Abstract
Background
The Veterans Affairs Frailty Index (VA-FI) is an electronic frailty index developed to measure frailty using administrative claims and electronic health records data in Veterans. An update to ICD-10 coding is needed to enable contemporary measurement of frailty.
Method
International Classification of Diseases, ninth revision (ICD-9) codes from the original VA-FI were mapped to ICD-10 first using the Centers for Medicaid and Medicare Services (CMS) General Equivalence Mappings. The resulting ICD-10 codes were reviewed by 2 geriatricians. Using a national cohort of Veterans aged 65 years and older, the prevalence of deficits contributing to the VA-FI and associations between the VA-FI and mortality over years 2012–2018 were examined.
Results
The updated VA-FI-10 includes 6422 codes representing 31 health deficits. Annual cohorts defined on October 1 of each year included 2 266 191 to 2 428 115 Veterans, for which the mean age was 76 years, 97%–98% were male, 78%–79% were White, and the mean VA-FI was 0.20–0.22. The VA-FI-10 deficits showed stability before and after the transition to ICD-10 in 2015, and maintained strong associations with mortality. Patients classified as frail (VA-FI > 0.2) consistently had a hazard of death more than 2 times higher than nonfrail patients (VA-FI ≤ 0.1). Distributions of frailty and associations with mortality varied with and without linkage to CMS data and with different assessment periods for capturing deficits.
Conclusions
The updated VA-FI-10 maintains content validity, stability, and predictive validity for mortality in a contemporary cohort of Veterans aged 65 years and older, and may be applied to ICD-9 and ICD-10 claims data to measure frailty.
Keywords: Claims-based frailty index, Electronic frailty index, Epidemiology, Frailty, ICD 9–ICD-10 transition
Electronic frailty indices (eFIs) estimate frailty in populations from readily available data in electronic health records and administrative databases (1,2). In the United States, the Veterans Health Administration (VHA) is an integrated national health care system where claims and health care data are captured and stored in a centralized database (3). In 2019, the Veterans Affairs Frailty Index (VA-FI) was developed using the International Classification of Diseases, ninth revision (ICD-9) diagnostic and procedural, Current Procedural Terminology (CPT), and Healthcare Common Procedure Coding System (HCPCS) (4). The VA-FI identifies Veterans at increased risk of mortality—independent of age and other predictors—and has been used as a predictor, confounder, and effect modifier (5–8).
In 2015, the United States transitioned from ICD-9 to ICD-10, requiring the VA-FI to be updated for contemporary use. Although the Centers for Medicare and Medicaid Services (CMS) provide general equivalence mappings (GEMs) to assist with conversion to ICD-10, these are not always accurate, given the increased granularity of the ICD-10 codebook (9). Moreover, national policy implemented in 2014 allows Veterans to receive care either within or outside VHA (10). It is largely unknown how this outside care utilization may impact the electronic measurement of frailty using the VA-FI. Finally, the original VA-FI used a 3-year assessment. Although other studies have shown that varying the length of assessment periods may change the prevalence and predictive performance of individual comorbidities (11,12), it is unclear whether varying assessment periods may impact an eFI such as the VA-FI.
The objectives of this study were to (i) update the VA-FI (VA-FI-9) using ICD-10 codes (VA-FI-10), while maintaining content validity in measuring claims representative of frailty and its clinical domains; (ii) assess whether VA-FI-10 maintains stability in measuring prevalence of frailty after the transition to ICD-10; and (iii) examine whether the VA-FI-10 maintains predictive validity in its association with mortality. We further sought to compare the distributions of VA-FI-10 and its associations with mortality with and without linking VHA to CMS data (measuring health deficits managed outside VHA) and when varying assessment periods in which health deficits are measured.
Method
Translation and Update of the VA-FI
As previously detailed (4), the VA-FI-9 was defined using the deficit accumulation criteria of frailty as the number of age-related health deficits spanning multiple domains of older adult health status (eg, cognition and function) observed in an individual patient. The presence of each health deficit was identified based on the presence of associated ICD-9, CPT, or HCPCS codes (4,13). We translated and updated the VA-FI-9 in 3 stages:
An initial translation was obtained by applying mappings from the 2018 CMS GEMs and the CMS Chronic Conditions Warehouse (CCW) (14) to map ICD-9 codes for each health deficit in VA-FI-9 to ICD-10 codes.
The initial translation was validated by manual review of whether candidate ICD-10 codes were correctly classified to each health deficit.
The full ICD-10 hierarchy and list of active CPT codes in the VHA Corporate Data Warehouse (CDW) were then manually reviewed to check whether additional ICD, CPT, or HCPCS codes should be included in the VA-FI-10.
In Stage 1, we mapped ICD-9 diagnosis and procedure codes for each deficit to ICD-10 codes using both the GEMs backward and forward mappings, accounting for possible novel concepts introduced in ICD-10. Mappings flagged to be either “exact” or “approximate” in the GEMs were included for a comprehensive initial translation. The CCW is a CMS research database that includes variables for chronic or disabling conditions defined based on ICD-10 and CPT/HCPCS codes (14). We augmented the GEM translations by including additional ICD-10 codes for CCW chronic conditions that were considered related to each deficit.
