qhet_mvmr — qhet_mvmr • MVMR

Fits a multivariable Mendelian randomization model adjusting for weak instruments. The functions requires a formatted dataframe using the format_mvmr() function, as well a phenotypic correlation matrix pcor. This should be obtained from individual level phenotypic data, or constructed as a correlation matrix where correlations have previously been reported. By default, confidence intervals are not produced but can be calculated by setting CI = TRUE. They are calculated using either a non-parametric bootstrap over the genetic variants (CI_method = "bootstrap", the default), with the number of bootstrap iterations specified using the iterations argument, or a leave-one-variant-out jackknife (CI_method = "jackknife"). Sanderson, Spiller and Bowden (2021) recommend the jackknife, as with a moderate number of variants bootstrap resamples can give very weak instruments. Note that calculating confidence intervals can take a substantial amount of time.

Usage

qhet_mvmr(
  r_input,
  pcor,
  CI,
  iterations,
  ncores = parallelly::availableCores(omit = 1),
  CI_method = c("bootstrap", "jackknife")
)

Arguments

r_input: A formatted data frame using the format_mvmr() function or an object of class MRMVInput from MendelianRandomization::mr_mvinput()
pcor: A phenotypic correlation matrix including the correlation between each exposure included in the MVMR analysis. It is used to construct the covariance between the estimated effects of each genetic variant on each exposure, as pcor multiplied by the outer product of the variant's standard errors. This is appropriate when the exposure GWAS were estimated in the same sample. If the exposure GWAS were estimated in non-overlapping samples these covariances are zero, and the identity matrix diag(K), where K is the number of exposures, should be supplied instead.
CI: Indicates whether 95 percent confidence intervals should be calculated.
iterations: Specifies number of bootstrap iterations for calculating 95 percent confidence intervals. Not used when CI_method = "jackknife".
ncores: Number of cores to use for parallel processing in the bootstrap or jackknife. Default is parallelly::availableCores(omit = 1). On Windows, this is automatically set to 1 regardless of user input. It is recommended to only set this to a maximum of parallelly::availableCores(omit = 1).
CI_method: Method used to calculate confidence intervals when CI = TRUE: "bootstrap" (the default) for bias-corrected and accelerated (BCa) non-parametric bootstrap intervals, or "jackknife" for normal-based intervals using the leave-one-variant-out jackknife standard error.

Value

A dataframe containing effect estimates with respect to each exposure, and when CI = TRUE their 95 percent confidence intervals. With CI_method = "jackknife" the jackknife standard errors are also returned.

Details

Effects are estimated by minimising a Q-statistic that includes an additive heterogeneity parameter tau-squared. Following Sanderson, Spiller and Bowden (2021), tau-squared is chosen so that the minimised Q-statistic equals L - K, where L is the number of genetic variants and K the number of exposures. Tau-squared is constrained to be non-negative, and is fixed at zero when the minimised Q-statistic without heterogeneity is already no greater than L - K.

References

Sanderson, E., et al., An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. International Journal of Epidemiology, 2019, 48, 3, 713–727. doi:10.1093/ije/dyy262

Sanderson, E., Spiller, W., and Bowden, J., Testing and correcting for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization. Statistics in Medicine, 2021, 40, 25, 5434–5452. doi:10.1002/sim.9133

Author

Wes Spiller; Eleanor Sanderson; Jack Bowden.

Examples

if (FALSE) { # \dontrun{
qhet_mvmr(r_input, pcor, CI = TRUE, iterations = 1000)
qhet_mvmr(r_input, pcor, CI = TRUE, CI_method = "jackknife")
} # }