Package 'dlim'

Title: Distributed Lag Interaction Model
Description: Collection of functions for fitting and interpreting distributed lag interaction models (DLIM). A DLIM regresses a scalar outcome on repeated measures of exposure and allows for modification by a continuous variable. Includes a dlim() function for fitting, predict() function for inference, and plotting functions for visualization. Details on methodology are described in Demateis et al. (2024) <doi:10.1002/env.2843>.
Authors: Danielle Demateis [aut, cre] , Kayleigh Keller [aut] , Ander Wilson [aut]
Maintainer: Danielle Demateis <[email protected]>
License: GPL (>= 3)
Version: 0.2.0
Built: 2025-02-25 04:31:40 UTC
Source: https://github.com/ddemateis/dlim

Help Index

Distributed Lag Interaction Model (DLIM)

Description

The package dlim contains functions to fit, perform inference and estimation on, and visualize a distributed lag interaction model (DLIM).

Modelling framework

A distributed lag interaction model (DLIM) is an extension of a distributed lag model. A DLIM regresses an outcome onto repeated measures of an exposure and allows for associations to vary by a single continuous modifier. More details on methodology are provided in the reference listed below.

Functions and data included in the package

To fit a DLIM using this package, use the dlim function, which calls the cross_basis function to create the cross-basis and estimates regression coefficients using gam from mgcv package.

The predict.dlim S3 function provides point-wise or cumulative effect estimates and uncertainty measures.

The plot_DLF and plot_cumulative functions provide plots of the modified distributed lag functions and the cumulative effect estimate curve.

Additional information

Additonal details on the package dlim are available in the vignette, available by typing:

vignette("dlimOverview")

The dlim package is available on the Comprehensive R Archive Network (CRAN). A development website is available on GitHub (github.com/ddemateis/dlim).

Please use citation("dlim") to cite this package.

Author(s)

Danielle Demateis, Kayleigh Keller, and Ander Wilson

Maintainer: Danielle Demateis <[email protected]>

References

Demateis et al. (2024) <doi:10.1002/env.2843>, avaibable at (arxiv.org/abs/2401.02939).

See Also

Type 'vignette(dlimOverview)' for a detailed description.

Build crossbasis

Description

Creates cross-basis using natural splines for regression in DLIM

Usage

cross_basis(
  x,
  M,
  L = NULL,
  argmod = list(),
  arglag = list(),
  model_type = "standard"
)

Arguments

x

a numeric time series vector of length n or matrix of lagged exposures (columns) for n individuals (rows)
M

vector of length n containing modifier values
L

a numeric vector of length 1 containing the number of lag terms. This is required if x is vector, and is not used if x is a matrix.
argmod

a list: $fun is the spline function for the modifier ("ps" or "cr" to penalize), $arg is a list of arguments for the spline function (must be named by argument), $df is the degrees of freedom, $sp is optional smoothing parameter
arglag

a list: $fun is the spline function for the lag ("ps" or "cr" to penalize), $arg is a list of arguments for the spline function (must be named by argument), $df is the degrees of freedom, $sp is optional smoothing parameter
model_type

"linear" for a DLIM with linear interaction, "quadratic" for a DLIM with quadratic interaction, "standard" for a DLIM with splines

Value

This function returns a list of 5 or 6 elements:

cb

cross-basis (matrix)
B_lag

lag basis (basis matrix)
B_mod

modifier basis (basis matrix)
df_l

lag degrees of freedom (numeric)
df_m

modifier degrees of freedom (numeric)
L

number of lags (numeric)
Slist

lag and modifier penalty matrices, if penalizing (list)

See Also

dlim

Type vignette('dlimOverview') for a detailed description.

