Secondary Logo

Journal Logo

Perioperative medicine

Frailty predicts 30-day mortality in intensive care patients

A prospective prediction study

De Geer, Lina; Fredrikson, Mats; Tibblin, Anna O.

Author Information
European Journal of Anaesthesiology: November 2020 - Volume 37 - Issue 11 - p 1058-1065
doi: 10.1097/EJA.0000000000001156



Frailty is a multidimensional syndrome characterised by the loss of physiological and cognitive reserves, leading to an increased vulnerability and a higher risk of negative health outcomes.1,2 Frailty is associated with multiple hospital admissions and peri-operative complications, with the need for institutionalisation, and with death.3,4 Thereby, frailty results in considerable consequences for patients and families as well as for the healthcare system.5,6

Although the concept of frailty was originally developed in geriatrics, there is now growing interest in the concept in other medical fields. In intensive care medicine, frailty is an independent predictor of adverse outcomes, not only in the elderly but also in the younger frail patient.7–10 With growing demands on intensive care services across all patient groups, the proportion of frail patients in intensive care, and the impact of frailty on outcomes, are likely to increase. However, while there is much interest in outcome prediction in intensive care medicine, prediction models in clinical use do not account for the level of frailty in patients.11,12

The current study was conducted to assess frailty in a large, unselected group of general intensive care patients, and its impact on outcome, defined as death within 30 days of ICU admission. Our hypothesis was that frailty could be used for outcome prediction, and that frailty would add prognostic accuracy to a well established and widely used model for outcome prediction in intensive care.


The current study was approved by the Regional Ethical Review Board in Linköping, Sweden (Dnr 2016/537-31). Due to the observational nature of study, the requirement for informed consent was waived. A formal license agreement with permission to use the clinical frailty scale (CFS) was obtained from the copyright holder.13 The study was performed and is reported according to the TRIPOD checklist (Supplementary file TRIPOD checklist,

Sources of data

The study was conducted as a prospective study from January 2017 to June 2018. Data were collected from patients or families, patient charts and registry sources.


Adult patients admitted to the ICU in a mixed, noncardiothoracic, tertiary general ICU in a university hospital were included. Patients could be included only once and, in cases of multiple ICU admissions, only their primary admission (Fig. 1). Data on patient characteristics and admission details were collected, as well as data on invasive mechanical ventilation and continuous renal replacement therapy. In addition, ICU length of stay, whether treatment was withheld or withdrawn, adverse events, diagnoses and death in or after the ICU, whereas also recorded.

Fig. 1:
Flow chart of patients.

Adverse events in the ICU were defined by the Swedish Intensive Care registry15: cases of ventilator associated pneumonia, clostridium difficile enterocolitis, severe hypoglycemia (plasma glucose <2.2 mmol l−1), central venous catheter-related infection, and pneumothorax requiring treatment.15 Also included were other events occurring in the ICU: dislocation or dysfunction of the endotracheal tube, unplanned re-intubation, postoperative meningitis, as well as discharge at night or unplanned readmission to the same ICU within 72 h of discharge.


Outcome was defined as death within 30 days of ICU admission: this was assessed from patient charts. Patients were followed up regarding survival for up to 180 days after ICU admission. Patient charts in Sweden are constantly linked to national registries covering all citizens, and thus provide information on whether a patient has died.


The predictors used in this study were frailty using the CFS as a categorisation tool (Supplementary File 1, and severity of illness using the simplified acute physiology score, third version (SAPS3). Frailty was defined as the level of frailty before the acute illness and hospital admission. The information necessary to perform this assessment was collected during the patient's stay in the ICU from the patient, members of the family or from patient charts. The CFS of an individual patient was decided by the senior treating physicians, all trained in the use of CFS before the start of the study. SAPS3 was assessed at the time of admission, and it was recorded in all cases.

The aim of this study was to compare a prediction model for mortality including frailty with a prediction model including SAPS3 only. The secondary outcome was to estimate a discrimination and calibration of a model including frailty and SAPS3.

Sample size considerations and treatment of missing data

No sample size calculation was performed for this study, and no imputation of data were performed. Patients for whom data on outcomes were missing were excluded from the study.

