Introduction

Epidemiological data and preclinical studies using animal models of lung injury show that advanced age is associated with increased susceptibility to develop the acute respiratory distress syndrome (ARDS) [1,2,3]. Age has also been recognized as one of the major determinants predicting morbidity and mortality in patients with ARDS [4, 5]. Even though elderly patients usually have more comorbidities, this does only partially explain their higher burden of disease [1]. In view of the growing number of elderly patients in intensive care units (ICUs) [6], recognition as well as understanding of the association between advanced age and adverse outcome in ARDS patients could improve prognostication and may even allow for development of age-specific treatment strategies.

Responses to tissue injury in general are modified by the process of aging [7, 8]. In healthy humans, age-related changes in the immune system, including an increased release of pro-inflammatory cytokines and chemokines [7,8,9], coagulation factors [10] and acute phase reactants, dysregulation of the activation and migration of inflammatory cells [11, 12], and endothelial dysfunction [13] have all been associated with functional decline and increased mortality. In addition, there is some evidence that this chronic activation of the “aged immune system” results in an uncontrolled host response to injury, an impairment to mount pathogens, and an inability to resolve tissue damage [8]. Little is known about the impact of this sometimes called “inflamm-aging” or “immunosenesence” on development and progression of ARDS [14]. Preclinical studies have shown a progressive pro-inflammatory status and an altered response to direct and indirect pulmonary insults in elderly animals [3, 15, 16], yet no studies in ARDS patients have been performed to confirm this in the human setting. Those clinical studies that did compare levels of inflammation and coagulation between age groups focused on specific subgroups of critically ill patients and showed conflicting results [17,18,19,20,21,22,23]. In particular, none of these studies investigated age-related differences in the host response as a potential mediator of the outcome. Studying the biological host response using a statistical mediation analysis can provide an etiological understanding of the association between age and outcomes in ARDS patients [24].

This study aimed to characterize and compared the systemic host response in ARDS patients in three age groups and investigated whether the association between age and mortality in ARDS patients is mediated through age-dependent differences in host responses. It was hypothesized that an aggravated host response, i.e., a host response with an excessive increase of one or more systemic biomarkers known to be involved in inflammation, endothelial activation, or coagulation, at least partially explains the increased mortality in elderly with ARDS.

Methods

Study design and setting

This study was performed as a preplanned secondary analysis of the “Molecular Diagnosis and Risk stratification of Sepsis” (MARS) Biobank project, a prospective observational cohort study performed in the mixed ICUs of two university-affiliated hospitals in the Netherlands (ClinicalTrials.gov identifier NCT01905033) [25,26,27,28,29]. The Institutional Review Board of both study centers approved the study protocol (protocol no. 10–056C) with an opt-out informed consent method. Part of the patients’ cohort and results of biomarkers’ measurements have been used in previous reports on the MARS study [25,26,27,28,29,30,31,32,33,34].

Inclusion and exclusion criteria

The parent MARS study included consecutive patients admitted to the participating ICUs if expected to stay in the unit beyond the following calendar day. The current study restricted participation to patients having acute lung injury (ALI) or ARDS, according to the American-European Consensus Conference (AECC) criteria [35] as used at the time the MARS study ran. Trained researchers collected data prospectively and screened all patients daily for the presence of ALI/ARDS. After closing the MARS study, patients were re-classified as having mild, moderate, or severe ARDS, according to the Berlin definition [36], based on the PaO2 to FiO2 ratio at the day of ARDS diagnosis. Patients with a limitation on life-sustaining measures at ICU admission were excluded, patients in whom biomarkers were not determined, as were patients aged under 18 years.

Age groups

The cohort of patients was divided into three groups using cut-off levels based on age-tertiles in the total cohort: young adults (18 to 54 years), middle-aged adults (55 to 67 years), and elderly (> 67 years and older).

