Enhancing-ARDS diagnostics for ICU patients: a retrospective, nested case-control study to develop a biomarker-based model


 Background: To investigate whether a series of biomarkers including club cell protein 16 (CC16), angiopoietin 2(Ang-2), soluble receptor for advanced glycation end-products (sRAGE), high-mobility group box 1 protein (HMGB1), and surfactant protein D (SPD) could be utilized for identifying patients, thereby increasing the diagnostic value of acute respiratory distress syndrome(ARDS) in intensive care unit (ICU). Methods: 211 ICU admissions were enrolled in this retrospective, nested case-control study. These patients were then divided into ARDS (n=79) and non-ARDS (n=132) groups according to the Berlin criteria on ICU day 1. Patient characteristics, vital signs, and laboratory examinations were collected within three hours of admission. Five inflammatory associated plasma biomarkers, as well as lung epithelial and endothelial injury which included CC16, Ang-2, sRAGE, HMGB1 and SPD were measured in the morning of day two in the ICU. Diagnostic values were analyzed with receiver operating characteristic (ROC) curves. Pearson’s product-moment correlation coefficient and multivariate logistic regression analysis were applied for predictive purposes. Results: C-reactive protein (CRP), NT-proBNP, and PH values for traditional indicators and five biomarkers were analyzed with an objective ARDS indicator, the PaO2/FiO2 ratio. Evidence suggests that only four of potential indicators analyzed here, and CRP hold high diagnostic value. The area under curve (AUC) for each were as follows: CC16 (AUC: 0.752; 95%CI0.680-0.824), Ang-2 (AUC: 0.695; 95%CI 0.620 -0.770), HMGB1 (AUC: 0.668; 95%CI 0.592-0.744), sRAGE (AUC: 0.665; 95%CI 0.588-0.743), CRP (AUC: 0.701; 95%CI 0.627-0.776). No single indicator surpassed the diagnostic capability of the PaO2/FiO2 ratio which had an AUC: 0.844(95%CI 0.789-0.898), especially in terms of sensitivity. However, when the binary logistic model was transformed and the model was built, the AUC increased from 0.647(95%CI 0.568-0.726) to 0.911(95%CI 0.864-0.946). Among the combinations tested, PaO2/FiO2+CRP+Ang-2+CC16+HMGB1 resulted in an AUC of 0.910 (95%CI 0.863-0.945), while PaO2/FiO2+CRP+Ang-2+CC16+HMGB1+sRAGE+SPD have an AUC of 0.911(95%CI 0.864-0.946). Conclusions: A combination of the assessed biomarkers could enhance ARDS diagnostics, which has obvious ramifications for patient care and prognosis. It may be possible to develop a predictive ARDS nomogram; however, of the combinations tested here, we would recommend PaO2/FiO2+CRP+Ang-2+CC16+HMGB1 for clinical practice. This is because of the cost implications in contrast with the benefit involved in utilizing the more elaborate model. Although, further health economics research is required to consider this opportunity cost for emergency care policy.


Introduction
Acute respiratory distress syndrome (ARDS) is a common disease characterized by permeability pulmonary edema and refractory hypoxemia in intensive care unit (ICU) 1 . ARDS is associated with high morbidity and mortality rates although, it remains underdiagnosed and therefore all too often untreated 2 . The traditional diagnostic methods mentioned in the Berlin definition 3 , such as the ratio of partial pressure in arterial oxygen over the fraction of inspired oxygen (PaO2/FiO2) and standard X-rays could perhaps be updated. The prevalence of ARDS and underdiagnosis with delays in identification of cases limits the effect of any intervention/s once administered. Physicians must have the most advanced diagnostics in order to improve clinical decision-making and at present this may not be the case for ARDS.
Biomarkers are indicators of pathophysiological processes, and prove insight into the biological responses to therapeutics. Biomarkers are increasingly becoming common place in both clinical research for participant selection but also for clinical practice such as triage. For example, biomarkers such as procalcitonin (PCT), Creactive protein (CRP), and interleukin-6 (IL-6) are frequently measured for sepsis, and Cardiac Troponin-I (cTnI) is now commonly used for acute myocardial infarction. These biomarkers provide timely information which can be effectively used to tailor therapeutic strategies around individual characteristics 4,5 . Therefore, having an appropriate biomarker (or combination of biomarkers) can help to determine disease characteristics as well as the sensitivity and specificity of an intervention/s.
To date, more than 20 biomarkers have proven useful for diagnosing or predicting ARDS 1 . Fremont 6 and Ware 7 have also proposed that combining biomarkers will enhance ARDS diagnostics compared to a single biomarker. This is, because a specific combination is likely to improve both the accuracy and reliability of diagnosis and therefore prognostics. However, when developing biomarker-based models, it is necessary to avoid repeating measures for those with dissimilar pathophysiologies. Therefore, biomarkers derived through basic research into different pathophysiological pathways for inflammation, lung epithelial and endothelial injury were chosen for this study. These included club cell protein 16 (CC16), angiopoietin 2(Ang-2), soluble receptor for advanced glycation end-products (sRAGE), high-mobility group box 1 protein (HMGB1) and surfactant protein D (SPD) 2 .

Study population
ICU patients were enrolled between March 2013 and March 2017. The following eligibility criteria were necessary for inclusion: 1) patient age > 18 and < 75; 2) expected ICU stay > 24 h; 3) blood samples were collected < 6 h after admission; 4) diagnosis had been conformed prior to discharge. Those who did not meet all of the criteria were excluded. Written informed consent was then requested from potential participants or their legal representatives. The institutional human ethics committee of the affiliated Baoan Hospital of Shenzhen, Southern Medical University approved our study protocols (BYL 20141007). Research involving human participants, human material, or human data have been performed in accordance with the Declaration of Helsinki. All methods were carried out in accordance with relevant guidelines and regulations.

Data collection and outcome of patients
Values at baseline were recorded within 3 h of admission to the ICU, including individual characteristics (i.e., age, gender, comorbidities, and risk factors of ARDS), Acute Physiology and Chronic Health Evaluation II score (APACHE II), vital signs (i.e., blood pressure, body temperature, respiratory rate, and heart rate).
Physiological variables for the PaO2/FiO2 ratio, as well as C-reactive protein (CRP), white blood cell count (WBC), the N-terminal of the prohormone brain natriuretic peptide (NT-proBNP), PH value, serum total protein (TP), D-Dimer, serum creatinine concentration (Scr), and lactic acid (Lac) were determined synchronously within 3 h of admission. Duration of mechanical ventilation (MV), mortality at Day-7, and Day-28 were recorded for all participants.

Diagnosis criteria and subgroups
ARDS is diagnosed according to the Berlin definition 3 which stipulates: 1)ARDS has an acute onset, of less than 7 days; with 2) bilateral opacity (consistent with pulmonary edema), as detected by CT or X-ray; and 3) PaO2/FiO2 ratio of less than 300 mmHg, with ventilation support (Positive End Expiratory Pressure or Continuous Positive Airway Pressure≥5 mmH2O). Two senior physicians make a diagnosis based on patients' conditions within the first 48 hours of admission. All participants were divided into an ARDS or non-ARDS groups for further retrospective analysis.

Measurement of serum CC16, Ang-2, sRAGE, HMGB1 and SPD
Blood samples at Day-1 and Day-2 were separately collected from the radial artery within 3 hours and 24 hours of admission to ICU. Blood samples were then centrifuged at 3000 rpm for 10 min and upper serum stored in EP tubes at an ultra-low temperature refrigerator (-80℃) until required for analysis. Serum CC16, Ang-2, sRAGE, and SPD concentrations were determined using ELISA kits (R&D Systems, Minneapolis, USA) and HMGB1 concentration were determined using ELISA kits (Elabscience Biotechnology Co., Ltd, Wuhan, Hubei, China), following the manufacturer's instructions. Each sample was measured in duplicate and assaying was conducted using the ELISA kits. The researchers who performed these analyses were blinded to group assignment.

Statistical analysis
Data are presented as the means with standard deviations or numbers (proportion) as indicated. Student's t-test was then performed to compare serum concentrations for Ang-2, CC16, sRAGE, SPD, and HMGB1 between the two groups, when distribution was considered to be normal. Conversely, when Guassian distribution was not evident, we adopted Mann-Whitney's U-test. Categorical data were compared using standard Chi-square tests. Correlations between Ang-2, CC16, sRAGE, SPD, HMGB1, PaO2/FiO2, PH, WBC, CRP, NT-proBNP, TP, D-Dimer, albumin, and Scr were estimated using Pearson's linear regression coefficients.
Receiver operating characteristic (ROC) curves were utilized to assess the optimal area under the curves (AUC) with corresponding 95% confidence intervals (CI). The optimal cut-off value, sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) were calculated after taking Youden's maximum (i.e., sensitivity+specificity-1). Statistical analysis was performed using R (v3.5.1, R Foundation for Statistical Computing, Vienna, Austria); A p-value <0.05 was considered the threshold for statistical significance.

