Translation of a Claims-Based Frailty Index From the International Classification of Diseases, Ninth Revision, Clinical Modification to the Tenth Revision

Abstract

The Faurot frailty index (FFI) is a validated algorithm that uses enrollment and International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)–based billing information from Medicare claims data as a proxy for frailty. In October 2015, the US health-care system transitioned from the ICD-9-CM to the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). Applying the Centers for Medicare and Medicaid Services General Equivalence Mappings, we translated diagnosis-based frailty indicator codes from the ICD-9-CM to the ICD-10-CM, followed by manual review. We used interrupted time-series analysis of Medicare data to assess the comparability of the pre- and posttransition FFI scores. In cohorts of beneficiaries enrolled in January 2015–2017 with 8-month frailty look-back periods, we estimated associations between the FFI and 1-year risk of aging-related outcomes (mortality, hospitalization, and admission to a skilled nursing facility). Updated indicators had similar prevalences as pretransition definitions. The median FFI scores and interquartile ranges (IQRs) for the predicted probability of frailty were similar before and after the International Classification of Diseases transition (pretransition: median, 0.034 (IQR, 0.02–0.07); posttransition: median, 0.038 (IQR, 0.02–0.09)). The updated FFI was associated with increased risks of mortality, hospitalization, and skilled nursing facility admission, similar to findings from the ICD-9-CM era. Studies of medical interventions in older adults using administrative claims should use validated indices, like the FFI, to mitigate confounding or assess effect-measure modification by frailty.

Abbreviations

ARIC

Atherosclerosis Risk in Communities

FFI

Faurot frailty index

GEM

General Equivalence Mapping

HCPCS

Healthcare Common Procedure Coding System

ICD

International Classification of Diseases

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

ICD-10-CM

International Classification of Diseases, Tenth Revision, Clinical Modification

PPF

predicted probability of frailty

SNF

skilled nursing facility

 

Frailty, a state of reduced ability to maintain physiological homeostasis following stress, is a distinct and important manifestation of the aging process (1). Frail individuals are vulnerable to physiological decline and disability and are at high risk of adverse health outcomes (2, 3). Although it is correlated with comorbidity and cognitive decline, frailty is a distinct clinical syndrome, with a unique etiology that remains largely unknown (3, 4). An estimated 15% of older adults (aged ≥65 years) in the United States are frail (5).

Frailty can be assessed using clinical evaluation tools, such as the Fried frailty phenotype, that require both physical assessment and patient-reported components (3). While these tools are useful for clinical and research purposes, they are underutilized in clinical practice (6). Even when they are administered, information from these assessments is not captured in administrative health-care data, such as insurance claims, which are increasingly used for generating real-world evidence on the effectiveness and safety of medical interventions. Medicare claims, which capture information on health-care encounters for the vast majority of adults aged 65 years or older in the United States, are a rich source of data for studying health and health-care delivery for older adult populations (7).

Several Medicare claims–based indices have been validated for prediction of frailty in older adults (8–12). Since their development, they have been widely used in pharmacoepidemiologic research for confounder adjustment and to assess effect-measure modification (13–16). One of these tools, developed by Faurot et al. in 2015 (8), predicts dependency in activities of daily living as a proxy for frailty based on demographic characteristics and the presence of 20 aggregated claims-based clinical constructs (“indicators”). These indicators are identified in Medicare claims data using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes and Healthcare Common Procedure Coding System (HCPCS) codes. The Faurot frailty index (FFI) has been validated against the frailty phenotype model, a gold standard clinical measure of frailty, and has been shown to predict frailty-related outcomes such as functional capacity, falls, and short-term mortality (5, 17).

On October 1, 2015, the US health-care system transitioned from the ICD-9-CM coding system to the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) coding system. The new system included an increased number of codes (approximately 68,000 codes vs. approximately 14,000 in the ICD-9-CM) and improved granularity for defining many health conditions and diagnoses (18). To our knowledge, no studies have translated the FFI to ICD-10-CM–era codes. Therefore, the primary aims of our study were 1) to translate the ICD-9-CM components of the original FFI to the ICD-10-CM coding system, 2) to assess consistency in the distribution of the composite predicted probability of frailty (PPF) measure across the ICD-9-CM and ICD-10-CM eras using the original and translated indices, and 3) to assess trends in the prevalence of the individual frailty indicators across the transition. Our secondary objective was to validate the new ICD-10-CM–era index by assessing the associations between predicted frailty and 1-year risk of 3 aging-related health outcomes: hospitalization, nursing home admission, and mortality.

