Study design, setting, and participants

This study was a prospective, observational follow-up study of a longer-stay (median ICU admission of 7 days) cohort. This randomised controlled trial (RCT) was conducted in a quaternary ICU in Melbourne, Australia, from 2007 to 2010 and detailed elsewhere [17]. RCT participants (n = 150) were screened and invited to participate, and informed consent sought for follow-up. Participants were included in the original RCT if they were > 18 years, ICU length of stay ≥ 5 days, understood English, resided within 50 km from the hospital and the intensive care specialist agreed to their participation. Patients were excluded if they had major disorders affecting the central nervous system or other conditions that would prevent participation in exercise, were approaching imminent death, length of stay > 5 days due to lack of general ward bed availability and unable to perform study physical outcome measures pre-morbidly.

The institutional ethics committee of Austin Health approved the study (H2012/04606), which is reported according to STROBE guidelines [23].

Procedure

From May 2012–December 2013, all patients enrolled in the RCT were screened for survival using hospital and general practitioner records. Patients not confirmed deceased were sent a letter describing the study and inviting participation, with an opt-out clause. Patients were contacted a week later via telephone to seek consent.

Outcome measures were performed at 4 to 5 years following ICU discharge in a standardised hospital environment and questionnaires completed in-person by a single assessor (KH). If participants were not within travelling distance (> 1 h via car journey) of the hospital, questionnaires were completed by phone interview. If participants were within travelling distance (< 1 h via car journey), but could not travel, the outcome assessor attended their home.

Outcome measurement

Statistical analysis

Descriptive data are presented as median [interquartile range (IQR)] and mean [standard deviation (SD)] as appropriate. Imputation of missing data for survivors was not undertaken as there was little missing data between baseline, 1-year, and 4–5-year time points for the outcomes of interest.

Multiple variable logistic regression was conducted as a post hoc analysis to investigate the factors associated with mortality at follow-up. A priori selected baseline variables were compared between survivors and non-survivors, using univariate analyses with p ≤ 0.1 used to determine which variables were entered into the final model. Six independent variables (baseline age, APACHE II scores, acute hospital length of stay, pre-ICU HRQoL (AQoL utility score), Physical Component Summary Score of the SF36v2, ability to perform the Physical Function in Intensive Care Test by day 10 of ICU admission [40]) were included in the final model. The Hosmer-Lemeshow test was used to assess goodness of fit, and pseudo R² statistics were calculated with Nagelkerke R-square. The Wald test was used to assess the significance of the association of the individual variables with mortality. Odds ratios (95% CIs) and sensitivity and specificity of the model are reported. A Kaplan-Meier analysis was conducted to investigate differences in survival between those without comorbidities and those with one or more.

For repeated measures at one with either 4 or 5 years (e.g. 6MWT, TUG), the paired t test was used for normally distributed data and Wilcoxon signed-rank test for non-normally distributed data. One-way repeated measures ANOVA was used to compare HRQoL (AQoL utility score, SF36v2 PCS, MCS and PF) across the three time points. Where appropriate, analyses are reported as mean change in scores with 95% CI and compared with reported minimal clinically important differences as a secondary analysis, where available.

Data analysis was performed using SPSS™ (Mac SPSS™ Statistical Version 20, IBM, New York, NY) and p < 0.05 was taken to indicate statistical significance.

Mortality

In the entire group of 150, 66 (44%) patients were deceased (cause of death listed in Additional file 1: Appendix E2). Date of death was only available for 43 (65%) patients, and mortality was highest (n = 19, 44%) during the first year following ICU discharge. In the multivariable regression, those who were deceased by follow-up were older (p = 0.05), with higher APACHE II scores (p = 0.001) and comorbidities (46% with one or more) compared to survivors (Table 2). Survival rates were significantly improved in the group who had no comorbidities compared to those who had one or more (Fig. 2, log rank p = 0.03).

	B	SE	Wald	df	P	OR	95% CI
Baseline MRC score	− 0.76	0.05	2.16	1	0.14	0.93	0.84 to 1.03
Baseline age	0.05	0.02	4.23	1	0.04*	1.05	1.00 to 1.10
APACHE II	0.17	0.07	6.46	1	0.01*	1.18	1.04 to 1.34
Acute hospital length of stay	0.12	0.01	1.59	1	0.21	1.01	0.99 to 1.03
Baseline AQoL utility score	− 2.16	1.28	2.85	1	0.09	0.12	0.01 to 1.41
Baseline SF36v2 PCS	− 0.04	0.03	1.58	1	0.21	0.96	0.91 to 1.02
Constant	0.41	3.47	0.01	1	0.91	1.51

