糖心vlog

Key Points

Question听 How does the updated version of the pediatric complex chronic condition (CCC) system (version 3 [V3]) compare with version 2 (V2) in capturing the range of clinical conditions represented in the International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM)?

Findings听 In this cross-sectional study of 7 186 019 hospitalizations within the Pediatric Health Information System from 2009 to 2019, V3 identified 38.3% of hospitalizations with a CCC compared with 40.1% identified by V2. The systems were similar in accounting for the number of CCC body-system categories per patient and in explaining variation in length of stay and mortality, and similar patterns were observed in the 2 999 420 hospitalizations from the Medicaid Merative MarketScan Research Databases.

Meaning听 The findings of this study suggest that the CCC V3 reflects the evolving ICD-10-CM system and should be used to identify CCCs.

Importance听 Since implementation of the International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) in the US, thousands of new or related codes have been added to represent clinical conditions. The widely used pediatric complex chronic condition (CCC) system required a major update from version 2 (V2) to version 3 (V3) to capture the range of clinical conditions represented in the ICD-10-CM.

Objective听 To update the CCC V3 system, creating V3, with new, missing, or retired codes; to reconceptualize the system鈥檚 use of technology codes; and to compare CCC V3 with V2.

Design, Setting, and Participants听 This repeated cross-sectional study examined US hospitalization data from the Pediatric Health Information System (PHIS) and the Medicaid Merative MarketScan Research Databases from January 1, 2009, to December 31, 2019, for all patients aged 0 to 18 years. Data were analyzed from March 1, 2023, to April 1, 2024.

Exposures听 The CCCs were identified in both data sources using the CCC V2 and V3 systems.

Main Outcomes and Measures听 The (1) percentage of pediatric hospitalizations associated with a CCC, (2) numbers of CCC body-system categories per patient, and (3) explanatory power for hospital length of stay and in-hospital mortality were compared over time for V3 vs V2.

Results听 Among 7 186 019 hospitalizations within PHIS, 54.3% patients were male, the median age was 4 years (IQR, 1-11 years), and 51.2% were aged 0 to 4 years). The CCC V2 identified 2鈥�878鈥�476 (40.1%) patients as having any CCC compared with 2鈥�753鈥�412 (38.3%) identified by V3. In addition, V2 identified 100鈥�065 (1.4%) patients with transplant status compared with 146鈥�683 (2.0%) by V3, and V2 identified 914鈥�835 (12.7%) as having technology codes compared with 805鈥�585 (11.2%) by V3. The 2 systems were similar in accounting for the number of CCC body-system categories per patient and in explaining variation in hospital length of stay and in-hospital mortality. For both V2 and V3, 10.0% of the variance in hospital length of stay and 12.0% of the variance in in-hospital mortality was explained by the presence of a CCC. Similar patterns were observed when analyzing the 2 999 420 Medicaid Merative MarketScan Research Databases鈥� hospitalizations (52.3% of patients were male, the median age was 1 year [IQR, 0-12 years], and 62.0% were 0 to 4 years old), except that the percentages of identified CCCs were all lower: V2 identified 758鈥�110 hospitalizations (25.3%) with any CCC compared with 718鈥�100 (23.9%) identified by V3.

Conclusions and Relevance听 These results suggest that, moving forward, V3 should be used to identify CCCs, and ongoing, frequent updates to V3, using a transparent, structured process, will enable V3 to accurately reflect the evolving spectrum of clinical conditions represented in the ICD-10-CM.

In the US, the International Classification of Diseases (ICD) system with clinical modifications (ICD-CM) is used to code and classify diagnoses, symptoms, and procedures associated with health care utilization.^1,2 The pediatric complex chronic condition (CCC) classification system uses ICD codes to determine the presence of 1 or more CCCs (10 body-system categories) and includes posttransplant and technology variables to signify additional complexity (Box).³ The open-source code CCC system is the most widely used classification system to study patients younger than 18 years with medical complexities. The CCC system has evolved alongside the ICD system and must account for updates in the underlying ICD codes: CCC version 1 (V1)⁴ corresponded to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and version 2 (V2) to the transition to the International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM).³ Since implementation of the ICD-10-CM in the US in 2015, however, thousands of new or related codes have been added to represent clinical conditions.⁵ The CCC system thus required a major update to capture the full spectrum of clinical conditions represented in the ICD-10-CM.

