Introduction

Although life-saving, the physiologic stress of hemodialysis initiation contributes to physical impairment in some patients. Mortality risk assessment following hemodialysis initiation is underdeveloped and does not account for change over time. Measures of physical resilience, the ability of a physiologic state to overcome physiologic stressors, may help identify patients at higher mortality risk and inform clinical management.

Methods

We created 3 resilience categories (improving, stable, and declining) for trajectories of 4 phenotypes (physical function [PF], mental health [MH], vitality [VT], and general health [GH]) using SF-36 data collected the first year after hemodialysis initiation in the Choices for Healthy Outcomes in Caring for ESKD (CHOICE) study on 394 adults aged more than 55 years. Using mixed effects and Cox proportional hazard modeling, we assessed mortality following the first year on dialysis by resilience categories for each phenotype, adjusting for baseline phenotype and other confounders defined a priori over 4 years average follow-up.

Results

Based on global Wald tests, statistically significant associations of PF (P = 0.03) and VT (P = 0.0004) resilience categories with mortality were found independent of covariates. Declining PF trajectory was associated with higher mortality risk (hazard ratio [HR] = 1.32; 95% confidence interval [CI], 1.05–1.66), whereas improving VT trajectory was associated with lower mortality risk (HR= 0.73; 95% CI, 0.53 to 1.00), each as compared to stable trajectory.

Conclusion

Decreased resilience in PF and VT was independently associated with mortality. Phenotypic trajectories provide added value to baseline markers and patient characteristics when evaluating mortality. Hence, resilience measures hold promise for targeting population health interventions to the highest risk patients.

