Circulating bioactive adrenomedullin as a marker of sepsis, septic shock and critical illness

Sepsis is a condition with high mortality and suffering, affecting millions of people yearly across all ages and backgrounds [1].

Since sepsis is a syndrome encompassing a variety of illnesses with multiple pathophysiologies, there is no broadly applicable single efficient treatment pathway.

New methods for stratification and classification of sepsis are warranted in order to better tailor the care of septic patients. The use of biomarkers can potentially help us understand and categorise sepsis into phenotypes [2] and thereby add value to existing risk and severity scoring systems as well as guiding treatment. Further, a better understanding of hormonal systems, which some biomarkers are derived from, can open up for new therapeutical pathways.

Adrenomedullin (ADM) is a 52-amino acid peptide hormone first discovered in human pheochromocytoma cells [3], but is produced by many different cell types [4]. ADM plays a part in the homeostasis of cardiovascular, endocrine, renal and immunological systems and has a role in the electrolyte balance [3‐6]. More specifically, ADM has vasodilatory properties [7, 8] by binding to receptors on both endothelial and smooth muscle cells [9]. Further, ADM is capable of modulating the endothelial barrier, where it has a stabilising effect [9].

Over the last fifteen years, the role of ADM in sepsis has been investigated. Several studies have reported an association of increased levels of ADM and poor outcomes among patients with sepsis and septic shock [10‐16]. The role of ADM in patients with a cardiopulmonary disease has also drawn attention [17‐24]. These studies have used two assays measuring different fragments from the ADM precursor, mid regional pro adrenomedullin (MR-proADM) [25] and circulating bioactive adrenomedullin (bio-ADM) [12], making results difficult to compare. A cut-off value of 70 pg/mL bio-ADM has been used, which originates from Marino and colleagues [12]. It is not clear how this threshold was chosen, but the authors reported a 100% 28-day survival rate in a minimal subgroup (n = 12) where a reduction of bio-ADM levels to below 70 pg/mL was observed.

In animal models of sepsis, however, exogenous ADM has led to improved outcomes[26‐28], why ADM has been referred to as a double-edged sword [29]. Further, modulation of the ADM hormonal system using antibodies against a non-ligand binding site of ADM has been suggested a potential therapy in sepsis [30]. This is currently investigated in a phase II clinical trial [31], where septic patients with initial levels of bio-ADM > 70 pg/mL are randomised to receive either the human ADM antibody adrecizumab or placebo [31]. Since ADM levels in non-septic and non-cardiopulmonary critical care patients are poorly investigated, we decided to perform this exploratory study.

The primary aim of this study was to investigate the association of admission bio-ADM with mortality in patients fulfilling the sepsis criteria and in a large mixed general ICU population. Secondary aims were to investigate the association of bio-ADM with organ failure in the ICU, measured as need of circulatory and renal support, and the ability to identify sepsis. Further, we aimed to perform a validation of the proposed cut-off value of 70 pg/mL.

The present study was a retrospective multicentre observational study of patients consecutively admitted to one of four general (mixed surgical and medical) ICUs in the Skåne Region (Scania county), Sweden, in 2016. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed [32].

All adult ICU admissions with valid admission blood samples were included. When direct transfers occurred between the participating ICUs, follow-up data were merged to form cohesive ICU admissions. Transfers from other ICUs were excluded since our aim was to limit our study to primary admissions to intensive care. Information was given to the patient or next of kin, and information letters were sent home to surviving patients 2–6 months after hospital discharge. Patient consent was on an opt-out basis. For deceased patients, consent was presumed.

The primary outcome was 30-day mortality in sepsis patients and the general ICU population. Secondary outcomes were: (1) need of cardiovascular support, defined as cardiovascular sequential organ failure assessment (SOFA) score \(\ge\) 3, at ICU admission, (2) need for continuous renal replacement therapy (CRRT) during ICU-stay and (3) identification of sepsis at ICU admission.

The process of identifying the sepsis population, and collection of background data for this cohort, has previously been described in detail [33].

