INTRODUCTION
The aging population is a critical global issue, with individuals aged ≥60 years expected to reach 2 billion by 2050, almost double the 900 million in 2015.1) In Japan, the aging process is remarkably rapid, with 29.4% of the population aged ≥65 years as of 2024, and this proportion is expected to rise to 37.1% in 2050.2) As societies age, the increasing prevalence of disability among older adults leads to substantial social, medical, and economic challenges.
Since 2000, Japan has implemented a long-term care insurance (LTCI) system to provide support based on the degree of physical and mental disability for adults aged ≥65 years. The LTCI system categorizes individuals into support levels 1–2, and care needs levels 1–5, with higher levels indicating a greater need for assistance.3) For example, support level 1 requires partial support with instrumental activities of daily living (IADL) while remaining independent in basic activities of daily living (ADL). Care level 1 involves declining IADL abilities beyond the support levels, with partial care required. In contrast, care level 5 indicates the need for assistance with all ADL tasks. The annual costs of LTCI have been steadily rising, reaching 10 trillion yen in 2018.4) As of 2024, approximately 14.0% of adults aged ≥65 years in Japan are certified as requiring long-term care under the LTCI system.5) Because population aging is expected to continue, the early identification of individuals at high risk for future disability is essential to reduce the social and financial burdens associated with LTCI certification.
Frailty is widely recognized for its association with adverse outcomes, including disability.6,7) In Japan, approximately 9.3% of community-dwelling older adults have been reported to be frail.8) Previous studies have shown an association between frailty and the incidence of disability.9-14) However, these studies have focused on relatively short follow-up periods of 3 to 6 years and were limited by small sample sizes. Moreover, these studies typically reported a single odds ratio (OR) or hazard ratio (HR) as the measure of association, overlooking the possibility that the relationship between frailty and the incidence of disability might vary over time.15,16) In the context of increasing life expectancy, it is critical to understand the long-term association between frailty and the incidence of disability, particularly how it varies over time. These insights could contribute to extending healthy life expectancy and reducing the social and financial burdens associated with disability.
Therefore, this study aimed to address a current evidence gap by examining how the association between baseline frailty and the incidence of disability changes over 10 years, by estimating the strength of the association across multiple follow-up intervals.
MATERIALS AND METHODS
Study Design
This cohort study was a longitudinal study conducted in Okayama City, the capital of Okayama Prefecture in western Japan. Okayama City had a population of 660,996 in 2006, including 125,954 adults aged ≥65 years (19.1%).17) This study aimed to investigate the relationship between lifestyle behaviors and certification for LTCI. Data were obtained from the municipal basic health examination program, which collects information on sociodemographic characteristics, medical history, self-rated health, lifestyle habits, and physical activity. Follow-up information on LTCI certification was obtained comprehensively from the municipal administrative database, which covers all city residents. Details of the study design and data collection methods have been described in a previous publication.12) Data were collected from 2006 to 2017.
Participants
The cohort was followed up for 10 years to obtain information on LTCI certification. The participants comprised 54,851 community-dwelling adults aged ≥65 years in Okayama City who underwent a basic health examination in 2006 or 2007. The inclusion criteria were age ≥65 years at the baseline assessment and residents of Okayama City. We excluded participants who were already certified in LTCI (care need) at baseline, had missing data on covariates, died before the first LTCI assessment, and were lost to follow-up before the first LTCI assessment (e.g., moving out of the city).
This study was approved by the Ethics Committee of Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences and Okayama University Hospital (No. K2106-038). The requirement for informed consent was waived due to the retrospective design of the study and data anonymization for analysis. This study followed the ethical guidelines for authorship and publication in the Annals of Geriatric Medicine and Research.18)
Exposure: Frailty Status
Frailty status was assessed using the Kihon Checklist (KCL), a screening tool developed in Japan to identify older adults at risk for requiring care.19) The KCL is a self-administered questionnaire with binary “yes” or “no” responses. It comprises 25 items categorized into seven domains: ADL, physical function, nutrition, oral function, housebound status, cognitive function, and depressive mood. Each item indicating a potential problem was scored as one point, with higher total scores reflecting an increased risk for disability. A previous study has shown that the KCL is associated with frailty status as defined by the Cardiovascular Health Study (CHS) frailty index, a widely recognized measure of frailty.19) Based on these results, we classified participants as frail with a total KCL score of ≥8 points and non-frail with scores of ≤7.