Participants and demographics
Between 1 st March, 2013, and 1 st March, 2017, 1,121 ICU patients were initially considered eligible. Please see Fig. 1 for a complete flowchart of the process from screening to analysis. Those discharged within 24 hours (n =253), those beyond the age range (n = 198) and those who did not provide informed consent within the prespecified 12-hour window (n =78), were excluded. A further 381 potential candidates were excluded due to incomplete blood samples (n =198), incomplete clinical data (n =152) and discharge with unclear diagnosis (n =31). A final cohort of 211 patients was recruited, of whom 37.4% (n=79) had been diagnosed with ARDS within the first day, and 62.6% (n=132) who did not meet the Berlin criteria ( Fig.1). Demographics and clinical characteristics of two groups are presented in Table 1.
The p values provided are the result of comparisons between the ARDS and Non-ARDS groups using Student's t test, or Chi-square test. The median age was 54.68 (±18.94) years and 63.3% (50/79) were male in ARDS group. At baseline, patients in the ARDS group had higher CRP levels and lower PaO2/FiO2 ratios than patients in non-ARDS group. Comorbidities were similarly distributed in both groups.
Pneumonia had a higher rate in ARDS group although; no other major risk factors were significantly different between the groups. APACHE II scores for the severity of illness, ventilation time, length of ICU stay and overall hospitalization were not significantly different between the two groups. However, 7-day mortality and 28-day mortality were higher in ARDS group.

Validation of a biomarker model for diagnosis of ARDS
Levels of the five plasma biomarkers recorded from the entire cohort are shown in Table 2. Bar graphs were used to compare CC16, Ang-2, sRAGE, HMGB1 and SPD levels between the ARDS and non-ARDS groups. Of the five biomarkers, four significantly differed between groups under univariate analysis at ICU Day 1 and Day 2, i.e., CC16, Ang-2, sRAGE and HMGB1. See Figure 2 for confirmation.  The p values were derived from using student's t-test or Mann Whitney U test.
A logistic regression model was developed for ARDS diagnosis using all 5 biomarkers, and model performance was assessed using AUCs. As can be seen in

Improvement of ROC value for the diagnosis of ARDS
By translating from a binary logistic model, it was possible to build a model using various biomarker combinations (  The PaO2/FiO2 ratio is the traditional indicator for the diagnosis of ARDS, whereas CRP is an index that is relatively easy to obtain. This evidence also confirms that CRP has a significant influence in the diagnosis of ARDS (Table   3), despite negatively correlating with PaO2/FiO2 (Figure 3). We tried to combine PaO2/FiO2, CRP, and panels of biomarkers (

Relationship of biomarkers with outcomes
Across the entire ICU sample, higher serum levels of CC16 (Fig.4a to Fig.4d) and Ang-2 ( Fig.4e to Fig.4h) were found in non-survivors. Consistently, CC16 at Day-1(r=0.189, Fig.5a), Ang-2 at Day-1 (r=0.158, Fig. 5b) and Day-2 (r=0.143, Fig. 5c) showed positive correlation with MV duration in the entire ICU patients. However, only CC16 on day 1 and day 2 showed higher mortality and statistical significance in 7-day and 28-day mortality in ARDS group. See Figure 4i to 4l for further details.  r values were accounted from using Pearson correlation coefficient.

