The incidence of ventriculostomy-related infections as diagnosed by 16S rRNA polymerase chain reaction: A prospective observational study

Background: Ventriculostomy-related infections (VRIs) are reported in about 10 % of patients with external ventricular drains (EVDs). VRIs are difficult to diagnose due to clinical and laboratory abnormalities caused by the primary neurological injury which led to insertion of the EVD. Polymerase chain reaction (PCR) of the cerebrospinal fluid (CSF) may enable more accurate diagnosis of VRI. We performed a prospective cohort study to measure the incidence of VRI as diagnosed by 16S rRNA PCR. Methods: Patients admitted to intensive care with a primary diagnosis of subarachnoid haemorrhage (SAH), traumatic brain injury (TBI), or intracerebral haemorrhage (ICH), who required an EVD, were assessed for inclusion in this study. Data were extracted from the electronic medical record, bedside charts, or from a prospectively collected database, the Neuroscience Outcomes in Intensive CarE database (NOICE). 16S rRNA PCR was performed on routinely collected CSF as per laboratory protocol. VRI was also diagnosed based on pre-existing definitions. Results: 237 CSF samples from 39 patients were enrolled in the study. The mean patient age was 55.7 years, and 56.4 % were female. The most common primary neurological diagnosis was SAH (61.5 %). The incidence of a positive PCR was 2.6 % of patients (1 in 39) and 0.8 % of CSF samples (2 in 237). The incidence of VRI according to pre-published diagnostic criteria was 2.6 % – 41 % of patients and 0.4 % – 17.6 % of CSF samples. 28.2 % of patients were treated for VRI. Pre-published definitions which relied on CSF culture results had higher specificity and lower false positive rates for predicting a PCR result when compared to definitions incorporating non-microbiological markers of VRI. In CSF samples with a negative 16S rRNA PCR, there was a high proportion of non-microbiological markers of infection, and a high incidence of fever on the day the CSF sample was taken. Conclusions: The incidence of VRI as defined as a positive PCR was lower than the incidence of VRI according to several published definitions, and lower than the incidence of VRI as defined as treatment by the clinical team. Non-microbiological markers of VRI may be less reliable than a positive CSF culture in diagnosing VRI.


Introduction
External ventricular drains (EVDs) are one of the most common neurosurgical devices used in modern practice [1].Ventriculostomyrelated infections (VRIs) are diagnosed in approximately 10 % of patients with an EVD [2,3].These infections are commonly diagnosed using routine microbiological tests (e.g.culture or gram stain of the cerebrospinal fluid (CSF)), analysis of the cytology or biochemistry of the CSF, or clinical parameters such as fever.However, primary neurological injuries which require the placement of an EVD may cause aberrations in clinical findings consistent with central nervous system (CNS) infection, such as raised CSF white cell counts, raised CSF protein, and low CSF glucose [4].Therefore, although these findings would typically indicate a CNS infection, they may not distinguish patients with VRI from patients without infection [5], and has led to uncertainty about how to best diagnose VRI [6].
Molecular diagnostic techniques such as polymerase chain reaction (PCR) may be useful for the diagnosis of VRI.PCR uses DNA primers to amplify target genes, for example the highly preserved 16S rRNA gene in bacterial species [7].The amplification process produces millions of copies of the target gene.Studies comparing PCR to conventional diagnostic parameters suggest PCR may detect bacterial DNA in the CSF of patients with negative CSF culture [8], and may outperform standard clinical and laboratory markers of CNS infection [9].Therefore, because non-microbiological markers of infection are difficult to interpret in the setting of acute neurological injury, PCR may be a more time-efficient microbiological technique when compared to CSF culture.
We performed a prospective observational cohort study to measure the incidence of VRI as defined as a positive 16S rRNA PCR in patients with subarachnoid haemorrhage (SAH), traumatic brain injury (TBI), and intracerebral haemorrhage (ICH), who required placement of an EVD.We also compared the incidence of VRI as diagnosed by 16S rRNA PCR to pre-published definitions of VRI and describe diagnostic parameters of VRI in this cohort.

