Purpose

An increasing number of very old (aged 80 or older) patients are being admitted to intensive care units (ICUs) [1, 2]. Uncertainty about outcomes after admission to ICU for these patients has prompted evaluation of their long-term mortality and health-related quality-of-life (HRQOL). Long-term mortality rates range from 55 to 90 % at 3 years [3–5], while in some studies, mortality is similar to old patients who are denied ICU admission [6, 7]. Longitudinal studies yielded mixed results, showing either acceptable [8–18] or poor long-term quality-of-life [19–22]. However, these investigations have been limited by single-center enrolment, small sample sizes, and use of non-validated functional outcome measures. The largest study of long-term HRQOL in old survivors of critical illness enrolled 299 patients [3]. No studies have comprehensively evaluated the determinants of long-term functional recovery after critical illness in these patients.

Other research has documented poor communication and decision-making regarding the use of life-sustaining treatments for these old, critically ill patients [23–25]. A major barrier to optimal decision-making is accurate prognostication in this group. Inadequately informed decision-making can have adverse clinical and psychological consequences for patients and families, and substantial economic consequences for healthcare systems [26, 27]. A clearer understanding of the determinants of long-term outcomes for critically ill patients aged 80 and older may lead to decision-making for these vulnerable individuals that is more consistent with their values.

Our primary objective was to determine the physical recovery of patients at least 80 years of age 1 year after their admission to Canadian ICUs. The secondary objective was to identify factors associated with return to baseline physical function 1 year after ICU admission. A priori, we hypothesized that baseline measures of frailty and physical function, and reason for admission to ICU, would be determinants of 12-month physical recovery, after adjusting for confounding variables, such as illness severity and comorbidities.

Methods

Study protocol

The primary purpose of this study was to develop a longitudinal cohort that we would follow for 12 months involving interviews with patients (if able) and their family members. Recognizing that the recruitment of such patients would reflect a select patient population, we collected ICU and hospital data from chart review on all consecutive patients admitted to the participating ICUs without requiring consent. For this analysis, we define 2 mutually exclusive cohorts: (1) the “longitudinal cohort” of consenting participants and their primary caregivers followed for 12 months, and (2) an unselected “hospital cohort” of consecutive patients not participating in the longitudinal cohort. In participating centers, all patients aged 80 years or older who were admitted to ICUs were eligible for the hospital cohort, but we excluded patients who remained in ICU for less than 24 h. Patients 80 years or older who were admitted to ICU for at least 24 h were eligible for the longitudinal cohort. From the longitudinal cohort, we excluded non-residents of Canada and patients who did not have available family members because we depended on family members for complete data collection. Family members (including partners, significant others, and friends) were eligible if they: (1) spoke English or French; (2) visited the patient at least once during the ICU admission; and (3) were at least 18 years old. For the longitudinal cohort, we obtained consent from the patient’s legal representative or participating family member before enrolment, and subsequently from competent, surviving patients.

Study procedures

Upon enrolment into the longitudinal cohort, trained research personnel administered the comprehensive geriatric assessment (CGA) questionnaire in-person to a family member [28]. For this questionnaire, family members were asked to consider the patient’s condition 2 weeks before hospitalization. The CGA questionnaire enabled calculation of the frailty index, which has been validated in population studies [29, 30]. The frailty index identifies 43 deficits at baseline (Supplementary Appendix 1 eTable 1), each coded as 0 (absent), 1 (present), or 0.5 (where intermediate values were possible). It was calculated as the cumulative proportion of deficits present (minimum score 0; maximum score 1.0), and then graded as mild (>0 and ≤0.2), moderate (>0.2 and <0.4) or severe frailty (≥0.4). Additional variables collected from family members and abstracted from hospital records are shown in Tables 1 and 2 and described in Supplementary Appendix 1.

Table 1 Baseline characteristics of study patients
Full size table
Table 2 Clinical outcomes
Full size table

In the longitudinal cohort, at 3, 6, 9 and 12 months after enrolment, hospital survivors (or family members) were contacted by telephone to record survival status and physical function using the Short Form (SF-36), a 36-item survey of general health status with summary physical component scale and mental component scales [31]. During follow-up, if patients were unable to participate in data collection, family members were asked to complete the SF-36 from the perspective of the patient. Surrogate assessment of the SF-36 has been shown to be reliable, particularly for assessment of physical function [32].