Endpoints

The primary endpoint was 90-day mortality, defined as death within 90 days after the onset of ARDS. Secondary outcome measures were 30-day mortality, 1-year mortality, ICU mortality, in-hospital mortality, ventilator-free days and alive at day 28 (VFD), ICU length of stay (ICU-LOS), ICU-free days and alive at day 30 (ICU-FD), in-hospital length of stay (hosp-LOS) and hospital-free days, and alive at day 90 (hosp-FD). The occurrence of death was recorded prospectively if patients died at the ICU. In addition, for all patients, we assessed the vital status in the government registration of persons at 1 year after admission to the ICU. In case a patient was deceased, the specific date of death was recorded. Other patient data was collected prospectively, for details on data collection and definitions, see Additional file 1.

Biomarker measurements

Daily left-over EDTA anti-coagulated plasma was harvested from blood obtained for regular patient care [28, 29]. For the current analysis, the sample nearest to the day of ARDS diagnosis was used. A panel of 20 biomarkers known to be involved in inflammation, endothelial activation, and coagulation pathways (e.g., interleukin (IL)-6, IL-8, IL-10, IL-1β, tumor necrosis factor-alpha (TNF-α), interferon-gamma (INF-γ), intracellular adhesion molecule (ICAM)-1, matrix metalloproteinase (MMP)-8, metallopeptidase inhibitor (TIMP)-1, fractalkine, E-selectin, P-selectin, angiopoietin-1 (ANG1) and ANG2, platelet factor 4 (PF4), protein C, plasminogen activator inhibitor (PAI)-1, antithrombin (AT), d-dimer, and tissue plasminogen activator (tPA) were measured [37,38,39]. For assays, see Additional file 1 and Additional file 1: Table S1.

Statistical analysis

First, clinical characteristics and outcomes were compared for the three age groups. Data are presented as absolute numbers with proportions, medians with interquartile ranges, or means with standard deviations, as appropriate. Differences between the age groups were analyzed using Kruskal-Wallis test for continuous variables and chi-squared for categorical variables. Survival differences between age groups were visualized by Kaplan-Meier plots and tested with a log-rank test.

Next, the association between age and mortality was analyzed by univariate and multiple logistic regression, the confounders included are described later in the “Methods” section. Then, the host response was compared for the levels of the 20 biomarkers between young adults and middle-aged adults, and young adults and elderly using Mann-Whitney U tests.

Finally, a mediation analysis was performed to investigate whether the association between the age groups and 90-day mortality was mediated through age-related differences in systemic biomarker levels (mediation step 1, Fig. 1). The mediation analysis is explained in more detail in Additional file 1. Briefly, first, the association between biomarker levels and 90-day mortality was tested using logistic regression models per each biomarker (mediation step 2). Then, the associations between the age-group and biomarker levels were determined with linear models (mediation step 3). In case of a significant association between age and a biomarker, and between that biomarker and mortality, the average direct effect (ADE) of age on a 90-day mortality and the effect of age on a 90-day mortality mediated by the biomarker were modeled (average causal mediation effect, ACME) (mediation step 4). Mediation was expressed as the ratio ACME to the total effect, so-called proportion of mediation. The 95% confidence intervals of the effects were obtained via bootstrapping. Mediation was only considered relevant if the point estimate of the mediated effect was in the same direction as the total effect because “negative mediation” can per definition not explain the increased mortality [40]. A priori < 5% was considered as a small proportion of mediation, 5–20% as a moderate and > 20% as large proportions of mediation [40, 41]. Young adults were defined as the reference group. Biomarkers were log-transformed to obtain normally distributed variables, and no outlines were removed from the analysis. All associations were determined with both univariate and multiple regression models, and effects were expressed as odds ratios (OR) or β-coefficients with 95% confidence intervals [95% CI], as appropriate. For multiple linear regression models, we reported the adjusted R2. In multiple logistic regression models, the goodness of fit was tested with the omnibus test. A p value < 0.05 was considered lack of fit.

Fig. 1
figure 1

Pathway analysis: stepwise mediation analysis assessing whether the association between age and mortality is mediated by age-dependent differences in biomarker levels. *Adjusted for ethnic background, gender, admission type, readmission, direct hit for ARDS, Charlson Comorbidity Index, APACHE-IV score adjusted for age, immunodeficiency, tidal volume per predicted body weight, positive end-expiratory pressure. C = average direct effect (ADE); A*B = average causal mediation effect (ACME). A’, B’, and C’—adjusted for confounders

Full size image

Because an etiological mediation model was built, it was important to adjust for confounders of the exposure–outcome relation (age and mortality) and the mediator–outcome relation (biomarker levels and mortality) [40]. Potential confounders were selected according to biological plausibility, including ethnic background, gender, admission type, readmission, direct-hit for ARDS, Charlson Comorbidity Index (CCI), Acute Physiology and Chronic Health Evaluation (APACHE)-IV score adjusted for age, immunodeficiency, tidal volume per predicted body weight, and positive end-expiratory pressure. These confounders were included in all models [42]. Collinearity was tested using the variance inflation factor. A value > 2 was considered to be collinear. If so, one of the covariates was restricted from the analyses. The handling of missing data is outlined in Additional file 1: Table S2. Of note, in a large proportion of patients the plasma concentrations of the cytokines TNF-α (64%), IL-1β (39%), and INF-γ (46%) were under the detection limit and were imputed with the LOD value which may introduce over- or underestimation of the estimates (see online Additional file 1: Table S1).

Several sensitivity analyses were performed. To potentially reduce the heterogeneity, the mediation analysis was repeated with a subgroup of patients with a direct hit for ARDS (i.e., pulmonary ARDS). In addition, because it can be argued that CCI and APACHE-VI are mediators instead of confounders, the adjusted mediation analysis was repeated without these variables as confounders. Furthermore, the mediation analysis was repeated with age as a continuous variable—for this analysis also, the goodness of fit was reported. Finally, the biomarker levels from a subset of patients with a sample collected at a later timepoint during ICU admission (4 to 10 days after the onset of ARDS) were explored, by comparing the median biomarker levels between the age-groups using Mann-Whitney U tests. Of note, the median time to sampling from the onset of ARDS did not differ according to age (see Additional file 1: Table S22).

No statistical power analysis was conducted prior to the study. The sample size was based on the available number of patients. All statistical tests were two-tailed and were performed in R statistics using the R-studio interface (www.r–project.org). The mediation analysis was performed using the “mediation package” [43]. In addition, a Benjamini-Hochberg correction for multiple comparisons was performed. A p value < 0.05 was considered as statistically significant.

Results

Of 818 ARDS patients included in MARS, two patients with an age under 18 years, 106 patients with a limitation on life-sustaining measures at ICU admission, and 92 patients without a blood sample were excluded, leaving 618 subjects for the full analysis (Fig. 2). Table 1 shows patient characteristics of 209 young, 213 middle-aged, and 196 elderly ARDS patients. In young patients, the prevalence of comorbidities was lower (see online Additional file 1: Table S3), but the severity of these chronic diseases among these patients was higher as reflected by a similar CCI among the age groups. In addition, young patients more frequently had an immune deficiency (see Additional file 1: Table S3).

Fig. 2
figure 2

Patient flow chart

Full size image
Table 1 Characteristics of the ARDS patients
Full size table

The 90-day mortality rate was 36% (225/618) for the entire cohort, 30% (63/209) in young patients, 37% (78/213) in middle-aged patients, and 43% (84/196) in elderly patients. Middle-aged and elderly patients had 1.5 to 2.1 times higher risk of dying within 90 days after the onset of ARDS compared to young patients, after adjustment for confounders (Table 2, see online Additional file 1: Figure S1). Similar age-related differences existed for 1-year mortality, in-hospital mortality, and hospFD (Table 2, Additional file 1: Figure S1, Table S4, Table S5). There were no significant differences in ICU mortality, 30-day mortality, VFD, ICU-LOS, ICU-FD, and hosp-LOS (see Additional file 1: Table S4, Table S5). Of note, the SOFA scores and the limitation on life-sustaining measures at the time of death of the patients that died in the ICU did not differ according to age (see Additional file 1: Table S6).

Table 2 Association between age and mortality in ARDS patients (mediation step 1)
Full size table

Except for lower plasma levels of E-selectin and IL-10, no statistically significant differences in biomarker levels in plasma were found between young and middle-aged patients (Fig. 3). However, compared to young patients, elderly patients had lower plasma levels of IL-6, IL-8, IL-10, INF-γ, fractalkine, ICAM-1, E-selectin, and higher plasma levels of PF4 and tPA (Fig. 3, see Additional file 1: Figure S2).

Fig. 3
figure 3

Systemic levels of biomarker in ARDS patients stratified by age group. Systemic levels of a interleukin (IL)-6, b IL-8, c IL-10, d Fractalkine, e intracellular adhesion molecule (ICAM)-1, f angiopoetin 2:angiopoetin 1 (ANG2:ANG1), g platelet factor (PF) 4, h tissue plasminogen activator (tPA), i protein C, at onset of ARDS. Young adults, middle-aged adults, and elderly. Box and whisker diagrams depict the median and lower quartile, upper quartile, and their respective 1.5 IQR as whiskers—as specified by Tukey. Group differences between young adults and middle-aged adults, and young adults and elderly were tested by a Mann-Whitney U test. A p value < 0.05 was considered as statistical significant

Full size image

Increased plasma levels of IL-8, IL-10, fractalkine, ANG2:ANG1, tPA, and PAI-1 were independently associated with 90-day mortality (see Additional file 1: Table S7). After adjusting for potential confounders, age was only statistically significant associated with increased levels of tPA and decreased levels of fractalkine and E-selectin (Tables 3, 4, and 5).

Table 3 Association between age and inflammatory marker levels in ARDS patients (mediation step 3)
Full size table
Table 4 Association between age and endothelial activation marker levels in ARDS patients (mediation step 3)
Full size table
Table 5 Association between age and coagulation marker levels in ARDS patients (mediation step 3)
Full size table

Because age was significantly associated with IL-8, IL-10, fractalkine, PF4, PAI-1, and tPA (mediation step 2), and these biomarkers were significantly associated mortality (mediation step 3), either univariate or adjusted, this subset of biomarkers was tested as potential mediators attributing to the association between age and mortality (see Additional file 1: Table S8, Additional file 1: Table S9). Only tPA was found to be a significant mediator, mediating 10 [1 to 28] % (p = 0.018) of the association between age and mortality (Fig. 4). This means that the increased risk of death in the elderly 90 days after the onset of ARDS is partially explained by their higher systemic levels of tPA. After a Benjamini-Hochberg correction for multiple testing, the proportion of mediation by tPA did not remain significant (p = 0.120).

Fig. 4
figure 4

Mediation analysis. a, b Tissue plasminogen activator is a mediator which partially explains the association between age and mortality. *Adjusted for ethnic background, gender, admission type, readmission, direct hit for ARDS, Charlson Comorbidity Index, APACHE-IV score adjusted for age, immunodeficiency, tidal volume per predicted body weight, and positive end-expiratory pressure. ADE, average direct effect; ACME, average causal mediation effect. A p value < 0.05 was considered as statistical significant

Full size image

The sensitivity analysis on patients with pulmonary ARDS and the adjusted analysis exclusion of CCI and APACHE-IV as covariates showed similar results, but no significant mediation in the subgroup of pulmonary ARDS (see Additional file 1: Table S10–S14 and Additional file 1: Table S15–S18). The sensitivity analysis with age as a continuous variable also showed a similar result, with statistically significant partial mediation of 9 [1 to 26] % (p = 0.027) by tPA after adjustment for confounders (see Additional file 1: Table S19–21).

Last, the subgroup analysis of 350 patients (119 young adults, 110 middle-aged adults, and 121 elderly) with a sample at a later timepoint showed that compared to young patients, elderly patients had again lower plasma levels of IL-6, IL-8, IL-10, INF-γ, fractalkine, ICAM-1, and MMP8 (Additional file 1: Figure S3), while plasma levels of PF4 and tPA were not found to be higher in elderly.

Discussion

In this prospective cohort of ARDS patients, a clear association was found between advanced age and higher mortality, which remained after adjustment for potential confounders including comorbidity and severity of illness scores. A mediation analysis showed only moderate (10%) mediation of this association by tPA. In contrast to the tested hypothesis, a comparison of the inflammatory mediators and endothelial activation markers revealed lower instead of higher plasma levels in elderly compared to young patients. Advanced age was only independently associated with increased plasma levels of tPA and decreased plasma levels of fractalkine and E-selectin.

The importance of the association between age and outcome in critically ill patients has been recognized over the past decades [4, 5]. Understanding of the underlying pathophysiological mechanisms explaining this association may provide novel therapeutic targets. To our best knowledge, this is the first study in humans that investigates age-related differences in systemic host response during ARDS as a potential mediator of the increased mortality in elderly by using a mediation analysis.

Although tPA was the only marker showing mediation and the higher levels of tPA in elderly explained just 10% of the association between age and outcome, this finding deserves attention. Increased levels of tPA were strongly associated with mortality and were significantly higher in elderly. Moreover, PAI-1, the endogenous inhibitor of tPA, tended to be lower in elderly, thereby shifting the balance tPA/PAI-1 towards fibrinolysis. Interesting in this context, recent studies have shown that both fibrinolysis shutdown and hyper-fibrinolysis are associated with increased mortality after acute injury [44, 45]. Besides its contribution to fibrinolysis, tPA has been shown to play a role in various other mechanisms involved in the pathophysiology of ARDS, such as the turnover of extracellular matrix components and extravasation of neutrophils [46, 47] [48]. Taking together, tPA could be a potential “modifiable mediator” of interest. However, it must be stressed that we performed multiple mediation models and the partial mediation of tPA did not remain significant after correction for multiple testing. Thus, this effect maybe a false discovery due to a type-1 error. Therefore, it is important that future studies confirm a role of tPA in the association between age and outcome.

The current data showed no evidence of mediation though an enhanced inflammatory mediator response or endothelial activation. Instead, the elderly had lower plasma levels of these biomarkers compared to young patients. Lower levels of systemic inflammatory markers in critically ill elderly have previously been described and could indicate an inadequate host response (so-called immunosenescence) [17]. Nevertheless, so far no association between immunosenescence and an adverse outcome of critically ill elderly has been described [17]. Of note, the mediation analysis showed that the lower biomarker levels found in elderly were not explanatory for the observed increased mortality in this age group. However, as multiple aspects of immunosenescence are known to be mediated via the cellular immune response, which was not included in the current analysis, a role of immunosenescence cannot be ruled out [49]. In contrast to immunosenescence, the lower levels of inflammatory markers in elderly may also reflect differences in the precipitating injury. But, beside transplantation, we found no significant differences in predisposing factors between the age groups. Furthermore, it is important to realize that the host response is a dynamic process. We assessed the host response at the onset of ARDS, i.e., in the early phase of ARDS, while the differences in mortality seem to occur much later (e.g., 90-day mortality and 1-year mortality). Potential age-dependent differences in the host response at a later time point could exist and may also explain outcome differences. Studies have suggested that a prolonged inflammatory response is associated with an adverse outcome [50, 51]. Moreover, there is evidence of age-dependent temporal differences in the host response in critically ill patients (e.g., a hyper-inflammatory response during the later phase of disease and a prolonged inflammation in elderly) [17, 20, 52,53,54,55,56]. The sensitivity analysis of a subset of patients with a sample at a later timepoint did not show any evidence for such prolonged inflammation in elderly. Nevertheless, serial sampling should be included in the design of future studies to assess the presence and implication of time-related differences and their relation to outcome in elderly ARDS patients. In addition, age-related differences in mechanisms involving resolution and repair of damage may also be important [57], especially, as the major differences between younger and elderly patients appear to concern the long-term outcome. Unfortunately, these mechanisms were not studied here due to the restriction in samples available for this analysis.

