Fronczek et al. Crit Care (2021) 25:231 https://doi.org/10.1186/s13054-021-03632-3 RESEARCH Open Access Relationship between the Clinical Frailty Scale and short‑term mortality in patients ≥ 80 years old acutely admitted to the ICU: a prospective cohort study Jakub Fronczek1, Kamil Polok1, Dylan W. de Lange2, Christian Jung3, Michael Beil4, Andrew Rhodes5, Jesper Fjølner6, Jacek Górka1, Finn H. Andersen7,8, Antonio Artigas9, Maurizio Cecconi10,11, Steffen Christensen6, Michael Joannidis12, Susannah Leaver13, Brian Marsh14, Alessandro Morandi15, Rui Moreno16, Sandra Oeyen17, Christina Agvald‑Öhman18, Bernardo Bollen Pinto19, Joerg C. Schefold20, Andreas Valentin21, Sten Walther22, Ximena Watson5, Tilemachos Zafeiridis23, Sigal Sviri24, Peter Vernon van Heerden25, Hans Flaatten26,27, Bertrand Guidet28, Wojciech Szczeklik1* and for the VIP1, VIP2 study group Abstract Background: The Clinical Frailty Scale (CFS) is frequently used to measure frailty in critically ill adults. There is wide variation in the approach to analysing the relationship between the CFS score and mortality after admission to the ICU. This study aimed to evaluate the influence of modelling approach on the association between the CFS score and short-term mortality and quantify the prognostic value of frailty in this context. Methods: We analysed data from two multicentre prospective cohort studies which enrolled intensive care unit patients ≥ 80 years old in 26 countries. The primary outcome was mortality within 30-days from admission to the ICU. Logistic regression models for both ICU and 30-day mortality included the CFS score as either a categorical, continu‑ ous or dichotomous variable and were adjusted for patient’s age, sex, reason for admission to the ICU, and admission Sequential Organ Failure Assessment score. Results: The median age in the sample of 7487 consecutive patients was 84 years (IQR 81–87). The highest fraction of new prognostic information from frailty in the context of 30-day mortality was observed when the CFS score was treated as either a categorical variable using all original levels of frailty or a nonlinear continuous variable and was equal to 9% using these modelling approaches (p < 0.001). The relationship between the CFS score and mortality was nonlinear (p < 0.01). Conclusion: Knowledge about a patient’s frailty status adds a substantial amount of new prognostic information at the moment of admission to the ICU. Arbitrary simplification of the CFS score into fewer groups than originally intended leads to a loss of information and should be avoided. *Correspondence: wojciech.szczeklik@uj.edu.pl 1 Department of Medicine, Center for Intensive Care and Perioperative Medicine, Jagiellonian University Medical College, ul. Skawińska 8, 31 – 066 Kraków, Poland Full list of author information is available at the end of the article © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​ mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Fronczek et al. Crit Care (2021) 25:231 Page 2 of 9 Trial registration NCT03134807 (VIP1), NCT03370692 (VIP2) Keywords: Intensive care units, Aged, 80 and over, Frailty, Prospective studies, Mortality Introduction pre-admission frailty and short-term mortality by means Progressive ageing is one of the leading issues in contem- of the CFS score using a large sample of very old inten- porary critical care [1]. Patients ≥ 80 years old account sive care unit patients (VIPs) acutely admitted to the ICU for 10 to 20% of all admissions to the intensive care unit within two international prospective cohort studies. We (ICU) , and this number is projected to increase sharply aimed to formally test the hypothesis that the association over next 3 decades [2]. The growing proportion of between frailty and ICU mortality is linear, describe the older adults admitted to ICUs will require substantial risk of ICU mortality across the entire spectrum of CFS resources, at the same time as many countries are already scores, and quantify the added prognostic value of frailty facing a shortage of ICU beds [3–5]. The high mortality in in patients ≥ 80 years old. the population of older critically ill adults has prompted researchers to identify factors which could help improve Methods the allocation of scarce intensive care resources and alle- Study design and patient population viate the suffering of patients and their families facing an The Very Old Intensive Care Patient (VIP1 and VIP2) uncertain prognosis [6]. studies were prospective, multicentre studies registered Among many geriatric syndromes, frailty may be on ClinicalTrials.gov (ID: NCT03134807, NCT03370692, the most useful concept in the context of severe illness respectively) [7, 11]. Both studies included patients requiring admission to the ICU [7]. Frailty is defined as aged 80 years or older admitted to the ICU with no spe- a clinically recognisable state of increased vulnerability cific exclusion criteria, and all acutely admitted patients resulting from ageing-associated decline in reserve and were eligible for this analysis. Sources of data, meth- function across multiple physiologic systems such that ods of measurement, and results of analyses performed the ability to cope with acute stressors is compromised previously on