METHODS

The Office of Human Research Ethics of the University of North Carolina at Chapel Hill (Chapel Hill, North Carolina) approved this study. All analyses were performed using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina).

Data source

We conducted our study using Medicare enrollment and fee-for-service claims data for a 0.2% random sample of Medicare beneficiaries between January 1, 2013, and December 31, 2017. These data captured information on enrollment, demographic characteristics, health-care encounter billing, and vital status for Medicare beneficiaries covered by Medicare Parts A and B, excluding those with Medicare Part C (or Medicare Advantage).

Translation of codes from the ICD-9-CM to the ICD-10-CM

The original FFI was developed using the 2006 Medicare Current Beneficiary Survey and was externally validated in the Atherosclerosis Risk in Communities (ARIC) Study (8, 17). The frailty indicators included in the final multivariable logistic regression model were selected from 57 candidate predicators using backward selection and bootstrapping of 1,000 samples to ascertain consistency of the backward selection results across samples. Parameter estimates from the original model are provided in Web Table 1 (available at https://doi.org/10.1093/aje/kwad151).

The final model included 23 indicators, of which 16 are identified using ICD-9-CM diagnosis codes. The remaining 7 indicators either are demographic variables (age, sex, race/ethnicity) or are identified using HCPCS codes (ambulance transport, home hospital bed, wheelchair, home oxygen), unaffected by the coding transition. We used a forward-backward General Equivalence Mapping (GEM) algorithm with a one-to-many approximate matching approach to translate the ICD-9-CM diagnosis codes used in the original Faurot index to ICD-10-CM codes. The GEM code lists were developed by the Centers for Medicare and Medicaid Services and the Centers for Disease Control and Prevention using input from clinical and coding experts (19). Each identified ICD-10-CM code in our study was then reviewed manually by one of the authors (E.D.D., S.S., K.R.F., J.P., A.M., or J.L.L.) to assess consistency with the original indicator definitions. Codes that were not related to the original construct were removed. We surveyed the literature and used a coding informatics database (20) to add additional codes related to the construct that were not identified in the initial GEM translation. Uncertainties were resolved by a second reviewer (E.D.D., S.S., K.R.F., or J.L.L.). We also reviewed the coding informatics database (20) and literature to confirm that HCPCS codes used to identify the receipt of ambulance transport, home hospital beds, wheelchairs, and home oxygen had not been added since development of the original model.

Assessment of frailty indicators

To assess the comparability of the individual components of the FFI between the ICD-9-CM and ICD-10-CM eras, we identified monthly cross-sectional cohorts of beneficiaries enrolled in Medicare Parts A and B during each respective calendar month between January 2013 and December 2017. Beneficiaries were required to be at least 65 years of age on the first day of the month and to stay enrolled through that calendar month. Individuals contributed to all months in which they met inclusion criteria.

Using these monthly cross-sectional cohorts, we estimated the monthly prevalence of each of the diagnosis-based indicators of the Faurot model for each month between January 2013 and December 2017. Monthly prevalence was defined as the number of individuals who had at least 1 claim with a diagnosis code for the indicator of interest (e.g., bladder dysfunction, podiatric care, heart failure) during the calendar month, divided by the number of Medicare beneficiaries enrolled in Medicare Parts A and B during that calendar month. Prevalence estimates were standardized by age and sex using direct standardization (21), with the age and sex distribution in December 2017 serving as the reference standard. Comparability between ICD-9-CM– and ICD-10-CM–era codes was assessed visually, by plotting the monthly prevalence estimates over time (22).

We used segmented linear regression of interrupted time-series data with an autoregressive correlation structure to assess the change in level and trend in the prevalence of each frailty indicator at the time of the International Classification of Diseases (ICD) transition (October 2015) (23, 24). We accounted for autocorrelated errors in our models using Durbin-Watson tests (α = 0.05) to specify autoregressive parameters lagged up to 2 years (25). Interrupted time-series analysis has previously been used to assess the effect of ICD-9-CM to ICD-10-CM coding system transitions on the identification of a claims-based comorbidity index (26).

