FormalPara Take-home message

In patients admitted to an intensive care unit with severe hypoxaemia, a lower oxygenation target PaO2 of 8 kPa as compared to a PaO2 of 12 kPa did not result in improved survival or health-related quality-of-life one year after randomisation. Survivors in both groups reported substantial impairments in several EQ-5D-5L dimensions, especially in mobility, usual activities, and pain.

Introduction

Acute hypoxaemic respiratory failure is a frequent and potentially life-threatening condition in patients admitted to the intensive care unit (ICU). In this population, the prevalence of acute respiratory distress syndrome (ARDS) increases with the degree of hypoxaemia, and mortality is high reaching rates of more than 50% among the most hypoxaemic patients [1, 2]. Supplemental oxygen is essential in the hypoxaemic patient; however, oxygen therapy may cause supranormal values of partial pressure of arterial oxygen (PaO2) (i.e. hyperoxaemia) [3,4,5], which may be harmful [6,7,8,9,10]. Hence, in the last decade there has been an increased focus on targeted oxygen administration in adult ICU patients, and several randomised clinical trials (RCTs) have been conducted with conflicting results on short-term mortality [11,12,13,14,15,16]. While survival is important, the growing awareness of morbidity in survivors has contributed to an increased attention on the long-term patient centred outcomes. Among these, health-related quality-of-life (HRQoL) is recognised as one of the most important and has been increasingly used in clinical trials within the ICU [13, 17, 18].

In 2021, we reported the results of the Handling Oxygenation Targets in the ICU (HOT-ICU) trial, which evaluated a lower versus a higher oxygenation target in ICU patients with acute hypoxaemic respiratory failure [15]. The trial found no between-group differences in neither the primary outcome being 90-day all-cause mortality nor in the secondary outcomes at 90 days (i.e. number of patients with one or more serious adverse events in the ICU, percentage of days alive without life-support, and percentage of days alive and out of hospital).

Here, we report three of the prespecified 1-year outcomes of the HOT-ICU trial being all-cause mortality and two measures of HRQoL [19, 20]. We a-priori hypothesised that the lower oxygenation target would result in improved survival and HRQoL at 1-year follow-up as compared to the higher oxygenation target.

Methods

Trial design

HOT-ICU was an investigator-initiated, pragmatic, multi-centre, randomised, outcome-assessor blinded, parallel-group trial of a lower versus a higher oxygenation target in adult patients acutely admitted to the ICU with hypoxaemic respiratory failure, defined as use of at least 10 L of oxygen per minute in an open system or a fraction of inspired oxygen (FiO2) of at least 0.50 in a closed system [15]. Patients were randomised 1:1 to a PaO2 target of 8 kPa versus 12 kPa, applied throughout the entire ICU stay, including readmissions, for up to 90 days. The trial protocol, statistical analysis plan, and results are available in the primary publication and elsewhere [15, 19, 20]. This report was prepared in accordance with the Consolidated Standards of Reporting Trials [21] [checklist is presented in the Electronic Supplement Material (ESM) 1]. The trial was approved by the local and national authorities as required (ESM 2).

Trial population and setting

The HOT-ICU trial enrolled 2928 patients between June 2017 and August 2020, in 35 ICUs across 7 countries. The intention-to-treat population, being all randomised patients except those for whom consent was withdrawn or unobtainable, was included in the 1-year assessments. As soon as possible after day 365 from randomisation, survivors were contacted by telephone by blinded and trained trial staff to perform the HRQoL evaluation. Interviewers could make several attempts for up to 30 days following day 365 to establish contact. By agreement with the managing centre, Finnish sites administered HRQoL evaluations through the self-complete paper version.

Outcomes and data source

The prespecified 1-year outcomes were all-cause mortality, EuroQol visual analogue scale score (EQ-VAS), and EuroQol five dimensions five level (EQ-5D-5L). EQ-VAS represented the primary HRQoL outcome.

Vital status at 1-year, including date of death for non-survivors, was assessed from the Danish National Patient Registry [22] and obtained by local investigators for non-Danish patients from patients’ medical records. HRQoL was assessed by the EQ-5D-5L questionnaire including EQ-VAS [23, 24]. If a patient was incapacitated, the next of kin or relevant caregiver was approached to complete HRQoL interview on behalf of the patient; in this case, the proxy version of the questionnaire was used. For the EQ-VAS, participants were asked to self-rate their perceived overall health on a scale from 0 (i.e. ‘the worst health you can imagine’) to 100 (i.e. ‘the best health you can imagine’). For the five dimensions of the EQ-5D-5L (i.e. mobility, self-care, usual activities, pain or discomfort, and anxiety or depression), the patients were asked to give each domain a five-level score (i.e. no problems, slight problems, moderate problems, severe problems, extreme problems) with higher scores indicating worse condition. The EQ-5D-5L index values were reported as a supplemental post-hoc outcome, using the Dutch and British set values for the patients enrolled in these countries, and the Danish set values for all other patients (as no value sets are currently available for Finland, Island, Norway, and Switzerland) [25].