In Stage 2, 2 geriatricians (C.D. and A.R.O.) independently reviewed the ICD-10 codes for each deficit translated from Stage 1 to verify whether each code correctly represented a frailty health deficit. The proportion of agreement and Cohen’s κ were calculated. Following individual review, discrepancies were reconciled to produce a harmonized set of mappings.
Stages 1 and 2 largely relied on mappings with ICD-9 codes. In Stage 3, the 2 geriatricians also reviewed the full ICD-10 hierarchy to ensure high level groupings of codes that do not directly map to ICD-9 codes were still captured if they represented a health deficit. Finally, as CPT and HCPCS codes are periodically updated, a list of active CPT and HCPCS codes used in CDW were reviewed to capture newly introduced codes that represented a health deficit.
Data Extraction
We extracted data on patient demographics, including age, sex, race, and associated VHA facility ID from structured data in CDW. The location of VHA station visited from recent outpatient records prior to the index date was used to identify the region of each patient. Smoking status was determined from the most recent clinical assessments of tobacco use prior to index date. Dates of mortality were extracted from VHA Vital Status Files and structured data in CDW. For each annual cohort, we extracted ICD diagnosis and procedure codes and CPT/HCPCS codes from tables in CDW. Health deficits managed outside VHA were captured from ICD and CPT/HCPCS codes from Medicare and Medicaid claims in CMS data. While data on patients’ administrative codes in CDW and Medicare claims were available through 2018, data on Medicaid claims were available only through 2014. Analyses involving CMS data included all codes that were available in the relevant time period.
Sample Selection
We evaluated the performance of VA-FI-10 over time among those born on or prior to December 31, 1953. We defined annual cohorts in which we recalculated the VA-FI-10 using a assessment period of 3 years prior to October 1 of that year. These annual cohorts allowed evaluation of VA-FI-10 consistency using codes before and after the transition to ICD-10 on October 1, 2015 and track its associations with outcomes across years. Each annual cohort was defined to include patients from the population who were alive by September 30 of that year (index date), were aged 65 years and older by index date, and had ≥1 outpatient visit for routine clinical care in the year prior. Patients were followed until mortality date or 1 month prior to the extraction of the mortality data, to account for lag in mortality reporting.
Evaluation of VA-FI-10
The presence of a health deficit in each patient of each annual cohort was defined by whether the patient received an associated diagnosis or procedure code in the 3 years prior to the index date, as previously defined (4). The final VA-FI was calculated as the sum of the number of deficits incurred divided by the total number of a possible 31 deficits (13). The prevalence rates of individual deficits over time were estimated based on the proportion of patients who had the deficit in each annual cohort. We repeated analyses similar to that of the validation for VA-FI-9, for which we estimated—in each of annual cohort—the survival curves from the index date by frailty category and adjusted associations between the VA-FI-10 and time to mortality. We fit Cox proportional models to estimate the hazard ratios (HRs) for VA-FI categories, adjusting for age, sex, race, smoking status, and region. Based on cut-points validated in our and others’ work (4,15–18,19), the VA-FI categories were nonfrail (VA-FI ≤ 0.1), prefrail (>0.1–0.2), mildly frail (>0.2–0.3), moderately frail (>0.3–0.4), and severely frail (>0.4). Beyond this base set of analyses, we repeated them in sensitivity analyses when calculating the VA-FI using administrative codes from CDW only. We also conducted analyses in which we varied the assessment period in which the VA-FI was calculated, for 1 and 3 years prior to the index date, using codes from both the CDW and CMS. Paired t tests were used to test for differences in the mean VA-FI for patients in the 2018 cohort when it is defined by the varying assessment periods and data sources.
Ethics
This study was approved, and the requirement for obtaining patient informed consent was waived, by the VHA Boston Institutional Review Board.
Results
Updated VA-FI: VA-FI-10
Prior to the update, the VA-FI included 714 unique codes (301 ICD-9 diagnosis, 18 ICD-9 procedure, and 395 CPT codes). Following the translation and update process, we identified 5593 additional codes (302 ICD-9 diagnosis, 18 ICD-9 procedure, 5163 ICD-10 diagnosis, 391 ICD-10 procedure, and 548 CPT codes; see Supplementary Data). Verifications of the initial translation using the GEMs and CCW mappings between the 2 reviewers had a proportion agreement of 98.5% (Cohen’s κ = 0.91).
Patient Characteristics of Cohorts
The annual cohorts included between 2 266 191 and 2 428 115 Veterans who were aged 65 years and older and alive as of October 1 of each year (Table 1). Mean age was 75.5–76.4 years, and majority were male (97.0%–97.8%) and White (77.9%–79.4%), with about 10% Black (8.4%–11.8%). There was fairly even representation of patients across different regions, with slightly more patients from the Midwest, North Atlantic, and Southeast regions. Nearly all had received CMS services. Most Veterans were current or former smokers (78.3%–78.9%). Mean VA-FI ranged from 0.20 to 0.22. The majority (77.3%–81.5%) were classified as either nonfrail, prefrail, or mildly frail.
Table 1.
