Clinical diagnostic accuracy of rapid detection of healthcare-associated bloodstream infection in critical care using multi-pathogen real-time polymerase chain reaction (RT-PCR) technology

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/13/16. The contractual start date was in May 2010. The draft report began editorial review in November 2013 and was accepted for publication in December 2014. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Geoffrey Warhurst, Paul Dark, Ronan McMullen and Daniel McAuley have received research funding from the Technology Strategy Board to develop point-of-care diagnostics for sepsis. Paul Chadwick has received lecture honoraria from Novartis and funding from Pfizer and AstraZeneca to support local and regional research meetings in Clinical Microbiology.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Warhurst et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Impact of health-care-associated infection

This report evaluates the diagnostic potential of a commercial multipathogen real-time polymerase chain reaction (PCR) platform for detection of health-care-associated bloodstream infection in the high-risk setting of critical care. Health-care-associated infection imposes a significant burden on health-care systems worldwide and is a major cause of morbidity and mortality. Figures from the UK suggest that over 300,000 patients are affected by health-care-associated infection annually, with an annual cost to the NHS in excess of £1B. 1,2 Comparative figures from the USA indicate that the costs associated with the five most common health-care-associated infections are approximately US$10B per annum. 3 Nosocomial bloodstream infections are among the most problematic in terms of excess hospital cost, increased length of stay and attributable mortality rates. 4 Improved infection prevention and control measures linked to organisational targets for reducing meticillin-resistant Staphylococcus aureus (MRSA) bacteraemia and Clostridium difficile infection have contributed to a significant reduction in overall health-care-associated infection prevalence. 5 Nevertheless, health-care-associated infection remains a significant burden, particularly in the critically ill, who, owing to the level of invasive interventional procedures required, in association with their compromised health state, are at high risk of developing severe infections and associated unfavourable health-care outcomes such as organ failure and death. 6

Current diagnostic standards in the critically ill

Consensus definitions of infection in a critical care setting require that diagnosis is confirmed by identification of live micro-organisms (pathogens) by microbiological culture of blood or other samples. 7 However, blood culture (BC), although reflecting the current gold standard, is subject to a significant time delay, requiring 24–48 hours before a positive result is available and at least 5 days to determine that a specimen is culture negative. 8 The significant rates of BC contamination by coagulase-negative staphylococci (CoNS) also add to the difficulties of judging the clinical significance of positive cultures. 9

Sepsis is a clinical syndrome resulting from a patient’s systemic inflammatory response to infection and is a major cause of mortality and increased health-care costs globally. 10,11 Sepsis can be difficult to diagnose and to differentiate from other common non-infectious causes of systemic inflammation. 12 Confirmation of sepsis, therefore, relies on objective diagnostic evidence for infection, including attempts to detect and identify live pathogens from blood samples by culture techniques. 7,11 Early confirmation of sepsis and administration of appropriate antimicrobial therapy is therefore critical to patient outcome but remains a difficult diagnostic problem owing to the time required for microbiological identification of pathogens.

Internationally recognised guidelines for the early management of sepsis advocate that antibiotics are administered within 1 hour of the initial clinical suspicion of infection and current evidence indicates that the correct initial choice of antibiotic saves more lives than virtually any other intervention in patients with severe sepsis. 13,14 In this context, BCs are insufficiently time critical and cannot assist in early management decisions. As a result, antimicrobial therapy should be guided by local pathogen surveillance and normally involves the administration of broad-spectrum, high-potency antibiotics as a ‘safety first’ strategy to cover the spectrum of likely pathogens in patients with suspected sepsis. 11 Although this approach is life-saving in patients with severe sepsis, an inevitable consequence of the temporal separation between initial suspicion and microbiological confirmation is a wasteful and potentially dangerous overuse of antimicrobial chemotherapy. This is particularly relevant given that initial suspicion of infection is based largely on the presence of two or more non-specific clinical signs of systemic inflammation [systemic inflammatory response syndrome (SIRS) criteria – detailed in Chapter 3, Methods, Participants]. SIRS can be induced by a variety of non-infectious aetiologies as well as infection including pancreatitis, major surgery and ischaemia–reperfusion after haemorrhagic shock. 15 As a result, the prevalence of SIRS is high in critical-care settings, with a recent study reporting that 93% of intensive care unit (ICU) patients had two or more SIRS criteria at some stage during their stay. 16 Although the presence of SIRS is associated with a higher risk of progression to severe sepsis or septic shock, the prevalence of confirmed infection is much lower. 17 The overuse of antibiotics to compensate for the diagnostic uncertainty, particularly broad-spectrum antimicrobials, creates a selection advantage for surviving organisms that inevitably facilitates the spread of antibiotic-resistant species. 1 In addition, broad-spectrum therapy disrupts the patient’s commensal flora, leaving them open to superinfections such as C. difficile. 18 The increasing threat from antibiotic resistance and the need for improved antibiotic stewardship within the health-care system has recently been the subject of a major report by the Chief Medical Officer. 2 The economic impact of inappropriate antibiotic use in terms of acquisition costs and higher expenditure resulting from avoidable adverse effects is also significant.

An urgent global challenge has emerged, therefore, to develop and translate techniques that could provide accurate diagnostic information within a short time frame of clinical signs appearing and so allow more informed use of antibiotic therapy at an early stage. This challenge has been championed by the World Health Organization,19 endorsed by national governments2 and highlighted within international consensus-derived sepsis guidelines. 11

Molecular approaches to diagnosis of health-care-associated bloodstream infection

There is a consensus that molecular technologies have the potential to provide rapid detection of pathogens in blood and other clinical samples in a much shorter time frame than is possible with conventional culture. 20–22 The majority of current approaches are based on detection of pathogen deoxyribonucleic acid (DNA) in the sample using nucleic acid amplification techniques – primarily real-time PCR – with results potentially available in 4–6 hours. 21 Real-time PCR allows detection of minute amounts of pathogen DNA by selective amplification of specific regions of bacterial or fungal DNA. Pathogen DNA amplified during the real-time PCR reaction is continuously monitored using either fluorescent dyes that bind non-specifically to double-stranded DNA or with fluorescently labelled probes that bind to specific sequences in the amplified pathogen DNA, the latter approach allowing direct identification of the microbial species present. The amplification of the pathogen signal during real-time PCR is crucially important owing to the low numbers of circulating bacteria [10 colony-forming units (CFUs) per ml] or fungi (1–10 CFU/ml) reported in adult sepsis. 23,24 The use of real-time PCR also facilitates detection of fastidious, difficult-to-culture organisms such as fungi. 25

Two basic approaches have been taken in the design of real-time PCR assays for sepsis pathogens, either (i) broad-range detection of bacterial or fungal DNA with universal primers followed by species identification using post-PCR techniques such as DNA sequencing, electrospray ionisation mass-spectrometry or high resolution melting analysis26–28 or (ii) multiplex assays utilising a panel of species-specific hybridisation probes that provide direct confirmation that a particular species is present. 29 Intuitively, the latter approach would appear to have the greatest clinical utility in terms of directing timely and appropriate antibiotic therapy, assuming that a pathogen panel with adequate coverage can be established. In both designs, assays are generally, although not exclusively, directed at conserved DNA sequences in the 16S, 23S or 16S–23S interspacer regions of the ribosomal ribonucleic acid (rRNA) gene of bacteria or the 18S and 28S rRNA gene regions of fungi. 30–33

The ability of real-time PCR to provide early information on likely antibiotic sensitivities of organisms detected is also of crucial importance in view of the developing crisis in antibiotic resistance. 2 Real-time PCR has the potential to detect the presence of several important antibiotic resistance genes including mecA (MRSA) and van subtypes (vancomycin-resistant enterococci). 34,35 However, reliable detection of resistances such as extended-spectrum beta-lactamases are more challenging given the wide range of genotypes involved and currently full profiling of antibiotic sensitivities can be achieved only by culture. 36

Although the laboratory analytical accuracy of these techniques for the detection of pathogen DNA in blood has been evaluated, there is a reported lack of clinical trial data to define the utility of such tests in patients. 21,36 This has been due in part to the lack of standardised technology platforms that meet accepted regulatory standards for clinical diagnosis.