Assessment of the composite frailty index

The original FFI parametrically estimated a composite measure for the PPF based on the presence of codes for the indicators during an 8-month look-back assessment period. We assessed the comparability of the PPF measure between the ICD-9-CM and ICD-10-CM eras, using monthly cross-sectional cohorts of Medicare beneficiaries. For each calendar month between January 2014 and December 2017, we identified cross-sectional cohorts of Medicare beneficiaries who were enrolled in Medicare Parts A and B during the 8 months prior to the first day of the month (look-back period). Individuals were required to be aged 65 years or older at the start of the look-back period (i.e., 65 years plus 8 months on the first day of the month).

The PPF was estimated using the intercept and parameter estimates from the original Faurot model and claims during the 8-month look-back period. We described the distribution of the PPF for each of the monthly cohorts visually and using descriptive statistics (mean and median values and interquartile ranges). We used segmented linear regression of interrupted time-series data with an autoregressive correlation structure to assess the change in level and trend for the mean, median, and interquartile range bounds of the PPF measure (23, 24). We considered 2 separate interruption time points: the date of the system transition (October 1, 2015) and the time when the look-back period for frailty assessment would not overlap with the ICD-9-CM era (June 1, 2016; i.e., 8 months after the transition).

Prediction of short-term outcomes

To further validate the ICD-10-CM–era index, we examined the association between PPF and 3 important aging-related outcomes: 1-year hospitalization, admission to a skilled nursing facility (SNF), and mortality. For this analysis, we created 3 cohorts indexed on January 1, 2015, 2016, or 2017, respectively. Each cohort consisted of all Medicare beneficiaries aged 65 years or older with continuous enrollment in Medicare Parts A and B for at least 8 months prior to the index date. These 3 cohorts were chosen so that their 8-month preindex look-back periods would fall entirely within the ICD-9-CM era (2015), partially in the ICD-9-CM and ICD-10-CM eras (2016), or entirely within the ICD-10-CM era (2017). The use of January 1 as the index date for each cohort also reduced the chance of seasonality’s affecting the measure. The cohorts are referred to hereafter as the ICD-9-CM, mixed, and ICD-10-CM cohorts, respectively.

We followed each cohort for up to 1 year to identify the outcomes of interest. Hospital and SNF admissions were identified using admission dates in the Medicare Provider Analysis and Review (MEDPAR) file. Individuals who had an ongoing hospitalization or SNF stay on the index date were excluded from these analyses. One-year mortality was defined on the basis of Medicare date of death, which was derived from the National Death Index. The 1-year cumulative incidences of hospital and SNF admissions were plotted using Aalen-Johansen estimators (27). Death was treated as a competing risk, and individuals were censored at the time of disenrollment from Medicare Part A or B. One-year cumulative incidence of mortality was plotted using the Kaplan-Meier estimator. In the mortality analysis, individuals were not censored at the end of Medicare enrollment, since date of death is captured regardless of insurance coverage. In all analyses, cumulative incidence curves were stratified by the PPF: less than 0.05 (low), 0.05–0.09 (low-medium), 0.10–0.19 (medium), 0.20–0.39 (medium-high), or 0.40 or more (high). These categories were the same as those used in the initial Faurot et al. analysis, which demonstrated an increase in the cumulative incidence of mortality for higher frailty strata (8).

For each of the cohorts, we also estimated associations between the PPF and 1-year hospitalization or SNF admission using Fine-Gray subdistribution hazards models to account for death as a competing event (28). The association between the PPF and 1-year mortality was assessed using Cox proportional hazards regression. Frailty was categorized using the same cutpoints as those used in the cumulative incidence analyses.

RESULTS

Translation of frailty indicators

The initial GEMs translation using forward-backward mapping resulted in consistent trends in monthly prevalence for the majority of the frailty indicators between the ICD-9-CM and ICD-10-CM eras. The exceptions were lipid abnormalities and rehabilitation care, which both experienced sharp declines in prevalence after the ICD transition, and weakness, which experienced an increase in prevalence following the transition. Manual review of code lists for these indicators, resulting in removal and addition of diagnosis codes, improved consistency in the capture of these indicators. For rehabilitation services, Current Procedural Terminology codes were also added to improve capture. Final code lists for each of the indicators are provided in Web Table 1.

