Bias Analysis of Current Approaches to Estimating Combined Population Attributable Risk for Multiple Risk Factors

• Activity Number: 534 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yibing Ruan* and Stephen D. Walter and Darren R. Brenner and Christine M. Friedenreich and on behalf of ComPARe Study Team
• Companies: Alberta Health Services and McMaster University and Alberta Health Services and Alberta Health Services and Alberta Health Services
• Keywords: population attributable risk; epidemiology; bias analysis

Abstract

When estimating combined population attributable risk (PAR) for multiple exposures, some researchers use Miettinen and Steenland's (M-S) method, which involves estimating the PARs of individual risk factors and calculating the combined PAR as 1-?(1-PARi). This approach assumes that the risk factors have 1) independent distributions, and 2) multiplicative relative risks (RR), i.e., no confounding or statistical interactions between factors. Although such assumptions are unrealistic in practice, no studies have so far addressed the degree of bias of the M-S approach when confounding and interactions are present. We demonstrate here that, when conditioned on given prevalences and RRs, the validity of the M-S approach involves a quadratic function that depends on the level of association and the magnitude of interaction between the risk factors. We further show that, under most circumstances, the bias in the M-S approach increases when the interaction departs from multiplicity, while it is less affected by the degree of association between the factors. We urge investigators to use the M-S approach cautiously, especially when interactions are present.