Graphical abstractKeywordsDialysis is a life-extending treatment for those with kidney failure. Dialysis can also, however, potentiate complications.1Saeed F. Moss A. Duberstein P. Fiscella K. Enabling patient choice: the “deciding notto decide” option for older adults facing dialysis decisions.,2Hemodialysis complications. Available mortality risk assessment models postdialysis initiation for older adults evaluate mortality risk over short time periods, have not been externally validated, and do not take into account change over time.3Ramspek C.L. Verberne W.R. Van Buren M. Dekker F.W. Bos W.J.W. van Diepen Predicting mortality risk on dialysis and conservative care: development and internal validation of a prediction tool for older patients with advanced chronic kidney disease.,4Wick J.P. Turin T.C. Faris P.D. et al.A clinical risk prediction tool for 6-month mortality after dialysis initiation among older adults. Risk assessment models also rarely account for patient reported information.5Anderson R.T. Cleek H. Pajouhi A.S. et al.Prediction of risk of death for patients starting dialysis.,6Ramspek C.L. Voskamp P.W.M. Van Ittersum F.J. Krediet R.T. Dekker F.W. Van Diepen M. Prediction models for the mortality risk in chronic dialysis patients: a systematic review and independent external validation study. Improved methods for identifying older adults at highest risk of death following dialysis initiation must be further explored to improve clinical decision-making and health outcomes.One potential way to improve mortality risk assessment is through physical resilience measurement.7Hadley E.C. Kuchel G.A. Newman A.B. Workshop Spealers and Participants. Report: NIA workshop on measures of physiologic resiliencies in human aging.,8Whitson H.E. Duan-Porter W. Schmader K.E. Morey M.C. Cohen H.J. Colón-Emeric C.S. Physical resilience in older adults: systematic review and development of an emerging construct. Physical resilience is defined as “the ability of a physiologic system to retain its function and identity following a powerful stressor.”9Fried L.P. Cohen A.A. Xue Q.L. Walston J. Bandeen-Roche K. Varadhan R. The physical frailty syndrome as a transition from homeostatic symphony to cacophony.,10Varadhan R. Walston J.D. Bandeen-Roche K. Can a link be found between physical resilience and frailty in older adults by studying dynamical systems?. It is hypothesized to arise from the fitness of one’s stress-response physiology, considered as a dynamical system.11Varadhan R. Seplaki C.L. Xue Q.L. Bandeen-Roche K. Fried L.P. Stimulus-response paradigm for characterizing the loss of resilience in homeostatic regulation associated with frailty. Because physical resilience is inherently dynamical, its measurement occurs over time, as trajectories rather than static measures. Resilience phenotype are health and functional markers thought to respond to experiencing specific stressors. Resilience trajectories of those phenotypes track such markers longitudinally, ideally beginning before the stressor onset and continuing through the stressor initiation and recovery phases, and are a crucial way to operationalize the resilience paradigm.12Colón-Emeric C. Pieper C.F. Schmader K.E. et al.Two approaches to classifying and quantifying physical resilience in longitudinal data. They allow for a dynamic understanding of the resilience phenomenon and have promise both to differentiate adverse outcome risk (e.g., postdialysis mortality) and to anchor discovery of prestressor factors that lead to a resilient response (or not).11Varadhan R. Seplaki C.L. Xue Q.L. Bandeen-Roche K. Fried L.P. Stimulus-response paradigm for characterizing the loss of resilience in homeostatic regulation associated with frailty.There is growing literature conceptualizing the study of physical resilience, but few studies have evaluated the clinical utility of the physical resilience concept in those undergoing hemodialysis. The initiation of dialysis requires a great deal of physiologic reserve to withstand the hemodynamic changes with each dialysis treatment, which is partially why those with physical frailty experience higher mortality and morbidity following dialysis initiation.13Johansen K.L. Delgado C. Kaysen G.A. et al.Frailty among patients receiving hemodialysis: evolution of components and associations with mortality. In this case, physical frailty is the underlying biologic vulnerability to stressors due to decreased physiologic reserve. High physical resilience may well reflect healthy reserve in physiologic mechanisms overlapping with those underlying frailty. The concept of physical resilience, however, is distinguished from frailty by addressing response to a specific stressor, as opposed to a global vulnerability. Therefore, distinct physiological mechanisms may be involved.10Varadhan R. Walston J.D. Bandeen-Roche K. Can a link be found between physical resilience and frailty in older adults by studying dynamical systems?. As such, physical resilience and frailty may involve different physiologic mechanisms.14Whitson H.E. Cohen H.J. Schmader K.E. Morey M.C. Kuchel G. Colon-Emeric C.S. Physical resilience: not simply the opposite of frailty. Exploring the role of physical resilience trajectories in an incident dialysis among an older adult population may add greater understanding to the physical resilience concept and greater insight into accurate mortality risk prediction. Currently, we are undertaking an observational study to evaluate underlying mechanisms of resilience in a small cohort of older adults with impending end-stage renal disease. This present work aimed to inform that study by characterizing resilience trajectories and their implications for long-term survival in a large observational cohort with end-stage renal disease.MethodsStudy Population and DesignThe CHOICE study was a prospective cohort of 1067 incident dialysis patients enrolled between 1995 and 1998; 73% (n = 783) initiated hemodialysis and 27% (n = 284) initiated peritoneal dialysis.15The CHOICE study: a National Study of dialysis outcomes in the United States.. Eligible participants were within 3 months following initial onset of dialysis treatment at study enrollment, aged 18 years and older, and English or Spanish speaking.15The CHOICE study: a National Study of dialysis outcomes in the United States.,16Powe N.R. Klag M.J. Sadler J.H. et al.Choices for healthy outcomes in caring for end-stage renal disease. Patients receiving hospice care were excluded. Following enrollment and baseline questionnaires, participants completed study visits at 3, 6, 9, and 12 months, followed by 6-monthly questionnaires for a mean time of 4 years.15The CHOICE study: a National Study of dialysis outcomes in the United States. The CHOICE cohort was followed for mortality through 2004.17Shafi T. Jaar B.G. Plantinga L.C. et al.Association of residual urine output with mortality, quality of life, and inflammation in incident hemodialysis patients: the choices for healthy outcomes in caring for end-stage renal disease (CHOICE) study.For the purposes of our study, participants were included if they initiated hemodialysis and were older than 55 years (n = 485); this is the eligibility criterion for our observational study. We chose 55 years or older (as opposed to 65 years or older) as our cut-off age because of the age accelerating nature of end-stage renal disease.18Kooman J.P. Kotanko P. Schols A.M.W.J. Shiels P.G. Stenvinkel P. Chronic kidney disease and premature ageing. Time point 0 was first study visit following dialysis initiation. Because we aimed to evaluate the potential for first year resilience trajectories to forecast subsequent survival, participants were excluded if they died within 365 days postdialysis initiation (n = 74). Descriptive statistics were examined to explore key baseline contributors to first-year mortality risk. Figure 1 shows the inclusion flow chart.

