4034 — Five Diverse Claims-Based Frailty Indices Predict Long-Term Institutionalization
Lead/Presenter: Huan Tianwen,
geriatrics and extended care data analysis center
All Authors: Huan T (Canandaigua VAMC), Wieland, D (retired VA) Intrator, O (Canandaigua VAMC) Kinosian, B (Philadelphia VAMC) Shubing, C (Canandaigua VAMC) Orkaby, A (Boston VAMC)
Long-term institutionalization (LTI) is a feared outcome for many older adults, most of whom wish to age in place, at home. LTI also carries a high-cost burden for healthcare institutions such as the Veterans Administration (VA). A claims-based tool that can readily identify older adults at risk of LTI at the population level is needed to target prevention strategies to those at highest risk. This study aims to assess the ability of five claims-based frailty indices (CFIs) to predict LTI compared to the benchmark VA Predictive Long-Term Institutionalization (PLI) score, designed to predict LTI.
VA and Medicare/Medicaid data were linked to retrieve diagnoses and calculate the Kim CFI, Segal CFI, JEN Frailty Index (JFI), VAFI, Figueroa CFI, and the PLI score. Each CFI is grounded in a different theory of frailty, including the Fried physical phenotype (Segal), Rockwood multiple-decrements (Kim, JFI, and VA-FI), and expert opinion (Figueroa). Each Rockwood CFI was developed using a different approach, e.g., LASSO regression for Kim and Geriatrician derived VA-FI. The VA PLI index overlaps with many variables included in the CFIs including variables related to morbidity, function, vital signs, labs, medication use, prior admissions, the JFI and the VA CAN score. CFI performance was examined with primary outcome any LTI stay >100 days in fiscal years 2015 to 2017. Since Segalâ€™s CFI and PLI already included age and sex, these variables were added to regression models for the Kim, JFI, VAFI, and Figueroa CFIs. Cox regression models calculated Harrellâ€™s concordance. Logistic regression calculated area under the receiver operating curve (AUC). Study sample included all Veterans seen at VA aged 65 and older in 2014, without LTI in the prior year.
In total 2.5 million veterans were included; mean age 74 with 98.2% male. The PLI had the highest concordance (0.85), followed by Kim-CFI (0.82), Segal-CFI (0.81), JFI (0.79), VA-FI (0.79), and Figueroa CFI (0.79). The same ranking was seen for AUC values: the PLI had the best performance (0.82), followed by Kim-CFI (0.79), Segal-CFI (0.78), JFI (0.76), VAFI (0.76), and Figueroa CFI (0.76).
Despite very different underlying constructs for frailty all five CFIs have very similar performance (range 0.04). A difference of 0.03-0.06 in CFI performance compared to a model dedicated to predicting LTI indicates very good performance. Although this cohort is predominantly male and may not generalized to non-VA cohorts, these data suggest there is no preference for which frailty index should be used, rather the most accessible CFI can be employed to identify high risk populations.
LTI is an important part of evaluation of population health. Tools that use clinical information to target patients at risk of LTI need further study. This study is an important first step to understanding which of the existing tools might be used clinically. Future work should be done to assess whether each CFI is identifying different groups and how CFIs may be used in tandem to identify populations at risk of LTI.