Package {RegCalib}

Title: Regression Calibration for Measurement Error Correction
Version: 0.1.0
Description: Corrects for measurement error in continuous exposures and covariates using regression calibration methods. Provides corrected coefficients, standard errors, p-values, confidence intervals, and variance-covariance matrices for linear and generalized linear outcome models under an external validation study design. Implements the deattenuation factor method (RegCalibDF) and the substitution approach (RegCalibSub). Supports single and multiple error-prone exposures.
License: GPL-3
Encoding: UTF-8
Imports: stats, dplyr, Matrix, matrixcalc
Suggests: knitr, rmarkdown, testthat (≥ 3.0.0)
Config/testthat/edition: 3
URL: https://github.com/JingyuCui639/RegCalib
Config/roxygen2/version: 8.0.0
Depends: R (≥ 3.5)
LazyData: true
VignetteBuilder: knitr
BugReports: https://github.com/JingyuCui639/RegCalib/issues
LazyDataCompression: xz
NeedsCompilation: no
Packaged: 2026-07-30 07:10:26 UTC; jc4428
Author: Jingyu Cui ORCID iD [aut, cre] (Affiliation: Department of Biostatistics, Yale School of Public Health), Wenze Tang [aut] (Affiliation: Vertex Pharmaceuticals), Molin Wang [aut] (Affiliation: Department of Biostatistics, Department of Epidemiology, Harvard T.H. Chan School of Public Health)
Maintainer: Jingyu Cui <jingyu.cui@yale.edu>
Repository: CRAN
Date/Publication: 2026-08-07 16:30:20 UTC

RegCalib: Regression Calibration for Measurement Error Correction

Description

Corrects for measurement error in continuous exposures and covariates using regression calibration methods. Provides corrected coefficients, standard errors, p-values, confidence intervals, and variance-covariance matrices for linear and generalized linear outcome models under an external validation study design. Implements the deattenuation factor method (RegCalibDF) and the substitution approach (RegCalibSub). Supports single and multiple error-prone exposures.

Implements regression calibration methods for correcting measurement error in continuous exposures and covariates using external validation data.

Author(s)

Maintainer: Jingyu Cui jingyu.cui@yale.edu (ORCID) (Affiliation: Department of Biostatistics, Yale School of Public Health)

Authors:

Jingyu Cui jingyu.cui@yale.edu

Wenze Tang

Molin Wang

See Also

Useful links:

Regression Calibration Using the Deattenuation Factor Method

Description

Corrects for measurement error in continuous exposures (and covariates) and returns corrected coefficients, standard errors, p-values, and variance-covariance matrices. Users may supply their own uncorrected estimates (e.g., from logistic regression model) instead of using the built-in outcome model. Only externalvalidation study design is supported. Based on Rosner, Spiegelman & Willett (1989, 1990) and Spiegelman, McDermott & Rosner (1997).

Usage

RegCalibDF(
  supplyEstimates = FALSE,
  ms,
  vs,
  sur,
  exp,
  covCalib = NULL,
  covOutcomePlus = NULL,
  outcome = NA,
  method = "lm",
  family = NA,
  link = NA,
  external = TRUE,
  pointEstimates = NA,
  vcovEstimates = NA
)

Arguments

supplyEstimates

Logical. If TRUE, uncorrected estimates are supplied by the user via pointEstimates and vcovEstimates, and ms is optional. Standard regression results are not returned in this case. Default FALSE.

ms

Main study data frame. Must contain all variables specified in sur, covCalib, and covOutcomePlus. Required when supplyEstimates = FALSE.

vs

Internal or external validation study data frame. Must contain all variables specified in exp, sur, and covCalib.

sur

Character vector of mismeasured exposure(s)/covariate(s) (surrogates) in the main study dataset.

exp

Character vector of correctly-measured exposure(s)/covariate(s) in the validation dataset. Must correspond one-to-one with sur and have the same length.

covCalib

Character vector of correctly-measured covariates to adjust for in both the calibration model and the outcome model. Default NULL.

