As a sub-study of the REALISM project, the REALIST score was developed as a pragmatic and clinically applicable stratification strategy to identify patients with occult immune dysfunction. In our cohort of mixed ICU patients, the REALIST score was able to identify patients at high risk of secondary infections, an association that was independent from major clinical risk factors for infection. As such, this approach demonstrated that genuinely occult immune dysfunction can be identified in ICU patients with tools that are quite potentially available to the frontline critical care physician outside expert research centers [27, 28].
Insights from the REALISM study
The REALISM study outlined how the immune response to injury engages all components of the immune system and does not significantly vary with the type of injury (infectious vs sterile). The initial response is not associated with increased risk of death and secondary infection, illustrating that the initial pro-inflammatory immune response induced by injury should not necessarily be seen as a deleterious factor per se but rather represents an adaptive response to the injury. As induction of the pro-inflammatory effector response is associated with the concomitant development of regulatory mechanisms to protect the host from such overwhelming immune response, this also illustrates the complex interplay between the effector and regulatory mechanisms of the immune system to set up a coordinated immune response to injury. This initial host response likely aims at controlling the aggression and at protecting the host from deleterious off-targets effects of this tremendous immune response.
Thus, after the initial physiologic immune response to injury, it is the persistence (or delayed recovery) of immune alterations that predisposes patients to deleterious infectious events, independently of usual confounding factors. In this subgroup of patients, this persistently dysregulated immune profile cannot be considered as part of the physiologic response to injury but rather as a maladaptive evolution of the immune response.
Therefore, as tempting it might be to try to predict subsequent infection in patients soon after ICU admission, this might not be neither practical nor pertinent. A promising approach to immune monitoring, therefore, seems to be to target the persistence of immune alterations at the end of the first week of ICU stay, identifying patients in which immune homeostasis is pathologically compromised. Knowledge of such occult immune dysfunction is not only interesting, it can also directly influence and hopefully optimize patients’ care, either through enhanced clinical surveillance, accelerated start of antimicrobials in case of infection suspicion, removal of potentially superfluous invasive device, and eventually through immune stimulation strategies [29]. In parallel, a REALIST score of 0 in an otherwise clinically stable patient would provide further reassurance and possibly support deescalating antimicrobial treatment in the right clinical context, for instance. Finally, such immune function scoring could be used to enrich a study population with patients at high risk for secondary infection in the context of an eventual immune stimulation randomized controlled trial.
Our study echoes the work of Conway-Moris et al.in the important INFECT study [14], in which the authors elegantly presented an immune score based on levels of mHLA-DR, Treg lymphocytes (CD4+/CD25++/CD127−) and dysfunctional neutrophils (nCD88) by flow cytometry after strict standardization within different centers. Their score was shown to predict secondary infection in critically ill patients with organ dysfunction (unadjusted HR 4.30 [CI 1.70–10.20] when measured at days 4–6), which interestingly is quite similar to the performance of the REALIST score (unadjusted HR 4.41 [CI 1.63–11.98] at D5–7). Even though Conway-Morris et al. selected different immune parameters than ours and even though patients in the INFECT study had higher illness severity than patients in REALISM, these similar results tend to validate the concept and pertinence of combining parameters to tackle immune monitoring in the ICU across a wide range of patients, even those whom clinical status might seem somewhat reassuring.
Recently, Fang et al. described and validated a similar immune dysfunction score performed at day 1 of ICU admission to help predict mortality in critically ill patients. In that study [30], the combination of monocyte HLA-DR, Il-10 levels, G-CSF levels and ratio of segmented neutrophils to monocyte allowed predicted 28 day mortality with an AUC of 0.789. Interestingly, even though the timepoint and outcome of interest are different than those in the REALIST score, there is significant similarities between chosen immune parameters, further reinforcing the rationale behind immune monitoring in the critically ill.
Choice of immune parameters
Besides mHLA-DR, which is the most studied and validated biomarker in the field with widespread standardization across laboratories [31, 32], Conway-Morris et al. also used the level of Treg (CD4+/CD25++/CD127−) and neutrophils surface expression of CD88. Neutrophil CD88, a receptor for complement anaphylatoxin C5a, has been relatively scarcely described in the critical care context and has only been reported in expert research centers [33,34,35]; it has not been performed in the REALISM study. In parallel, the phenotypical identification of Treg lymphocytes is notoriously problematic and the standardization of their staining by flow cytometry is challenging even with modern techniques in expert centers [36,37,38]. Of note, in the REALISM study, percentage of Treg lymphocytes was not associated with occurrence of secondary infection at D30 through univariate analysis, whether measured at days 1–2, 3–4 or 5–7 (best HR at D3–4: 1.08 (0.84–1.38), p = 0.552).
As immunophenotyping has historically suffered from lack of standardization and reproducibility [39], particular importance must be attributed to these aspects if one hopes for immune monitoring tools to permeate into clinical practice. Thus, for the REALIST score, we purposely and pre-emptively chose immune parameters based on their immediate applicability in clinical practice outside expert centers. As such, technical robustness and reproducibility were major drivers for selecting otherwise relevant immune parameters from the REALISM study.
To complement mHLA-DR, we selected the expression of CD10 and CD16 on neutrophils as a technically simple marker of dysregulated granulopoiesis and inadequate granulocyte maturation. In other words, CD10lowCD16low neutrophils are immature and quite probably the immunophenotypic equivalent of band cells [26, 40, 41], although variability and impreciseness in band cell measurement [42] has precluded such a definite association. Like band cells [43], an increase in CD10lowCD16low neutrophils has been associated with poor outcomes in sepsis patients, namely, occurrence of secondary infection and death [41], and might also directly contribute to impaired T cell function [40]. Our study supports these past findings, as higher proportion of circulating CD10lowCD16low neutrophils at days 5–7 was independently associated with occurrence of secondary infection at day 30.
Lymphopenia has been found to be associated with poor outcome after sepsis and in other clinical illnesses, and therapeutic interventions to increase lymphocytes levels after sepsis have been proposed and are under investigation [44]. Surprisingly, low lymphocytes levels were not associated with secondary infection or mortality at any timepoint in the REALISM cohort, a finding that might be due to lower severity of illness and relatively low event rate.
IL-6 and IL-10 levels were both associated with occurrence of secondary infection in our study, although they brought redundant information in pairwise analysis. As IL-6 is a pro-inflammatory cytokine and IL-10 a globally anti-inflammatory cytokine, it was somewhat expected that elevated IL-10 levels would be associated with immune dysfunction, which was confirmed in this study. Of note, IL-10 levels are strongly correlated with IL-6 levels at all timepoints, as shown in the REALISM study [17], a finding that reflects the intricate and immediate interplay between effector and regulatory limbs of the immune system. As such, IL-6 levels that fail to return towards normal at days 5–7 suggest either immune dysfunction through impaired homeostasis mechanisms and/or uncontrolled inflammatory focus with associated higher disease severity. Of note, the latter hypothesis is supported by the finding that the association between higher IL-6 levels at days 5–7 and secondary infection was not statistically significant after controlling for SOFA score and presence of invasive device.
Strengths and limitations
In our study, we assessed the performance of multiple parameters at multiple timepoints in a relatively large cohort of mixed critically ill patients. We tailored our score to be easily applicable in clinical practice, with a fixed and clear timepoint, reliable and technically robust parameters with a strong track record, and simple computation by bedside clinicians. We also demonstrated strong association with secondary infections even after controlling for SOFA score and presence of invasive mechanical ventilation. We elected to control for both of these variables, even though they are partially redundant (as respiratory support is included in the SOFA score) because of the strong and clinically important association between invasive mechanical ventilation and risk of infection (namely, pneumonia). This thus represents a more stringent statistical correction than seen in other similar studies, further supporting the claim that our score is not a mere marker of disease severity but really an immune monitoring tool that provides genuinely new and previously occult information to the clinician.
Among significant limitations, our study was single-centre and suffers from a relatively low disease severity and low event rate. Although this reduced the strength of association between parameters and outcome, it also suggests that the score is applicable to a wide array of patients with varying disease severity, as secondary infections can occur even in patients with low disease severity with genuinely occult immune dysfunction. Importantly, the REALIST score will have to be further validated in a separate multicentric cohort.