Outcome: Incidence of Disability
We defined the outcome as the incidence of disability based on LTCI certification at care need levels 1–5. Only the first LTCI certification was included, and changes in certification levels were excluded. Details about the LTCI certification levels are described above. We obtained data on LTCI certifications at the end of each year. To analyze the trends in the association between frailty and the incidence of disability by each time period, we divided the time into five intervals: years 1–2, 3–4, 5–6, 7–8, and 9–10. Participants were classified as having a disability if it occurred at any point during each time period.
Other Variables
Baseline characteristics included age, sex, comorbidities, alcohol consumption, and smoking status. Comorbidities consisted of heart disease, hypertension, renal disease, diabetes, liver disease, anemia, and hyperlipidemia. The dataset did not allow us to distinguish between type 1 and type 2 diabetes. These variables were selected as covariates based on previous studies, suggesting their potential role as confounders in the association between frailty and future disability.20-22)
Statistical Analysis
For descriptive analysis, continuous variables were presented as medians with interquartile ranges (IQRs), and categorical variables were presented as numbers with percentages. To handle missing values for KCL, we used multiple imputation using the chained equation (MICE), assuming missing at random. The imputation model included the baseline covariates (age, sex, comorbidities, alcohol consumption, and smoking status) and the outcome variables, including the binary incidence of disability and its time-to-event data, and generated 30 imputed datasets.23) If any item in the KCL was missing, the total score was considered missing and imputed using multiple imputation. The imputed total score was used to categorize the participants as frail or non-frail.
For the main analysis, we used the marginal structural model (MSM) to validly estimate the association between frailty and the incidence of disability by time period. The MSM creates a pseudo-population at each time period and compares participants' estimators of outcome events as if they had been frail versus as if they had not been frail. To implement the MSM, we used two types of weights: the inverse probability of treatment weight (IPTW) and the inverse probability of censoring weight (IPCW).24) First, we used the baseline covariates described above to estimate the propensity score (PS), defined as the probability of being classified as frail conditional on baseline covariates. The IPTW was calculated as the inverse of the PS. To reduce the variability of the weights, we stabilized the IPTW by replacing its numerator with the marginal probability of being classified as frail in the overall population.24) The distribution of PS among the frail and non-frail participants before and after applying IPTW is illustrated in Fig. S1. Second, we calculated the IPCW to account for potential selection bias due to censoring. In this study, censoring included participants who were lost to follow-up due to death or other reasons and those who received LTCI certification before the end of the respective time period. The IPCW was calculated as the inverse of the probability of remaining uncensored, conditional on the frailty status and baseline covariates. We stabilized the IPCW by using the probability of remaining uncensored conditional only on frailty status as the numerator.24) All probabilities were estimated using logistic regression models. The final weight for each participant was calculated by multiplying the stabilized IPTW and stabilized IPCW. Using these weights, we performed weighted logistic regression to estimate ORs and 95% confidence intervals (CIs) for the association between frailty and the incidence of disability by time period.
To evaluate whether the competing risk of death might influence the declining association between frailty and disability, we additionally conducted competing-risk analyses using Fine–Gray subdistribution hazard models. Because the censoring mechanism in this framework differs from that in the main MSM analysis, we recalculated interval-specific IPCWs for loss to follow-up within each time period. These IPCWs were then multiplied by the stabilized IPTW from the main analysis to construct the final weights. Using these weighted Fine–Gray models, we estimated subdistribution hazard ratios (sHRs) for incident disability within each interval while accounting for death as a competing event.
Using the standardized mean difference (SMD), we assessed the balance of covariates before and after applying IPTW and separately evaluated the balance before and after applying IPCW by time period. The SMD indicates the degree of imbalance between groups, with values closer to 0 reflecting improved balance and absolute values below 0.1 suggesting that covariates are well-balanced.25,26) To evaluate the robustness of the association between frailty and the incidence of disability to potential unmeasured confounding, the E-value was calculated. The E-value quantifies the minimum strength of association that an unmeasured confounder would need to have with both the exposure (frailty) and the outcome (incidence of disability), conditional on the measured covariates, to fully explain away the observed association.27) A higher E-value indicates that substantial unmeasured confounding would be necessary to explain away the observed association. Because this study was exploratory, no adjustments were made for multiple comparisons.28) All statistical analyses were performed using Stata/SE 18.0 (Stata Corp, College Station, TX, USA).
