Selection Bias and Baseline Imbalances in Randomized Trials
- PDF / 1,582,066 Bytes
- 2 Pages / 573.839 x 774.72 pts Page_size
- 38 Downloads / 166 Views
(ORRESPONDENCE
3
AUTHOR’S RESPONSE
Selection Bias and Baseline Imbalances in Randomized Trials (DrugInformation Journal, 2003;3Z293-308) 0005M38 Damion 1. McEnte~ort,MSc Statistics Group, ClinPhone Group Ltd, Nottingham, United Kingdom
Correspondence Address Damian McEntegart, Statistics Group, ClinPhone Ltd, Lady Bay House, Meadow Grove, Nottingham NG2 3Hi5 United Kingdom @-mail: dmcen&@ dinphone.com).
Essentially, my article is in agreement with Dr. Berger’sconcern about selection bias in clinical trials. Minimization of bias in all its forms is one of the principles underlying clinical trials. In my article I discuss both accidental bias caused by chance imbalance on baseline covariates and selection bias. I devote a whole section of the paper to the discussion of the predictability of various randomization schemes and advocate that it is a factor to consider in the choice of randomization scheme. Dr. Berger highlights the statements I made concerning the nature of any imbalances observed on prognostic factors. I am happy to clarify that my use of the word ‘systematic’was intended to refer to an allocation scheme that was carried out without subversion or any attempt by the investigator to predict the treatment allocation for the current patient presenting for allocation. In the case that such selection bias exists, I agree with Dr. Berger that baseline imbalances can be wholly or partially attributable to this mechanism, that is, they are not random. I am grateful for the opportunity to make this clarification. Dr. Berger takes issue with my statements regarding the inappropriate use of hypothesis testing of baseline imbalances. As he notes, I am in good company here with some influential authors taking this position, including Altman (l),
Senn ( 2 ) ,and the Consort Group (3). I believe that the usefulness of baseline hypothesis testing to detect selection bias in multicenter trials (the main focus of my article) is debatable. The primary focus of baseline comparisons in most industry trials is the overall comparison pooled across centers. Investigators may have many different reasons for selection bias that may or may not be consistent over investigators. One reason could be the investigator’s perception of the subject’s ability to respond, which is the focus of Dr. Berger’s work (4). Other motivations could include utility (younger patients may gain more quality of life years from a successful treatment than older patients), perceived tolerance of adverse events, social considerations based on income or family circumstances, prejudice, friendship, and a host of other reasons that one could imagine. These many different reasons add noise to the data and may counter-balance each other. Thus, baseline comparisons over the whole trial are unlikely to detect anything in terms of possible selection bias. Comparisons within a specific center, or at least the larger centers, may be more informative in this respect but, if one considers hypothesis testing, there is then the problem of false positives caused by
Data Loading...
