Skip to contents

Computes three group missing data summary diagnostics

smdi_diagnose.Rd

This function bundles and calls all three group diagnostics and returns the most important summary metrics. For more information and details, please refer to the individual functions.

Important: don't include variables like ID variables, ZIP codes, dates, etc.

Usage

smdi_diagnose(
  data = NULL,
  covar = NULL,
  median = TRUE,
  includeNA = FALSE,
  train_test_ratio = c(0.7, 0.3),
  set_seed = 42,
  ntree = 1000,
  n_cores = 1,
  model = c("logistic", "linear", "cox"),
  form_lhs = NULL,
  exponentiated = FALSE
)

Arguments

data

dataframe or tibble object with partially observed/missing variables

covar

character covariate or covariate vector with partially observed variable/column name(s) to investigate. If NULL, the function automatically includes all columns with at least one missing observation and all remaining covariates will be used as predictors

median

logical if the median (= TRUE; recommended default) or mean of all absolute standardized mean differences (asmd) should be computed (smdi_asmd())

includeNA

logical, should missingness of other partially observed covariates be explicitly modeled for computation of absolute standardized mean differences (default is FALSE)

train_test_ratio

numeric vector to indicate the test/train split ratio for random forest missingness prediction model, e.g. c(.7, .3) is the default

set_seed

seed for reproducibility of random forest missingness prediction model, defaults to 42

ntree

integer, number of trees for random forest missingness prediction model (defaults to 1000 trees)

n_cores

integer, if >1, computations will be parallelized across amount of cores specified in n_cores (only UNIX systems)

model

character describing which outcome model to fit to assess the association between covar missingness indicator and outcome. Currently supported are models of type logistic, linear and cox (see smdi_outcome)

form_lhs

string specifying the left-hand side of the outcome formula (see smdi_outcome)

exponentiated

logical, should results of outcome regression to assess association between missingness and outcome be exponentiated (default is FALSE)

Value

smdi object including a summary table of all three smdi group diagnostics:

Group 1 diagnostic:

  • asmd_mean or asmd_median: average/median absolute standardized mean difference (and min, max) of patient characteristics between those without (1) and with (0) observed covariate

  • hotteling_p: p-value of hotelling test. Rejecting the H0 means that Hotelling's test detects a significant difference in the distribution between patients without (1) and with (0) the observed covariate

Group 2 diagnostic:

  • rf_auc: The area under the receiver operating curve (AUC) as a measure of the ability to predict the missingness of the partially observed covariate

Group 3 diagnostic:

  • estimate_univariate: univariate association between missingness indicator of covar and outcome

  • estimate_adjusted: association between missingness indicator of covar and outcome conditional on other fully observed covariates and missing indicator variables of other partially observed covariates

Details

Wrapper for individual diagnostics function.

References

TBD

See also

smdi_asmd smdi_hotelling smdi_little smdi_rf smdi_outcome

Examples

library(smdi)

smdi_diagnose(
 data = smdi_data,
 covar = "egfr_cat",
 model = "cox",
 form_lhs = "Surv(eventtime, status)"
 )
#> smdi summary table:
#> # A tibble: 1 × 6
#>   covariate asmd_median_min_max  hotteling_p rf_auc estimate_univariate      
#>   <chr>     <chr>                <chr>       <chr>  <glue>                   
#> 1 egfr_cat  0.243 (0.010, 0.485) <.001       0.629  0.06 (95% CI -0.03, 0.15)
#> # ℹ 1 more variable: estimate_adjusted <glue>
#> 
#> p_little: <.001