Patient Characteristics for Each Annual Cohort
| Annual Cohort | |||||||
|---|---|---|---|---|---|---|---|
| Patient Characteristic | 2012 | 2013 | 2014 | 2015 | 2016 | 2017 | 2018 |
| Size, n | 2 266 191 | 2 340 247 | 2 408 477 | 2 428 115 | 2 425 350 | 2 395 838 | 2 359 207 |
| Age, mean (SD) | 76.40 (8.35) | 75.92 (8.52) | 75.61 (8.58) | 75.52 (8.54) | 75.58 (8.48) | 75.67 (8.39) | 75.79 (8.28) |
| Sex, % | |||||||
| Male | 2 215 272 (97.8) | 2 288 134 (97.8) | 2 354 486 (97.8) | 2 371 190 (97.7) | 2 364 608 (97.5) | 2 330 701 (97.3) | 2 288 687 (97.0) |
| Female | 50 845 (2.2) | 52 055 (2.2) | 53 941 (2.2) | 56 838 (2.3) | 60 654 (2.5) | 65 037 (2.7) | 70 392 (3.0) |
| Unknown | 74 (0.0) | 58 (0.0) | 50 (0.0) | 87 (0.0) | 88 (0.0) | 100 (0.0) | 128 (0.0) |
| Race, % | |||||||
| Black | 190 665 (8.4) | 210 008 (9.0) | 229 071 (9.5) | 243 011 (10.0) | 255 239 (10.5) | 265 852 (11.1) | 277 238 (11.8) |
| White | 1 766 026 (77.9) | 1 841 310 (78.7) | 1 907 579 (79.2) | 1 928 219 (79.4) | 1 924 319 (79.3) | 1 894 816 (79.1) | 1 855 171 (78.6) |
| Other | 46 077 (2.0) | 49 458 (2.1) | 52 596 (2.2) | 54 591 (2.2) | 55 941 (2.3) | 56 786 (2.4) | 57 313 (2.4) |
| Unknown | 263 423 (11.6) | 239 471 (10.2) | 2 19 231 (9.1) | 202 294 (8.3) | 189 851 (7.8) | 178 384 (7.4) | 169 485 (7.2) |
| Region, % | |||||||
| Continental | 358 555 (15.8) | 371 712 (15.9) | 384 536 (16.0) | 389 642 (16.0) | 390 936 (16.1) | 389 864 (16.3) | 387 827 (16.4) |
| Midwest | 535 299 (23.6) | 552 962 (23.6) | 569 977 (23.7) | 575 798 (23.7) | 573 554 (23.6) | 564 862 (23.6) | 553 160 (23.4) |
| North Atlantic | 556 880 (24.6) | 566 927 (24.2) | 573 833 (23.8) | 570 545 (23.5) | 564 311 (23.3) | 551 721 (23.0) | 537 693 (22.8) |
| Pacific | 363 401 (16.0) | 381 175 (16.3) | 397 249 (16.5) | 403 503 (16.6) | 406 695 (16.8) | 403 103 (16.8) | 399 604 (16.9) |
| Southeast | 452 056 (19.9) | 467 471 (20.0) | 482 882 (20.0) | 488 627 (20.1) | 489 854 (20.2) | 486 288 (20.3) | 480 923 (20.4) |
| Smoking, % | |||||||
| Current or former | 1 773 711 (78.3) | 1 838 509 (78.6) | 1 897 481 (78.8) | 1 915 106 (78.9) | 1 911 356 (78.8) | 1 883 647 (78.6) | 1 846 359 (78.3) |
| Never | 401 422 (17.7) | 415 065 (17.7) | 428 570 (17.8) | 434 310 (17.9) | 438 107 (18.1) | 438 934 (18.3) | 439 007 (18.6) |
| Unknown | 91 058 (4.0) | 86 673 (3.7) | 82 426 (3.4) | 78 699 (3.2) | 75 887 (3.1) | 73 257 (3.1) | 73 841 (3.1) |
| VA-FI category (%) | |||||||
| Nonfrail | 605 860 (26.7) | 633 037 (27.1) | 647 959 (26.9) | 635 441 (26.2) | 580 567 (23.9) | 562 507 (23.5) | 584 405 (24.8) |
| Prefrail | 756 056 (33.4) | 790 365 (33.8) | 814 839 (33.8) | 822 800 (33.9) | 793 631 (32.7) | 770 760 (32.2) | 754 754 (32.0) |
| Mildly frail | 474 277 (20.9) | 484 158 (20.7) | 498 390 (20.7) | 507 485 (20.9) | 524 784 (21.6) | 520 150 (21.7) | 495 700 (21.0) |
| Moderately frail | 248 857 (11.0) | 249 767 (10.7) | 255 664 (10.6) | 262 672 (10.8) | 287 193 (11.8) | 289 244 (12.1) | 276 684 (11.7) |
| Severely frail | 181 141 (8.0) | 182 920 (7.8) | 191 625 (8.0) | 199 717 (8.2) | 239 175 (9.9) | 253 177 (10.6) | 247 664 (10.5) |
| VA-FI | |||||||
| Mean (SD) | 0.20 (0.13) | 0.20 (0.13) | 0.20 (0.13) | 0.20 (0.13) | 0.21 (0.13) | 0.22 (0.14) | 0.22 (0.14) |
| ≥.5 (%) | 63 390 (2.8) | 65 570 (2.8) | 69 901 (2.9) | 74 095 (3.1) | 94 402 (3.9) | 10 3371 (4.3) | 102 700 (4.4) |
| ≥.7 (%) | 2710 (0.1) | 2964 (0.1) | 3291 (0.1) | 3640 (0.1) | 5680 (0.2) | 6815 (0.3) | 6938 (0.3) |
Notes: VA-FI = Veterans Affairs Frailty Index. Patient characteristics of each annual cohort. Age denotes age of patients by the index date (10/1) of each year. The location of the VHA station visited from recent outpatient records prior to the index date was used to identify the region of each patient. Smoking status was determined from the most recent provider assessments of tobacco use prior to the index date as recorded in clinical reminders in the VHA electronic health records.
