Abstract
Objective. To construct a Frailty Index (FI) as a measure of vulnerability to adverse outcomes among patients with systemic lupus erythematosus (SLE), using data from the Systemic Lupus International Collaborating Clinics (SLICC) inception cohort.
Methods. The SLICC inception cohort consists of recently diagnosed patients with SLE followed annually with clinical and laboratory assessments. For this analysis, the baseline visit was defined as the first study visit at which sufficient information was available for construction of an FI. Following a standard procedure, variables from the SLICC database were evaluated as potential health deficits. Selected health deficits were then used to generate a SLICC-FI. The prevalence of frailty in the baseline dataset was evaluated using established cutpoints for FI values.
Results. The 1683 patients with SLE (92.1% of the overall cohort) eligible for inclusion in the baseline dataset were mostly female (89%) with mean (SD) age 35.7 (13.4) years and mean (SD) disease duration 18.8 (15.7) months at baseline. Of 222 variables, 48 met criteria for inclusion in the SLICC-FI. Mean (SD) SLICC-FI was 0.17 (0.08) with a range from 0 to 0.51. At baseline, 27.1% (95% CI 25.0–29.2) of patients were classified as frail, based on SLICC-FI values > 0.21.
Conclusion. The SLICC inception cohort permits feasible construction of an FI for use in patients with SLE. Even in a relatively young cohort of patients with SLE, frailty was common. The SLICC-FI may be a useful tool for identifying patients with SLE who are most vulnerable to adverse outcomes, but validation of this index is required prior to its use.
Systemic lupus erythematosus (SLE) is an autoimmune disease with diverse manifestations and an unpredictable clinical course1. Despite advances in diagnosis and treatment2, many patients with SLE accumulate organ damage3, and the mortality risk remains high4,5. Given this variability in health trajectories, it would be advantageous to identify those patients with SLE at increased risk for adverse outcomes. However, instruments that accurately predict longterm outcomes in SLE are limited6.
In geriatric medicine7, and increasingly in other disciplines8,9,10,11,12,13, differences in susceptibility to adverse outcomes are quantified using the construct of frailty, defined as a state of increased vulnerability due to degradation of homeostatic mechanisms, resulting in diminished ability to respond to physiologic stressors14. Although often linked to advanced age, frailty can be observed across the life course15, including among individuals with acquired vulnerability states16,17.
Two different conceptual approaches inform the measurement of frailty18. One approach uses rules-based tools, where specific criteria must be met to classify an individual as frail18. The most common example of this approach is the Fried frailty phenotype, which defines frailty as a clinical syndrome with at least 3 of 5 specific health deficits: weight loss, exhaustion, physical inactivity, slow walking speed, and reduced grip strength19.
The second approach to measuring frailty is the Frailty Index (FI)20, which conceptualizes frailty as a loss of physiologic reserve arising from the accumulation of health deficits across multiple systems21. Individuals who possess few deficits are considered relatively fit, while those with an increasing number of health problems are considered increasingly frail18. Prior studies have consistently shown an association between higher FI values and increased risk of negative health outcomes, including hospitalizations, morbidity, and mortality15,22,23,24. Although used in many different clinical contexts22,23,25, the deficit accumulation approach has yet to be applied in SLE.
Health deficits in SLE may occur because of the disease, its treatment, other comorbidities, or aging. Evaluating frailty through deficit accumulation could improve our understanding of the heterogeneous health outcomes in SLE. The aim of the present study was to use the deficit accumulation approach to construct an FI as a novel health measure in SLE, using data from an international inception cohort. Future studies are required to validate the Systemic Lupus International Collaborating Clinics (SLICC)-FI, including its predictive validity for adverse health outcomes.