these data were described in detail else- [8, 9]. The presence of frailty is associated with increased where [7, 11]. In summary, all participating ICUs were morbidity in older adults, and the chances of 30-day sur- asked to include 20 consecutive VIPs over a 3-month vival after admission to the ICU are significantly lower in (VIP1) or 6-month (VIP2) period. Patients’ vital status frail patients compared to their fit counterparts [10–12]. within 30 days of admission to the ICU was assessed The Clinical Frailty Scale (CFS) is a widely used by inspecting hospital records, direct contact with the frailty assessment tool which has gained recognition in patient, or checking a national registry. Patients were a variety of clinical settings over the past decade [13]. recruited between October 2016 and February 2017 Unlike many other instruments describing the degree (VIP1 study) and between May 2018 and May 2019 of frailty, the CFS score was shown to be reliable in the (VIP2 study). All procedures performed in studies involv- highly dynamic context of acute admission to the ICU ing human participants were in accordance with the in patients ≥ 80 years old [14, 15]. Frailty is most often ethical standards of the institutional and/or national defined as a CFS score greater than or equal to 5, with research committee and with the 1964 Declaration of a CFS score of 4 labelled as pre-frailty (vulnerability, liv- Helsinki and its later amendments or comparable ethi- ing with mild frailty) and CFS scores between 1 and 3 cal standards. Each country had a national coordinator referred to as fitness. However, recent evidence shows responsible for application for national or regional ethi- that no clinically relevant worsening in prognosis occurs cal and regulatory study approval. Institutional research at the transition of pre-frailty to frailty, suggesting that ethic board approval was obtained from each study site. such categorisation does not accurately reflect the rela- Some countries were allowed to recruit patients with- tionship between the CFS and patients’ outcomes [16]. out prior written informed consent, while others had Analysing the CFS score as a continuous variable pro- to collect informed consent from patient or their legal vides a solution to the problem of arbitrary stratification representative. of patients according to pre-defined thresholds, but at the same time makes a strong assumption that the risk of an Outcomes and procedure outcome of interest increases linearly with increasing lev- Outcomes in this study were ICU mortality and 30-day els of frailty. mortality. Demographic data which included patients’ The objective of this study was to establish the opti- age, sex, and reason for admission to the ICU were col- mal approach to analysing the association between lected in each eligible participant. Severity of organ Fronczek et al. Crit Care (2021) 25:231 Page 3 of 9 dysfunction within the first 24 h after admission to the ing for nonlinear relationships and avoiding the ICU was assessed using the cumulative Sequential Organ known problems resulting from categorization of Failure Assessment (SOFA) score (from 0 to 24 points continuous variables (e.g. loss of statistical power, assigned in the cardiovascular, respiratory, renal, neu- increased risk of false positive findings, underestima- rologic, hepatic, and coagulation system with increasing tion of variation in outcomes between groups) [20]. number of points corresponding to more severe organ Wald tests were used to assess departures from lin- failure; the highest summary score observed within the earity of log-odds. first 24 h was reported) [17]. Necessity for organ support 4. Simplified grouping: we labelled patients as fit (vasoactive drugs, intubation and mechanical ventilation, (CFS < 4), pre-frail (CFS = 4), or frail (CFS ≥ 5) and and renal replacement therapy) was documented during constructed the model using such created categorical the ICU stay. variable with three levels. We used the Clinical Frailty Scale (CFS) to describe 5. Dichotomisation: we defined frailty using a single patients’ frailty before admission to the hospital with 9 threshold (CFS ≥ 5) and included a binary indicator possible classes from very fit prior to the acute illness to of patient’s frailty in the model. terminally ill [18]. Information necessary to perform this assessment was given by patients or proxy or obtained The reference category for calculating odds ratios from patient records. The CFS visual along with a sim- (ORs) was CFS = 1 for the categorical, linear, and nonlin- ple description was used with permission. We performed ear approach. Simplified grouping and dichotomisation separate analyses using the 8-point variant of the scale used CFS < 4 and CFS < 5 as a reference to calculate ORs, (from fit to very severely frail) and the 9-point scale, respectively. where patients in the last category are not expected to Age was modelled using restricted cubic splines with survive more than 6 months but are not otherwise evi- four knots at the 5th, 35th, 65th, and 95th quantile of its dently frail. Primary analyses were based on the 8-point distribution in all analyses. We summarized the added version of this tool [16]. prognostic value of frailty