DETECTING SEPSIS

Abstract

A method for predicting sepsis or diagnosing systemic inflammatory response syndrome (SIRS) and/or sepsis in a subject comprises determining levels of at least three markers selected from CCL23, A1AT, CRP, sICAM, PLA2, IL-6, procalcitonin, MMP8, TNFalpha, AcPGP, enzymatic MMP activity, TIMP1, sRAGE and desmosine in a sample taken from the subject. The combined levels of the at least three markers are used to predict or diagnose SIRS and/or sepsis. The methods may be performed on a subject with SIRS and which is used to identify an infection in the subject. A preferred panel of markers includes CCL23, A1AT, sICAM, sICAM/VCAM-1 and CRP. Corresponding products, methods of treatment and medical uses are provided.

Claims

1-49. (canceled)

50. A method of treating sepsis in a subject identified as having sepsis on the basis of PLA2G2A, CRP, sICAM, and IL-6 levels, the method comprising: detecting the levels of PLA2G2A, CRP, sICAM, and IL-6 in a sample from the subject; identifying the subject as having sepsis based on an increased level of PLA2G2A, CRP, sICAM, and IL-6 in the sample as compared to healthy control; selecting the subject so identified for treatment with an antibiotic; and administering the antibiotic to the selected subject, thereby treating sepsis.

51. The method of claim 50, further comprising detecting the level of at least one additional marker selected from A1AT, procalcitonin, MMP8, TNFalpha, AcPGP, —enzymatic MMP activity, TIMP1, sRAGE, and desmosine in the sample from the subject, and wherein identifying the subject as having sepsis is further based on an increased level of at least one additional marker selected from A1AT, procalcitonin, MMP8, TNFalpha, enzymatic MMP activity, TIMP1, sRAGE, and desmosine, as compared to a healthy control, and/or a decreased level of AcPGP as compared to a healthy control.

52. The method of claim 50, wherein the levels of PLA2G2A, CRP, sICAM and IL-6 are determined using a lateral flow assay.

53. The method of claim 50, wherein the subject is a hospitalized patient and/or an immunocompromised patient.

54. The method of claim 50, wherein the sample is a whole blood, plasma or serum sample.

55. The method of claim 50, wherein the antibiotic is selected from an aminoglycoside, a cephalosporin, a glycopeptide, a penicillin, a quinolone, aztreonam, clindamycin, imipenem-cilastin, linezolid, metronidazole, rifampin and an antifungal.

56. A method of selecting a subject for treatment with an antibiotic comprising: determining levels of at least five markers selected from CRP, sICAM, PLA2G2A, IL-6, and A1AT in a sample taken from the subject, wherein the combined levels of the at least five markers are used to predict or diagnose sepsis; selecting the subject for treatment where sepsis is predicted or diagnosed; and administering an antibiotic to the subject where sepsis is predicted or diagnosed.

57. The method of claim 56, comprising determining the levels of at least one additional marker, wherein the at least one additional marker is selected from CCL23, procalcitonin, MMP8, TNFalpha, AcPGP, enzymatic MMP activity, TIMP1, sRAGE, and desmosine.

58. The method of claim 56, wherein the level of the at least five markers is determined using a lateral flow assay.

59. The method of claim 56, wherein the subject is a hospitalized patient and/or an immunocompromised patient.

60. The method of claim 56, wherein the sample is a whole blood, plasma or serum sample.

61. The method of claim 56, wherein the antibiotic is selected from an aminoglycoside, a cephalosporin, a glycopeptide, a penicillin, a quinolone, aztreonam, clindamycin, imipenem-cilastin, linezolid, metronidazole, rifampin and an antifungal.

62. A method of treating sepsis in a subject, the method comprising administering an antibiotic to the subject suffering from sepsis, wherein the subject displays, in a sample, an altered level of PLA2G2A and A1AT and at least one marker selected from CRP, sICAM, IL-6, CCL23, procalcitonin, MMP8, TNFalpha, AcPGP, enzymatic MMP activity, TIMP1, sRAGE and desmosine, and wherein the subject was diagnosed as having sepsis based on the altered level of PLA2G2A and A1AT and the at least one marker.

63. The method of claim 62, wherein the at least one marker is selected from CRP, sICAM, and IL-6.

64. The method of claim 62, wherein the level of the markers is determined using a lateral flow assay.

65. The method of claim 62, wherein the subject is a hospitalized patient and/or an immunocompromised patient.

66. The method of claim 62, wherein the sample is a whole blood, plasma or serum sample.

67. The method of claim 62, wherein the antibiotic is selected from an aminoglycoside, a cephalosporin, a glycopeptide, a penicillin, a quinolone, aztreonam, clindamycin, imipenem-cilastin, linezolid, metronidazole, rifampin and an antifungal.

Description

DESCRIPTION OF THE FIGURES

[0353] FIG. 1 shows pie chart representations of the nature of infections detected in the clinical samples

[0354] FIG. 2A shows the origins of infections in sepsis samples

[0355] FIG. 2B shows the origins of infections in sepsis samples infected with gram negative bacteria

[0356] FIG. 2C shows the origins of infections in sepsis samples infected with gram positive bacteria

[0357] FIG. 3 shows multiple ROC curves for the various assays performed on the samples. The source of each curve is indicated.

[0358] FIG. 4 is a scatter plot showing the ability of the CRP, sICAM and TNFalpha marker combination to distinguish sepsis from controls in the D samples

[0359] FIG. 5 is a ROC curve showing the ability of the CRP, sICAM and TNFalpha marker combination to distinguish sepsis from controls in the D samples

[0360] FIG. 6A is a scatter plot showing the ability of the CRP, sICAM and TNFalpha marker combination to distinguish sepsis from controls in the A samples

[0361] FIG. 6B is a box plot showing the ability of the CRP, sICAM and TNFalpha marker combination to distinguish sepsis from controls in the A samples

[0362] FIG. 7 is a ROC curve showing the ability of the CRP, sICAM and TNFalpha marker combination to distinguish sepsis from controls in the A samples

[0363] FIG. 8A is a scatter plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from SIRS

[0364] FIG. 8B is a box plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from SIRS

[0365] FIG. 8C is a ROC curve showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from SIRS

[0366] FIG. 9A is a scatter plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish SIRS from controls

[0367] FIG. 9B is a box plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish SIRS from controls

[0368] FIG. 9C is a ROC curve showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish SIRS from controls

[0369] FIG. 10A is a scatter plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from controls

[0370] FIG. 10B is a box plot showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from controls

[0371] FIG. 10C is a ROC curve showing the ability of the sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α marker combination to distinguish sepsis from controls

[0372] FIG. 11A shows levels of CRP over time in sepsis versus controls

[0373] FIG. 11B shows levels of IL-6 over time in sepsis versus controls

[0374] FIG. 11C shows levels of TIMP1 over time in sepsis versus controls

[0375] FIG. 11D shows levels of sICAM-1 over time in sepsis versus controls

[0376] FIG. 12 shows a point of care lateral flow device useful in the present invention

[0377] FIG. 13 sets out a possible personalised testing strategy that may be adopted according to the invention