Statistics

All analyses were performed according to the analysis plan using Stata (StataCorp. 2021. Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC).

All-cause 1-year mortality

All-cause 1-year mortality was analysed in the intention-to-treat population, defined as the 2928 patients randomised excluding those for whom consent for the use of data was withdrawn. We compared the 1-year mortality in the two trial groups, using a generalised linear model with a log-link or identity-link and binomial error distribution with adjustment for stratification variables (i.e. trial site of randomisation, known chronic obstructive pulmonary disease (COPD), and active haematological malignancy). Analysis of the 1-year mortality was supplemented with crude Kaplan–Meier plots and hazard ratio from a Cox-proportional-hazards model with adjustment for stratification variables. Post hoc analyses were conducted using a logistic regression model with adjustments for stratification variables only, and a model with stratification variables together with important prognostic baseline factors being age, active metastatic cancer, type of ICU admission (medical, elective surgical or emergency surgical) and the Sequential Organ Failure Assessment score [26]. We also post-hoc evaluated 1-year mortality in subgroups; further details are provided in the ESM 2. Results are presented as absolute risk differences, risk ratios (RR), or odds ratios (OR) as appropriate. Since we expected that the majority of deceased patients would have be dead at 90-day follow-up, we considered the 1-year mortality highly dependent on the primary 90-day mortality outcome, and so, no multiplicity adjustments were performed. Hence, we used a confidence interval (CI) of 95%, and a p value below 5% was considered statistically significant [19].

EQ-VAS and EQ-5D-5L

Patients alive at 1-year follow-up who had filled in the HRQoL questionnaire were included in the primary analysis of EQ-VAS and EQ-5D-5L. The van Elteren test adjusting for trial site only was used in the EQ-VAS analysis since the assumptions of a normal distribution were not met. Adjustment due to multiplicity was performed as previously specified according to the procedure specified by Jakobsen et al., and significance was indicated by a p value below 1.25% [19, 27]. Van Elteren test adjusting for trial site only was also used to compare EQ-5D-5L scores in each dimension, with a p value below 5% considered statistically significant. We also conducted the prespecified analyses in the entire intention-to-treat population. We assumed death as the worst possible health state in terms of self-rated scores, therefore, we assigned to non-survivors the worst possible scores for EQ-VAS (i.e. zero) and EQ-5D-5L dimensions (i.e. five) [19]. EQ-5D-5L index values were reported in the intention-to-treat population as well, assigning the score of zero to non-survivors. Analyses of the outcomes in the intention-to-treat population were performed using the van Elteren test adjusted trial site only, and considering a p value below 5% statistically significant. Since the 90-day mortality in the HOT-ICU trial was twice as high as hypothesised, non-survivors’ scores would dominate the HRQoL estimates, and as the 1-year mortality did not differ between the two trial groups, we post hoc chose to present the results of HRQoL assessments within the population that survived at 1 year as the primary analyses. For the same reasons, and due to the presence of 41 patients with missing data for both 1-year outcomes, which would complicate multiple imputation, we also post hoc decided to present the multiple imputation analysis as a sensitivity analysis in survivors only.

Multiple imputation and sensitivity analysis

Since missing data for the EQ-VAS exceeded 5% of the intention-to-treat population, we performed a multiple imputation analysis within the population of survivors at 1 year, using a general linear model. Additional sensitivity analyses of the EQ-VAS in survivors were also conducted, providing best–worst and worst-best case scenarios to assess the potential impact of any pattern of missing data. Further details about multiple imputation and sensitivity analyses are explained in the ESM 2.

Results

Figure 1 shows the flow of participants from randomisation to 1-year follow-up. Baseline characteristics for all HOT-ICU participants have been presented previously [15]. Table S1 in ESM 2 presents baseline characteristics for survivors, for those lost to HRQoL follow-up, and for those who had died at 1 year.