Figure 1Flow Chart for Sample Size

Show full caption

GH, general health; MH, mental health; PF, physical function; VT, vitality.

Measures to Characterize ResilienceThe resilience trajectory measures in this study were derived from the PF, emotional well-being (henceforth “mental health”; MH); energy or fatigue (henceforth “vitality”; VT), and GH subscales of the 36-item Short-Form Health survey (SF-36), considered as our resilience phenotypes.19Ware J.E. Sherbourne C.D. The MOS 36-item Short-Form Health survey (SF-36): I. Conceptual framework and item selection.,20Wu A.W. Fink N.E. Cagney K.A. et al.Developing a health-related quality-of-life measure for end-stage renal disease: the CHOICE health experience questionnaire. These 4 measures were chosen because of their clinical relevance within both hemodialysis21Avramovic M. Stefanovic V. Health-related quality of life in different stages of renal failure. and general populations.22McHorney C.A. Ware J.E. Raczek A.E. The MOS 36-item Short-Form Health survey (Sf-36):II. Psychometric and clinical tests of validity in measuring physical and mental health constructs. They were also the closest surrogates available to our physical resilience theoretical framework, wherein we hypothesize phenotypes such as PF, VT, GH and MH to be resilient manifestations of the complex biological dynamics resulting from the stressor of hemodialysis initiation.10Varadhan R. Walston J.D. Bandeen-Roche K. Can a link be found between physical resilience and frailty in older adults by studying dynamical systems?. Our 4 resilience phenotypes included the following: (i) the 10-item PF subscale, which contained items such as “Does your health now limit you in climbing several flights of stairs?” using a 3-item response scale of “yes, limited a lot,” “yes, limited a little” or “no, not limited at all”; (ii) the 5-item MH subscale, which includes items such as “In the past 4 weeks, how much of the time have you felt downhearted and blue?” using 6-point response options ranging from “all of the time” to “none of the time”; (iii) the 4-item VT subscale, which included items such as “In the last 4 weeks, how much of the time have you felt full of pep?” with the same answer option range as MH; and (iv) the 5-item GH subscale, which includes items such as “My health is excellent” using 5-point response options ranging from “definitely true” to “definitely false.” Each subscale was scored from 100 (highest possible score indicating best health) to 0 (lowest possible score).Socio-demographicsA standard baseline questionnaire assessed characteristics including the following: age (in years); sex (male or female); race or ethnicity (White, African American, or Other) because of the established racial differences in survival on dialysis;23Racial and ethnic disparities in end-stage kidney failure-survival paradoxes in African-Americans.,24Crews D.C. Sozio S.M. Liu Y. Coresh J. Powe N.R. Inflammation and the paradox of racial differences in dialysis survival. education (less than high school, high school or some college, or college or higher); and insurance status (dichotomized as any inclusion of Medicaid insurance or no insurance vs. private insurance and/or Medicare without Medicaid).CovariatesCovariates were included based on associations with mortality in previous dialysis research. Body mass index (in kg/m2) was based on height and weight reported on the Health Care Financing Administration Medical Evidence Report (HCFA Form 2728).17Shafi T. Jaar B.G. Plantinga L.C. et al.Association of residual urine output with mortality, quality of life, and inflammation in incident hemodialysis patients: the choices for healthy outcomes in caring for end-stage renal disease (CHOICE) study. Kidney disease specific markers selected included the following: (i) type of vascular access, by category: arteriovenous fistula or graft, or central venous catheter or unknown;25Astor B.C. Eustace J.A. Powe N.R. Klag M.J. Fink N.E. Coresh J. Type of vascular access and survival among incident hemodialysis patients: the choices for healthy outcomes in caring for ESRD (CHOICE) study. (ii) predialysis initiation nephrology consult timing (early [>12 months], intermediate [4–12 months], late [26Kinchen K.S. Sadler J. Fink N. et al.The timing of specialist evaluation in chronic kidney disease and mortality.; (iii) laboratory values for serum albumin (g/dl) obtained from HCFA Form 2728;27Miskulin D.C. Meyer K.B. Athienites N.V. et al.Comorbidity and other factors associated with modality selection in incident dialysis patients: the CHOICE study. and (iv) estimated glomerular filtration rate, calculated using the 4-variable Modification of Diet in Renal Disease Study equation.28Levey A.S. Bosch J.P. Lewis J.B. Greene T. Rogers N. Roth D. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. We modeled comorbidities using the Index of Coexistent Disease (ICED), a validated index of presence and severity of comorbid conditions used extensively in mortality risk prediction in patients with kidney failure. ICED scores included 0 and 1 (mild), 2 (moderate), and 3 (severe).27Miskulin D.C. Meyer K.B. Athienites N.V. et al.Comorbidity and other factors associated with modality selection in incident dialysis patients: the CHOICE study.,29Nicolucci A. Cubasso D. Labbrozzi D. et al.Effect of coexistent diseases on survival of patients undergoing dialysis.,30Athienites N.V. Miskulin D.C. Fernandez G. et al.Comorbidity assessment in hemodialysis and peritoneal dialysis using the index of coexistent disease.Statistical AnalysisOur first aim was to characterize resilience trajectories based on SF-36 subscales over the baseline and follow-up assessments at 3, 6, and 12 months, with 1 per each resilient phenotype (PF, MH, VT, and GH). We conceptualized key features in terms of trends over time as well as the potential nonlinearity or variability relative to a straight-line trend. As a first step, repeated measures from each resilient phenotype were characterized using linear mixed effects models with random intercept and slope terms. We then estimated each participant’s intercept and slope, thus a straight-line approximation to their trajectory, using best linear unbiased predictor values.31McDonald S.R. Heflin M.T. Whitson H.E. et al.Association of integrated care coordination with postsurgical outcomes in high-risk older adults: the perioperative optimization of senior health (POSH) initiative. Next, we categorized trajectories into a few subgroups with similar characteristics based on the slope of their straight-line approximation as increasing or decreasing as well as the closeness of fit of a straight-line approximation to their actual trajectory values (details are provided in the Supplemental Material). This process identified 3 mutually exclusive groups who exhibited generally improving, declining, or stable (flat) trajectories (henceforth, “resilience categories”), which were utilized for further analysis.Participant characteristics were compared for those with survival of less than 12 months versus those surviving more than 1 year. We next characterized post-12-month survival by resilience categories using Kaplan-Meier curves and Cox proportional hazards models adjusted for potential confounders. One set of analyses, each, was conducted per resilience phenotype (PF, VT, MH, and GH). In Cox modeling, models adjusted for random intercept estimates from the initial phenotype-specific trajectory modeling, so that we could address the value added by trajectory information beyond measures of baseline resilience phenotype values. A robust variance correction was applied to address CHOICE data clustering by dialysis clinics. Multiple imputation using sequential imputation with chained equations was performed to address missingness in serum albumin, type of vascular access, nephrology referral time, body mass index, estimated glomerular filtration rate, urine albumin-creatinine ratio and ICED score.32White I.R. Royston P. Wood A.M. Multiple imputation using chained equations: issues and guidance for practice.,33Lepkowski J.M. Raghunathan T.E. Solenberger P. Van Hoewyk J. A multivariate technique for multiply imputing missing values using a sequence of regression models-ARCHIVED. We hypothesized lower mortality across improving (lowest), stable and declining trajectories for each resilience phenotype.