RESULTS
Fig. 1 presents the flow chart of the study participants. Of the 54,851 participants, we excluded 9,560 because of certified LTCI (care need) at baseline (n=6,984) and missing covariate data (n=1,368). Additionally, 994 participants died, and 214 were lost to follow-up before the first LTCI certification. The final analysis included 45,291 participants.
Table 1 shows the baseline characteristics of the included 45,291 participants. Among them, 8,652 (19.1%) participants had missing KCL values. Participants with missing KCL values did not show remarkable differences in baseline characteristics compared to the overall participants. The median age of the participants was 74 years (IQR, 70−78), and 16,348 (36.1%) were male. The prevalence of frail was 17.3% (n=7,828) and non-frail was 63.6% (n=28,811). Table 2 presents the number at risk, the incidence of disability, and the incidence proportion by time period.
Table 3 summarizes the ORs of frailty and the incidence of disability, estimated by weighted logistic regression, by time period. The adjusted ORs gradually decreased, but the associations persisted throughout the follow-up period. The adjusted ORs and 95% CI were 2.90 (2.55–3.29) for years 1–2, 2.40 (2.09–2.75) for years 3–4, 2.03 (1.76–2.34) for years 5–6, 1.75 (1.52–2.01) for years 7–8, and 1.58 (1.38–1.82) for years 9–10. After applying IPTW and IPCW, the balance of covariates was improved, as shown in Tables S1 and S2.
A competing-risk analysis using Fine–Gray models showed a similar temporal pattern, with sHRs displaying an overall decline over the 10-year follow-up and broadly aligning with the findings from the main analysis. Detailed results are shown in Table S3.
The E-values for the OR point estimate and the lower limits (LL) of the 95% CIs were as follows: 5.25 (4.54) for years 1–2, 4.23 (3.60) for years 3–4, 3.48 (2.92) for years 5–6, 2.90 (2.41) for years 7–8, and 2.54 (2.10) for years 9–10. These results indicate that, for all time periods, explaining away the observed association would require substantial unmeasured confounding strongly associated with both frailty and the incidence of disability, indicating that the observed association is robust to unmeasured confounding.
DISCUSSION
The key finding of this study was that baseline frailty was associated with the incidence of disability and that this association gradually declined over a 10-year follow-up period. This suggests that frailty has a stronger impact on disability risk in the earlier years following assessment, though the association remains substantial. While the overall association between frailty and disability has been established, our study adds a longer-term perspective that enhances understanding of how this association extends over time. By applying a MSM, we estimated this association from a causal perspective, providing robust evidence beyond a simple association.
Our findings extend previous evidence, primarily focusing on shorter follow-up periods of 3 to 6 years.9-14) The persistence of the association over 10 years suggests that frailty is not only related to the development of disability in the short term but also has long-term consequences. This long-term association may be explained by the cumulative effects of frailty, where progressive declines in physiological reserves and functional capacity increase vulnerability to disability over time. Given that frailty is often a progressive condition, individuals classified as frail at baseline may have continued to experience functional decline, contributing to the persistent association with disability.29) Previous studies have suggested that transitioning to a more frail state is associated with an increased risk of mobility impairment and mortality.29) Although we could not assess changes in frailty status over time, the observed long-term association highlights the importance of early identification and intervention for frailty to mitigate its long-term impact on the incidence of disability.
The association between frailty and disability gradually decreased, from an OR of 2.90 in years 1–2 to 1.58 in years 9–10. This declining magnitude can be attributed to several factors. First, the competing risk for death may have contributed to the observed decline in the association because frail older adults who are at a higher risk for mortality may not have survived sufficiently long to develop disability.19) Nevertheless, the competing-risk analysis using Fine–Gray models indicated a generally declining pattern similar to that observed in the main analysis. Although sHRs and ORs target different estimands—and the absolute sHR values differ from the ORs, likely because death data were available only as annual status updates rather than exact dates—the Fine–Gray results align with the overall time-varying trend. This suggests that the observed attenuation of the association between frailty and disability is unlikely to be fully explained by censoring due to death. Second, the natural aging process and the accumulation of new health problems over time may have attenuated the association between baseline frailty and subsequent disability. As time passes, even individuals who were initially non-frail are likely to develop frailty or other disabling conditions, thereby reducing the contrast in disability risk between the frail and non-frail groups. Third, the baseline assessment of frailty status might not fully capture the dynamic nature of frailty because individuals’ frailty status could change over time.29) As a result, our findings likely reflect the cumulative impact of initial frailty rather than the effects of time-updated frailty status. Despite these factors, the persistence of the association over 10 years highlights the long-term impact of frailty on disability, underscoring the importance of early identification and intervention.