Statistical analysis

Patient characteristics and hospital outcomes were compared between the longitudinal cohort and the hospital cohort. Continuous variables were described by means, standard deviations (SD) and ranges, except for the skewed length of stay variables which were described by medians, quartiles and ranges. Categorical variables were described by counts and percentages. Differences between the hospital and longitudinal cohorts were tested by the Mann–Whitney U test for continuous variables and the chi-square test for categorical variables.

Our primary outcome was physical recovery from critical illness, defined as being alive with SF-36 physical function score of at least 10, and no more than 10 points below baseline. The SF-36 physical functioning domain ranges from 0 (worst score) to 100 (best score) and raw scores are reported for survivors and compared to Canadian age- and sex-matched community controls [33]. A 10-point change is considered small but clinically important; we required patients to have a score greater than 10 to avoid floor effects [34]. We combined survival with the physical function domain score to more comprehensively describe outcomes relevant to this population; for many older persons, quality-of-life and functional capacity are more important than just survival [35]. Patients who died were not considered to have recovered. When estimating physical recovery, we employed fully conditional multiple imputation to reduce potential bias due to loss to follow-up [36]. As described in the Supplementary Appendix 2, the imputation model utilized baseline patient characteristics as well as available 3-, 6- and 9-month SF-36 physical function scores. The estimates of the physical recovery rate and all regression models account for clustering by center by treating center as the primary sampling unit as implemented by the survey procedures of SAS [37].

Unadjusted associations between baseline characteristics and physical recovery at 12 months were estimated by odds ratios (OR) derived from single predictor logistic regression models. We then performed multivariable logistic regression including all baseline characteristics presented in Table 3. Continuous predictors were scaled so that one unit was approximately equal to the interquartile range. This analysis was repeated focusing on 12-month mortality and a separate linear regression model was developed for the 12-month physical recovery of survivors. We evaluated for multicollinearity among the independent variables using variance inflation factors; the largest was 2.6, indicating no such difficulties.

Table 3 Logistic regression model predicting physical recovery 12 months after ICU admission
Full size table

To assess representativeness of the longitudinal cohort, we re-estimated our models of 12-month survival and physical recovery using inverse probability weighting (IPW) so the weighted longitudinal cohort more closely reflected the unselected hospital cohort. Details of the IPW and treatment of missing data (including multiple imputation and sensitivity analyses) are provided in Supplementary Appendix 2.

Analyses were done using SAS Version 9.4 (SAS Inc., Cary, NC., USA).

Role of the funding source

The funding agency had no role in data collection, analysis, or interpretation, writing of the report or decision to submit the paper for publication.

Results

Of 3064 consecutive patients 80 years or older screened for eligibility, 2450 met the inclusion criterion of ICU admission for at least 24 h. From these, 610 patients comprise our longitudinal cohort and an additional 894 patients comprise our hospital chart review cohort (Figure e1a). By the 12-month assessment in the longitudinal cohort, 40 patients had withdrawn consent or refused to respond and 63 were missed or lost to follow-up, leaving 507 patients with known 12-month status. (eFigure 1b). The primary outcome of recovered physical function at 12 months was available in 505 patients (2 patients had missing baseline SF-36 scores). Tables 1 and 2 of Supplementary Appendix 2 compare the baseline characteristics and hospital outcomes between patients with a known versus unknown primary outcome.

Patients in the longitudinal cohort were on average 84 years old (range 80–99), 55 % were men, 62 % had a medical admission diagnosis, 87 % lived at home before hospital admission, 19 % were moderately to severely frail and the average (SD) baseline physical function score was 40 (30) (Table 1). Characteristics of the longitudinal cohort were similar to those of the unselected hospital cohort, but the longitudinal cohort had a higher proportion of patients who had CABG or valve replacement.

Compared to the hospital cohort, the longitudinal cohort had significantly longer ICU (median of 6 vs. 4 days) and hospital (median of 21 vs. 16 days) stays but significantly lower ICU (14 vs. 24 %) and hospital mortality (26 vs. 39 %) rates (Table 2). In patients in the longitudinal cohort, of 526 patients who lived at home before ICU admission, only 179 (34 %) were discharged home and the proportion of patients discharged home decreased with increasing frailty index (Table 2).

12-month follow up data

The 12-month survival rate in the longitudinal cohort sample was 56 %. Using inverse probability weighting (IPW) to reflect the unselected hospital cohort, 12-month survival was estimated to be 50 %.

