Patients and data
This study was a single-center, retrospective review of all patients with CSF mNGS testing at Children’s Hospital of Hebei Province from April 5, 2019 to December 5, 2020. The Children’s Hospital is a large tertiary teaching hospital. Most of the patients were transferred from other hospitals. Patients were enrolled based on a particular exposure (i.e., suspected CNS infectious patients without pathogenic evidence and received empirical treatment (Fig. 1)). Patients who were older than 18 years old were excluded. The patients’ CSF was collected when a lumbar puncture was performed after admission. All of the patients’ details were obtained from medical records. Clinical details were obtained, including demographic information, symptoms, clinical diagnosis, C-reactive protein (CRP), complete blood count, examinations of cerebrospinal fluid (routine biochemical, culture, smear, PCR, antibody, mNGS), and treatment. Details about CSF mNGS test (BGI China) including results, interval (in days) from illness onset to CSF collection, lengths of empirical treatment (duration of antimicrobial use before the mNGS test), turnaround time (time from collection to return result), and clinical impacts (including new antimicrobial, antimicrobial de-escalation, no change) were evaluated. Clinically relevant organism identified from CSF mNGS was assessed relative to final overall diagnosis (CNS infection versus non-CNS infection). All information was collected by two independent researchers. Any discrepancies between the two researchers were resolved by consensus or, if necessary, a third-party clinician adjudicated by consulting the medical record. All of patients were classified into two groups in the final diagnosis: CNS infection group and non-CNS infection group. mNGS test was defined as positivity if it identified an organism. Otherwise, the mNGS test was defined as negative. In this way, CNS infection patients were divided into two groups (mNGS positive group and mNGS negative group). This retrospective study was approved by Children’s Hospital of Hebei Province ethics committee (Approval number: 2018014). A signed, written informed consent form was attained from the patient's guardian. All methods were performed in accordance with the relevant guidelines and regulations.
mNGS of CSF
CSF specimens (0.5ml) were collected from patients according to standard aseptic procedures, snap frozen, stored at −20◦C, and subjected to mNGS within 24h. The process of CSF mNGS consisted sample processing, nucleic acid (DNA and RNA) extraction, followed by library generation and bioinformatic pipeline analysis. After samples were received in the clinical laboratory, sample were processed by adding glass beads to CSF samples, followed by vigorous agitation. Then extracted DNA/ RNA was fragmented to yield 150-bp to 200-bp fragments, and DNA fragments were constructed through an end-repair method. Quality-controlled libraries were sequenced on a BGISEQ-500/50 platform (BGI-Tianjin, Tianjin, China). For each sample, an average of 20 million reads was obtained. High-quality sequencing data were generated by removing low-quality and short (length <35 bp) reads, after that computational subtraction of human host sequences were mapped to a human reference genome (YH sequences and hg19) using Burrows-Wheeler alignment. Nonhuman sequences were mapped to categorization reference databases downloaded from NCBI (ftp://ftp.ncbi.nlm.nih.gov/genomes/), which included the genome sequences of 3446 bacterial species (45 species of mycoplasma/chlamydia and 104 species of Mycobacterium tuberculosis), 206 fungal species, 1515 viral species, and 140 parasites connected to human diseases.
Diagnostic assessment of mNGS
Definitive clinical diagnoses of the participants who completed the study were adjudicated by retrospective. The diagnosis of CNS infection was done according to the diagnosis of CNS infections written by Singhi et al . The diagnosis of autoimmune encephalitis was carried out in accordance with the clinical diagnosis of autoimmune encephalitis published by Graus et al . The diagnosis of epilepsy was made in accordance with the clinical diagnostic criteria for epilepsy .
The clinical utility of mNGS was evaluated through the following steps. Firstly, clinical characteristics were compared between CNS infections patients with positive and negative mNGS groups. Second, to evaluate whether mNGS test had a clinical impact. Clinical impact was defined as the following conditions: 1) added new targeted antimicrobial, 2) de-escalation of antibiotic therapy, according mNGS test results. Finally, diagnostic performance of mNGS results was calculated on the basis of the following two approaches . In the absence of a gold standard for mNGS results, positive percent agreement (PPA: agreement between mNGS test and CNS infection diagnosis) and negative percent agreement (NPA: agreement between negative mNGS test and diagnosis of non-CNS infection) were reported instead of sensitivity and specificity. In addition, the proportion of true positives out all positive mNGS findings and proportion of true negatives out of all negative mNGS tests were described. mNGS tests often identified two or more organisms, which can be divided into two categories: clinically relevant and clinically irrelevant organisms depending on whether the organism had been recognized as causative agents. One method we counted one CSF sample as one test. An organism in one CSF sample test was clinically relevant, mNGS test was defined as true positive. Another method was adopted: each organism identified was counted and assessed independently. This method provides more detail for mNGS findings by separately assessing each organism.
For categorical data, Fisher’s exact test or chi-square test was used as appropriate. Continuous data were compared by Mann–Whitney U test or Student's t-test. For baseline characteristics and CSF laboratory tests, continuous variants were described by means when they conformed to the Kolmogorov–Smirnov test and by medians when not. A p value<0.05 was considered significantly, and all tests were 2-tailed. All statistical analyses were performed using SPSS version 26 (IBM SPSS Statistics for Windows, Armonk, NY).