as the fraction of new informa- tion (proportion of explainable variation that is explained Statistical analysis by frailty or ratio of variances of predicted values before We used the likelihood ratio test to answer the ques- and after adding frailty to the model containing only tion whether frailty adds new information about the risk baseline variables described above) [19]. Descriptive of death in the ICU and at 30-day follow-up to baseline statistics on baseline variables are presented as median variables available at admission in a logistic regression (interquartile range [IQR]) or count and percentage. model. (i.e. age, sex, reasons for admission to the ICU, All analyses were performed using R version 3.6.0 (R and the SOFA score) [19]. Statistical models were built Project). based on clinical rationale and utilised all baseline vari- ables available in the merged dataset. We decided to per- Results form a complete-case analysis as the overall amount of The combined number of acutely admitted patients in missing data was relatively small. A detailed inspection of both the VIP 1 and VIP 2 studies was equal to 8173. The missing data was performed and presented in Additional median number of subjects recruited per country was file 1. 166 (IQR 33–482), and the two most actively recruiting We used five different approaches to analysing frailty in countries were UK (1619) and France (971). Survival sta- logistic regression models: tus within 30 days from admission to the ICU was known in 94.5% of the patients, and 99.5% of the patients had 1. Categorical: we modelled the CFS score as a categori- complete data on ICU survival. Among patients whose cal variable using all original levels (i.e. 8 or 9 catego- survival status was known, 96.9% had complete data on ries in the main and supplementary analysis, respec- all remaining variables. Patterns of missing data as well as tively). a comparison of patients with complete and incomplete 2. Linear: CFS was treated as a continuous variable data on survival status are provided in Additional file 1: without allowing for nonlinearities in the log-odds Figures S1–S3 and Tables S1 and S2. for mortality (i.e. a standard technique used as a Descriptive statistics, overall and among different cate- default setting in statistical software). gories of the CFS score of the analysed sample (n = 7487), 3. Nonlinear: the relationship between the CFS and the are presented in Table 1. Patients classified as very frail outcome was modelled using restricted cubic splines (CFS = 8) and terminally ill (CFS = 9) had the highest with five knots at the 5th, 27.5th, 50th, 72.5th, and SOFA scores and the highest incidence of vasoactive 95th quantile of the distribution of CFS, thus allow- drugs and mechanical ventilation initiation compared Table 1 Sample characteristics Variable Overall CFS = 1 CFS = 2 CFS = 3 CFS = 4 CFS = 5 CFS = 6 CFS = 7 CFS = 8 CFS = 9 n 7487 319 756 1723 1446 1097 1127 730 246 43 Male sex (n, %) 3934 (52.5) 200 (62.7) 455 (60.2) 988 (57.3) 783 (54.1) 530 (48.3) 504 (44.7) 333 (45.6) 122 (49.6) 19 (44.2) Fronczek et al. Crit Care Age, years (median [IQR]) 84.0 [81.0, 87.0] 82.0 [81.0, 85.0] 83.0 [81.0, 86.0] 83.0 [81.0, 86.0] 84.0 [81.0, 86.0] 84.0 [82.0, 87.0] 84.0 [82.0, 88.0] 85.0 [82.0, 88.0] 84.0 [82.0, 87.0] 83.0 [81.0, 87.0] SOFA score (median [IQR]) 7.0 [4.0, 10.0] 6.0 [3.0, 8.0] 6.0 [4.0, 10.0] 6.0 [4.0, 10.0] 6.0 [4.0, 10.0] 7.0 [4.0, 10.0] 7.0 [4.0, 10.0] 8.0 [5.0, 10.0] 9.0 [6.0, 12.0] 9.0 [7.5, 12.5] Reason for admission (n, %) Respiratory failure 1809 (24.2) 67 (21.0) 154 (20.4) 358 (20.8) 374 (25.9) 278 (25.3) 300 (26.6) 191 (26.2) 73 (29.7) 14 (32.6) Circulatory failure 1057 (14.1) 32 (10.0) 111 (14.7) 240 (13.9) 221 (15.3) 157 (14.3) 166 (14.7) 100 (13.7) 23 (9.3) 7 (16.3) (2021) 25:231 Combined respiratory/ 874 (11.7) 17 (5.3) 58 (7.7) 165 (9.6) 183 (12.7) 137 (12.5) 141 (12.5) 116 (15.9) 50 (20.3) 7 (16.3) circulatory failure Sepsis 992 (13.2) 26 (8.2) 80 (10.6) 242 (14.0) 159 (11.0) 162 (14.8) 152 (13.5) 119 (16.3) 45 (18.3) 7 (16.3) Emergency surgery 900 (12.0) 49 (15.4) 96 (12.7) 246 (14.3) 166 (11.5) 131 (11.9) 137 (12.2) 60 (8.2) 14 (5.7) 1 (2.3) Non-traumatic cerebral 477 (6.4) 38 (11.9) 67 (8.9) 104 (6.0) 84 (5.8) 52 (4.7) 68 (6.0) 48 (6.6) 12 (4.9) 4 (9.3) pathology Multiple trauma without 141 (1.9) 12 (3.8) 28 (3.7) 39 (2.3) 25 (1.7) 14 (1.3) 18 (1.6) 5 (0.7) 0 (0.0) 0 (0.0) head injury Multiple trauma with head 128 (1.7) 15 (4.7) 24 (3.2) 36 (2.1) 18 (1.2) 18 (1.6) 9 (0.8) 7 (1.0) 1 (0.4) 0 (0.0) injury Isolated head injury 180 (2.4) 10 (3.1) 29 (3.8) 43 (2.5) 30 (2.1) 20 (1.8) 29 (2.6) 13 (1.8) 4 (1.6) 2 (4.7) Other 929 (12.4) 53 (16.6) 109 (14.4) 250 (14.5) 186 (12.9) 128 (11.7) 107 (9.5) 71 (9.7) 24 (9.8) 1 (2.3) NIV (n, %) 1854 (24.8) 71 (22.3) 155 (20.5) 383 (22.2) 375 (25.9) 298 (27.2) 306 (27.2) 200 (27.4) 53 (21.5) 13 (30.2) Intubation (n, %) 3804 (50.8) 165 (51.7) 421 (55.7) 885 (51.4) 731 (50.6) 510 (46.5) 551 (48.9) 357 (48.9) 159 (64.6) 25 (58.1) Vasoactive drugs (n, %) 4400 (58.8) 173 (54.2) 436 (57.7) 1004 (58.3) 837 (57.9) 646 (58.9) 677 (60.1) 436 (59.7) 162 (65.9) 29 (67.4) Renal replacement therapy 820 (11.0) 24 (7.5) 72 (9.5) 208 (12.1) 165 (11.4) 101 (9.2) 126 (11.2) 76 (10.4) 43 (17.5) 5 (11.6) (n, %) Life sustaining treatment limitation None 4930 (65.8) 239 (74.9) 571 (75.5) 1236 (71.7) 982 (67.9) 688 (62.7) 669 (59.4) 370 (50.7) 154 (62.6) 21 (48.8) Withdrawal 1108 (14.8) 39 (12.2) 107 (14.2) 234 (13.6) 194 (13.4) 166 (15.1) 183 (16.2) 135 (18.5) 39 (15.9) 11 (25.6) Withholding 1449 (19.4) 41 (12.9) 78 (10.3) 253 (14.7) 270 (18.7) 243 (22.2) 275 (24.4) 225 (30.8) 53 (21.5) 11 (25.6) Length of stay in the ICU, 3.00 [1.00, 7.93] 5.10 [1.87, 0.82] 3.00 [0.67, 8.85] 3.30 [1.00, 8.00] 3.80 [1.10, 9.00] 2.50 [1.00, 7.20] 3.00 [1.00, 7.00] 2.20 [0.80, 6.05] 3.00 [1.00, 8.20] 2.30 [0.80, 5.47] days, non-survivors (median [IQR]) Length of stay in the ICU, 3.30 [1.70, 7.00] 3.80 [1.80, 8.00] 3.50 [1.80, 6.93] 3.60 [1.80, 7.82] 3.50 [1.70, 7.00] 3.15 [1.50, 6.53] 3.00 [1.60, 6.23] 3.00 [1.35, 6.25] 6.20 [2.00,15.00] 3.70 [1.00,12.50] days, survivors (median [IQR]) Death in the ICU (n, %) 2024 (27.0) 60 (18.8) 188 (24.9) 403 (23.4) 369 (25.5) 273 (24.9) 331 (29.4) 259 (35.5) 113 (45.9) 28 (65.1) Death in 30-day follow-up 2985 (39.9) 91 (28.5) 254 (33.6) 576 (33.4) 545 (37.7) 416 (37.9) 513 (45.5) 406 (55.6) 149 (60.6) 35 (81.4) (n, %) IQR interquartile range, ICU intensive care unit, NIV non-invasive ventilation, SOFA Sequential Organ Failure Assessment Page 4 of 9 Fronczek et al. Crit Care (2021) 25:231 Page 5 of 9 to participants with lower CFS scores. The length of stay The magnitude of the estimated effect of frailty on was by far the longest in very frail patients who survived mortality was greater in the 30-day mortality logistic until discharge from the ICU. regression compared to the ICU mortality model, and the Results from both univariate (Additional file 1: Figure slope of increase in the odds of death was steeper with S4) and multivariable logistic regression models provided increasing CFS scores. When patients were dichotomised evidence that the relationship between the CFS score and as either frail (CFS ≥ 5) or not frail (CFS < 5), frailty was mortality within 30 days is not linear (Fig. 1, Additional associated with an OR for mortality in the ICU and file 1: Figure S5). within 30 days equal to 1.22 (95% CI 1.09–1.36) and 1.50 The OR for ICU mortality was statistically significantly (95% CI 1.35–1.66), respectively. different from unity at a CFS score of 6 using both the The fraction of new information from the categorical categorical and nonlinear modelling approach based on CFS score over and above data on patient’s age, sex, rea- restricted cubic splines. In case of 30-day mortality, an son for admission to the ICU, and the SOFA score across apparent increase in the odds of death at a CFS score different modelling strategies is shown in Table 2. equal to 6 was also apparent, though the lower bound- Categorical and nonlinear modelling of the CFS was ary of the 95% confidence interval (CI) surpassed unity associated with the highest information yield, and already at a CFS score of 4 and 5 using the categorical and dichotomising patients into either frail or not frail led nonlinear approach respectively. Evidence that vulner- to the greatest loss of information in both the ICU and ability (CFS 4 vs. CFS < 4) is associated with an increased 30-day mortality models. Likelihood ratio tests indicated mortality was sufficient only in 30-day follow-up. that the addition of frailty significantly improves model Modelling approach Categorisation Linear Nonlinear Simplified grouping Modelling approach Categorisation Linear Nonlinear Simplified grouping Odds ratio for death in 30-day follow-up (95% CI) Odds ratio for death in the ICU (95% CI) 3.0 3.0 1.5 1.5 1.0 1.0 1 2 3 4 5 6 7 8 Vulnerable Frail 1 2 3 4 5 6 7 8 Vulnerable Frail Clinical Frailty Scale (CFS) score Clinical Frailty Scale (CFS) score Fig. 1 The impact of modelling approach on the association between frailty and short-term mortality. Simplified grouping used CFS < 4 as a reference, with vulnerability defined as a CFS = 4 and frailty as a CFS ≥ 5; CFS = 1 was used as a reference in the three other modelling approaches shown in the figure. Odds ratios obtained after dichotomisation of the CFS score are described in the text. The relationship between the CFS score and mortality was nonlinear in both models (p values < 0.01). Adjusted for age, sex, reason for admission to the ICU, and admission SOFA score Table 2 Fraction of new prognostic information from including frailty in the model Fraction of new information p value (likelihood ratio test) ICU mortality model (%) 30-day mortality model (%) ICU mortality model 30-day mortality model Categorisation 3 9 < 0.001 < 0.001 Linear 2 8 < 0.001 < 0.001 Nonlinear 3 9 < 0.001 < 0.001 Simplified grouping 1 6 0.002 < 0.001 Dichotomisation 1 5 < 0.001 < 0.001 Fronczek et al. Crit Care (2021) 25:231 Page 6 of 9 fit in all cases (all p values < 0.01). Results of sensitivity a CFS ≥ 5 threshold to guide ICU admission [16]. The analyses based on the 9-point version of the CFS score authors noted little difference between vulnerable, mild, are presented in Additional file 1: Figures S6 and S7 and and moderate frailty categories in terms of the adjusted Table S3. The adjusted probability of death across differ- risk of mortality. Another interesting finding was that the ent CFS scores is shown