ICU survivors had significantly lower physical function and physical component SF-36 scores compared to age- and sex-matched community controls at all time points (eFigure 2). By 12 months, an estimated 26 % (95 % CI 21–31 %) of the longitudinal cohort had physically recovered (Fig. 1). Using inverse probability weighting to reflect the unselected hospital cohort, the 12-month physical recovery rate was estimated at 24 (18–30 %). Table 6 of the Supplementary Appendix 2 shows the minimal impact of IPW, multiple imputation and the method for accounting for site clustering on estimates of physical recovery. eFigure 3 shows the physical recovery rates for patients with low baseline PF (<40) and high baseline PF (≥40).

Fig. 1
figure 1

Combined outcomes of mortality and change of physical function from baseline at different time points in the longitudinal cohort. Each section of the bar graph represents the proportion of patients who died or returned to varying levels of baseline function as denoted by the legend (PF physical function)

Full size image

In both the single predictor and multivariable models, younger age, lower APACHE II score, lower Charlson comorbidity index, lower frailty index, and admission diagnosis of CABG/valve replacement were significantly associated with a greater likelihood of physical recovery. Conversely, after adjustment for other covariates, patients admitted for stroke had one-ninth the odds of returning to baseline physical function compared to other patients (Table 3). After adjusting for all other patient characteristics, a higher baseline physical function score was associated with a lower probability of physical recovery.

In the multivariable model, independent associations with survival included lower APACHE II score, admission diagnosis of CABG/valve replacement, lower Charlson comorbidity index, lower frailty index, and family preferences for more than comfort measures (eTable 2). When considering 12-month physical function score as the dependent variable amongst survivors, lower APACHE II scores, admission diagnosis of CABG/valve replacement, higher baseline physical function, lower frailty index, and family preference for more than comfort care were significantly associated with better physical function, whereas admission diagnosis of stroke was associated with significantly worse physical function (eTable 3).

We obtained similar estimates and the same conclusions for all three outcomes using inverse probability weighting so that the longitudinal cohort more closely represented the unselected hospital cohort (see Tables 4, 5, 6 of Supplementary Appendix 2).

Discussion

In this multicenter cohort study, we followed 610 very old patients admitted to 22 Canadian ICUs, and determined their survival and physical recovery 1 year after ICU admission. On average, patients spent just under 1 week in ICU and 3 weeks in hospital. Approximately 20 % died in the ICU, and an addition 12–15 % of patients discharged from ICU died prior to hospital discharge. At 1 year, 50 % had died, and survivors had reduced physical function compared to community controls, and only 26 % of patients were alive and had recovered back to, or near, their pre-hospital level of physical functioning. Only one-third of surviving patients who lived at home before ICU admission returned home; one-quarter of those who were moderately or severely frail at baseline returned home. In multivariable models, primary ICU diagnostic category, baseline physical function, pre-hospital frailty index and APACHE II score were the key predictors of recovery.

Prior studies have addressed long-term outcomes of very old people after critical illness, but most do not report an endpoint that meaningfully combined survival and return of physical function [4, 7, 9, 12, 15, 19]. In an observational study of 2646 patients aged 80 years or more, Boumendil and colleagues reported that 63 % either died or declined (defined as a one-point loss in at least one dimension of the Katz functional status scale) 6 months after enrollment [25]. Given our definition of recovery, this would translate to a recovery rate of approximately 35 % at 6 months.

Our most notable finding is that approximately one-quarter of patients returned to near their baseline physical state 1 year after ICU admission, and the determinants of recovery included variables related to both baseline function and acute illness. Higher baseline physical function was associated with higher 12-month survival, but not after adjusting for other baseline characteristics. Not surprisingly, higher baseline physical function was associated with significantly higher 12-month physical function both before and after covariate adjustment. It is important to note however, baseline physical function was not associated with our composite measure of physical recovery before adjustment and was negatively associated with physical recovery after controlling for the other baseline patient characteristics. This is because physical recovery required a return to within 10 points of baseline PF score which is more difficult to achieve for patients who start with a higher baseline score than a lower baseline score, especially after controlling for other characteristics. The clinical message here is that older patients with higher baseline physical function are more likely to survive, and, on average, if they survive will have a better PF long-term compared to patients with low baseline PF. This is the case even though their chances of returning to the prior level of physical function, after controlling for their other baseline characteristics, are reduced compared to patients who had low baseline PF prior to their critical illness.

