growthcleanr

R package for cleaning data from Electronic Health Record systems, focused on cleaning height and weight measurements.

This package implements the Daymont et al. algorithm, as specified in Supplemental File 3 within the Supplementary Material published with that paper.

Carrie Daymont, Michelle E Ross, A Russell Localio, Alexander G Fiks, Richard C Wasserman, Robert W Grundmeier, Automated identification of implausible values in growth data from pediatric electronic health records, Journal of the American Medical Informatics Association, Volume 24, Issue 6, November 2017, Pages 1080–1087, https://doi.org/10.1093/jamia/ocx037

This package also includes an R version of the SAS macro published by the CDC for calculating percentiles and Z-scores of pediatric growth observations and utilities for working with both functions. As of summer 2021, it also supports cleaning anthropometric measurements for adults up to age 65. The adult algorithm has not yet been published in a peer-reviewed publication, but is described in detail at Adult algorithm.

Bug and current fix

There is an error in the recentering tables in the current CRAN version that will causes the algorithm to inappropriately exclude birth observations in datasets with <5K observations. Until we are able to make changes on CRAN, we recommend using devtools to install growthcleanr from GitHub. We are working on updates as soon as possible.

Installation

To install the stable version from CRAN:

install.packages("growthcleanr")

Or from Github directly:

library(devtools)
install_github("carriedaymont/growthcleanr")

Summary

The growthcleanr package processes data prepared in a specific format to identify biologically implausible height and weight measurements. It bases these evaluations on techniques which use patient-specific longitudinal analysis and variations from published growth trajectory charts for pediatric subjects. These techniques are performed in a specific order which refines and improves results throughout the process.

Results from growthcleanr include a flag for each measurement indicating whether it is to be included or excluded based on plausibility, with a variety of specific types of exclusions identified distinctly. These flags can be analyzed further by researchers studying anthropometric EHR data to determine which measurements to include or exclude in their own studies. No values are deleted or otherwise removed; each is only flagged in a new column.

To start running growthcleanr, an R installation with a variety of additional packages is required, as is a growth measurement dataset prepared for use in growthcleanr.

The rest of this documentation includes:

Getting started:

• Quickstart, a brief tour of using growthcleanr, including data preparation
• Installation, options for installing growthcleanr, with notes on specific platforms and source-level installation for developers
• Usage, examples of cleaning data, multiple options, example data

Advanced topics:

• Configuration options, changing growthcleanr operational settings
• Understanding growthcleanr output, the exclusion types growthcleanr identifies
• Adult algorithm, a detailed description of how growthcleanr assesses observations from adult subjects
• Computing BMI percentiles and Z-scores, additional functions for common data transforms and determining percentiles and Z-scores using the CDC method
• Working with large datasets, notes and suggestions for running growthcleanr with large data sources
• Next steps, notes on potential enhancements to the pediatric and adult algorithms
• Developer guidelines, advice for contributors to this package, including a CRAN release checklist

Changes

For a detailed history of released versions, see the Changelog orNEWS.md. Tagged releases, starting with 1.2.3 in January 2021, are listed at GitHub.