To explore adequacy of our primary analysis, several sensitivity analyses were implemented. To ensure that findings were not unduly influenced by individuals having only 1 measurement occasion and no trajectory information in the first year, an analysis excluding these individuals was performed. To further explore dependence of trajectory on persons’ baseline phenotypic values, analyses adding interactions between the random intercept estimate and resilience category, and replacing random intercept estimates with persons’ measured baseline values were performed. To ensure that there was not undue information loss in categorizing resilience trajectories, an analysis replacing resilience categories with random slope estimates for first year functional change was performed. To further evaluate whether phenotypic variability, and not only mean trend, may have survival implications, analyses were performed adding the logarithm of the residual variance of individuals’ phenotype values about their estimated trend to the analysis described in the previous sentence. All analyses that employed continuously scaled covariates were checked for linearity and undue influence using Martingale residual plots. Proportionality of hazards was evaluated using Schoenfeld residual plots.

ResultsOur study sample had average age of 68.4 years with nearly equal gender distribution, diversely distributed education levels, and 23% on Medicaid (Table 1). Body mass index, ICED category, and clinical characteristics relating to nephrology care and dialysis type were heterogeneously distributed. Mean resilience phenotype scores at baseline were low compared to a community dwelling sample of older adults34Walters S.J. Munro J.F. Brazier J.E. Using the SF-36 with older adults: a cross-sectional community-based survey. at 41.2 for PF, 42.5 for GH, and 42.1 for VT (all out of 100). The average baseline MH score was 70.6.Table 1Sociodemographic and medical characteristics comparing those Surviving more than 1 yeara

Survival > 1 year (n = 394) is sample used in resilience trajectory analyses.

versus 1 year or less

BMI, body mass index; eGFR, estimated glomerular filtration rate; ICED, index of coexistent diseases

Measures are displayed as mean (SD), except where indicated. Sample sizes may differ across covariates. For continuous variables, Kruskal-Wallis test was used; for categorical variables, Chi-squared test was used.

Individual characteristics were similarly distributed, in many respects, by survival status of less than or equal to 1 year versus greater than 1 year (Table 1). Characteristics exhibiting statistically significant differences (survival ≤ 1 year vs. >1 year), included race (84% vs. 65% White), ICED category (46% vs. 27% category 3), and access type (73% vs. 51% catheter). Mean resilience phenotypes at baseline were 6 to 8 points lower among those dying earlier versus those surviving longer for all subscales except PF, where the deficit exceeded 11 points.For each resilience phenotype, substantial heterogeneity in baseline to 12 month trajectories was observed, encompassing both baseline function level and patterns of change over 1 year. Figure 2 displays, for each phenotype, trajectories according to the improving, stable and declining “resilience categories” defined earlier. In each case, despite heterogeneity, our categorization distinguished trajectories that were improving, stable and declining overall. In addition (data not shown), individuals’ trajectories tended to be positively associated across domains. Analytical sample characteristics are tabulated by resilience categories, and multiple regression analyses of resilience categories by personal characteristics are presented in the Supplementary Materials and Supplementary Table S1 to S4.

Figure 2First 150 Resilience Trajectories for each Resilient Phenotype

Show full caption

GH, general health; MH, mental health; PF, physical function; VT, vitality.

Post-12-month survival was followed on average for over 4 years. In fully adjusted Cox models to evaluate determinants of this outcome (Table 2), both PF (P = 0.03) and VT (P = 0.0004) resilience categories were associated with mortality risk (adjudicated by global Wald tests). For the PF phenotype, mortality risk for the declining versus the stable trajectory category was higher by 32% (HR = 1.32, 95% CI = 1.05–1.66) after adjustments for socio-demographics, body mass index, comorbidities, and nephrology specific gold standard measures. Likewise, the risk of mortality was lower by 27% among those in the improving VT resilience category compared with those in the stable VT resilience category after adjustments for the same covariates (HR = 0.73, 95% CI = 0.53–1.00). Importantly, these associations were present after adjusting for many well-established variables that predict mortality, including the baseline phenotype (random intercept) itself. Kaplan-Meier plots characterizing crude mortality associations with resilience categories were consonant with Cox model findings for each resilience phenotype for the first 5 to 6 years of follow-up, after which some crossover of survival curves was observed (Supplementary Figure S1). These plots lend insight into the counterintuitive, albeit not statistically significant, HR in the direction of a slight mortality deficit for those with increasing versus stable PF trajectory category (HR = 1.15, 95% CI = 0.90–1.48). Kaplan-Meier curves mirrored this trend in the first 6 years of follow-up, but a survival benefit for those with improving trajectory category was observed thereafter. Interestingly, baseline phenotypic status estimated by random intercepts was not associated with mortality after adjustments for any of the 4 resilience phenotypes.