in Fig. 2. CFS score behaved differently in patients ≥ 65 years old than in younger adults. The observed increase in mortal- Discussion ity with increasing CFS scores was apparently linear in We found that the relationship between frailty and the whole cohort and gradually accelerating in the older 30-day mortality is not linear in patients ≥ 80 years old subgroup. acutely admitted to the ICU. Using separate logistic Results from a combined database of patients regression models, we showed that the prognostic con- ≥ 80 years old enrolled in the VIP1 and VIP2 studies cor- tribution of frailty was larger in case of 30-day mortality roborate the results of Darvall and colleagues in the sense compared to ICU mortality. The association between the that there is little evidence of a difference in outcomes CFS score and mortality was highly dependent on how between patients whose CFS score ranges between 1 the score was handled in statistical analysis. Defining and 5 after adjustment for basic patient characteristics frailty as a CFS ≥ 5 led to a substantial underestimation and that a threshold value of CFS ≥ 5 is not justified of the odds of death for patients whose CFS score was as a marker of frailty in the context of mortality within equal to 7 or 8 and resulted in a loss of more than 50% of 30 days from admission to the ICU. Possibly thanks to new prognostic information compared to using all levels the larger sample size, we were able to confirm that a CFS of the 9-point Clinical Frailty Scale. The categorical and score of 6, not 7, is already associated with an increased nonlinear approach to modelling frailty best captured the risk of death. This minor discrepancy may either reflect relationship between the CFS score and mortality and a higher statistical power in current analyses or stem should be preferred over the other inspected methods of from the fact that our study enrolled a broad spectrum analysing frailty in future studies evaluating mortality in of patients admitted to the ICU and was not focused this population of patients. exclusively on patients with pneumonia. Unlike Darvall A scoping review of the utilisation of the CFS score in and colleagues, we found evidence that the relationship the literature showed that 40 out of 58 papers (68.9%) between frailty as described by the CFS score and mortal- using a threshold value of the CFS score to define frailty ity is nonlinear, which is in line with previous theoretical were based on a cut-off of 5, with 25.9% of publica- considerations about the expected character of increase tions based on a threshold of 4 and only 3.4% labelling in mortality with accumulating deficits [21]. patients with a CFS score ≥ 6 as frail [13]. Results of large The association between frailty and mortality was more registry-based study describing differences in outcomes pronounced in 30-day follow-up than when assessed up of critically ill patients with pneumonia depending on to the moment of death or discharge from the ICU. Sev- their level of frailty conducted recently did not support eral factors may underpin this finding, yet the most likely Modelling approach Categorisation Linear Nonlinear Modelling approach Categorisation Linear Nonlinear Probability of death in 30-day follow-up (95% CI) 60% 60% Probability of death in the ICU (95% CI) 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 Clinical Frailty Scale score Clinical Frailty Scale score Fig. 2 The impact of modelling approach on the model-calculated risk of death. Adjusted probabilities displayed for an 84-year-old male admitted due to respiratory reasons with a SOFA score equal to 7 Fronczek et al. Crit Care (2021) 25:231 Page 7 of 9 explanation is that despite surviving the acute phase of using the CFS score as either a categorical variable critical illness, frail patients ultimately reach a point of comprising all the original levels or a continuous vari- exhaustion of their functional reserves and deteriorate as able while allowing for nonlinearities in the estimated a result of the initial stressor. It needs to be highlighted effect of frailty on outcome is likely to lead to the least that despite not finding a clear association between lower amount of residual confounding and improve inference CFS scores and mortality, it is reasonable to believe that compared to simpler approaches to modelling frailty patients with wild milder frailty fare worse in terms of such as dichotomisation or forced linearity [26, 27]. functional outcomes and quality of life in long-term This study has several limitations. First, ICUs observation after being admitted to the ICU [22–24]. included in the VIP1 and VIP2 studies participated A more general claim that can be made based on the voluntarily and were not sampled at random. Second, results of our study is that any attempts to define frailty informed consent from patients or their legal repre- as a condition that is either present or absent are destined sentatives had to be collected before enrolment in most to fail. Despite the conceptual ease and convenience of participating countries due to national differences in treating patients as fit or frail both from a researcher’s data protection regulations which may affect the gen- and practicing clinician’s