Evaluation of claims-based frailty measurements in older patients with cancer: a retrospective cohort study

Abstract

Background

Frailty is associated with poor outcomes in older adults with cancer. Several efforts have been made to assess frailty using the administrative claims data based on the number of clinical diagnosis codes, yet the literature reporting on this is scarce. This study aimed to evaluate the impact of frailty measures using administrative databases in Japan.

Design

A retrospective cohort study.

Setting and participants

5176 patients with cancer aged ≥65 years who underwent cancer treatment in hospitals.

Methods

The Electronic Frailty Index (eFI) and Veterans Affairs Frailty Index (VA-FI), based on diagnostic codes recorded were calculated. We plotted Kaplan–Meier survival curves and calculated hazard ratios (HR) using Cox regression analyses. The primary outcome was mortality, whereas the composite secondary outcome included a decline in care-need level, admission to a long-term care facility (LTCF) or mortality.

Results

The Kaplan–Meier survival curve demonstrated a significant association between the eFI and VA-FI and each research outcome. Compared to the lowest frailty group, the highest frailty group exhibited an HR of 2.59 [95% confidence interval (CI), 1.66–4.06] for eFI and 2.45 (95%CI, 1.02–5.91) for VA-FI in relation to a decline in care-need level, an LTCF admission and mortality. The trend test indicated a significant increase in the rate of each outcome with higher frailty levels.

Conclusions

Higher frailty levels are associated with an increased risk of composite outcomes in older adults with cancer. This study suggests the potential application of frailty measurements in oncology care settings.

Key Points

• Frailty assessed using administrative data could be used to identify older individuals with cancer.

• Increasing severity of frailty identifies older adults who are at an elevated risk of care need level decline and mortality.

• This study suggests the potential application of frailty measurements in oncology care settings.

Introduction

With the rapid increase in the aging population, the number of older adults with cancer has also increased. An estimated 2.3 million new cancer cases were reported in individuals aged 80 years or older worldwide, accounting for 13.3% of all new cancer cases diagnosed globally in 2018 [1]. Japan is one of the countries with the highest percentage of total cancer cases in older individuals, with more than 75% of patients with cancer being aged 65 years or older [2]. One of the primary challenges in caring for older patients with cancer is determining the most appropriate therapy for each patient [3, 4]. As the prevalence of frailty increases with age, recent guidelines on cancer treatment recommend assessing individual frailty to guide and support treatment decisions for older adults with cancer [5, 6].

Frailty is an age-related clinical condition characterised by an accumulation of vulnerability following a stressful event [7]. Older individuals with frailty are at an increased risk of adverse outcomes, such as mortality [8, 9], activities of daily living (ADL) decline [10] and admissions to long-term care facility (LTCF) [7, 11]. Therefore, cancer treatment should be determined by assessing health status rather than solely relying on chronological age [12, 13]. In clinical settings, frailty screening can be challenging because it requires a wide range of assessments and manual processing [14]. In recent years, various efforts have been made to assess frailty by utilising accessible administrative claims data, leveraging various clinical diagnostic codes [15–19], including the Electronic Frailty Index (eFI) and Veterans Affairs Frailty Index (VA-FI).

The eFI is a frailty scale that has been widely used in general practice in the UK, and the VA-FI has been developed in the USA, requiring an urgent need to promote healthier aging. The VA-FI is an internally and externally validated measurement using items on US claims data, including the VA database [17, 20], and the Surveillance, Epidemiology, and End Results-Medicare linked database [21]. The construct validity of the eFI and VA-FI has been demonstrated [20, 22]. To verify the applicability in specific cohort groups, such as older individuals with cancer, additional international validation studies are necessary using the highest percentage of total cancer cases in older individuals.

Routinely accumulated claims data provide standardised and feasible methods for identifying cumulative deficits. Given that claims-based frailty measures facilitate the identification of older adults with frailty, they enhance decision-making support in oncology settings by enabling standardised, evidence-based practice. Although the concept of frailty is recognised to predict outcomes in patients with cancer, the literature reporting on this is scarce. Patients with cancer often require admission to an LTCF for ADL assistance [23], and LTCFs frequently serve as settings for end-of-life care. Therefore, this study aimed to assess whether the eFI and VA-FI can be applied to administrative databases in Japan. Additionally, we evaluated the association between frailty, as measured by the eFI and VA-FI, and outcomes such as all-cause mortality, decline in care-need level and LTCF admission.

