Superior accuracy of mid-regional proadrenomedullin for mortality prediction in sepsis with varying levels of illness severity

This study was a prospective observational cohort of patients recruited consecutively from two intensive care units (ICU) in Spain and France. Adult patients with age ≥18 years and admitted to the ICU from April 2013 to January 2016 were enrolled within 12 h after meeting criteria for severe sepsis or septic shock, based on the SEPSIS-2 definition by the American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference [14]. Enrolled patients also had SOFA score ≥2 and therefore met criteria for the new SEPSIS-3 definition for sepsis [15]. Patients with human immunodeficiency virus (HIV) infection and those undergoing radiotherapy or receiving immunosuppressive drugs, including chemotherapy or systemic steroids, in the 3 months prior to admission to the ICU were considered to be immunosuppressed. Exclusion criteria were age <18 years, the presence of pregnancy, the absence of a blood sample available for biomarker profiling within the first 12 h following ICU admission, or lack of informed consent. Clinical data recorded from the medical records included demographics, comorbidities, laboratories, microbiology, and biomarker levels. The severity of illness was assessed on admission by calculating the Sequential Organ Failure Assessment (SOFA) score.

Plasma samples for biomarker profiling were collected as close as possible to the moment of ICU admission, and always within the first 12 h. Plasma MR-proADM measurement was performed by TRACE technology (Time Resolved Amplified Cryptate Emission) using a new sandwich immunoassay (Kryptor Compact Plus Analyser, BRAHMS, Hennigsdorf, Germany); limit of detection 0.05 nmol/L. PCT measurement was performed by electrochemiluminescence immunoassay (ECLIA) on a chemistry analyser (Cobas 6000, Roche Diagnostics, Meylan, France); limit of detection 0.02 ng/ml. Serum CRP and lactate were measured by particle-enhanced immunoturbidimetric and colorimetric assay, respectively (e501 Module Analyser, Roche Diagnostics, Meylan, France); limit of detection 0.15 mg/dL and 0.2 mmol/L, respectively.

Differences in demographic and clinical characteristics between survivors and non-survivors were assessed using the Chi-square test for categorical variables. Student’s t test or Mann–Whitney U test were, respectively, used to compare continuous variables based upon the presence or absence of normal distribution. The association between biomarkers and the risk of mortality was assessed by Cox regression analysis, adjusted by confounding variables. Time was censored at 28 days following admission to the ICU. The first 24 h of ICU admission was considered as day 1 in the analysis. Variables yielding a p < 0.05 in the univariate regression analysis were further included in the multivariate analysis. Biomarkers were log transformed in order to reach a normal distribution. The impact of biomarkers on mean survival time was assessed by using Kaplan–Meier curves and the Mantel–Haenszel log-rank test. Similar to the Cox regression analysis, time was censored at 28 days following admission to the ICU. Accuracy and predictive values of the biomarkers for mortality were evaluated by calculating the area under the receiver operating characteristic (AUROC) curve. Patients were distributed into three groups depending on disease severity as assessed by the SOFA score using two predefined cut-offs, one with a sensitivity close to 90% and the other showing a specificity close to 90% for detecting non-survivors at 28 days (Additional file 1: Figure S1). Data were analysed by using the IBM SPSS 20.0 software (SPSS, Chicago, Ill).