Web Figure 1 presents monthly point prevalences of the 16 diagnosis-based indicators of the FFI before and after the ICD-9-CM to ICD-10-CM transition, after finalization of the code list. The prevalence of several claims-based frailty indicators (bladder dysfunction, dementia, hypotension/shock, paralysis, psychiatric illness, cancer screening) increased over time. However, the change in prevalence for each indicator at the time of the ICD transition was less than 1 percentage point for all indicators (Web Table 2). The change in slope for all indicators was also close to 0, indicating that trend lines were consistent across the ICD transition.

Translation of the composite measure

Results from the interrupted time-series analysis of the PPF measure are presented in Figure 1, and parameter estimates are provided in Web Tables 3–6. Across all calendar months, the distribution of the PPF was right-skewed, with the mean being larger than the median and 75th percentile. Over the period from January 2014 to December 2017, the median PPF was stable, ranging from 0.033 to 0.039; the lower quartile was also stable over this period (range, 0.023–0.024). The mean PPF (range, 0.092–0.109) and the upper quartile (range, 0.072–0.095) increased steadily over the study period. None of the evaluated statistics (mean, median, and bounds of the interquartile range) exhibited a sudden increase or decrease in value at either interruption point. In addition, the general trends following the transition were similar to the trends during the period prior to the transition. These results indicated that the composite PPF measure was largely consistent across the ICD transition.

Figure 1.

Results from an interrupted time-series analysis of the predicted probability of frailty in cross-sectional cohorts of Medicare beneficiaries, 2014–2017.

A SAS macro for identifying the claims-based frailty indicators and for estimating the composite PPF in the ICD-9-CM and ICD-10-CM eras is available online (29).

One-year outcomes

The January 1 cohorts included 36,537, 37,286, and 37,487 older adults in the ICD-9-CM (2015), mixed (2016), and ICD-10-CM (2017) cohorts, respectively. Across cohorts, the median age of the Medicare beneficiaries on the index date was 74 years, and 60% were female; 87% were White, 7% were Black, 2% were Hispanic, and 4% were of other race/ethnicity. The proportions of individuals classified as having low (<0.05), low-medium (0.05–0.09), medium (0.10–0.19), medium-high (0.20–0.39), and high (≥0.40) frailty were similar for each cohort (Web Table 7).

The cumulative incidence of 1-year mortality, stratified by the PPF, is presented in Figure 2A. The PPF was strongly and positively associated with 1-year mortality risk. In the ICD-10-CM cohort, the 1-year cumulative incidences of mortality among individuals with low, low-medium, medium, medium-high, and high PPF were 1.7%, 5.3%, 10.2%, 17.8%, and 31.2%, respectively (Table 1). Higher PPF was also associated with increased 1-year cumulative incidence of hospitalization and SNF admission (Figures 2B and 2C, respectively; Table 1). For each of the study outcomes, the cumulative incidence curves were similar in the ICD-9-CM and mixed cohorts.

Figure 2.

Cumulative incidence of 1-year mortality (A), hospitalization (B), and admission to a skilled nursing facility (SNF) (C) according to the predicted probability of frailty in Medicare beneficiaries, 2014–2017. The line style depicts the cohort, and the line shading depicts the stratum of predicted probability of frailty (a darker shade denotes a higher frailty stratum). ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification.

Table 1.