Methods

Study design and data source

This retrospective cohort study was conducted using an administrative database in Shizuoka Prefecture, Japan. For this study, we used claims data from the National Health Insurance System (<75 years old), Late-Stage Medical Care System for the elderly (≥75 years old), and Long-term Care Insurance data from 32 municipalities. The database includes information such as age, gender, observation period, reason for withdrawal and death dates during the study period. It also contains health insurance data (e.g. insurer code and disease code) and care insurance data (e.g. care levels and care service codes) [24]. The databases were linked using unique identifiers. Patients were defined as individuals aged ≥65 years who underwent initial cancer treatment in a hospital between 1 December 2012 and 30 September 2017. In the claims data, the month in which a new diagnosis of cancer and cancer treatment initiation were concurrently recorded was used as the index month. We included patients who had no prior history of cancer diagnosis or treatment in the 6-month period leading up to the index month. We excluded patients who died or used an LTCF during the index month (Figure 1).

Patients with cancer were identified using the International Classification of Diseases, 10th Revision (ICD-10) codes, along with procedure codes for endoscopic treatment, surgery, systemic therapy and radiotherapy. Claims-based definitions utilising diagnostic and procedure codes demonstrated high validity in a previous study [25]. This study was approved by the Institutional Review Board of Institute of Science Tokyo in Japan (M2020–240).

Frailty measures

We used the eFI and VA-FI in this study. The eFI was developed using UK primary care Read codes based on the deficit accumulation approach. This is measured using 36 equally weighted deficits, including ICD-10 codes and polypharmacy [16]. In the current study, polypharmacy was defined as the prescription of ≥5 drugs during 1 month before the index month [26]. The VA-FI was mainly developed using the sum of 31 equally weighted deficits using ICD-10 codes, Current Procedural Terminology and the Healthcare Common Procedure Coding System [18]. The full lists of eFI and VA-FI items are provided in Appendices 1 and 2.

To calculate the eFI and VA-FI, we used the ICD-10 codes recorded in medical claims during the baseline period. Both eFI and VA-FI were measured as equally weighted deficits, each of which was calculated as the sum of the number of deficits incurred divided by the total number of each possible deficit. For example, if there are 36 deficits for the eFI, and 9 of them are present in a given person, that person’s frailty index would be calculated as 9/36 = 0.25. We then categorised patients into four groups for the eFI, fit (0–0.12), mild frailty (>0.12–0.24), moderate frailty (>0.24–0.36) and severe frailty (>0.36), and into five groups for the VA-FI, nonfrail (≤0.1), prefrail (>0.1–0.2), mildly frail (>0.2–0.3), moderately frail (>0.3–0.4) and severely frail (>0.4) according to the previous studies [16, 17].

Outcomes

The primary outcome of this study was all-cause mortality. Death was defined as the month in which the reason for withdrawal was recorded as death, and the death dates were available. As the secondary outcomes, we employed the following composite outcomes: the first observation of a decline in care-need level or all-cause mortality, and the first observation of a decline in care-need level, the first observation of LTCF admission or all-cause mortality. Patients with cancer experience a rapid decline in ADL 1 month before death [27, 28]. There is a good correlation between Barthel index scores, which assess ADL, and a decline in care-need level [29]. In the current study, we defined a decline in care-need levels as a worsening of one or more care-need levels during the follow-up period in this study. Additionally, patients with cancer often require admission to an LTCF for ADL assistance [23], and LTCF is one of the places where death occurs [30]. To evaluate the events impacting frailty from multiple perspectives, including risks of death, decline in care-need level and LTCF admission, we employed these outcomes for this study. Care-need level and LTCF admission were defined based on the month in which the corresponding care service codes were recorded. Details regarding care-need level and LTCF admissions are provided in Appendix 3. The maximum possible follow-up period was 69 months.

Statistical analyses

The patients’ baseline characteristics were summarised. The distributions of the eFI and VA-FI were assessed using the Shapiro–Wilk test. Spearman’s correlation coefficient was used to evaluate the association among age, gender and continuous frailty scores. Kaplan–Meier survival curves were used to estimate each research outcome according to the categories of the two frailty measurements, and Log-rank tests were performed to examine the distribution between the populations. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards models. All models were adjusted for age, gender and types of cancer treatment as they reflect the cancer types. Trend tests were employed using Cox regression models to evaluate the association between the levels of each frailty index and the research outcomes. Two-sided P < 0.05 was considered statistically significant. Statistical analyses were performed using Stata version 18.0 (Stata Corporation, College Station, Texas, USA).

Results

Participant characteristics

Overall, 5176 patients with cancer underwent two frailty measurements. The mean age was 74.73 [standard deviation (SD) 6.68] years, and 43.5% were female. Types of cancer treatment included endoscopy 3464 (67.7%), surgery 1082 (21.2%), systemic therapy 487 (9.5%) and radiotherapy 84 (1.6%) (Table 1).