[0378] FIG. 14 is a box plot showing the ability of sRAGE to distinguish sepsis from controls in the tested samples

[0379] FIG. 15A is a schematic representation of an ELISA assay format for measuring levels of CCL23 in a sample

[0380] FIG. 15B shows a representative calibration curve for the ELISA assay format for measuring levels of CCL23 in a sample

[0381] FIG. 16 is a schematic representation of a lateral flow assay format for measuring levels of CCL23 in a sample

[0382] FIG. 17A is a schematic representation of an ELISA assay format for measuring levels of CRP in a sample

[0383] FIG. 17B shows a representative calibration curve for the ELISA assay format for measuring levels of CRP in a sample

[0384] FIG. 18A is a schematic representation of a lateral flow assay format for measuring levels of CRP in a sample

[0385] FIG. 18B shows a representative calibration curve for the lateral flow assay format for measuring levels of CRP in a sample

[0386] FIG. 19A is a schematic representation of an ELISA assay format for measuring levels of A1AT in a sample

[0387] FIG. 19B shows a representative calibration curve for the ELISA assay format for measuring levels of A1AT in a sample

[0388] FIG. 20A is a schematic representation of a lateral flow assay format for measuring levels of A1AT in a sample

[0389] FIG. 20B shows a representative calibration curve for the lateral flow assay format for measuring levels of A1AT in a sample

[0390] FIG. 21A is a schematic representation of an ELISA assay format for measuring levels of TNFalpha in a sample

[0391] FIG. 21B shows a representative calibration curve for the ELISA assay format for measuring levels of TNFalpha in a sample

[0392] FIG. 22A is a schematic representation of a lateral flow assay format for measuring levels of TNFalpha in a sample

[0393] FIG. 22B shows a representative calibration curve for the lateral flow assay format for measuring levels of TNFalpha in a sample

[0394] FIG. 23A is a schematic representation of an ELISA assay format for measuring levels of IL-6 in a sample

[0395] FIG. 23B shows a representative calibration curve for the ELISA assay format for measuring levels of IL-6 in a sample

[0396] FIG. 24A is a schematic representation of a lateral flow assay format for measuring levels of IL-6 in a sample

[0397] FIG. 24B shows a representative calibration curve for the lateral flow assay format for measuring levels of IL-6 in a sample

[0398] FIG. 25A is a schematic representation of an ELISA assay format for measuring levels of PLA2G2A in a sample

[0399] FIG. 25B shows a representative calibration curve for the ELISA assay format for measuring levels of PLA2G2A in a sample

[0400] FIG. 26A is a schematic representation of a lateral flow assay format for measuring levels of PLA2G2A in a sample

[0401] FIG. 26B shows a representative calibration curve for the lateral flow assay format for measuring levels of PLA2G2A in a sample

[0402] FIG. 27A is a schematic representation of an ELISA assay format for measuring levels of sICAM1 in a sample

[0403] FIG. 27B shows a representative calibration curve for the ELISA assay format for measuring levels of sICAM1 in a sample

[0404] FIG. 28A is a schematic representation of a lateral flow assay format for measuring levels of sICAM1 in a sample

[0405] FIG. 28B shows a representative calibration curve for the lateral flow assay format for measuring levels of sICAM1 in a sample

[0406] FIG. 29 shows a decision tree analysis of sepsis markers.

[0407] FIG. 30A and FIG. 30B show diagnostic performance of the LR1 markers in terms of distinguishing the infection+SIRS group from the SIRS only group. FIG. 30A is a scatter plot and FIG. 30B is a ROC curve.

[0408] FIG. 31A and FIG. 31B show diagnostic performance of the LR2 markers in terms of distinguishing the infection+SIRS group from the SIRS only group based on patients with SOFA scores of 2-6. FIG. 31A is a scatter plot and FIG. 31B is a ROC curve.

[0409] FIG. 32A and FIG. 32B show diagnostic performance of the LR2 markers in terms of distinguishing the infection+SIRS group from the SIRS only group across all patients, irrespective of SOFA score. FIG. 32A is a scatter plot and FIG. 32B is a ROC curve.

[0410] FIG. 33A and FIG. 33B show diagnostic performance of the LR3 markers in terms of distinguishing high SOFA score from low SOFA score in the infected group to indicate severity (sepsis). FIG. 33A is a scatter plot and FIG. 33B is a ROC curve.

[0411] FIG. 34 is a scatter plot showing that the LR2 algorithm was able to positively identify the infected patients in the validation set (based on day 0 samples).

[0412] FIG. 35 is a timecourse showing that the LR2 algorithm could also be used to discriminate survivors from non-survivors.

EXAMPLES

[0413] The invention will be further understood with reference to the following experimental examples.

Example 1— Improving on CRP for Sepsis Diagnosis

SUMMARY

Background

[0414] Although many new biomarkers for sepsis diagnosis have been investigated, to be useful in clinical practice, and to improve sensitivity and specificity it may be necessary to combine them with traditional markers, particularly C-reactive protein (CRP). This study compared the diagnostic accuracy of CRP used alone and in combination with selected complementary markers for the diagnosis of sepsis.

[0415] Methods

[0416] One hundred and two patients diagnosed with sepsis based on strict clinical criteria including positive blood cultures (51 with Gram-positive, 49 Gram-negative and 2 mixed Gram-positive and Gram-negative microorganisms) were compared to 102 patients with no evidence of sepsis. Serum levels were measured for CRP and procalcitonin (PCT), as well as seven selected potential markers. These comprised: two inflammatory cytokines, interleukin-6 (IL-6) and tumour necrosis factor alpha (TNFα); the vascular marker soluble intercellular adhesion molecule (sICAM-1); a chemokine, C-C motif ligand 23 (CCL23, also known as macrophage inflammatory protein 3, MIP-3); secreted type IIA phospholipase (sPLA2g2a); matrix metalloprotease 8 (MMP8) and the serpin protease inhibitor, a-1 antitrypsin (A1AT). The diagnostic accuracy of each marker alone was evaluated by receiver operator curve (ROC). Combinations of markers giving improved diagnostic performance were identified by logistic regression. Cut-off values and diagnostic algorithms for marker combinations were developed by classification and regression tree analysis.

[0417] Findings

[0418] When markers were measured following confirmation of sepsis by positive blood culture, the accuracy of CRP for sepsis diagnosis was superior to that of the other markers investigated. sPLA2g2a was as sensitive as CRP, but its specificity was lower. Combining levels of sICAM-1 with CRP improved diagnostic accuracy of CRP alone. When levels were measured earlier, at the time the patient presented with symptoms of sepsis, sPLA2g2a displayed diagnostic accuracy equivalent to that of CRP. An increase in the diagnostic accuracy of CRP could be achieved by combining measurement of CRP with sPLA2g2a.

[0419] Interpretation

[0420] Potential exists for improving the diagnostic accuracy of CRP in sepsis by combining measurement of its level with those of sPLA2g2a and sICAM-1. This approach would have value in supporting choice of treatment options prior to confirmation by culture and in culture-negative cases where sepsis is suspected.