perspective, we must not forget eralizability of our results to the entire population of that frailty is a continuum. Loss of prognostic informa- acutely admitted patients. Third, a small fraction of tion is inevitable whenever arbitrary grouping of patients patients had still been hospitalized in the ICU after takes place, particularly when trying to transform a 30 days from admission, thus increasing the uncer- robust score such as the CFS into a binary screening tainty about actual mortality rates observed within tool. In practice, decisions such as whether to admit a our study. Fourth, the pragmatic nature of the VIP1 patient to the ICU or not have to be made at some point and VIP2 studies precluded routine collection of more based on a context-specific threshold [25]. Nevertheless, detailed data on patients’ chronic health status prior to efforts should be made to present the clinician with the admission to the ICU, potentially leaving a substantial most accurate and comprehensive picture of the patient’s part of variation in the modelled outcome unexplained prospects and let her or him utilise the full potential of and possibly inflating the fraction of new prognostic analysed data (i.e. the probability of an event based on information from frailty. Fifth, we focused exclusively a granular assessment of patient’s frailty, not merely the on mortality and did not trace post-discharge health information whether an arbitrary CFS threshold had trajectories of VIPs (e.g. in terms of functional capacity, been crossed). cognitive impairment, and mental health) in long-term Even with the best available estimates of prognosis observation. based on robust statistical models, the decision-making The strength of our study is that we prospectively process in the context of intensive care provision for enrolled a large, contemporary cohort of consecutive older adults will remain a multidimensional challenge patients ≥ 80 years old admitted to ICUs in 26 coun- shaped by a plethora of contextual and temporal factors tries for various reasons. The collection of standardized that are difficult to capture in a set of static regression data from nearly 8000 VIPs and the use of contempo- coefficients. The evidence to inform risk–benefit assess- rary statistical modelling techniques provided an ample ment on admission to the ICU and further into the course opportunity to answer the question about the opti- of critical illness in vulnerable populations is accumulat- mal approach to analysing frailty in this population of ing, with more detailed data on functional outcomes in patients. long-term observation to come in the near future. It is in our patients’ interest that we transform this evidence into policies efficiently, understanding the consequences of Conclusions choices made at the stage of data analysis. If the adjusted In conclusion, frailty contributes up to 9% of new prog- probability of death within 30 days from admission to the nostic information about 30-day mortality in the pres- ICU with a CFS score of 5, 6, and 7 is estimated at 39%, ence of basic patient characteristics. We showed that the 47%, and 55%, as illustrated in our study, where should increase in the risk of death within 30 days from admis- we draw a demarcation line for prognostically important sion to the ICU is mild at best in lower categories of the frailty, if at all? Before attempting to answer this question, CFS score and more abrupt in patients whose CFS score one must first acknowledge the variability in outcomes is equal to 6 or greater. To accurately describe the rela- between particular categories of the CFS score that arbi- tionship between frailty and short-term mortality, future trary cut-offs unfortunately discard altogether. studies should analyse the CFS score either as a nonlinear In observational studies attempting to adjust for con- continuous variable or a categorical variable while retain- founding by frailty or quantify its impact on outcomes, ing all the original levels of frailty. Fronczek et al. Crit Care (2021) 25:231 Page 8 of 9 Supplementary Information Heinrich-Heine-University, Düsseldorf, Germany. 4 Medical Intensive Care Unit, Hadassah Medical Center, Jerusalem, Israel. 5 St George’s University Hospitals The online version contains supplementary material available at https://​doi.​ NHS Foundation Trust, London, London, UK. 6 Department of Intensive Care, org/​10.​1186/​s13054-​021-​03632-3. Aarhus University Hospital, Århus, Denmark. 7 Department of Anaesthe‑ sia and Intensive Care, Ålesund Hospital, Ålesund, Norway. 8 Department Additional file 1. Figures S1–S3 and Tables S1 and S2. of Circulation and Medical Imaging, NTNU, Trondheim, Norway. 9 Critical Care Department, Corporacion Sanitaria Universitaria Parc Tauli, CIBER Enferme‑ dades Respiratorias, Autonomous University of Barcelona, Sabadell, Spain. 10 Acknowledgements Department of Anesthesia and Intensive Care Medicine, Humanitas Clinical The members of VIP1 and VIP2 study groups are listed in Additional file 1 and and Research Center – IRCCS, Via Alessandro Manzoni 56, 20089 Rozzano, MI, are listed in their individual PubMed records. Italy. 11 Department of Biomedical Sciences, Humanitas University, Pieve Ema‑ nuele, Rozzano, MI, Italy. 