We performed a prospective cohort study in a medical-surgical ICU in hospital (level II of care complexity) that included all patients admitted from Oct 1, 2016, to May 1, 2019, regardless of age. Only the first ICU admission was considered, and patients whose follow-up was lost due to transfer to another center or without signed informed consent were excluded.

The registry and data collection were approved by the Research Ethics Committee of Francisco de Vitoria University (N°44).

Frailty identified at ICU admission was assessed according to the status before the prompt admission, following the recommendations of the Canadian Health and Aging Clinical Frailty Scale (CFS) 9 points. The CFS was further categorized as nonfrail if the CFS score was ≤3, prefrail if the CFS score was 4, and frail if the CFS score was ≥5.

Before the start of the study, a training period on frailty and the CFS tool was conducted through a training session. The CFS was obtained by reviewing the medical records or interviewing the participant or their proxy decision maker.

The data were collected prospectively on a standardized form and entered into an anonymised electronic database.

The demographic factors, comorbid conditions, origin of ICU admission, admission shift, diagnostic category, severity of illness at ICU admission (Simplified Acute Physiologic Score [SAPS] III) severity of organ dysfunction at ICU admission (Sequential Organ Failure Assessment Score (SOFA)),6 and type of organ failure that developed during ICU admission (at least two points according to SOFA score for each organ) were included; the total number of organ failures, intensity of treatment and categorized ICU complications were also included. The following variables related to short-term prognosis were also collected: length of stay (hospital, ICU, and pre-ICU), number of readmissions to the ICU and outcome at hospital discharge (alive/dead).

The primary outcomes were the prevalence of frailty and prefrailty in the ICU and in-hospital mortality. The secondary outcomes were length of stay (hospital, pre-ICU and ICU) and clinical variables related to the patient's clinical course.

We used frequencies and percentages for categorical variables and medians and interquartile ranges (IQRs) or means and standard deviations for continuous variables.

Three frailty categories were created according to the instructions of the CFS questionnaire: nonfrail patients (scores 1–3), prefrail patients (score 4), and frail patients (scores 5–9). Differences were analyzed using the chi-square test or Fisher's exact test for categorical variables and Student's t-test or its nonparametric alternative, the Mann‒Whitney test, for numerical variables. We constructed a multinomial logistic regression model (with the nonfrail group serving as a reference) using variables with P values less than .05 in the univariate analysis.

In addition, we investigated the factors associated with a higher risk of in-hospital mortality in prefrail and frail patients. First, we performed univariate logistic regressions with all potential predictors of in-hospital mortality. Factors with P values less than .05 were included in a subsequent multivariable model. We used the best subset selection method and the R2 metric to select the model factors. We accounted for homoscedasticity and multicollinearity in the final model with variance inflation factor values >5. In addition, we calculated the optimal cutoff points associated with a higher risk of in-hospital mortality for SAPS 3 patients and for the days before ICU stay using the nearest method; these cutoff points had perfect sensitivity and specificity on the receiver operating characteristic curve. The estimated cutoff point was determined by bootstrapping (100 replications) to calculate confidence intervals.

Finally, we present the differences between the groups of frail patients younger and older than 65 years according to days of hospitalization, in-hospital mortality, need for readmission, presence of delirium and use of supportive therapies. Differences were analyzed using the chi-square test or Fisher's exact test for categorical variables and Student's t-test or its nonparametric alternative, the Mann‒Whitney test, for numerical variables.

Statistical analyses were performed using R 3.6 (R Project for Statistical Computing) and STATA 15.0 (Stata Corporation, College Station, TX, USA). All P values were two-sided; P<.05 was considered to indicate statistical significance.

The total cohort spans an age range from 15 to 93 years, with a median of 66 years (IQR 55–74 years). The CFS score ranged from 1 to 8 points, with a median of 3 (IQR 2–3). Although young frail patients (<65 years) have been reported, frailty is prone to develop with increasing age (Fig. 1). Categorization according to the CFS showed 78% (n=1141) nonfragile, 17% (n=250) prefragile and 4.8% (n=71) fragile patients. Three percent (n=47) of the patients were readmitted, with a median ICU stay of 2 days (IQR 1–4) and a hospital stay of 6 days (IQR 3–12). The mean ICU survival rate was 96.8% (n=1415), and the mean hospital survival rate was 93.8% (n=1371). The remaining data are shown in the bivariate analysis performed after frailty categorization (Tables 1–3).

Figure 1.

Distribution of the three frailty categories according to age.

Notes: CFS: Clinical Frailty Scale; Nonfrail: CFS 1–3; Prefrail: CFS 4; Frail CFS>=5.

(0.11MB).

The results of the baseline characteristics of the bivariate analysis on categorization into nonfrail, prefrail and frail groups showed a tendency to be admitted on a weekend shift as the category increased (20% vs 27% vs 41%; P<.001), indicating a more significant presence of comorbidities and severity on admission (Table 1).

During ICU admission, frail and prefrail patients developed more organ failure, with respiratory failure being the most frequent (31% vs 56% vs 66%; P<.001). A higher therapeutic intensity was administered to higher categories, with an increased incidence of complications. The length of stay was longer, and mortality was greater, with no readmission differences (Table 2).

Variables with statistical significance were included in the multinomial logistic regression analysis (Table 3). Notably there was a high risk of in-hospital mortality in the prefrail group (RR 4.04, 95% CI 2.11–7.74; P<.001), and in the frail group (RR 5.88, 95% CI 2.45–14.10; P<.001). Other findings have included a higher risk of colonization by multidrug-resistant bacteria in patients older than 65 years and using noninvasive mechanical ventilation.

After prefrail and frail patients were grouped into a single group (CFS≥4), a univariate model was performed, followed by a second multivariate analysis of in-hospital mortality. Both models are shown in parallel, and the statistical significance of each variable is shown in Table 4.

The following factors were identified among the deceased patients: higher SAPS 3 score (RR 1.07, 95% CI 1.03–1.12; P=.002), days before ICU stay (RR 1.08, 95% CI 1.01–1.16; P=.029), hematological status (RR 11.46, 95% CI 2.21–59.42; P=.004) and neurological failure (RR 4.14, 95% CI 1.23–13.98; P=.022).

Following these findings, optimal cutoff points for prefrail and frail mortality were identified for the SAPS 3 score (64.5 points; 95% CI=63.6–65.4) and pre-ICU length of stay (4.5 days; 95% CI=1.6–7.4).

The results of the bivariate analysis, which included prefrail and frail patients (CFS≥4) compared against two age categories, nonelderly (<65 years) vs elderly (≥65 years), are summarized in Table 5. The only difference found was the greater severity level at ICU admission in older adults (SAPS3 59.8±13.9 vs 54.9±15.7; P value=.017).

The prevalence of prefrailty and frailty in our cohort was 17% and 5%, respectively. This value varies according to the type of population and screening tools; similar data were observed in the study from the ANZICS,7 with a prevalence of 19%. The mean severity of the cohort at admission (SAPS 3, 48.3±14.5) was comparable to that documented in a European multicenter study.8

This study has limitations and strengths derived from its single-center observational design. On the one hand, this approach allowed us to control and directly supervise the recollection of our independent variable (CFS). On the other hand, the cohort size prevented us from analyzing the frailty spectrum (9 points), limiting the results. However, we consider it a strength to have included a categorization into three groups instead of two and to be able to identify the effect of prefrailty in patients admitted to the ICU without age limits.