Table 2Adjusted risk of mortality by resilience phenotype and resilience category

CI, confidence interval; HR, hazard ratio; ref, reference.

Adjusted GEE model results with imputed data (Adjustment included: age, sex, race, body mass index, education, random intercept, Index of Coexistent Diseases, estimated glomerular filtration rate, serum albumin, access type, nephrology consult timing, and insurance type).

Analytical sample characteristics overall and by PF resilience categories are presented in Table 3. Many characteristics were similarly distributed across resilience categories. Nevertheless, the percentage of African American or Other participants among those with improving resilience category (46%) was significantly higher than in stable (30% African American or Other) and declining (35% African American or Other) categories (chi-squared P = 0.021). Analogous data on other phenotypes are presented in Supplementary Table S1 to S3. Higher education was significantly associated with improving MH and declining GH resilience categories. For MH, differentiation by race was observed with African American or Other participants making up a larger percentage of the MH stable and improving resilience categories compared to White participants (32% stable, 47% improving, chi-squared P = 0.049). For VT, there was an association with serum albumin at dialysis initiation, which was lower among those with improving or declining VT categories versus the stable VT category (Kruskal-Wallis P = 0.004). In mutually adjusted multinomial logistic regression models to characterize trajectory category by covariates (Supplementary Table S4), only education consistently sustained statistical significance. The most striking pattern was that those with college education were less likely to have the declining trajectory category for both MH and VT, but also less likely to have improving GH category.

Table 3Sociodemographic and medical characteristics by physical function resilience phenotype

BMI, body mass index; eGFR, estimated glomerular filtration rate; ICED, index of coexistent diseases

Measures are displayed as mean (SD), except where indicated. Sample sizes may differ across covariates, for continuous variables, Kruskal-Wallis test is used; for categorical variables, Chi-squared test is used.

Sensitivity Analysis

Sensitivity analyses both supported and added nuance to our primary analysis. Findings were slightly strengthened in analyses excluding individuals with only 1 phenotypic measurement over time. In addition in these analyses, the declining (vs. stable) resilience category for GH was significantly associated with a 49% increased risk of mortality (95% CI = 5.6% to 110%, P = 0.023; n = 364).

Analyses evaluating interactions between baseline resilience phenotype (random intercept) and resilience phenotype trajectories did not identify significant moderation for either the VT or GH phenotypes (VT: P = 0.650, GH: P = 0.401). For the PF phenotype, a slight synergistic survival benefit of improving PF trajectory together with higher baseline PF was observed (global P-value for interaction = 0.039 with 15.1% reduction in the relative hazard for mortality for improving vs. stable trajectory per 1 SD higher baseline score, 95% CI = 0.5%–27.4%). For the MH phenotype, where no main effect was shown, the mortality association with a declining MH trajectory was considerably increased among persons with low MH to start (global P-value for interaction < 0.0001 with 30.9% increase in the relative hazard for declining vs. stable trajectory per 1 SD lower baseline score, 95% CI = 22.9%–38.0%).

Fully adjusted analysis representing resilience trajectories as subject specific random slopes for first year phenotypic change, rather than categorizing, amplified findings from our primary analysis as follows: improving trajectory was significantly associated with decreased mortality for PF (HR = 0.758, 95% CI 0.616–0.932) and VT (HR = 0.628, 95% CI 0.461–0.856) as well as GH (HR = 0.749, 95% CI 0.623–0.902) phenotypes. Adding one