Statistical analysis

Data are presented as median [IQR] and number (%). The Mann–Whitney U test and χ2 test were used for comparisons between groups, as appropriate. The Pearson's correlation analysis was used to test the relation of CFS to SAPS3. The ability of CFS and SAPS3 to predict death was analysed using receiver-operating characteristic (ROC) curves. The cut-off CFS was defined as the value providing the best sensitivity and specificity for predicting death and determined with the likelihood ratio test as well as using the Youden index test.16 The combined predictive ability of SAPS3 and CFS was calculated as the odds of the hazard ratios of the two and the areas under the ROC curves were compared using χ2 tests.17 A bootstrap sampling with 1000 replications was used for internal validation, and the calibration of the prediction models was evaluated with Hosmer–Lemeshow goodness-of-fit test and calibration plots. The ability of the predictors was further analysed using the net reclassification index test.18 Groups of patients were compared regarding baseline characteristics, treatments and outcomes using the Mann–Whitney U test and the χ2 test. Survival curves were plotted using Kaplan–Meier curves with log-rank tests, and unadjusted and adjusted Cox proportional hazards regression analyses were performed to analyse survival. A P value less than 0.05 was considered statistically significant for all comparisons. Statistical analyses were performed using IBM SPSS 25.0 (IBM Corp, Armonk, New York, USA) and Stata 15.1 (StataCorp LLC, College Station, Texas, USA).


During the study period, 1181 patients were admitted to the ICU. After excluding children, patients lost to follow-up, and multiple admissions for the same patient, 872 patients remained for inclusion in the study. The median [IQR] age was 64 [46 to 73] years, 512 (59%) of the patients were male, and 807 (93%) were acute admissions to the ICU. At the time of ICU admission, the median SAPS3 was 56 [43 to 67]. Similar proportions of patients were admitted from the emergency department (ED) (30%, n=257), hospital wards (26%, n=223) and from the operating theatre or the postoperative high dependency unit (24%, n=211). The median length of stay in the ICU was 26 [13 to 70] h, and during that time, 425 (49%) patients required invasive ventilation and 40 (5%) required continuous renal replacement therapy. The most common ICU diagnoses were sepsis or septic shock (22%, n=190), respiratory insufficiency (13%, n=117), cardiac arrest (10%, n=90) and multiple trauma (10%, n=88). All patients were assessed regarding their CFS preceding the acute illness and hospital admission, and the median CFS was 4 [3 to 6]. The proportions of patients corresponding to different CFS levels are shown in Fig. 2.

Fig. 2:
Prevalence of clinical frailty scale scores and corresponding proportion of deaths within 30 days.

The correlation of CFS to SAPS3 corresponded to an r of 0.4. The ability of CFS to predict death within 30 days was analysed using a ROC curve (Fig. 3). The area under the curve (AUC) was 0.74 [95% confidence interval (CI), 0.69 to 0.79], and a CFS of 5 corresponded to a sensitivity of 76% and specificity of 66%, defining CFS at least 5 as the cut-off point. The predictive ability of SAPS3 corresponded to an AUC of 0.79 (95% CI, 0.75 to 0.83), and there was no significant difference between the two (P = 0.53; Fig. 3). Combining CFS and SAPS3 increased the AUC to 0.82 (95% CI, 0.79 to 0.86), corresponding to an improved predictive ability (CFS + SAPS3 vs. CFS alone, P < 0.001; CFS + SAPS3 vs. SAPS3 alone, P = 0.02) (Fig. 3), and an integrated discrimination index of 0.009 (P = 0.02). Calibration with a Hosmer–Lemeshaw goodness-of-fit test showed a good agreement between the observed and predicted values for the model with CFS and SAPS3 (χ2 = 8.41, P = 0.39). Calibration plots are shown in Fig. 4. A bootstrap replication for internal validation gave a result very similar to the original logistic regression analysis [odds ratio 1.45 (95% CI, 1.25 to 1.68) vs. 1.45 (95% CI, 1.28 to 1.65)].

Fig. 3:
The receiver-operating characteristics curves of clinical frailty scale (red), simplified acute physiology score, third version (green) and a composite of the two (clinical frailty scale + simplified acute physiology score; blue), plotted against death within 30 days.
Fig. 4:
Calibration plots for the predictive capacity of clinical frailty scale, simplified acute physiology score, third version and a composite of the two (clinical frailty scale + simplified acute physiology score).