MATERIALS AND METHODS
Data source
This was a secondary analysis of longitudinal data from the SLICC inception cohort. SLICC comprises 52 investigators at 43 academic centers in 16 countries. From 1999 to 2011, a cohort of 1826 recently diagnosed patients with SLE was recruited from 31 SLICC sites in Europe, Asia, and North America. Patients were enrolled within 15 months of SLE diagnosis, based on ≥ 4 revised American College of Rheumatology (ACR) classification criteria for SLE26. At enrollment and annually thereafter, data were collected per a standardized protocol, submitted to the coordinating centers at the University of Toronto (Toronto, Ontario, Canada) and Dalhousie University (Halifax, Nova Scotia, Canada), and entered into centralized databases. The study was approved by the institutional research ethics boards of the Nova Scotia Health Authority central zone (#1020396) and of participating centers in accordance with the Declaration of Helsinki’s guidelines for research in humans. All participants provided written informed consent.
Clinical assessments
Demographic features included age, sex, race/ethnicity, geographic location, and years of post-secondary education. Corticosteroid, antimalarial, and immunosuppressive use was documented. Medical comorbidities prior to SLE diagnosis and between followup visits were recorded. The revised ACR classification criteria for SLE26 and neuropsychiatric events27 were documented at enrollment and between followup visits28. SLE disease activity was measured using the SLE Disease Activity Index 200029, cumulative organ damage using the SLICC/ACR Damage Index (SDI)30, and health-related quality of life using the Medical Outcomes Survey Short Form-36 (SF-36)31. Blood pressure (in mmHg), height (m), and weight (kg) were also recorded.
Laboratory data
Investigations for the assessment of SLE disease activity and organ damage were performed at each visit: anti-dsDNA, C3 and C4, serum creatinine, urinalysis, fasting glucose, lipid profile, and inflammatory markers (erythrocyte sedimentation rate and C-reactive protein).
Standard procedure for FI construction
An FI can be constructed from any existing health dataset using a standard procedure described by Searle, et al (Table 1)20. These methods have been shown to be valid and reliable15,22,23,32,33,34. Briefly, potential health deficits are first identified. A health deficit is any symptom, physical sign, disease process, functional impairment, or laboratory abnormality that is acquired, associated with adverse health outcomes, and associated with chronological age20,35. If deficits are either too infrequent or too common, they are unlikely to provide meaningful information in an FI, and are respectively combined or excluded20,35. Finally, if a single item is missing values for > 5% of individuals, it is excluded20,35.
The totality of health deficits in an FI must represent several organ systems. Of note, frailty not only measures irreversible damage but also measures an individual’s potential for recovery. Therefore, an FI also includes measures of function and mobility20,35. Finally, an FI requires a minimum of 30–40 health deficits to produce stable and precise estimates of frailty22,33,35,36.
Each health deficit is assigned a score from 0 to 1, with 0 representing no deficit and 1 representing the deficit fully expressed20. Health deficit scores are combined to produce an FI score between 0 and 1, calculated as the sum of deficit scores for an individual divided by the total number of deficits considered20,35.
Establishing a baseline dataset for SLICC-FI construction
Given the importance of the SDI and the SF-36 for the construction of the SLICC-FI, each patient’s baseline visit was defined as the first at which both an SDI and an SF-36 were completed. Patients were excluded if they had never had an SDI recorded, never had an SF-36 recorded, or never had both instruments recorded at the same visit.
Selecting health deficits for the SLICC-FI
Potential health deficits were evaluated using the criteria in Table 1. Variables judged to be innate, as opposed to acquired, were excluded. Age-relatedness was assessed by reviewing the literature to determine whether each variable is observed more frequently with increasing age in SLE populations. While a health deficit should generally increase in prevalence with increasing age, this relationship may not exist for all deficits, in part because of survivor effects20. Variables were retained in the SLICC-FI even if there was attenuation of this relationship at advanced ages.
The association of each health deficit with increased risk of adverse health outcomes in SLE was also determined through literature review. Variables not clearly associated with adverse outcomes were excluded. If literature specific to SLE was not available, evidence from the general population was sought and extrapolated to SLE populations.