INTRODUCTION

[0421] Early and accurate diagnosis of sepsis is essential for initiation of appropriate therapy. C-reactive protein (CRP) and procalcitonin (PCT) are the most widely used markers, but they have limited ability to distinguish between sepsis and systemic inflammatory response due to non-infectious causes. Many other serum biological markers have been evaluated for this purpose, including pro- and anti-inflammatory cytokines, chemokines, cell receptors for microbial toxins and cellular adhesion molecules..sup.1,2 The underlying pathobiology of sepsis is poorly understood and, owing to potential involvement of many organ systems, finding a specific marker that is reproducibly quantifiable in all septic patients has not been achieved. Currently, very few serum-based tests are routinely used in the diagnosis of sepsis. CRP, and to a lesser extent PCT, are used in the diagnosis of sepsis but are without the high levels of sensitivity and specificity that are required to fully/confidently support clinical decisions. When considering a pathology such as sepsis that has heterogeneous clinical presentations, the sensitivity and specificity of CRP and PCT might be improved by combination with other biological markers, reflecting different aspects of the host response to infection..sup.3,4

[0422] In this study we investigated whether combining a number of potential markers with CRP would improve its diagnostic accuracy in patients with blood culture-confirmed systemic infection, using a group of age- and sex-matched hospital outpatients free from infection as controls. The study was designed to select the optimum combination of markers, to establish the cut-off levels for interpretation and re-assess their accuracy for the early diagnosis of sepsis. Serum was collected following patient consent and confirmation of systemic infection by a positive blood culture and, for a subgroup of patients, sera were also analysed when symptoms of sepsis were first observed or when a septic patient presented to our hospital. The markers studied comprised two cytokines: IL6 5 and TNFα;.sup.56 a chemokine (CCL23, also known as MIP3);.sup.7 the soluble intercellular adhesion molecule (sICAM1) reflecting the vascular endothelial activation in infection;.sup.8 the tissue-degrading enzymes, matrix metalloprotease 8 (MMP8) and secreted type IIA phospholipase A2, PLA2g2a;.sup.9,10 and the protease inhibitor, al-antitrypsin (A1AT)..sup.11 The use of each marker in sepsis diagnosis has been previously reviewed by Pierrakos and Vincent..sup.1

[0423] Here, diagnostic accuracy of each marker was determined by area under the receiver operator curve (AUROC), sensitivity and specificity. Potential for improving diagnostic accuracy by combining markers was explored by logistic regression analysis. Effective cut-off levels and their application in diagnostic algorithms were investigated by classification and regression tree analysis. We examined the levels of each marker in serum collected following patient consent and confirmation of sepsis by a positive blood culture. To investigate the value of the markers in detecting sepsis at an early stage, we also measured marker levels in serum taken at intervals of one to eight days (median four days) prior to the time that symptoms of sepsis presented. This was carried out for 44 patients where surplus samples were available from measurement of other clinical parameters.

[0424] Methods

[0425] Patients

[0426] Serum samples were obtained from evaluable adult patients who had been suffering from SIRS..sup.12 All the patients had been acutely unwell for a duration of no more than five days and had positive blood cultures. All the patients' clinical details were reviewed independently by two Consultant Microbiologists (TE and MD) at the University Hospitals Birmingham NHS Foundation Trust and the blood culture isolates were considered to represent significant bacteraemia. Sepsis was defined as SIRS in the presence of a confirmed infection. Criteria for SIRS were temperature <36.6° C. or 38° C. and white cell count <4.0×10.sup.9/L or >12.0×10.sup.9/L. Bacteremia was confirmed in all sepsis patients by positive blood cultures. The causative microorganism was identified by Gram-staining and subculture onto appropriate media followed by routine biochemical identification tests. Serum samples were collected from 98 patients following patient consent and confirmation of sepsis by a positive blood culture. In addition, sera were obtained from 44 patients at various time intervals during the course of their sepsis. These were samples that had been taken for analysis of other blood markers as part of their routine clinical management. Samples were collected on the day of onset of symptoms or admittance to hospital (when the initial blood samples were taken for culture). Controls comprised an age- and sex-matched group of patients attending an ophthalmic outpatient clinic at UHBFT. None had any evidence of sepsis in the preceding 12 weeks and no evidence of an inflammatory condition nor immunosuppression.

[0427] Study Approvals

[0428] Research Ethics committee (NRES West Midlands—Coventry and Warwickshire, REC ref: 12/WM/0251) and R&D Department (The University Hospitals Birmingham NHS Foundation Trust) approvals were obtained prior to commencing the study. Informed written consent was received from all the study patients prior to participation in this study.

[0429] Marker Assays

[0430] Serum levels for IL-6, TNFα, sICAM-1, CCL23 and MMP8 were determined in triplicate using commercial ELISA kits (Duosets, R&D Systems, Abingdon, UK). PCT, PLA2 and A1AT were measured in triplicate by ELISA using assays developed by Mologic Ltd. CRP levels were determined by particle-enhanced immunoturbidimetry (Roche Diagnostics, Burgess Hill, West Sussex, UK) by Clinical Biochemistry at University Hospitals Birmingham NHS Foundation Trust.

[0431] Statistical Analysis

[0432] Descriptive statistics were calculated using Prism vs6 (GraphPad). The D'Agostino-Pearson omnibus normality test was used to check whether marker values followed a Gaussian distribution. The diagnostic accuracy of each marker to distinguish between sepsis and controls was determined as the area under the receiver operator curve (AUROC, sensitivity vs 1-specificity). Optimum cut-off levels for each marker used alone were calculated from the ROC curves at the maximum value of the Youden index (J=sensitivity+specificity−1). Forward stepwise logistic regression was used to investigate the value of combinations of markers in the prediction of sepsis (SPSS vs20.0, IBM). Classification and regression tree analysis was used to construct diagnostic algorithms by sequential application of marker cut-off values (SPSS).