12 Division of Intensive Care and Emergency Medicine, Authors’ contributions Department of Internal Medicine, Medical University Innsbruck, Innsbruck, JF, HF, BG, KP, DWdL, CJ, and WS contributed to the conception and design Austria. 13 Research Lead Critical Care Directorate St George’s Hospital, London, of the study. JF, KP, DWdL, CJ, MB, AR, JF, JG, FHA, AA, MC, SC, MJ, SL, BM, AM, UK. 14 Mater Misericordiae University Hospital, Dublin, Ireland. 15 Department RM, SO, SAO, BBP, JCS, AV, SW, XW, TZ, SS, PVvH, HF, BG, and WS were involved of Rehabilitation Hospital Ancelle di Cremona Italy, Geriatric Research Group, in acquisition of the data. JF and KP contributed to statistical analysis. JF, KP, Brescia, Italy. 16 Faculdade de Ciências Médicas de Lisboa (Nova Médical DWdL, CJ, MB, AR, JF, JG, FHA, AA, MC, SC, MJ, SL, BM, AM, RM, SO, SAO, BBP, School), Unidade de Cuidados Intensivos Neurocríticos e Trauma, Hospital de JCS, AV, SW, XW, TZ, SS, PVvH, HF, BG, and WS were involved in interpretation of São José, Centro Hospitalar Universitário de Lisboa Central, Lisbon, Portugal. 17 data. JF, KP, DWdL, CJ, MB, AR, JF, JG, FHA, AA, MC, SC, MJ, SL, BM, AM, RM, SO, Department of Intensive Care 1K12IC, Ghent University Hospital, Ghent, SAO, BBP, JCS, AV, SW, XW, TZ, SS, PVvH, HF, BG, WS contributed to drafting the Belgium. 18 Karolinska University Hospital, Stockholm, Sweden. 19 Department article or revising it critically for important intellectual content. JF, KP, DWdL, CJ, of Anaesthesiology, Pharmacology and Intensive Care, Geneva University MB, AR, JF, JG, FHA, AA, MC, SC, MJ, SL, BM, AM, RM, SO, SAO, BBP, JCS, AV, SW, Hospitals, Geneva, Switzerland. 20 Department of Intensive Care Medicine, XW, TZ, SS, PVvH, HF, BG, and WS were involved in final approval of the version Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland. 21 to be submitted. Kardinal Schwarzenberg Hospital, Schwarzach, Austria. 22 Department of Cardiothoracic Surgery, Anesthesia and Intensive Care, Linköping University Funding Hospital and Department of Medical and Health Sciences, Linköping Uni‑ None. versity, Linköping, Sweden. 23 Intensive Care Unit, General Hospital of Larissa, Larissa, Greece. 24 Department of Medical Intensive Care, Hadassah Medical Availability of data and materials Center and Faculty of Medicine, Hebrew University of Jerusalem, Jerusalem, Data collected for the study (anonymized database) as well as additional Israel. 25 Department of Anesthesia, Intensive Care and Pain Medicine, Hadas‑ related documents (study protocol, statistical analysis plan, informed consent sah Medical Center and Faculty of Medicine, Hebrew University of Jerusalem, form) will be made available for the reviewers and editors of the Journal upon Jerusalem, Israel. 26 Department of Anaesthesia and Intensive Care, Haukeland request. Relevant documents will be made available by study investigators via University Hospital, Bergen, Norway. 27 Department of Clinical Medicine, password-protected data-sharing web page. University of Bergen, Bergen, Norway. 28 UPMC Univ Paris 06, INSERM, UMR_S 1136, Institut Pierre Louis d’Epidémiologie et de Santé Publique, Equipe: Epidémiologie Hospitalière Qualité et Organisation des Soins, Sorbonne Uni‑ Declarations versités, Assistance Publique - Hôpitaux de Paris, 75012 Paris, France. Ethics approval and consent to participate Received: 25 April 2021 Accepted: 6 June 2021 All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Each country had a national coordinator responsible for application for national or regional ethical and References regulatory study approval. Institutional research ethic board approval was 1. Flaatten H, de Lange DW, Artigas A, Bin D, Moreno R, Christensen S, et al. obtained from each study site. Some countries were allowed to recruit The status of intensive care medicine research and a future agenda for patients without prior written informed consent, while others had to collect very old patients in the ICU. Intensive Care Med. 2017;43:1319–28. informed consent from patient or their legal representative. 2. Nguyen Y-L, Angus DC, Boumendil A, Guidet B. The challenge of admit‑ ting the very elderly to intensive care. Ann Intensive Care. 2011;1:29. Consent for publication 3. Stelfox HT, Hemmelgarn BR, Bagshaw SM, Gao S, Doig CJ, Nijssen-Jordan Not applicable. C, et al. Intensive Care Unit bed availability and outcomes for hospi‑ talized patients with sudden clinical deterioration. Arch Intern Med. Competing interests 2012;172:467–74. Joerg C. Schefold declares that the Dept. of Intensive Care Medicine Bern has/ 4. Simchen E, Sprung CL, Galai N, Zitser-Gurevich Y, Bar-Lavi Y, Gurman G, had research and/or development/consulting contracts with (full disclosure): et al. Survival of critically ill patients hospitalized in and out of intensive Orion Corporation, Abbott Nutrition International, B. Braun Medical AG, CSEM care units under paucity of intensive care unit beds. Crit Care Med. SA, Edwards Lifesciences Services GmbH/SA, Kenta Biotech Ltd, Maquet Criti‑ 2004;32:1654–61. cal Care AB, Omnicare Clinical Research AG, and Nestlé. Educational grants 5. Town JA, Churpek MM, Yuen TC, Huber MT, Kress JP, Edelson DP. Relation‑ were received from Fresenius Kabi; GSK; MSD; Lilly; Baxter; Astellas; AstraZen‑ ship between ICU bed availability, ICU readmission, and cardiac arrest in eca; B. Braun Medical AG, CSL Behring, Maquet, Novartis, Covidien, Nycomed, the general wards. Crit Care Med. 2014;42:2037–41. Pierre Fabre Pharma (Roba Pharma); Pfizer, Orion Pharma. The money went 6. Guidet B, Vallet H, Boddaert J, de Lange DW, Morandi A, Leblanc G, et al. into departmental funds. No personal financial gain applies. All other authors Caring for the critically ill patients over 80: a narrative review. Ann Inten‑ do not have any conflict of interest to declare related to this manuscript. sive Care. 2018;8:1–15. 