Frailty was a more significant independent predictor than age, illness severity, or comorbidity, which are commonly considered important determinants of long-term outcomes (Table 3). Frailty is a multi-dimensional state characterized by loss of physiologic and cognitive reserves which accumulate in older patients and predict adverse events and unfavorable outcomes [29]. Frailty has only recently been recognized as an important determinant of mortality among older critically ill patients [38, 39], and was not included in previous longitudinal studies that examined functional recovery or quality-of-life outcomes of very old people who were admitted to ICU. The frailty index used in this study is based on an accumulation of deficits, reflecting that frailty (a multiply determined greater risk of adverse health outcomes compared with others of the same age [29]) will have different manifestations in different contexts. We measured more than 30 variables from the family and hospital record to determine this index. The frailty index is robust, such that many of the properties of the index are preserved despite differences in how the index is calculated across studies. In fact, within studies, it is possible to randomly select variables that make up an FI and get closely comparable results [40]. This property is a manifestation of the redundancy of biological systems. In short, not every item in a failing system need to be gathered to know that the system is failing [41]. Awareness of baseline frailty can inform discussions about the goals of care and expected outcomes of treatment.

Our longitudinal cohort study found lower short-term and 12-month mortality rates than those reported in other studies of old ICU patients [9, 10, 42, 43]. These findings may relate to enrollment of patients who were admitted to the ICU and who survived for at least 24 h. Therefore, our cohort may have a better prognosis than all old patients admitted to ICU. Furthermore, our longitudinal cohort required the presence of a caregiver, which may have been less likely for patients who were in the ICU very briefly; this hypothesis is supported by the higher mortality and shorter length of stay (ICU and hospital) in the hospital cohort compared to the longitudinal cohort. Also, differences in medical and surgical case-mix may explain differences in long-term outcomes between our study and other similar studies. Thus, compared to studies that enrolled consecutive, very old critically ill people, our study underestimates mortality and overestimates recovery of physical function. Our hospital cohort data gives us a sense of the magnitude of the potential differences in outcome between those enrolled and not enrolled in the longitudinal study. Using inverse probability weighting so that the longitudinal cohort would better reflect the unselected hospital cohort, we obtained 6 % higher 12-month mortality (from 44 to 50 %) and 2 % lower rate of physical recovery compared to the unweighted longitudinal cohort.

There are several strengths of this study. This is the largest longitudinal cohort study of very old patients admitted to ICU that evaluated 12-month outcomes and the determinants of those outcomes. Our study was unique in combining survival and physical recovery as a composite endpoint, recognizing the importance of function in these patients [35], and the utility of predicting future functional recovery for clinical decision-making in the ICU [44]. Our baseline evaluation of patients was comprehensive, and we used validated instruments including a structured assessment of frailty. Our results offer important insights about the role of frailty and other conditions and characteristics that are associated with 12-month outcomes in old critically ill patients. In contrast to previous single-center studies, our multicenter design enhances the generalizability of our findings to those old patients who survive more than 24 h after ICU admission.

Limitations of this study include the fact that a relatively large proportion of patients could not be followed, primarily due to withdrawal of consent and inability to track patients for 1 year after ICU admission. This may have influenced estimates of mortality and recovery, but should not have substantially affected the estimates of association between these outcomes and baseline patient characteristics. Multiple imputation and inverse probability weighting only minimally altered our estimate of 12-month physical recovery, and sensitivity analyses indicate that the estimated rate is fairly robust. We used a proxy family assessment for some baseline variables which is a limitation of most similar studies. This approach has been validated using a comprehensive geriatric assessment as the referent standard [28], which is unavoidable for such studies of critically ill participants, and is consistent with how these data will be used in ‘real-life’ practice. Most of the patients in our sample were Caucasian and are from a single health care system; this may limit the applicability of our results. Finally, we did not record complications, adverse events, or healthcare encounters after critical illness which may have influenced outcomes.

In conclusion, we observed that very old patients who are admitted to the ICU have a low probability of being alive and returning to baseline levels of physical function at 1 year after ICU admission. Clinicians should be aware of the determinants of these outcomes and consideration should be given to recording these measurements for older patients admitted to the ICU. Further research is needed to investigate strategies to systematically include this type of information for real-time decision-making, so that old patients, family members and clinicians become more familiar with the role of baseline frailty and physical function on the trajectory of critical illness.