National Cancer Institute (NCI) methods workshop

Didier Brassard

Introduction

This workshop is an introduction to random measurement errors in nutrition research. While it is obvious that a systematic error (difference) between the “true” value and its measurement can be a problem, the impact of random errors is often more subtle. However, in many cases, random errors can be as problematic as systematic errors if they are ignored.

Before diving into measurement error correction methods, I recommend two introduction blog posts on the topic: ‘Statistical concept you should know’: random and systematic measurement errors and Impact of random errors: two nutrition examples. Useful reference includes Using Short-Term Dietary Intake Data to Address Research Questions Related to Usual Dietary Intake among Populations and Subpopulations: Assumptions, Statistical Techniques, and Considerations by Kirkpatrick et al. (2022) as well as the STRATOS guidance document on measurement error and misclassification of variables in observational epidemiology: Part 1-Basic theory and simple methods of adjustment by Keogh et al. (2020)

Do I need measurement error correction?

Food-frequency questionnaires or screeners reflect long-term intakes and measurement error correction methods - at least for within-individual random errors - are not needed. On the contrary, measurement error correction methods may be required when using short-term dietary intake instruments, e.g., food records or 24-hour dietary recalls.

The flow chart below is a general guide to decide whether measurement error correction (e.g., with the NCI methods) may be required when using dietary intakes data from short-term instruments. Please note that in the context of this repository, measurement error correction refers to random errors.

%%{init: {'theme': 'neutral' } }%%
flowchart TD
  Q1("What is the statistic of interest?")
  S1{"Mean<br>only"}
  S2{"Distribution or<br>prevalence"}
  S3{"Regression<br>coefficient"}
  S1-->Q4("Is there a <b>scoring</b> <br>(eg, diet quality score)<br>or a <b>ratio</b><br>(eg, % of total energy)?")
  Q4--"Yes"-->F1
  Q4--"No"-->F2
  F1{{"Use the<br>population ratio<br>method"}}
  Q1-->S1
  Q2("Are <b>repeated data</b><br>available for n>50?")
  Q1-->S2
  Q1-->S3
  S3-->Q3("Is dietary intake<br>the <b>independent</b> variable?")
  Q3--"Yes"-->Q2
  Q3--"No"-->F2
  F2{{"Measurement error<br>correction not needed"}}
  S2-->Q2
  Q2--"Yes"-->F3
  F3{{"Apply measurement error<br>correction methods"}}
  Q2--"No"-->F4
  F4{{"Use raw data,<br>assess bias with simulation,<br> acknowledge limitation"}}

Loading

For more details about the population ration method, see Analyzing the Canadian Community Health Survey (CCHS) 2015 data with R: mean diet quality score

NCI univariate method, distribution

SAS codes (01, 11, 12, …) are provided as an introduction to basic aspects of random measurement errors and application of the NCI univariate method to generate a distribution of usual dietary intakes.

The 12_Univariate_2part_exercise.sas code is intended as an exercise to practice using the univariate method. The answers are provided in 12_Univariate_2part.sas.

A companion presentation is available here.

The original NCI macros and further examples are available on the Biometry Research Group’s Software for Measurement Error in Nutrition Research website

Schematic overview of the NCI methods

%%{init: {'theme': 'neutral' } }%%
flowchart TB
subgraph data[<b>Repeated dietary assessment</b>]
    R1("24-hour recall #1")
    R2("24-hour recall #2")
    R3("24-hour recall #J")
    RT("Dietary constituents")
    R1 & R2 & R3-->RT
end
subgraph C[<b>Covariates</b>]
    N(Sequence,<br>weekend,<br>season)
    Z(Subgroups)
end
subgraph A[<b>Assumptions</b>]
    S1(Statistical<br>assumptions)
    S2(Classical error<br>model assumptions)
end
    data --"Box-Cox<br>transformation"-->M[<b>Step 1<br>Measurement error model</b>]
    C-->M
    A-->M
    M-->P1["Model parameters"]
    P1-->P2["Predicted intakes"]
    M-->P3[Within- & between-<br>individual variance]
    P3-->P2
    P1 & P2-->MC["<b>Step 2<br>Monte Carlo simulations</b>"]
    MC--"<I>M</I> simulations<br>per individual"-->PI["<b>Step 3<br>'Usual intakes' distribution<br>calculation among<br>(<I>M*n</I>) pseudo-individuals"]
    

Loading