|
Activity Number:
|
277
- Recent Advances in Methods to Address Measurement Error
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
|
Abstract #326889
|
Presentation
|
|
Title:
|
Recent Developments in Modeling Nonlinear Relationships in the Presence of Measurement Error
|
|
Author(s):
|
Ruth Keogh* and Christen Gray
|
|
Companies:
|
London School of Hygiene & Tropical Medicine and London School of Hygiene & Tropical Medicine
|
|
Keywords:
|
Alcohol intake;
Epidemiology;
Fractional polynomial;
Measurement error;
Non-linear model;
Regression calibration
|
|
Abstract:
|
It is well known that random error in a single exposure measurement results in an attenuated estimate of its association with an outcome. However, this applies only when the exposure outcome association is linear (on the appropriate scale). Random measurement error causes non-linear associations to appear 'flattened'. I will discuss methods for correcting for measurement error in the non-linear setting. The most measurement error correction method is regression calibration; although this extends to non-linear associations it becomes more difficult to apply. Correction methods for non-linear models based on splines have also been developed. Fractional polynomials are an increasingly popular way of modelling non-linear associations. I will describe recent work on correcting for measurement error in fractional polynomial models using a Bayesian approach. The methods will be illustrated in a study of the association between alcohol intake and all-cause mortality in the EPIC-Norfolk cohort, using measures of alcohol intake from 7-day diet diaries. This example inludes the interesting extra challenge of complex measurement error in the form of excess zeros due to 'episodic consumers'.
|
Authors who are presenting talks have a * after their name.