In brief, the sepsis-3 criteria [34] were used to identify patients with sepsis, defined as a SOFA score \(\ge\) 2 on ICU admission with a suspicion of infection within 24 h before or 24 h after ICU admission. A suspected infection was defined by blood culture sampling and concomitant administration of oral or intravenous antibiotics (24 h before to 72 h after blood culture), as suggested by the sepsis-3 task force [34].

The predefined exclusion criteria for sepsis admissions were: (1) elective ICU admission after elective surgery, and (2) cardiac arrest within 6 h before or 1 h after ICU admission.

Septic shock was defined as the need of a vasopressor, identified by either a cardiovascular SOFA score \(\ge\) 3 or after a medical record review, and a lactate level of \(\ge\) 2 mmol/L among those fulfilling sepsis criteria on ICU admission.

Background and survival data were extracted from the patient administrative system for Intensive care units (PASIVA). PASIVA is the portal by which the treating physician and nursing staff submit prospectively collected laboratory and physiological data to the Swedish Intensive Care Registry. PASIVA is synchronised with the Swedish population register, which contains survival data.

Medical records were reviewed retrospectively by trained data collectors to identify sepsis criteria and additional background data [33].

Blood samples, used for the analysis of bio-ADM, were collected on ICU admission and then centrifuged, aliquoted, frozen, and stored in the SWECRIT biobank at Region Scania (BD-47, SC-1922). Samples collected later than 6 h after ICU admission were excluded. If the sampling time was missing, samples were included if the time of freezing was within 6 h. Frozen plasma samples were shipped, and batch analysis of bio-ADM was performed on thawed samples in March 2019 at the laboratory of SphingoTec GmbH (Hennigsdorf, Germany). The assay has previously been described elsewhere [35].

The study size was not predetermined but rather a convenience sample. All adult ICU admissions from 2016, with valid admission blood samples and consent, in the SWECRIT biobank constituted our study material.

For all hypothesis tests, we considered p values < 0.05 as significant. To assess a difference in the location of two independent variables, we used the Wilcoxon rank-sum test (Mann–Whitney U test). Differences in proportions were assessed using Pearson’s \(\chi ^2\) test. Medians were reported with their corresponding interquartile ranges (IQR). The Swedish 2016 calibration of the Simplified Acute Physiology Score III (SAPS3) was used to calculate the estimated 30-day mortality risk (EMR_30-day) [36, 37]. Multivariable binary logistic regression, adjusted for age, was used to analyse outcomes. The results of the regression analyses are reported as odds ratios (OR) with 95% confidence intervals (CI). The regression models were evaluated with the Hosmer–Lemeshow goodness-of-fit test with ten groups, and models resulting in significant tests were marked [38]. To adjust for severity of disease, SAPS3 was included in the regressions. If a parameter, due to skewness, needed transformation, the base 2 logarithm was used. The difference in Kaplan–Meier curves was evaluated with the log-rank test [39]. Areas under the curve (AUC) were derived from receiver operating characteristic (ROC) curves[40]. Differences in AUCs were tested with the method of DeLong et al. [41]. Youden’s index derived thresholds were reported [42]. Admissions with missing data for any variable were excluded for mean and median calculations. If a variable had missing values, the number of observations available was specified.

There were 2724 adult admissions in 2016. After merging and exclusion, 1867 admissions with valid samples remained, constituting our study population, shown in Fig. 1. The ICU study population was then divided into a sepsis and a non-sepsis cohort, with 632 and 1235 admissions, respectively.

Patients in the sepsis cohort were generally older and sicker on admission with higher illness severity scores than patients in the non-sepsis cohort, as seen in Table 1. Septic patients were, to a greater extent, admitted from within the hospital, while non-septic patients more often were admitted from the emergency department and directly after surgery. The suspected focus of infection for the sepsis patients is shown in Additional file 1: Table S1. Positive blood cultures with the most common pathogens are displayed in Additional file 1: Fig. S1.