Box Section Ref ID

For this update from CCC V2 to V3, we sought to address 4 additional important issues. First, we needed to update the CCC V3 system with new, missing, or retired codes to reflect the evolving ICD-10-CM system. A substantial amount of manual review was required to transition the CCC system from V1 to V2 because of the fundamentally different code formats and structures of the ICD-9, ICD-9-CM, ICD-10, and ICD-10-CM. More recently, the Centers for Medicare & Medicaid Services (CMS) began to regularly release digital ICD-10-CM update lists and comprehensive mapping files, which raised the possibility of using a semiautomated review process to identify new or missing codes.⁵ Second, because implanted medical technological support (hereinafter technology) is becoming more common and may not indicate a level of complexity consistent with the CCC operational definition, technology codes alone may not reliably identify CCC status (eg, the presence of a gastrostomy tube in a patient younger than 18 years with nutritional concerns but no additional medical complexity). We thus disaggregated CCC status from the presence of technology. A positive CCC status now depends on CCC diagnosis or procedure codes, while a separate technology variable indicates additional medical complexity. Third, we aimed to improve the transparency of decisions to include or exclude codes by creating open-source documentation of all CCC code determinations and changes. Finally, to address whether the change from V2 to V3 would alter the study results, we needed to compare the 2 versions.

In this study, we aimed to (1) describe the process used to update the CCC V3 system with new, missing, or retired codes (2) describe a new strategy of handling technology codes, and (3) compare the 2 versions using 2 inpatient pediatric administrative datasets.

This cross-sectional study was deemed exempt from review by the Colorado Multiple Institutional Review Board because it did not constitute human participants research and used deidentified data; the requirement for informed consent was therefore waived. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology () reporting guideline.⁶

We used a comprehensive, supervised semiautomated strategy to identify new, missing, or retired ICD-10-CM codes. First, we remapped the previously published CCC V2 ICD-9-CM and ICD-10-CM codes using the 2018 General Equivalence Mappings electronic mapping tool from the CMS.⁵ All CCC V2 ICD-9-CM codes were forward remapped, and all CCC V2 ICD-10 codes were remapped backward using the General Equivalence Mappings, which allowed for identification of new or missing codes. Second, we manually compared the remapped codes with annual update lists of new ICD-10-CM codes for 2019 to 2023 using the hierarchical ICD-10 code structure to identify related codes that might represent a CCC.⁵ Third, 2 complex care pediatricians (J.A.F. and C.F.) independently reviewed all identified candidate codes for inclusion or exclusion in the CCC V3 system based on the CCC definition used to create the previous versions: 鈥淎ny medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or 1 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.鈥�^3,4 Discrepancies between reviewers were resolved by joint review of the code or codes in question. Fourth, the final list of CCC V3 codes was created, reviewed for quality, and finalized for use. To enable transparency of code determinations, we created a list of all codes reviewed for inclusion or exclusion for CCC V3 and their determinations, including retired codes that were used in previous CCC versions (eTable 1 in Supplement 1, eTable 2 in Supplement 2, and eTable 3 in Supplement 3).

In CCC V2, certain transplant codes and technology codes were used both to identify CCCs within the 10 CCC categories and to indicate additional medical complexity using transplant and technology variables. Whereas V3 implements CCC transplant codes identically to the V2 system, V3 implements CCC technology codes differently than the V2 system. This is because, unlike the transplant codes (which always signify end-organ failure corresponding to a CCC category), many new technology codes do not indicate body-system dysfunction and do not necessarily signify medical complexity. Accordingly, for V3, we separated CCC diagnosis and procedure codes from all CCC technology codes.

Because of the changes to the handling of CCC technology codes, we reconceptualized how CCC status (ie, yes or no) and CCC category were assigned using a multistage process (Figure 1). The major difference between this CCC assignment strategy and prior versions is that in V3, to be classified as having a CCC, the patient (or encounter) must have a positive CCC status based only on the CCC diagnosis or procedure codes (and not based only on CCC technology codes). First, we assessed whether the patient had at least 1 CCC diagnosis or procedure code. If yes, the patient was classified as having a positive CCC status. Next, for patients with a positive CCC status, we assessed for the presence of technology using the technology codes. Finally, we identified the CCC categories that the patient had using CCC diagnostic, procedure, and technology codes.