7. Guidet B, de Lange DW, Boumendil A, Leaver S, Watson X, Boulanger Author details C, et al. The contribution of frailty, cognition, activity of daily life and 1 Department of Medicine, Center for Intensive Care and Perioperative Medi‑ comorbidities on outcome in acutely admitted patients over 80 years in cine, Jagiellonian University Medical College, ul. Skawińska 8, 31 – 066 Kraków, European ICUs: the VIP2 study. Intensive Care Med. 2020;46:57–69. Poland. 2 Department of Intensive Care Medicine, University Medical 8. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, et al. Center, University Utrecht, Utrecht, The Netherlands. 3 Division of Cardiol‑ Frailty in older adults: Evidence for a phenotype. Journals Gerontol - Ser A ogy, Pulmonology and Vascular Medicine, University Hospital Düsseldorf, Biol Sci Med Sci. 2001;56:146–57. Fronczek et al. Crit Care (2021) 25:231 Page 9 of 9 9. Pel-Littel RE, Schuurmans MJ, Emmelot-Vonk MH, Verhaar HJJ. Frailty: 20. Altman DG, Royston P. The cost of dichotomising continuous variables. Br defining and measuring of a concept. J Nutr Heal Aging. 2009;13:390–4. Med J. 2006;332:1080. 10. Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in elderly peo‑ 21. Mitnitski AB, Mogilner AJ, MacKnight C, Rockwood K. The mortality rate ple. Lancet. 2013;381:752–62. as a function of accumulated deficits in a frailty index. Mech Ageing Dev. 11. Flaatten H, De Lange DW, Morandi A, Andersen FH, Artigas A, Berto‑ 2002;123:1457–60. lini G, et al. The impact of frailty on ICU and 30-day mortality and the 22. Bagshaw SM, Stelfox HT, McDermid RC, Rolfson DB, Tsuyuki RT, Baig N, level of care in very elderly patients (≥ 80 years). Intensive Care Med. et al. Association between frailty and short- and long-term outcomes 2017;43:1820–8. among critically ill patients: a multicentre prospective cohort study. 12. Jung C, Flaatten H, Fjølner J, Bruno RR, Wernly B, Artigas A, et al. The CMAJ. 2014;186:E95-102. impact of frailty on survival in elderly intensive care patients with COVID- 23. Heyland DK, Garland A, Bagshaw SM, Cook D, Rockwood K, Stelfox HT, 19: the COVIP study. Crit Care. 2021;25:149. et al. Recovery after critical illness in patients aged 80 years or older: a 13. Church S, Rogers E, Rockwood K, Theou O. A scoping review of the Clini‑ multi-center prospective observational cohort study. Intensive Care Med. cal Frailty Scale. BMC Geriatr. 2020;20:1–18. 2015;41:1911–20. 14. Pugh RJ, Thorpe CM, Subbe CP. A critical age: can we reliably measure 24. McIsaac DI, Taljaard M, Bryson GL, Beaulé PE, Gagne S, Hamilton G, et al. frailty in critical care? Crit Care. 2017;21:121. Frailty and long-term postoperative disability trajectories: a prospective 15. Flaatten H, Guidet B, Andersen FH, Artigas A, Cecconi M, Boumendil A, multicentre cohort study. Br J Anaesth. 2020;125:704–11. et al. Reliability of the Clinical Frailty Scale in very elderly ICU patients: a 25. Flaatten H, Beil M, Guidet B. Prognostication in older ICU patients: mission prospective European study. Ann Intensive Care. 2021;11:22. impossible? Br J Anaesth. 2020;125:655–7. 16. Darvall JN, Bellomo R, Bailey M, Paul E, Young PJ, Rockwood K, et al. Frailty 26. Kow CS, Hasan SS, Thiruchelvam K, Aldeyab M. Association of frailty and outcomes from pneumonia in critical illness: a population-based and mortality in patients with COVID-19: a meta-analysis. Br J Anaesth. cohort study. Br J Anaesth. 2020;125:730–8. 2021;126:e108–10. 17. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, Bruining 27. Sanchez D, Brennan K, Al Sayfe M, Shunker S-A, Bogdanoski T, Hedges S, H, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to et al. Frailty, delirium and hospital mortality of older adults admitted to describe organ dysfunction/failure. Intensive Care Med. 1996;22:707–10. intensive care: the Delirium (Deli) in ICU study. Crit Care. 2020;24:609. 18. Rockwood K, Song X, MacKnight C, Bergman H, Hogan DB, McDowell I, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ. 2005;173:489–95. Publisher’s Note 19. Harrell Jr FE. Statistically efficient ways to quantify added predictive value Springer Nature remains neutral with regard to jurisdictional claims in pub‑ of new measurements. Statistical Thinking. 2020. https://​www.​fharr​ell.​ lished maps and institutional affiliations. com/​post/​addva​lue/. Accessed 20 April 2021. Ready to submit your research ? Choose BMC and benefit from: • fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations • maximum visibility for your research: over 100M website views per year At BMC, research is always in progress. Learn more biomedcentral.com/submissions