A prospective multicenter cohort study of frailty in younger critically ill patients

Study design, participants and setting

We conducted a planned substudy of a prospective multicenter cohort study that has been described previously [11]. Adults admitted to six closed multisystem medical-surgical ICUs located in two tertiary/academic and four community hospitals in Alberta, Canada, between February 2010 and July 2011 were screened for enrollment [11]. The study was approved by the research ethics board at the University of Alberta (Pro00007628). All participants or their designated surrogate decision-makers provided written informed consent to participate.

Frailty definition

Frailty was operationalized using the Canadian Study of Health and Aging Clinical Frailty Scale (CFS) score, which was modified to an 8-point tool designed to categorize patients as fit, vulnerable, or frail [6]. The CFS is a subjective judgment-based screening tool for frailty that has been proven to be valid, reliable, and simple to perform. We defined patients as frail if their CFS score was ≥5 (moderate to severe frailty for CFS score 6–8), as vulnerable if their CFS score was 4, and as fit if their CFS score was ≤3 [11]. Trained coordinators interviewed participants and/or their surrogate decision-makers and reviewed each participant’s medical record.

Outcomes

The primary outcome was all-cause mortality 1 year following enrollment. Secondary outcomes were focused on (1) patient-centered outcomes, including ICU, hospital, 90-day, and 6-month mortality; HRQoL at 6 and at 12 months, captured using the EuroQol (EQ-5D) Health Questionnaire (including the EQ-5D visual analogue scale [EQ-5D-VAS]) [13]; and discharge disposition; and (2) health service use, including ICU and hospital lengths of stay, ICU readmission, and hospital readmission in the 1-year period following enrollment.

Data collection and management

Data were prospectively captured on standardized forms and entered into an electronic database. These data elements included sociodemographic factors, baseline functional status and disability (e.g., basic and instrumental activities of daily living), comorbid conditions defined and summated using the Elixhauser comorbidity scale [18], prescription medications, source of ICU admission (e.g., ward, operating theater, emergency department), diagnostic category, illness severity (e.g., defined according to the Acute Physiology and Chronic Health Evaluation [APACHE] II score [19]), presence and severity of organ dysfunction (e.g., defined according to the Sequential Organ Failure Assessment score [20]), treatment intensity (e.g., mechanical ventilation, vasoactive support, renal replacement therapy), and preferences for life-sustaining therapy (e.g., full ICU support, limitation in therapy). Participants were contacted at 6 and 12 months to ascertain long-term outcomes, including vital status, disposition, and HRQoL.

Statistical analyses

Among nonelderly patients, the distribution of CFS scores was presented and descriptive statistics were calculated according to the presence or absence of frailty. Independent associations between baseline sociodemographic and clinical characteristics as well as frailty status were evaluated using multivariable logistic regression analysis. Clinically important variables (i.e., sex, comorbidity, case mix, APACHE II score, site) and those found to be significant in univariate analysis (p = 0.20) were entered into the multivariable model. Model calibration and discrimination were assessed using the Hosmer-Lemeshow goodness-of-fit test, Brier score, and the AUC (c-statistic). The independent percentage contribution of each variable in the model’s explanatory power was estimated by dividing differences in log-likelihoods of the nested models by the difference in the log-likelihoods of null and full (final) models [14, 21]. Sensitivity analysis was performed by adding the variables age, surgical status, and sepsis to the multivariable model. Survival curves were plotted using Kaplan-Meier curves with log-rank tests. Multivariable Cox proportional hazards regression analyses were performed to analyze survival. Similar to the analyses described above, clinically important variables (i.e., sex, comorbidity, case mix, APACHE II score, site) and those found to be significant in univariate analysis (p = 0.20) were entered into the multivariable model. Proportional hazards model assumptions were checked by comparing log (−log) plots of survival probabilities over time of frail and nonfrail patients and testing interactions between frailty status and logarithmic scales of follow-up times in the model. A p value <0.05 was considered statistically significant for all comparisons. All analyses were performed using STATA 12.1 software (StataCorp, College Station, TX, USA).

Factors associated with frailty among nonelderly patients

Association between frailty and clinical course among nonelderly patients

Association between frailty and mortality among nonelderly patients

Unadjusted mortality at 1 year was not significantly greater for frail patients than for nonfrail patients (32.7 % vs. 20.4 %; OR 1.90, 95 % CI 0.95–3.78). Unadjusted mortality in the ICU, in the hospital, at 90 days, and at 6 months was similarly not statistically different between frail and nonfrail patients (Table 3). In multivariable analysis, frail patients were found to have a higher adjusted risk of death at 1 year than nonfrail patients (adjusted HR 1.83, 95 % CI 1.03–3.25; p = 0.039) (Fig. 2 and Fig. 3). The burden of comorbid illness, acute illness severity, and frailty represented 81 % of the model’s final explanatory power for 1-year mortality. There was evidence of a dose-response increase in adjusted risk of death associated with greater CFS score (Table 4 and Fig. 4). This increase was most apparent when we compared fit (CFS score 1–3) with moderate to severe frailty (CFS score 6–8), while there was overlap among those classified as vulnerable (CFS score 4) and mildly frail (CFS score 5). In an exploratory analysis using the entire study cohort (n = 421), there was an effect modification between frailty and age with respect to all-cause mortality (mortality among frail patients age <65 years 36 % vs. age ≥65 years 59 %; OR 2.0,3 95 % CI 1.30–3.16; p < 0.001), although in multivariable analysis the interaction term was not statistically significant (OR 0.84, 95 % CI 0.42–1.71; p = 0.68) (Additional file 3).