Methods
This manuscript has been structured according to the STROBE guidelines [10].Ethics approval for this study was granted by the Northern Sydney Local Health District Human Research Ethics Committee (RESP/18/099).Funding for the 16S rRNA PCR testing was provided by a grant from the Northcare foundation.

Study design and setting
This prospective observational cohort study was conducted in the Neurosciences Intensive Care Unit (ICU) at Royal North Shore Hospital, a tertiary referral centre in Sydney, Australia.This is a 13 bed ICU which specialises in the management of neurosurgical patients.Patient recruitment occurred from August 2018 until June 2019.Due to the COVID-19 outbreak and consequent redirection of laboratory resources, sample analysis was delayed until 2023.

Participants
Patients admitted to the ICU with a primary diagnosis of SAH, TBI, or ICH, who required placement of an EVD for their primary diagnosis, and had at least one CSF sample taken, were eligible to consent for this study.Patients were excluded if they had evidence of CNS infection prior to EVD insertion.If the eligible patient was unable to be consented, a person responsible was approached for consent.Patients were followed until 2 weeks after removal of their EVD.

Variables and data sources
Variables collected for this study included demographics, primary neurological diagnosis, illness severity scoring [11][12][13][14][15], and daily clinical and laboratory data.16S rRNA PCR was performed in batches by the RNSH microbiology lab on stored CSF samples taken as part of routine clinical practiceat this centre, CSF samples were taken twice a week, and when clinically indicated.An in-house 16S rRNA in vitro diagnostics method has been validated for use at RNSH New South Wales Health Pathology National Association of Testing Authorities accredited laboratory, using the MagNA Pure96 (MP96) Extraction Platform and Thermal Cycler BioRad CT1000 Touch.
A volume of 200 µL CSF drain samples was used for extraction on the Roche MagNA Pure automated extractor.The CSF drain samples were run in duplicate with a spiked control to monitor for inhibition to ensure no false negative results were issued.An in-house positive control, negative control and no-template control were included in each run.Two set of primers were used in this protocol, and each patient had two sets of master mix.Primer sets 1 and 2 were (forward) 5′-GAC TCC TAC GGG AGG CAG CAG-3′ and (reverse) 5′-CTG ATC CGC GAT TAC TAG CGA TTC-3′.Primer sets 1 and 2 generate 1020 base pairs [16].Primer sets 3 and 4 were (forward) 5′-CCT AAC ACA TGC AAG TCG ARCG-3′ and (reverse) 5′-CGT ATT ACC GCG GCT GCT-3′.Primer sets 3 and 4 generate 490 base pairs [17,18].5 µL of extracted DNA samples was mixed with 20 µL of master mix and amplified in the thermal cycler.The obtained amplicon was purified by using a Qiaquick PCR Purification Kit, sequenced with the Applied Biosystems Hitachi 3500 Genetic Analyzer with Big Dye protocol, and compared to a public DNA database [19].When multiple CSF samples were taken in a day, the first CSF sample was included in analysis.Samples were frozen at − 80 degrees Celsius between collection and PCR analysis.
Clinical data from eligible patients were prospectively collected in a registry called the Neurological Outcomes in Intensive CarE (NOICE) registry [20], which was accessed for this study.Data not included in NOICE such as CSF cytology and biochemistry were extracted separately from the electronic medical record.
Four pre-published definitions [21][22][23][24] were selected to measure the incidence of VRI.These definitions are shown in Supplementary Table 1.The Centre for Disease Control/National Healthcare Safety Network (CDC/NHSN) criteria [22] contains subjective criteria (i.e.meningeal signs, headache) which were excluded from the definition used for this study, because these findings may not be reliably reported in an intubated or sedated patient.Antibody titres were excluded as they are not routinely measured at this site.Normal ranges of CSF criteria were defined according to the Royal College of Pathologists of Australasia [25].Treatment of VRI was extracted from NOICE, and was defined as initiation of antibiotic therapy for VRI based on the clinical team's judgement.The clinical teams involved in the care of the patient were not involved in this study.

