This is an R package to aid in determining if observational or two-stage least square (instrument variable or in genetics Mendelian randomization) analysis have a non-linear relationship between expoosure and outcome.

devtools::install_github("hughesevoanth/glsmr")

• There are two functions that are most useful

  • glsmr()
  • plot_glsmr()

• an example for using glsmr

    myexample = glsmr( wdata = mydata,
          outcome = "trait",
          exposure = "bmi",
          instrument = "bmi_grs",
          linear_covariates = c("batch", "sex"),
          smooth_covariates =  c("age"),
          # strata = 4, ## for quartiles
          strata = c(10,18.5,25,30,45),
          rnt_outcome = TRUE,
          weights_variable = NA,
          outlier_method = "iqr",
          outlier_cutoff = 5,
          messages = FALSE,
          return_models = TRUE)
  

  • NOTE: 'smooth_covariates' will be modeled as smooths or non-linear variables in the GAM, but as typical parametric variables in the linear and tsls analyses.

• an example for using plot_glsmr()

    plot_glsmr(myexample,
    	add_strata_2_curves = FALSE,
    	add_strata_2_points = TRUE,
    	brewer_col = "Set1",
    	old_plot_scheme = TRUE,
    	old_GAM_smooths = TRUE,
    	plot_obs_res_betas = FALSE,
      	pval_thresh = 0.05)
  

wiki page here