mrcieu / measurement_error_adjustment

Code accompanying paper on exploring regression dilution bias using repeat measures in UK Biobank

Measurement Error Adjustment using UK Biobank Phenotype data

This code relates to the article: Rutter CE, Millard LAC, Borges MC and Lawlor DA. Exploring regression dilution bias using repeat measurements of 2858 variables in up to 49 000 UK Biobank participants (submitted to IJE)

There are two components:

Part 1 An illustrative example to adjust for random measurement error in the main exposure and confounders of the association between C-reactive protein and mortality. Can be adapted for other variables and associations.

Part 2 Calculating Intraclass correlation coefficients (ICC) and associated correction factor data for all repeatedly measured, continuous variables in UK Biobank.

Both parts require Stata version 16, run in batch.

Set directories at the top of the Stata do files with your own path and folders as follows:

global DATA "/your/data/path/and/folder"

global RESULTS "/your/results/path/and/folder"

Must have the following data files:

• Tab delimited text file of UK Biobank phenotype data

• Text file list of withdrawn participants ids

• Data dictionary in csv format from the UK Biobank Showcase (www.ukbiobank.ac.uk>Data Showcase>Essential Information>Requesting data and using the UK Biobank showcase>Data Dictionary)

• An extra file containing UK Biobank date fields, which were not in the main file, but this is not necessary if all the required data is in the main file.

Part 1 – Illustrative example of accounting for measurement error in exposure and confounders.

1. getmodeldata.sh

This selects the relevant data for the example from the main files using shell script only. It contains the variable list for relevant exposures and confounders that can be adapted for other models.

• Requires – ukbdata_new.txt

• Main output – modelfile.txt

2. modelexcl.sh

This calls Stata to run modelexcl.do, which excludes withdrawn participants, renames and formats variables. Creates mortality outcomes and confounders.

• Requires - modelfile.txt and withdrawal file w16729_20220222.csv

• Main output - modelfile_all.dta and modelfile_comp.dta

3. bootmodels1.sh to bootmodels10.sh

These create bootstrap replicates of the hazard models using regression calibration.

• Requires – modelfile_comp.dta

• Main output – boots_1.dta to boots_10.dta

4. runmodels.sh

This calls Stata to run runmodels.do, which fits an unadjusted proportional hazard model, along with models adjusted for random measurement error using ICC correction factors and regression calibration.

• Requires – modelfile_all.dta, modelfile_comp.dta and boots_1.dta to boots_10.dta

• Main output – results in runmodels.log

Part 2 - Calculating ICC and correction factors for all UK Biobank continuous variables.

Process for baseline and online diet variables fields

1. varlist_01.sh

This calls Stata to run varlist_01.do, which creates a variable list by running initial exclusions on data dictionary.

• Requires – Data_Dictionary_Showcase.csv

• Main output – fields01.txt

2. getdata_01.sh

This extracts required variables from UK Biobank file using shell script only.

• Requires – fields01.txt, ukbdata_new.txt

• Main output – mainfile01.txt

3. rundata_01.sh

This calls Stata to run rundata_01.do which calls the following do files in turn:

withdrawals_01.do – excludes the ids of participants who have withdrawn consent

• Requires – withdrawal csv file e.g. w16729_20220222.csv

• Main output – mainfile01.dta

dates_01.do - adds on the dates that were missing from the main file, renames and reformats date fields.

• Requires – mainfile01.dta, extra dates file (new_dates_data.csv)

• Main output – mainfile01.dta

arrays_01.do – deals with specific fields with multiple values in one visit.

• Requires – mainfile01.dta

• Main output – mainfile01.dta

recodes_01.do - amends some of the special coded non-numeric values

• Requires – mainfile01.dta

• Main output – mainfile01.dta

extraexcl_01.do – checks and removes

a. variables with <20 distinct values per instance (for observations non missing in both instances)

b. variables with fewer than 100 observations non missing in both instances

c. variables with >=20% of non-missing observations having one single value (should be categorical)

• Requires - mainfile01.dta

• Main output - finalfile01.dta

concord_01.do – run concordance analysis and get ICC for all records

• Requires – finalfile01.dta, used_vars_list01.dta

• Main output – concordance_output01.dta, concordance_output.xlsx

4. graphs.sh

This calls Stata to run graphs.do which creates the graphs included in the paper.

• Requires – concordance_output.xlsx

Process for imaging visit fields:

Same as above but replace end of script file names with 23 instead of 01.

The imaging programs are similar but split the file into three due to size.