Attributable Fractions and Excess Fractions with Multiple Exposure Level: The Relations and Bounds

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Yasutaka Chiba*+
• Keywords: Causal Inference ; Potential Outcome ; Principal Stratification

Abstract

Suzuki et al. (American Journal of Epidemiology 2012; 175: 567-575) discussed the relations among attributable and excess fractions in the case of a binary exposure, and showed that (i) attributable and excess caseloads (AC and EC) are equivalent to attributable and excess proportions (AP and EP) under the monotonicity assumption, respectively, and (ii) when the target population is the exposed group, AC and AP are equivalent to EC and EP, respectively. In this presentation, we extend their results to the case of a multiple exposure, and show that their results are also achieved in the case of a multiple exposure. In addition, we derive bounds for these four fractions. The bounds are derived with (i) no assumption, (ii) the assumption of monotone treatment response, (iii) the assumption of monotone treatment selection, which were introduced by Manski and Pepper (Econometrica 2000; 68: 997-1010), and (iv) their combination. Although AC and EC can be identified under the assumption of no unmeasured confounder, the AP and EP cannot be identified even under this assumption. Therefore, we also derive the bounds for AP and EP under this assumption.