Frailty and survival in elderly intensive care patients in Norway

Today, 10%‐15% of Norwegian intensive care patients are ≥80 years. This proportion will increase significantly over the next 20 years, but it is unlikely that resources for intensive care increase correspondingly. Thus, it is important to establish which patients among elderly people will benefit from intensive care. The main objective of the study was to investigate the relationships between geriatric scoring tools and 30‐day mortality.

and conversely, if there are particular groups who have little or no benefit from such treatment. This is both an ethically and healthpolitically important issue.
Advanced age, comorbidity, and disease severity at ICU admission predict partially the risk of death, but traditional intensive care scoring tools are inaccurate when applied in elderly ICU patients. 10,11 Hence, it is conceivable that other geriatric factors are more important for outcome than acute illness characteristics, and that geriatric scoring tools, therefore, may perform better than traditional intensive care scoring tools. That was the research question of two large European studies, VIP1 12 of VIP2. 13 However, intensive care medicine is differently organized in European countries, leading to a different patient population among countries. Thus, results from international studies cannot directly be transferred to each national context. We, therefore, present the specific result from the Norwegian population.
Based on a Norwegian cohort of intensive care patients aged 80 years or older at the time of ICU admission, we have studied the association among frailty, cognitive failure, comorbidity, the general health condition, and 30-day mortality.

| Design of study population
Data presented in this paper are based on two prospective observational studies (VIP1 12 and VIP2 13 ), which included intensive care patients ≥80 years in Europe. Norwegian participants from these studies were merged into one study population. Outcome in the studies was ICU survival and survival 30 days after ICU admission.
Inclusion criteria in both studies were acute ICU admissions of patients ≥80 years of age. VIP1 included also elective ICU admissions, but since this is not common practice in Norway, we consid-

| Data collection
For all patients in both studies, we collected demographic data (age, sex, and place of living before hospital admission), ICU length of stay (ICU-LOS), ICU treatment (vasoactive medication, mechanical ventilation (invasive and non-invasive), and renal replacement treatment (RRT)), and life-sustaining treatment (LST) limitation (withhold/withdrawal) during the ICU stay. The frailty status before the actual critical illness was scored with clinical frailty scale (CFS) 14 by either a physician or a nurse. Sources for CFS scoring were patients, proxies, or patient records. The following definitions for CFS were used: not frail = CFS 1-3, vulnerable = CFS 4, and frail = CFS ≥ 5. SOFA (Sequential Organ Failure Assessment) score was used for assessment of severity of illness. 15 The primary reason for ICU admission was grouped according to a list consisting of the following 11 categories: respiratory failure, circulatory failure, combined respiratory/circulatory failure, sepsis (severe sepsis in VIP1 and sepsis in VIP2), multitrauma without head injury, multitrauma with head injury, isolated head injury, postemergency surgery, intoxication, non-traumatic cerebral failure, and other causes (Table 1).
In VIP2, there was also collected information on place of living before hospital admission, duration of ICU treatment, and geriatric syndromes. For collection of data regarding the geriatric syndromes, the following scores were used: comorbidity and polypharmacy score (CPS 16

| Statistics
Patient characteristics were analyzed as percentages for categorical variables, while continuous variables were measured as means with standard deviation if the distribution was symmetrical, and medians with quartiles if the distribution was skewed. The main analysis was performed with logistic regression with 30-day mortality as outcome that can be used to visualize the association between variables in a data set with many variables. In a data set with only two variables, one may plot them against each other, but because we have six variables, we would need a six dimensional plot. However, with PCA we may still plot these variables using only two axes, but the cost is that some nuances are lost. This cost can be described as "explained variance." The six-dimensional plot mentioned earlier would have 100% explained variance. Missing information in the PCA analyses was imputed by using the R-packages FactoMineR and missMDA to regularized iterative PCA.