Statistical methods
Demographics data are presented as mean and standard deviation for normally distributed data, and median and interquartile interval for non-normally distributed data.Nominal data, such as the incidence of VRI, is presented as counts and proportions.For each definition of VRI, the diagnostic accuracy was calculated both by patient and by CSF sample.For the latter, diagnostic accuracy was calculated only on samples where 16 s rRNA PCR was done AND samples where it was possible to assess the presence or absence of VRI based on available data.Receiver operating characteristic (ROC) analysis was used to obtain the measures of diagnostic accuracy (sensitivity, specificity, area under the ROC (AuROC) curve, and percentage of correctly classified patients/CSF samples) of the four definitions of VRI against the reference, 16 s rRNA PCR.Diagnostic markers for VRI were presented as mean and standard deviation, and as proportions of CSF samples with diagnostic markers between specific thresholds.16S rRNA PCR is not a test routinely performed at RNSH, so sample size was limited by funding provided from a grant from the Northcare Foundation.Statistical analyses were conducted using Stata BE V17.0 for Windows (StataCorp LLC, College Station, TX).

Participants
There were 42 patients admitted to ICU who met inclusion criteria during the enrolment period, and 39 patients consented to be included S. Chadwick et al. in the study.16 s rRNA PCR was performed on 237 CSF samples.Demographics data for enrolled patients is included in Table 1.

The incidence of VRI
The incidence of VRI as measured by 16S rRNA PCR, clinicianinitiated treatment of VRI, and four pre-published definitions of VRI, is described in Table 2.
The clinical team initiated treatment for VRI in 11 of 39 patients (28.2 %).VRI as defined as a positive 16S rRNA PCR was diagnosed in 2.6 % of patients (1 of 39 patients).The incidence of VRI according to the pre-published definitions ranged from 2.6 % of patients to 41 % of patients.The incidence of VRI according to diagnostic criteria relying on CSF culture results [21,23] was closest to the incidence of VRI as defined as a positive 16S rRNA PCR.Definitions incorporating nonmicrobiological criteria [22,24] measured a higher incidence of VRI compared to 16S rRNA PCR.

The accuracy of published definitions of VRI
Measures of diagnostic utility for pre-published definitions of VRI when compared to 16S rRNA PCR are described in

Diagnostic parameters of VRI
Diagnostic parameters of VRI in CSF samples with a negative 16S rRNA PCR are described in Table 5.One CSF sample with a positive CSF culture was PCR negative.A fever of 37.6 • C or greater was measured on the same day as 61.1 % of CSF samples.The red cell count to white cell count ratio (RCC:WCC) in 202 PCR negative CSF samples was 200 or less and 47 % of samples.An elevated protein was measured in 64 % of CSF samples, and a CSF:blood glucose ratio was less than 0.5 in 25.9 % of CSF samples.Of 32 CSF samples in which a lactate was measured, lactate was greater than 2.0 mmol/L in 81.2 % of samples.