Prevalence of Health Deficits
The prevalence of individual deficits stratified by domain is presented in Figure 1 (numeric rates in Supplementary Table S1). For most deficits, the prevalence was stable over the years before and after the transition from ICD-9 to ICD-10 in 2015. After the transition, there were increases in the prevalence of atrial fibrillation (12.3% in 2015 to 20.7% in 2016) and peripheral neuropathy (9.3%–16.0%), cancer (23.3%–27.6%), and durable medical equipment (15.1%–18.3%). There was a small decrease in the prevalence of vision comorbidity (27.7%–26.3%). The prevalence of deficits using a 1-year assessment period or only CDW data was similar (Supplementary Table S1 and Supplementary Figure S3). Figure 2 displays the distribution of VA-FI-10 in 2018, using both 1- and 3-year assessment periods and when including or excluding CMS codes. There is a greater impact on the distribution when excluding CMS codes than when using a shorter assessment period for defining the deficits (p value <.05 for all differences in mean VA-FI measured from varying data sources and assessment periods; Supplementary Table S2).
Figure 1.
Prevalence of VA-FI deficits defined by the index date of year, using 3-y assessment periods. Afib = atrial fibrillation; CAD = coronary artery disease; CVA = cerebral vascular accident; Depress = depression; DuraMed = durable medical equipment; FtoThrive = failure to thrive; GaitAb = gait abnormality; HF = heart failure; HTN = hypertension; ICD = International Classification of Diseases; Incont. = incontinence; Osteo= osteoporosis; PD = Parkinson’s disease; PerNeuro = peripheral neuropathy; PVD = peripheral vascular disease; VA-FI = Veterans Affairs Frailty Index; WgtLoss = weight loss.
Figure 2.
Distribution of VA-FI for 2018 annual cohort using 3- and 1-y assessment periods and using 3-y assessment with CDW codes only. Density of VA-FI-10 for the 2018 annual cohort fit to a gamma distribution, using CDW with CMS data vs CDW data only with 3-y assessment period (left) and using 3- vs 1-y assessment period (right). CDW = Corporate Data Warehouse; CMS = Centers for Medicaid and Medicare Services; VA-FI = Veterans Affairs Frailty Index.
Survival by VA-FI Categories and Adjusted Associations With VA-FI-10
Figure 3 shows survival estimates from the index date for the 2014–2016 annual cohorts by frailty category. The VA-FI-10 differentiates individuals with varying risk of mortality, with nonfrail patients having the highest survival rates. The 5-year survival in 2012 was 85.4% (95% CI: 85.3%–85.5%) for nonfrail patients versus 30.9% (30.7%–31.1%) for frail patients. The degree of risk stratification remained consistent before and after the transition. Adjusted HRs for frailty categories relative to nonfrail across annual cohorts are shown in Figure 4. The hazard of mortality increased among patients with greater levels of VA-FI after adjustment. Hazard ratios for prefrail, mildly frail, and moderately frail versus nonfrail patients were mostly stable across years. There was an increasing trend in the HR for severely frail versus nonfrail patients over time. However, this trend began before the ICD-10 transition. The associations with frailty were attenuated when repeating these analyses using deficit defined using codes from only the CDW and somewhat stronger when using codes based on a 1-year assessment period (Supplementary Figure S4). However, the overall patterns of increasing risk by frailty category remained the same.
Figure 3.
Survival stratified by VA-FI category in 2014–2016 annual cohorts. Kaplan–Meier estimates of survival from index date of each year for 2014 (A), 2015 (B), and 2016 (C) annual cohorts by VA-FI category. CDW = Corporate Data Warehouse; CMS = Centers for Medicaid and Medicare Services; VA-FI = Veterans Affairs Frailty Index.
Figure 4.
Adjusted HRs for VA-FI categories for 2012–2018 annual cohorts. Estimates of the HR for each VA-FI category relative to nonfrail patients from Cox proportional hazard models for survival across the annual cohorts and pointwise 95% confidence intervals. The estimates are obtained from a Cox model that stratifies on age and sex and is adjusted for race smoking status, and region. HR = hazard ratio; VA-FI = Veterans Affairs Frailty Index.
Discussion
In this study, we updated the VA-FI-9 to VA-FI-10, incorporating ICD-10 diagnostic and procedural codes and new CPT and HCPCS codes representing age-related health deficits. Two geriatricians’ review of these codes ensured content validity of the VA-FI-10, with high interrater reliability. The VA-FI-10 showed overall stability in measuring individual health deficits and summative frailty over periods before and after the ICD-10 transition. It also maintained strong associations with mortality, reinforcing its predictive validity.
