Abstract
This project develops computationally efficient software to generate classification of pediatric complex chronic conditions using a free, open-source statistical environment.
Identification of children with complex chronic conditions (CCCs) is necessary to improve health care delivery and perform clinical research, because this patient population uses significant inpatient and outpatient medical resources.1 The original CCC classification was published in 2000.2 A second version was published in 2014 to reflect additions to the International Classification of Diseases system and the US adoption of the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision.3 The CCC classification is widely used in research (currently cited in more than 100 peer-reviewed journal publications). However, the current approach to assigning the CCC categories in health care–related data sets is limited by proprietary software and computational inefficiency. SAS and Stata software to assign CCC categories were published as appendices to the 2014 update,3 but not all investigators have access to these statistical packages. In addition, increasingly large data sets are available to investigators. Although the data processing capability of individual computers continues to improve, the SAS and Stata software can take significant time to run on data sets with millions of observations. The objective of this project was to develop computationally efficient software to generate the CCC categories using R, a free, open-source statistical environment.4 We then compared the SAS, Stata, and R software with respect to accuracy and speed of classification on a typical desktop system.
Methods
We developed the pccc R package based on the 2014 version 2 CCC system.3 To maximize computational efficiency, we leveraged the ability to call C++ from within R using the Rcpp package.5 We used standard software engineering practices, including distributed version control, issue tracking, and unit testing. We tested the pccc package using the same Healthcare Cost and Utilization Project data sets from the Agency for Healthcare Research and Quality used to develop the 2014 software (2009 Kids’ Inpatient Database [KID] and 2010 Nationwide Emergency Department Sample [NEDS]).6 On the same desktop system (i7 dual-core, 16-GB RAM), we classified each record using the SAS, Stata, and R software and compared the results. We tested the accuracy (percentage correctly classified) of the R software using SAS as the criterion standard. To test the relative speed of the 3 implementations, we compared processing time (in minutes) for the 3 407 146-record KID data set and the 28 584 301-record NEDS data set. The latest release of the R package is available on the Comprehensive R Archive Network (https://cran.r-project.org/web/packages/pccc/index.html), and the developmental version is on GitHub (https://github.com/CUD2V/pccc). Institutional review board approval was not required for this study using publicly available data sets.
Results
Unit testing of the new pccc package revealed several different types of issues present in the 2014 SAS and Stata software (Table 1). We collaborated with the authors of the 2000 (C.F.) and 2014 (J.A.F., C.F., and D.D.) CCC systems to resolve those issues. Subsequently, the R package and the updated SAS and Stata software yielded identical patient CCC categorizations when run on each row of patient data in the KID and NEDS data sets. Processing the same data, the R package was comparable to SAS and significantly more efficient than Stata (Table 2). The updated SAS and Stata software packages are available at https://feudtnerlab.research.chop.edu/ccc_version_2.php.
Table 1. Issues Revealed by the Unit Testing Process.
