Estimation of Measurement Error for Self-Reported Physical Activity
Principal Investigator
Name
Mithat Gonen
Degrees
PhD
Institution
Memorial Sloan Kettering Cancer Center
Position Title
Attending
Email
About this CDAS Project
Study
IDATA
(Learn more about this study)
Project ID
IDATA-6
Initial CDAS Request Approval
Oct 19, 2016
Title
Estimation of Measurement Error for Self-Reported Physical Activity
Summary
The goal of this project is to obtain estimates of measurement error for physical activity self-reported by the participants using as gold standard the objective monitors..
This is important for two reasons: First, obtaining objective measure is costly, inconvenient and time consuming, as a result it is not possibly to routinely have objective measures in clinical trials or observational studies where physical activity is recorded either as an outcome or as a confounder. Understanding how much self-reported activity deviates from objective monitors will give researchers a tool in deciding what measures can be incorporated with confidence in future studies.
Second reason is that statistical methods for analyzing data which are subject to measurement error are developed over the past decade. These methods have the advantage of incorporating a measurement error model into the analysis and producing conclusions that adjust for the presence of measurement error. To use these methods, one has to assume or derive a measurement error model. This project will enable us to develop one or more models that can inform future analyses. While most measurement error models used for this purpose are simple models of convenience (such as assuming normal distributions), use of IDATA will allow us to more precisely estimate the distribution of measurement error, free from most of the assumptions that are typically made.
Our plan is to first evaluate the differential between self-reported and objectively measured physical activity levels. Then we will aggregate this information across subjects, and possibly across time points, to estimate a distribution for measurement error. We will investigate if the measurement error is a function of patient characteristics, objective measures or other factors. This may lead to development of multiple measurement error models.
This is important for two reasons: First, obtaining objective measure is costly, inconvenient and time consuming, as a result it is not possibly to routinely have objective measures in clinical trials or observational studies where physical activity is recorded either as an outcome or as a confounder. Understanding how much self-reported activity deviates from objective monitors will give researchers a tool in deciding what measures can be incorporated with confidence in future studies.
Second reason is that statistical methods for analyzing data which are subject to measurement error are developed over the past decade. These methods have the advantage of incorporating a measurement error model into the analysis and producing conclusions that adjust for the presence of measurement error. To use these methods, one has to assume or derive a measurement error model. This project will enable us to develop one or more models that can inform future analyses. While most measurement error models used for this purpose are simple models of convenience (such as assuming normal distributions), use of IDATA will allow us to more precisely estimate the distribution of measurement error, free from most of the assumptions that are typically made.
Our plan is to first evaluate the differential between self-reported and objectively measured physical activity levels. Then we will aggregate this information across subjects, and possibly across time points, to estimate a distribution for measurement error. We will investigate if the measurement error is a function of patient characteristics, objective measures or other factors. This may lead to development of multiple measurement error models.
Aims
1. Estimate measurement error for self-reported physical activity by using objective monitors as the gold standard
2. Investigate if measurement error is a function of personal characteristics and other factors.
3. Develop measurement error model or models for various self-reported measures.
Collaborators
Mithat Gonen, Memorial Sloan Kettering Cancer Center
Lee Jones, Memorial Sloan Kettering Cancer Center
Chaya Moskowitz, Memorial Sloan Kettering Cancer Center