Fit DLIM

Description

Fit distributed lag interaction model

Usage

dlim(
  y,
  x,
  modifiers,
  z = NULL,
  df_m = NULL,
  df_l,
  penalize = TRUE,
  pen_fn = "ps",
  mod_args = NULL,
  lag_args = NULL,
  fit_fn = "gam",
  model_type = "standard",
  ID = NULL,
  ...
)

Arguments

y

vector of response values (class "numeric")
x

matrix of exposure history (columns) for individuals (rows) (class "matrix")
modifiers

vector of modifying values (class "numeric")
z

matrix of covariates, not including the modifier (class "matrix")
df_m

degrees of freedom for modifier basis. Cannot specify for linear modification (model_type = "linear") (class "numeric")
df_l

degrees of freedom for exposure time basis (class "numeric")
penalize

TRUE to penalize model (class "logical")
pen_fn

if penalizing, can specify "ps" for penalized B-splines or "cr" for cubic regression splines with penalties on second derivatives
mod_args

a list of additional arguments for the spline function (must be named by argument)
lag_args

a list of additional arguments for the spline function (must be named by argument)
fit_fn

specify "gam" to use the gam function for data sets that are not very large, and specify "bam" to use the bam function for data sets that are very large. Default will fit using gam. (class "character")
model_type

"linear" for a DLIM with linear interaction, "quadratic" for a DLIM with quadratic interaction, "standard" for a DLIM with splines (class "character")
ID

group identifier for random intercept, only supported for penalized models
...

Other arguments to pass to model fitting function

Value

This function returns a list that is an object of class "dlim" with the following components

cb

cross-basis (class "matrix")
fit

model object (class "lm", "glm", "gam")
modifiers

modifying values (class "numeric")
call

model call

See Also

Type vignette('dlimOverview') for a detailed description.

predict.dlim

plot_cumulative

plot_DLF

Examples

library(dlim)
data("ex_data")
dlim_fit <- dlim(y = ex_data$y, 
                 x = ex_data$exposure, 
                 modifier = ex_data$modifier, 
                 z = ex_data$z, 
                 df_m = 10, 
                 df_l = 10, 
                 method = "REML")
dlim_pred <- predict(dlim_fit, 
                     newdata = 0.5, 
                     type="CE")

Example data set

Description

Data set for examples

Usage

ex_data

Format

List of response, exposure, modifiers, covariates

Source

Simulated

Examples

data(ex_data) # lazy load

Exposure data set for simulation

Description

Data set of PM 2.5 exposure history for 1000 individuals over 37 weeks

Usage

exposure

Format

A data frame of 1000 rows and 37 columns

Source

Data source??

Examples

data(exposure) # lazy load

Model Comparison

Description

Compare models to test for interaction

Usage

model_comparison(fit, null = "DLM", x, B, conf.level = 0.95)

Arguments

fit

dlim object (must be fit with REML)
null

"DLM", "linear" to specify the null model
x

exposure
B

number of bootstrap samples
conf.level

The confidence level (class "numeric")

Value

The function returns a decision to either reject or fail to reject the null model

See Also

Type vignette('dlimOverview') for a detailed description.

dlim

Plot Distributed Lag Function

Description

Plot estimated distributed lag function values from a DLIM object, can also compare those of a DLM

Usage

plot_cumulative(
  new_modifiers,
  mod_fit,
  dlm_fit = NULL,
  mod_name = NULL,
  mod_trans = NULL,
  link_trans = NULL
)

Arguments

new_modifiers

a vector of new modifier values for prediction (class "numeric")
mod_fit

DLIM model object (class "dlim")
dlm_fit

a list containing a crossbasis object from the dlnm package as the first element and a DLM model object as the second element (class "list")
mod_name

modifier name (character)
mod_trans

if modifiers are transformed, specify back transformation function (class "character")
link_trans

if family for glm is not Gaussian, specify back transformation to undo link function (class "character")

Value

This function returns a ggplot for cumulative effects, including for a DLM if specified

See Also

dlim

Type vignette('dlimOverview') for a detailed description.