At the time of application for Health Technology Assessment (HTA) funding for the current trial (early 2008), the only real-time PCR platform with regulatory approval [Conformité Européenne (CE) mark] for simultaneous detection of bacterial and fungal pathogens in suspected bloodstream infection was the LightCycler^® SeptiFast kit (Roche Diagnostics GmbH, Mannheim, Germany). The system uses a multiplex approach, which allows detection of the most common pathogen species causing bloodstream infection in a single blood sample29 without the need for pre-culture and is described in detail below. To date, a further three PCR-based platforms have gained CE mark approval for this purpose, although data on clinical diagnostic performance remain limited. SepsiTest™ CE IVD (Molzym, Bremen, Germany) is a pan-bacterial, pan-fungal real-time PCR assay able to detect DNA from more than 345 bacterial and fungal species with subsequent species identification by amplicon sequencing. 26,37 VYOO^®, introduced by SIRS-Lab, Jena, Germany, gained a CE mark in late 2008 and takes a multiplex approach to detect a panel of sepsis pathogens by electrophoretic separation of species-specific PCR amplicons. 38,39 PLEX-ID™ (Abbott, Wiesbaden, Germany) also uses universal PCR to detect a very broad range of bacterial and fungal pathogens, but utilises electrospray ionisation mass spectrometry for post-PCR species identification. 40

LightCycler SeptiFast real-time polymerase chain reaction

The LightCycler SeptiFast real-time PCR system uses a multiplex approach, allowing detection of a panel of 25 of the most common pathogen species associated with bloodstream infection in a single blood sample. The SeptiFast Master List (SML) panel is shown in Table 1. The assay involves three steps: (i) mechanical lysis and DNA extraction from the blood specimen; (ii) real-time PCR amplification and detection of bacterial/fungal DNA based on the use of species-specific hybridisation probes targeting the internal transcribed region between the 16S and 23S bacterial rRNA gene and between the 18S and 5.8S rRNA region of the fungal genome; and (iii) identification of species based on melting point analysis of real-time PCR products with automated reporting using dedicated software. 29 SeptiFast has been assessed at the laboratory level on clinical isolates and has shown to have good analytical specificity and exclusivity, confirming its analytical validity. 29 When tested against spiked blood samples, SeptiFast showed minimal analytical sensitivity of 3–30 CFU/ml for the pathogen species on the SML and excellent specificity and exclusivity. 29 At the time of submission of the funding application for the current study, information on the clinical diagnostic accuracy of the LightCycler SeptiFast platform for detection of bloodstream infection was limited. An unpublished European Union registration study of 278 critically ill patients with suspected sepsis from Denmark, Germany and Italy was undertaken as part of the CE-marking process.

In addition, three full studies had been published, covering a total of 262 patients and 353 samples across a mix of clinical settings, including general medicine wards, intensive care and emergency departments. 41–43 During the course of the current National Institute for Health Research (NIHR) HTA trial, further studies investigating the diagnostic performance of LightCycler SeptiFast technology have been published, predominantly focusing on suspected sepsis. None of studies to date has specifically examined the performance of the test in the context of suspected health-care-associated bloodstream infection, which is the focus of the current study. A systematic review of the published data on the diagnostic accuracy of LightCycler SeptiFast forms the basis of Chapter 2 of this report.

Overview of the study

It is widely acknowledged that PCR-based nucleic acid amplification techniques have the potential to deliver real improvements in sensitivity and speed of diagnosis of life-threatening infection in the setting of critical care. Although a number of commercial platforms are available for clinical use, there has been no systematic adoption of these technologies in health-care systems, including the NHS. This is primarily due to a paucity of information on clinical utility and a lack of formal HTA. The overall aim of this study is to provide independent clinical accuracy data along with a systematic analysis of current published studies to allow decisions to be made on the utility of the LightCycler SeptiFast real-time PCR for early diagnosis of suspected sepsis-related health-care-associated bloodstream infection in critical care.

The three components of the evidence synthesis are:

1. a systematic review of published diagnostic accuracy studies of the SeptiFast platform compared with BC in the setting of suspected sepsis-associated bloodstream infection (see Chapter 2)

2. the results of an independent and systematic clinical accuracy study of SeptiFast in a clearly defined clinical population of critically ill patients suspected of developing health-care-associated bloodstream infection (see Chapter 3)

3. statistical analysis of the potential impact on diagnostic accuracy of error in the BC gold standard and the inclusion of commonly used circulating inflammatory biomarkers as instrumental variables using latent class modelling (see Chapter 4).

The detailed objectives, methods and results of each of these analyses are reported in the ensuing chapters. The final chapter includes a discussion on the implication of our findings for the real-time PCR-based diagnosis of health-care-associated bloodstream infection including a recommendation on whether or not this technology has sufficient clinical diagnostic accuracy to move forward to efficacy testing during the provision of routine critical care. The priorities for future research are also discussed.

Aim

Chapter 3 of this report presents the results of a large independent, multicentre Phase III clinical diagnostic accuracy study to assess the SeptiFast multipathogen real-time PCR platform in detection and identification of health-care-associated bloodstream infection funded by the NIHR HTA programme. 44 As part of the background to this independent HTA, we report here a systematic review that was designed to focus on the diagnostic test accuracy of SeptiFast real-time PCR for detection and simultaneous identification of pathogens in the blood of patients with suspected sepsis. This systematic review was piloted and registered with PROSPERO, the International Prospective Register of Systematic Reviews, in 2011 (www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42011001289) and the study protocol was published in 2012. 45

Methods

Results

Study inclusion

We identified 2129 citations in total, of which 61 were considered potentially suitable (Figure 1). After full-text review, and having contacted corresponding authors for any extra data in the case of conference abstracts and journal correspondences, 23 studies were excluded as it proved impossible to derive a 2 × 2 table to calculate required diagnostic metrics. In addition, one abstract was removed as the study data were coreported in a full paper and another abstract was replaced by a full paper that was sent to us by the authors. In total, 37 studies were included in the final analysis (26 papers, nine conference abstracts and two correspondences – summarised in Table 2).

FIGURE 1.

Flow diagram of study selection.

TABLE 2 - Summary of studies included in review

Haemato-oncology study cohorts.

First author	Year	Manuscript type	Study country	Sepsis patient setting	Age category	Diagnostic study evidence level	Number of patients recruited with suspected sepsis	Number of paired blood tests/episodes	Bacteraemia prevalence (%)
Raglio52	2006	Abstract	Not stated	Not stated	Not stated	III	74	114	15
Klemm53	2007	Abstract	Germany	Intensive/critical care	Not stated	III	44	56	37
Bingold54	2007	Abstract	Germany	Intensive/critical care	Not stated	III	21	134	15
Vince55	2008	Correspondence	Croatia	In hospital and intensive/critical care	Not stated	III	36	39	21
Mancini42	2008	Paper	Italy	In hospital and unclear if intensive/critical carea	Adults	III	34	103	20
Louie41	2008	Paper	USA	Emergency department, in hospital and intensive/critical care	Adults	III	200	200	21
Lodes56	2008	Abstract	Germany	Intensive/critical care	Not stated	III	137	358	13
Westh57	2009	Paper	Germany	Not stated	Not stated	III	359	558	13
Varani58	2009	Paper	Italy	In hospital and unclear if intensive/critical carea	Adults and children	III	100	130	29
Palomares59	2009	Abstract	Spain	Intensive/critical care	Not stated	III	73	76	13
Lodes60	2009	Paper	Germany	Intensive/critical care	Adults	III	52	258	12
Dierkes61	2009	Paper	Germany	Intensive/critical care	Adults	III	77	99	23
Dark62	2009	Correspondence	UK	Intensive/critical care	Adults	III	50	90	12
Yanagihara63	2010	Paper	Japan	In hospital and emergency department	Not stated	III	212	400	8
Wallet64	2010	Paper	France	Intensive/critical care	Adults	III	72	102	10
Tsalik65	2010	Paper	USA	Emergency department	Adults	III	306	306	22
Sóki66	2010	Abstract	Hungary	In hospital and intensive/critical care	Not stated	III	159	162	24
Regueiro67	2010	Paper	Spain	In hospital and intensive/critical care	Adults	III	72	106	25
Maubon68	2010	Paper	France	In hospital and unclear if intensive/critical carea	Not stated	III	110	110	29
Lehmann69	2010	Paper	Germany	Intensive/critical care	Adults	III	108	453	13
Bloos70	2010	Paper	Germany, France	Intensive/critical care	Adult	III	142	236	17
Berger71	2010	Abstract	Austria	Neonatal unit	Neonates	III	38	38	45
Avolio72	2010	Paper	Italy	Emergency department	Adult	III	144	144	30
Vrioni73	2011	Abstract	Greece	Not stated	Not stated	III	33	33	24
Sitnik74	2011	Abstract	Brazil	Intensive/critical care	Not stated	III	114	114	14
Obara75	2011	Paper	Japan	Emergency department, in hospital and intensive/critical care	Adults	III	54	78	15
Lucignano76	2011	Paper	Italy	In hospital and intensive/critical care	Neonates and children	III	811	1553	10
Josefson77	2011	Paper	Sweden	In hospital	Adults and children	III	1093	1141	12
Hettwer78	2011	Paper	Germany	Emergency department	Adults	III	153	113	45
Bravo79	2011	Paper	Spain	In hospital and intensive/critical care	Adult	III	53	53	47
Tschiedel80	2012	Paper	Germany	In hospital and intensive/critical care	Adults and children	III	75	110	17
Rath81	2012	Paper	Germany	Intensive/critical care	Adults	III	170	225	36
Pasqualini82	2012	Paper	Italy	In hospital and unclear if intensive/critical care	Not stated	III	391	391	15
Mauro83	2012	Paper	Italy	In hospital and unclear if intensive/critical carea	Adult and children	III	79	79	41
Lodes84	2012	Paper	Germany	Intensive/critical care	Adults	III	104	148	20
Guido85	2012	Paper	Italy	In hospital and unclear if intensive/critical carea	Adults	III	166	166	14
Grif86	2012	Paper	Austria	In hospital and intensive/critical care	Not stated	III	61	71	7

Study quality

Our independent external reviewers reported variable study quality and, although studies reported as full papers were associated with the best quality measures, there were important deficiencies overall in study design and reporting (Figure 2). Reviewers agreed that all of the studies selected aimed to compare test results from SeptiFast with BC and that the reported blood sampling for these tests was such that disease progression or regression bias would have been avoided. BC, as the reference standard, appeared to have been applied to patients equally in a way that both partial (work-up) bias and verification bias were probably avoided. However, the reference standard was not always adequately described, including blood sampling methods and the prevalence of defined contamination as a potential source of false-positive culture results. BC results were often difficult to adjudicate by reviewers, with no clear standards of reporting followed. The chance of misclassification when comparing the reference and index tests was therefore thought to be likely, impacting on the derived diagnostic accuracy metrics. In some studies, it was not clear how well the CE-marked protocol for SeptiFast real-time PCR had been followed, including how blood samples had been stored/handled prior to assay delivery. Assay failure rates were rarely reported. There was a universal lack of reported blinding of both reference standard and index tests such that reviewers believed that incorporation bias was highly likely. Overall, reviewers agreed that none of the included studies, as reported, met the standards for the reporting of diagnostic accuracy studies (STARD) criteria in full,87 and in some cases there were significant deficiencies (see Figure 2).