| Type of Issue by Specific Code Affected | Resolution |
|---|---|
| Duplicatesa | |
| Neuromuscular | |
| 343 | Duplicate deleted |
| G253 | Duplicate deleted |
| Technology dependence | |
| T84498A | Duplicate deleted |
| T86890 | Duplicate deleted |
| T86891 | Duplicate deleted |
| T86899 | Duplicate deleted |
| Transplantation | |
| T86890 | Duplicate deleted |
| T86891 | Duplicate deleted |
| T86899 | Duplicate deleted |
| Deletions and additionsb | |
| Neuromuscular | |
| 331 | Added |
| 3311 | Dropped from Stata only |
| 3318 | Dropped from Stata only |
| 35921 | Added |
| 35922 | Added |
| 35923 | Added |
| 35929 | Added |
| 9782 | Dropped from Stata only |
| E750 | Dropped, matched by E75 |
| E751 | Dropped, matched by E75 |
| E752 | Dropped, matched by E75 |
| E754 | Dropped, matched by E75 |
| G3189 | Dropped, matched by G318 |
| G3289 | Added in Stata only |
| G4735 | Added |
| G800 | Added |
| G804 | Added |
| G808 | Added |
| Q851 | Added in SAS only |
| Cardiovascular | |
| 4160 | Added |
| Q219 | Added in SAS only |
| Q258 | Added in SAS only |
| Q259 | Added in SAS only |
| Q268 | Added |
| T82121A | Added in Stata only; also flags technology dependence |
| Hematologic/immunologic | |
| D869 | Dropped, already matched by D86 |
| Metabolic | |
| D841 | Deleted |
| Respiratory | |
| 4160 | Added, previously only in Stata |
| 51630 | Added |
| 51637 | Added |
| Errorsc | |
| Respiratory | |
| 9620 | Changed to J9620 in SAS only |
| G4753 | Changed to G4735 in SAS only |
| Metabolic | |
| 2359 | Changed to 2539 |
| Substring errorsd | |
| Neuromuscular | |
| 359 | Now uses exact matching |
| 3592 | Now uses exact matching |
| G80 | Now uses exact matching |
| Respiratory | |
| 5163 | Now uses exact matching |
| Cardiovascular | |
| 416 | Now uses exact matching; previously in Stata only |
| Metabolic | |
| 624 | Now uses exact matching |
| Shift in categorizationse | |
| Cardiovascular and respiratory | |
| I43 | Now only in cardiovascular category |
| Hematologic/immunologic and metabolic | |
| D84 | Now only in heme/immunologic category |
| Metabolic | |
| E75 | Now in neuromuscular category |
Abbreviation: CCCs, complex chronic conditions.
Includes duplicate codes that were present in the SAS CCC version 2 software.
Includes codes that should (or should not) have been classified as CCCs.
Includes erroneous codes due to, for example, typos and keystroke errors.
Includes issues with codes where matching on a substring led to erroneous inclusion of more specific codes that did not correspond to a CCC.
Includes codes that were misclassified or erroneously included in ≥2 CCC categories.
Table 2. Processing Time by Software Type.
| Software | KID (N = 3 407 146) | NEDS (N = 28 584 301) |
|---|---|---|
| R | 4 min 48 s | 18 min 21 s |
| SAS | 3 min 1 s | 14 min 57 s |
| Stata | 22 min 45 s | 69 min 11 s |
Abbreviations: KID, Kids’ Inpatient Database; NEDS, Nationwide Emergency Department Sample.
Discussion
The free and open-source pccc R package provides accurate, efficient, and reproducible pediatric CCC categorization for large files of administrative records. The ability of R to call C++ directly can improve computational efficiency and is an advantage for package developers. Software development practices, including unit testing, can identify errors before code release. Code in the pccc package was developed collaboratively and that process, including issue tracking, is publicly visible in the GitHub repository. Suggestions or improvements can be submitted through GitHub’s pull request mechanism.
References
- 1.Cohen E, Berry JG, Camacho X, Anderson G, Wodchis W, Guttmann A. Patterns and costs of health care use of children with medical complexity. Pediatrics. 2012;130(6):e1463-e1470. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Feudtner C, Hays RM, Haynes G, Geyer JR, Neff JM, Koepsell TD. Deaths attributed to pediatric complex chronic conditions. Pediatrics. 2001;107(6):E99. [DOI] [PubMed] [Google Scholar]
- 3.Feudtner C, Feinstein JA, Zhong W, Hall M, Dai D. Pediatric complex chronic conditions classification system version 2. BMC Pediatr. 2014;14:199. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.The R Foundation for Statistical Computing The R Project for Statistical Computing. https://www.r-project.org/. 2017. Accessed January 10, 2018.
- 5.Rcpp.org Rcpp for Seamless R and C++ Integration. http://www.rcpp.org/. Accessed January 10, 2018.
- 6.Healthcare Cost and Utilization Project. Overview of Nationwide Emergency Department Sample (NEDS). http://www.hcup-us.ahrq.gov/nedsoverview.jsp. December 2017. Accessed January 10, 2018.