Updating International Classification of Diseases 9th Revision to 10th Revision of a Claims-Based Frailty Index

A claims-based frailty index (CFI) is useful to define frailty subgroups for confounding adjustment and subgroup analysis by frailty levels in claims-based observational studies. We have previously developed and validated a CFI using the 2006 Medicare Current Beneficiary Survey (1–3). This index estimates a deficit-accumulation frailty index using 93 claims-based variables derived from International Classification of Diseases, ninth revision (ICD-9) codes, Current Procedural Terminology codes, and Healthcare Common Procedure Coding System codes assessed over 12 months (1). As the ICD-10 system was adopted in the United States on October 1, 2015, it is necessary to adapt the CFI using ICD-10 codes. To this end, we redefined the original ICD-9 code-based variables using ICD-10 codes and examined the distribution and reliability of CFI before and after the ICD transition in three health care databases. We used the 2014 (ICD-9 system) and 2016 data (ICD-10 system) from two commercial insurers’ databases, Optum Clinformatics (N = 5,830,809) and IBM MarketScan (N = 3,619,937), and two disease-specific Medicare fee-for-service cohorts (N = 18,669,955). The study population included individuals 65 years or older who were continuously enrolled in the entire calendar year. Two Medicare samples were selected based on (a) diabetes, heart failure, or stroke (cardiometabolic disease cohort) and (b) rheumatoid arthritis, psoriatic arthritis, or gout (arthritis cohort) within each year. We applied the forward mapping using the General Equivalence Mapping (GEM) algorithm developed by the Centers for Medicare and Medicaid Services (4). We examined the prevalence of each claim-based variable in the 2014 and 2016 data within each dataset. When there were large discrepancies in the prevalence, three authors (NG, LB, and DHK) manually reviewed individual ICD-10 codes to ensure that ICD-9 and ICD-10 code-based variables are measuring similar constructs (SAS program available at https://doi.org/10.7910/DVN/HM8DOI). Finally, we removed duplicates (one ICD-9 code mapped to more than one ICD-10 codes). After iterative reviews and reconciliation, a CFI was created for the 2014 and 2016 data using the coefficients from the original CFI model (1). We examined the distribution, Cronbach’s alpha (5), and intraclass correlation between the two CFI versions within each database. Statistical analysis was performed using the SAS software version 9.4 and Aetion Evidence Platform version 4.3. The combined population from the three databases included over 28 million continuously enrolled patients. Commercial database populations were younger than the disease-specific Medicare populations (Table 1). The two commercial populations have similar mean and median values of CFI, which were lower than the corresponding values in the Medicare cohorts. The maximum CFI ranged from 0.48–0.49 (Medicare arthritis cohort) to 0.69–0.72 (Medicare cardiometabolic disease cohort). Within each population, the two versions of CFI showed similar distributions, internal consistency by a Cronbach’s alpha of 0.73–0.80, and concordance by intraclass correlation of 0.69–0.78 (Table 1). Distribution and Internal Consistency of ICD-9 Code-Based and ICD-10 Code-Based Claims-Based Frailty Indexa Notes: CFI = claims-based frailty index; ICD = International Classification of Diseases. aICD-9 code-based CFIs were calculated from the 2014 data and ICD-10 code-based CFIs were calculated from the 2016 data. bCronbach’s alpha between ICD-9 code-based CFI and ICD-10 code-based CFI was calculated from patients who had a 12-mo continuous enrollment in both years 2014 and 2016 in Optum database (n = 1,884,101), IBM MarketScan database (n = 1,375,534), Medicare cardiometabolic disease cohort (n = 4,292,698), and Medicare arthritis cohort (n = 1,980,837). Distribution and Internal Consistency of ICD-9 Code-Based and ICD-10 Code-Based Claims-Based Frailty Indexa Notes: CFI = claims-based frailty index; ICD = International Classification of Diseases. aICD-9 code-based CFIs were calculated from the 2014 data and ICD-10 code-based CFIs were calculated from the 2016 data. bCronbach’s alpha between ICD-9 code-based CFI and ICD-10 code-based CFI was calculated from patients who had a 12-mo continuous enrollment in both years 2014 and 2016 in Optum database (n = 1,884,101), IBM MarketScan database (n = 1,375,534), Medicare cardiometabolic disease cohort (n = 4,292,698), and Medicare arthritis cohort (n = 1,980,837). The strengths of our study include use of multiple databases and manual review of individual ICD-10 codes to generate ICD-9 to ICD-10 code mapping for CFI calculation. Limitations include use of disease-specific Medicare populations instead of a random Medicare sample, which may influence the prevalence of frailty, and lack of clinical frailty assessments to directly assess the validity of ICD-10 codes-based CFI. Nonetheless, our results show that the CFI, which was originally developed using ICD-9 codes, can be calculated using the ICD-10 codes mapped using the GEM algorithm and rigorous review process. Although health status of patients may have changed from 2014 to 2016, we found generally good internal consistency and concordance between the two CFI versions. This information is useful for claims-based studies of health outcomes in older adults by allowing and comparing frailty levels across the years of ICD-9 and ICD-10 systems. As the coding system evolves over time, regular updating and validation of CFI are necessary. This study was funded by R01AG062713 from the National Institute on Aging (NIA) to DHK. The funding sources had no role in the design, collection, analysis, or interpretation of the data, or the decision to submit the manuscript for publication. DHK contributed to conception and design. NG, LB, and RL conducted statistical analysis. NG and DHK drafted the manuscript. All authors interpreted data, critically revised the manuscript for important intellectual content, and read and approved the final manuscript for submission. DHK provides paid consultative services to Alosa Health, a nonprofit educational organization with no relationship to any drug or device manufacturers. The other authors declare no competing interests.