CMA_sliding_window {AdhereR}R Documentation

CMA_sliding_window constructor.

Description

Applies a given CMA to each sliding window and constructs a CMA_sliding_window object.

Usage

CMA_sliding_window(
  CMA.to.apply,
  data,
  ID.colname = NA,
  event.date.colname = NA,
  event.duration.colname = NA,
  event.daily.dose.colname = NA,
  medication.class.colname = NA,
  medication.groups = NULL,
  flatten.medication.groups = FALSE,
  medication.groups.colname = ".MED_GROUP_ID",
  carry.only.for.same.medication = NA,
  consider.dosage.change = NA,
  followup.window.start = 0,
  followup.window.start.unit = c("days", "weeks", "months", "years")[1],
  followup.window.start.per.medication.group = FALSE,
  followup.window.duration = 365 * 2,
  followup.window.duration.unit = c("days", "weeks", "months", "years")[1],
  observation.window.start = 0,
  observation.window.start.unit = c("days", "weeks", "months", "years")[1],
  observation.window.duration = 365 * 2,
  observation.window.duration.unit = c("days", "weeks", "months", "years")[1],
  sliding.window.start = 0,
  sliding.window.start.unit = c("days", "weeks", "months", "years")[1],
  sliding.window.duration = 90,
  sliding.window.duration.unit = c("days", "weeks", "months", "years")[1],
  sliding.window.step.duration = 30,
  sliding.window.step.unit = c("days", "weeks", "months", "years")[1],
  sliding.window.no.steps = NA,
  return.inner.event.info = FALSE,
  date.format = "%m/%d/%Y",
  summary = "CMA per sliding window",
  event.interval.colname = "event.interval",
  gap.days.colname = "gap.days",
  force.NA.CMA.for.failed.patients = TRUE,
  return.mapping.events.sliding.window = FALSE,
  parallel.backend = c("none", "multicore", "snow", "snow(SOCK)", "snow(MPI)",
    "snow(NWS)")[1],
  parallel.threads = "auto",
  suppress.warnings = FALSE,
  suppress.special.argument.checks = FALSE,
  ...
)

Arguments

CMA.to.apply

A string giving the name of the CMA function (1 to 9) that will be computed for each treatment episode.

data

A data.frame containing the events used to compute the CMA. Must contain, at a minimum, the patient unique ID, the event date and duration, and might also contain the daily dosage and medication type (the actual column names are defined in the following four parameters).

ID.colname

A string, the name of the column in data containing the unique patient ID; must be present.

event.date.colname

A string, the name of the column in data containing the start date of the event (in the format given in the date.format parameter); must be present.

event.duration.colname

A string, the name of the column in data containing the event duration (in days); must be present.

event.daily.dose.colname

A string, the name of the column in data containing the prescribed daily dose, or NA if not defined.

medication.class.colname

A string, the name of the column in data containing the medication type, or NA if not defined.

medication.groups

A vector of characters defining medication groups or the name of a column in data that defines such groups. The names of the vector are the medication group unique names, while the content defines them as logical expressions. While the names can be any string of characters except "}", it is recommended to stick to the rules for defining vector names in R. For example, c("A"="CATEGORY == 'medA'", "AA"="{A} & PERDAY < 4" defines two medication groups: A which selects all events of type "medA", and B which selects all events already defined by "A" but with a daily dose lower than 4. If NULL, no medication groups are defined. If medication groups are defined, there is one CMA estimate for each group; moreover, there is a special group __ALL_OTHERS__ automatically defined containing all observations not covered by any of the explicitly defined groups.

flatten.medication.groups

Logical, if FALSE (the default) then the CMA and event.info components of the object are lists with one medication group per element; otherwise, they are data.frames with an extra column containing the medication group (its name is given by medication.groups.colname).

medication.groups.colname

a string (defaults to ".MED_GROUP_ID") giving the name of the column storing the group name when flatten.medication.groups is TRUE.

carry.only.for.same.medication

Logical, if TRUE, the carry-over applies only across medication of the same type.

consider.dosage.change

Logical, if TRUE, the carry-over is adjusted to also reflect changes in dosage.

followup.window.start

If a Date object, it represents the actual start date of the follow-up window; if a string it is the name of the column in data containing the start date of the follow-up window either as the numbers of followup.window.start.unit units after the first event (the column must be of type numeric) or as actual dates (in which case the column must be of type Date or a string that conforms to the format specified in date.format); if a number it is the number of time units defined in the followup.window.start.unit parameter after the begin of the participant's first event; or NA if not defined.

followup.window.start.unit

can be either "days", "weeks", "months" or "years", and represents the time units that followup.window.start refers to (when a number), or NA if not defined.

followup.window.start.per.medication.group

a logical: if there are medication groups defined and this is TRUE, then the first event considered for the follow-up window start is relative to each medication group separately, otherwise (the default) it is relative to the patient.

followup.window.duration

either a number representing the duration of the follow-up window in the time units given in followup.window.duration.unit, or a string giving the column containing these numbers. Should represent a period for which relevant medication events are recorded accurately (e.g. not extend after end of relevant treatment, loss-to-follow-up or change to a health care provider not covered by the database).