Conclusions
We performed a prospective cohort study to measure the incidence of VRI as defined as a positive 16 s rRNA PCR in patients admitted to ICU with a diagnosis of SAH, TBI or ICH.We found that the incidence of VRI as defined as a positive PCR was less than the incidence of VRI diagnosed by the treating team.The incidence of VRI according to pre-published definitions ranged from 2.6 % to 41 % of patients.A positive 16S rRNA PCR was best predicted by diagnostic criteria which required a positive CSF culture.The diagnostic parameters described in Tables 3 and 4 should be interpreted with caution due to the low incidence of VRI in this cohort.We also found that high numbers of CSF samples with a   Variations in the incidence of VRI based on different definitions is a well described phenomenon.One systematic review [6] found 17 different definitions of VRI, and studies which compare the incidence of VRI based on different definitions find different results based on the diagnostic criteria in use [6,26,27].In part, this stems from the absence of a validated gold standard diagnostic tool for VRI.For example, CSF cultures have been reported to have a false negative rate of between 20-70 % [28][29][30][31], which is further confounded by the use of antibiotic impregnated EVDs [32].Molecular microbiological technologies such as PCR or metagenomic sequencing may have potential as gold standard diagnostic tools, but their use in clinical practice has yet to be validated in a large enough cohort.
Although PCR has been purported to diagnose culture negative VRI [8], there were no cases of culture negative VRI in this study.Whilst only one patient had a positive 16S rRNA PCR result, 11 patients were treated for VRI by the clinical team, and 16 patients met the definition for VRI as published by Holloway et al. [24], which allowed for culture negative VRI.On the other hand, a positive PCR was best predicted by the definition which relied exclusively on a positive CSF culture taken after the first 24 h of EVD placement [21].This is reflected by the high specificity and low false positive rates in the definitions relying on CSF culture results.
The compromised diagnostic utility of definitions which use nonmicrobiological markers is consistent with previous studies demonstrating limited utility of these markers of CNS infection for the diagnosis of VRI [4,5].CSF samples with a negative PCR had a high incidence of results commensurate with CNS infection, such as increased white cells, decreased RCC:WCC ratio, or increased protein.These samples were also associated with a high incidence of fever on the day the sample was taken.On the other hand, the incidence of positive gram stain or culture was low in these samples.In this study, the single PCR-negative CSF sample with a positive culture was taken at the time of EVD insertion.Both PCR and repeat culture of the culprit sample were negative, suggesting that the positive culture was a contaminant.However, because the positive culture led to commencement of treatment for VRI in this patient, the culture has been considered positive in the study analysis.This accounts for the reduced diagnostic accuracy of the definition published by Gozal et al. [23].
The high incidence of clinical markers suggestive of infection, and the low incidence of positive gram stain or culture in PCR-negative CSF samples, indicates that a positive CSF culture is the most reliable marker of VRI.It also suggests that reliance on routine non-microbiological markers of CNS infection may lead to overdiagnosis of VRI.In a 2016 systematic review of definitions of VRI [6], 13 of 16 definitions included clinical or laboratory criteria, and 8 of 16 did not include a CSF culture result.The earliest definition in this systematic review is from a 1984 paper [33] which used prospective data to conclude that VRI should be diagnosed on the basis of a positive CSF culture due to the low predictive value of CSF white cell counts or fever.40 years later, our findings support this assertion.
The lack of a consensus definition, the variability demonstrated by different diagnostic criteria, and the lack of a gold standard diagnostic tool, have consequences on accuracy of diagnosis and treatment of VRI.Whilst CNS infections in patients presenting from the community have poor morbidity and mortality [34], the same has not been clearly demonstrated in patients with VRI [3].One explanation for this is that a patient admitted to ICU, who is at risk of VRI, will have more prompt recognition and treatment of CNS infection compared to an undifferentiated patient presenting from the community.This would reduce delays in treatment.Therefore, the impact of CNS infection on mortality and functional neurological outcome would be minimised.However, another explanation is that VRI is over diagnosed, and patients with aseptic CNS inflammation in the setting of severe neurologic pathology are being analysed in the same group as patients with true CNS infection.This would create a bias in analysis due to the formation of an unreliable denominator.Therefore, any studies examining the incidence, treatment, prevention, or prognosis of patients with VRI are more difficult to design or interpret.Moreover, these diagnostic uncertainties have direct effects on patient care.Overdiagnosis of VRI will necessarily lead to overtreatment with broad spectrum antibiotics, exposing a vulnerable patient group to potential adverse reactions associated with these medications, and increasing the risk of antibiotics resistance in our communities.Overdiagnosis of VRI may also lead to delayed diagnosis and treatment of alternative pathology which explains the patient's clinical findings.
Although we acknowledge the small cohort size as a limitation of this study, this study represents a contribution to our understanding of the value of PCR in the diagnosis of VRI which is comparable to other similar studies [8,9,35,36].The four studies found during our literature review which used PCR on CSF samples only from EVDs included between 28 and 67 patients and between 24 and 86 CSF samples (our study included 39 patients and 237 CSF samples).Other limitations of this study include the low incidence of a positive 16S rRNA PCR (which makes the AUROC, sensitivity, and false negative rates in Tables 3 and 4 less reliable, although they are included for completion), and that the precise reasons for initiating treatment for VRI by the clinical teams in these patients was not available from the NOICE database.This study recruited patients from a specialised ICU in a major tertiary centre with a high neurosurgical patient load and nursing staff specifically trained in the management of patients with EVDs, so these findings may not be generalisable to centres with less exposure to patients with EVDs.Further research would involve multi-centre studies to detect more cases of VRI, and to increase the generalisability of the study.
In this prospective cohort study, the incidence of VRI as diagnosed by 16S rRNA PCR was 2.6 % of patients and 0.8 % of CSF samples.This is closest to the incidence measured by diagnostic criteria relying on a positive CSF culture.Diagnostic criteria for VRI using nonmicrobiological markers of CNS infection had a lower specificity and higher false positive rate compared to definitions relying on a positive CSF culture.The incidence of abnormal CSF results and fevers associated with CSF samples with a negative PCR was high.These findings suggest that the incidence of VRI is lower than currently estimated.