covOutcomePlus

Character vector of correctly-measured risk factors for the outcome that are not associated with the exposure or surrogate. These are included in the outcome model only and must not overlap with covCalib. Default NULL.

outcome

Character. Name of the outcome variable. Required when method is "lm" or "glm".

method

Character. Outcome modelling method: "lm", "glm". Default "lm".

family

Family function for glm (e.g., binomial). Not a character string. Required when method = "glm".

link

Character. Link function for glm (e.g., "logit"). Required when method = "glm".

external

Logical. TRUE (default) for an external validation study; FALSE for an internal validation study. When external = FALSE, vs must contain the outcome variable and supplyEstimates must be FALSE.

pointEstimates

Named numeric vector of uncorrected point estimates from standard regression (intercept excluded). Names must match the (expanded) covariate names from covCalib followed by covOutcomePlus. Required when supplyEstimates = TRUE.

vcovEstimates

Named square matrix of uncorrected variance-covariance estimates (intercept excluded). Column names must match those of pointEstimates. Required when supplyEstimates = TRUE.

Value

A named list containing:

correctedCoefTable

Data frame of corrected estimates, standard errors, Z-values, p-values, and 95% confidence intervals.

correctedVCOV

Variance-covariance matrix of the corrected estimates.

standardCoefTable

(When supplyEstimates = FALSE) Results from the uncorrected standard regression.

standardVCOV

(When supplyEstimates = FALSE) Variance- covariance matrix from the uncorrected standard regression.

calibrationModelCoefTable

Calibration model slope estimates.

calibrationModelVCOV

Residual variance-covariance matrix from the calibration model.

References

Rosner B, Willett WC, Spiegelman D (1989). Correction of logistic relative risk estimates and confidence intervals for systematic within-person measurement error. Statistics in Medicine 8:1051–1069.

Rosner B, Spiegelman D, Willett WC (1990). Correction of logistic regression relative risk estimates and confidence intervals for measurement error: the case of multiple covariates measured with error. American Journal of Epidemiology 132:734–745.

Spiegelman D, McDermott A, Rosner B (1997). The many uses of the 'regression calibration' method for measurement error bias correction in nutritional epidemiology. American Journal of Clinical Nutrition 65:1179S–1186S.

Spiegelman D, Carroll RJ, Kipnis V (2001). Efficient regression calibration for logistic regression in main study/internal validation study designs with an imperfect reference instrument. Statistics in Medicine 20:139–160.

Examples

data("main_data_sim", package = "RegCalib")
data("valid_data_sim", package = "RegCalib")

set.seed(123)
selected_rows <- sample(
  seq_len(nrow(main_data_sim)),
  1000
)

main_example <- main_data_sim[
  selected_rows,
  ,
  drop = FALSE
]

result <- RegCalibDF(
  ms = main_example,
  vs = valid_data_sim,
  sur = c("fqtfatinc", "fqcalinc", "fqalcinc"),
  exp = c("drtfatinc", "drcalinc", "dralcinc"),
  covCalib = "agec",
  outcome = "case",
  method = "glm",
  family = binomial,
  link = "logit",
  external = TRUE
)

result$correctedCoefTable

Regression Calibration Using Substitution Method

Description

Corrects for measurement error in continuous exposures (and covariates) and returns corrected coefficients, standard errors, p-values, and variance-covariance matrices using the Carroll-Ruppert-Stefanski-Crainiceanu (CRS) sandwich variance estimator. Supports linear and generalized linear outcome models under external validation study design. Standard errors are derived analytically via the sandwich estimator (not bootstrap). Non-linear terms such as interaction terms should be pre-computed as permanent columns in the input datasets rather than specified in the formula.