Next, variables were evaluated for duplications. Items were excluded from the SLICC-FI if they represented constructs that were already better accounted for by another variable in the database. Where appropriate, multiple related variables were combined to produce single health deficits. Variables whose prevalence in the dataset was < 1% were excluded if there were no similar deficits with which they could be reasonably combined. Finally, variables were excluded if their prevalence in the dataset was > 80%, or if there were missing values for > 5% of observations.
Coding of individual health deficits for the SLICC-FI
Binary variables were assigned a score of 0 (absence of the deficit) or 1 (presence of the deficit). Ordinal variables were coded by converting the number of possible ranks into equally spaced scores ranging from 0 to 1. Continuous variables were coded using established cutpoints from the SLE literature.
SLICC-FI calculation
Individual health deficit scores were combined to produce a SLICC-FI score for each patient. For example, with 48 health deficits in the SLICC-FI, an individual in whom 24 of these deficits were fully present would have a SLICC-FI score of 24/48 = 0.50. SLICC-FI scores were not calculated for individuals with missing values for > 20% of health deficits36.
Statistical analysis
Descriptive statistics were calculated for demographic and clinical characteristics. For quantitative variables, measures of central tendency (means and medians) and dispersion (SD and interquartile ranges) were reported, as appropriate. For categorical variables, absolute and relative frequencies were reported. Descriptive statistics were calculated for SLICC-FI values and the distribution of SLICC-FI scores was visualized. Using cutpoints derived in the general population15,37,38, we classified patients as robust (SLICC-FI ≤ 0.03), relatively less fit (0.03 < SLICC-FI ≤ 0.10), least fit (0.10 < SLICC-FI ≤ 0.21), or frail (SLICC-FI > 0.21), and reported the prevalence of frailty with 95% CI.
To evaluate for bias due to varying SLE disease durations, analyses were repeated in patients with baseline assessments within 2 years of SLE diagnosis. Finally, to evaluate the effect of a given variable on the SLICC-FI, an iterative, resampling procedure was used20,39. One hundred iterations were performed in which each iteration calculated SLICC-FI values using 80% of health deficits and then reevaluated the descriptive statistics of the SLICC-FI. Data analysis was conducted using STATA-IC Version 14 (StataCorp).
RESULTS
Patient characteristics
There were 1683 patients (92.2% of cohort) with study visits at which both the SDI and SF-36 were recorded. The first such visit was included in our baseline dataset, and for most patients this occurred early in their disease course [1390/1683 patients (82.6%) within 2 yrs of SLE diagnosis]. Demographic and clinical characteristics are shown in Table 2.
Excluded patients
There were 143 patients (7.8% of cohort) excluded, most (n = 90) of whom had a single visit within 6 months of diagnosis, which precluded scoring the SDI. Other reasons for exclusion were no SF-36 recorded (n = 32), no SDI recorded (n = 6), and no visit with both SF-36 and SDI recorded (n = 15). At enrollment, excluded patients were similar to non-excluded patients in age, sex, education, SLE disease activity, and SLE manifestations (data not shown). Hispanic patients were more likely to be excluded compared to patients of other races/ethnicities, largely owing to higher rates of missing SF-36 data and early loss to followup (data not shown).
SLICC-FI construction: selection of health deficits
Of the 222 candidate variables identified as potential health deficits (Figure 1), 18 were excluded for failing to meet the first 3 health deficit criteria (Table 1) and 46 were excluded as duplicates. The remaining 158 SLICC variables were used to construct health deficits. There were 36 variables that were directly converted into 36 health deficits. In other cases, several variables representing varying aspects of the same condition were combined to create a single health deficit. For example, the health deficit “coronary artery disease,” defined as “any history of angina, myocardial infarction, or coronary revascularization ever,” used information from 12 different variables. Thus, information from the remaining 122 variables was combined to form 32 health deficits. In total, 68 distinct health deficits were generated for further evaluation. Of these, 9 were excluded owing to low baseline prevalence (< 1%), one owing to high baseline prevalence (> 80%), and 10 because of missing data in > 5% of observations. Forty-eight health deficits met all required criteria for inclusion in the SLICC-FI.