followup.window.duration.unit

can be either "days", "weeks", "months" or "years", and represents the time units that followup.window.duration refers to, or NA if not defined.

observation.window.start, observation.window.start.unit, observation.window.duration, observation.window.duration.unit

the definition of the observation window (see the follow-up window parameters above for details).

sliding.window.start, sliding.window.start.unit, sliding.window.duration, sliding.window.duration.unit

the definition of the first sliding window (see the follow-up window parameters above for details).

sliding.window.step.duration, sliding.window.step.unit

if not missing (NA), these give the step (or "jump") to the right of the sliding window in time units.

sliding.window.no.steps

a integer specifying the desired number of sliding windows to cover the observation window (if possible); trumps sliding.window.step.duration and sliding.window.step.unit.

return.inner.event.info

Logical specifying if the function should also return the event.info for all the individual events in each sliding window; by default it is FALSE as this information is useful only in very specific cases (e.g., plotting the event intervals) and adds a small but non-negligible computational overhead.

date.format

A string giving the format of the dates used in the data and the other parameters; see the format parameters of the as.Date function for details (NB, this concerns only the dates given as strings and not as Date objects).

summary

Metadata as a string, briefly describing this CMA.

event.interval.colname

A string, the name of a newly-created column storing the number of days between the start of the current event and the start of the next one; the default value "event.interval" should be changed only if there is a naming conflict with a pre-existing "event.interval" column in event.info.

gap.days.colname

A string, the name of a newly-created column storing the number of days when medication was not available (i.e., the "gap days"); the default value "gap.days" should be changed only if there is a naming conflict with a pre-existing "gap.days" column in event.info.

force.NA.CMA.for.failed.patients

Logical describing how the patients for which the CMA estimation fails are treated: if TRUE they are returned with an NA CMA estimate, while for FALSE they are omitted.

return.mapping.events.sliding.window

A Logical, if TRUE then the mapping between events and sliding windows is returned as the component mapping.windows.to.events, which is a data.table giving, for each sliding window, the events that belong to it (an event is given by its row number in the data).

parallel.backend

Can be "none" (the default) for single-threaded execution, "multicore" (using mclapply in package parallel) for multicore processing (NB. not currently implemented on MS Windows and automatically falls back on "snow" on this platform), or "snow", "snow(SOCK)" (equivalent to "snow"), "snow(MPI)" or "snow(NWS)" specifying various types of SNOW clusters (can be on the local machine or more complex setups – please see the documentation of package snow for details; the last two require packages Rmpi and nws, respectively, not automatically installed with AdhereR).

parallel.threads

Can be "auto" (for parallel.backend == "multicore", defaults to the number of cores in the system as given by options("cores"), while for parallel.backend == "snow", defaults to 2), a strictly positive integer specifying the number of parallel threads, or a more complex specification of the SNOW cluster nodes for parallel.backend == "snow" (see the documentation of package snow for details).

suppress.warnings

Logical, if TRUE don't show any warnings.

suppress.special.argument.checks

Logical parameter for internal use; if FALSE (default) check if the important columns in the data have some of the reserved names, if TRUE this check is not performed.

...

other possible parameters

Details

CMA_sliding_window first computes a set of fixed-size (possibly partly overlapping) sliding windows, each sliding to the right by a fixed timelag, and then, for each of them, it computes the given "simple" CMA. Thus, as opposed to the "simple" CMAs 1 to 9, it returns a set of CMAs, with possibly more than one element.

It is highly similar to CMA_per_episode which computes a CMA for a set of treatment episodes.

Value

An S3 object of class CMA_sliding_window with the following fields:

Please note that if medication.groups are defined, then the CMA and event.info are named lists, each element containing the CMA and event.info corresponding to a single medication group (the element's name).

See Also

CMA_per_episode is very similar, computing a "simple" CMA for each of the treatment episodes. The "simple" CMAs that can be computed comprise CMA1, CMA2, CMA3, CMA4, CMA5, CMA6, CMA7, CMA8, CMA9, as well as user-defined classes derived from CMA0 that have a CMA component giving the estimated CMA per patient as a data.frame. If return.mapping.events.sliding.window is TRUE, then this also has an attribute mapping.windows.to.events that gives the mapping between episodes and events as a data.table with the following columns:

Examples

## Not run: 
cmaW <- CMA_sliding_window(CMA="CMA1",
                           data=med.events,
                           ID.colname="PATIENT_ID",
                           event.date.colname="DATE",
                           event.duration.colname="DURATION",
                           event.daily.dose.colname="PERDAY",
                           medication.class.colname="CATEGORY",
                           carry.only.for.same.medication=FALSE,
                           consider.dosage.change=FALSE,
                           followup.window.start=0,
                           observation.window.start=0,
                           observation.window.duration=365,
                           sliding.window.start=0,
                           sliding.window.start.unit="days",
                           sliding.window.duration=90,
                           sliding.window.duration.unit="days",
                           sliding.window.step.duration=7,
                           sliding.window.step.unit="days",
                           sliding.window.no.steps=NA,
                           date.format="%m/%d/%Y"
                          );
## End(Not run)

[Package AdhereR version 0.8.1 Index]