Discussion
The primary objective was to consider whether intercalating series of biomarkers into traditional ARDS diagnostics would increase diagnostic accuracy for critically ill patients. We initially sought to confirm whether biomarkers such as CC16, Ang-2, HMGB1 and sRAGE would prove useful for patients with ARDS. We then began to analyze potential correlations between theafore mentioned biomarkers and traditional indicators. We found that these biomarkers can supplement routine indicator such as the PaO2/FiO2 ratio and could prove useful in clinical practice. We also found that CC16, Ang-2 and HMGB1 detection when combined with PaO2/FiO2 and CRP can improve the diagnostic accuracy of traditional methods. Our evidence also suggests that CC16 and Ang-2 may prove useful for assessing lung function recovery, which positively correlates with mechanical ventilation time. Further, we speculate that CC16 might be useful as a prognostic indicator for ARDS patients who  We also found that there are physiological interrelations between some of the investigated biomarkers and traditional indicators. Evidence from our previous research 19 and other 20,21 suggests that CC16 positively correlates with Scr, Lac and NT-proBNP. This might related to the anti-inflammatory function, immune system activation, and lactate metabolism pathway related to prephosphorylation. Likewise, the relationship between Ang-2 and albumin, and CRP, is partly consistent with previous results 22 , most of which were obtained in clinical samples. It was found that the pro-inflammatory properties may affect the leukocyte adhesion mechanism in inflammation. The association between sRAGE and NT-proBNP we observed might be also affected by cardiac remodeling and anti-inflammatory effects. Both BNP and NT-pro-BNP levels elevate in patients with chronic renal insufficiency, which are closely related to left ventricular hypertrophy and abnormal systolic function.
Therefore, both have the potential to predict heart failure and mortality. This hypothesis seems to be supported by the results obtained in other types of patients 23-26 , where elevated levels of sRAGE were considered a sign of worsening cardiac function and mortality.
From a prognostic perspective, the association between serum Ang-2 and mortality is consistent with previous findings in patients with sepsis 27,28 . Consistently, multivariate regression analysis has shown that non-survivors are more likely to have biological dysfunction when admitted, which may help integrate biomarker-based predictive models to support clinicians assessing critically ill patients before confirming diagnosis. SPD appears to have diagnostic value for ARDS and shows superiority in prognostic value of interstitial lung disease 29 . However, previous research has found that the level of SPD is not related to lung contusion associated with ARDS 30 . In that study, the researchers also found that serum CC16 level is related to the volume of lung contusion and may not be affected by the overall severity of injury, age, gender or ventilation. This supports our findings that SPD should not be used alone to diagnose and evaluate the prognosis of ARDS.
However, we found that the CC16 levels on Day-1 and Day-2 correlated with 7-day and 28-day mortality. Based on the pathophysiology of ARDS, prognosis appears to be reflected in CC16 levels which we suggest related to the blood gas barrier repair process. In our previous study 31 , we discovered the prognostic value of CC16 for non-invasive ventilation in critically ill patients. The findings of this study elaborate on this; suggesting that serum CC16 and Ang-2 in critically ill patients are closely related to the duration of mechanical ventilation. However, this is a tentative notion because subsequent lung infections such as hospital-acquired pneumonia, or ventilator-associated pneumonia, may have influenced our findings.

Strength and limitations
To the best of our knowledge, this is one of the few studies which combine plasma-based biomarkers and traditional indicators to evaluate ARDS in more than 200 patients 32,33 . We also think by recruiting various critically ill patients, our research can be repeated at the local emergency and critical care center. The recommended combination could help clinicians identify ARDS in critically ill people although the findings will need to be externally validated across a larger Chinese population.
The patients enrolled in this study were also pretreated, prior to receiving ICU treatment. This of course creates, inconsistencies across an otherwise genetically population. Future research might try to develop a nomogram for this population although effort should be taken to reduce selection bias, where possible. There are a number of prospective advantages in intercalating plasma-based biomarkers, yet remain difficult to adopt these in clinical practice without commercially produced instanteous testings. Given this, we were only able to perform testing at the early stage of patient admission, not throughout the entire period of hospitalization.
However, we hope this evidence adds to a necessary, growing body of evidence and will help those considering the design of large-scale, prospective clinical studies in this field.

Conclusion
We investigated the detection of serum CC16, Ang-2, sRAGE, SPD and HMGB1 to identify patients with ARDS in those considered to be critically ill. Using an iterative approach to developing a combined detection method, we found that Ang-2+CC16+HMGB1+CRP +PaO2/FiO2 enhances ARDS diagnostics substantially.
We also found that CC16 and Ang-2 might effect mortality and mechanical ventilation time. Further research into the underlying mechanisms is currently underway, which we hope will improve our understanding of ARDS physiology. We would also suggest there may be a need to develop a predictive nomogram as well as for health economics research to consider the trade-offs between the more basic model recommended here and the more sophisticated model, which is marginally more accurate.

Declarations Ethics approval and consent to participate
Research involving human participants, human material, or human data have been performed in accordance with the Declaration of Helsinki. All methods were carried out in accordance with relevant guidelines and regulations. The Institutional Human Ethics Committee of affiliated Baoan Hospital of Shenzhen, Southern Medical University approved the study protocols employed in this observational study. Written informed consent was obtained from each subject or their legal guardians.

Consent for publication
Not applicable.

Availability of data and material
The datasets used and analyzed during the current study are available from the corresponding author in response to reasonable requests.

Competing interests
The authors declare that they have no competing interests. All of funders equally contribute to this investigation.