Plot Cumulative Effects

Description

Plot estimated cumulative effects from a DLIM object, can also compare estimated cumulative effects between a DLM and DLIM

Usage

plot_DLF(
  new_modifiers,
  mod_fit,
  mod_name,
  dlm_fit = NULL,
  plot_by,
  time_pts = NULL,
  mod_trans = NULL,
  link_trans = NULL
)

Arguments

new_modifiers

a vector of new modifier values for prediction (class "numeric")
mod_fit

DLIM model object (class "dlim")
mod_name

modifier name that follows variable name nomenclature (class "character")
dlm_fit

a list containing a crossbasis object from the dlnm package as the first element and a DLM model object as the second element (class "list")
plot_by

choose to create plots for particular modifier values, "modifier", or particular time points, "time", (class "character")
time_pts

a set of time points if plotting by time (class "numeric")
mod_trans

if modifiers are transformed, specify back transformation function (class "character")
link_trans

if family for glm is not Gaussian, specify back transformation to undo link function (class "character")

Value

This function returns a ggplot for point-wise effects isolated by either time points or modifier, including a DLM if specified

See Also

dlim

Type vignette('dlimOverview') for a detailed description.

DLIM Predictions

Description

Predicted values based on a dlim object.

Usage

## S3 method for class 'dlim'
predict(
  object,
  newdata = NULL,
  type = c("DLF", "CE", "response"),
  conf.level = 0.95,
  ...
)

Arguments

object

an object of class "dlim"
newdata

a vector of new modifier values for prediction (class "numeric")
type

Type of prediction. "DLF" for the estimated distributed lag functions, "CE" for cumulative effects, "response" for fitted values, or any combination of these in a vector (class "character")
conf.level

The confidence level (class "numeric")
...

additional arguments affecting the predictions produced

Value

This function returns a list of 3 elements:

est_dlim

cumulative and/or point-wise estimates, standard errors, and confidence intervals (class "list")
cb

cross-basis object (class "cross-basis")
model

model object (class "gam")

See Also

dlim

Type vignette('dlimOverview') for a detailed description.

Simulated DLIM Predictions

Description

This function estimates cumulative and non-cumulative lag/modifier coefficients from a model in which the response is regressed on a cross-basis generated by the cross_basis() function.

Usage

## S3 method for class 'sim_dlim'
predict(object, newdata = NULL, type = c("DLF", "CE", "response"), ...)

Arguments

object

an object of class "dlim"
newdata

vector of modifiers for inference (class "numeric")
type

Type of prediction. "response" for predicted responses, "DLF" for the estimated distributed lag functions, "CE" for cumulative effects (class "character")
...

additional arguments affecting the predictions produced

Value

This function returns a list of 4 or 7 elements:

est_dlim

est_dlim element from predict.dlim (class "list")
cb

cross-bais from object (class "cross-basis")
fit

fit from object (class "lm", "glm", "gam")
true_betas

true_betas from object (class "matrix")
cb_dlm

cb_dlm from object (class "crosspred")
model_dlm

model_dlm from object (class "lm", "glm", "gam")
est_dlm

cumulative and/or point-wise estimates, standard errors, and confidence intervals for the DLM (class "list")

See Also

predict.dlim

Type vignette('dlimOverview') for a detailed description.

Print DLIM Information

Description

prints information about an object of class dlim

Usage

## S3 method for class 'dlim'
print(x, ...)

Arguments

x

a dlim object
...

further arguments passed to or from other methods

Value

This function returns information about an object of class dlim

See Also

Type vignette('dlimOverview') for a detailed description.

Simulate Data

Description

Simulate data to use with the dlim package. There are different effect modification scenarios to choose for simulation.