One-Year Cumulative Incidence of Aging-Related Health Outcomes Among Medicare Beneficiaries According to Frailty Stratum and Revision of the International Classification of Diseases in Use at the Time, 2014–2017^a

	ICD Revision Cohort (Year)
	ICD-9-CM (2015)		Mixed (2016)		ICD-10-CM (2017)
Health Outcome and   Frailty Stratum	Cumulative   Incidence, %	95% CI	Cumulative   Incidence, %	95% CI	Cumulative   Incidence, %	95% CI
Mortality
Low (<0.05)	1.6	1.5, 1.8	1.6	1.4, 1.8	1.7	1.5, 1.8
Low-medium (0.05–0.09)	4.9	4.4, 5.5	4.6	4.1, 5.2	5.3	4.7, 5.9
Medium (0.10–0.19)	10.8	9.7, 12.0	9.6	8.6, 10.7	10.2	9.2, 11.3
Medium-high (0.20–0.39)	18.4	16.8, 20.2	17.6	16.1, 19.4	17.8	16.2, 19.5
High (≥0.40)	29.5	27.6, 31.5	26.5	24.7, 28.3	31.2	29.4, 33.1
Hospitalization
Low (<0.05)	13.3	12.9, 13.8	13.0	12.6, 13.4	12.9	12.4, 13.3
Low-medium (0.05–0.09)	25.0	23.9, 26.2	24.2	23.1, 25.4	24.0	22.9, 25.1
Medium (0.10–0.19)	33.9	32.1, 35.6	34.5	32.8, 36.2	33.3	31.7, 35.0
Medium-high (0.20–0.39)	39.5	37.3, 41.6	41.7	39.6, 43.8	38.9	36.8, 41.0
High (≥0.40)	50.5	48.3, 52.5	47.6	45.6, 49.6	47.9	45.9, 49.9
SNF admission
Low (<0.05)	2.8	2.6, 3.0	2.5	2.3, 2.7	2.2	2.1, 2.4
Low-medium (0.05–0.09)	8.8	8.1, 9.6	7.7	7.0, 8.4	7.7	7.0, 8.4
Medium (0.10–0.19)	14.0	12.8, 15.4	14.8	13.5, 16.1	13.3	12.1, 14.5
Medium-high (0.20–0.39)	19.3	17.6, 21.0	20.0	18.3, 21.8	18.9	17.3, 20.7
High (≥0.40)	29.2	27.3, 31.1	28.0	26.2, 29.9	26.3	24.5, 28.1

Abbreviations: CI, confidence interval; ICD, International Classification of Diseases; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification; SNF, skilled nursing facility.

^a Distribution of frailty strata in the ICD-9-CM cohort: low, 65.7%; low-medium, 14.9%; medium, 7.7%; medium-high, 5.6%; high, 6.0%. Distribution in the mixed cohort: low, 65.3%; low-medium, 15.0%; medium, 7.9%; medium-high, 5.6%; high, 6.2%. Distribution in the ICD-10-CM cohort: low, 65.2%; low-medium, 14.8%; medium, 8.0%; medium-high, 5.5%; high, 6.4%.

Results from the Fine-Gray and Cox proportional hazards regression models are presented in Figure 3. Increasing frailty was consistently associated with an increased hazard of each of the study outcomes across cohorts. For the 1-year mortality and hospital admission outcomes, the log hazard ratios for the ICD-9-CM and ICD-10-CM cohorts did not differ by more than 0.1 within each frailty stratum. In the analysis of 1-year SNF admission, there was slightly more variation in the hazard ratios between the ICD-9-CM and ICD-10-CM cohorts. The largest difference in the log hazard ratio between the ICD-9-CM and ICD-10-CM cohorts occurred in the comparison between the medium-high and low frailty strata.

Figure 3.

Associations between predicted probability of frailty and 1-year mortality (A), hospitalization (B), and admission to a skilled nursing facility (SNF) (C) in Medicare beneficiaries, 2014–2017. Low predicted probability of frailty (<0.05) served as the reference group. Bars, 95% confidence intervals (CIs). HR, hazard ratio; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ICD-10-CM, International Classification of Diseases, Tenth Revision, Clinical Modification.

DISCUSSION

We translated the Medicare claims–based FFI from the ICD-9-CM era to the ICD-10-CM era and evaluated the validity of the updated index. Overall, the FFI translated well. We found consistent trends in the individual claims-based frailty indicators and in the composite PPF measure. The updated index was strongly predictive of aging-related outcomes, and associations were consistent with those observed using claims data from the ICD-9-CM era.