FIGURE 2.

Summary of independent review of quality of included studies.

Estimated summary diagnostic accuracy of SeptiFast

Figure 3 shows the coupled forest plots of sensitivity and specificity for each study and Figure 4 displays the scatterplot in ROC space (plotting sensitivity against 1 – specificity for each study). Summary sensitivity and specificity for SeptiFast compared with BC, estimated using the bivariate model method, were 68% (95% CI 62% to 74%) and 86% (95% CI 84% to 89%), respectively, suggesting that a positive blood test at genus/species level returned by SeptiFast in the setting of a patient with suspected sepsis could have higher diagnostic value (rule in) than a negative test result (rule out) when compared with BC.

FIGURE 3.

Forest plot of included studies. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

FIGURE 4.

Summary ROC with 95% confidence region (dotted) and 95% prediction region (dashed).

Discussion

Our comprehensive systematic review was designed to help understand the estimated combined diagnostic accuracy of SeptiFast real-time PCR in detecting and identifying bacterial and fungal pathogens in the blood of patients with suspected sepsis. We included 37 studies reporting on a total of 8437 SeptiFast tests when compared with BC.

Estimated combined results for sensitivity and specificity suggested that SeptiFast has a higher specificity than sensitivity. For the health-care team, this implies that positive blood tests returned by SeptiFast at pathogen genus/species level could have higher diagnostic utility (as a rapid rule-in test) than negative results (as a rapid rule-out test), at least when compared with BC. The apparent confidence in this statement is greater for specificity than sensitivity (see Figures 3 and 4) because the median event rate of 17% BC positivity for the studies means that the majority of reference tests performed were negative.

The interpretation of these combined diagnostic accuracy results is that negative SeptiFast tests could reasonably be false negatives, explained in part by pathogens detected in BC that were not on the PCR test panel. In addition, despite a higher estimated combined specificity, when compared with sensitivity, the upper CI did not reach 90%. Specificity of SeptiFast real-time PCR, when compared with culture, will be limited by the presence of false-positive results – a positive PCR in the setting of a negative BC. In some studies, a proportion of these false-positive results were reported to be concordant with culture positivity from samples other than blood, suggesting that in some cases of suspected sepsis a positive SeptiFast result may reflect infection not detectable by BC. However, there are no clear explanations for these false-positive and false-negative SeptiFast results because no systematic investigation has been undertaken linking laboratory performance with clinical diagnostic accuracy. In addition, it remains extremely difficult to speculate what implications these diagnostic accuracy results may have for direct clinical care because SeptiFast does not report antibiotic sensitivity data (other than identifying the mecA gene, confirming meticillin resistance following detection of Staphylococcus aureus) and there have been no systematic interventional clinical trials reported to date on the efficacy and effectiveness of supplementing or replacing BC with SeptiFast real-time PCR.

All diagnostic metrics were reported using BC as the reference standard. We accept that there could be limitations to this standard, particularly in the setting of intercurrent antimicrobial therapy. However, we do not know the full extent of this problem, or indeed whether studies have deliberately included or excluded such patients. In addition, SeptiFast real-time PCR was developed to simultaneously detect and identify a panel of the most common pathogens based on reported international BC surveillance data. 29 Therefore, in the absence of an internationally agreed approach to an alternative reference standard for pathogen detection from blood samples at present, we believe that a culture-based reference standard provided the most robust approach for this review and is consistent with methods used in our independent clinical diagnostic accuracy study described in Chapter 3.

Our review identified diagnostic accuracy studies performed within routine clinical care. Clinical diagnostic accuracy studies, in general, are challenging to perform well and often fall short in terms of study quality. Our independent reviewers found a similar trend for these Phase III clinical diagnostic accuracy studies reporting on SeptiFast real-time PCR compared with BC. When assessing the quality of study design and reporting, significant deficiencies were discovered. For both papers and abstracts, the application of the reference and index tests were the only elements that were reported consistently, raising significant concern about the possible effects of numerous sources of inherent study bias on the derived summary estimates of SeptiFast test performance. Indeed, the 95% prediction region in ROC space in Figure 4 shows considerable uncertainty about the likely true sensitivity and specificity of a future study.

During the preparation of this review, another systematic review has reported on the diagnostic accuracy performance of SeptiFast real-time PCR. Chang and colleagues used a basic keyword search strategy for journal papers only, risking publication bias, and they included a number of studies that were judged not to warrant inclusion in our own review on the basis of our inclusion criterion of ‘suspected sepsis’. 88 It appears that Chang reported pooled estimates of sensitivity and specificity when comparing SeptiFast real-time PCR with BC results as composite events, not at a genus/species level – a key feature of our own review – which may have contributed to the improved diagnostic accuracy metrics reported in Chang’s review. 88 However, we would emphasise, based on our comprehensive systematic review reported here, that clinical diagnostic studies in this field vary in quality with the real risk of biases that could impact seriously on the reported estimates of summary diagnostic metrics. Despite the considerable international effort in determining the likely diagnostic accuracy of SeptiFast real-time PCR in the setting of suspected sepsis, we are not confident in the current body of evidence because of the weaknesses in study design and reporting outlined in our systematic review. In particular, we do not agree with Chang and colleagues, who state that:

in the presence of a positive SeptiFast result in a patient with suspected bacterial or fungal sepsis, a clinician can confidently diagnose bacteremia or fungemia and begin appropriate antimicrobial therapy, while forgoing unnecessary additional diagnostic testing. 88

Furthermore, we do not agree that returning a negative SeptiFast result, even within a low-prevalence population, ‘may justify withholding antibiotics’. 88 Our views, evidenced by our systematic review presented here, supports current international guidelines on diagnosis and treatment of sepsis which state that there is limited clinical experience with non-culture-based diagnostic methods, such as real-time PCR, and that high-quality clinical studies are needed before any firm recommendations can be made about their potential utility. 11

We recommend that future clinical studies incorporating SeptiFast should include well-designed and -reported clinical diagnostic accuracy elements measured against all of the features of the STARD criteria. 87 Based on the evidence reviewed here, we are concerned that clinical decisions about treatment interventions/adjustments (notably antimicrobial chemotherapy) based on SeptiFast real-time PCR results, potentially delivered within hours of the suspicion of sepsis, could expose patients to risk because sepsis is associated with high mortality and rapid appropriate antimicrobial choices are crucial for survival. 11

Aim

The primary aims of this Phase III clinical diagnostic accuracy study, designed to meet all the features of the STARD criteria,87 were to (i) determine the accuracy of a multipathogen real-time PCR technology (SeptiFast) for detection and identification of sepsis-related health-care-associated bloodstream infection in adult critical-care patients against the current service standard of microbiological BC and (ii) assess further the preliminary evidence that detection of pathogen DNA in the bloodstream using SeptiFast real-time PCR may have value in detecting the presence of culture-proven infection elsewhere in the body.

Methods

The detailed protocol for this study was approved by the Trial Steering Committee and published in advance by NIHR HTA (www.nets.nihr.ac.uk/projects/hta/081316 and in open access). 44

Results

Recruitment

A total of 1006 new consecutive episodes of suspected health-care-associated bloodstream infection, where simultaneous blood samples were taken for analysis by the reference and index tests, from 853 patients were recruited into the study based on confirmed consent/assent from the four recruitment centres (Figure 5). Recruitment commenced in August 2010 in two centres; the third and fourth centres commenced recruitment in May 2011 and January 2012 respectively. The end date for recruitment from all centres was 31 January 2013. Analyses were performed on 922 episodes (91.7%) from 795 patients; the distribution of analysed episodes across the four centres is shown in Table 4.

FIGURE 5.

Trial accrual: cumulative recruitment for the trial as a whole and for each centre individually. Also shown is the timing of the key amendments to the trial protocol around the eligible clinical setting and resampling of patients who developed new episodes of infection.

