Development and validation of a frailty index for use in the osteoarthritis initiative

Abstract

Background

The Osteoarthritis Initiative (OAI) evaluates the development and progression of osteoarthritis. Frailty captures the heterogeneity in aging. Use of this resource-intensive dataset to answer aging-related research questions could be enhanced by a frailty measure.

Objective

To: (i) develop a deficit accumulation frailty index (FI) for the OAI; (ii) examine its relationship with age and compare between sexes, (iii) validate the FI versus all-cause mortality and (iv) compare this association with mortality with a modified frailty phenotype.

Design

OAI cohort study.

Setting

North America.

Subjects

An FI was determined for 4,755/4,796 and 4,149/4,796 who had a valid FI and frailty phenotype.

Methods

Fifty-nine-variables were screened for inclusion. Multivariate Cox regression evaluated the impact of FI or phenotype on all-cause mortality at follow-up (up to 146 months), controlling for age and sex.

Results

Thirty-one items were included. FI scores (0.16 ± 0.09) were higher in older adults and among females (both, P < 0.001). By follow-up, 264 people had died (6.4%). Older age, being male, and greater FI were associated with a higher risk of all-cause mortality (all, P < 0.001). The model including FI was a better fit than the model including the phenotype (AIC: 4,167 vs. 4,178) and was a better predictor of all-cause mortality than the phenotype with an area under receiver operating characteristic curve: 0.652 vs. 0.581.

Conclusion

We developed an FI using the OAI and validated it in relation to all-cause mortality. The FI may be used to study aging on clinical, functional and structural aspects of osteoarthritis included in the OAI.

Key points

• A frailty index (FI) quantifies the age-related physiological damages that result in the accumulation of health deficits.

• The Osteoarthritis Intiative is a resource-intensive cohort study that may be useful for studying frailty and osteoarthritis

• We developed an FI based on variables included in the Osteoarthritis Initiative and demonstrate that it increases with age and is higher among females.

• The FI was a better predictor of all-cause mortality at 146-month follow-up than a modifided frailty phenotype in this cohort

• The developed FI will be useful in studying the impact of frailty on osteoarthritis

Introduction

Frailty is a state of increased vulnerability to adverse outcomes, such as frequent use of healthcare services, disability, risk of falls and osteoporotic fractures [1–3]. Frailty arises from age-related damages across various physiological systems that contribute to the accumulation of medical and social issues, which can interact and exacerbate one another [4]. Among older adults, osteoarthritis and particularly knee osteoarthritis, is the leading cause of disability [5]. While the prevalence of osteoarthritis and frailty level increase with age and overlap with mobility-related factors, considerably less is known regarding the link between frailty and osteoarthritis development.

The Osteoarthritis Initiative (OAI) is a multi-center, North American longitudinal observational study in adults 45–79 years of age designed to enable researcher with the tools to better understand osteoarthritis [6–8]. Since its inception in 2001, the publicly available cohort has included ~5,000 participants with longitudinal clinical assessments, patient reported outcomes, and detailed radiographs and magnetic resonance imaging. Using frailty phenotype in the European Project of Osteoarthritis, the odds ratio of exhibiting frailty is three times higher in those with osteoarthritis versus those without [9]. The existing work investigating osteoarthritis and frailty using the OAI is based on physical frailty [10, 11], using a modified version of the frailty phenotype [12], but a well-defined tool based on the outcomes collected in OAI has not been developed. An alternative measurement of frailty, a frailty index (FI), considers the multi-dimensional aspects of aging and accumulation of health deficits [13] and has been applied extensively in clinical and research settings [2, 14–17]. An FI condenses 30+ variables that reflect multiple aspects of health (e.g. symptoms, signs, diseases and functional limitations) into one ratio to represent the health of that individual [18]. FIs that are typically higher in females versus males [19] have been validated with mortality records [16, 20, 21] and are associated with numerous diseases (e.g. cancer, cardiovascular disease, diabetes, dementia, rheumatoid arthritis, etc.) [22–24]. FIs have been developed using retrospectively collected data from cohort studies [19, 25, 26], but the differences in the type and number of items measured prevent a uniform FI being applicable to all cohort studies. The development and validation of an FI based on the OAI would be advantageous for studying the link between osteoarthritis and frailty using the well-detailed interdisciplinary osteoarthritis-related outcomes collected.