Despite the unambiguous evidence from various animal models of acute injury, clinical studies—including the current study—failed to detect a robust age-related difference in the systemic host response system that can explain the adverse outcome of elderly with ARDS (see Additional file 1: Table S23). Instead, the data from these clinical studies indicates that elderly ARDS patients die of other reasons than an overwhelming systemic host response. An important contributor to outcome in elderly may be “frailty,” which is characterized by the vulnerability to an acute stressor [58], and its interaction with physiological reserve, i.e., the ability to maintain and restore vital functions [59, 60]. Therefore, the relation between the biomarker levels and outcome may differ according to age. This could have important consequences for the use of biomarkers and biomarker profiles for prognostication, prediction, and selection of patients for clinical trials.

One strength of this study is its prospective design and the use of a broad range of biomarkers characterizing the three main pathways involved in the host response to injury during ARDS. However, study limitations also need to be considered. First, age-groups cut-offs were based on age-tertiles of the total ICU population. These chronological boundaries may not reflect the accurate biological stages of aging [61], although the sensitivity analysis with age as a continuous variable showed similar results. Second, the current study is a secondary analysis of a subset of patients in the parental MARS project, performed on the available dataset. Thus, the absence of statistical significance may be due to a lack of power. In addition, the included ARDS patients formed a heterogeneous patient population. This heterogeneity may have increased the variation in the host response, thereby limiting the ability to detect subtle influences of age-related differences on outcome. However, the subgroup analysis of patients with pulmonary ARDS did not change the results. Moreover, our cohort is relatively large compared to previous studies (see Additional file 1: Table S23), which implies more precision of the estimates. Strikingly, some of the biomarkers showed significant “negative mediation.” This could be a statistical artifact (e.g., unidentified confounding or a result of multiple testing), but a biological cause of this so-called inconsistent mediation (i.e., a suppressor effect) cannot be excluded [62]. Still, because negative mediation can per definition not explain the increased mortality in elderly further exploration of this negative mediation is beyond the scope of our study. Third, despite the strict protocol for timely handling and storage of the daily left-over EDTA, the use of discarded blood by itself could have influenced the biomarker level. It should also be noted, though, that all samples were handled in the same way and irrespective of the age of patients. Fourth, we investigated only the systemic biomarkers. The pulmonary response might be different from the systemic response and more representative for mechanisms contributing to the outcome of ARDS [16]. In addition, it was not possible to determine indices of the activation of the cellular immune response as we had only daily left-over EDTA anti-coagulated plasma samples available for analysis. This is another limitation of the study because there is growing evidence that the cellular immune response plays an important role in the regulation of the inflammatory response and the progression of ARDS [63]. Moreover, instead of looking at single markers, we may have to use a broader approach like genomics and proteomics or combine biomarkers into profiles—by methods like cluster or class analysis—to identify the biological subtypes, so-called “phenotypes” [29, 64]. It has been shown that clinical outcomes and responses to treatment depend on the phenotype of the patient [64]. Whether or not such phenotypes are depending on age or differ between different age groups need further attention.

Finally, the lack of information on the causes and circumstances of death is an important limitation of our study. We assessed the host response at an early stage of disease while death occurs much later in time. Consequentially, patients may have died of other factors than the progression of their initial disease. Previous studies have indicated that age is one of the determinants in end-of-life decisions [65]. Age-related differences in the limitation on life-sustaining measures may have introduced selection bias, which can prevent the detection of a biological relation between the host response and outcome. To minimize this influence of age-related decision-making patients with a limitation on life-sustaining measures at admission were excluded. Furthermore, the available data from patients that died in the ICU showed high respiratory and cardiovascular SOFA scores independent of age, and no age-related differences in the limitation on life-sustaining measures on the day the patients died. Still, selection bias may have influenced the outcome of the study. Last, 92 of the 818 patients with ARDS had to be excluded because there was no blood sample available at the onset of ARDS, this may have introduced selection bias. However, there was no difference in the proportion of missing samples among the age groups.

Conclusion

In this cohort of ARDS patients, the observed association between advanced age and increased mortality was unrelated to alterations in systemic inflammation and endothelial cell activation. Only tPA was found to be a significant partial mediator of the association between advanced age and increased mortality, a mediation effect that disappeared after correction for multiple testing.