Fig. 1
figure 1

Patient flow in the HOT-ICU trial. a1 patient in the lower oxygenation group with missing data at 90-day follow-up was included in the -year follow-up. *45/598 (7.5%) were completed by-proxy in the lower oxygenation group and 39/591 (6.6%) in the higher oxygenation group

Full size image

1-year mortality

We obtained 1-year mortality data in 2887/2928 (98.6%) patients. One year after randomisation 707/1442 patients (49%) in the lower oxygenation group and 704/1445 (48.7%) in the higher oxygenation group had died. There was no significant difference between the intervention groups in the primary analysis adjusted for the stratification variables (adjusted RR 1.00; 95% CI 0.93 to 1.08, p = 0.92) (Table 1). The results were in line with the secondary analysis adjusted for stratification variables and important baseline risk factors (adjusted OR 1.02, 95% CI 0.87–1.19; p = 0.79) (Table 1). Figure 2 shows crude Kaplan–Meier plots for the probability of survival between randomisation and 1-year follow-up supplemented with a stratification variable-adjusted hazard ratio. In the subgroup analyses we did not observe heterogeneities in the effects of a lower versus a higher oxygenation target on 1-year mortality (Table S2, ESM 2).

Table 1 1-year all-cause mortality
Full size table
Fig. 2
figure 2

Kaplan–Meier estimates of survival. Shown are the results of Kaplan–Meier analysis of data regarding survival, which was administratively censored at 365 days (adjusted hazard ratio 1.03; 95% confidence interval 0.92–1.14). The Cox proportional-hazards model was adjusted for the trial site, and for the presence or absence of chronic obstructive pulmonary disease and of active haematological malignancy

Full size image

Health-related quality of life

A total of 2600/2928 patients (88.8%) were included in the HRQoL analysis of the intention-to-treat population. Among patients alive at 1-year follow-up, a total of 1189/1476 (80.4%) participated in HRQoL interviews. In survivors at 1 year after randomisation, the median EQ-VAS [interquartile range (IQR)] was 65 (50–80) in the lower oxygenation group vs. 67 (50–80) in the higher oxygenation group (p = 0.98) [Tables 2 and S3 (ESM2)]. The multiple imputation analysis showed similar results (Table S4, ESM 2). In the best–worst and worst-best case analyses, statistically significant differences between the two intervention groups were found in both scenarios, albeit in opposite directions (Table S5, ESM 2). The 5 dimensions of the EQ-5D-5L in survivors are presented in Fig. 3 and in Table 2. No between-group differences in any of HRQoL dimensions were found. The analyses of EQ-VAS and EQ-5D-5L in the intention-to-treat population showed similar results as the primary analyses (Table S6, ESM 2). In the intention-to-treat population the median EQ-5D-5L index value (IQR) was 0 (0–0.7) in both trial groups (p = 0.73) (Tables S6 and S7, ESM 2).

Table 2 EQ-VAS and EQ-5D-5L in survivors 1 year after randomisation
Full size table
Fig. 3
figure 3

Distribution of EQ-5D-5L among survivors at 1 year from randomisation. EQ-5D-5L denotes EuroQol five dimensions five-level questionnaire [23, 24]. Values are from the responding survivors (n = 598 in the lower oxygenation group; n = 591 in the higher oxygenation group). The corresponding numeric data are presented in Table 2

Full size image

Discussion

In this long-term follow-up of the HOT-ICU trial, investigating oxygenation targets in acutely admitted ICU patients with severe hypoxaemia, we found that targeting a PaO2 of 8 kPa, as compared to a PaO2 of 12 kPa, did not result in improved survival or HRQoL at 1 year after randomisation.