The purpose of our study was to: (i) develop a deficit accumulation FI for the OAI, (ii) examine its relationship with age and compare values between sexes, (iii) validate the FI versus all-cause mortality and (iv) compare this association with mortality with a modified frailty phenotype.

Methods

The OAI

We performed secondary analysis of the Osteoarthrosis Initiative baseline and follow-up data (up to 146 months). The Osteoarthrosis Initiative is a public–private partnered multicenter, longitudinal and prospective cohort study sponsored jointly by the National Institutes of Health and four pharmaceutical companies (GlacoSmithKline, Merck Sharp & Dohme, Inc., Novartis, Pzizer Inc.) whose aim is to generate a public domain resource to evaluate osteoarthritis biomarkers as indicators of disease onset and progression [27]. The cohort consists of 4,796 participants and is in its 14th year of follow-up. To be included, participants must be (i) between 45 and 79 years of age, and (ii) with or at risk for symptomatic femoral-tibial knee osteoarthritis. Exclusion criteria include: (i) inflammatory arthritis, (ii) contraindication to 3 T MRI or (iii) bilateral end-stage knee osteoarthritis. The FI is commonly used in middle-aged and older adults. As a matter of policy, the Nova Scotia Health Research Ethics Board does not review research involving secondary analyses of datasets that contain deidentified individual-level data.

Development of FI

An FI approach operationalizes frailty as an accumulation of health deficits, which confer susceptibility to adverse health outcomes [18]. Self-reported comorbidities and functional assessments were identified in the available health measures within the OAI that could be used in an FI. Items that were determined at baseline were considered for our FI. Guidelines for retrospectively constructing an FI were followed [26], and previously developed FIs from cohort studies [19, 25] were used as guides for selecting items.

All candidate FI items were recoded so that the absence of the deficit had a score of 0 and the presence of the deficit resulted in a score of 1 (worse frailty level). Variables with more than two response options were coded as a proportion of the complete deficit. Some items were calculated from available measures (e.g. pulse pressure from systolic minus diastolic blood pressure), as done previously [20]. A data dictionary of variables included in the FI is available in Supplemental File 1.

Candidate items were screened following a standard protocol [18] to ensure they were (i) health-related, (ii) age-associated, (iii) representative of a broad range of health systems, (iv) neither overly rare (present in less than 1% of participants) nor overly common (present in greater than 80% of participants), (v) comprised of fewer than 5% missing values and (vi) not highly correlated with other candidate items (r < 0.95). Not being associated with age was the most common reason for being excluded (see section Results). Items directly related to the presence or severity of osteoarthritis, disability and fall history were not considered for inclusion in the FI due to these variables representing potential outcomes that may be used to examine the relationship with frailty in future studies.

An FI score was calculated as the ratio of deficits present to the total deficits considered, resulting in a continuous score of 0 to 1 representing the continuum of total fitness to total frailty. While frailty indices are meant to be considered as continuous score, for illustrative purposes where a categorical variable is required (i.e. Kaplan–Meier survival analysis), degree of frailty was also categorized as follows: non-frail (FI ≤ 0.10), very mild (0.1 < FI ≤ 0.2), mild frailty (0.2 < FI ≤ 0.3) and moderate to severe frail (FI > 0.3) [19, 28].

