Skip to contents

Single-site cross-check against standard summary-data MR

validateAgainstTwoSampleMR.Rd

Runs Medusa's single-site profile-likelihood estimate and compares it with TwoSampleMR estimates computed from the same site's per-SNP summary statistics. This is a calibration check: agreement is expected when the profile is close to quadratic and the allele-score approximation aligns with standard summarized-data MR assumptions. This helper requires the optional TwoSampleMR package.

Usage

validateAgainstTwoSampleMR(
  siteProfile,
  instrumentTable,
  methods = c("IVW", "MREgger", "WeightedMedian"),
  outcomeSampleSize = NULL,
  exposureSampleSize = NULL,
  outcomeType = "binary",
  ciLevel = 0.95
)

Arguments

siteProfile

A single medusaSiteProfile object produced by fitOutcomeModel with analysisType = "perSNP".

instrumentTable

Data frame. Output of getMRInstruments. Used both for the Medusa Wald ratio and, if needed, to backfill allele columns missing from older perSnpEstimates objects.

methods

Character vector of summary-data methods to compare. Default is c("IVW", "MREgger", "WeightedMedian").

outcomeSampleSize

Optional integer. Passed to runSensitivityAnalyses.

exposureSampleSize

Optional integer. Passed to runSensitivityAnalyses.

outcomeType

Character. Outcome summary-statistic scale represented by siteProfile$perSnpEstimates. Default is "binary".

ciLevel

Numeric. Confidence interval level for the Medusa estimate. Default is 0.95.

Value

A data frame with one row for the Medusa profile-likelihood estimate and one row for each requested TwoSampleMR method. The delta_vs_medusa column reports each summary-data estimate minus the Medusa estimate.

Details

Validate a Single-Site Medusa Estimate Against TwoSampleMR

See also

fitOutcomeModel, computeMREstimate, runSensitivityAnalyses

Examples

if (requireNamespace("TwoSampleMR", quietly = TRUE)) {
  simData <- simulateMRData(n = 3000, nSnps = 5, trueEffect = 0.3)
  siteProfile <- fitOutcomeModel(
    cohortData = simData$data,
    instrumentTable = simData$instrumentTable,
    analysisType = "perSNP",
    betaGrid = seq(-2, 2, by = 0.05)
  )
  comparison <- validateAgainstTwoSampleMR(siteProfile, simData$instrumentTable)
  comparison
}
#> Fitting outcome model at site 'site_1' (1725 cases, 1275 controls)...
#> Site 'site_1': beta_ZY_hat = 0.0426 (SE = 0.1182).
#> Pooling profile likelihoods from 1 site(s)...
#> Pooling complete: 1 sites, 1725 total cases, 1275 total controls.
#> MR estimate: beta = 0.2369 (95% CI: -0.7107, 1.1845), p = 6.72e-01
#> Odds ratio: 1.267 (95% CI: 0.491, 3.269)
#> Running sensitivity analyses with 5 SNPs...
#>   Engine: TwoSampleMR
#> Harmonising exposure (medusa_exposure) and outcome (medusa_outcome)
#>   IVW...
#> Analysing 'medusa_exposure' on 'medusa_outcome'
#>   MR-Egger...
#> Analysing 'medusa_exposure' on 'medusa_outcome'
#>   Weighted Median...
#> Analysing 'medusa_exposure' on 'medusa_outcome'
#> Sensitivity analyses complete.
#>        source             method   beta_MR     se_MR     ci_lower  ci_upper
#> 1      Medusa Profile likelihood 0.2369024 0.5603280 -0.710707109 1.1845118
#> 2 TwoSampleMR                IVW 0.2194567 0.1167514 -0.009376005 0.4482894
#> 3 TwoSampleMR           MR-Egger 0.6304650 0.2779381  0.085706258 1.1752237
#> 4 TwoSampleMR    Weighted Median 0.2426033 0.1419169 -0.035553826 0.5207604
#>         pval delta_vs_medusa
#> 1 0.67244679     0.000000000
#> 2 0.06014999    -0.017445678
#> 3 0.10808801     0.393562607
#> 4 0.08736312     0.005700936