Three hundred and twenty-six patients (326) with severe sepsis (21.7%) or septic shock (79.3%) were enrolled with a 28-day mortality rate of 25.5 and 34.9% in Valladolid and Dijon, respectively, and an overall mortality rate of 31.0% across both sites (Table 1). The median age was 65 years and 54.3% of patients were male. Compared to survivors, non-survivors were older and presented with higher SOFA scores, and an increased incidence of septic shock, mechanical ventilation, renal replacement therapy, neoplasia, cardiovascular disease, chronic renal failure, immunosuppression, and respiratory disease (all p < 0.05). The most common source of infection was of respiratory and urologic origin, regardless of outcome. Mortality rates depending on the source of infection were as follows: 37.2% in patients suffering from a respiratory infection, 32.4% in those with an urological infection, 28.6% in patients with an abdominal infection, 35.5% in those showing a primary or secondary bacteremia and 25.6% in those patients with an infection of other origin. Regarding microbiological identification, both survivors and non-survivors showed a similar presence of Gram−, Gram+ and virus pathogens. Fungal infections were more frequent in non-survivors. The most common cause of death was multi-organ dysfunction syndrome (n = 58; 57.4%), followed by refractory shock (n = 9; 8.9%) and refractory hypoxemia (n = 8; 7.9%). A limitation of therapeutic effort was applied to 21 patients. MR-proADM, PCT and lactate concentrations were all significantly elevated in non-surviving patients compared to survivors (all p < 0.01), whereas CRP levels remained similar in both groups. Levels of MR-proADM depending on the source of infection were as follows [median (interquartile range)] respiratory infection [3.6 nmol/L (5.6)], urological infection [4.6 nmol/L (5.4)], abdominal infection [4.9 nmol/L (6.5)], bacteremia [3.8 nmol/L (5.1)], and [3.5 nmol/L (5.8)] in infections of other origin. Levels of MR-proADM depending on the infecting microbe were [median (interquartile range)]: fungal infection [6.1 nmol/L (5.6)], Gram − bacteria [4.9 nmol/L (5.9)], Gram + bacteria [4.1 nmol/L (6.2)] or viruses [1.2 nmol/L (3.4)].

MR-proADM, PCT and lactate showed a significant association with mortality in the univariate Cox regression analysis (Table 2). After adjusting for confounders and compared to PCT, CRP, and lactate, MR-proADM showed the strongest independent association with the risk of mortality (hazard ratio 8.5; 95% confidence interval 4.2–17.4; p < 0.001; Table 2). In addition, Kaplan–Meier analysis showed that no patients with a MR-proADM value ≤0.88 nmol/L died in the first 28 days following ICU admission (Fig. 1). This cut-off was selected since it provided a sensitivity of 100% in identifying non-survivors in the AUROC (Fig. 2).

Adjusting variables were: age, septic shock, cardiovascular disease, immunosuppression, chronic renal failure, neoplasia, respiratory source of infection, renal replacement therapy, hospital (Valladolid/Dijon), presence of fungal infection, limitation of therapeutic effort.

MR-proADM showed the best AUROC for mortality prediction at 28 days in the analysis over the entire cohort, even better than that of SOFA score (Fig. 2). When patients were stratified by the degree of organ failure, MR-proADM was the only biomarker able to discriminate non-survivors from survivors at 28 days in those patients with the lowest degree of disease severity (SOFA score ≤ 6), (AUROC [95% confidence interval (95% CI)] 0.75 [0.61–0.88]), (p = 0.006) (Fig. 3). In the moderately severe patients (SOFA score 7–12), MR-proADM showed a higher AUROC than that observed with lactate (0.74 [0.66–0.83] vs. 0.61 [0.52–0.71], respectively, Fig. 3). In the most severe patients (SOFA score ≥13), MR-proADM and lactate had a similar AUROC (0.73 [0.59–0.86] vs. 0.72 [0.59–0.86], respectively, Fig. 3). Neither CRP nor PCT was predictive of 28-day mortality in any severity group based on the SOFA score, with AUROCs ranging from 0.43 to 0.60.

The threshold values (cut-off) of MR-proADM were determined by having the highest specificity with a pre-fixed sensitivity of at least 0.80 for identifying non-survivors. For patients with SOFA scores ≤6, 7–12, and ≥13, the MR-proADM cut-offs were 1.79, 3.25, and 5.58 nmol/L, respectively (Table 3).

The length of ICU stay in each severity group was [mean, (SD)]: SOFA ≤ 6: 11.0 days (18.3); SOFA 7–12: 12.4 days (16.0) and SOFA ≥ 13: 8.4 days (7.6). The mortality rates for each severity group were 12.8, 30.6 and 59.7%, respectively.

We evaluated the combination of MR-proADM and SOFA score in predicting mortality, such that patients with MR-proADM concentrations >1.79 nmol/L were considered to have a 1 point increase in the SOFA score. In patients with SOFA ≤ 6, the MR-proADM modified SOFA score (ADM-SOFA) showed an increased ability to identify non-survivors compared to SOFA alone, AUROC [95% CI] SOFA 0.70 [0.58–0.82] and ADM-SOFA 0.77 [0.66–0.88].