[0433] Results

[0434] The sepsis group comprised 102 patients, age range 20-97 years, mean 59.0 years, M/F 52/48. The sources of sepsis were; intravascular (IV) catheters (32), abdominal and biliary (21), renal and urinary (17), skin and soft tissue (15), respiratory (7), non-IV catheter vascular (5) and other/unknown (5). Causative organisms identified from blood were Gram-negative bacteria (51), Gram-positive bacteria (49) and 2 mixed Gram-positive and Gram-negative bacteria (2). Bacterial species are given in Table 1. The control group comprised 102 individuals, age range 23-88 years, mean 60.6 years, M/F 55/47. Median marker levels in sera taken from 98 patients following patient consent and confirmation of sepsis by a positive blood culture, and from controls are shown in Table 2. Because the marker levels were not normally distributed, the significance of the difference in median marker levels between sepsis and controls was determined by the Mann-Whitney U test. Performance of each marker in the diagnosis of sepsis was summarised as the area under the ROC curve (AUROC). The cut-off levels giving optimum diagnostic accuracy for each marker were determined at the maximum value of the Youden index on the ROC curves, sensitivity and specificity at this cut-off are listed in Table 2. Individually, CRP gave the clearest diagnosis of sepsis, followed by sPLA2g2a and sICAM-1, each marker being more accurate than PCT in terms of AUROC, sensitivity and specificity. IL6 and CCL23 gave poor specificity and sensitivity respectively, whilst the other markers were of very limited diagnostic value. To investigate whether different diagnostic performance would be obtained from samples taken earlier in the septic episode, marker levels were measured in sera at the time of the onset of symptoms of sepsis. These marker levels were compared with the marker levels from the serum taken following patient consent and confirmation of sepsis by a positive blood culture. Only IL6, PLA2g2a, TNFα and CCL23 showed significant differences (Mann Whitney p values <0.0001, 0.0191, 0.0384 and 0.0025 respectively), suggesting that these markers might provide better diagnostic accuracy when measured in samples taken early in the course of sepsis. Diagnostic accuracy of each of the markers was therefore assessed as before, using the earlier time samples for the 44 patients and 102 controls. The results are shown in Table 3. The median time between the early samples (Table 3) and those taken when blood cultures were positive (Table 2) was four days (mean=4.038, SD=1.311, SEM=0.2571), minimum one day (one patient), maximum eight days (one patient), 25% percentile 3.75 days, 75% percentile five days. CRP and PLA2g2a showed the best diagnostic accuracy in terms of sensitivity at the optimum cut-off. Comparing the data in Tables 2 and 3 shows a notable improvement in diagnostic accuracy of IL6 and CCL23 when their levels are measured at the earlier stage of sepsis. This indicates their early production and subsequent reduction during the course of sepsis.

[0435] The value of combining markers with CRP to improve its diagnostic performance was explored by binary logistic regression. Logistic regression models were derived by entering each marker separately with CRP. The predictive performance of each model was then compared with that of CRP alone (Table 4). Using the marker levels from serum following patient consent and confirmation of sepsis by a positive blood culture, only sICAM-1 provided a logistic regression model that improved on the performance of CRP alone in both sensitivity and specificity. Whereas CRP alone correctly identified 92/98 sepsis patients and 99/102 controls, combination of sICAM-1 with CRP in the logistic regression model correctly identified 95/98 sepsis and 100/102 controls. When all markers were submitted to forward stepwise (likelihood ratio) logistic regression a model giving 96% sensitivity and 99% specificity was produced, based on combination of CRP, sICAM-1, IL6 and sPLA2g2a. Using the same approach for marker levels measured at the onset of sepsis symptoms, combination of PLA2 with CRP gave the greatest improvement in performance. Addition of this marker improved the diagnostic performance of CRP from 40/44 sepsis patients and 101/102 controls correctly identified to 43/44 sepsis patients and 101/102 controls. Separate combinations of either PCT, sICAM-1, IL6 or CCL23 with CRP also improved sensitivity from 40/44 sepsis to 41/44 without loss of specificity. When all markers were submitted to forward stepwise (likelihood ratio) logistic regression a model was produced based on combination of CRP, sICAM-1, IL6 and sPLA2g2a.

[0436] Combination of markers with CRP in diagnostic algorithms was also investigated by classification and regression tree analysis (CART). For the marker levels taken following patient consent and confirmation of sepsis by a positive blood culture, where combination of sICAM-1 with CRP improved the sensitivity by logistic regression, the initial application of sICAM-1 at a cut-off level of 302 pg/L correctly identified 68/98 sepsis patients with no false positives (i.e. no controls above this cut-off level). Application of a CRP cut-off level of 20.5 mg/L to the remaining samples (those with sICAM-1 levels below 302 pg/L, comprising 32 sepsis patients and 102 controls) correctly identified a further 29/32 sepsis patients with one false positive, leaving 101/102 controls with marker levels below the cut-off values for both markers. Overall sensitivity for this group of samples was therefore 99% with 99% specificity, compared with 92% and 97% for CRP alone. For the smaller number of samples analysed at the time of onset of sepsis symptoms, where logistic regression showed sPLA2g2a to improve the accuracy of CRP, CART analysis produced an algorithm involving application of CRP at a cut-off level of 16.5 mg/L, followed by sPLA2g2a at a cut-off of 57.6 pg/L to those samples falling below this CRP cut-off. This algorithm correctly identified 42/44 sepsis patients with CRP above the cut-off level and three false positives. Application of the PLA2 cut-off to those samples falling below the CRP cut-off correctly identified the two remaining sepsis patients with no additional false positives. The overall sensitivity was therefore 100% with a specificity of 97%, compared with 95% and 97% respectively with CRP alone.

[0437] When marker levels in patients with sepsis caused by Gram-positive and Gram-negative bacteria were compared, only CRP showed a significant difference in median levels. For the early onset samples, the CRP median was 113 mg/L (range 3-547) for Gram-positive sepsis (30), compared with a median CRP of 46 mg/L (range 4-42) for Gram-negative sepsis (12) (p=0.0005, Mann-Whitney U test). For samples taken following patient consent and confirmation of sepsis by a positive blood culture, the CRP median levels were: 112 mg/L (range 7-550) for Gram-positive sepsis (49), compared with 69 mg/L (range 5-278) for Gram-negative sepsis (48) (p=0.0102). None of the other markers showed a significant difference between marker levels for Gram-positive and Gram-negative sepsis.

TABLE-US-00010 Mixed Gram-negative Gram-positive Gram-negative and -positive infection (n = 51) (n = 49) (n = 2) Staphylococcus aureus (25) Escherichia coli (28) E. coli and E. faecium (1) ‘Viridans’ streptococci (5) Pseudomonas aeruginosa (6) P. aeruginosa and ‘viridans’ Coagulase-negative staphylococci (4) Klebsiella oxytoca (3) streptococci (1) Streptococcus pneumoniae (4) Stenotrophamonas maltophilia (3) Streptococcus pyogenes (4) Enterobacter aerogenes (1) Enterococcus faecium (1) Acinetobacter ursingii (1) Enterococcus gallinarum (1) Klebsiella pneumoniae (1) Streptococcus agalactiae (2) Serratia marcescens (1) Group G Streptococci (2) Bacillus fragilis (1) Streptococcus constellatus (1) Salmonella sp. (1) Cardiobacterium hominis (1) E. coli and K. oxytoca (1) Coagulase-negative staphylococci E. coli and K. pneumoniae (1) and Enterococcus faecalis (1) K. pneumoniae and Enterobacter cloacae (1)

