PVBcorrect (version 0.3.1)

acc_em: PVB correction by EM-based logistic regression method

Description

Perform PVB correction by EM-based logistic regression method.

Usage

acc_em(
  data,
  test,
  disease,
  covariate = NULL,
  mnar = TRUE,
  ci = FALSE,
  ci_level = 0.95,
  ci_type = "basic",
  R = 999,
  seednum = NULL,
  show_t = TRUE,
  t_max = 500,
  cutoff = 1e-04,
  t_print_freq = 100,
  return_t = FALSE,
  r_print_freq = 100,
  description = TRUE
)

Value

A list object containing:

boot_data

An object of class "boot" from boot. Contains Sensitivity, Specificity, PPV, NPV and t (i.e. EM iteration taken for convergence). Use acc_em_object$boot_data$t[,5] to check the t.

boot_ci_data

A list of objects of type "bootci" from from boot.ci. Contains Sensitivity, Specificity, PPV, and NPV.

acc_results

The accuracy results.

Arguments

data

A data frame, with at least "Test" and "Disease" variables.

test

The "Test" variable name, i.e. the test result. The variable must be in binary; positive = 1, negative = 0 format.

disease

The "Disease" variable name, i.e. the true disease status. The variable must be in binary; positive = 1, negative = 0 format.

covariate

The name(s) of covariate(s), i.e. other variables associated with either test or disease status. Specify as name vector, e.g. c("X1", "X2") for two or more variables. The variables must be in formats acceptable to GLM.

mnar

The default is assuming missing not at random (MNAR) missing data mechanism, MNAR = TRUE. Set this to FALSE to obtain results assuming missing at random (MAR) missing data mechanism. This will be equivalent to using acc_ebg.

ci

View confidence interval (CI). The default is FALSE.

ci_level

Set the CI width. The default is 0.95 i.e. 95% CI.

ci_type

Set confidence interval (CI) type. Acceptable types are "norm", "basic", "perc", and "bca", for bootstrapped CI. See boot.ci for details.

R

The number of bootstrap samples. Default R = 999.

seednum

Set the seed number for the bootstrapped CI. The default is not set, so it depends on the user to set it outside or inside the function.

show_t

Print the current EM iteration number t. The default is TRUE.

t_max

The maximum iteration number for EM. Default t_max = 500. It is recommended to increase the number when covariates are included.

cutoff

The cutoff value for the minimum change between iteration. This defines the convergence of the EM procedure. Default cutoff = 0.0001. This can be set to a larger value to test the procedure.

t_print_freq

Print the current EM iteration number t at each specified interval. Default t_print_freq = 100.

return_t

Return the final EM iteration number t. This can be used for the purpose of checking the EM convergence. The default is FALSE, but is set to TRUE when ci = TRUE.

r_print_freq

Print the current bootstrap sample number at each specified interval. Default r_print_freq = 100.

description

Print the name of this analysis. The default is TRUE. This can be turned off for repeated analysis, for example in bootstrapped results.

References

1. Kosinski, A. S., & Barnhart, H. X. (2003). Accounting for nonignorable verification bias in assessment of diagnostic tests. Biometrics, 59(1), 163–171.

Examples

# For sample run, test with low R boot number, low t_max, low cutoff
# The results will not be good

# without covariate
em_out = acc_em(data = cad_pvb, test = "T", disease = "D", ci = TRUE, seednum = 12345,
                R = 2, t_max = 100, cutoff = 0.01)
em_out$acc_results
em_out$boot_data$t  # bootstrapped data, 1:5 columns are Sn, Sp, PPV, NPV,
                    # t (i.e. EM iteration taken for convergence)
em_out$boot_ci_data