Sensitivity analysis for an unmeasured confounder: a review of two independent methods

Revista Brasileira de Epidemiologia

Luiz, Ronir Raggio²  Cabral, Maria Deolinda Borges¹ 

[1] Instituto Brasileiro de Geografia e Estatística;Universidade Federal do Rio de Janeiro

关键词: Sensitivity analysis;    Unmeasured confounder;    Confounding;    Observational studies;   

DOI  :  10.1590/S1415-790X2010000200002

学科分类：过敏症与临床免疫学

来源: SciELO

【 摘 要 】

One of the mainpurposes of epidemiological studies is to estimate causal effects. Causal inferenceshould be addressed by observational and experimental studies. A strong constraintfor the interpretation of observational studies is the possible presence ofunobserved confounders (hidden biases). An approach for assessing the possibleeffects of unobserved confounders may be drawn up through the use of a sensitivityanalysis that determines how strong the effects of an unmeasured confoundershould be to explain an apparent association, and which should be the characteristicsof this confounder to exhibit such an effect. The purpose of this paper is toreview and integrate two independent sensitivity analysis methods. The two methodsare presented to assess the impact of an unmeasured confounder variable: onedeveloped by Greenland under an epidemiological perspective, and the other developedfrom a statistical standpoint by Rosenbaum. By combining (or merging) epidemiologicaland statistical issues, this integration became a more complete and direct sensitivityanalysis, encouraging its required diffusion and additional applications. Asobservational studies are more subject to biases and confounding than experimentalsettings, the consideration of epidemiological and statistical aspects in sensitivityanalysis strengthens the causal inference.

【 授权许可】

CC BY