Tracking Frailty and Its Deficits in Older U.S. Veterans Beyond the ICD-9 Era
Previous epidemiologic studies have used the VA-FI to estimate frailty and its impact in older U.S. Veterans (5–8). Patel et al. found frailty based on the VA-FI to predict mortality in 3807 older Veterans with multiple myeloma (7). Similarly, Ganta et al. found the VA-FI predictive of mortality in 16,761 older community-dwelling Veterans (5). Griffith et al. evaluated the VA-FI in a predictive model for mortality in older Veterans with diabetes (6), and La et al. used the VA-FI to evaluate treatment effects in subgroups of frailty in Veterans with cancer (8). The VA-FI-10 can now be used as a predictor, confounder, and effect modifier in cohorts spanning the ICD-10 era. It can be studied as a continuous or categorical variable, using cut-points previously adopted that have been validated again in our analyses (4,15–18,19).
Our study provides a contemporary view of health-related deficits in U.S. Veterans. In the comorbidity domain, for example, increasing prevalence of diabetes and related conditions such as kidney disease reflect trends in the general population, with higher rates observed in Veterans (20,21). Similarly, increasing burden of cancer may reflect higher national trends for screening and better treatments and supportive care (22–25). In the cognitive and mood domains, dementia, depression, and anxiety prevalence all increased gradually. Studies in the general population are mixed regarding trends in dementia (26), but U.S. Veterans experience high rates of traumatic brain injury, posttraumatic stress disorder, and depression—all of which contribute to higher risk of mood disorders and cognitive impairment (27,28). In the functional domain, increased use of durable medical equipment may be secondary to rises in comorbidities, which alone or in synergy can impair physiologic systems necessary for independent mobility and functioning (29).
Important Considerations When Updating eFIs Across ICD Transitions
The prevalence of certain health deficits changed more significantly than others during the ICD-9 to ICD-10 transition. For example, atrial fibrillation increased from 12.3% in 2015 to 20.7% in 2016, which may be in part due to improved detection (30). However, the near doubling in prevalence over 1 year during the transition suggests other influences, such as changes in clinician coding behaviors. The more granular coding schema for atrial fibrillation (from 1 ICD-9 code to 4 ICD-10 codes), provides additional opportunities to bill for atrial fibrillation. Instead of just “atrial fibrillation,” clinicians can now choose from “atrial fibrillation, unspecified,” or “chronic” and “paroxysmal atrial fibrillation.”
On the other hand, the more granular coding schema in ICD-10 may underlie some declines in deficits we observed. The prevalence of arthritis decreased from 44.1% in 2012 to 39.1% in 2018. Clinicians may not have been familiar with the new codes for osteoarthritis and either failed to code entirely or coded the condition inaccurately, as others have suggested (31–33). Moreover, evolving reimbursement incentives may have influenced physician billing patterns over the study period (34).
Other authors who have updated risk indices, including a U.S. Medicare-based frailty index, have found stability across transition periods, but have also commented on the increased plurality of codes in ICD-10 and resultant challenges (9,33,35–37). Two studies used GEMs to map ICD-9 to ICD-10 and showed the necessity of manual review of mappings to ensure accuracy as only a small portion was defined as “exact” mappings, whereas others were defined as “approximate” (9,37). In our study, for example, “baby stroller colliding with stationary object” (ICD-10 = V00.822) was approximately mapped from “accidental fall from other furniture” (ICD-9 = E884.5)—we excluded this code from our VA-FI-10 as it did not meet age-related health deficit criteria.
Impact on VA-FI-10 of Varying Data Sources and Assessment Periods
Linking VHA data to CMS shifted the distribution of VA-FI-10 to more severe frailty. After CMS linkage, in 2018, 1 in 4 Veterans were reclassified to a more severe frailty category. This shift is expected as the addition of CMS data allows broader capture of health deficits managed outside VHA. This is important as VHA policies allow Veterans to receive care either in VHA or in civilian practices (10). Therefore, not including CMS data may underestimate a Veteran’s frailty, suggesting caution when making treatment decisions based on frailty cutoffs using VHA data alone. Including CMS data also strengthens the association between the VA-FI and mortality, particularly for the severely frail. However, using VHA data alone, the VA-FI still demonstrates a clear relationship with mortality over increasing severity of frailty.
Using a 1-year assessment period to measure health deficits versus a 3-year period, the distribution of VA-FI-10 shifted to less severe frailty, although fewer Veterans were reclassified than when CMS data were added to VHA data. The association with mortality was strengthened with a 1-year period, especially in higher-frailty categories. This may reflect increased health care utilization in the last year of life. Other studies have demonstrated that associations between individual variables and outcomes change over varying assessment periods; the magnitude and direction of this change depend on the individual variable (11,12,38,39). Investigators and clinicians using eFIs should consider these impacts on the frailty distribution and association with outcomes when considering data sources and the length of assessment period in which to measure health deficits.