Usage

RegCalibSub(
  ms,
  vs,
  sur,
  exp,
  vsIndicator,
  covCalib = NULL,
  covOutcome = NULL,
  outcome = NA,
  method = "lm",
  family = NA,
  link = NA,
  external = TRUE
)

Arguments

ms

Main study data frame. Must contain all variables specified in sur, outcome, and covOutcome (if any).

vs

External validation study data frame. Must contain all variables specified in exp, sur, and covCalib (if any). Required when external = TRUE.

sur

Character vector of mismeasured exposure(s)/covariate(s) (surrogates) in the main study dataset.

exp

Character vector of correctly-measured exposure(s)/covariate(s) in the validation dataset. Must correspond one-to-one with sur and have the same length.

vsIndicator

Character. Name of the indicator variable in the main study dataset that identifies subjects with a validation record (1 = has validation record). Required when external = FALSE.

covCalib

Character vector of correctly-measured covariates to adjust for in the calibration model, including any non-linear terms. Default NULL.

covOutcome

Character vector of correctly-measured covariates to adjust for in the outcome model (with corrected exposure), including any non-linear terms. Default NULL.

outcome

Character. Name of the outcome variable. Required.

method

Character. Outcome modelling method: "lm" or "glm". Default "lm".

family

Family function for glm (e.g., binomial). Not a character string. Required when method = "glm".

link

Character. Link function for glm (e.g., "logit" or "log"). Required when method = "glm".

external

Logical. TRUE (default) for an external validation study; FALSE for an internal validation study embedded in ms.

Value

A named list containing:

correctedCoefTable

Matrix of corrected estimates, standard errors, Z-values, p-values, and 95% confidence intervals.

correctedVCOV

Variance-covariance matrix of the corrected estimates.

References

Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu CM (2006). Measurement Error in Nonlinear Models, 2nd ed. Chapman & Hall/CRC.

Examples

data("main_data_sim", package = "RegCalib")
data("valid_data_sim", package = "RegCalib")

set.seed(123)
selected_rows <- sample(
  seq_len(nrow(main_data_sim)),
  1000
)

main_example <- main_data_sim[
  selected_rows,
  ,
  drop = FALSE
]

result <- RegCalibSub(
  ms = main_example,
  vs = valid_data_sim,
  sur = c("fqtfatinc", "fqcalinc", "fqalcinc"),
  exp = c("drtfatinc", "drcalinc", "dralcinc"),
  covCalib = "agec",
  covOutcome = "agec",
  outcome = "case",
  method = "glm",
  family = binomial,
  link = "logit",
  external = TRUE
)

result$correctedCoefTable
result$correctedVCOV

Simulated Main-Study Data

Description

A simulated main-study dataset for demonstrating regression calibration with error-prone dietary exposures.

Usage

main_data_sim

Format

A data frame with 89,538 observations and the following variables:

id

Participant identifier.

fqcal

Food-frequency questionnaire estimate of caloric intake.

fqcalinc

Transformed caloric intake.

fqtfat

Food-frequency questionnaire estimate of total fat intake.

fqtfatinc

Transformed total fat intake.

fqalc

Food-frequency questionnaire estimate of alcohol intake.

fqalcinc

Transformed alcohol intake.

age

Participant age.

agec

Categorized participant age.

case

Binary outcome indicator.

Source

Simulated data created for the RegCalib package.

Simulated External Validation Data

Description

A simulated external validation dataset containing surrogate and reference measurements of dietary exposures.

Usage

valid_data_sim

Format

A data frame with 173 observations and variables containing food-frequency questionnaire measurements, reference measurements, and age.

id

Participant identifier.

fqcal

Surrogate caloric intake measurement.

fqcalinc

Transformed surrogate caloric intake.

fqtfat

Surrogate total-fat intake measurement.

fqtfatinc

Transformed surrogate total-fat intake.

fqalc

Surrogate alcohol intake measurement.

fqalcinc

Transformed surrogate alcohol intake.

drcal

Reference caloric intake measurement.

drcalinc

Transformed reference caloric intake.

drtfat

Reference total-fat intake measurement.

drtfatinc

Transformed reference total-fat intake.

dralc

Reference alcohol intake measurement.

dralcinc

Transformed reference alcohol intake.

age

Participant age.

agec

Categorized participant age.

Source

Simulated data created for the RegCalib package.