Optimal Multi-Wave Validation for Secondary Use Data with Outcome and Exposure Misclassification

Lotspeich, Amorim, Shaw, Tao, and Shepherd

The complete R package auditDesignR and code for the simulation settings included in the paper. Art for the hex sticker was made with the help of Dall-E 3.

Install

To install the package, run the following in your R console: devtools::install_github("sarahlotspeich/auditDesignR", ref = "main").

Validation Study Designs

• Simple random sampling (SRS): All subjects in Phase I have equal probability of inclusion in Phase II.
• Unvalidated case-control sampling (CC):* Subjects are stratified on Y* and separate random samples of size n/2 are drawn from each stratum.
• Unvalidated balanced case-control sampling (BCC):* Subjects are jointly stratified on (Y*, X*) and separate random samples of size n/4 subjects drawn from each stratum.
• Optimal design (optMLE): Subjects are jointly stratified on (Y*,X*), and stratum sizes are chosen following Section 2.2. The optMLE is included as a "gold standard" design since it requires knowing the parameters θ.
• Two-wave approximate optimal design (optMLE-2): Subjects are jointly stratified on (Y*,X*). In the first wave, n/2 subjects are selected using BCC*, and in the second wave the remaining subjects are chosen following the design in Section 2.4.

Simulation settings

Inside the simulations subdirectory, you will find the following:

• YXboth_SimSetup.R: simulations with outcome and exposure misclassification, defining sampling strata on error-prone outcome and exposure, Y* and X* (Table 1 in the Manuscript)
• YXboth_inclZ_SimSetup.R: simulations with outcome and exposure misclassification, defining sampling strata on error-prone outcome and exposure, Y* and X*, and error-free binary covariate Z (Table S2 in the Supplemental Materials for the Manuscript).
• YXboth_inclZ_misspec_SimSetup.R: simulations considering model misspecification at the design stage (Table 2 in the Manuscript).
• Yonly_SimSetup.R: simulations with outcome misclassification alone, defining sampling strata on error-prone outcome Y* and error-free exposure X (top of Table S3 in the Supplemental Materials for the Manuscript).
• Xonly_SimSetup.R: simulations with exposure misclassification alone, defining sampling strata on error-free outcome Y and error-prone exposure X* (bottom of Table S3 in the Supplemental Materials for the Manuscript).

See the preprint for more details.

Example data

Example datasets for numerous error settings can be found in data.