CFS category	Unadjusted HR	95 % CI	p Value	Adjusted HRa	95 % CI	p Value
Fit (CFS score 1-3)	1.0	–	–	1.0	–	–
Vulnerable (CFS score 4)	3.67	1.55–8.69	0.003	2.89	1.19–7.02	0.019
Mild frailty (CFS score 5)	2.82	0.95–8.38	0.063	2.54	0.82–7.90	0.107
Moderate to severe frailty (CFS score 6–8)	4.93	1.91–12.73	0.001	4.41	1.62–12.06	0.004

CFS Clinical Frailty Scale

^aAdjusted by sex, Elixhauser comorbidity score, Acute Physiology and Chronic Health Evaluation II score, primary diagnostic criteria, and hospital type (tertiary care/academic vs. community hospital)

Association between frailty and nonfatal outcomes among nonelderly patients

Among hospital survivors, 69.8 % of frail and 76.2 % of nonfrail patients were living at home (difference 6.4 %, 95 % CI −9.9 % to 22.1 %); however, only 32.8 % of frail patients were independent (absolute decrease from prehospital level −9.2 %, 95 % CI −28.4 % to 9.9 %), while 50.8 % of nonfrail patients were independent (absolute decrease from prehospital −35.8 %, 95 % CI −46.3 % to −25.3 %) at the time of hospital discharge. EQ-5D-VAS scores were similar for frail and nonfrail patients at 6 months (58.8 [18.7] vs. 63.4 [20.4]; p = 0.25) and at 12 months (63.0 [20.2] vs. 68.3 [17.6]; p = 0.18). There was no significant difference between 6 and 12 months in the EQ-5D-VAS for frail and nonfrail patients (4.0 [95 % CI −1.4 to 9.5] vs. 5.3 [95 % CI 1.8–8.8]; p = 0.72) or in the proportion achieving a minimal clinically important improvement in EQ-5D-VAS (37.0 % vs. 37.7 %; p = 0.95) by 12 months. A greater proportion of frail patients had problems across all EQ-5D domains compared with those who were not frail (Table 3).

Association between frailty and health services use among nonelderly patients

There were no statistical differences in ICU or hospital lengths of stay or rates of ICU readmission between frail and nonfrail patients (Table 3). The hospital readmission rate in the year following enrollment was significantly greater for frail patients than for nonfrail patients.

Main findings

First, we showed that frailty was relatively common among younger patients admitted to the ICU in our study, evident in more than one in four. This was significantly greater than the estimated prevalence of frailty among a contemporaneous general population [17]. Second, as expected, several sociodemographic factors correlated with prehospital frailty. Frailty was more common among women, those with less than a secondary school education, those receiving disability insurance, and those requiring assistance at home. In addition, in the 1-year period before the study, health service use was higher among those classified as frail than among those who were not frail. Third, frail patients in this younger cohort where characterized by a high burden of comorbid disease, in particular CTD and heart failure (HF), and were prescribed a greater number of medications than those who were not frail. Fourth, preferences for life support, reasons for ICU admission, and treatment intensity were similar between frail and nonfrail patients. Finally, in adjusted analyses, long-term mortality and rehospitalization were greater among those with prehospital frailty.

Context with prior literature

Most published literature is focused on describing frailty among older persons [12, 14, 15]; few authors have characterized its epidemiology among nonelderly persons [17], in particular in the context of critical illness [22]. This is likely attributable to most frailty research being performed in the domain of geriatric medicine [6, 16] and using administrative databases with age thresholds (e.g., U.S. Medicare beneficiaries database) [14, 15] or being based on the misconception that frailty is solely a product of chronologic aging [12]. More recently, researchers in a number of studies have evaluated the prevalence and impact of frailty among specialized cohorts of much younger patients, including lung transplant candidates (median age 59 years [23]), end-stage liver disease (mean age 55 years [24]), end-stage kidney disease (mean age 55 years [25]), and kidney transplant (mean age 53 years [26]). A greater rate of deficit accumulation in selected younger persons may accelerate the development of frailty, and, as such, these persons may have “age-shifted” vulnerability to major stressors (i.e., critical illness) [27]. These patients manifest frailty and “age” more rapidly relative to chronologic age [28]. Similarly, the probability of survival after critical illness appears higher for younger than for older frail patients [11]; however, younger patients may still experience a longer and/or more complex post-ICU course and recovery [29–32]. Accruing evidence, along with our data, suggests that selected younger patients, such as those with advanced chronic inflammatory diseases (e.g., CTD) or end-stage organ diseases (e.g., HF, cirrhosis) may benefit from screening for frailty.