Frailty phenotype

The frailty phenotype considers frailty as a clinical syndrome and is quantified by five items representing physical frailty, including unintentional weight loss, weakness or poor handgrip strength, exhaustion, slow walking speed and low physical activity [12]. A modified frailty phenotype was calculated as previously described [10] resulting in a score out of five where 0 denotes no frailty, 1–2 denotes an pre-frail status and 3–5 denotes frailty. This modified phenotype was previously utilized in the OAI [10, 11]. The criteria considered included weight loss, exhaustion, weakness, gait speed and physical activity. Self-reported weakness was used instead of grip strength, and it was unclear whether weight loss was unintentional. Included variables are detailed in Table 1. Only participants who had complete frailty phenotype scores were included in analysis (N = 4,149; 86.5%).

Statistical analysis

Statistical calculations were performed in SPSS 28 (IBM, Armonk, NY, USA). Demographics of the sample were expressed as a percentage, and the mean FI for each sub-population were determined. A one sample binomial test was used to evaluate the distribution of sex within the sample whereas a one-sample t-test was used to evaluate the distribution of age. A 99% estimate was calculated to determine the limit of near-lethal frailty in this dataset.

The distribution of FI score and mean FI relative to age were plotted. Mean FI scores were compared between sexes, age groups, cohort and races with either independent sample t-test or two-way ANOVA with post-hoc Bonferroni tests where appropriate.

To examine the relationship between baseline frailty (FI and frailty phenotype) and all-cause mortality, Kaplan–Meier curves were generated to evaluate survival based on the Mantel-Cox log-rank test. Multivariate Cox regression with age and sex as covariates was conducted on participants that had both a valid frailty phenotype score (all five items) and FI score (<20% missingness). Akaike information criteria (AIC) was calculated to assess goodness of fit for both Cox regression frailty models (AIC = −2logLikelihood + 2 k, where k = degrees of freedom). Receiver operating characteristic (ROC) curves were generated to compare the ability of the FI and frailty phenotype to predict death at 146 months. The area under the curve (AUC) was determined for the FI and frailty phenotype. An α = 0.05 was considered statistically significant.

Results

Development of an FI

Screening of 59 potential FI items derived from available clinical measures resulted in a 31-item FI that includes variables related to comorbidity, physical function, mental health and vital signs (Table 1). The excluded variables, specific wording for each included variable and the scoring are presented in Supplemental Table 1 and Supplemental File 1. Reasons for variable exclusion included prevalence not increasing with age (n = 27), being too rare (n = 3) and high missingness (n = 1). Three variables were retained due to their association with age in larger datasets (everything feels like an effort, could not “get going” and loneliness) [20, 25]. Only 41/4,796 participants (<1%) of participants were missing ≥20% of FI items, resulting in their exclusion from analysis.

Impact of age and sex

Of the 4,755 individuals for which an FI score was calculated, the distribution of sex was 58.5% female (95% CI: 0.57–0.60) and the average age was 61.2 years of age (95% CI: 60.93–61.45). The mean FI score was 0.155 (95% CI: 0.153–0.158) corresponding to a population living with very mild frailty (Table 2). FI scores were greater in females (P < 0.001), different between all ages grouped by decade (all P < 0.001, except non-frail ~ very mild frailty p = 0.001), and between the three cohorts (all P < 0.001; incidence, progression and control groups). FI scores were lower in individuals who self-identified as Caucasian or Asian versus other non-white (P = 0.002 and P = 0.008, respectively) or African American or Black (both, P < 0.001; see Table 1). The FI score distribution skewed to the right (Supplemental Figure 1) and increased with age in both sexes (Table 2). The observed 99% FI limit, considered a near lethal level of frailty, was approximately 0.42 in our participants (FI: 0.42 in males, 0.41 in females).

Frailty phenotype

The frailty phenotype could be calculated for 4,149 individuals with a valid FI score. The average frailty phenotype score (out of 5) was 0.86 (95% CI: 0.84–0.89) and was greater in females than males (P < 0.001). Participants in their 40s (P = 0.045), 50s (P < 0.001) and 60s (P < 0.001) had a lower score than participants in their 70s, but score was not different between individuals in their 40s–60s (all, P = 1.00). Frailty phenotype scores were significantly greater in the progression than incidence and control cohorts, as well as the incidence compared to the control cohort (all, P < 0.001). Frailty phenotype score was lower in participants who identified as Caucasian (P < 0.001) or Asian (P = 0.001) compared to Black or African Americans.

