Type: Package
Title: Lagged Weighted Quantile Sum Regression
Version: 0.5.0
Description: Wrapper functions for the implementation of lagged weighted quantile sum regression, as per 'Gennings et al' (2020) <doi:10.1016/j.envres.2020.109529>.
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
Encoding: UTF-8
LazyData: true
RoxygenNote: 7.1.1
Imports: data.table, ggplot2, plyr, gridExtra, gWQS, gamm4
Suggests: knitr, rmarkdown
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2021-03-03 16:42:30 UTC; PCPC
Author: Paul Curtin [aut, cre], Stefano Renzetti [aut], Chris Gennings [aut]
Maintainer: Paul Curtin <paul.curtin@mssm.edu>
Depends: R (≥ 3.5.0)
Repository: CRAN
Date/Publication: 2021-03-04 10:00:02 UTC

Function to extract time-varying mixture (wqs) index from lWQS object

Description

Function to extract time-varying mixture (wqs) index from lWQS object

Usage

extract_mixture(lobj)

Arguments

lobj

An object returned from lWQS function

Value

Data frame containing the time index, wqs index estimated at each repeated measure, subject ID, and the outcome variable.

Examples


# identify predictor variables used in mixture
mixvars=names(lwqs_data)[5:9]

# run model. Note for example run-time only 1 bootstrap (b=1) is used. Set b to be >50
model=lwqs(data=lwqs_data,
           timevar="time",
           wqs_parms=list(formula=out ~ wqs,
              data = lwqs_data,
              mix_name=mixvars,
              b1_constr = TRUE,
              b1_pos=FALSE,
              b = 5,
              q = 5,
              validation = 0,
              family = "gaussian",
              seed = 1),
              outcome="out",
              ID="ID")

# use extract_mixture to access time-varying wqs index
mixtime=extract_mixture(model)

Function to extract time-varying weights from lWQS object

Description

Function to extract time-varying weights from lWQS object

Usage

extract_weights(lobj)

Arguments

lobj

An object returned from lWQS function

Value

A (long-form) data frame containing the time index and corresponding variable weights estimated in an lWQS

Examples


# identify predictor variables used in mixture
mixvars=names(lwqs_data)[5:9]

# run model
model=lwqs(data=lwqs_data,
           timevar="time",
           wqs_parms=list(formula=out ~ wqs,
              data = lwqs_data,
              mix_name=mixvars,
              b1_constr = TRUE,
              b1_pos=TRUE,
              b = 5,
              q = 5,
              validation = 0,
              family = "gaussian",
              seed = 1),
              outcome="out",
              ID="ID")

# use extract_weights to access time-varying predictor weights
timeweights=extract_weights(model)

Wrapper function for the implementaion of lagged WQS.

Description

Wrapper function for the implementaion of lagged WQS.

Usage

lwqs(
  data,
  timevar,
  wqs_parms,
  outcome,
  ID,
  rDLM_parms = list(formula = wqs ~ s(time, by = y, bs = "cr"), random = ~(1 | id))
)

Arguments

data

Data frame containing observations in long format.

timevar

Enquoted variable name identifying the repeated measure / time variable

wqs_parms

A list containing parameters to be passed to the WQS algorithm. See gWQS package for details.

outcome

An enquoted variable name identifying the outcome measure

ID

An enquoted variable name identifying the subject identifier

rDLM_parms

(optional). A list containing parameters to be passed to the GAM algorithm. See gamm4 package for details. Parameters wqs, time, by, and id (see above) are created by the lwqs function and passed to the gamm4 function automatically.

Value

The lwqs function returns a list containing final model output and time-specific model parameters.

parameters

This list contains several objects summarizing different stages of the lagged ensemble model. The first object, res, contains output from the gWQS algorithm applied to each discreet repeated measure in the overall model; see package gWQS for details. The second output, wqstime, provides the mixture index, identified as "wqs", estimated for each subject at each discrete time point. The third item, weightstime, provides the weights estimated for each predictor at each discrete time point.

plot

This list contains two plots (as grobs) which summarize output of the lwqs algorithm.

Examples

 # identify predictor variables used in mixture
mixvars=names(lwqs_data)[5:9]

model=lwqs(data=lwqs_data,
           timevar="time",
           wqs_parms=list(formula=out ~ wqs,
              data = lwqs_data,
              mix_name=mixvars,
              b1_constr = TRUE,
              b1_pos=TRUE,
              b = 5,
              q = 5,
              validation = 0,
              family = "gaussian",
              seed = 1),
              outcome="out",
              ID="ID")

Simulated dataset for accompanying vignette

Description

Simulated dataset for accompanying vignette

Usage

data(lwqs_data)

Value

A data frame containing simulated data to explore the lwqs algorithm. Variables included are as follows:

ID

Variable identifying each simulated subject. Data reflect 30 successive measures per subject.

Sex

A simulated binary covariate, either 1 or 0.

time

Variable identifying the successive timing of each repeated measure

out

Simulated outcome on standardized scale

pred1

First simulated time-varying predictor. This has a large positive association with "out" from times 11-20.

pred2

Second simulated time-varying predictor. This has a moderate positive association with "out" from times 11-20.

pred3

Third simulated time-varying predictor. This has a moderate negative association with "out" from times 1-10.

pred4

Fourth simulated time-varying predictor. This has a strong negative association with "out" from times 1-10.

pred5

Fifth simulated time-varying predictor. This has no significant association with "out".