TABLE 4 - Numbers recruited and analysed by hospital site

	Hospital site				Overall
	Site 1	Site 2	Site 3	Site 4
Total patient recruitment	398	301	142	12	853
Total episode recruitment	481	343	170	12	1006
Episodes withdrawn from analysis
Clinical issues
Inclusion criteria not met	1	5	2	0	8
BCs lost	0	3	0	0	3
Lack of clinical information	4	0	0	0	4
Assay failures
Reagent control failure	2	4	0	0	6
IC failure	26	19	11	0	56
Other reasons	2	3	2	0	7
Analysed episodes	446	309	155	12	922

Patient cohort

Of the 922 new episodes of suspected sepsis-related health-care-associated bloodstream infection, 60% occurred in male patients, which is typical of current national critical care in the UK. 93 The median age of patients was 58 years (IQR 45–68 years) compared with 70–74 years from national statistics during this period. 93 The gender (Figure 6) and age (Figure 7) distributions relating to patient episodes were similar across the recruitment sites. Patients were exposed to a median of 8 days (IQR 4–16 days) of hospital care prior to each of the suspected sepsis episodes (Figure 8), which is longer than the median stay in critical care during this period nationally. 93 The range and number of patients recorded by primary specialty are shown in Table 5, suggesting an inclusive, generalisable study cohort. The number of organ support activities was captured at the time of blood sampling using the Critical Care Minimum Data Set (CCMDS) national reporting system, the ranges of which are shown for the whole episode cohort and for each centre in Figure 9. 94 These data show that our patient cohort was receiving far more organ support interventions than was recorded nationally during this period, where nationally two-thirds of patients receive two or fewer organ support interventions and 11% have no documented organ support. 93 Survival to 28 days was 87% (95% CI 84% to 89%) and to hospital discharge was 79% (95% CI 77% to 83%) (Figure 10). These outcomes compare favourably with national audit figures for sepsis outcomes and for critical care more generally. 95 Of note, the vast majority of patients (86%) were receiving antimicrobial and/or, on occasion, antiviral pharmaceutical therapies, within the 48 hours prior to the suspected new sepsis episode (Figures 11–13). Antimicrobial drugs were often delivered in combinations and included the most potent, broad-spectrum antibiotics available, usually reserved for hard-to-treat infections.

FIGURE 6.

Gender distribution for the analysed episodes (n = 922).

FIGURE 7.

Age distribution for the analysed episodes. (a) All sites combined; (b) site 1; (c) site 2; and (d) site 3. Site 4 data are not shown separately due to the small number of episodes (n = 12) contributed but are included in the all-sites graph.

FIGURE 8.

Days in hospital prior to research blood sample. Distribution is shown for individual sites (sites 1–3) and combined for all sites. Site 4 data are not shown separately because of the small number of episodes (n = 12) contributed but is included in the all-sites column. The solid line indicates the median value in each group. Median and IQR values are as follows: all sites, median 8 days, IQR 4–16 days; site 1, median 7 days, IQR 3.5–14 days; site 2, median 7 days, IQR 4–13 days; site 3, median 13 days, IQR 6–32 days.

TABLE 5 - Number of patients contributing suspected sepsis episodes recorded by primary specialty

Specialty	Number of patients
Surgery
General surgery	97
Colorectal surgery	30
Hepatobiliary and pancreatic surgery	23
Upper gastrointestinal surgery	52
Vascular surgery	28
Trauma and orthopaedics	17
Ear, nose and throat	3
Maxillofacial surgery	9
Plastic surgery	3
Burns care	7
Cardiothoracic surgery
Cardiothoracic surgery	31
Cardiac surgery	17
Thoracic surgery	18
Cardiothoracic transplantation	3
Neurosurgery	113
Medical
General medicine	32
Critical-care medicine	148
Gastroenterology	28
Endocrinology	5
Hepatology	4
Cardiology	6
Respiratory medicine	19
Genitourinary medicine	9
Nephrology	31
Medical oncology	3
Neurology	6
Rheumatology	1
Care of the elderly	1
Haematology
Clinical haematology	86
Blood and marrow transplantation	8
Obstetrics and gynaecology
Obstetrics	1
Gynaecology	7
Others	7
Total	853

FIGURE 9.

Organ support activities for patients when a new sepsis episode was suspected. (a) All sites; (b) site 1; (c) site 2; (d) site 3; and (e) site 4. Distribution of activities is shown for individual sites (sites 1–4) and combined for all sites, and is an ordinal scale representing organ support therapy over seven organ systems with a possible score range of 0–9 because cardiovascular and respiratory includes a level of organ support from 0 to 2. 94

FIGURE 10.

Survival to (a) 28 days and (b) hospital discharge: combined (all sites) and individual (sites 1–4) site data are shown.

FIGURE 11.

Application of antimicrobial/antiviral therapy prior to blood sampling. Data show the percentage of analysed episodes in which antimicrobial or antiviral therapy was commenced within the 48 hours prior to blood sampling for culture and SeptiFast real-time PCR.

FIGURE 12.

Number of antimicrobial/antiviral drugs administered in each patient episode. Data are expressed as a percentage of the total analysed episodes for all sites combined and for individual sites and show the numbers of drugs administered within the 48 hours prior to blood sampling for culture and SeptiFast real-time PCR. Site 4 data are not shown separately because of the small number of episodes (n = 12) contributed but are included in the all-sites analysis.

FIGURE 13.

Most common antimicrobial/antiviral drugs administered. (a) All sites; (b) site 1; (c) site 2; (d) site 3; and (e) site 4. Figure shows the most common drugs used for all sites combined and individual sites. Data are shown as a percentage of the total analysed episodes for drugs administered within the 48 hours prior to blood sampling for culture and SeptiFast real-time PCR.

Patients studied in this report were included if a new sepsis episode was suspected, defined by clinical criteria. Circulating CRP and PCT concentrations were measured at the time of each suspected new sepsis episode and are reported in Figure 14. For all sites combined, the mean concentrations of CRP and PCT were 160 mg/l (95% CI 152.3 to 167.6 mg/l) and 7.75 ng/ml (95% CI 6.12 to 9.39 ng/ml) respectively. These data strongly suggest that our patient cohort had objective evidence of systemic inflammation at the time of blood sampling. 7 No serious adverse incidents involving patients or staff were reported that were attributable to the research protocol as implemented.

FIGURE 14.

Circulating (a) CRP and (b) PCT levels in the study population: combined (all sites) and individual (sites 1–4) site data are shown; data represent mean ± 95% CI for each biomarker. Total numbers analysed for CRP and PCT at all sites were 898 and 877 respectively.

Pathogen spectrum

Culture

The pathogen range and prevalence detected in the primary blood sample by culture in association with independently adjudicated bloodstream infection, is shown in Table 14. Eleven out of 88 (12.5%) pathogens in positive BC from adjudicated bloodstream infections were not found on the SML, a greater proportion than that predicted by the manufacturer (quoted as < 5%). 29 How the pathogens detected varied by hospital site is shown in Figures 15 and 16 and Table 15. Figure 17 shows the incidence of polymicrobial infections in positive BC. The 88 pathogens identified by culture were associated with 74 cultures containing a single organism and 7 cultures containing two organisms, a rate of polymicrobial infection of 8.8%.

TABLE 14 - Pathogens detected: BC

Pathogen species	Number of cases detected
	BC only	PCR only	BC and PCR
Pathogens on the SML
Gram-negative bacteria
Escherichia coli	4	19	12
Klebsiella (pneumoniae/oxytoca)	3	28	10
Serratia marcescens	0	3	1
Enterobacter (aerogenes/cloacae)	2	12	1
Proteus mirabilis	0	0	0
Acinetobacter baumanii	1	0	0
Pseudomonas aeruginosa	3	1	5
Stenotrophomonas maltophilia	0	0	1
Gram-positive bacteria
Staphylococcus aureus	1	19	1
CoNS	3	6	2
Streptococcus pneumoniae	0	2	0
Streptococcus species	4	5	3
Enterococcus faecium	6	7	3
Enterococcus faecalis	1	5	3
Fungi
Candida albicans	2	9	2
Candida tropicalis	0	1	0
Candida parapsilosis	0	0	0
Candida glabrata	3	2	0
Candida krusei	0	3	0
Aspergillus fumigatus	0	1	0
Total	33	123	44
Pathogens not on the SML
Bacteroides fragilis	1	0	0
Citrobacter koseri	1	0	0
Clostridium perfringens	1	0	0
Enterobacter amnigenus	1	0	0
Leclercia adecarboxylata	1	0	0
Morganella morganii	2	0	0
Staphylococcus capitis	1	0	0
Salmonella enterica	1	0	0
Raoultella planticola	1	0	0
Candida lipolytica	1	0	0
Total	11	0	0

FIGURE 15.

Number of (a) Gram-negative and (b) Gram-positive bacteria from the SML isolated from BCs at each hospital site (sites 1–3). No positive BCs were identified at site 4.

FIGURE 16.

Number of fungal organisms from the SML isolated from BC at each hospital site (sites 1–3). No positive BCs were identified at site 4.

TABLE 15 - Organisms detected by BC not present on SML

Organisms isolated from BC at each hospital site (sites 1–3) not on SML. No positive BCs were identified at site 4.

Hospital site 1	Hospital site 2	Hospital site 3
Bacteroides fragilis	Staphylococcus capitis	Clostridium perfringens
Citrobacter koseri		Raoultella planticola
Enterobacter amnigenus		Candida lipolytica
Leclercia adecarboxylata
Morganella morganii (n = 2)
Salmonella enterica

FIGURE 17.

Numbers of pathogens detected per infection episode. Data show the number of infection episodes in which one, two or three pathogens were detected by BC and by SeptiFast real-time PCR.

Table 16 describes similar information for the enhanced reference standard. This table includes pathogens isolated from blood and other sites, but does not include pathogens from adjudicated bloodstream infections on sample day, which are shown in Table 14. Notably, a higher percentage of infections (23%, 59 out of 261 positive sample cultures from adjudicated infection) were due to pathogens not on the SML, indicating that cultures from sites other than the bloodstream have a higher incidence of off-panel organisms. The sites of infection from which pathogens included in the enhanced reference standard were cultured are shown in Table 17; these are in addition to the pathogens isolated from BC on sample day (shown in Table 14) which were also included in the enhanced reference standard.

