calc_nb_ratioci: Calculate incidence rate ratio and confidence limits from a...

In jenniferthompson/JTHelpers: Personal Helper Functions, Some Related to rms Package

Description

The most common way to express incidence rate ratios is simply exp(beta), representing a one-unit increase in the covariate value; however, in the case of continuous covariates, this is often not a practically meaningful difference (one year of age among adults, eg, or a one-unit change in mean arterial pressure). This function allows you to specify a clinically meaningful comparison in the case of continuous covariates, defaulting to the 75th vs the 25th percentiles of values in the data frame specified.

Usage

calc_nb_ratioci(nbObj = NULL, ...)

## Default S3 method:
calc_nb_ratioci(nbObj = NULL, nbCoefs, nbVcov, predVar,
  adjustTo = NULL, df, alpha = 0.05)

## S3 method for class 'mira'
calc_nb_ratioci(nbObj, predVar, adjustTo = NULL, df,
  alpha = 0.05)

## S3 method for class 'negbin'
calc_nb_ratioci(nbObj, predVar, adjustTo = NULL, df,
  alpha = 0.05)

Arguments

nbObj	Model fit of class glm.nb, or mice::mira object fit using glm.nb.
nbCoefs	Vector of coefficients from a glm.nb.
nbVcov	Variance-covariance matrix from a glm.nb fit.
predVar	Character string; name of main predictor variable.
adjustTo	Numeric vector of length 2; values to adjust predVar to if predVar is continuous. Defaults to c(25th, 75th percentiles).
df	Data frame used to determine type of predVar, get defaults for adjustTo and to calculate nonlinear terms if applicable.
alpha	Alpha level for confidence limits. Defaults to 0.05.

Details

Currently does not handle interaction terms.

Value

matrix with one row per comparison, containing columns for point estimate, confidence limits, and reference and comparison levels used.

Methods (by class)

• default: Method used when passing coefficients and variance-covariance matrix, vs original model fit

  Default method uses given coefficients and variance-covariance matrix. Intended use cases include bootstrapped negative binomial models.

• mira: Method for glm.nb models fit with mice objects.

• negbin: Method for glm.nb models (no imputation).

See Also

glm.nb; mice, pool; rcspline.eval for nonlinear terms.

glm.nb; rcspline.eval for nonlinear terms.

glm.nb; mice rcspline.eval for nonlinear terms.

glm.nb; rcspline.eval for nonlinear terms.

Examples

## Create data frame
df <- data.frame(y = round(rexp(n = 100, rate = 0.5)),
                 v1 = sample(1:5, size = 100, replace = TRUE),
                 v2 = rnorm(n = 100))

## Fit negative binomial model
mymod <- MASS::glm.nb(y ~ v1 * v2, data = df)

calc_nb_ratioci(mymod, predVar = 'v1', df = df)


## Create data frame
df <- data.frame(y = round(rexp(n = 100, rate = 0.5)),
                 v1 = sample(1:5, size = 100, replace = TRUE),
                 v2 = rnorm(n = 100))

## Fit negative binomial model
mymod <- MASS::glm.nb(y ~ v1 * v2, data = df)
mycoefs <- coef(mymod)
myvcov <- vcov(mymod)

calc_nb_ratioci(nbCoefs = mycoefs, nbVcov = myvcov, predVar = 'v1', df = df)


## Create data frame
df <- data.frame(y = round(rexp(n = 100, rate = 0.5)),
                 v1 = sample(c(1:5, NA), size = 100, replace = TRUE),
                 v2 = rnorm(n = 100))

## Impute missing data using mice
mids.df <- mice(df)

## Fit negative binomial model
mymod <- with(mids.df, MASS::glm.nb(y ~ v1 * v2))

calc_nb_ratioci(mymod, predVar = 'v1', df = df)


## Create data frame
df <- data.frame(y = round(rexp(n = 100, rate = 0.5)),
                 v1 = sample(1:5, size = 100, replace = TRUE),
                 v2 = rnorm(n = 100))

## Fit negative binomial model
mymod <- MASS::glm.nb(y ~ v1 * v2, data = df)

calc_nb_ratioci(mymod, predVar = 'v1', df = df)

jenniferthompson/JTHelpers documentation built on May 19, 2019, 4:04 a.m.