We compared the CCC V2 and V3 systems using encounter-level data from the Pediatric Health Information System (PHIS) of the Children鈥檚 Hospital Association and the Medicaid Merative MarketScan Research Databases (hereinafter referred to as Medicaid) between January 1, 2009, to December 31, 2019. Both contain ICD-9-CM and ICD-10-CM diagnosis and procedure codes necessary for the analysis; PHIS includes up to 41 diagnosis and 41 procedure codes per encounter, and Medicaid includes up to 15 diagnosis and 15 procedure codes per encounter. All inpatient and observation encounters for patients 0 to 18 years old were included, except for 鈥渘ormal newborns鈥� (MS-DRG 795) and patients who were pregnant (MS-MDC 14), according to the ICD-10-CM codes. Over the 11-year period, we compared (1) the percentage of patients identified with 1 or more body-system CCCs; (2) the percentage of patients by numbers of body-system CCCs per patient; and (3) the explanatory power of the CCC systems, as measured by the R² value for length of inpatient stay and the pseudo-R² value for in-hospital mortality. Patient race and ethnicity categories included Hispanic, non-Hispanic Black, non-Hispanic White, and other (including American Indian, Asian, Native Hawaiian, multiracial, other race, and missing). To enable comparisons between CCC V3 and previously published studies using the CCC system, race and ethnicity, as recorded in the PHIS and Medicaid databases, were reported.

All CCC lookup tables were created as Microsoft Excel comma-separated value spreadsheets to enable compatibility with an end user鈥檚 preferred database or statistical software. These lookup tables are open-source and available from the Children鈥檚 Hospital Association.⁷ Descriptive statistics and temporal graphs were used to compare the CCC V2 with the V3 systems. To assess relative changes for each CCC category from V2 to V3, percentage change was calculated using the following formula: 100鈥壝椻��(number of V3 CCC hospitalizations鈥夆垝鈥塶umber of V2 CCC hospitalizations)/number of V2 CCC hospitalizations. The McNemar test was used to compare differences in percentages between V2 and V3 by CCC category. Annual linear regressions for V2 and V3 were fit with the presence of a CCC as the only covariate and log-transformed length of stay (due to nonnormality) in days as the outcome. The R² value and the Akaike information criterion (AIC) were used as the measures of fit for these models. Second, annual logistic regressions for V2 and V3 were fit with the presence of a CCC as the only covariate and mortality as the outcome. The McFadden pseudo-R² and the AIC were used as the measures of fit. Data were analyzed from March 1, 2023, to April 1, 2024. All analyses were conducted using SAS, version 9.4 (SAS Institute Inc), and significance was set at a 2-tailed P鈥�&濒迟;鈥�.05.

Overall, PHIS contained 7鈥�186鈥�019 hospitalizations (45.7% of patients were female, and 54.3% were male; the median age was 4 years [IQR, 1-11 years]), and Medicaid contained 2鈥�999鈥�420 hospitalizations (47.7% of patients were female, and 52.3% were male; the median age was 1 year [IQR, 0-12 years]). In PHIS, 51.2% of patients were between 0 and 4 years old and in Medicaid, 62.0% were included in this age group (the age range for both databases was 0 to 18 years). Of the total patients from the PHIS database, 19.4% were Hispanic, 19.0% were non-Hispanic Black, 49.0% were non-Hispanic White, and 12.5% were in the other race and ethnicity category. Among those patients from the Medicaid database, 6.1% were Hispanic, 28.5% were non-Hispanic Black, 43.3% were non-Hispanic White, and 22.1% were in the other race and ethnicity category. Additional demographics for both datasets are listed in eTable 4 in Supplement 1.

The percentages of hospitalizations with CCCs classified by V2 and V3 are listed in the Table. The CCC V2 identified 2鈥�878鈥�476 (40.1%) of the 7鈥�186鈥�019 encounters within PHIS as having any CCC compared with 2鈥�753鈥�412 (38.3%) identified by V3 (鈭�4.3% change). Similarly, V2 identified 100鈥�065 (1.4%) patients with transplant status compared with 146鈥�683 (2.0%) by V3 (46.6% change). Finally, V2 identified 914鈥�835 patients (12.7%) as having technology codes compared with 805鈥�585 (11.2%) by V3 (鈭�11.9% change). The Table contains the overall V2 results by CCC category and reports the V3 results based only on the presence of a CCC diagnosis or procedure code, excluding technology codes. The CCC categories (listed verbatim in alphabetical order) include any CCC, cardiovascular disease, congenital or genetic, gastrointestinal, hematologic or immunologic, malignancy, metabolic, neonatal, neuromuscular, renal, respiratory, tech dependence, and transplant. Comparing these data, several CCC categories remained stable (eg, malignancy: V2, 504鈥�072 [7.0%] vs V3, 507鈥�283 [7.1%]), while others demonstrated larger differences (eg, gastrointestinal: V2, 741鈥�298 [10.3%] vs V3, 256鈥�313 [3.6%]). The Table also reports V3 technology codes by CCC category among patients with a positive CCC status. For example, among any of the 7鈥�186鈥�019 total encounters, 422鈥�072 (5.9%) had a positive CCC status and technology codes attributed to the gastrointestinal category. Finally, the Table reports the overall V3 results by CCC category for patients with a positive CCC status and technology codes present for that category. For example, 256鈥�313 (3.6%) had either a gastrointestinal CCC diagnostic or procedure code, 422鈥�072 (5.7%) had a gastrointestinal technology code, and a total of 678鈥�385 (9.4%) were categorized by V3 as having a positive CCC status and gastrointestinal technology codes. Similar patterns were observed when analyzing Medicaid data (eTable 5 in Supplement 1), except that the percentages of identified CCCs were all lower in magnitude (V2 identified 758鈥�113 [25.3%] hospitalizations with any CCC compared with 718鈥�100 [23.9%] identified by V3).

