The Mayo Lung Project lung cancer mortality findings are unlikely to be biased by a volunteer effect

Autor: Pamela M. Marcus, Eric J. Bergstralh, Barnett S. Kramer, Robert S. Fontana
Rok vydání: 2011
Předmět:
Zdroj: European Journal of Epidemiology
ISSN: 1573-7284
0393-2990
Popis: ‘‘The final MLP report stated that 655 randomized non-volunteers refused screening and were excluded from the study, thus documenting violation of the rule that no selection should occur after randomization. After such selection and exclusion of candidates, an important methodological weakness, there is no guarantee that the screening and control groups at baseline were well matched for all known and unknown lung cancer risks.’’ We reviewed Woolner et al. [3] (the reference given for the final MLP report) and found that 655 participants were excluded for the reason ‘‘prevalence screening refused or incomplete’’ rather than ‘‘refused screening’’. The statement ‘‘655 randomized non-volunteers refused screening and were excluded from the study’’ should not be interpreted to mean that the 655 participants were excluded from only the intervention arm. If that had been the case, serious bias indeed would have been introduced. However, internal MLP documents indicate that the 655 participants were equally divided between the two trial arms (intervention arm—329; control arm—326). The reason for exclusion given in the internal documents was ‘‘incomplete initial screen, otherwise eligible’’. The reason for exclusion, while unclear, is inconsequential. The relevant question concerns whether the two arms, after exclusions, were balanced on potential confounders of the screeningmortality relationship. A previous report addresses that point and indicates that the MLP arms were balanced at baseline with respect to the two most powerful lung cancer risk factors, age and cigarette smoking history [4]. Furthermore, adjustment using multivariate models for those variables, as well as other potential confounders, did not change the finding of no mortality reduction with screening [4]. Confounding by unmeasured factors is a possibility, but we believe it to be quite unlikely that exclusion from the trial is a surrogate for a factor that is related to both screening and lung cancer mortality, and that the factor, should it exist, is differentially represented in 4,618 intervention arm and 4,593 control arm participants who were analyzed in the incidence portion of the MLP. P. M. Marcus (&) Division of Cancer Prevention, National Institutes of Health, National Cancer Institute, Bethesda, MD 20892-7354, USA e-mail: marcusp@mail.nih.gov
Databáze: OpenAIRE
Externí odkaz: