One-year patient outcomes based on lung morphology in acute respiratory distress syndrome: secondary analysis of LIVE trial

We conducted a secondary preplanned analysis on the long-term outcomes of patients enrolled in the LIVE study, as described in the original protocol [13]. The LIVE study was a prospective, multicentre, stratified, parallel-group, single-blind randomised controlled trial in 20 university and non-university French ICUs [10]. Patients with moderate-to-severe ARDS, according to the Berlin definition, were characterised by local site investigators as having focal ARDS (presence of consolidations localised only in the lower and back parts of the lungs) or non-focal ARDS [14]. Lung morphology was assessed before randomisation using a CT scan of the whole lung or chest X-ray when the severity of the patient was not compatible with transport. Patients were randomly assigned to a ventilation protocol adjusted on the basis of lung morphology (personalised group) or to a standard strategy in line with traditional care (control group). The detailed protocol is available in the online supplement.

In the LIVE trial, 24% of patients were misclassified as focal or non-focal ARDS before randomisation (see LIVE trial for more explanation [10]). We performed an intention-to-treat analysis that included all participants who were randomly assigned to treatment, except those who withdrew consent and those who were found to be ineligible because they met the exclusion criteria. Next, we performed a per-protocol analysis in which misclassified patients in the personalised group were excluded due to a breakdown in the ventilation protocol. Misclassified patients in the control group were not excluded because they were not misaligned with ventilator strategy, which, by definition, is not related to lung morphology.

The study took place in two steps. First, long-term outcomes were assessed based on lung morphology phenotypes: focal versus non-focal ARDS. Then, the impact of a personalised ventilation protocol (i.e. personalised vs. control group) was assessed on the same outcomes. Only the per-protocol analysis was used to compare focal and non-focal ARDS to avoid any bias induced by misclassified patients. However, both the intention-to-treat and the per-protocol analyses were used to compare the personalised and control groups. The results of the personalised and control groups are reported in the Supplemental Data available online.

According to current French law, the original trial was approved by the “Institutional Review Board of Clermont-Ferrand, France” (ID RCB 2013-A01756-39) and registered on ClinicalTrials.gov (NCT 02149589) [15]. Before any inclusion, written informed consent was obtained from patients or their relatives. They could refuse to participate at any time, and their decisions were recorded in patient files.

Patient’s survival was assessed, after a 1-year follow-up, between focal and non-focal ARDS and between personalised and control groups. Early mortality was defined as mortality in the first 90 days. Late mortality was defined as mortality between day 90 and 1 year. Further, at the 1-year follow-up, survivors were evaluated by trained research staff who were blinded to treatment allocation. Patient-reported outcome measurements included age- and sex-adjusted physical function and mental health domain scores of the Medical Outcomes Study Short Form 36 (SF-36) instrument [16]; anxiety and depression symptoms from the Hospital Anxiety and Depression Scale (HAD) subscale scores [17]; and effects of fatigue on quality of life from the Modified Fatigue Impact Scale (MFIS) [18]. SF-36 ranges from 0 to 100 with 8 categories and sums up into the physical component summary and the mental component summary. HAD score is divided into anxiety and depression subscales, both ranging from 0 to 21. The presence of anxiety or depression was defined as a subscale over 7 [19]. MFIS total score ranges from 0 to 84 with three subscales: physical (range 0–36), cognitive (range 0–40), and psychosocial (range 0–8). Based on previous studies, patients with an MFIS of over 38 were described as having fatigue [20].

Given that this was a secondary analysis of the LIVE study, no sample size calculation was required a priori. A descriptive analysis was performed on all patients and survivors. Quantitative variables are expressed as mean (standard derivation [SD]) or median (interquartile range [IQR], 25–75%) and compared using Student’s t test or the Mann–Whitney U test, as appropriate. Categorical variables are expressed as numbers (%) and compared using the chi-square test or the Fisher’s exact test as appropriate.

Survival curves were plotted using the Kaplan–Meyer method, and comparisons were made using the log-rank test and a Cox model. To assess the impact of lung morphology phenotypes (focal vs. non-focal ARDS) on mortality, significant variables during the univariable analysis (threshold of p < 0.10) or any variable known to be associated with mortality during ARDS were included in a Cox model. Multivariable analysis was performed on complete cases. A sensitivity analysis with multiple imputations was conducted to deal with missing data. Additional information about the process use for multiple imputations is reported in Additional file 1.

Statistical analyses used to assess the impact of a personalised ventilation protocol adjusted on the basis of lung morphology (personalised vs. control group) are reported in Additional file 1.

To compare the functional outcomes of the SF-36 instrument with previous reports [3, 21, 22], z-scores were created by standardising the values (mean = 50, standard deviation = 10; range 0–100, with a higher score indicating better function) using previously described methods [23] and a French cohort of healthy patients [24]. A Mann–Whitney U test was used to assess the difference between the z-scores and 50. Because 35% of patients missed the functional outcomes assessment, a sensitivity analysis with imputation was conducted to deal with missing data. Additional information about the process use for imputation is reported in Additional file 1. A P value of ≤ 0.05 was considered to be statistically significant. Statistical analyses were carried out using R version 4.1.2 for macOS® (https://​www.​r-project.​org, accessed November 2021).

From June 2014 to February 2017, 420 patients were included in the LIVE study. Flow charts of the study are reported in Fig. 1 and Additional file 1: Fig. S1. Sixty patients were excluded from the per-protocol analysis, as described in Additional file 1: Fig. S1. Of the remaining 360 patients, 124 (34%) and 236 (66%) had focal and non-focal ARDS, respectively (Fig. 1). Information for survival at 1 year was available for 334 patients (93%) (114 and 220 patients with focal and non-focal ARDS, respectively). Data on the population included to compare the personalised and control groups are reported in the online supplement.

Patient baseline data for the per-protocol cohort and for the focal and non-focal ADRS groups are summarised in Table 1. There were more males, and BMI was higher in the focal ARDS group. Plateau pressure, PEEP, and driving pressure were higher in non-focal ARDS. Ventilator-free days to day 28 and ICU length of stay were not different between the groups.

The 1-year mortality was higher in non-focal ARDS patients compared with focal ARDS patients (81 [37%] vs. 27 [24%], respectively; log-rank test: p = 0.012) (Fig. 2A). In the multivariate analysis, non-focal ARDS (hazard ratio (HR), 3.44; 95% confidence interval (95%CI), 1.80–6.59; p < 0.001), age (HR, 1.04; 95%CI, 1.02–1.06; p < 0.001), McCabe score (HR, 1.51; 95%CI, [1.04–2.19]; p = 0.029), haematological cancers (HR, 2.24; 95%CI, 1.02–4.97; p = 0.045), SAPS II (HR, 1.02; 95%CI, 1.00–1.04; p = 0.047), and renal replacement therapy (HR, 1.71; 95%CI, 1.04–2.79; p = 0.033) were independently associated with 1-year mortality (Additional file 1: Table S1).

The difference in 1-year mortality was driven by early mortality (65 non-focal ARDS patients [28%] vs. 19 focal ARDS patients [16%]; log-rank test: p = 0.010) but not by late mortality (16 non-focal ARDS patients [7%] vs. 8 focal ARDS patients [6%]; log-rank test: p = 0.591) (Fig. 2B). Non-focal ARDS, age, SAPS II, and renal replacement therapy were independent predictors of early mortality (Table 2). Only the McCabe score was associated with late mortality (Table 2). During the first 90 days, death was directly related to ARDS in 35% and 37% of patients with focal and non-focal ARDS, respectively (p = 0.89). Underlying disease was accountable for 65% and 63% of deaths in focal and non-focal ARDS, respectively.

There were no missing data for the included variables, except for plateau pressure, PEEP, and driving pressure, which were unavailable for 83 (23%), 14 (4%), and 88 (24%) patients, respectively. A sensibility analysis with multiple imputations for missing data confirmed the previous multivariate analysis. (Additional file 1: Table S2).

A total of 226 patients survived after a 1-year follow-up and were eligible for functional outcomes assessment. We noted that 79 patients (35%) missed the functional outcomes assessment (27 [31%] and 52 [37%] patients with focal and non-focal ARDS, respectively) (Fig. 1). There was no difference between patients who did and did not complete the functional outcomes assessment at the 1-year follow-up except for age and McCabe score (Additional file 1: Table S3). Focal ARDS patients who completed the functional outcomes assessment were older (62 [14] vs. 54 [16] years; p = 0.005) and had a lower McCabe score (p = 0.027) than those who did not. However, there was no difference between focal and non-focal ARDS among patients who completed the functional outcomes assessment.

After a 1-year follow-up, the per-protocol cohort had significant impairment in patient-reported outcomes when compared against an age- and sex-matched population (Table 3) [24]. The median values for the standardised SF-36 physical and mental component summary were 37 (p < 0.001) and 41 (p < 0.001), respectively. Compared against an age- and sex-matched population [24], both focal (p < 0.001) and non-focal ARDS (p < 0.001) had impairment in the SF-36 physical component summary, with no differences between both groups (p = 0.201).

Regarding the HAD scale, 41% and 35% of the per-protocol cohort had anxiety and depression, respectively. The median MFIS component summary of the per-protocol cohort was 43 (Table 3), and 91 (62%) patients experienced fatigue. There was no difference in anxiety and depression symptoms and rate of fatigue between the focal and non-focal ARDS groups (Table 3).

A sensitivity analysis dealing with patients that missed the 1-year functional assessment confirmed the previous analysis (Additional file 1: Table S4).

At the 1-year follow-up, there was no difference in survival and quality of life between the personalised and the control groups regarding both the intention to treat and per-protocol analysis (Additional file 1: Fig. S2 and Tables S5 to S8).

A prior study demonstrated a significant difference in early mortality between focal and non-focal ARDS without assessing long-term outcomes [8]. Beyond confirming these results, our study found that the difference in survival after a 1-year follow-up is only explained by short-term mortality. The impact of morphological phenotypes on mortality may be explained by differences in respiratory mechanics, with a higher driving pressure and elastance in non-focal ARDS [25]. Indeed, in our study, non-focal ARDS had higher plateau and driving pressures at baseline.

As with other intensive care variables (i.e. SAPS II, renal replacement therapy), which are independently associated with short-term mortality (90-day mortality), morphological phenotypes had no impact on mortality thereafter. Only comorbidities and previous health status seemed to impact long-term survival. These results are consistent with previous studies reporting that the severity of illness is a strong predictor of hospital and short-term mortality, whereas age, serious comorbidities, and previous health status are associated with long-term mortality in ARDS [11, 26]. Nevertheless, morphological phenotypes may be considered one of baseline intensive care variables, along with others such as respiratory mechanics and renal replacement therapy, which may only impact short-term mortality [27, 28]. These findings are consistent with a recent study that focused on the impact of the inflammatory phenotype on long-term survival in ARDS patients [12]. In that study, the inflammatory phenotype had an impact on short-term survival, with little implication after 90 days.

In our trial, beyond survival, quality of life was altered after a 1-year follow-up. Physical and mental health were impaired compared with a healthy population [24] which is consistent with previous studies [3, 21, 22]. Age and previous comorbidities have been reported to be associated with impairment of the SF-36 instrument [3]. Even though the specific contribution of ARDS is contested [29], the severity of the initial ARDS, the rapidity of its resolution, and the pulmonary dysfunction thereafter were all correlated with long-term outcomes [21, 30]. Anxiety and depression affected one-third of the ARDS survivors in our study. These results were similar to those of previous trials that reported between 11 and 42% of anxiety and 9 to 39% of depression using the HAD scale [22, 31]. Female sex, unemployment, alcohol misuse, and greater opioid use in the ICU were significantly associated with psychiatric symptoms [32]. Baseline intensive care variables, severity of illness, and mechanical ventilation duration were not associated with anxiety or depression. Eventually, fatigue was reported in 62% of ARDS survivors using the MFIS score in our study. Using another fatigue scale, a recent study reported 70% and 66% of ARDS survivors with significant symptoms of fatigue at 6-month and 1-year follow-ups, respectively [31]. Worse physical, cognitive, and mental health but not baseline critical care variables were associated with greater fatigue. Thus, one could wonder about the impact of morphological phenotypes on survivors’ later quality of life. However, in our study, there was no influence of morphological phenotypes on any functional score.

In the age of personalised medicine, a personalised ventilation protocol adjusted on the basis of morphological phenotypes was of concern. The LIVE trial failed to show any effect on 90-day mortality [10]. Similarly, we could not find any improvement in mortality or functional outcomes even with the exclusion of patients with a ventilation protocol break. The LIVE trial was unfortunately biased by a high number of misclassified patients who had worse outcomes than the control group. Morphological phenotyping of ARDS patients is a major problem because the misclassification of some patients makes it impossible to determine the value of phenotyping over the standard of care. Recently, some machine learning algorithms have been inconsistent in their ability to identify clusters of ARDS patients involved in significant heterogeneity of treatment effects [33]. Future studies may be needed to provide physicians with effective solutions to avoid misclassification, with or without the help of new technologies.

The prospective and multicentre aspects of this study are some of its strengths. Compared with most studies, phenotyping was performed before randomisation, and the large cohort allowed the use of multivariable analysis in the statistical protocol, providing a more detailed interpretation of the data. However, this study has some limitations. First, it is a secondary analysis of a previous trial with an a priori sample size calculation based on the primary outcome and with numerous misclassified patients. This might have induced an underpower analysis, which can explain the absence of differences induced by the personalised ventilation protocol. Second, 24% of patients were misclassified into the focal and non-focal ARDS groups in the LIVE study, which may influence the different analyses and their interpretations. Indeed, misclassified patients have worse outcomes than correctly classified patients in both the LIVE trial and study [10]. Third, there were about 20% of missing data concerning plateau pressure, PEEP, and driving pressure. Given that we used a complete case analysis in our multivariable models, this may have induced a selection bias. However, our results were confirmed by a sensitive analysis with multiple imputations for missing data. Eventually, about 30% of survivors missed the functional outcome assessment after a 1-year follow-up. This may have induced a bias, especially as the patients were not similar in terms of age and McCabe score. However, our results are similar to those of previous studies, and there was no difference between the two phenotypes.