SLICC-FI construction: health deficit coding
The majority of SLICC-FI health deficits were binary, with values of either 0 or 1. Examples included “diabetes” and “active nephritis.” Ordinal health deficits included those derived from the SF-36. For example, for “self-rated health,” the self-reported SF-36 responses were coded as “excellent = 0,” “very good = 0.25,” “good = 0.5,” “fair = 0.75,” and “poor = 1.” For continuous variables, existing literature was used to define clinically significant cutpoints. For example, the “body mass index” (BMI) cutpoints were derived from published data regarding the association between BMI and mortality in the general population (BMI 18.5–24.9 kg/m2 = 0; BMI 25–29.9 kg/m2 = 0.5; BMI < 18.5 or ≥ 30 kg/m2 = 1)40.
The SLICC-FI
Of the 48 health deficits in the SLICC-FI (Table 3, and Supplementary Table 1, available with the online version of this article), 14 were related to organ damage, before or after the diagnosis of SLE (e.g., congestive heart failure and chronic kidney disease). Another 14 deficits reflected active inflammation (e.g., serositis and inflammatory arthritis), while 6 items reflected comorbid conditions (e.g., hypertension and obesity). Finally, there were 14 variables related to function, mobility, health attitude, and mental health.
SLICC-FI values
SLICC-FI scores were calculated for 1682 patients in the baseline dataset. In 1 patient, a SLICC-FI score could not be calculated because of missing data for 12 (25%) health deficits. The distribution of baseline SLICC-FI scores (Figure 2) ranged from 0 to 0.51, with a median (IQR) of 0.16 (0.11–0.22) and a mean (SD) of 0.17 (0.08).
Based on SLICC-FI values > 0.21, 27.1% (95% CI 25.0–29.2) of patients with SLE were classified as frail at baseline (Table 4). The prevalence of frailty increased with increasing age, from 19.3% (95% CI 16.4–22.6) among patients < 30 years of age, to 28.1% (95% CI 24.6–31.8) for patients aged 30–45 years, and 38.5% (95% CI 33.7–43.5%) among patients aged 45 years or older. Very few patients (n = 28; 1.7%) were classified as robust (SLICC-FI ≤ 0.03). These individuals were combined with the relatively less fit patients (0.03 < SLICC-FI ≤ 0.10) into a single category (“relatively fit”).
Compared to the relatively fittest patients, those who were classified as frail were older, less well educated, and more likely to be current smokers (Table 4). There was a trend toward a higher prevalence of frailty among women (27.5%; 95% CI 25.3–29.9%) compared to men (23.7%; 95% CI 17.8–30.4%). There was also a trend toward shorter SLE disease duration among frail patients when compared to relatively fit patients.
Sensitivity analysis
Our results were similar when only patients with baseline assessments within 2 years of SLE diagnosis (n = 1390) were considered (data not shown). Finally, SLICC-FI scores showed little sensitivity to which health deficits were included. In 100 iterations in which the SLICC-FI was recalculated using 80% of the 48 total deficits selected at random, the descriptive statistics and distribution of SLICC-FI scores were largely unchanged.
DISCUSSION
In this secondary analysis of data from the SLICC inception cohort, we have demonstrated the feasibility of constructing the first FI for patients with SLE. We have described the process for constructing the SLICC-FI in detail, including the selection of health deficits, and how these deficits were operationalized to calculate SLICC-FI values. We found a high prevalence of frailty among patients with SLE, the majority of whom were early in their disease course. A similar approach can be applied to investigate frailty in other SLE cohorts. However, additional studies are needed to demonstrate the validity of the SLICC-FI, including its association with the risk of future adverse health outcomes.