Strengths and Limitations
Our decision to update the VA-FI using clinical knowledge to include codes representing health deficits is similar to other existing eFIs but stands in contrast to data-driven, automated approaches (2,37,40–43). An advantage of using an a priori selection of codes meeting the criteria of frailty by deficit accumulation ensures broad coverage of domains relevant to older adult health. A data-driven approach may exclude important variables that represent clinically relevant conditions but perform poorly in automated procedures (44,45). Disadvantages to our approach are that it is labor-intensive, less parsimonious, and introduces subjectivity.
Limitations of our study include disproportionally fewer women in the U.S. Veteran population in comparison with the general population, and thus results of the VA-FI may not generalize to other populations. However, the VA-FI was developed and now updated using a nationwide sample of millions of Veterans, making external validation in other populations unnecessary for its intended use in VHA. The VA-FI is based on the well-validated deficit accumulation definition of frailty, but other approaches such as the physical phenotype exist (46). Different approaches can lead to different variables in the frailty index, leading to differences in frailty classifications and in prediction of outcomes (2,47).
Conclusions
This study provides the VA-FI-10 as an updated eFI incorporating ICD-10 and current procedural codes, extending the ability estimate frailty beyond 2015. Although linkage to CMS data leads to more accurate classification and stronger associations with mortality, the VA-FI-10 can also be measured in VHA data alone as a risk factor, confounder, and effect modifier. Future efforts should investigate the VA-FI not only in epidemiologic studies, but also as an automated measure to be used at the point-of-care by VA clinicians to identify frailty in Veterans.
Supplementary Material
Funding
This work was support by the Harvard Translational Research in Aging Training Program, National Institute on Aging of the National Institutes of Health (T32AG023480 to C.D.); the Health Care Systems Research Network (HCSRN)-Older Americans Independence Centers (OAICs) AGING Initiative (R33AG057806 to C.D. and N.F.); the National Institute of Diabetes and Digestive and Kidney Diseases (T32-DK007199 to M.Z.); the National Institute on Aging (R01AG062713 to D.H.K.); the Department of Veterans Affairs (CSR&D CDA-2 IK2-CX001800 to A.R.O.); and National Institute on Aging (R03-AG060169 to A.R.O.). Support for VA/CMS data is provided by the Department of Veterans Affairs, VA Health Services Research and Development Service, VA Information Resource Center (project numbers SDR 02-237 and 98-004). The funding sources 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. The views expressed are those of the authors and do not represent the views of VA or the U.S. Government.
Conflict of Interest
D.H.K. provides paid consultative services to Alosa Health, a nonprofit educational organization with no relationship to any drug or device manufacturers. All other authors declare no conflict of interest.
Author Contributions
Concept and design: D.C., C.D., J.M.G., N.D., M.B., K.C., J.A.D., N.F., and A.R.O.;
acquisition, analysis, or interpretation of data: D.C., C.D., C.Y., B.C., C.H., M.Z., J.P., E.Y., D.H.K., J.A.D., N.F., and A.R.O.; drafting of the manuscript: D.C., C.D., N.F., and A.R.O.; critical revision: all authors; statistical analysis: D.C., C.Y., B.C., and N.F.; obtained funding: C.D., K.C., N.F., and A.R.O.; and supervision: N.F. and A.R.O.
Data Availability Statement
The data underlying this article were accessed from the VA Corporate Data Warehouse. The derived data generated in this research will be shared on reasonable request to the corresponding author as permitted by VA policy. We have also uploaded code to compute the VA Frailty Index here: https://github.com/bostoninformatics/va_frailty_index.
References
- 1.Buta BJ, Walston JD, Godino JG, et al. Frailty assessment instruments: systematic characterization of the uses and contexts of highly-cited instruments. Ageing Res Rev. 2016;26:53–61. doi: 10.1016/j.arr.2015.12.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kim DH, Patorno E, Pawar A, Lee H, Schneeweiss S, Glynn RJ. Measuring frailty in administrative claims data: comparative performance of four claims-based frailty measures in the U.S. medicare data. J Gerontol A Biol Sci Med Sci. 2020;75:1120–1125. doi: 10.1093/gerona/glz224 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Price LE, Shea K, Gephart S. The Veterans Affairs’s Corporate Data Warehouse: uses and implications for nursing research and practice. Nurs Adm Q. 2015;39:311–318. doi: 10.1097/NAQ.0000000000000118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Orkaby AR, Nussbaum L, Ho YL, et al. The burden of frailty among U.S. veterans and its association with mortality, 2002–2012. J Gerontol A Biol Sci Med Sci. 2019;74:1257–1264. doi: 10.1093/gerona/gly232 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ganta N, Sikandar S, Ruiz SJ, et al. Incidence of frailty in community-dwelling United States older veterans. J Am Med Dir Assoc. 2021;22:564–569. doi: 10.1016/j.jamda.2020.10.019 [DOI] [PubMed] [Google Scholar]