Regarding the 647 individuals excluded for missing either FI or frailty phenotype score, their FI was 0.20 (95% CI: 0.19–0.20), whereas their frailty phenotype score was 1.23 (95% CI:1.13–1.33) based on the available information. The FI and phenotype scores of participants excluded due to missingness was greater than included individuals (both, P < 0.001). Missingness was predominantly driven by missing follow-up weight (550/647), a variable required to calculate the weight-loss item in the modified frailty phenotype.

Validation with all-cause mortality

Figure 1 illustrates the association of the baseline degree of frailty adjusted for sex with all-cause mortality censored at the time of the last reported death at 146 months after baseline for participants where both FI and frailty phenotype scores could be calculated. In this population and during this period, 264 deaths occurred. Log-rank tests indicated that the distribution of deaths across degrees of frailty measured by both FI and frailty phenotype are significantly different (P < 0.001). Multivariable adjusted hazard ratios were calculated for sex (female), age (1-year increase), and baseline degree of frailty (0.01 increase in FI and 1-point increase in frailty phenotype score) (Table 3). In both models, being female decreased and age increased risk of all-cause mortality (both P < 0.001). An increase in both FI and frailty phenotype scores increased risk of all-cause mortality (both P < 0.001).

Comparison of FI and frailty phenotype

The Cox model including FI (with covariates age and sex) had a lower AIC and therefore had better model fit than the model including frailty phenotype (Table 3). AUC analysis of the ROC curves (Figure 2) indicated that the FI (AUC = 0.652 ± 0.018 standard error (SE); 95%CI: 0.618–0.687) was a better predictor of all-cause mortality at 164 months than frailty phenotype (AUC = 0.581 ± 0.018 SE; 95%CI: 0.545–0.617).

Discussion

Consistent with the objective of this study, we created an FI that met the standard criteria [18] of at least 30-items and validated it against mortality records in the OAI, whereby participants with higher baseline frailty levels were more likely to die in the 146-month longitudinal follow-up. The predictive capabilities of the developed FI were better than the currently used modified version of the frailty phenotype in the OAI.

Inherent to the items included in the FI, frailty was positively correlated with chronological age among males and females. Consistent with other self-report based FIs [19, 25], females report greater frailty levels than males at the same age on frailty questionnaire-based frailty indices, whereas indices based on laboratory assessments do not consistently observe a sex difference [29]. While the male–female frailty-survival paradox is not completely understood, females may be more likely to discuss physical and psychological symptoms that may lead to earlier diagnoses and better health care management decisions than males [30, 31], as evident by the higher rates of healthcare utilization among females versus males [32]. Males may be less likely to report minor and major health issues that might subsequently develop into more serious healthcare issues. Notably, this notion is consistent with males exhibiting a higher risk of mortality in this OAI cohort in our age-adjusted model. When age-adjusted, prior evidence in non-osteoarthritis-specific groups similarly indicates that males tend to exhibit higher mortality rates than females [33]. In general, females exhibit a higher prevalence of osteoarthritis risk and overt condition than males [34], which negatively impacts physical function and results in disability. While there were more females included in this cohort than males, the Kaplan–Meier curves relating all-cause mortality to frailty level tend to be steeper in males versus females. In our age- and sex-adjusted analyses, mild and moderate-to-severe levels of frailty were associated with a greater risk of all-cause mortality relative to the non-frail group, validating the FI as a useful tool for distinguishing an important health outcome.