B beta coefficient, SE standard error, Wald Wald test, df degrees of freedom, OR odds ratio, CI confidence interval MRC score Medical Research Council score, APACHE II Acute Physiological and Chronic Health Evaluation II, AQoL Assessment of Quality of Life, SF36v2 PCS Short Form 36 Health Survey Version 2 Physical Component Score

*Statistically significant p ≤ 0.05

Physical function

At longer-term follow-up, 48 of the 56 survivors (86%) completed the 6MWT with data unable to be collected on 8 survivors due to the travel distance outside defined inclusion criteria. Whilst there was an improvement in the survivors’ 6MWT distance between 1 year (mean 465.8 m, SD 148.9) and 4–5 years (mean 507.5 m, SD 118.2), this difference was not statistically significant p = 0.15 (mean difference = − 24.5 m, CI − 58.3, 9.2). Survivors’ 6MWT distance at 4–5 years was 70% of the predicted distance for Australian age and gender-matched norms [41]. In comparison, survivors’ scores were 89% of predicted North American normative values [42], derived from a sample size more than double that of the Australian reference equation.

More than a third of survivors had an improvement in their walk distance greater than the reported minimal clinically important difference (MCID) of 20 m as reported for ICU survivors [43] and similarly for the previously reported MCID of 30 m for patients with chronic respiratory disease [44, 45]. The frequency distribution of distances for the 6MWT is displayed in Fig. 3.

The survivors had an improvement in their TUG time from a median (IQR) of 7.5 s (6.0–9.0) at 1 year to 6.5 s (6.0–9.0) at 4–5 years. This improvement was statistically significant (p = 0.001) and survivors outperformed normative values for their age range (mean (CI) 8.1 (7.1–9.0) [46] although did not improve by one of the few available reports of MCID for the TUG in acutely hospitalised older medical patients of 9.5 s [47].

Baseline outcome measures for strength included Medical Research Council (MRC) scores and diagnosis of ICU-AW. Survivors’ mean (SD) MRC score was 51/60 (8.0), and 73% did not have ICU-AW as measured during their ICU admission. At longer-term follow-up, grip strength in males was 76% of age-matched normative values [48] at mean (SD) 34 (12.5) kg. Females had a mean (SD) grip strength of 20 (9.9) kg, which was 77% of age-matched normative values [48].

Psychological outcomes

As measured by the CES-D, depressive symptoms were low with a median (IQR) CES-D score of 7.0 (1.0–15.0). Forty (71%) survivors had no depression, 10 (18%) had mild depression and 6 (11%) had major depression. As measured by the HADS, survivors’ symptoms of anxiety and depression were within normal ranges with respective median (IQR) scores of 3.0 (1.0–6.0) and 1.0 (0–4). Forty-five (80%) survivors had no symptoms of anxiety, whilst 11 (20%) had clinically significant symptoms. Forty-eight (86%) survivors reported no symptoms of depression whilst 8 (14%) had clinically significant symptoms. The incidence of PTSD was also sub-clinical with a median (IQR) score of 1.0 (0–11.0) as measured by the IES. Nine (16%) survivors had ‘clinically significant’ symptoms for PTSD.

Health-related quality of life

At longer-term follow-up survivors’ mean (SD), AQoL utility score was 0.74 (0.23), below age-matched normative values of 0.79 (0.19) [49]. There were no differences in AQoL scores over time (p = 0.14, n = 38). Between pre-ICU and 1-year follow-up, the change in the survivors’ mean difference in AQoL scores exceeded the reported MCID of 0.06 [49]. Between 1 year and 4–5 years, there was a smaller clinically insignificant improvement. Between 1 and 4 or 5-year follow-up, 11 (23%) survivors improved more than the AQoL MCID [49], 22 (45%) deteriorated and 16 (32%) did not differ compared to the MCID.

Return to work

Twenty (69%) survivors who had been working prior to ICU (n = 29, 52% of original cohort) had returned to work. Five (17%) survivors had not, reporting poor health as the reason. Twenty-seven (48%) survivors were not working prior to ICU, with 21 (81%) being retired whilst only 4 (15%) survivors were not working due to ill health.