Figure 2 displays the percentage of hospitalizations within the PHIS database with any CCC, 1 CCC, and multiple CCCs over time, as classified by either V2 or V3. The CCC V2 consistently identified a higher percentage of patients with a CCC compared with V3, although the difference was smaller after 2015. A subanalysis of the most recent year of PHIS data (eFigure 1 in Supplement 1) demonstrates how ICD-10 code additions, deletions, and the reconceptualization of technology codes affected the observed net difference between V2 and V3. For both V2 and V3, the percentage of patients with 1 CCC decreased over time, although the percentages were consistently higher in V2 compared with V3. The latter was due to removal of nonspecific codes (eg, scoliosis, unspecified, other problems with newborn, and gastrostomy status) that were not alone indicative of a CCC. For example, removal of the diagnosis code Z931 (for gastrostomy status) in V3 resulted in 38鈥�499 fewer CCC cases in a single year of data. Of those, 34鈥�649 (90.0%) cases were still identified as having a CCC by a different V3 diagnosis code, resulting in 3850 cases no longer classified with a CCC. For both V2 and V3, the percentage of patients with multiple CCCs increased over time, and the percentage of hospitalizations with 3 CCCs and 4 or more CCCs was higher for V3 compared with V2 after 2015 (when the CMS stopped updating the ICD-9-CM and started releasing additional new ICD-10-CM codes that were only available for inclusion in V3 but not in V2). Similar patterns were observed when analyzing Medicaid data (eFigure 2 in Supplement 1).

For both V2 and V3, 10.0% of the variance in hospital length of stay within the PHIS database was explained by the presence of a CCC (Figure 3A). The R² value for length of stay remained stable over time, although the values were consistently higher for V2 than V3. For both V2 and V3, 12.0% of the variance in in-hospital mortality was explained by the presence of a CCC (Figure 3B). The pseudo-R² value for mortality increased slightly over time for both V2 and V3. The values were higher between 2009 and 2015 for V3 than for V2 and then converged after 2015. eTable 6 in Supplement 1 reports the AIC fit statistics for both models. Similar patterns were observed in Medicaid data (eFigure 3 in Supplement 1).

In this repeated cross-sectional study, CCC V2 identified 40.1% of the 7 186 019 patients within PHIS as having any CCC compared with 38.3% of the 7 186 019 patients identified by V3. Similarly, V2 identified 1.4% of patients with transplant status compared with 2.0% by V3, and V2 identified 12.7% of patients as having technology codes compared with 11.2% by V3. Over the 11-year study period, V2 consistently identified a higher percentage of patients with a CCC compared with V3, although the difference was smaller after鈥塮ull implementation of the ICD-10-CM in 2015. The 2 systems were comparable in accounting for the number of CCC body-system categories per patient and in explaining variation in inpatient length of stay and in-hospital mortality. Similar patterns were observed when analyzing Medicaid data, except that the percentages of identified CCCs were lower in magnitude (eg, 38.3% of patients in PHIS had a CCC using V3 vs 23.9% in Medicaid). Overall, CCC V3 and V2 were comparable across all outcome measures, and the updated CCC V3 system should be used to identify CCCs.