- 6.Griffith KN, Prentice JC, Mohr DC, Conlin PR. Predicting 5- and 10-year mortality risk in older adults with diabetes. Diabetes Care. 2020;43:1724–1731. doi: 10.2337/dc19-1870 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Patel BG, Luo S, Wildes TM, Sanfilippo KM. Frailty in older adults with multiple myeloma: a study of US veterans. JCO Clin Cancer Inform. 2020;4:117–127. doi: 10.1200/CCI.19.00094 [DOI] [PubMed] [Google Scholar]
- 8.La J, Cheng D, Brophy MT, et al. Real-world outcomes for patients treated with immune checkpoint inhibitors in the Veterans Affairs system. JCO Clin Cancer Inform. 2020;4:918–928. doi: 10.1200/CCI.20.00084 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Feudtner C, Feinstein JA, Zhong W, Hall M, Dai D. Pediatric Complex Chronic Conditions Classification System version 2: updated for ICD-10 and complex medical technology dependence and transplantation. BMC Pediatr. 2014;14:199. doi: 10.1186/1471-2431-14-199 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Aggarwal NK. Ramifications of the VA MISSION Act of 2018 on mental health: potential implementation challenges and solutions. JAMA Psychiatry. 2020;77:337–338. doi: 10.1001/jamapsychiatry.2019.3883 [DOI] [PubMed] [Google Scholar]
- 11.Gilbertson DT, Bradbury BD, Wetmore JB, et al. Controlling confounding of treatment effects in administrative data in the presence of time-varying baseline confounders. Pharmacoepidemiol Drug Saf. 2016;25:269–277. doi: 10.1002/pds.3922 [DOI] [PubMed] [Google Scholar]
- 12.Nakasian SS, Rassen JA, Franklin JM. Effects of expanding the look-back period to all available data in the assessment of covariates. Pharmacoepidemiol Drug Saf. 2017;26:890–899. doi: 10.1002/pds.4210 [DOI] [PubMed] [Google Scholar]
- 13.Searle SD, Mitnitski A, Gahbauer EA, Gill TM, Rockwood K. A standard procedure for creating a frailty index. BMC Geriatr. 2008;8:24. doi: 10.1186/1471-2318-8-24 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Centers for Medicare and Medicaid Services Chronic Conditions Data Warehouse: Condition Categories. https://www2.ccwdata.org/web/guest/about-ccw. Accessed November 1, 2019. [Google Scholar]
- 15.Armstrong JJ, Andrew MK, Mitnitski A, Launer LJ, White LR, Rockwood K. Social vulnerability and survival across levels of frailty in the Honolulu-Asia Aging Study. Age Ageing. 2015;44:709–712. doi: 10.1093/ageing/afv016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Blodgett J, Theou O, Kirkland S, Andreou P, Rockwood K. Frailty in NHANES: comparing the frailty index and phenotype. Arch Gerontol Geriatr. 2015;60:464–470. doi: 10.1016/j.archger.2015.01.016 [DOI] [PubMed] [Google Scholar]
- 17.Hoover M, Rotermann M, Sanmartin C, Bernier J. Validation of an index to estimate the prevalence of frailty among community-dwelling seniors. Health Rep. 2013;24:10–17. [PubMed] [Google Scholar]
- 18.Pajewski NM, Williamson JD, Applegate WB, et al. Characterizing frailty status in the systolic blood pressure intervention trial. J Gerontol A Biol Sci Med Sci. 2016;71:649–655. doi: 10.1093/gerona/glv228 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Orkaby AR, Hshieh TT, Gaziano JM, Djousse L, Driver JA. Comparison of two frailty indices in the physicians’ health study. Arch Gerontol Geriatr. 2017;71:21–27. doi: 10.1016/j.archger.2017.02.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Afkarian M, Zelnick LR, Hall YN, et al. Clinical manifestations of kidney disease among US adults with diabetes, 1988–2014. J Am Med Assoc. 2016;316:602–610. doi: 10.1001/jama.2016.10924 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Liu Y, Sayam S, Shao X, et al. Prevalence of and trends in diabetes among veterans, United States, 2005–2014. Prev Chronic Dis. 2017;14:E135. doi: 10.5888/pcd14.170230 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Shapiro CL. Cancer survivorship. N Engl J Med. 2018;379:2438–2450. doi: 10.1056/NEJMra1712502 [DOI] [PubMed] [Google Scholar]
- 23.DeSantis CE, Miller KD, Dale W, et al. Cancer statistics for adults aged 85 years and older, 2019. CA Cancer J Clin. 2019;69:452–467. doi: 10.3322/caac.21577 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Howlader N, Forjaz G, Mooradian MJ, et al. The effect of advances in lung-cancer treatment on population mortality. N Engl J Med. 2020;383:640–649. doi: 10.1056/NEJMoa1916623 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Moye J, Schuster JL, Latini DM, Naik AD. The future of cancer survivorship care for veterans. Fed Pract. 2010;27:36–43. [PMC free article] [PubMed] [Google Scholar]
- 26.Alzheimer’s disease facts and figures. Alzheimer Dement. 2020;16:391–460. doi: 10.1002/alz.12068 [DOI] [Google Scholar]
- 27.Hoge CW, McGurk D, Thomas JL, Cox AL, Engel CC, Castro CA. Mild traumatic brain injury in U.S. soldiers returning from Iraq. N Engl J Med. 2008;358:453–463. doi: 10.1056/NEJMoa072972 [DOI] [PubMed] [Google Scholar]
