Skip to contents

Estimate the association between an exposure and a continuous outcome

linear_model.Rd

This function is used to estimate the (adjusted) linear association between an exposure variable and a continuous outcome variable using a linear regression model. It is a wrapper for the lm function for convenience and provides results from some additional analysis steps if requested.

Usage

linear_model(
  data,
  exposure,
  outcome,
  adjust_for = NULL,
  exclude = NULL,
  exposure_term = exposure,
  additional = "",
  return = c("model", "reduced-model", "coefs", "exposure"),
  return_fn = identity,
  nuisance = character(0),
  ...
)

Arguments

data

A data frame.

exposure

Name of column containing exposure variable.

outcome

Name of column containing outcome variable.

adjust_for

Character vector of terms to include in the model formula.

exclude

Names of columns that identify observations that should be excluded.

exposure_term

The term(s) to include in the model for the exposure. This can be a character vector or a function that take the exposure name as the argument and returns a character vector. (Default: exposure)

additional

Extra results to include. See "Additional results" for details.

return

What should the function return:

  • "model" (default): The model object (object of class "lm").

  • "reduced-model": The model object with some large components removed.

  • "coefs": A data frame of all estimated coefficients and other key model statistics.

  • "exposure": A data frame of estimated coefficient(s) for the exposure and other key model statistics.

return_fn

If return="model", name of function to apply to model before it's returned. (Default: identity)

nuisance

If "emmeans" is requested via the additional argument, use this to specify predictors to omit from the reference grid. See the help for emmeans and ref_grid.

...

Arguments passed to stats::lm()

Value

Either a model object, reduced model object or data frame of coefficients.

Additional results

The additional argument is used to add results from additional analyses. It must be a character vector containing one or more of:

  • "anova": ANOVA F-test comparing the full model to one without the exposure. Output will have additional columns "F" (F test statistics), "Df" (degrees of freedom), "p_F" (p-value).

  • "float": Standard errors for the exposure calculated by float. Output will have additional columns "float_estimate", "float_se" and "limits".

  • "qvcalc": Standard errors for the exposure calculated by qvcalc. Output will have additional columns "qvcalc_estimate", "quasi_se" and "worstErrors".

  • "emmeans": Estimated marginal means for the exposure calculated using emmeans with proportional weights for factor variables that are averaged over. Output will have additional columns "emmean", "emmean_se" and "emmean_quasi_se".

  • "" (default): Nothing.

These will only be included when relevant e.g. standard errors from float when the exposure has more than two levels. This only has an effect if return is "coefs" or "exposure".

See float and qvcalc for details about these results. Note that "limits" and "worstErrors" do not correspond to particular groups but accuracy other all contrasts.

Examples

linear_model(ckbtools_participant_data,
             exposure = "age_at_study_date",
             outcome = "sbp_mean",
             adjust_for = "sex",
             return = "coefs")
#> # A tibble: 3 × 7
#>   term              estimate std.error statistic  p.value     n formula         
#>   <chr>                <dbl>     <dbl>     <dbl>    <dbl> <int> <chr>           
#> 1 age_at_study_date    0.125   0.00945      13.2 1.42e-39 25000 sbp_mean ~ age_…
#> 2 (Intercept)        102.      0.516       198.  0           NA NA              
#> 3 sexFemale           -2.74    0.205       -13.4 7.53e-41    NA NA              

# factor exposure, calculate estimated marginal means
linear_model(ckbtools_participant_data,
             exposure = "smoking",
             outcome = "sbp_mean",
             adjust_for = c("age_at_study_date", "sex"),
             additional = "emmeans",
             return = "coefs")
#> # A tibble: 7 × 11
#>   term   estimate std.error statistic   p.value emmean emmean_se emmean_quasi_se
#>   <chr>     <dbl>     <dbl>     <dbl>     <dbl>  <dbl>     <dbl>           <dbl>
#> 1 smoki…   NA      NA          NA     NA          108.     0.157           0.197
#> 2 smoki…   -0.673   0.352      -1.91   5.61e- 2   107.     0.305           0.296
#> 3 smoki…   -0.297   0.352      -0.845  3.98e- 1   108.     0.304           0.295
#> 4 smoki…   -1.98    0.299      -6.64   3.15e-11   106.     0.224           0.220
#> 5 (Inte…  103.      0.555     186.     0           NA     NA              NA    
#> 6 age_a…    0.125   0.00944    13.2    9.49e-40    NA     NA              NA    
#> 7 sexFe…   -3.66    0.256     -14.3    2.55e-46    NA     NA              NA    
#> # ℹ 3 more variables: n <int>, n_group <dbl>, formula <chr>

# return the model object
linear_model(ckbtools_participant_data,
             exposure = "bmi_grp",
             outcome = "sbp_mean",
             adjust_for = c("age_at_study_date", "sex"),
             return = "model")
#> 
#> Call:
#> stats::lm(formula = model_formula, data = data)
#> 
#> Coefficients:
#>       (Intercept)     bmi_grp18.5-25       bmi_grp25-30         bmi_grp30+  
#>           95.9089             4.7895             9.7153            15.7819  
#> age_at_study_date          sexFemale  
#>            0.1386            -2.9989  
#>