TABLE-US-00011 TABLE 2 Marker levels in sera from patients with sepsis (n = 98) and healthy controls (n = 102). Markers were measured in sera taken following patient consent and confirmation of sepsis by a positive blood culture. sensitivity, 25% 75% specificity Marker group min percentile median percentile max p AUROC (cut-off) CRP control 0 0 0 3 29 <0.0001 .992 94%, 97% mg/L sepsis 5 41.5 96.5 162.3 550 (18 mg/L) PCT control 0 0 0 0 2 <0.0001 .810 72%, 86% μg/L sepsis 0 0 2 6.5 75 (0.6 ng/L) sICAM-1 μg/L control 0 129.5 174.5 215.5 299 <0.0001 .883 85%, 87% sepsis 0 265 374.5 545 1039 (245 μg/L) IL6 control 0 0 3 21 982 <0.0001 .772 84%, 67% ng/L sepsis 0 15.25 39.5 124 1779 (11 ng/L) sPLA2g2a control 0 8 14 22 80 <0.0001 .959 94%, 88% μg/L sepsis 8 47 102.5 266.8 806 (32 μg/L) A1AT control 0 782 1360 2152 5831 0.0016 .628 66%, 49% μg/L sepsis 0 913 2028 3927 7989 (1314 mg/L) TNF control 0 0 0 27 2857 0.0127 .590 50%, 69% ng/L sepsis 0 0 0.5 154 2764 (3.2 ng/L) CCL23 control 0 75 294 491 2969 <0.0001 .720 56%, 81% ng/L sepsis 0 282.5 716 1374 7684 (570 ng/L) MMP8 control 0 6.25 11 18.75 49 0.0031 .623 63%, 58% μg/L sepsis 0 6 17 37.5 43 (12.5 μg/L) p = probability, Mann-Whitney U test. AUROC = area under the receiver operator curve (sensitivity vs 1-specificity). Sensitivity and specificity determined at the optimum cut-off (maximum Youden index on ROC curve).

TABLE-US-00012 TABLE 3 Marker levels and diagnostic accuracy in sera from patients with sepsis (n = 44) taken at the time blood was taken for culture, i.e. the time of symptom onset vs healthy controls (n = 102) sensitivity, 25% 75% specificity Marker group min percentile median percentile max p AUROC (cut-off) CRP control 0 0 0 3 29 <0.0001 .988 95%, 98% mg/L sepsis 3 60 104 200 547 (17.5 mg/L) PCT control 0 0 0 0 2 <0.0001 .801 69%, 81% μg/L sepsis 0 0 2 6.5 80 (0.5 ng/L) sICAM-1 μg/L control 0 129.5 174.5 215.5 299 <0.0001 .922 73%, 96% sepsis 146 253.5 408 581 1059 (284 μg/L) IL6 control 0 0 3 21 982 <0.0001 .928 98%, 78% ng/L sepsis 0 113 395 1200 1664 (28.5 ng/L) sPLA2g2a control 0 8 14 22 80 <0.0001 .994 96%, 99% μg/L sepsis 37 63 145 523.5 899 (54.5 μg/L) A1AT control 0 782 1360 2152 5831 <0.0001 .768 76%, 70% μg/L sepsis 0 1764 2788 4446 10048 (1883 mg/L) TNFα control 0 0 0 27 2857 <0.0001 .692 60%, 79% ng/L sepsis 0 0 64 263 2014 (49 ng/L) CCL23 control 0 75 294 491 2969 <0.0001 .911 89%, 81% ng/L Sepsis 77 978 1833 3422 6271 (599 ng/L) MMP8 control 0 6.25 11 18.75 49 0.0011 .686 50%, 95% μg/L sepsis 0 8.5 31 40 45 (27.5 μg/L) p = probability, Mann-Whitney U test. AUROC = area under the receiver operator curve (sensitivity vs 1-specificity). Sensitivity and specificity determined at the optimum cut-off (maximum Youden index on ROC curve).

TABLE-US-00013 TABLE 4 Sensitivity and specificity of CRP used alone or in combination with other markers in serum samples taken following patient consent and confirmation of sepsis by a positive blood culture and at the time of symptom onset. Marker levels at time Marker levels at time blood cultures were positive of symptom onset (n = 98 sepsis, (n = 44 sepsis, 102 controls) 102 controls) Marker Sensiti- Specifi- Sensiti- Specifi- combinations vity % city % vity % city % CRP alone 94 97 91 99 CRP + PCT 95 97 95 99 CRP + sICAM-1 97 98 93 99 CRP + IL6 94 97 93 98 CRP + sPLA2g2a 96 97 98 99 CRP + A1AT 96 97 91 99 CRP + TNF 94 97 91 99 CRP + CCL23 94 97 93 98 CRP + MMP8 94 97 91 99

DISCUSSION

[0438] When markers were measured at the time following patient consent and confirmation of sepsis by a positive blood culture, the accuracy of CRP for sepsis diagnosis was superior to that of the other markers investigated, including PCT. sPLA2g2a was as sensitive as CRP, but its specificity was lower. When levels were measured earlier, at the time of onset of symptoms/admittance of patient to hospital, sPLA2g2a displayed diagnostic accuracy equivalent to that of CRP. Using both logistic regression and CART, we showed that an increase in the diagnostic accuracy of CRP could be achieved by combining measurement of CRP with sPLA2g2a. When measured at the time of onset of symptoms/admittance of patient to hospital, combination of PLA2 and CRP levels in a logistic regression model gave 100% sensitivity and 100% specificity. Applying CART analysis to the combination we generated a diagnostic algorithm involving the application of CRP at a cut-off of 16.5 mg/L followed by sPLA2g2a at a cut-off of 57.6 pg/L. This gave an accuracy of 100% sensitivity and 97% specificity. If these two markers were measured at the time blood is taken for culture, their interpretation, either by the logistic regression model or the diagnostic algorithm, could provide support for patient management by predicting the blood culture result as either positive or negative. Most importantly, prediction of a negative blood culture result might support a decision to withhold antibiotic treatment and thereby avoid unnecessary therapy reducing the risk of emerging multi antimicrobial resistant microorganisms. Other markers identified for potential combination with CRP are PCT and sICAM-1, both of which improved the sensitivity of CRP without compromising its selectivity in the early samples. For serum samples taken following patient consent and confirmation of sepsis by a positive blood culture, sICAM-1 gave a greater improvement in the accuracy of CRP than either sPLA2g2a or PCT when used in combination with CRP. If a single additional marker were to be chosen for combination with CRP, sICAM-1 would be recommended for application to samples taken over the time course of sepsis before confirmation by positive blood culture. In practice, many blood cultures taken from patients with sepsis are negative due to prior use of antibiotics or inadequate blood volume sampling. In such cases, positive prediction of infection from a marker combination would also support initiation or continuation of antibiotic therapy. Similarly, the use of such combinations of markers may identify when positive blood cultures represent contamination rather than true sepsis. This study points the way to those markers that have greatest potential for combination with CRP. Clearly, the combination of sPLA2g2a, sICAM-1 or PCT as complementary markers with CRP must be rigorously tested in multicentre studies involving other patient groups, including sepsis of fungal aetiology and non-infectious conditions causing elevated inflammatory and vascular markers. Assays for sPLA2g2a and sICAM-1 could easily be adapted for use in a routine clinical laboratory.