The mean eFI score was 0.10 (SD, 0.09), and the VA-FI score was 0.06 (SD, 0.06). There was a right-skewed distribution [W, 0.967 (P < 0.001) for the eFI and 0.968 (P < 0.001) for the VA-FI]. Histograms of the eFI and VA-FI are shown in Figure 2. Histograms of the eFI and VA-FI by age, gender and cancer treatment are shown in Appendices 4–6. Of the included study participants, 3407 (65.8%) were categorised as fit, 1391 (26.9%) as mildly frail, 335 (6.5%) as moderately frail and 43 (0.8%) as severely frail. The VA-FI proportions of patients classified as nonfrail, prefrail, mildly frail and moderately frail were 3794 (73.3%), 1253 (24.2), 122 (2.4%) and 7 (0.1%), respectively. Severe frail for the VA-FI was not observed in the current study.

Frailty, age and cancer treatments

The proportion of patients categorised as mild for the eFI increased among patients aged ≥75 years (34.5%). This trend was similar to prefrailty for the VA-FI among patients aged ≥75 years (30.9%). Spearman’s correlation coefficients between frailty scores and age were 0.31 (P < 0.001) for the eFI and 0.25 (P < 0.001) for the VA-FI.

For the eFI, the deficit of “fall” was absent among the 36 deficits, whereas for the VA-FI, the Japanese claim codes did not detect “fall,” “failure to thrive,” and “durable medical equipment” among the 31 deficits. The proportions of each item are shown in Appendices 1 and 2. For cancer treatment, patients with mild, moderate and severe frailty according to the eFI and those classified as prefrail, mildly frail and moderately frail according to the VA-FI underwent endoscopic treatment (i.e. endoscopic mucosal resection or endoscopic submucosal dissection), surgery, systemic therapy and radiotherapy (Appendix 7).

Association with mortality, care-need level decline or LTCF admissions

There was a total of 1088 deaths (21.0%), 712 (13.8%) cases of a decline of care-need level and 125 (2.52%) LTCF admissions at the end of the study period. The Kaplan–Meier survival curve demonstrated a significant difference in the eFI and VA-FI on each research outcome (Figure 3). The 60-month survival rate decreased with increasing frailty severity, ranging from 75.8% (95% CI, 73.8–77.6%) in the fit group to 57.3% (95% CI, 32.7–75.8%) in the severe frailty group according to the eFI. Similarly, in the VA-FI, the survival rate declined from 75.0% (95% CI, 73.2–76.8%) in the nonfrail group to 50.1% (95% CI, 36.6–62.2%) in the mild frailty group (Figure 3). The Kaplan–Meier survival curves for care-need level decline and LTCF admission stratified by frailty measures are provided in Appendix 8.

In the Cox regression analysis, the eFI categories of mild, moderate and severe frailty were not associated with all-cause mortality. However, the trend test indicated significance (P < 0.001), suggesting a tendency for mortality rates to increase as the degree of frailty progresses. For the composite secondary outcome including a decline in care-need level or mortality, the HR for severe frailty compared to fit was 2.48 (95% CI, 1.59–3.88) in the eFI. The HR for the mild frailty group was 2.49 (95% CI, 1.89–3.28) when compared to the non-frail group in VA-FI. For another composite secondary outcome, a decline in care-need level, LTCF admission, or mortality, the HR for severe frailty compared to fit was 2.59 (95% CI, 1.66–4.06) in the eFI. Similar results were observed for moderate frailty compared to non-frailty in the VA-FI (HR 2.45, 95% CI, 1.02–5.91). The trend test indicated significance, suggesting the composites’ secondary outcome rates (i.e. a decline in care-need level, mortality, or a combination of a decline in care-need level, LTCF admission and mortality) increase as the degree of frailty progresses in the eFI and VA-FI (Table 2). Unadjusted and adjusted hazard ratios for outcomes of care need level decline and LTCF admission based on eFI and VA-FI are provided in Appendix 9.

Discussion

In this study, we evaluated the impact of eFI and VA-FI on all-cause mortality, a decline in care-need level and LTCF admission using administrative databases in Japan. The risk of these outcomes was increased in those with mild, moderate and severe frailty for the eFI and in those with pre, mild and moderate frailty for the VA-FI.

The distributions of the eFI and VA-FI were right-skewed, similar to the results of previous studies [16–18]. Regarding the eFI, the proportion of nonfrail individuals was higher in the original and external validation studies [16]. Because the eFI was developed in primary care settings in the UK and the primary care doctor system was being introduced in Japanese community care settings to promote an integrated community care system, the current population exhibited a similar trend to previous original studies. The trend in the VA-FI was also lower than that reported in previous studies involving patients with cancer [18]. Given that veterans are more likely to be influenced by the burden of chronic conditions [31], the current results are reasonable.