Mortality rates for the ICU population, sepsis cohort and non-sepsis cohort are shown in Table 1. The sepsis cohort had worse survival data, a greater need for organ support with significantly higher cardiovascular SOFA scores and a higher proportion of CRRT, and a longer ICU stay. A more detailed description of sepsis patients, divided into 30-day survivors and non-survivors, is shown in Table 2. Sepsis patients who did not survive were older and sicker, but with similar pre-existing comorbidities and a similar degree of septic shock on ICU admission, as survivors. Forty-two per cent of sepsis patients fulfilled the septic shock criteria on admission. This subgroup had a 30-day mortality rate of 30.1%, compared to 25.2% in non-shock patients (p = 0.15). EMR_30-day among septic shock patients was 40.3% (22.5–58.9), while non-shock patients had an EMR_30-day of 24.2% (12.2–44.5).

The range of bio-ADM was 8–4689 pg/mL, and since the distribution was highly skewed, a logarithmic transformation was used, see Fig. 2. The median time from admission to sampling was 25 min (15–40).

Dividing patients by quartiles of bio-ADM resulted in significant survival separation in the sepsis cohort as well as in the entire ICU population, as seen in Fig. 3.

Within the sepsis cohort, non-survivors had significantly higher levels of bio-ADM compared to survivors, shown in Table 2.

The associations of bio-ADM in the regression models for 30-day mortality were almost identical in the sepsis cohort and in the entire ICU population, as in Table 3. A doubling of bio-ADM generated a 22–23% increased OR for death.

The predictive accuracy for bio-ADM and 30-day mortality in the sepsis cohort, presented as AUC, in addition to c-reactive protein (CRP) and lactate are shown in Additional file 1: Table S2.

The association of bio-ADM with CRRT was strong in the sepsis cohort with OR 1.97 (1.64–2.36) but even stronger in the general ICU population, OR 2.28 (2.01–2.59). The ORs for a cardiovascular SOFA score 3 or 4 were 1.33 for both the septic (1.17–1.50) and the general ICU patients (1.23–1.42), as in Table 3.

The median bio-ADM in the sepsis cohort was more than twice as high as the median in the non-sepsis group, as in Table 1. The median bio-ADM in the septic shock subgroup was 107 pg/mL (58–188) compared to 62 pg/mL (35–116) in sepsis patients not presenting with shock (\(p< 0.001\)). In Table 3, the association of increased bio-ADM levels and the risk of having sepsis and septic shock is presented. The OR of having sepsis in the entire ICU population was 1.78 (1.64–1.94) after adjustment for severity of disease.

In the ICU population, the AUC (95% CI) of bio-ADM to identify sepsis was 0.76 (0.73–0.78), see Additional file 1: Table S2. A Youden’s index derived threshold of 37 pg/mL for detecting sepsis resulted in a sensitivity and specificity of 61% and 80%, respectively.

The cut-off of 70 pg/mL separated the ICU population into high and low bio-ADM, as shown in Table 1. The same information is shown graphically in Fig. 2. The sensitivity for 30-day mortality using a cut-off of 70 pg/mL was 42% with a corresponding specificity of 73% in the ICU population. For the sepsis cohort, the sensitivity and specificity were 60% and 50% for 30-day mortality, respectively. Kaplan–Meier curves and results from log-rank tests for bio-ADM levels above or below 70 pg/mL are displayed in Fig. 4. Youden’s index identified a threshold for survival prediction of 45 pg/mL in the ICU population and 108 pg/mL in the sepsis cohort. A separate Kaplan–Meier curve for the sepsis cohort using the Youden’s index-derived cut-off of 108 pg/mL is shown in Fig. 4. Sensitivity, specificity, positive predictive values, negative predictive values, positive and negative likelihood ratios for all cut-offs are displayed in Additional file 1: Table S2.

Our sepsis cohort was identified using a structured method where ICU admissions were manually screened for sepsis-3 and septic shock criteria within a narrow time window at ICU admission. Hence, the sepsis diagnosis was not based on discharge diagnose coding, which has been shown to be misleading [33, 43, 44]. We applied predefined exclusion criteria in order to ensure that our sepsis cohort would represent clinically relevant sepsis patients requiring intensive care.

Interestingly, bio-ADM on admission was associated with mortality in sepsis patients and in the general ICU population in a similar fashion. When included in the same regression model for 30-day mortality, lactate and bio-ADM both contributed independently of each other, indicating that bio-ADM carries additional information in sepsis. In line with this, Blet and colleagues, reported added prognostic value of bio-ADM in addition to lactate among septic patients [45].

Bio-ADM has repeatedly been shown to be associated with increased morbidity [15, 16], which also was evident in our study. Sepsis patients were generally sicker and had significantly higher bio-ADM than the general ICU population. Further, patients with septic shock had significantly higher levels of bio-ADM, which is in agreement with previous reports [15, 16, 46].

The association of bio-ADM with sepsis remained after adjusting for severity of disease, implying that elevated levels of bio-ADM on ICU admission makes it more likely that a patient has sepsis.

To our knowledge, there have been no previous reports on the sepsis discriminating properties of bio-ADM in a general ICU population. The ability of bio-ADM to identify sepsis patients was modest with an AUC of 0.76. A Youden’s index derived cut-off of 37 pg/mL generated a sensitivity of 61% and a specificity of 80%, which indicates limited clinical utility of that cut-off.

The finding that bio-ADM could be broadly applicable to critically ill patients has been reported previously [46]. Lemasle and colleagues studied a large population of patients requiring vasopressor or invasive ventilation for more than 24 h and found an association of bio-ADM with mortality and need for organ support. Their patient population was, however, sicker in comparison with ours. In addition, the bio-ADM samples were not admission samples, which could explain the lower bio-ADM median level in our study (40 pg/mL versus 66 pg/mL).

In spite of the questionable rationale of using a cut-off of 70 pg/mL for bio-ADM in sepsis, it has been used in several studies since it was first proposed [12].

In the present study, the 70 pg/mL cut-off managed to separate survivors from non-survivors, but a Youden’s index derived cut-off of 108 pg/mL performed better in sepsis patients, see Fig. 4. Interestingly, Mebazaa et al. reported a similar Youden’s index cut-off of 102 pg/mL from their sepsis cohort in a recent study [15]. For the entire ICU population, the Youden’s index identified the cut-off 45 pg/mL, which is a novel finding for bio-ADM.

There are several limitations to this study.

The study was designed to focus on bio-ADM levels in sepsis patients. All ICU admissions were initially screened for sepsis-3 criteria, and the aim of our data retrieval was primarily to collect detailed data from this cohort. For the remaining ICU population, collection of data was by necessity limited to the PASIVA database, which resulted in different data availability for the sepsis and non-sepsis cohorts. We did, for example, not collect data on comorbidities systematically nor lactate or c-reactive protein levels in the non-sepsis cohort.

We did not have information on the volume status of the patients, nor whether adequate volume resuscitation measures were taken before vasopressor treatment was commenced, a diagnostic criterion for septic shock. However, this limitation is a common feature of studies aiming at identifying septic shock. Initiation of vasopressor therapy in the ICU would usually imply that adequate fluid resuscitation was done, assuming adherence to the Surviving Sepsis Guidelines [47].

We used a strict time frame in which we identified the sepsis and non-sepsis patients and did not investigate the development of sepsis or septic shock beyond that time. Our time constraint may have underestimated the diagnostic value of bio-ADM in sepsis. On the other hand, our method of retrospectively identifying patients fulfilling the sepsis criteria has probably identified patients who were not considered clinically septic by the treating physician.

We were confined to admission samples only, and could not investigate dynamic changes in bio-ADM levels and the impact these may have had on reported outcomes.

The mortality rate in our sepsis and septic shock subgroups was somewhat lower than expected, which could make our results difficult to generalise to patient populations outside of Scandinavia.