The OAI is uniquely positioned to study persons at risk for or have osteoarthritis. While this cohort has investigated the impact of aging on knee pain [35], recurrent falls [36] and radiographic images of knee osteoarthritis [36], the consideration of “biological age” [37] as reflected in the accumulation of health deficits has not been considered. Accordingly, the development of an FI for this cohort study was crucial for utilizing the full extent of the rigorous data collected as it relates to studying aging and osteoarthritis. This may be utilized to understand how frailty may impact joint health and separately in among the three cohorts included in the initiative over time. Similarly, future work investigating the relationship between frailty and quality of life outcomes (e.g. health-related quality of life, disability, etc.) among people with osteoarthritis is now possible. As well, the OAI has conducted a rigorous imaging protocol [38] in a relatively large sample of participants over a long period of time as well as the integration of accelerometry outcomes in some waves [6] that are otherwise not often measured in other cohort studies. The included FI may be used in conjunction with the unique outcomes measures to answer novel questions linking frailty, physical function and MRI-determined knee structure. Future studies implementing our validated FI in the OAI to better understand factors that contribute to osteoarthritis are needed.

The frailty phenotype is a well-established measure of physical frailty that characterizes frailty as a syndrome [12]. We demonstrated that the developed FI predicted mortality at follow-up better than a modified version of the frailty phenotype that has been previously utilized in the OAI [10, 11]. These observations are consistent with prior work in the Cardiovascular Health Study, whereby a 48-item FI was more precise at predicting the chance of death than the frailty phenotype [39]. Similarly, in comparing a 46-item FI and modified frailty phenotype in the National Health and Nutrition Examination Survey, almost all individuals determined as “frail” by the phenotype were dependent for instrumental activities of daily living, but the FI was better at distinguishing individuals at very-mild and mild frailty levels [25]. The superiority of the FI in our study may be attributed to the modified variables used in the frailty phenotype. The phenotype used in previous OAI studies [10, 11] is a creation of the available information collected and is meant to provide proxy of the five original phenotype items, but these questions may not truly reflect the original, particularly for unintentional weight loss or grip strength. The comparisons herein are with the modified phenotype and may differ if the original phenotype items were included in the OAI. Nevertheless, the developed FI addresses these concerns and may be useful for operationalizing frailty in the OAI.

The study is strengthened by the characterization of the impact of age and sex, as well as the validation with relatively long-term mortality records. While mortality records were examined up until 146 months (~12 years), a longer tracking of mortality records would strengthen our ability to detect differences between frailty groups on all-cause mortality (e.g. non-frail vs. vulnerable groups). Related, the specific cause of mortality and exact time of death was not assessed. The ~12% of participants for who a phenotype could not be calculated for based on missing follow-up weight measurement tended to be frailer. It should be acknowledged that our FI was based primarily on self-report outcomes with some physical function tests, with self-report measures possibly subject to variable answers for self-perception-based questions. Including blood or objective examination outcomes may improve the characterization of frailty [20, 21, 40].

In conclusion, the FI developed from age-related variables and validated using mortality records may help researchers interested in frailty as it relates to osteoarthritis pursue novel research questions.

Acknowledgements:

This manuscript was prepared using the OAI public use data set and does not necessarily reflect the opinions or views of OAI investigators, the NIH or the private funding partners. The authors wish to thank the participants, investigators and research staff who contributed to the OAI, making this excellent dataset possible. MWO and SPM were supported by a CIHR Post-Doctoral Fellowship Award (#181747 and #494621) and a Dalhousie University Department of Medicine University Internal Medicine Research Foundation Research Fellowship Award.

Declaration of Conflict of Interest:

KR has asserted copyright of the Clinical Frailty Scale and KR and OT have asserted copyright of the Pictorial Fit-Frail Scale which are both made freely available for education, research, and not-for-profit health care. Licences for commercial use of the Clinical Frailty Scale and the Pictorial Fit-Frail Scale are facilitated through the Dalhousie Office of Commercialization and Industry Engagement. Additionally, KR is the co-founder of Ardea Outcomes (until 2021 DGI Clinical) which has held contracts with Novartis, among other pharma and device manufacturers, within the last three years.

Declaration of Sources of Funding:

This research was undertaken, in part, thanks to funding from the Canada Research Chairs Program.

References

Author notes