REFERENCES

[0439] 1. Pierrakos C, Vincent J-L. Sepsis biomarkers: a review. Crit Care 2010; 14: R15. [0440] 2. Tsalik E L, Jaggers L B, Glickman S W, et al. Discriminatory value of inflammatory biomarkers for suspected sepsis. J Emerg Med 2012; 43: 97-106. [0441] 3. Gaini S, Koldjaer O G, Pedersen C, Pedersen SS (2006). Procalcitonin, lipopolysaccharide-binding protein, interleukin-6 and C-reactive protein in community-acquired infections and sepsis: a prospective study. Crit Care 2006; 10: R53. [0442] 4. Xing K, Murthy S, Liles W C, Singh J M. Clinical utility of biomarkers of endothelial activation in sepsis-a systematic review. Critical Care 2012; 16: R7. [0443] 5. Uusitalo-Seppala R, Koskinen P, Leino A, Peuravuori H, Vahlberg T, Rintala E M. Early detection of severe sepsis in the emergency room: diagnostic value of plasma C-reactive protein, procalcitonin, and interleukin-6. Scand J Infect Dis 2011; 43: 883-90. [0444] 6. Mera S, Tatulescu D, Cismaru C, et al. (2011). Multiplex cytokine profiling in patients with sepsis. Acta Path Microbiol Immunol Scand 2011; 119: 155-63. [0445] 7. Ginde A A, Blatchford P J, Trzeciak S, et al. (2014). Age-related differences in biomarkers of acute inflammation during hospitalization for sepsis. Shock 2014; 42: 99-107. [0446] 8. Kung C T, Hsiao S Y, Su C M, et al. Serum adhesion molecules as predictors of bacteremia in adult severe sepsis patients at the emergency department. Clinica Chimica Acta. 2013; 421: 116-20. [0447] 9. Yazdan-Ashoori P, Liaw P, ToItl L, et al. Elevated plasma matrix metalloproteinases and their tissue inhibitors in patients with severe sepsis. J Crit Care 2011; 26:556-65. [0448] 10. Rintala E M, Aittoniemi J, Laine S, Nevalainen T J, Nikoskelainen J. Early identification of bacteremia by biochemical markers of systemic inflammation. Scand J Clin Lab Invest 2001; 61: 523-30. [0449] 11. Bossink A W, Groeneveld A B, Thijs L G. Prediction of microbial infection and mortality in medical patients with fever: plasma procalcitonin, neutrophilic elastase-alpha1-antitrypsin, and lactoferrin compared with clinical variables. Clinical Infectious Diseases. 1999; 29: 398-407. [0450] 12. Bone R C, Balk R A, Cerra F B, et al. Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. The ACCP/SCCM Consensus Conference Committee. American College of Chest Physicians/Society of Critical Care Medicine. 1992. Chest 2009; 136(5suppl): e28.

Example 2—UHB Study and Results

[0451] Clinical samples made available from a clinical study performed by University Hospital Birmingham (UHB) were tested. They were categorised according to the time of sample:

[0452] A=admission/onset of symptoms

[0453] B=time blood culture (BC) taken *values in bold represent (Number of A/B samples=44)

[0454] C=time BC positive=32 samples

[0455] D=time of consent (large volume)=88 samples

[0456] E=last sample prior to discharge=18 samples

[0457] Controls=102 samples

[0458] They were tested using the following assays:

TABLE-US-00014 Marker Assay Type Format Supplier CRP ELISA Sandwich R&D Systems CRP Lateral flow Sandwich Mologic sICAM1 ELISA Sandwich R&D Systems sICAM1 Lateral flow Sandwich Mologic CRP/sICAM1 Duplex Lateral flow Sandwich Mologic IL-6 ELISA Sandwich R&D Systems IL-6 Lateral flow Sandwich Mologic PLA2 ELISA Sandwich Mologic PLA2 Lateral flow Sandwich Mologic IL-6/PLA2 Duplex Lateral flow Sandwich Mologic A1AT ELISA Sandwich Mologic A1AT Lateral flow Sandwich Mologic TNFα ELISA Sandwich R&D Systems TNFα Lateral flow Sandwich Mologic CCL23 ELISA Sandwich R&D Systems Ac-PGPv3 ELISA Competition Mologic MMP activity Substrate Fluorescent Mologic Creatinine Substrate Colourmetric R&D Systems Desmosine ELISA Competition Mologic MMP8 ELISA Sandwich R&D Systems Procalcitonin ELISA Sandwich Abcam Procalcitonin ELISA Sandwich Mologic

[0459] Patient demographics were as follows:

TABLE-US-00015 All sepsis Gram Negative Gram Positive Control All Male Female All Male Female All Male Female All Male Female (n = 105) (n = 59) (n = 46) (n = 53) (n = 29) (n = 24) (n = 50) (n = 103) (n = 103) (n = 103) (n = 28) (n = 22) Median age 60 64 55.5 64 69 60.5 54 54.5 46.5 63 63 65 Min Age 20 28 20 20 20 33 20 28 20 23 29 23 Max Age 97 97 80 88 88 80 97 97 76 88 88 86

[0460] The nature of the infections is shown in more detail in FIG. 1.

[0461] The origins of the infections are summarised in FIG. 2A for all sepsis patients and in FIG. 2B for gram negative infections and FIG. 2C for gram positive infections respectively.

[0462] For each assay logistic regression analysis was performed on the D samples versus controls. ROC curves for each of the assays performed are shown in FIG. 3. The results are also summarised in the table below.

TABLE-US-00016 Area Under the Curve Test Result Variable(s) Area CRPLF .993 CRPduplex .964 PLA2G2AELISA .950 sICAMELISA .935 CRPELISA .900 sICAMLF .897 PLA2G2ALF .827 PLA2G2Aduplex .819 ProcalcitoninELISA .791 sICAMduplex .774 IL6ELISA .731 CCL23ELISA .681 A1ATELISA .657 MMP8ELISA .627 A1ATLF .613 MMPsubstrate .609 TNFaLF .546 TNFaELISA .541 AcPGPELISA .530 IL6duplex .493 IL6LF .482 desmosineELISA .471 creatinine .354

[0463] As can be seen, some markers displayed good performance when used alone.

[0464] Applying a logistic regression analysis of D samples against control samples, the following markers were selected for use in combination: CRP, sICAM and TNFalpha. The classification table is provided below:

TABLE-US-00017 Classification Table.sup.a Predicted VAR00001 Percentage Observed Control Sepsis D Correct Step 1 VAR00001 Control 89 2 97.8 Sepsis D 5 77 93.9 Overall Percentage 96.0 Step 2 VAR00001 Control 89 2 97.8 Sepsis D 4 78 95.1 Overall Percentage 96.5 Step 3 VAR00001 Control 90 1 98.9 Sepsis D 4 78 95.1 Overall Percentage 97.1 aThe cut value is .500

[0465] As can be seen from the scatter plot in FIG. 4 and the ROC plot in FIG. 5, this combination of markers was able to sensitively and specifically detect sepsis (as compared to control samples).

[0466] The logistic regression model was also applied to A samples verus controls and the scatter and boxplot results using this model are shown in FIGS. 6A and 6B respectively.

[0467] As can be seen from the ROC plot in FIG. 7, this combination of markers was also able to sensitively and specifically detect sepsis in the admission/onset of symptoms samples (as compared to control samples).

Example 3—DSTL Study and Results

[0468] 910 samples were received and tested on 10 assays at Mologic. The samples were collected from patients admitted for elective surgery and daily samples were collected for up to 7 days post-surgery. The patients were stratified into 3 different groups: [0469] 1. Control group (n=70)— those that recovered with no SIRS symptoms [0470] 2. SIRS group (n=66)— those that presented with SIRS symptoms within the 7 days [0471] 3. Sepsis group (n=70)— those that developed sepsis within the 7 days

[0472] Some samples were missing where patients refused consent or the research nurses failed to get good venous access, and as for the sepsis group, focus was on days −1 , −2 and −3 pre-sepsis diagnosis.