This study assessed frailty using the KCL, a screening tool developed for older adults in Japan. The KCL has been widely used in community-based health assessments to identify individuals at risk for requiring long-term care. It incorporates physical, cognitive, and social factors for comprehensive frailty evaluation. Compared to other frailty measures, such as the CHS frailty scale, the KCL captures a broader range of factors contributing to frailty, which may result in differences in the prevalence of frailty and its association with disability. While a previous study has demonstrated that the KCL is associated with the CHS frailty index, variations in frailty classification criteria and assessment methods may influence the strength of associations.19) Future research should compare different frailty assessment tools to evaluate their associations with the long-term incidence of disability.
In a previous study using the same cohort, we examined the association between frailty and different levels of LTCI certification and found that frailty was associated with both mild and severe disability levels.12) These findings suggest that the magnitude of the association may vary by disability severity. However, in the present analysis, we treated LTCI certification levels 1–5 collectively as incident disability because the LTCI levels are administratively determined and may be influenced by contextual factors, making them less reliable as strict ordinal measures.30) Moreover, stratifying by LTCI levels would have substantially increased model complexity and made the results more difficult to interpret under the IPTW/IPCW framework. Based on these considerations, we regarded any LTCI certification as incident disability, consistent with previous large-scale Japanese cohort studies.9,10)
The findings of this study underscore the importance of early frailty detection and intervention for frailty to mitigate its long-term impact on disability. Given that frailty is a potentially modifiable condition, preventive strategies targeting at-risk older adults may play a crucial role in delaying or preventing disability.31) Multidimensional interventions, including physical activity programs, nutritional support, and cognitive interventions, have shown promise in improving the frailty status and preventing functional decline.32,33) Routine frailty screening, such as through the KCL, should be integrated into community-based healthcare systems to facilitate the timely identification and management of frailty. Additionally, the persistence of the association between frailty and disability over 10 years highlights the need for long-term interventions that extend beyond short-term care strategies. Future research should evaluate the long-term effectiveness of targeted interventions to alter the progression of frailty and reduce the risk for disability.
This study has several strengths. First, the 10-year follow-up period enabled a comprehensive assessment of the long-term association between frailty and the incidence of disability, extending previous evidence focused on shorter follow-up periods. Second, by applying the MSM with IPTW and IPCW, this study appropriately accounted for potential confounding and selection bias, thus enabling a more valid estimation of the time-varying association between frailty and disability over the follow-up period. These strengths enhance the reliability of our findings and contribute to a better understanding of the long-term implications of frailty. However, this study has several limitations. First, the potential for unmeasured confounding could not be ruled out. For example, additional comorbidities (e.g., dementia and stroke) and social factors, which are known to influence the incidence of disability, were not included due to the retrospective design.34,35) Nevertheless, our E-value analysis demonstrated that substantial unmeasured confounding would be required to explain away the observed associations, supporting the robustness of our findings. Second, approximately 19% of participants had missing KCL values, handled using multiple imputations under the assumption of missing at random (MAR). While this assumption cannot be directly verified, participants with missing KCL values did not differ substantially from those with complete data regarding baseline characteristics (Table 1), supporting the possibility of the MAR assumption. Bias may still exist if missingness was related to unmeasured factors such as frailty severity or functional limitations. Third, we did not account for time-varying covariates or changes in frailty status because of limited data. This may have led to misclassification of frailty and underestimation of the association, particularly in later years. Future studies with longitudinal data are needed. Fourth, our results might have underestimated the association because participants who died without LTCI certification were not included. This could lead to selection bias, particularly among those with severe frailty. To address this issue, we applied IPCW to reduce potential bias arising from attrition during follow-up and, in addition, performed a competing-risks analysis that treated death as a competing event. However, some degree of selection bias may remain despite these analytical approaches. Fifth, we could not identify the specific causes of LTCI certification. Frailty is associated with dementia, falls, and stroke—major contributors to disability.36-39) While these mechanisms could not be evaluated, frailty likely contributes to disability through multiple clinical pathways. Finally, because this study was conducted in Japan using the KCL, generalizability to other populations remains uncertain.
In conclusion, this study demonstrated that frailty was consistently associated with the incidence of disability over a 10-year follow-up period, with a stronger association in the earlier years and a gradual decline over time. These findings highlight the long-term impact of frailty and the need for early identification and intervention to mitigate its consequences.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
agmr-25-0118-Supplementary-Fig-S1.pdf