TABLE 16 - Pathogens detected at other sites: enhanced reference standard

Enhanced reference standard is defined as any positive culture in blood and/or cultures from other specimens including culture-proven central venous or arterial line colonisation occurring 48 hours either side of the primary blood sample. Figures do not include pathogens from adjudicated bloodstream infections on sample day, which are shown in Table 14.

Pathogen species	Number of episodes detected
	Culture only	PCR only	Culture and PCR
Pathogens on the SML
Gram-negative bacteria
Escherichia coli	29	11	8
Klebsiella (pneumoniae/oxytoca)	19	24	4
Serratia marcescens	5	2	1
Enterobacter (aerogenes/cloacae)	5	11	1
Proteus mirabilis	2	0	0
Acinetobacter baumanii	0	0	0
Pseudomonas aeruginosa	12	1	0
Stenotrophomonas maltophilia	5	0	0
Gram-positive bacteria
Staphylococcus aureus	33	16	3
CoNS	9	5	1
Streptococcus pneumoniae	2	2	0
Streptococcus species	9	5	0
Enterococcus faecium	7	7	0
Enterococcus faecalis	4	4	1
Fungi
Candida albicans	29	6	3
Candida tropicalis	1	1	0
Candida parapsilosis	1	0	0
Candida glabrata	5	2	0
Candida krusei	0	3	0
Aspergillus fumigatus	3	1	0
Total	180	101	22
Pathogens not on the SML
Anaerobes	2	0	0
Candida lipolytica	1	0	0
Clostridium difficile	8	0	0
Unidentified aerobic Gram-negative Bacillus	14	0	0
Enterococci	3	0	0
Penicillin-resistant Enterococcus	1	0	0
Vancomycin-resistant Enterococcus species	1	0	0
Haemophilus influenzae	7	0	0
Hafnia alvei	1	0	0
Influenza A (H1N1)	3	0	0
Moraxella catarrhalis	1	0	0
Morganella morganii	2	0	0
Proteus species	2	0	0
Pseudomonas species	7	0	0
Raoutella planticola	1	0	0
Streptococcus parasanguinis	1	0	0
Yeast (unidentified)	4	0	0
Total	59	0	0

TABLE 17 - Pathogens cultured from other sites: enhanced reference standard

Figures do not include pathogens from adjudicated bloodstream infections on sample day, which are shown in Table 14.

New bloodstream infection adjudicated as occurring 48 hours either side of the PCR sample day but not on the sample day.

Pathogen species	Infection site
	Blooda	Lung	Urinary tract	Surgical site	Other
Pathogens on the SML
Gram-negative bacteria
Escherichia coli	9	5	9	10	4
Klebsiella (pneumoniae/oxytoca)	3	7	4	1	8
Serratia marcescens	0	3	0	0	3
Enterobacter (aerogenes/cloacae)	0	3	0	1	2
Proteus mirabilis	1	0	1	0	0
Acinetobacter baumanii	0	0	0	0	0
Pseudomonas aeruginosa	0	5	0	1	6
Stenotrophomonas maltophilia	0	4	0	0	1
Gram-positive bacteria
Staphylococcus aureus	3	19	1	1	12
CoNS	1	0	0	2	7
Streptococcus pneumoniae	1	1	0	0	0
Streptococcus species	4	1	0	2	2
Enterococcus faecium	2	0	0	5	0
Enterococcus faecalis	0	1	0	0	4
Fungi
Candida albicans	1	24	0	4	3
Candida tropicalis	0	0	0	0	1
Candida parapsilosis	0	1	0	0	0
Candida glabrata	2	3	0	0	0
Candida krusei	0	0	0	0	0
Aspergillus fumigatus	0	3	0	0	0
Total	27	80	15	27	53
Pathogens not on SML
Anaerobes	0	0	0	1	1
Candida lipolytica	1	0	0	0	0
Clostridium difficile	0	0	0	0	8
Unidentified aerobic Gram-negative Bacillus	0	4	6	3	1
Enterococci	0	0	1	2	0
Penicillin-resistant Enterococcus	1	0	0	0	0
Vancomycin-resistant Enterococcus species	0	0	0	0	1
Haemophilus influenzae	0	5	0	0	2
Hafnia alvei	0	0	0	0	1
Influenza A (H1N1)	0	3	0	0	0
Moraxella catarrhalis	0	0	0	0	1
Morganella morganii	1	0	0	0	1
Proteus species	0	0	0	2	0
Pseudomonas species	0	1	3	0	3
Raoutella planticola	0	1	0	0	0
Streptococcus parasanguinis	0	0	0	1	0
Yeast (unidentified)	0	3	1	0	0
Total	3	17	11	9	19

SeptiFast polymerase chain reaction

The pathogen range and prevalence detected in blood specimens by SeptiFast is shown in Table 14. For pathogens on the SML, PCR detected 167 pathogens compared with 77 by BC, with fungal species representing only 18 (11%) of the PCR-detected pathogens.

Detection of multiple organisms was more common with SeptiFast, with PCR detecting 167 organisms in 144 blood samples with 124 blood samples containing a single organism, 20 samples containing two organisms and one sample containing three organisms, a rate of polymicrobial PCR of 14.5% (see Figure 17). The pathogen range detected by PCR in the enhanced reference standard appeared broadly similar to that observed in BCs (see Table 16).

Blood culture contamination

Blood culture contamination was adjudicated independently at each site without any knowledge of the SeptiFast results and is shown in Figure 18. These contamination rates were reported within NHS, forming part of continuous quality improvements for BC sampling at each site. The overall BC contamination rate was about 2% for all BCs taken within this study and provides independent evidence for blood sampling quality when compared with national standards. 96 Following study completion and calculation of the clinical diagnostic accuracy metrics it became apparent, post hoc, that none of the adjudicated BC contaminations were associated with a positive SeptiFast result.

FIGURE 18.

Blood culture contaminants. Percentage of BC adjudicated as contaminants for (a) all sites combined; (b) site 1; (c) site 2; and (d) site 3. No contaminants were identified in the 12 BCs taken at site 4. (e) Species identified as a percentage of total contaminants.

Discussion

The comprehensive systematic review of the likely diagnostic accuracy performance of SeptiFast (see Chapter 2) highlights significant deficiencies in the quality of design and reporting of clinical diagnostic accuracy studies in this field. As a result, the current evidence base that underpins any recommendations on the suitability of the test for adoption into frontline sepsis clinical care pathways is unsound. Our aim in this chapter was to report the clinical diagnostic accuracy of SeptiFast compared with BC as the gold standard test in a defined setting of suspected sepsis-related health-care-associated bloodstream infection employing study design, performance and reporting that meet all the features of the STARD criteria. 87 In addition, we investigated a broader (‘enhanced’) reference standard in this study population that included adjudicated culture-proven health-care-associated infection more inclusively than just bloodstream.

We have reported clinical diagnostic accuracy measures of SeptiFast real-time PCR against BC at two levels, event level (i.e. concordance of PCR positivity with BC positivity) and species concordance level (i.e. PCR detection of the same organism at species/genus level as identified by BC). Intuitively, the latter metric would seem to be a more robust measure of the potential clinical utility of the SeptiFast test given that organism identification at species/genus level is a routine element of culture-based diagnosis in clinical practice and the SeptiFast test only delivers results at species level. As discussed in Chapter 2, although some previous studies have specified the metrics used to evaluate SeptiFast real-time PCR, in other cases it is uncertain whether data was reported at event or species level.

In the present study, sensitivity of the SeptiFast real-time PCR assay, measured at both event and species level, was poor (58.8% and 50.0% respectively), suggesting that the assay may have little value as a ‘rule-out’ test in suspected sepsis-related health-care-associated bloodstream infection. However, our patients had low pre-test probabilities for bloodstream infection at an event rate (8.7%) and at species level (9.2%). Returning a negative SeptiFast test for a patient in our study would result in no more than a 5.9% and 7.4% chance of bloodstream infection at an event and species level respectively. In addition, this information would probably be returned to clinical practice earlier than a BC result – although this was not formally tested in this study. These relatively low levels of post-test probability might suggest that returning a negative SeptiFast test result in one of our patients with suspected sepsis could lead to ruling out bloodstream infection quickly and, therefore, stopping antimicrobial therapies. However, based on these post-test probabilities, this would result in the significant risk of undertreating up to six to eight patients with sepsis-related bloodstream infection in every 100 with suspected sepsis. Effective management of sepsis-related bloodstream infection requires timely and adequate antimicrobial therapy and, based on these metrics, stopping antimicrobial therapy early in this high-risk population could not be advised on the basis of a negative SeptiFast result.

The specificity of the SeptiFast real-time PCR test was higher, 88.5% at event level and 85.8% at species level, suggesting that the test has potential value as a ‘rule-in’ test, allowing reasonable confidence that a positive test indicates the presence of a specific pathogen. However, as described above, our patients had low pre-test probabilities of bloodstream infection. Returning a positive SeptiFast test for a patient in our study was an infrequent event (less than 1 in 10 on average) but was associated with a 41% and 34% chance of bloodstream infection at an event and species level respectively. Therefore, returning a positive SeptiFast test in one of our patients with suspected sepsis does substantially increase the chance of patients having a bloodstream infection, but not to a high level. A further significant limitation of SeptiFast that is likely to restrict its impact on clinical practice is that the test does not provide information on the antibiotic sensitivities of the identified pathogen, with the exception of meticillin resistance in S. aureus. In the absence of this information, currently provided only by culture, clinicians may not be willing to refine antibiotic therapy on the basis of an organism identified by SeptiFast real-time PCR alone. 36 Development of molecular platforms with the ability to detect a broad range of resistance genes direct from blood would be an important step towards clinical utility. However, challenges remain including the broad range of genotypes involved in some resistances (e.g. extended-spectrum beta-lactamases) and the precise relationship between genotype and functional antibiotic resistance. Therefore, in our patients with suspected sepsis-related health-care-associated infection, directing antimicrobial treatment based on a positive SeptiFast result is problematic in terms of both the post-test uncertainty in the positive test and the lack of information about likely antimicrobial drug responses.