The process for constructing the SLICC-FI has many strengths. First, we followed a standard protocol20 to derive health deficits and their cutpoints from existing instruments that are well validated in SLE. With 48 items, the number of health deficits in the SLICC-FI is sufficient to provide stable and reliable estimates of frailty22,33,35,36. Last, the deficits in the SLICC-FI cover multiple organ systems and embrace both fixed and reversible health domains20.
That many small effects can aggregate to produce larger ones is well recognized in other disciplines. Applying this principle in medicine allows for the cumulative effect of multiple small deficits, which individually might not be statistically or clinically significant41. Some may be concerned about redundancy within the SLICC-FI and desire a more parsimonious list of items. However, each item contributes additional information, regardless of the correlation between them. One strength of the deficit accumulation approach to quantifying vulnerability is its ability to embrace the complexity of human systems, by placing less emphasis on specific items, and instead focusing on the overall effect of multiple health problems18. Indeed, similar to the results of prior work in other populations20,37, our sensitivity analysis demonstrated that SLICC-FI scores were not driven by a small number of specific variables but reflected the global effect of deficit accumulation.
The relationships that exist between deficits within the SLICC-FI are critical to its performance20. For example, the equal weighting of transient ischemic attacks and debilitating strokes in the “cerebrovascular disease” health deficit may appear to lack face validity, because these events clearly differ in their effect on overall health. However, an individual with a disabling stroke is more likely to have additional deficits related to their functional performance that will be reflected in their SLICC-FI score. Thus, including deficits related to functional status ensures that the health effect of different medical problems is accurately represented. Further, the potential reversibility of such deficits means that individuals may transition in and out of a frail state during followup, enabling the SLICC-FI to record improvements in a patient’s status over time and distinguishing this instrument from the SDI30. Future work will examine the trajectories of SLICC-FI values during followup. Given that frailty is potentially treatable7, the SLICC-FI may be useful as an outcome measure for future intervention studies.
An alternative conceptual approach to the measurement of frailty is the Fried frailty phenotype19, which was recently evaluated in a prevalent cohort of 152 women with SLE42. In this study, 20% of the sample was classified as frail42. The presence of frailty was associated with increased risk of functional decline and mortality42, emphasizing its relevance in SLE. However, the authors also found that 2 of the 5 components of the frailty phenotype, as defined in geriatric medicine, had limited utility in SLE42, suggesting that measures with more relevance in SLE may be needed to better quantify frailty in this population.
There are several other challenges associated with applying the frailty phenotype in SLE that are overcome using the deficit accumulation approach. First, the frailty phenotype requires physical performance data18,19,42 that is not routinely collected in SLE and is unavailable in the SLICC inception cohort. In contrast, the variables in the SLICC-FI are derived from existing validated instruments that are commonly used in SLE cohorts and rheumatology clinics, allowing the SLICC-FI to be easily implemented in other clinical and research settings. Another limitation of the frailty phenotype is its lack of granularity, because individuals are assigned to 1 of 3 risk categories18,19. Meanwhile, the SLICC-FI identifies a full spectrum of vulnerability, and studies using this approach in other populations have demonstrated a dose-response relationship between FI values and risk of adverse outcomes20,22,23,33. Finally, with only 5 variables included in the frailty phenotype, modifying how the phenotypic criteria are defined can alter the prevalence estimates for frailty considerably43. In contrast, the properties of the FI remain remarkably consistent regardless of the number or type of variables included20,22,23,24,33. While the FI and the frailty phenotype have shown reasonable agreement in geriatric populations34,37, it is unclear whether this correlation exists in SLE. Future work should investigate agreement between the SLICC-FI and the Fried phenotype for the identification of frailty in SLE.
