Skip to contents

Check Pmetrics Inputs for Errors

Source: R/PMcheck.R
PMcheck.Rd

[Superseded]

This function is largely superseded as it is called automatically when data are initialized as a PM_data object. It can still be called independently of this route and will check for data and optionally model errors.

Usage

PMcheck(data, model, fix = F, quiet = F)

Arguments

data

The name of a Pmetrics .csv matrix file in the current working directory, the full path to one not in the current working directory, or a data.frame containing the output of a previous PMreadMatrix command.

model

The filename of a Pmetrics model file in the current working directory. This parameter is optional. If specified, and the data object has no errors, the model file will be evaluated.

fix

Boolean operator; if TRUE, Pmetrics will attempt to fix errors in the data file. Default is FALSE.

quiet

Boolean operator to suppress printed output. Default is false.

Value

If fix=TRUE, then PMcheck returns

  • The original data if no errors are found, or

  • A PMmatrix data object which has been cleaned of errors as much as possible, displaying a report on the console.

If fix=FALSE, then PMcheck creates a file in the working directory called "errors.xlsx". This file can be opened by Microsoft Excel or any other program that is capable of reading .xlsx files. This file contains highlighted areas that are erroneous, with clarifying comments. You can correct the errors in the file and then re-save as a .csv file.

When fix=FALSE, the function also returns a list of objects of class PMerr. Each object is itself a list whose first object ($msg) is a character vector with "OK" plus a brief description if there is no error, or the error. The second object ($results) is a vector of the row numbers that contain that error.

  • colorder The first 14 columns must be named id, evid, time, dur, dose, addl, ii, input, out, outeq, c0, c1, c2, and c3 in that order.

  • maxcharCol All column names should be less than or equal to 11 characters.

  • maxcharID All id values should be less than or equal to 11 characters.

  • missEVID Ensure that all rows have an EVID value.

  • missTIME Ensure that all rows have a TIME value.

  • doseDur Make sure all dose records are complete, i.e. contain a duration.

  • doseDose Make sure all dose records are complete, i.e. contain a dose.

  • doseInput Make sure all dose records are complete, i.e. contain an input number.

  • obsOut Make sure all observation records are complete, i.e. contain an output.

  • obsOuteq Make sure all observation records are complete, i.e. contain and outeq number.

  • T0 Make sure each subject's first time=0.

  • covT0 Make sure that there is an non-missing entry for each covariate at time=0 for each subject.

  • timeOrder Ensure that all times within a subject ID are monotonically increasing.

  • contigID Ensure that all subject IDs are contiguous.

  • nonNum Ensure that all columns except ID are numeric.

  • noObs Ensure that all subjects have at least one observation, which could be missing, i.e. -99.

  • mal_NA Ensure that all NA values are ".", not ". ", " .", "..", or other malformations.

Details

It will check a .csv file or a data frame containing a previously loaded .csv file (the output of PMreadMatrix for errors which would cause the analysis to fail. If a model file is provided, and the data file has no errors, it will also check the model file for errors. Note that as of Pmetrics Version 2, this function is called automatically when a new PM_data object is created, and users generally no longer need to call the function directly. In PM_data$new(), the data object is first standardized to contain all required columns, since only "ID", "TIME", "DOSE" and "OUT" are required at minimum, and then checked with PMcheck.

If calling PMcheck directly, either a filename or a Pmetrics data object in memory are accepted as data. Because there is no standardization with direct calls, in this case the format of the .csv matrix file is fairly rigid. It must have the following features. Text is case-sensitive.

  • A header in row 1 with the appropriate version, currently "POPDATA DEC_11"

  • Column headers in row 2. These headers are: #ID, EVID, TIME, DUR, DOSE, ADDL, II, INPUT, OUT, OUTEQ, C0, C1, C2, C3.

  • No cell should be empty. It should either contain a value or "." as a placeholder.

  • Columns after C3 are interpreted as covariates.

  • All subject records must begin with TIME=0.

  • All dose events (EVID=1) must have entries in ID, EVID, TIME, DUR, DOSE and INPUT. ADDL and II are optional, but if ADDL is not 0 or missing, then II is mandatory.

  • All observation events (EVID=0) must have entries in ID, EVID, TIME, OUT, OUTEQ. If an observation is missing, use -99; otherwise use a "." as a placeholder in cells that are not required (e.g. INPUT for an observation event).

  • If covariates are present in the data, there must be an entry for every covariate at time 0 for each subject.

  • All covariates must be numeric.

  • All times within a subject ID must be monotonically increasing.

  • All subject IDs must be contiguous.

  • All rows must have EVID and TIME values.

  • All columns must be numeric except ID which may be alpha-numeric.

  • All subjects must have at least one observation, which could be missing, i.e. -99.

  • Cells which are not needed (e.g. dose on an observation event, EVID=0), should contain ".".

To use this function, see the example below.

After running PMcheck and looking at the errors in the errors.xlsx file, you can fix the errors manually directly in the errors.xlsx file and resave it as a .csv file. Alternatively, you could then try to fix the problem(s) with mdata2 <- PMcheck(mdata,fix=T). Note that we are now returning a PMmatrix data object called mdata2 (hopefully cleaned of errors) rather than the PMerr object returned when fix=FALSE. Pmetrics handles each of the errors in the following ways.

  • If the columns are simply out of order, they will be reordered. If some are missing, the fix must be done by the user, i.e. manually.

  • All id and covariate values are truncated to 11 characters.

  • Missing observations are set to -99 (not ".").

  • Incomplete dose records are flagged for the user to fix manually.

  • Incomplete observation records are flagged for the user to fix manually.

  • Subjects without an EVID=1 as first event are flagged for the user to fix manually.

  • Subjects with TIME != 0 as first event have dummy dose=0 events inserted at time 0.

  • Subjects with a missing covariate at time 0 are flagged for the user to fix manually.

  • Non-numeric covariates are converted to numeric (via factor()).

  • Non-ordered times are sorted within a subject if there are no EVID=4 events; otherwise the user must fix manually.

  • Non-contiguous subject ID rows are combined and sorted if there are no EVID=4 events; otherwise the user must fix manually.

  • Rows missing an EVID are assigned a value of 0 if DOSE is missing, 1 otherwise.

  • Rows missing a TIME value are flagged for the user to fix manually.

  • Cells with malformed NA values are attempted to be fixed.

  • Columns that are non-numeric which must be numeric are flagged for the user to fix manually. These are EVID, TIME, DUR, DOSE, ADDL, II, INPUT, OUT, OUTEQ, C0, C1, C2, and C3. Covariate columns are fixed separately (see above).

See also

PMwriteMatrix, PMreadMatrix

Author

Michael Neely and Patrick Nolain

Examples

if (FALSE) {
err <- PMcheck(badData)
#look at the errors.xlsx file in the working directory
#try to automatically fix what can be fixed
goodData <- PMcheck(badCSV,fix=T)
PMcheck(goodData)
#you have to fix manually problems which require data entry
}