Medicare enrollment and claims data are a valuable resource for studying health and health-care delivery in older adults. Confounding by frailty is an important potential source of systematic bias in observational studies of older adult populations, since frail older adults may be less likely to receive certain therapeutic interventions or procedures and may be more likely to experience aging-related outcomes (30). However, Medicare claims data do not contain clinical measures of frailty, such as the frailty phenotype (3). Several claims-based frailty algorithms, including the FFI, have been developed to assess frailty or poor physical function using Medicare data (8–12). The Kim frailty index has previously been updated for use in the ICD-10-CM era (31). Prior work has demonstrated that the original FFI can reduce bias due to confounding by frailty in studies using Medicare data (13, 14). Since its development, the original FFI has been used in numerous epidemiologic studies for confounder adjustment and has been used to emulate clinical trial enrollment criteria (15, 32–34). Our work builds upon this line of research by validating an updated FFI for use in the ICD-10-CM era.

After standardization for age and sex, we observed increases in prevalence over time for many of the diagnosis-based frailty indicators, including bladder dysfunction, dementia, hypotension/shock, paralysis, psychiatric illness, and cancer screening services. However, changes in slopes for all of the frailty indicators were near 0 (i.e., indicated no change) after the ICD transition. These increasing trends warrant further investigation to deduce whether observed increases in prevalence were due to true changes in the prevalence of these conditions over time. Since we observed no meaningful difference in the associations between claims-based frailty and short-term aging-related outcomes across the ICD transition, the trends may be more likely to be due to actual changes in prevalence, increased engagement with the health-care system, or other changes in billing practices.

We sought to translate the existing Medicare claims-based FFI and used a standard process to update the code lists and assess their consistency with the original indicators. We used the intercept and parameter estimates from the original ICD-9-CM model to estimate the PPF. However, improved granularity of ICD-10-CM diagnosis codes may warrant development of a new algorithm for predicting frailty in older Medicare beneficiaries. For example, while sarcopenia (an important indicator of frailty) does not have a distinct ICD-9-CM code, it has a unique ICD-10-CM diagnosis code (M62.84) (35, 36). While we have included this new code in our indicator for weakness, future models developed using ICD-10-CM codes may be able to better differentiate between weakness due to sarcopenia and weakness due to other muscular atrophy to improve frailty prediction. In addition, updated frailty models could consider additional elements, such as the timing and frequency of codes, to improve frailty ascertainment in Medicare claims data. While the FFI only considers demographic information and procedure- and diagnosis-based codes, additional types of codes, such as Diagnosis Related Group codes, or medication-based codes, such as Anatomical Therapeutic Chemical codes, could also improve the validity of claims-based frailty indices.

Results from our study should be interpreted in light of its limitations. First, we did not assess the criterion validity of the FFI (comparability with a gold standard) against a clinical measure of frailty, such as the frailty phenotype. Future work using Medicare linkages with cohort studies that capture clinical measures of frailty (e.g., linkage of ARIC or National Health and Aging Trends Study data with Medicare) may be used to assess this important aspect of validity or to update the weights (17, 37). The mean PPF estimated in our analysis represented the average claims-based score in our study cohort, rather than the underlying prevalence of frailty. A cutpoint would be required to estimate the prevalence of frailty using the FFI. Prior research has estimated the validity of using various cutpoints to identify frailty (17). In addition, while our work demonstrated the suitability of the updated index for use in Medicare beneficiaries with fee-for-service Medicare coverage (Parts A and B), its suitability may differ in other populations and settings. This is particularly relevant in the United States, where Medicare billing for certain conditions and procedures may differ substantially from that for other payers. In addition, results may not be generalizable to older adults covered by Medicare Part C (Medicare Advantage), which currently accounts for approximately 50% of all Medicare beneficiaries (38). Finally, we chose to focus on 1-year outcomes given the acute effect of frailty on short-term outcomes. Associations with longer-term outcomes were not evaluated and could differ across the ICD-9-CM and ICD-10-CM cohorts. In future work, investigators should consider evaluating the associations between the updated index and longer-term outcomes.

In conclusion, we translated the Medicare claims-based FFI for use in the ICD-10-CM era. The updated index remained strongly predictive of aging-relevant outcomes and behaved similarly to the original ICD-9-CM–era index. Studies of drug effects in older adults should use validated indices, such as the updated FFI, to mitigate confounding or assess effect-measure modification by frailty.