Usage

sim_data(
  x,
  L = NULL,
  modifiers,
  noise = 1,
  type = 2,
  SNR,
  ncovariates = 0,
  gamma = 1
)

Arguments

x

a time series vector of length n or matrix of lagged exposures for n individuals (class "numeric", "matrix")
L

a vector of length 1 containing the number of lag terms. This is required if x is vector, and is not used if x is a matrix (class "numeric")
modifiers

vector of length n containing modifying values (class "numeric")
noise

a vector of length 1 containing the standard deviation for a normal distribution with mean 0 used to add noise to the simulated response values. Must proivde if SNR is not provided (class "numeric")
type

a vector containing the number 1, 2, 3, or 4 for simulation modification type: none, linear, non-linear shift, non-linear shift with linear scale (class "numeric")
SNR

The signal-to-noise ratio. If SNR is provided, but noise is not, noise is reset to be the standard deviation of the response, before adding noise. (class "numeric")
ncovariates

number of covariates to add to the model, numeric vector of length 1.
gamma

True coefficient for the main effect of the modifier (class "numeric")

Value

This returns a list of 8 items:

x

a lagged exposure matrix. If x was a matrix, it is unchanged. (class "matrix")
L

a numeric vector of length 1 containing the number of lag terms (class "numeric")
modifiers

the modifiers argument (class "numeric")
y

a numeric vector of length nrow(x) containing the perturbed simulated response values. (class "numeric")
betas

a matrix containing true coefficients for each lag/modifier combination, with each row representing a lag and each column a modifier (class "matrix")
betas_cumul

a numeric vector of length L+1 containing cumulative true coefficients for the lag terms, summed over modifiers (class "numeric")
Z

covariates (class "matrix")
gammas

true coefficients for the covariates (class "numeric")

See Also

sim_dlim

Type vignette('dlimOverview') for a detailed description.

Simulate Distributed Lag Functions

Description

generate true distributed lag function values for a given type of simulation

Usage

sim_dlf(L, modifiers, type)

Arguments

L

Number of lags minus 1
modifiers

Vector of modifiers
type

Effect modification simulation type: 1 is no modification, 2 is linear scale modification, 3 is non-linear shift modification, 4 is types 2 and 3 combined

Value

This function returns the true distributed lag function values (class "numeric")

See Also

sim_data

Type vignette('dlimOverview') for a detailed description.

Fit DLIM for simulation

Description

Fit DLIM for simulation

Usage

sim_dlim(
  data,
  df_m,
  df_l,
  penalize = TRUE,
  pen_fn = "ps",
  mod_args = NULL,
  lag_args = NULL,
  fit_dlm = FALSE,
  model_type = "standard",
  ...
)

Arguments

data

output from sim_data
df_m

degrees of freedom for modifiers
df_l

degrees of freedom for lags
penalize

True to penalize model
pen_fn

if penalizing, can specify "ps" for penalized B-splines or "cr" for cubic regression splines with penalties on second derivatives
mod_args

a list of additional arguments for the spline function (must be named by argument)
lag_args

a list of additional arguments for the spline function (must be named by argument)
fit_dlm

True to additionally fit dlm for comparison
model_type

"linear" for a DLIM with linear interaction, "quadratic" for a DLIM with quadratic interaction, "standard" for a DLIM with splines
...

arguments to pass to model fitting function

Value

This function returns an object of class "sim_dlim"

cb

DLIM cross-basis (class "cross-basis")
fit

DLIM model fit (class "lm", "glm", "gam")
cb_dlm

DLM cross-basis (class "crossbasis")
model_dlm

DLM model fit (class "lm", "glm", "gam")
true_betas

true linear effect of the exposure on the response for each individual and time point (class "matrix")
modifiers

modifiers from numeric
data

data (class "list")

See Also

dlim

sim_data

Type vignette('dlimOverview') for a detailed description.

Summarizing DLIM

Description

prints summary of object of class dlim

Usage

## S3 method for class 'dlim'
summary(object, ...)

Arguments

object

a dlim object
...

additional arguments affecting the summary produced

Value

This function returns a summary for an object of class dlim

See Also

Type vignette('dlimOverview') for a detailed description.