cited by Gotzsche is also problematic.2 It is stated as fact that albumin is harmful. This very narrow and questionable interpretation of a controversial systematic review. The limitations of the study, as described by the authors and in the extensive subsequent correspondence as well as the history of albumin in resuscitation are ignored.3,4 Why? With regard to hypovolaemia others have commented on the inclusion of patients who clearly were not hypovolaemic and this casts doubt on the clinical eligibility criteria of these studies.5 The statistical methodology used is also questionable in that there was no adjustment for multiple comparisons of retrospective data. Various authorities also opine that, for these reasons, a P value of 0.05 is not conservative enough in meta-analysis.6-8 Moreover, it has been suggested that definitive conclusions should not be made unless the number of patients in the meta-analysis exceed the number of patients that would be required in a single, properly powered randomized controlled trial.8 The data for hypovolaemia fall well short of these criteria using standard statistical methods far less Bayesian approaches to data interpretation.9 This useful but imperfect systematic review can therefore only be one small piece in the jigsaw and alone should not drive clinical practice.10 Clinicians must critically appraise and weigh the totality of evidence to best serve their patients. Appraisal and weighting of evidence necessarily remains a subjective hybrid of art and science. It is not completely objective and individuals or organisations are deluding themselves if they think otherwise. A truly broad perspective therefore remains paramount. Dr David W Noble Consultant in Anaesthesia and Intensive Care Department of Anaesthetics, Intensive Care Medicine & Hyperbaric Medicine, Aberdeen Royal Infirmary, Aberdeen AB25 2ZN, Scotland Email David.Noble@arh.grampian.scot.nhs.uk References. [1]. Gotzsche P. Why we need a broad perspective on meta-analysis. BMJ 2000; 321: 585-6. [2]. Cochrane Injuries Group Albumin Reviewers. Human albumin administration in critically ill patients. BMJ 1998; 317: 235-40. [3]. Peters M. Albumin in meningococcal disease. Hospital Medicine 2000; 61: 80-1. [4]. Judkins K. Burns resuscitation: what place albumin? Hospital Medicine 2000; 61: 116-9. [5]. Is albumin harmful? Drummond GB, Ludlam CA. British Journal of Haematology 1999; 106: 266-9. [6]. Peto R. Why do we need systematic overviews of randomized trials? Statistics in Medicine 1987; 6: 233-44. [7]. Goodman SN. Towards evidence-based statistics. 2: The Bayes factor. Annals of Internal Medicine 1999; 130: 1005-13. [8]. Pogue J, Yusuf S. Overcoming the limitations of current meta- analysis of randomised controlled trials. Lancet 1998; 351: 46-52. [9]. Spiegelhaler D J, Myles JP, Jones DR, Abrams KR. An introduction to Bayesian methods in health technology assessment. BMJ 1999; 319: 508-512. [10]. Barton S. Which clinical studies provide the best evidence? BMJ 2000; 321: 255-6.