[0473] Differentiation Between SIRS and Sepsis

[0474] Samples were grouped according to DSTL defined SIRS diagnosis based on heart rate, respiratory rate, WCC and temperature and other parameters. The first step was to analyse all samples from patients with sepsis (with SIRS) (n=117) and SIRS (n=173) to come up with the best combination to discriminate between the two groups. The markers identified were sICAM-1 (ELISA and LF), CCL23, A1AT, CRP, IL-6 and TNF-α.

[0475] Logistic regression analysis was applied to generate a model for stratifying the patients, as shown in the table below:

TABLE-US-00018 sICAM ELISA/LF, CCL23, A1AT, CRP, IL-6 and TNFα Predicted VAR00001 Percentage Observed SIRS SEPSIS Correct Step 1 VAR00001 SIRS 147 26 85.0 SEPSIS 26 91 71.8 Overall Percentage 82.1 a. The cut value is .500

[0476] Using the logistic regression model scatter plots and ROC curves were created to show comparisons with all groups. Results for SIRS versus sepsis are shown in FIG. 8, for SIRS v controls in FIG. 9 and for sepsis versus controls in FIG. 10. As can be seen, the model was very effective in distinguishing sepsis from SIRS and also sepsis from controls.

[0477] A variety of statistical methods was used to test the significance of any differences between the three groups. The logistic regression derived predictive algorithm developed in this report to distinguish individuals with sepsis compared to SIRS or no SIRS symptoms identified seven biomarkers, these being CRP, A1AT, IL-6, sICAM-1 (ELISA and LF) and TNFα. With these biomarker combinations 85% sensitivity for SIRS and 71.8% for sepsis was achieved with a p value of <0.0001 (Mann-Whitney t test) and an AUC value of 0.8754. Overall, these results provide a very strong basis for the construction of meaningful algorithms and strongly support the feasibility of accurate patient monitoring/prediction of sepsis through the assay of selected biomarkers.

Example 4—Rat Study and Results

[0478] Rats were housed in a metabolic monitoring cage for acclimatisation purposes. They were anaesthetised for 30 minutes for a transthoracic heart scan, and vascular lines were inserted. This was done by making an incision in the centre of the neck and inserting one line into the right jugular vein and one into the left carotid artery. Sepsis animals were given an injection of fecal slurry into the peritoneal cavity, but naïve animals were not. Blood was taken from the animals from the inserted lines.

[0479] There were eight rats in the study, four Naïve and four “Sepsis” following time-points: 0, 3, 6, 12 and 24 hours.

[0480] The following assays were run:

TABLE-US-00019 Marker Assay Type Dilution IL-6 Quantikine 1 in 2 CRP Duoset 1 in 40k sICAM Duoset 1 in 100, 1in 200 PLA myBiosource Neat TIMP1 Duoset 1 in 20, 1 in 1000 MMP8 In-house Neat a1AT myBiosource 1 in 50k

[0481] These experiments enabled time courses to be analysed, for example to detect early markers of sepsis which may be of value in predictive methods. Marker time courses are shown for selected markers in FIGS. 11A-D. Marker time course features: [0482] IL6 and PLA2 peak at 6 h, rapid decline to 24 h [0483] TIMP1 peaks at 6h, slow decline to 24 h [0484] sICAM-1 peaks 12h, slow decline to 24 h [0485] CRP peaks >24 h

Example 5—Sepsis Prototype Point of Care Test

[0486] The prototype device is shown schematically in FIG. 12 and consists of a sample pad to which 80 μL of a diluted or neat serum sample is added, a conjugate pad containing an optimised mixture of gold colloid-conjugated antibodies for both the test line(s) and control line. The sample containing unknown concentrations of analyte bind to the respective antibodies and travels up the nitrocellulose membrane towards the capture lines. The test line consists of a second antibody to the target analyte and if present in the sample will form a complex resulting in a visible test line. The control line consists of an antigen to the control line antibody conjugated to gold i.e. BSA-biotin to anti-biotin gold, this will tell the user that the test has run successfully. All lines are quantifiable within 10 minutes with a lateral flow device reader such as the Cube (Optricon, Germany).

[0487] A personalised testing strategy is set out schematically in FIG. 13.

Example 6—sRAGE in Sepsis

[0488] Sixteen samples collected from sepsis patients and 16 samples collected from control subjects were analysed to determine levels of soluble receptor for advanced glycation end products (sRAGE). Higher levels of sRAGE was found in sepsis patients with a median of 1.425 (IQR 1.154-2.337) compared to the control subjects with a median of 1.028 (IQR 0.7678-1.212). A significant Mann-Whitney test of 0.0062 was obtained where a value <0.5 is deemed significant. Box plot results are shown in FIG. 14

[0489] ELISA Method:

[0490] Reagents: Disposable 96-well polystyrene plates were obtained from Costar (9018 flat bottomed). Human sRAGE was supplied by Novoprotein (Cat No, C423). Capture antibody sheep anti-sRAGE was obtained from Orygen antibodies, (Cat No. SA056). Detection antibody Rabbit anti-sRAGE (Cat No, RA040, Orygen antibodies) conjugated to alkaline phosphatase (Innova bioscience 702-0005). PNPP solution was obtained from Biopanda (Cat No. PNPP-001). Sample diluent prepared at Mologic consisted of PBS, pH6.9, supplemented with 1% (v/v) Tween 20) 1% BSA. Wash buffer prepared at Mologic consisted of 50 mM tris buffered saline pH8, supplemented with 0.1% (v/v) Tween 20.

[0491] Pilot production process: Microtitre plates were coated overnight with 100 μL of sheep anti-sRAGE SA056 (1 pg/mL in PBS) per well. Plates were washed 3 times with wash buffer between the blocking step and each of the following incubation steps. Each microtitre well was blocked with 120 μl of sample diluent for 1 h at room temperature, to minimise non-specific binding.

[0492] Basic assay process: sRAGE was diluted in the sample diluent to give concentrations between 5 ng/mL and 0.078 ng/mL to generate the dose—response curve. Samples were added neat. 100 μL Standards and/or samples were added to microtitre wells and incubated for 1 h at room temperature with gentle agitation. Alk-phos conjugated a Rabbit anti-sRAGE RA040 was diluted 1 in 6000 in sample diluent and 100 μL subsequently added to microtitre wells and incubated for 1 hours at room temperature with gentle agitation. 100 μL of pNPP solution was added and incubated for a further 20 minutes. The absorbance was measured at 405 nm using a BMG omega plate reader.

Example 7—Decision Tree Analysis of Sepsis Markers

[0493] Sera from patients suffering from post-surgical sepsis (n=211) and post-surgical SIRS (n=288) was analysed by decision tree analysis with a CRT growing method. The combination of markers is CRP, IL-6 and sICAM1.