The identified cut-off CFS level was used to classify patients as frail/nonfrail (Table 1). In patients where treatment was limited (treatment withheld or withdrawn at admission or during their ICU stay), the proportion who died in the ICU did not differ [55 of 122 frail patients (45%) vs. 14 of 34 nonfrail (41%), P = 0.45]. Among those discharged from the ICU, the proportion of patients with continuing limitations of care was higher in the frail than in the nonfrail [67 of 308 frail patients (22%) vs. 24 of 527 (4%), P < 0.001]. After excluding patients with therapeutic limitations, mortality remained higher in the frail than in the nonfrail [13 of 299 frail patients (4%) without therapeutic limitations died in the ICU vs. three of 210 nonfrail patients (1%), P = 0.016].

Table 1:
Baseline characteristics, clinical course and outcome of patients

The median ICU length of stay was longer in frail patients than in the nonfrail in ICU survivors (39 [18 to 92] vs. 22 [11 to 50] h, P < 0.001) but did not differ among those who died in the ICU (46 [16 to 129] h in the frail vs. 31 [9 to 128] h in the non frail, P = 0.67).

The proportion of deaths within 30 days increased with the level of frailty (Fig. 2), and the hazard for death within 30 days from ICU admission increased with CFS in a similar pattern (Table 2). The unadjusted survival estimates of frail and nonfrail patients, categorised as described above and plotted as Kaplan–Meier curves, are displayed in Fig. 5. The corresponding hazard for death within 30 days in frail vs. nonfrail patients was also markedly increased when adjusted by SAPS3, limitations of treatment, comorbidities, age and sex (Table 2).

Table 2:
Survival analysis of patients (Cox model; all patients censored at 30 days) unadjusted, and adjusted by severity of illness score (simplified acute physiology score, third version), comorbidities, limitations of treatment, age and sex
Fig. 5:
Unadjusted survival estimates by frailty status (frail in red, nonfrail in blue) in intensive care patients up to 180 days after ICU admission.


In this study, we found that premorbid frailty is a predictor of death in ICU patients. The study also suggests a strengthened predictive ability of severity of illness scores in clinical use (in this case SAPS3) when combined with an assessment of a patient's degree of frailty. Even when adjusted for severity of illness and comorbidities as reflected by SAPS3, limitations of treatment, age and sex, the risk of death remained increased in frail patients.

A number of different classification systems have been developed for identifying frailty, all focusing on different aspects of the frail patient.1,5,19,20 One of these is the clinical frailty scale (CFS), originally a seven-step scale, later revised to nine steps, corresponding to increasing levels of frailty.5 The CFS is based on clinical judgement and focuses on how increasing degrees of frailty translate into a patient's everyday life, and it correlates strongly with the other more detailed scores available. The CFS has gained much interest for its ease of application in the clinical context, and it is now well validated in several fields of medicine.5 In intensive care the CFS is the most-studied tool for assessing a patient's level of frailty, and was therefore chosen for this study. Our results support previous studies on frailty in intensive care patients, in that a patient's degree of frailty can be assessed using the CFS as a simple bedside tool.

In our study, the cut-off CFS level providing the best sensitivity and specificity for predicting death at 30 days was identified at CFS = 5. This is in line with most studies on CFS, where a CFS of 5 or higher is the usual threshold used for classification of patients as frail.5,7,9,10 However, when used for categorisation, a larger proportion of the patients were defined as frail than in previous studies.10,21 This may be due to our study being conducted in a tertiary level hospital where only the most severely ill are admitted and patients with less complex medical needs are treated in intermediate care units. Although not always defined as such, frailty in an individual patient may be one of the factors that affect the priorities regarding which patients are admitted to intensive care. Thus frailty and ICU bed availability may have biased the selection of patients for this study and thus affected the results.

The cohort of patients in this study differs from that of other studies in several ways. While others have demonstrated a higher proportion of female patients among the frail, we found no such difference. Furthermore, we found no difference between frail and nonfrail patients regarding adverse events in the ICU, despite the higher severity of illness scores and longer length of stay in frail patients. Also, the sources of admission differed from other studies in that frail patients were more often admitted from general wards, whereas a larger proportion of the nonfrail were admitted from the ED: this is also reflected by the longer length of in-hospital stay before ICU admission in the frail patients. This may indicate a higher risk of frail patients becoming more ill during their hospital stay, and may reflect a need for intensive care earlier in the course of the disease in the frail than in the nonfrail.22

Most studies on frailty in intensive care have focused on elderly patients, and it has been suggested that frailty be added to the clinical assessment in this patient group.7 However, among younger critically ill patients, the prevalence of frailty is higher than expected for their chronological age, and frailty is associated with a negative outcome in this patient group as well as in the elderly.10,23,24 We therefore included all adult patients admitted to our ICU and chose not to stratify them according to age, aiming to study frailty across the whole age range seen by clinicians. In line with previous data, frail patients were older than the nonfrail, but frailty predicted an increased risk of death even when adjusted for age. Our results therefore support previous suggestions on recognising frailty as an important prognostic marker in intensive care, and not only in the elderly patients.