- 28.Lwi SJ, Barnes DE, Xia F, Peltz C, Hoang T, Yaffe K. Ten-year prevalence of cognitive impairment diagnoses and associated medical and psychiatric conditions in a national cohort of older female veterans. Am J Geriatr Psychiatry. 2019;27:417–425. doi: 10.1016/j.jagp.2018.12.015 [DOI] [PubMed] [Google Scholar]
- 29.Ryan A, Wallace E, O’Hara P, Smith SM. Multimorbidity and functional decline in community-dwelling adults: a systematic review. Health Qual Life Outcomes. 2015;13:168. doi: 10.1186/s12955-015-0355-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Morillo CA, Banerjee A, Perel P, Wood D, Jouven X. Atrial fibrillation: the current epidemic. J Geriatr Cardiol. 2017;14:195–203. doi: 10.11909/j.issn.1671-5411.2017.03.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Horsky J, Drucker EA, Ramelson HZ. Accuracy and completeness of clinical coding using ICD-10 for ambulatory visits. AMIA Annu Symp Proc. 2017;2017:912–920. [PMC free article] [PubMed] [Google Scholar]
- 32.Yoon J, Chow A. Comparing chronic condition rates using ICD-9 and ICD-10 in VA patients FY2014–2016. BMC Health Serv Res. 2017;17:572. doi: 10.1186/s12913-017-2504-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Khera R, Dorsey KB, Krumholz HM. Transition to the ICD-10 in the United States: an emerging data chasm. J Am Med Assoc. 2018;320:133–134. doi: 10.1001/jama.2018.6823 [DOI] [PubMed] [Google Scholar]
- 34.Geruso M, Layton T. Upcoding: evidence from Medicare on squishy risk adjustment. J Polit Econ. 2020;12:984–1026. doi: 10.1086/704756 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Glasheen WP, Renda A, Dong Y. Diabetes Complications Severity Index (DCSI)-update and ICD-10 translation. J Diabetes Complications. 2017;31:1007–1013. doi: 10.1016/j.jdiacomp.2017.02.018 [DOI] [PubMed] [Google Scholar]
- 36.Glasheen WP, Cordier T, Gumpina R, Haugh G, Davis J, Renda A. Charlson Comorbidity Index: ICD-9 update and ICD-10 translation. Am Health Drug Benefits. 2019;12:188–197. [PMC free article] [PubMed] [Google Scholar]
- 37.Gautam N, Bessette L, Pawar A, Levin R, Kim DH. Updating International Classification of Diseases 9th revision to 10th revision of a claims-based frailty index. J Gerontol A Biol Sci Med Sci. 2021;76(7):1316–1317. doi: 10.1093/gerona/glaa150 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Savitz DA, Barón AE. Estimating and correcting for confounder misclassification. Am J Epidemiol. 1989;129:1062–1071. doi: 10.1093/oxfordjournals.aje.a115210 [DOI] [PubMed] [Google Scholar]
- 39.Brunelli SM, Gagne JJ, Huybrechts KF, et al. Estimation using all available covariate information versus a fixed look-back window for dichotomous covariates. Pharmacoepidemiol Drug Saf. 2013;22:542–550. doi: 10.1002/pds.3434 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Davidoff AJ, Zuckerman IH, Pandya N, et al. A novel approach to improve health status measurement in observational claims-based studies of cancer treatment and outcomes. J Geriatr Oncol. 2013;4:157–165. doi: 10.1016/j.jgo.2012.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Faurot KR, Jonsson Funk M, Pate V, et al. Using claims data to predict dependency in activities of daily living as a proxy for frailty. Pharmacoepidemiol Drug Saf. 2015;24:59–66. doi: 10.1002/pds.3719 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Segal JB, Chang H-Y, Du Y, Walston JD, Carlson MC, Varadhan R. Development of a claims-based frailty indicator anchored to a well-established frailty phenotype. Med Care. 2017;55:716–722. doi: 10.1097/MLR.0000000000000729 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Kundi H, Valsdottir LR, Popma JJ, et al. Impact of a claims-based frailty indicator on the prediction of long-term mortality after transcatheter aortic valve replacement in Medicare beneficiaries. Circ Cardiovasc Qual Outcomes. 2018;11:e005048. doi: 10.1161/CIRCOUTCOMES.118.005048 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Steyerberg EW, SpringerLink (Online service). Clinical prediction models a practical approach to development, validation, and updating . In: Statistics for Biology and Health. 2nd ed. Springer; 2019;207–225. [Google Scholar]
- 45.Harrell FE.Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; 2001. [Google Scholar]
- 46.Segal JB, Huang J, Roth DL, Varadhan R. External validation of the claims-based frailty index in the National Health and Aging Trends Study cohort. Am J Epidemiol. 2017;186:745–747. doi: 10.1093/aje/kwx257 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Shi SM, McCarthy EP, Mitchell S, Kim DH. Changes in predictive performance of a frailty index with availability of clinical domains. J Am Geriatr Soc. 2020;68:1771–1777. doi: 10.1111/jgs.16436 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this article were accessed from the VA Corporate Data Warehouse. The derived data generated in this research will be shared on reasonable request to the corresponding author as permitted by VA policy. We have also uploaded code to compute the VA Frailty Index here: https://github.com/bostoninformatics/va_frailty_index.