Comorbid disease and frailty are not mutually exclusive and may show greater correlation among younger patients with significant and/or advanced chronic illness [1]. As examples, we identified both CTD and HF as predictors of prehospital frailty. Few studies have evaluated the prevalence and impact of frailty among patients with CTD [33]. Numerous factors may predispose these patients to frailty, including chronic disease progression and persistent inflammation, disease-specific therapy (i.e., immunosuppressive or other disease-modifying antirheumatic drugs), nutritional alterations (i.e., cachexia), and sarcopenia [34]. Indeed, sarcopenia and nutritional deficiencies before critical illness may amplify early skeletal muscle loss in critical illness, further prolonging recovery and exacerbating risk of incident or worsening disability [35, 36]. Likewise, an estimated 18–54 % of patients with HF are clinically frail [37]. Frailty in HF may have important prognostic implications, including reduced likelihood of self-management, impaired HRQoL, greater hospitalizations, being declined for transplant, and death [37, 38]. We believe our finding of associations between frailty, selected comorbid conditions, and critical illness requires confirmation and evaluation in larger studies.

Prehospital frailty portends a greater risk of death and impaired recovery following the stress of critical illness. This finding would appear robust regardless of age; however, there is likely some additive effect between frailty, older age, and risk of death [11, 12, 16]. Our study suggests that the association between frailty and mortality may be attenuated in younger compared with older patients; however, those with moderate to severe frailty still exhibit significantly higher risk [11]. This implies that frailty among younger patients may be less likely a terminal event or imminent end-of-life trajectory. That said, these relatively young frail survivors have greater opportunity and time at risk for the multifarious physical (i.e., disability) and psychosocial (i.e., depression, posttraumatic stress, impaired HRQoL, inability to work, lost income, social isolation) complications increasingly described after critical illness [29–32, 39]. Indeed, at 6 months, frail survivors in our study described far greater problems with mobility, self-care, usual activities, and pain. We found that more than half of frail survivors described issues with depression and/or anxiety [40]. Interestingly, despite all EQ-5D domains being generally worse among frail patients, EQ-5D-VAS scores, while impaired, were not significantly different at 6 and 12 months between frail and nonfrail patients. Thus, frail survivors’ self-rated global HRQoL was not significantly different from that of those who were not frail. There may be a number of explanations for these findings. First, it may imply that frail survivors adapt with time to their new vocation and/or disposition, despite the high prevalence of residual “problems.” Second, this substudy may have had limited capability to detect significant clinical differences in HRQoL due to being relative small and within the context of the operative features of the EQ-5D. Regardless, this also translated into greater rates of rehospitalization in the subsequent 1-year period, implying greater health service use and healthcare costs for those screened as frail [5].

Implications for clinicians and future research

Our study has relevance for intensivists and all the other clinicians who care for survivors of critical illness. It also reinforces the potential value of frailty screening among selected younger patients admitted to the ICU. Recognition and acknowledgement of prehospital frailty can serve to inform survivorship expectations after critical illness, as well as steer the mobilization of customized recovery needs, both in the hospital and in the community, across physical, emotional, and social domains [41]. The ideal content and implementation of multifactorial, interprofessional post-ICU interventions to improve recovery remain uncertain and a challenge for providers [42, 43]. Importantly, we believe further rigorous research in larger cohorts is needed to confirm our findings, to further characterize those younger patients most likely to benefit from frailty screening, and to develop translatable interventions aimed at enhancing recovery (i.e., preserving autonomy, slowing health status deterioration, societal reintegration and engagement) and informing clinical decision-making [13]. Moreover, future work should ideally evaluate the comparative performance of additional screening instruments in addition to the CFS score (e.g., frailty index, physical performance measures). In addition, one of the most important implications of our work is that future research related to recognizing and mitigating frailty should not necessarily be age-restricted. Indeed, on the basis of the importance of prior hospitalizations and the presence of selected comorbid conditions, we believe that the cohort of younger frail patients is only likely to increase over time.

Limitations

Our study has important limitations that must be considered. First, while the study was preplanned, we recognize that it entails a secondary analysis focused on a smaller subgroup with limited statistical power. Second, while this study was focused on younger critically ill patients, the age range for this subgroup was still limited to patients aged 50–64.9 years. We therefore cannot comment on the prevalence or implications of frailty among those younger than 50 years old. Third, we recognize that our study is susceptible to selection bias, given that all participants were recruited following ICU admission. Fourth, the CFS score was intended as a screening tool for frailty that was previously validated in older patients. We recognize there is no “gold standard” for the diagnosis of frailty among younger patients; however, we believe that the CFS is simple, has face validity, and was able to discriminate a subgroup at increased risk for adverse outcomes [6, 9]. We also did not capture additional highly correlated surrogates for frailty, such as sarcopenia, that were shown to have similar predictive capacity as measures of frailty for adverse outcomes among critically ill surgical patients [44]. Finally, as previously described [13], we did not capture physical performance measures such as mobility, grip strength, or cognition before critical illness, which may have particular relevance among younger frail survivors of critical illness.