Our data are in line with previous studies demonstrating how the risk of death increases with increasing degrees of frailty. A striking finding was that the hazard for death at 30 days was higher in patients classified as CFS 1 than in those classified as CFS 2, 3 or 4, although statistically significant only in comparison with CFS 3. Among the previously fit patients classified as CFS 1, there were deaths from devastating bacterial meningitis, multiple trauma, and cardiac arrest, leading to a disproportionately large number of deaths in CFS 1 patients. However, the group was small, and the causes of death were such that, presumably, the premorbid degree of frailty plays a minor role. Furthermore, and in line with previous studies, we demonstrate an increased risk of death in frail patients compared with the nonfrail, even after adjustment for severity of illness and comorbidities, as well as limitations of care, and age. This further highlights the need to account for a patient's degree of frailty in clinical decision-making, since frailty appears to extend beyond other well known markers of adverse outcome.24

Outcome prediction is a field of much interest in intensive care medicine. Models used for prognostication have relied primarily on measures of the acute physiological derangement close to ICU admission, which have been used for an estimated probability of survival. A number of different scoring systems have been in clinical use, one of the most widely used presently being the SAPS3.11,12 Although SAPS3 does include comorbidities, this translates to rather severe disease and does not account for the collective burden of several, by themselves, less severe comorbidities. Furthermore, SAPS3 includes no significant measure of premorbid functional status, disability or frailty. Frailty often overlaps with comorbidity, and the distinction between the two has been stressed by other authors.1,25 While an interaction between SAPS3 and CFS as predictors is possible, we found only a moderate correlation between CFS and SAPS3, indicating that CFS mirrors factors unaccounted for by SAPS3. Our results indicate a need to extend prediction scores beyond the more serious comorbidities and acute derangements, and suggest that frailty provides added value to intensive care prognostication.

A strength of the study is that we included all ICU admission, not only those with certain admissions or certain lengths of stay, and that those where data on outcomes were missing were excluded. The risk of bias should therefore be reasonably small. An additional strength is the fact that the study was based on a large number of patients.


Our study has limitations that need consideration. First, this was a single-centre study and local practices regarding ICU admission and treatment may have affected the results. Second, although we aimed to estimate the patient's degree of frailty before the acute illness, the assessment may have been affected by the events leading up to admission to, and the patient's status in, the ICU. We tried to lessen this by ensuring there was sufficient information available for making the assessment, often avoiding completing it on the day of admission. Third, as the patient's level of frailty was assessed by one of the treating physicians rather than dedicated research personnel, events surrounding the patient's admission, response to therapy, and any decisions on limitations of care, were not blinded. Fourth, the CFS has elements of inherent subjectivity, and the level of interrater variability was not tested in this study. Nonetheless, it is a well validated bedside assessment tool in various medical fields, and it has demonstrated a good interrater reliability.5,26 Last, while this study suggests that frailty could be included in prognostication in intensive care, further model calibration and performance tests remain to be undertaken.


Premorbid frailty is common among unselected general ICU patients, and predicts a higher mortality in and after the ICU stay. Frailty needs to be recognised as a prognostic marker, and may be a valuable addition to established models for outcome prediction in intensive care. A better understanding of the implications of frailty may contribute to better informed decision-making before, during and after intensive care, and help moderate the expectations on intensive care outcome.

Acknowledgements relating to this article

Assistance with the study: none.

Financial support and sponsorship: departmental funding only.

Conflicts of interest: the authors have no competing interests.

Authors's contributions: study design and conception (LDG and AOT). Data acquisition, analysis and/or interpretation (LDG, MF and AOT). Article writing, critical revision and final approval (LDG, MF and AOT).

Presentation: none.