To date, no clinical trials have been conducted on the likely impact and safety of acting on the results of SeptiFast real-time PCR in patients in any setting.

Based on the results reported here, the investigators have significant reservations about the likely efficacy and safety of such a study in the setting of sepsis-related health-care-associated infection. However, this conclusion is based solely on the assumption that culture results are an error-free reference standard, an assumption we will explore further in Chapter 4.

Our study population focused on critically ill patients developing new clinical evidence for suspected sepsis who had been hospitalised for at least 48 hours in the NHS in the north-west of England. Despite the pragmatic regional focus of this study, our patients were drawn from a wide range of primary specialties, were broadly representative in terms of age and sex of critical care patients throughout England and Wales compared with national audit figures, but were more likely to be receiving higher levels of invasive organ supportive interventions and had been exposed to a high level of broad-spectrum antibiotics in the immediate period leading into study recruitment. Within this complex study-setting, there was evidence of differences in diagnostic accuracy metrics for SeptiFast between hospital sites, particularly assay specificity at the level of pathogen concordance. Specificity against BC across the three main contributor sites ranged from 91.3% (95% CI 88.2% to 93.9%) at site 1 to 74.3% (95% CI 66.5% to 81.2%) at site 3. Intriguingly, these differences were maintained when the PCR assay was compared with the enhanced reference standard, specificity ranging from 92.6% (95% CI 89.3% to 95.2%) at site 1 to 71.4% (95% CI 61.4% to 80.1%) at site 3. The reasons for these differences are unknown; however, site differences in antimicrobial use, which would be expected to impact on the level of false-positive results, and/or case mix are potential explanations.

What factors limit the diagnostic accuracy of the SeptiFast test? Poor diagnostic sensitivity is driven by the relatively high number of false-negative results produced by SeptiFast real-time PCR. A significant proportion of false negatives (12.5%) were associated with organisms not present on the SML. This finding is consistent with several previous studies57,63,76 and suggests that restricting the SeptiFast panel to the most common bloodstream pathogens has a limiting effect on the diagnostic performance of the test in practice. False negatives associated with organisms present on the SML appeared to be spread broadly across all organisms, although there were exceptions. For example, none of three cases of Candida glabrata found by BC were detected by SeptiFast real-time PCR, although, interestingly, two cases of infection with this organism were detected by SeptiFast real-time PCR in the absence of confirmatory BC. The most likely explanation of the false negatives for common, on-panel organisms is that the levels of circulating DNA may be below the limits of detection of the assay in some infection episodes. The reported detection limit for SeptiFast real-time PCR is 3–100 CFU/ml29 which compares relatively favourably to the microbial burden often associated with adult sepsis (< 10 CFU/ml). 23,97 However, this varies considerably between species with some organisms including C. glabrata, being undetectable at 3 CFU/ml blood. A similar explanation could be proposed for the performance of the test in detecting CoNS. A PCR signal was detected in 40% of cases where a culture-confirmed bloodstream infection with CoNS was adjudicated, while no PCR signal was detected in the 27 cases of positive BCs adjudicated as CoNS contaminants. Contamination during the blood collection procedure is likely to be associated with low bacterial load and may therefore be below the minimum analytical sensitivity for CoNS (30 CFUs/ml) reported for SeptiFast real-time PCR. 29

In the current study, SeptiFast real-time PCR also produced a significant number (n = 123) of false-positive results (i.e. pathogens detected by PCR in blood that are not found by BC), which lowered the specificity of the assay. Previous studies have suggested that, in many cases, the organism detected by SeptiFast real-time PCR can be found only on culture of specimens obtained from other sites of infection outside the bloodstream such as urine, bronchioalveolar lavage fluid, peritoneal fluid, etc. 41,69 This has led to the suggestion that pathogen DNA in the bloodstream may have clinical relevance in terms of the infection status of the patient even when corresponding BCs are negative and that concordance with an enhanced reference standard that includes cultures from blood and other significant specimens may be of value in assessing the diagnostic utility of SeptiFast real-time PCR. 36 To our knowledge, this is the first prospective study designed around the STARD criteria87 to assess the diagnostic accuracy of SeptiFast against an enhanced reference standard. The data show that the SeptiFast assay sensitivity was markedly lower (33.1% and 18.1% at event and species level respectively) when other microbiological findings were taken into account as part of blinded, independent infection adjudication, indicative of the high number of pathogens detected at sites outside the bloodstream that were not found by SeptiFast real-time PCR. Assay specificity was maintained using the enhanced reference standard (enhanced reference standard vs. BC: 90.8% vs. 88.5% at event level; 85.9% vs. 85.8% at species level). These data suggest that a positive SeptiFast result in the setting of suspected sepsis-related health-care-associated infection may have some diagnostic ‘rule-in’ utility as an indicator of the overall infection status of the patient. However, a negative test does not appear to have any utility in excluding infection in this setting.

The independent review and meta-analysis (see Chapter 2) of the current literature studies on SeptiFast diagnostic accuracy indicated significant deficiencies that impact on the derived diagnostic accuracy metrics with none, as reported, meeting the STARD criteria87 in full. The strengths of the present study are the systematic approach taken in the design and conduct of the study. These include assessment of the test in a defined, focused patient cohort in the setting of suspected health-care-associated bloodstream infection, prior publication of the trial protocol,1 blinding of SeptiFast real-time PCR and culture results, independent blinded clinical adjudication of bloodstream and other infections, strict adherence to clinical and CE-marked laboratory protocols and reporting of assay failure rates. The investigators recognise, however, that there remain significant weaknesses principally associated with the use of BC as the gold standard reference test. This approach provides diagnostic accuracy measures that can be directly compared with other studies in this field which have all used BC as the stated reference standard (reported in Chapter 2). However, in contrast to other studies internationally, we have also taken a Bayesian approach to describing and arguing the likely clinical utility of SeptiFast real-time PCR in the setting of a critically ill patient developing new signs of suspected sepsis while receiving complex health care – and, for the first time, this has been performed within the NHS.

It is, however, important to point out that the performance of SeptiFast real-time PCR in terms of clinical diagnostic accuracy is predicated on the assumption that BC is 100% error free, which seems unlikely. Potential sources of error include low sample volume, lack of replicate BC sets, delays in incubation and contamination during sampling. 9,98 In this study, BC was taken as part of routine clinical service and, although training in approved procedures for taking and processing BC were promoted, we cannot be certain that errors did not occur, although the mean level of BC contaminants across the four centres was consistent with current quality performance standards. A further potentially important source of error affecting the gold standard is the high level of antimicrobial use in this patient cohort. One or more antimicrobial drugs had been given within 48 hours prior to blood sampling in > 85% of the suspected bloodstream infection episodes included in the study. The impact on blood and other specimen culture performance cannot be quantified, but it is highly likely that this will be a significant factor. This study was commissioned specifically to assess the diagnostic utility of SeptiFast real-time PCR in the setting of sepsis-related health-care-associated infection where empirical broad-spectrum antibiotic therapy is frequently employed. This setting challenges the utility of culture in supporting complex clinical decisions. However, given that levels of free pathogen DNA and the intact pathogen load in the circulation are likely to be interrelated, this may also present a challenging environment for SeptiFast real-time PCR.

In Chapter 4, we will undertake a preliminary assessment of the potential impact of error in the BC gold standard through the use of latent class statistical modelling. 99 In addition, the impact on diagnostic performance of including the commonly used circulating inflammatory biomarkers white cell count, CRP and PCT as instrumental variables will be considered.

Rationale and aims

The implicit assumption underlying the standard method of analysis as reported in Chapter 3 is that the sensitivity and the specificity of the laboratory culture-confirmed diagnosis of bloodstream infection are both 100%. It is assumed that there are no misclassification errors – that is, if an infection is present it will be detected and if it is absent then it is correctly recorded as being absent. Current evidence suggests that BC has limitations that are likely to mean that both specificity and sensitivity of this test will be less than 100%. The wide range of factors that are known to influence the diagnostic accuracy of BC, and the supporting evidence base, are covered in detail elsewhere96 as part of the latest consensus-derived standard operating procedures that guide NHS practice. Therefore, although we believe there is a widespread understanding of the limitations of BC in the clinical community, BC is in routine use worldwide and the formal definitions and guidelines relating to the diagnosis of suspected bloodstream infection are based on this test with no currently accepted alternatives. It is therefore not surprising that the diagnostic validity of new technologies such as PCR is judged against BC.