[0494] This decision tree, shown in FIG. 29 has four terminal nodes with SIRS defined as either CRP<97.582 mg/L or as IL-6<133.910 pg/mL when CRP>97.582 μg/mL and sICAM1 >872.650 ng/mL. Sepsis is defined as sICAM1<872.650 ng/mL when CRP is >97.582 or IL-6 >133.910 pg/mL when sICAM1 >872.650 ng/mL and CRP >97.582 pg/mL. This decision tree accurately identifies 80.2% of SIRS patients and 72.0% of Sepsis patients, as shown in the table below:

TABLE-US-00020 Predicted SIRS Sepsis % correct observed SIRS 231 57 80.2 Sepsis 59 152 72.0

Example 8—Diagnostic Performance Using Sofa Scores to Define Sepsis

[0495] This study was designed to recruit patients who had undergone elective major surgery; these individuals are at risk of developing post-surgical infections and sepsis.

[0496] One serum sample was taken pre-surgery as well as 7 days following surgery. The day of onset of SIRS symptoms is designated day 0, preceding days termed day −1, day −2 etc. and days post symptoms of SIRS termed day 1, day 2 etc.

[0497] 3 cohorts were recruited to the study: [0498] i) Individuals that developed 2 or more SIRS symptoms+confirmed or suspected infection [0499] ii) Individuals that developed 2 or more SIRS symptoms with no evidence of infection [0500] iii) Individuals that did not develop any SIRS symptoms

[0501] Our analysis here is limited to only include groups (i) an (ii) and limited to the first day of SIRS symptoms. [0502] The markers panel that were tested on these samples: [0503] CRP, acute phase response [0504] IL6 and TNFα, inflammatory cytokines [0505] CCL23 (MIP3), a chemokine [0506] sICAM-1, vascular endothelial activation* [0507] secreted type IIA phospholipase A2 (sPLA2g2a), pro-inflammatory enzyme [0508] acetylated-PGP and desmosine, products of collagen and elastase respectively [0509] al-antitrypsin (A1AT), protease inhibitor

[0510] *sICAM-1 was measured by ELISA and LF. Both these methods of analysis do not correlate to a high degree, suggesting that they are recognising different (albeit related) targets. These immunoassays contain different antibody pairs and it is believed that the LF assay is detecting different forms of ICAM including ICAM-1 and VCAM-1.

[0511] The patients were differentiated into 2 groups based on having sepsis or not having sepsis (day one presentation of symptoms): [0512] 1. Infection+SIRS (Infection+2 SIRS criteria (RR, WBC, HR, Temp)) [0513] 2. SIRS: (2+SIRS criteria (RR, WBC, HR, Temp))

[0514] The groups were further defined according to SOFA scores: those based on having SOFA 2-6 were grouped: [0515] 1. Infection+SIRS group: Number of patients n=23 [0516] 2. SIRS group: Number of patients n=25

[0517] The median SOFA score for both groups was 4.

[0518] SOFA Scores for all Patients:

TABLE-US-00021 INFECTION + SOFA SIRS SIRS Scores N = 52 N = 44 0 22 15 1 5 6 2 5 4 3 5 6 4 6 3 5 4 1 6 2 3 8 0 2 9 1 1 10 0 1 13 1 0 14 0 1 15 1 1

[0519] Diagnostic Performance with Logistic Regression (LR) Models

[0520] Diagnostic performance was initially measured in relation to distinguish the infection+SIRS group from the SIRS only group.

[0521] A first LR model, LR1, was generated using 10 markers: Desmosine, TNF, IL-6, AcPGP, PLA2g2A, CCL23, A1AT, sICAM1 (ELISA), sICAM1 (LF), CRP. LR1 gave a Sensitivity of 90.5% and specificity of 88.0% as shown in the table below and in FIGS. 30A and 30B.

TABLE-US-00022 Classification Table - LR1 Predicted VAR00001 Infection + Percentage Observed SIRS SIRS Correct Step 1 VAR00001 SIRS 22 3 88.0 Infection + 2 19 90.5 SIRS Overall Percentage 89.1

[0522] A second model, LR2, was generated using a backwards conditional Logistic regression function. Five markers were selected: CCL23, A1AT, CRP, sICAMLF and sICAM ELISA. LR2 gave a sensitivity of 85.7% and specificity of 88.0% as shown in the table below and in FIGS. 31A and 31B.

TABLE-US-00023 Classification Table - LR2 Predicted VAR00001 Infection + Percentage Observed SIRS SIRS Correct Step 5 VAR00001 SIRS 22 3 88.0 Infection + 3 18 85.7 SIRS Overall Percentage 87.0

[0523] LR2 was also shown to reliably detect SIRS+infection across all patients, irrespective of SOFA score in sensitive and specific fashion. Results are presented in FIGS. 32A and 32B.

[0524] Diagnosing Severity

[0525] Diagnostic performance was further measured in relation to distinguishing the infection group with SOFA scores of at least 2 from the infection group with SOFA score less than 2.

[0526] When tested, LR2 was not able to differentiate SOFA scores of at least 2 from SOFA scores less than 2 in either group (data not shown). Accordingly, a third model, LR3, was generated. Seven markers were selected: CCL23, A1AT, sICAM, desmosine, TNF alpha, IL-6 and PLA2g2A. LR3 was shown to reliably distinguish high SOFA score from low SOFA score in the infected group to indicate severity (sepsis). Results are presented in FIGS. 33A and 33B.

Example 9—Validation Study

[0527] The validation study was a prospective, observational cohort study of critically ill adult patients admitted to the ICU. Following study approval from an ethics committee, all patients were screened on a daily basis to assess those meeting the following inclusion and exclusion criteria:

[0528] Inclusion criteria [0529] Multi-organ failure (at least 2 organ systems involved) [0530] Initial SOFA score >3 [0531] Predicted length of ICU stay >3 days

[0532] Following consent, patients were enrolled into the study and data were recorded on a daily basis for the first week and weekly thereafter. Blood sampling was undertaken on the day of admission (Day 0) and, subsequently, on Days 1, 2, 3, 5, 7 and weekly thereafter.

[0533] Patient Recruitment

[0534] A total of 556 patients were screened for enrolment. Of 177 eligible for inclusion, 51 consented to taking part in the study. Of these, 24 patients with either fecal peritonitis or community acquired pneumonia had samples analysed by Mologic.

[0535] Results

[0536] As shown in FIG. 34, the LR2 algorithm was able to positively identify the infected patients in the validation set (based on day 0 samples).

[0537] It was also shown that the LR2 algorithm could also be used to discriminate survivors from non-survivors, see FIG. 35.

[0538] The present invention is not to be limited in scope by the specific embodiments described herein. Indeed, various modifications of the invention in addition to those described herein will become apparent to those skilled in the art from the foregoing description and accompanying figures. Such modifications are intended to fall within the scope of the appended claims. Moreover, all aspects and embodiments of the invention described herein are considered to be broadly applicable and combinable with any and all other consistent embodiments, including those taken from other aspects of the invention (including in isolation) as appropriate. Various publications are cited herein, the disclosures of which are incorporated by reference in their entireties.