These results are consistent with the primary report, showing no differences in 90-day all-cause mortality nor in the secondary outcomes at 90 days between the intervention groups [15]. Thus, targeting a PaO2 of 8 kPa during the ICU stay did not improve either short- or long-term outcomes as compared to a PaO2 of 12 kPa. The wide confidence intervals around the 1-year mortality point estimates did not preclude potentially important clinical benefit or harm of the lower oxygenation strategy, emphasising the need of even larger trials to inform clinical recommendations and guidelines. In the secondary Bayesian analysis of 90-day mortality in the HOT-ICU trial, harm of a lower oxygenation strategy with higher degrees of shock (measured as higher administered doses of continuously infused norepinephrine at baseline) was suggested [28]. Also, in the subgroup analysis of patients with sepsis included in the ICU-ROX trial, point estimates for the treatment effects indicated possible harm of the lower oxygenation strategy, although this was not statistically significant [29]. Importantly, neither of our subgroup analyses on 1-year mortality showed any heterogeneities in the effects of a lower compared to a higher oxygenation target, including the analysis subgrouping patients according to norepinephrine dose at baseline. Our results lend weight to the utility of a lower oxygenation target in adult ICU patients admitted with acute hypoxaemic respiratory failure, which makes our findings more remarkable in times of a pandemic. Several health care systems have been challenged by an increase in oxygen demand due to the outbreak of the severe acute respiratory syndrome coronavirus 2 [30]. Hence, using a lower oxygenation target in the hypoxaemic patient might help in sparing the available oxygen stockages, and additional interventions such as prone positioning [31]. The high short-term as well long-term all-cause mortality of our population, which consisted of non-selected hypoxaemic ICU patients, is comparable to that observed in a cohort of mechanically ventilated patients with ARDS [2], highlighting the disease severity of the trial population. This is further confirmed by the fact that HOT-ICU survivors had very low EQ-VAS compared to the Danish population norms (mean score 82.4; 95% CI 81.5–83.4) [32], and that an important proportion of survivors—between 17 and 44%—reported moderate to extremely severe problems in several dimensions of the EQ-5D-5L. This poor self-reported HRQoL is similar to those presented in another ICU population of septic shock survivors at 6 months [18]. Among RCTs investigating lower versus higher oxygenation strategies in the ICU, only the ICU-ROX trial has reported HRQoL in survivors at 180-day follow-up [13]. Survivors in this trial had low scores of EQ-VAS without between-group differences, and a consistent decrement in many dimensions of the EQ-5D-5L was found, particularly in respect of mobility, usual activities, and pain. The 1-year survivors in our trial showed even lower scores in the same EQ-5D-5L dimensions, despite a longer follow-up. This may be explained by differences in the trial populations. The ICU-ROX trial had a higher percentage of patients included after surgery and patients with acute brain disease at baseline, but only 64.6% of the intention-to-treat population had acute hypoxaemic respiratory failure at randomisation, as confirmed by a PaO2:FiO2 ratio twice as high as that in the HOT-ICU trial, which in turn was equivalent to moderate to severe ARDS [13, 15]. Consequently, the severity of hypoxaemic respiratory failure in our population may have significantly contributed to the higher observed mortality, and potentially to the more severe long-term HRQoL impairment in survivors as compared to the ICU-ROX trial. Our findings of low self-reported HRQoL align with prior studies conducted in ARDS survivors [33,34,35]. Remarkably, an ARDS cohort study showed both low HRQoL scores and cognitive impairment at 1-year follow-up, finding an association between a poorer cognitive performance and a lower PaO2 [35]. Finally, in an observational study of ARDS survivors a decrement in the physical scale of HRQoL was found to persist after 5 year remaining approximately one SD below the mean score for an age- and a sex-matched control population [34]. This could imply that 1 year may be an appropriate time-point to investigate the long-term outcomes in survivors after hypoxaemic respiratory failure.

Our study has several strengths. It represents the largest long-term mortality and HRQoL assessment in a RCT of lower versus higher oxygenation strategies in critically ill adults acutely admitted to the ICU. Both endpoints of this study were prespecified secondary outcomes of the HOT-ICU trial, and data were collected in the context of a large pragmatic RCT [15, 19, 20]. The setup of the HOT-ICU trial, with inclusion of patients at 35 ICUs across 7 countries, increases the generalisability of our findings. The follow-up was conducted by trained and blinded research personnel, and we had a high follow-up rate for 1-year mortality (98.6%). The HRQoL follow-up-rate was 80.4%, which is similar to other trials [13, 18]. To take the missing data into account, we conducted a multiple imputation analysis confirming the results of the primary analysis, and also performed best–worst and worst-best sensitivity analyses of the EQ-VAS in the cohort of survivors. The latter two scenarios detected statistically significant differences between the trial groups in opposite directions. This is not surprising, although it emphasises that missingness may have negatively affected the trial´s power to draw definitive conclusions. Moreover, secondary analyses of HRQoL, accounting for patients who were dead at 1-year follow-up by assigning them the worst scores, confirmed the primary results, thus increasing the validity of our findings. However, due to the higher than expected mortality, which was equally distributed between the intervention groups, we primarily focused on presenting the results of survivors. Furthermore, the use of EQ-5D-5L questionnaire with EQ-VAS also represents a strength, since the tool is well validated, available in more than 130 languages [24], and it is recommended for HRQoL assessment in critical care trials [36]. Some limitations must also be considered. We did not collect data on concurrent illnesses, readmissions to the hospital, or aftercare needs following the first 90 days from randomisation. Moreover, HRQoL was not assessed at baseline preventing comparison of the long-term outcome with pre-randomisation scores. However, obtaining baseline HRQoL scores in the context of a RCT including acutely ill patients would only have been possible through a retrospective assessment by proxy or by survivors, which would likely be biased by the severity of the patient´s acute condition or by outcome. Therefore, collection of HRQoL at baseline was not deemed meaningful.

Conclusion

In the multi-centre, randomised HOT-ICU trial, no long-term survival benefit and no HRQoL benefit of a lower oxygenation target of 8 kPa as compared to a higher oxygenation target of 12 kPa was found. Also, a lower oxygenation target did not result in improved quality of live. Survivors at 1 year had low HRQoL with substantial impairments in several EQ-5D-5L dimensions, particularly mobility, usual activities, and pain, underlining the disease severity of the HOT-ICU trial cohort.