The standard method of analysis is merely a convention and our aim in the present chapter is to (i) challenge current assumptions concerning the performance of BC, and their use as a reference standard to assess new technologies, using alternative statistical methodology100–103 – which has been discussed recently by Waikar et al. 104 – and (ii) illustrate how assumptions that are more realistic (together with additional data) may lead to different conclusions that could ultimately provide a more realistic basis for assessing the diagnostic accuracy of emerging technologies. The guiding motivation for the work reported in this chapter is that, if the prior assumptions concerning the performance of the reference standard (BC) are invalid, then inferences concerning the performance of the new competitor (SeptiFast real-time PCR) will be biased, quite possibly severely so. Full details of the proposed statistical methodology for the comparisons of two or more fallible indicators of disease have been provided previously by one of the co-authors of the present report. 99

The key to the use of the statistical methodology to be described in Statistical models is the acknowledgement that all diagnostic indicators (including microbiologically proven BC) are almost certainly fallible (that is, subject to classification errors). BCs might, for example, fail to detect certain infections (false negatives) or may be subject to contamination (false positives). Instead of basing our analyses of the characteristics of the SeptiFast real-time PCR results on unrealistic assumptions concerning the BCs, we relax these assumptions and use the data to estimate the sensitivity and specificity of both the BC and SeptiFast real-time PCR results. However, in order for this to be possible, we need further data on the likely infection status of the patient (in the examples below we use infections at additional sites and/or levels of circulating immune-inflammatory biomarkers). We also need to assume that the measurement or classification errors for the different indicators of infection are statistically independent (uncorrelated). The latter is a strong assumption but likely to hold true if the different indicators are founded on different biological processes. The statistical methods used are known as latent class analyses,99 the latent classes being the presence or absence of a bloodstream infection and ‘latent’ because we have acknowledged that none of the indicators of infection allow us to be 100% confident concerning the true infection status of the patient.

Statistical models

Model fitting and parameter estimation

The statistical model provided by description in Statistical models is an example of a latent class or finite mixture model. Unless we are prepared to assume that one of the tests is providing the truth, presence or absence of infection cannot be directly observed but is inferred from the observed pattern of test results. Fitting the model was carried out using maximum likelihood – specifying the Maximum Likelihood Robust option – via the expectation–maximisation algorithm using Mplus version 7.0 software (Muthén & Muthén, Los Angeles, CA, USA). 105 All models allowed for the prevalence of infection to vary with hospital, thus eliminating the latter as a source of confounding. Appendix 16 provides a listing of a typical Mplus command (input) file. Dunn99 provides full details of the rationale and technical aspects of latent class/finite mixture models for diagnostic test data. Data from hospital 4 (12 subjects) were excluded from all analyses on the grounds of too few data (resulting n = 910).

Results

First, the model was fitted using data from only two indicators (BC and SeptiFast real-time PCR), allowing infection prevalence to vary with hospital. Assuming that the laboratory culture is infallible, the following parameter estimates for SeptiFast real-time PCR are obtained: sensitivity, α_PCR, = 58.8% (95% CI 49.7% to 67.8%) and specificity, β_PCR, = 88.8% (95% CI 87.0% to 90.6%) and the prevalence of infection is estimated to be 8.7%. Assuming the specificities of both BC and SeptiFast real-time PCR to be 100% yields the following estimates: α_BC = 33.6% (95% CI 27.0% to 40.1%) and α_PCR = 58.8% (95% CI 49.7% to 67.8%). The estimated prevalence of infection has now risen to 26.2%. Note that, as indicated earlier, the data cannot be used to distinguish these two models, or allow the relaxation of the competing assumptions, without the use of further information as illustrated below.

Second, the model was fitted for data from three different indicators of infection (BC, SeptiFast real-time PCR and infection at ASs). As there are now three indicators, assumed to be conditionally independent (i.e. with uncorrelated misclassification errors) the model can be fitted without any unrealistic assumptions concerning the performance of any of the three tests. Six parameters representing the sensitivities and specificities of the three tests (plus the prevalence of infection) can be estimated. The results are shown in Table 20(a).

Finally, the potential utility of circulating levels of CRP and PCT measurements was investigated. The summary statistics for each of the two measures is presented according to BC status and again by SeptiFast real-time PCR status. The results are shown in Table 21. CRP does not appear promising but the PCT measurements are associated (in the right direction) with both BC and SeptiFast real-time PCR outcomes. To use PCT as an instrumental variable it is assumed that this association arises solely from PCT’s association with the underlying infection status. The advantage of using this measure as an instrument as opposed to ASs is that it is a marker determined by a completely different process from those used to assess BC and PCR positivity. Although it is not absolutely certain that PCT is a valid instrument (i.e. its measurement errors are not associated with the misclassification errors for BC and/or SeptiFast real-time PCR), the assumption does have a degree of face validity.

When PCT was included in the model for the two binary indicators, the results (not shown) indicated that the estimated mean PCT for those without infection (as predicted by the model) was 4.0 ng/ml (standard error 0.5 ng/ml) and was, for the infected participants, very much higher at 129.5 ng/ml (standard error 17.5 ng/ml). Estimated sensitivities for the BC and SeptiFast real-time PCR were 18.8% (95% CI 2.5% to 35.0%) and 50.0% (95% CI 29.6% to 70.3%) respectively. Corresponding specificity estimates were 91.5% (95% CI 89.7% to 93.4%) and 85.7% (95% CI 83.4% to 88.0%). Accordingly, the PCT measurements were dichotomised to create a binary indicator of infection in the setting of suspected sepsis (positive if PCT > 6 ng/ml – reflecting the median PCT level that indicates systemic inflammation resulting from severe systemic infection such as in the bloodstream;106 negative if PCT ≤ 6 ng/ml). No attempt was made to optimise the value of this cut-point – the idea being to simply illustrate the power of the method. The latent class/finite mixture model was refitted using dichotomised PCT measurements as a third, independent indicator of infection status. As before, both the sensitivity and specificity of the BC and of the SeptiFast real-time PCR were left free to be estimated. The results are given in Table 20(b). Note that the model fitting does not require that all 910 participants have PCT data – the model is fitted to all available data. Concordances between the BC and dichotomised PCT, and between SeptiFast real-time PCR and dichotomised PCT, are shown in Tables 22 and 23, respectively.

A latent class/finite mixture model was then fitted to all four indicators of infection (BC, SeptiFast real-time PCR, AS and PCT). The resulting estimates are shown in Table 20(c).

How might one develop a better biomarker of infection? Clearly, a combination of the diagnostic test results has the potential to have a considerably higher sensitivity than when they are used alone, although specificity might be lowered. In order to illustrate the promise of this approach the SeptiFast real-time PCR and PCT results were combined: an infection was indicated if the SeptiFast real-time PCR result was positive or if the PCT result was positive, or if both were positive. If neither test was positive then the combined test result was negative. Table 20(d) provides estimates of the sensitivity and specificity of the combined SeptiFast real-time PCR/PCT marker obtained through fitting a latent class model using BC and infection at an additional site as the other two conditionally independent indicators. Table 24 summarises the estimated diagnostic accuracy characteristics of SeptiFast real-time PCR incorporating different instrumental variables. The performance of the combined SeptiFast real-time PCR/PCT marker looks very promising as both sensitivity and specificity are at, or very close to, 100%. We stress, however, that this is only a preliminary result – further work needs to be done on algorithms for the optimal combination of indicators of infection, using both existing data to generate well-performing algorithms and new data for a thorough validation of their performance.

Conclusion and implications

Overall, the results of the latent class/finite mixture modelling lead us to conclude the following:

1. The reference standard (BC) used routinely to assess the diagnostic accuracy of new tests of bloodstream infection, including in the present study, should not be assumed to be a perfect – infallible – indicator of infection. In particular, BC appears to have a worryingly low sensitivity (≈ 40%, at most).

2. The sensitivity of the SeptiFast test of 60–80% is also less than ideal for a useful diagnostic test but appears much better than that of the reference standard (BC).

3. The specificities of BC and SeptiFast PCR are both high (> 95%) with blood samples identified as positive by either method from our study population of suspected sepsis-related health-care-associated bloodstream infection having a high probability of infection. Data from additional body sites and from circulating biomarkers such as PCT appear to be additional indicators of infection and a dichotomised PCT measurement appears to be as sensitive as BC and only marginally less specific.

4. It appears likely that the lone use of the reference (BC) in clinical settings could lead to seriously misleading estimates of the prevalence of infection. Data reported here suggest that in the clinical setting of suspected health-care-associated sepsis, BC might be failing to detect 60–70% of all infections.

5. The less than ideal clinical diagnostic sensitivity of SeptiFast real-time PCR will impact on the associated likelihood ratios and predictive values, even in the presence of high specificity, resulting in more limited diagnostic utility when translating evidence derived from our study population to an individual patient episode. These results indicate that SeptiFast real-time PCR should not replace traditional BC but could be added to the diagnostic test battery together with data on infections from other body sites and on levels of biomarkers such as PCT. Algorithms for the optimal use of such a test battery should be the subject of further work as illustrated by the final latent class model to assess the performance of a combined SeptiFast real-time PCR/PCT marker, above.

6. Other investigators should seriously consider abandoning the assumption that BC is an infallible gold standard for infection.

7. Based on the provisional analysis results reported in this chapter, there are two situations that might be worthy of consideration for future intervention studies based on the potential for SeptiFast real-time PCR as a more accurate rapid ‘rule-in’ test for pathogens in blood samples:

   1. In the setting of suspected sepsis-related bloodstream infection, what is the impact on antimicrobial stewardship in the setting of a positive SeptiFast test?

   2. In the setting of interventional studies investigating bloodstream infection (e.g. a clinical trial investigating the effectiveness of different durations of antimicrobial therapy), what is the utility of identifying patient cohorts rapidly based on SeptiFast testing rather than relying solely on the results of BC?