In our study, 27.1% of patients were classified as frail. This is higher than expected for similarly aged individuals in the general population15,32,44. For example, among patients with SLE who are < 30 years of age, 19.3% were classified as frail, compared with an estimated frailty prevalence of 2.0% among Canadian adults in the same age group15. SLICC-FI values (mean FI 0.17) were substantially lower than FI scores reported in other clinical cohorts, including patients with human immunodeficiency virus (mean FI 0.31)22 and systemic sclerosis (mean FI 0.33)23. This could be partially explained by the higher mean age in these other cohorts, as deficits accumulate with increasing age35. Overall, our findings support those of prior studies in non-SLE populations that have demonstrated older age, female sex, lower educational attainment, and cigarette smoking to be associated with higher prevalence of frailty15,20,33.
There is biological plausibility to our findings. The link between chronic inflammation and frailty is well established, with elevated markers of systemic inflammation observed among frail older adults compared with those who are not frail45. Further, certain inflammatory cytokines, such as interleukin 6, have been implicated in the pathogenesis of both frailty and SLE45,46. While more work is required to fully elucidate the role of immune dysregulation in the development of frailty, this could represent a potential mechanism for accelerated aging in SLE.
Our study has important limitations. Because of missing data, we were unable to calculate SLICC-FI scores at enrollment for some patients. Despite this, 82.6% of eligible patients had their baseline assessment for SLICC-FI construction within 2 years of SLE diagnosis. Second, our sample size is small compared with some other FI studies15,20,33, but is still sufficient for FI construction23. Third, we used FI cutpoints derived from general population samples to estimate the prevalence of frailty in our dataset15,37,38. It is possible that a different cutoff for SLICC-FI scores should be used to define phenotypic frailty in SLE. This is an area for future research. Last, we have constructed the SLICC-FI in a cohort of relatively young, recently diagnosed patients with SLE. It remains unclear whether these findings can be generalized to older patients with longstanding SLE. Prior to use, validation of the SLICC-FI is required, including external validation in prevalent SLE cohorts and confirmation of its association with the risk of future adverse health outcomes.
Evaluating frailty through deficit accumulation provides a novel approach to the quantification of vulnerability among patients with SLE. We identified a high prevalence of frailty among patients with SLE, which warrants additional investigation. The SLICC-FI requires validation prior to its use as a tool to identify patients with SLE who are at increased risk for adverse outcomes.
ONLINE SUPPLEMENT
Supplementary material accompanies the online version of this article.
Footnotes
Funding for this work was provided by the Nova Scotia Health Authority Research Fund. Dr. Bae’s work was supported in part by NRF-2017M3A9B4050335, Republic of Korea. Dr. Gordon’s work was supported by Lupus UK and the NIHR/Wellcome Trust Clinical Research Facility. The Montreal General Hospital Lupus Clinic is partially supported by the Singer Family Fund for Lupus Research. Dr. Clarke holds The Arthritis Society Chair in Rheumatic Diseases at the University of Calgary. Dr. Fortin presently holds a tier 1 Canada Research Chair on Systemic Autoimmune Rheumatic Diseases at Université Laval, and part of this work was done while he was still a Distinguished Senior Investigator of The Arthritis Society. Dr. Bruce is an NIHR Senior Investigator and is supported by Arthritis Research UK, the NIHR Manchester Biomedical Centre, and the NIHR/Wellcome Trust Manchester Clinical Research Facility. The views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. Dr. Isenberg and Dr. Rahman are supported by the NIHR University College London Hospitals Biomedical Research Center. The Hopkins Lupus Cohort is supported by the US National Institutes of Health (NIH; grant AR43727 and 69572). Dr. Dooley’s work was supported by the NIH (grant RR00046). Dr. Ramsey-Goldman’s work was supported by the NIH (grants 5UL1TR001422-02, formerly 8UL1TR000150 and UL-1RR-025741, K24-AR-02318, and P60AR064464 formerly P60-AR-48098). Dr. Ruiz-Irastorza is supported by the Department of Education, Universities and Research of the Basque Government. Dr. Jacobsen is supported by the Danish Rheumatism Association (A3865) and the Novo Nordisk Foundation (A05990). Dr. Hanly is supported by the Canadian Institutes of Health Research (grant MOP-88526).
- Accepted for publication March 27, 2019.