| Ethics
Both VIP1 and VIP2 included patients who usually lacked the ability to give an informed consent, and the ethical committees waived the need for consent at inclusion. Survivors later received information about the studies and were given the possibility to decline participation in the studies. Both studies were approved by the Regional

| RE SULTS
In total, 451 patients from Norway were included in VIP1 and VIP2 (VIP1: N = 215; VIP2: N = 236). Median age was 85 years, and 51.9% were women. Respiration and/or circulatory failure and sepsis were the main reasons for ICU admission in 67.1% of the cases. The percentage of patients admitted to university hospitals was 28.6%. The majority of patients lived in their own homes before hospital admission (81.3%; of whom 9.7% lived with their family) (  (Table 3).
We found no significant differences between patients with a CFS of 4 and CFS 5-9. Among patients in the group CFS 5-9, 152 of 234 patients had values 5 or 6.

1.41)). When all the scores were included in a multiple regression
analysis, we found no significant impact in the model. Figure 1 shows the different variables in the PCA relative to the first two principal components. These two components contribute with a total of 40.53% + 17.64% = 58.17% explained variance. As mentioned, including six dimensions would yield 100% explained variance. In other words, the first two principal components can be said to approximate the full data set very well. The angles between the arrows show the degree of correlation between the variables (0° means 100% positive correlation, 90° means no correlation, while 180° means 100% negative correlation). The length of the arrows shows how much weight each variable has on each principal component.

Katz ADL and IQCODE have more points because these variables
lacked several observations that had to be imputed. Because IQCODE was scored by proxies, and Katz ADL was scored by proxies or patients, we believe that the missingness was random in the sense that the score itself was not the reason for the missingness.
As seen, CFS and the other geriatric scores are correlated with each other, while CFS has the longest arrow, and this finding corresponds well with the regression analyses. on an individual basis for patients aged 80 years or older, preferably on the basis of established risk factors for poor outcome.

| D ISCUSS I ON
The term "frailty" has been established in geriatrics decades ago, and is evaluated in two different ways; a phenotypical model 23 and a cumulative deficit model. 24,25 Recently, interest in frailty has increased in intensive care medicine through the simple scoring tool "Clinical frailty scale" (CFS), especially regarding intensive care for elderly patients. One reason for the emergence of the CFS might be due to its visual simplicity. The scoring tool has also correlated well with clinically important outcomes in large studies. 26,27 In our study, we found that the majority of the patients were categorized as frail. Increasing frailty showed to be a strong prognostic factor for 30-day mortality, which again corresponded to the main conclusions from both VIP studies. A Canadian study has also found that frailty correlates with health-related quality of life. 28 Since frailty showed strong correlation with survival in VIP1, other geriatric variables were tested in VIP2. Findings from our study were that cognitive deficit and functional decline, in addition to frailty, were strongly associated with 30-day mortality, but that CFS was a better predictor than the other scores. However, registration of cognitive deficit and functional decline did not provide additional prognostic value when CSF was used. All of these fac-  This study has several strengths. Prospective data have been obtained from both local and university hospitals in Norway, where all parts of the country were represented. Information was also obtained from relatives regarding cognitive function and functional status. There are also some limitations. Firstly, we cannot rule out a certain form of selection bias, as we have no information about those patients who were not admitted to ICUs. The patients' caregivers may think that the patient in any case will not be accepted for intensive care treatment due to a full ICU capacity and, therefore, probably will not be prioritized, a so-called "hidden triage."

P-
Also, no information was obtained about the selection process of which patients were accepted for intensive care by an ICU physician, and which were rejected. Such a process is often referred to as the "ICU triage." Furthermore, we cannot rule out information bias regarding misclassifications of the scores. However, a recent substudy of the VIP2 study showed a high reliability in the scoring of CFS. 35  Fourthly, we used a composite score for comorbidity, which do not identify specific chronic diseases as risk factors.

| CON CLUS ION
In this study of intensive care patients aged 80 years or older, we