Minor differences between the 2 systems raise questions that deserve discussion. Why did V2 consistently identify a higher percentage of patients with CCCs than did V3? Why did this gap shrink after 2015? To address the first question, we must consider code deletions. In creating V3, we removed ICD-10-CM codes present in V2 that, when used alone, are not indicative of CCCs, such as scoliosis, unspecified (code M419), or other problems with newborn (code P84). The removal of these common nonspecific ICD-10-CM codes and corresponding pre-2015 ICD-9 codes from V3 explains the sustained differences over time, including the pattern of V2 identifying a higher percentage of patients with 1 CCC than that of V3. For the second question, we must consider code additions. After 2015, the CMS stopped updating the ICD-9-CM and started releasing additional new ICD-10-CM codes that were available for inclusion in V3 (but not V2). For example, many new ICD-10-CM codes corresponding to stem cell transplants and organ transplant complications were added after 2015. However, many of these codes could not be reverse mapped to corresponding pre-2015 ICD-9 codes, meaning that identification of patients with these CCC codes could only happen after 2015. These types of code additions likely explain the relative increase of patients with a CCC identified by V3 after 2015 by the increases in patients with multiple CCCs after 2015. The net result, therefore, of the transition from V2 to V3 is that slightly fewer patients younger than 18 years with only 1 CCC or only a technology-dependent CCC in V2 were classified as having a CCC in V3, whereas those with more than 1 CCC were present in nearly identical percentages.

These small differences notwithstanding, 2 aspects of the CCC V3 system warrant highlighting. First, end users must consider how to implement the CCC V3 system to meet their analytic needs in the context of the changes to V3. Some body-system categories may perform differently due to relocations of diagnoses within the CCC system. For example, the increase in neuromuscular CCCs was mostly attributable to the movement of cerebral infarction codes (433 and I63) from the cardiovascular to the neuromuscular CCC type. Similarly, the decrease in congenital or genetic CCCs was due to greater specificity of the included CCC codes and removal of nonspecific codes like constitutional short stature (E34.31). End users must also be aware that technology codes no longer are used to assign CCC status; instead, positive CCC status is based on the presence of CCC diagnosis or procedure codes only. Due to these changes in how technology codes are handled, some differences are immediately apparent, like the shift in the gastrointestinal CCC category by uncoupling CCC status from the presence of a gastrostomy tube.

Second, the CCC V3 system will require ongoing curation and maintenance, as ICD codes have been added, modified, or deleted from the overall ICD-10-CM system, and end users provided feedback after implementing CCC V3 for use in clinical and research settings. With this update, we moved from a CCC code identification process based entirely on manual identification and review to a transparent semiautomated identification and review process. With future updates, we anticipate moving toward a more automated process that leverages objective data-driven thresholds for identifying and including CCC codes. This process may be further enabled by the World Health Organization鈥檚 rollout of the ICD-11, which is a fully digital ICD system with linkages to medical terminology systems that are already highly integrated with clinical electronic health record systems.²

Our study must be considered in the context of 2 limitations. First, while we compared the systems using 2 common pediatric data sources (PHIS and Medicaid), we only analyzed inpatient hospitalization data. The CCC system could perform differently when applied to other data sources representing other settings of care (eg, Medicaid claims data for outpatient or emergency care settings) and depending on the number of diagnostic and procedure codes available per encounter. Second, if history is a guide, a limited number of specific codes will either have been included in or excluded from the CCC system by what will come to be viewed as errors in judgment. Moving toward a transparent system of informing users about code inclusion and exclusion decisions and incorporation of subsequent end user feedback will be extremely important to continued definitional calibration and improvement of the CCC system.

In this cross-sectional study using 2 large databases of hospitalizations to compare CCC V2 with CCC V3, within PHIS, CCC V3 identified 38.3% as having any CCC vs with 40.1% by V2. The updated V3 system reflects the evolving ICD-10-CM system, and V3 should be used to identify CCCs. Ongoing frequent updates to V3 using a transparent structured process will enable V3 to accurately reflect the evolving spectrum of clinical conditions represented by the ICD system.

Accepted for Publication: April 25, 2024.

Published: July 15, 2024. doi:10.1001/jamanetworkopen.2024.20579

Open Access: This is an open access article distributed under the terms of the CC-BY License. 漏 2024 Feinstein JA et al. 糖心vlog Open.

Corresponding Author: James A. Feinstein, MD, MPH, ACCORDS, University of Colorado Anschutz Medical Campus, 1890 N Revere Ct, Ste 300, Aurora, CO 80045 (james.feinstein@cuanschutz.edu).

Author Contributions: Dr Hall and Ms Davidson had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Feinstein, Hall, Feudtner.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Feinstein.

Critical review of the manuscript for important intellectual content: All authors.

Statistical analysis: Feinstein, Hall.

Obtained funding: Feinstein.

Administrative, technical, or material support: Feinstein, Davidson.

Supervision: Feudtner.

Conflict of Interest Disclosures: None reported.

Funding/Support: Dr Feinstein was supported by grant K23HD091295 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health (NIH).

Role of the Funder/Sponsor: The NIH had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The content is solely the responsibility of the authors and does not represent the official views of the NIH or the US government.

Data Sharing Statement: See Supplement 4.