Statements and declarations
Funding for the 16S rRNA PCR was provided from a grant from the Northcare Foundation.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 3 (
[23]atient) and Table 4 (by sample).When compared to 16S rRNA PCR, the definitions by Chi et al.[21]and Gozal et al.[23], which relied on CSF culture results, correctly classified >99 % of patients and CSF samples.The definitions which included non-microbiological criteria were less accurate at predicting a positive 16S rRNA PCR.The definition published by the CDC/NHSN [22] correctly classified 79.49 % of patients and 96.17 % of CSF samples.The definition published by Holloway et al. [24] correctly classified 61.54 % of patients and 82.69 % of CSF samples.The definitions published by the CDC/NHSN and Holloway et al. had lower specificity and higher false positive rates than the definitions by Chi et al. and Gozal et al.

Table 1
Patient characteristics.

Table 2
Number of positive results from 16S rRNA polymerase chain reaction (PCR) test and existing definitions of ventriculostomy-related infection (VRI), by cerebrospinal fluid (CSF) sample and by patient.

Table 3
Measures of diagnostic accuracy for pre-existing definitions of ventriculostomy-related infection (VRI) when compared with 16S rRNA polymerase chain reaction (PCR) (by patient).Area Under Receiver Operating Characteristic; CDC = Centre for Disease Control; CSF = Cerebrospinal Fluid; NHSN = National Healthcare Safety Network; PCR = Polymerase Chain Reaction; VRI = Ventriculostomy-Related Infection

Table 4
Measures of diagnostic accuracy for pre-existing definitions of ventriculostomy-related infection (VRI) when compared with 16S rRNA polymerase chain reaction (PCR) (by CSF sample).Area Under Receiver Operating Characteristic; CDC = Centre for Disease Control; CSF = Cerebrospinal Fluid; NHSN = National Healthcare Safety Network; PCR = Polymerase Chain Reaction; VRI = Ventriculostomy-Related Infection.

Table 5
Analysis of diagnostic markers of ventriculostomy-related infection from cerebrospinal fluid (CSF) samples with a negative 16 s-rRNA polymerase chain reaction (PCR).