Main findings

Limitations

Implications for practice

When compared with NHS service culture standards, the clinical diagnostic accuracy of SeptiFast appears unlikely to result in sufficient diagnostic utility in the setting of suspected sepsis-related health-care-associated infection, despite its potential to deliver results more rapidly.

Using preliminary analyses that take into account the possibility that culture standards are not completely error free, SeptiFast real-time PCR may have a greater ability to rule in infection than may be apparent from conventional analyses – although its diagnostic sensitivity remains inadequate such that clinical utility may remain significantly compromised.

In further preliminary analyses, other circulating biomarkers, such as PCT, when used in combination with SeptiFast real-time PCR, may improve clinical diagnostic accuracy to levels where clinical utility is far more likely.

Based on the present study, we found no evidence that SeptiFast real-time PCR should replace traditional BC but the assay could be added to the diagnostic test battery (together with data on infections from other body sites and levels of biomarkers such as PCT). Algorithms for the optimal use of such a test battery should be the subjects of further work.

We do not know whether our findings are relevant to patients with suspected community-acquired sepsis as they were not investigated and there is a lack of high-quality diagnostic evidence in relation to this setting.

Recommendations for research

Health Technology Assessment trial investigators

Professor Geoffrey Warhurst.

Dr Paul Dark.

Dr Paul Chadwick.

Professor Gordon Carlson.

Dr Andrew Bentley.

Professor Graham Dunn.

Dr Duncan Young.

Health Technology Assessment Trial Steering Committee

Dr Neil Pendleton (chair), Rachel Georgiou (sponsor representative), Professor John Wain (independent member), Rosemary Martin (patient representative), Professor Geoffrey Warhurst, Dr Paul Dark, Professor Gordon Carlson, Dr Duncan Young, Dr Andrew Bentley, Professor Graham Dunn and Dr Paul Chadwick.

We wish to thank:

The NIHR HTA programme for funding this study.

The Intensive Care Foundation (UK), which provided support to Professor Daniel McAuley and Professor Gavin Perkins in their roles as co-directors of the foundation.

Staff in the Research and Development department at SRFT (the study sponsors) for their support and advice throughout the study.

Clinical and nursing staff at participating hospitals

SRFT: Dr Murad Ghrew, Consultant in Respiratory and Intensive Care Medicine; Tracey Evans, Senior Research Nurse and Trial Recruitment Co-ordinator; and Elaine Cooke, Research Nurse.

University Hospital of South Manchester NHS Foundation Trust: Dr John Hopper, Consultant in Anaesthesia and Intensive Care; Alison Royle, Research Nurse; and Alison Middleton, Research Nurse.

Central Manchester Foundation Trust: Dr John Moore, Consultant in Anaesthesia and Intensive Care; Dr Mike Sharman, Consultant in Respiratory and Intensive Care Medicine; Dr Jane Eddlestone, Consultant in Anaesthesia and Intensive Care; and Elaine Coghlan, Research Nurse.

East Lancashire Hospitals Trust: Dr Paul Dean, Consultant in Anaesthesia and Intensive Care; Lynne Bullock, Research Nurse; Martin Bland, Research Nurse; and Donna Harrison-Griggs, Research Nurse.

Laboratory scientists and support staff in the Biomedical Research Facility at Salford Royal NHS Foundation Trust

Dr Satanayarana Maddi, Postdoctoral Research Assistant; Claire Wilson, NIHR Trial Intern; Daniel Graham, NIHR Trial Intern; Kate Timms, NIHR Trial Intern; Matthew Langley, NIHR Trial Intern; Pamela Davies, Senior Technician and Laboratory Manager; and Norman Higgs, Senior Research Technician.

Health-care workers and laboratory staff at the acute care NHS trusts named in this report who facilitated patient participation in this study.

Contributions of authors

Professor Geoffrey Warhurst (chief investigator), Consultant Biomedical Scientist, SRFT, Honorary Professor of Biomedicine, University of Salford: conceived the study, had overall responsibility for laboratory aspects of the study, was involved in design of the diagnostic accuracy study and systematic review, and prepared final report with PD.

Dr Paul Dark (principal clinical investigator), Consultant in Intensive Care Medicine, Reader in Institute of Inflammation and Repair, University of Manchester: conceived the study, had overall clinical responsibility for the design and conduct of the diagnostic accuracy study and systematic review, and prepared final report with GW.

These authors contributed exclusively to the NIHR-HTA funded diagnostic accuracy study as follows:

Professor Graham Dunn (Trial Statistician), Professor of Biomedical Statistics, University of Manchester: contributed to the design of the study, led on the statistical design and analysis of the data, and reviewed a draft of the report.

Dr Paul Chadwick, Consultant Microbiologist, Salford Royal NHS Foundation Trust: contributed to the design of the study, provided microbiological expertise and reviewed a draft of the report.

Dr Andrew Bentley, Consultant in Respiratory and Intensive Care Medicine, University Hospital of South Manchester: contributed to the design and conduct of the study, and reviewed a draft of the report.

Dr Duncan Young, Senior Lecturer and Consultant in Anesthaesia and Intensive Care, John Radcliffe Hospital, Oxford: contributed to study design as expert in clinical trials and reviewed a draft of the report.

Professor Gordon L Carlson, Consultant Gastro-intestinal Surgeon, Salford Royal NHS Foundation Trust: contributed to study design as an expert in the management of health-care-associated sepsis and reviewed a draft of the report.

These authors contributed exclusively to the systematic review as follows:

Dr Bronagh Blackwood, Senior Lecturer, Centre for Infection and Immunity, Queen’s University Belfast: helped developed the initial architecture, and contributed to production and revision of the systematic review report.

Professor Daniel McAuley, Centre for Infection and Immunity, Queen’s University Belfast: provided critical care and clinical trial input, and production and revision of the systematic review report.

Professor Gavin D Perkins, Clinical Professor in Critical Care Medicine, Warwick University: provided critical care and clinical trial input, and production and revision of the systematic review report.

Dr Ronan McMullen, Consultant in Medical Microbiology, Royal Victoria Hospital, Belfast: provided microbiological input, and production and revision of the systematic review report.

Professor Simon Gates, Professor of Clinical Trials, Warwick University: provided statistical leadership and contributed to the production and revision of the systematic review report.

Publications

Dark P, Blackwood B, Gates S, McAuley D, Perkins GD, McMullan R, et al. Accuracy of LightCycler – SeptiFast for the detection and identification of pathogens in the blood of patients with suspected sepsis: a systematic review and meta-analysis. Intensive Care Med 2015;41:21–33.

Warhurst G, Maddi S, Dunn G, Ghrew M, Chadwick P, Alexander P, et al. Diagnostic accuracy of SeptiFast multi-pathogen real-time PCR in the setting of suspected health-care associated bloodstream infection. Intensive Care Med 2015;41:86–93.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

MEDLINE search strategy

URL: http://ovidsp.uk.ovid.com/sp-3.14.0b/ovidweb.cgi

Date range searched: January 2006 to November 2012.

Date of search: 31 October 2011 (updated 30 November 2012).

#1 sepsis.mp. or exp Sepsis/

#2 septic shock.mp. or Shock, Septic/

#3 fung?emia.mp. or Fungemia/

#4 bacter?emia.mp. or Bacteremia/

#5 blood?stream infection$.mp.

#6 blood poison$.mp.

#7 Systemic Inflammatory Response Syndrome/ or SIRS.mp.

#8 septic?emia.mp.

#9 “severe sepsis”.mp.

#10 (presumed adj4 sepsis).mp.

#11 (suspected adj4 sepsis).mp.

#12 #1 or #2 or #3 or #4 or #5 or #6 or #7 or #8 or #9 or #10 or #11

#13 PCR.mp. or Polymerase Chain Reaction/

#14 SeptiFast.mp.

#15 LightCycler.mp.

#16 multiplex PCR.mp.

#17 real time PCR.mp.

#18 real?time PCR.mp.

#19 Molecular Diagnostic Techniques/ or molecular diagnosis.mp.

#20 molecular identification.mp.

#21 #13 or #14 or #15 or #16 or #17 or #18 or #19 or #20

#22 blood cultur$.mp.

#23 Bacteriological Techniques/mt [Methods]

#24 Blood/mi [Microbiology]

#25 #22 or #23 or #24

#26 #12 and #21 and #25

#27 Animals/

#28 #26 not #27

#29 Viruses/

#30 #28 not #29

#31 limit #30 to (humans and yr=“2006 -Current”)

(mp=protocol supplementary concept, rare disease supplementary concept, title, original title, abstract, name of substance word, subject heading word, unique identifier)

Molecular diagnosis of hospital infection research project

Patient consent at commencement: patient information sheet

Patient consent at recovery: patient information sheet

Clinical case record form

(PDF download)

Title

SeptiFast PCR trial

Data

File is SeptiFast.dat;

Variable

Names are LC PCR AS Hospital WCC CRP PCT H1 H2;

! H1 and H2 are binary dummy variables distinguishing the three hospitals

Missing are all (–9999);

Classes C(2);

Categorical LCC PCR AS;

Usevariables LC PCR AS H1 H2;

Analysis

Type = mixture;

Estimator = mlr;

Starts = 1000 20;

Model:

%overall%

C#1 on H1 H2;

%C#1%

[LC$1*3 PCR$1*3 AS$1*3];

%C#2%

[LC$1*–3 PCR$1*–3 AS$1*–3];

Output

Tech10; Cinterval;