The distribution of frailty was consistent across different cancer treatments for the eFI and VA-FI in this study. Among patients who received systemic therapy, 24. 6% had mild frailty, 6.0% moderate frailty and 0.4% severe frailty according to the eFI. These proportions of frailty are similar to the results of a previous study that evaluated the eFI among patients receiving systemic therapy, which reported that 19% had mild frailty, 7.7% moderate frailty and 3.6% severe frailty [32].

Older patients with cancer represent a large and heterogeneous group that is often underrepresented in clinical research [33], highlighting the need for new approaches to characterise their risk of adverse outcomes. Given that frailty can influence care-need level decline, LTCF admission and mortality, incorporating frailty assessments into future treatment decisions may be beneficial. For example, the eFI and VA-FI can be used to evaluate frailty prior to cancer treatment, enabling enhanced precautions or post-treatment care, such as rehabilitation for ADL decline. In addition, frailty may affect the adverse outcomes of cancer treatment [34], which suggests that assessments using a claims database can be helpful for identifying patients with frailty. Since ICD-10 codes are integrated into hospital medical records, incorporating the algorithm used to calculate the frailty index in this study into hospital record systems could facilitate rapid frailty assessments. This integration may enable future clinical application, such as timely decision-making and personalised care strategies.

The current study used two frailty measures in patients with cancer and demonstrated that those with frailty were at risk of mortality for Kaplan–Meier survival estimates. This was consistent with previous results [18]. The secondary outcomes of care-need level decline or all-cause mortality and, a decline in care-need level, LTCF admission or all-cause mortality, showed significant association in unadjusted and adjusted models. Considering the ADL trajectory in patients with cancer 1 month before death [28], the current results were reasonable. Although the eFI lacks one deficit and the VA-FI lacks three deficits in the administrative data, this study suggests the probability of their application in oncology care settings in Japan. Both eFI and VA-FI can be readily calculated from the claims database, and this would enhance decision-making regarding cancer treatments. Since fall are linked to gait and balance deficits [35], related code could have been a proxy indicator eFI and VA-FI (ICD-10 code R26; Abnormalities of gait and mobility). Additionally, the eFI included codes for fragility fractures, such as fractures of the proximal humerus (ICD-10 code S42; fracture of the shoulder and upper arm), distal radius (ICD-10 code S52; fracture of the forearm) and femoral neck and trochanteric fractures (ICD-10 code S72; fracture of the femur). These codes might have acted as substitutes for “falls” in our study, addressing the issue of missing data related to falls. This possibility has also been suggested in a previous study [36].

Strength and limitations

This study has some limitations. First, owing to the characteristics of the administrative data, patients with a gap between their cancer diagnosis and treatment initiation (e.g. those awaiting surgery) were not included in the dataset. In addition, the absence of socioeconomic data may have introduced unmeasured confounding factors. Second, owing to the nature of claims data, numerical data, such as vital signs and blood tests of each patient, were not included because they could not be linked to the hospital data. Third, ICD-10 codes were employed to measure frailty. A limitation of using ICD-10 codes is that they do not fully capture disease severity. For example, in Japanese medical care settings, ICD-10 codes associated with prescriptions are more likely to be registered. Therefore, future validation studies may be required. Fourth, this study used data from a single region, potentially leading to a regional effect that may not be generalizable to other regions. Fifth, the small sample size of patients with severe frailty may not fully reflect the variability and complexity of this high-risk population. Sixth, the omission of falls—a significant factor in frailty—may have influenced the findings. Future research exploring alternative data sources or methodologies could provide more comprehensive insights for capturing data on these patients. Lastly, we employed composite outcomes for this study. Composite outcomes allow us to understand the overall effect of the frailty index. However, they may limit the interpretation of research results. Despite these limitations, this study highlights the importance of assessing frailty using administrative data of patients with cancer.

Conclusions

This study was conducted to evaluate the eFI and VA-FI in patients with cancer using administrative databases in Japan. Applying the eFI and VA-FI, the increasing severity of frailty identifies older individuals with cancer who are at an increased risk of future a decline in care-need level, LTCF admissions and mortality. This study suggests the potential application of frailty measurements in oncology care settings.

Declaration of Conflicts of Interest

SF received grant support from the NEC Corporation. The other authors declare no conflicts of interest.

Declaration of Sources of Funding

This study was supported by the Health Labour Sciences Research Grants (grant number: 21GA2003), and the NEC Corporation (grant number: 21AB100085).

References