Introduction to PROscorerTools

Overview

PROscorerTools provides tools to score patient-reported outcome (PRO) measures and other quality of life (QoL) and psychometric instruments. PROscorerTools also provides the building blocks of the functions in the PROscorer package.

PROscorerTools contains several “helper” functions, each of which performs a specific task that is common when scoring PRO-like instruments (e.g., reverse coding items). But most users will find that the scoreScale() function alone can address their scoring needs.

The scoreScale() Function

The workhorse function in PROscorerTools is the scoreScale() function. Its basic job is to take a data frame containing responses to some items, and output a single score for those items.

The scoreScale() function has simple, flexible arguments that enable it to handle nearly all scoring situations.

Features:

• Reverse Coding: Before calculating a score, scoreScale() can reverse code all of the items, only some specific items, or none of the items (no reverse coding is the default).

• Different Types of Scores: Some instruments need to be scored by summing item responses, others by taking the mean of item responses, and others by re-scaling the sum or mean scores to range from 0 to 100. All 3 of these score types are available in the scoreScale() function.

• Calculation of Scores with Missing Items: For most instruments, valid scores can be obtained despite a certain number of missing item responses. For example, on the EORTC QLQ-C30, a score can be calculated as long as at least 50% of items on a given scale are non-missing. The scoreScale() function allows the user to specify the proportion of missing items that is allowed, and the score is prorated to be comparable to scores with no missing items. If a respondent has more than the allowed proportion of missing items, then that respondent will be assigned a missing score (e.g., NA).

• Scoring Instruments with Multiple Scores: More complex instruments that yield more than a single score can be scored by combining multiple calls to the scoreScale() function. In fact, most of the functions in the PROscorer package call scoreScale() multiple times.

Installation and Basic Usage

Install the stable version from CRAN (recommended):

install.packages("PROscorerTools")

If you want to contribute to the development of the PROscorerTools or PROscorer packages, then you can install the development version from GitHub (generally NOT recommended):

devtools::install_github("MSKCC-Epi-Bio/PROscorerTools")

Load PROscorerTools in your R workspace:

library(PROscorerTools)

As an example, we will use the makeFakeData() function to make a data frame of responses to 6 fake items from 20 imaginary respondents. The created data set (named “dat”) has an “id” variable, plus responses to 6 items (named “q1”, “q2”, etc.) from 20 imaginary respondents. There are also missing responses (“NA”) scattered throughout.

dat <- makeFakeData(n = 20, nitems = 6, values = 0:4, id = TRUE)

Below we use the scoreScale function to score the fake responses in “dat”. We use the items argument to tell scoreScale which variables are the items we want to score. We will score the items by summing the responses (type = "sum"). We will save the scores from the fake questionnaire in a data frame named “dat_scored”.

dat_scored <- scoreScale(df = dat, items = 2:7, type = "sum")
dat_scored

By default, scoreScale will score the items for a given respondent as long as the respondent is missing no more than 50% of the items. This can be changed with the okmiss argument. Above, okmiss = 0.50 by default, so a respondent could be missing 3 of the 6 items and still be assigned a score (if missing 4 or more items, they were assigned a score of NA). Below, we again score the items, but this time we allow less than half of the items to be missing to be scored (okmiss = 0.49).

dat_scored <- scoreScale(df = dat, items = 2:7, type = "sum", okmiss = 0.49)
dat_scored

By default, scoreScale will score the items for a given respondent as long as the respondent is missing no more than 50% of the items. This can be changed with the okmiss argument. Above, okmiss = 0.50 by default, so a respondent could be missing 3 of the 6 items and still be assigned a score (if missing 4 or more items, they were assigned a score of NA). Below, we again score the items, but this time we allow less than half of the items to be missing to be scored (okmiss = 0.49).

dat_scored <- scoreScale(df = dat, items = 2:7, type = "sum", okmiss = 0.49)
dat_scored

For more information on the scoreScale function, you can access its “help” page by typing ?scoreScale into R.

Future Development Plans

The PROscorer family of R packages includes PROscorer, PROscorerTools, and FACTscorer. My priorities for developing these 3 packages are:

1. Streamline how the packages check arguments and processes input to scoreScale and other custom-written scoring functions. The current system gets the job done, but it is not very pretty. I believe that a more elegant, easy-to-use system for performing these tasks (possibly using the assertive package) will speed up the expansion of the PROscorer package (which contains custom scoring functions for specific, commonly-used PROs). I hope to have a stable version of this system in the next major update of PROscorerTools.

2. Make the unit testing framework of PROscorer and PROscorerTools more comprehensive. Most of the code underlying the PROscorer functions will be already be tested by the PROscorerTools tests; however, I intend to come up with a standard set of tests for PROscorer functions to make it easier for me and others to add unit tests to their scoring functions.

3. Expand PROscorer with more scoring functions for specific PROs.

4. Finalize the collaborative infrastructure (e.g., on GitHub) by which users can use PROscorerTools to write scoring functions for their favorite PROs and submit them for inclusion in PROscorer. A major component of this is to add a few instructional vignettes, including a step-by-step guide for writing the scoring functions, guidelines for writing the instrument descriptions, and templates for writing the function documentation.

5. Update the FACTscorer package. FACTscorer scores the FACT (Functional Assessment of Cancer Therapy) and FACIT (Functional Assessment of Chronic Illness Therapy) family of measures.
   Before writing PROscorerTools, I had completely re-written most of the underlying FACTscorer code to be more foolproof and easier to update in the future. I also wrote an “Instrument Descriptions” vignette, similar to what is included with PROscorer. I will put the finishing touches on the FACTscorer update and release it as soon as the above work is done.

6. Add capability to generate IRT-based scores for PROs that use that scoring method. I know many researchers that use various PROMIS measures. They would prefer to use the IRT-based scoring method, but find it too difficult to integrate into their research workflow. PROscorer could make IRT-based scores accessible to a much wider group of researchers.

Resources for More Information

• You can access the “help” page for the PROscorerTools package by typing ?PROscorerTools into R.

• If you have any feature requests, or you want to report bugs or other strange behavior in PROscorerTools, please submit them to me on the PROscorerTools GitHub page:

• Check out the PROscorerTools vignettes.

• For examples on how to use the scoreScale function within a more complex scoring function, check out the source code for some of the functions in the PROscorer package.