Mortality

The long-term mortality rate of 44% was unexpectedly high compared to previous reports of 19% [6] and 30% [8] in landmark studies at 5-year follow-up. Differences observed in our study may be attributable to increased baseline age and higher APACHE II scores comparative to these previous reports [6, 8]. The Kaplan-Meier curve highlights the contribution of comorbidity to mortality although these analyses were not adjusted for age and APACHE II scores. Overall, the original cohort had a high prevalence of comorbid illness [53]. Almost half of the non-survivors had one or more comorbidities compared to a third of survivors. Pre-existing comorbid illness may be an important consideration for post-ICU trajectories of recovery [53] with worse outcomes attributed to pre-existing illness, particularly for HRQoL [54, 55]. We hypothesise our observed mortality rate could be influenced by local healthcare system factors including physician decision-making regarding ICU admission and rationing and socioeconomic factors. For example, the Australian healthcare model may be relatively well resourced compared to other regions, with a greater ratio of ICU beds to ward beds, a ‘closed’ ICU model and 1:1 nurse to patient ratios [11]. As a result, the threshold for ICU admission may be lower in Australian units than more resource-limited regions.

Physical function

The majority of survivors had minimal decrements in their physical function as evidenced by their 6MWD and TUG values relative to population normative data. Most recovery appears to be gained by 1 year although this was a heterogeneous group, with some survivors still experiencing physical impairment at follow-up. This demonstrates the variability in trajectories of recovery [56] that may influence response to targeted intervention [57, 58] and the importance of stratification according to disability [59]. The greatest deficits were seen in the survivors’ grip strength compared to normative data although this is difficult to interpret as the majority did not have an earlier diagnosis of ICU-AW and grip strength was not measured during the original RCT. Reasons for observed differences in our physical function outcomes could be related to a high proportion of patients admitted for cardiac causes. These patients have a relatively unimpeded recovery following surgery and critical care with resolution of some premorbid comorbidities [10, 60]. The majority of our cohort did not have a diagnosis of ICU-AW during their ICU admission, and this may explain the overall level of high performance in the physical function tests. Comparatively, Herridge and colleagues hypothesised that in their younger cohort with a lower prevalence of comorbid illness, the adverse physical outcomes observed in their cohort likely stemmed from persistent weakness although the incidence of ICU-AW was not specifically reported [6].

HRQoL

The HRQoL findings of our study were comparable to the patterns demonstrated in the study by Cuthbertson and colleagues [8]. In both of these studies, by 5 year follow-up, survivors’ HRQoL scores were comparable if not better than their premorbid scores. However, in the study by Cuthberston and colleagues, the survivors’ HRQoL remained lower than population norms at 5 years whereas in our study, scores were comparable to Australian population norms at this time point which is consistent with the findings of more recent research [9, 61]. The similarity in trends in the SF36 scores over time between these two studies is likely attributable to similarities in demographics (both were conducted in mixed older age cohorts with comparable APACHE II scores) and consistency in the administration of the SF36 to capture premorbid HRQoL.

Psychological outcomes

We provide one of the first reports at 4 to 5 years of directly measured, comprehensive psychological outcome data. Consistent with the other outcomes we have described for this cohort, the incidence and prevalence of psychological morbidity was low. Although not evaluated, these survivors may have possessed higher levels of attributes such as resiliency and self-efficacy as well as access to greater familial and social support. This may have assisted their overall high level of recovery. This concept has been demonstrated in other ICU survivors where resilience has an inverse correlation with neuropsychological impairment and other outcomes such as pain and self-care [62].

Our data are influenced by survivor bias and loss to follow-up inherent in longitudinal studies. The findings suggest most recovery occurs within the first year, and this may be important to consider in the design of future interventional studies. Patients and their families may be at greatest risk of adverse outcomes during this time, and promising interventions such as peer support may assist their recovery transition [63]. A limitation of this study is the lack of follow-up from 1 to 4 and 5 years. This may have assisted in improved follow-up rates through repeated contact with participants although overall attrition was comparable if not better than previous studies. We approached the measurement of psychological outcomes using screening tools rather than diagnostic tools although this was consistent with other reports [21, 22, 29, 33, 64]. Further, the IES was selected as the best measure to screen for PTSD at the time of study design, although more recent reports support the use of the IES-Revised [65] which may limit comparability of our findings.

This study highlights the need for a co-ordinated and collaborative international approach to describe the spectrum of critical care survivorship, particularly as we are starting to see better outcomes reported in other regions such as Europe [9]. In order to improve the outcomes for critical care survivors, consensus is required between clinicians, researchers and policy-makers regarding time points for follow-up and which outcome measures to use. Further, there may be other important factors to evaluate that mediate recovery such resilience [62] and post-traumatic growth and the role of caregivers and their ability to provide support following exposure to critical illness [66, 67]. By establishing large international datasets for a range of patient and family outcomes, we may be able to better understand survivorship from critical illness and develop interventions that will be sensitive to these specific domains.