1. Fried LP, Ferrucci L, Darer J, et al. Untangling the concepts of disability, frailty, and comorbidity: implications for improved targeting and care. J Gerontol A Biol Sci Med Sci 2004; 59:255–263.
2. Clegg A, Young J, Iliffe S, et al. Frailty in elderly people. Lancet 2013; 381:752–762.
3. Dasgupta M, Rolfson DB, Stolee P, et al. Frailty is associated with postoperative complications in older adults with medical problems. Arch Gerontol Geriatr 2009; 48:78–83.
4. Makary MA, Segev DL, Pronovost PJ, et al. Frailty as a predictor of surgical outcomes in older patients. J Am Coll Surg 2010; 210:901–908.
5. Rockwood K, Song X, MacKnight C, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ 2005; 173:489–495.
6. Ekerstad N, Swahn E, Janzon M, et al. Frailty is independently associated with 1-year mortality for elderly patients with non-ST-segment elevation myocardial infarction. Eur J Prev Cardiol 2014; 21:1216–1224.
7. Flaatten H, De Lange DW, Morandi A, et al. The impact of frailty on ICU and 30-day mortality and the level of care in very elderly patients (>/=80 years). Intensive Care Med 2017; 43:1820–1828.
8. Bagshaw SM, Webb SA, Delaney A, et al. Very old patients admitted to intensive care in Australia and New Zealand: a multicentre cohort analysis. Crit Care 2009; 13:R45.
9. McDermid RC, Stelfox HT, Bagshaw SM. Frailty in the critically ill: a novel concept. Crit Care 2011; 15:301.
10. Bagshaw M, Majumdar SR, Rolfson DB, et al. A prospective multicenter cohort study of frailty in younger critically ill patients. Crit Care 2016; 20:175.
11. Metnitz PG, Moreno RP, Almeida E, et al. SAPS 3 – from evaluation of the patient to evaluation of the intensive care unit. Part 1: Objectives, methods and cohort description. Intensive Care Med 2005; 31:1336–1344.
12. Moreno RP, Metnitz PG, Almeida E, et al. SAPS 3 – from evaluation of the patient to evaluation of the intensive care unit. Part 2: Development of a prognostic model for hospital mortality at ICU admission. Intensive Care Med 2005; 31:1345–1355.
13. Rockwood K. Permission for use agreement. 2018.
14. Collins GS, Reitsma JB, Altman DG, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ 2015; 350:g7594.
15. The Swedish Intensive Care Registry.
16. Ruopp MD, Perkins NJ, Whitcomb BW, et al. Youden Index and optimal cut-point estimated from observations affected by a lower limit of detection. Biom J 2008; 50:419–430.
17. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988; 44:837–845.
18. Van Calster B, Vickers AJ, Pencina MJ, et al. Evaluation of markers and risk prediction models: overview of relationships between NRI and decision-analytic measures. Med Decis Making 2013; 33:490–501.
19. Rockwood K, Stadnyk K, MacKnight C, et al. A brief clinical instrument to classify frailty in elderly people. Lancet 1999; 353:205–206.
20. Pugh RJ, Ellison A, Pye K, et al. Feasibility and reliability of frailty assessment in the critically ill: a systematic review. Crit Care 2018; 22:49.
21. Bagshaw SM, Stelfox HT, McDermid RC, et al. Association between frailty and short- and long-term outcomes among critically ill patients: a multicentre prospective cohort study. CMAJ 2014; 186:E95–E102.
22. Bagshaw SM, Muscedere J. Is this intensive care unit patient frail? Unraveling the complex interplay between frailty and critical illness. Am J Respir Crit Care Med 2017; 196:4–5.
23. McDermid RC, Bagshaw SM. Scratching the surface: the burden of frailty in critical care. Intensive Care Med 2014; 40:740–742.
24. Muscedere J, Waters B, Varambally A, et al. The impact of frailty on intensive care unit outcomes: a systematic review and meta-analysis. Intensive Care Med 2017; 43:1105–1122.
25. Singh M, Alexander K, Roger VL, et al. Frailty and its potential relevance to cardiovascular care. Mayo Clin Proc 2008; 83:1146–1153.
26. Ekerstad N, Swahn E, Janzon M, et al. Frailty is independently associated with short-term outcomes for elderly patients with non-ST-segment elevation myocardial infarction. Circulation 2011; 124:2397–2404.

Supplemental Digital Content

Copyright © 2020 European Society of Anaesthesiology. All rights reserved.