odds_ratio_function {EpiForsk}

Easier to perform logistic and log-linear regressions giving a standardized output table

Description

odds_ratio_function analyses specified data given user specifications, including outcome, exposures and possible weights. It can handle survey-data, but not complex sampling schemes (if specified as survey-data, the model will create a simple survey-object from the data, using weights as specified - if not specified, the weights are 1 for each observation) The standard regression is logistic regression (yielding Odds Ratios=OR) but it is possible to perform a log-linear regression (yielding Risk Ratios=RR) instead, if specified and requirements are met.

Usage

odds_ratio_function(
  normaldata,
  outcomevar,
  expvars,
  number_decimals = 2,
  alpha = 0.05,
  regtype = c("logistic", "log-linear"),
  matchgroup = NULL,
  matchtiemethod = c("exact", "approximate", "efron", "breslow"),
  values_to_remove = NULL,
  weightvar = NULL,
  surveydata = FALSE,
  textvar = NULL,
  model_object = FALSE
)

Arguments

normaldata

A data frame or data frame extension (e.g. a tibble).

outcomevar

A character string naming of factor variable in normaldata to use as the outcome.

expvars

A character vector with the names of the exposure variables (either numeric or factors). Any transformations or interactions to be included must also be specified, e.g. c("Var1", "I(Var1^2)", "Var2", "Var3*Var4").

number_decimals

An integer giving the number of decimals to show in the standardized output (default is two decimals).

alpha

A scalar, between 0 and 1 specifying the desired significance level of the confidence intervals (default is 0.05 which will yield the usual 95% confidence interval).

regtype

A character string specifying the analysis method. Can either be "logistic" for logistic regression (the default) or "log-linear" for log-linear regression. Log-linear regression can only be used with binomial, unconditional analysis.

matchgroup

Character string specifying a variable in normaldata to condition the analysis on. Can only be used in binomial logistic regression models (default is NULL).

matchtiemethod

Character string specifying the method for ties when using a matched/conditional analysis. The default options is "exact", however this option does not take weights into account for the analysis, so if weights (other than 1) are used, another option should be selected. Other options are "approximate", "efron", and "breslow" - for further explanations, see documentation for clogit.

values_to_remove

A Character vector specifying values to remove from ALL variables used in the regression before the analysis (default is NULL). This is useful if some value(s) are used consistently to encode missing/irrelevant in the data (e.g. c("888", "987") - normal missing (NA) don't need to be specified as it will be removed automatically. Do NOT remove the reference values as this will lead to unexpected results!

weightvar

A character string specifying a numeric variable in normaldata with pre-calculated weights for observations in the analysis. The default value NULL corresponds to weight 1 for all observations.

surveydata

A Boolean specifying whether the data comes from a survey (default is FALSE).

textvar

A character string with text (like a note) to be added to the output. The default value NULL corresponds to no added note.

model_object

A Boolean. If TRUE, returns the raw output object from the analysis instead of the standard output. This might be useful to see information not included in the standardized output (default is FALSE).

Value

A standardized analysis object with results from a model.

Author(s)

ASO

See Also

odds_ratio_function_repeated() to perform several analysis in one go.

Examples

### Binomial outcome
data(logan, package = "survival")

resp <- levels(logan$occupation)
n <- nrow(logan)
indx <- rep(1:n, length(resp))
logan2 <- data.frame(
  logan[indx,],
  id = indx,
  tocc = factor(rep(resp, each=n))
)
logan2$case <- (logan2$occupation == logan2$tocc)
logan2$case <- as.factor(logan2$case)
logan2$case <- relevel(logan2$case, ref = "FALSE")

# Standard binomial logistic regression but using interaction for exposures:
func_est1 <- odds_ratio_function(
  logan2,
  outcomevar = "case",
  expvars = c("tocc", "education", "tocc:education")
)


# Conditional binomial logistic regression with some extra text added:
func_est2 <- odds_ratio_function(
  logan2,
  outcomevar = "case",
  expvars = c("tocc", "tocc:education"),
  matchgroup = "id",
  textvar = "Testing function"
)


# Standard binomial logistic regression as survey data with no prepared
# weights:
func_est3 <- odds_ratio_function(
  logan2,
  outcomevar = "case",
  expvars = c("tocc", "education"),
  surveydata = TRUE
)

# Example changing significance level and the number of decimals in fixed
# output and adding some text:
func_est4 <- odds_ratio_function(
  logan2,
  outcomevar = "case",
  expvars = c("tocc", "education"),
  number_decimals = 5,
  alpha = 0.01,
  textvar = "Testing function"
)

# Getting raw output from the regression function:
func_est5 <- odds_ratio_function(
  logan2,
  outcomevar = "case",
  expvars = c("tocc", "education"),
  model_object = TRUE
)

### Polytomous/multinomial outcome
data(api, package = "survey")

# As normal data, but using weights:
func_est6 <- odds_ratio_function(
  apiclus2,
  outcomevar = "stype",
  expvars = c("ell", "meals", "mobility", "sch.wide"),
  weightvar = "pw"
)

# As survey data with weights:
func_est7 <- odds_ratio_function(
  apiclus2,
  outcomevar = "stype",
  expvars = c("ell", "meals", "mobility"),
  weightvar = "pw", surveydata = TRUE
)

# Binomial logistic regression with same data (by removing all observations
# with a specific value of outcome):
func_est8 <- odds_ratio_function(
  apiclus2,
  outcomevar = "stype",
  expvars = c("ell", "meals", "mobility"),
  weightvar = "pw",
  values_to_remove = c("E")
)