Cell-free DNA as diagnostic and prognostic biomarkers for adult sepsis: a systematic review and meta-analysis

Although cell-free DNA (cfDNA) is an emerging sepsis biomarker, the use of cfDNA, especially as diagnostic and prognostic indicators, has surprisingly not been systemically analyzed. Data of adult patients with sepsis that conducted cfDNA measurement within 24 h of the admission was collected from PubMed, ScienceDirect, Scopus, and Cochrane Library until October 2022. The Quality in Prognosis Studies (QUIPS) and Quality Assessment in Diagnostic Studies-2 (QUADAS-2) tools were used to reduce the risk of biased assessment. The mean difference (MD) of cfDNA concentration and the standardized mean difference (SMD) between populations was calculated using Review Manager (RevMan) version 5.4.1 package software. Pooled analysis from 18 included studies demonstrated increased serum cfDNA levels in sepsis when compared with healthy control (SMD = 1.02; 95% confidence interval (CI) 0.46–1.57) or non-sepsis patients in the intensive care unit (ICU) (SMD = 1.03; 95% CI 0.65–1.40), respectively. Meanwhile, a slight decrease in the statistical value was observed when compared with non-sepsis ICU patients with SIRS (SMD = 0.74; 95% 0.41–1.06). The lower cfDNA levels were also observed in sepsis survivors compared to the non-survivors (SMD at 1.43; 95%CI 0.69–2.17) with the pooled area under the receiver operating characteristic curve (AUC) of 0.76 (95% CI 0.64–0.87) for the mortality prediction. Levels of cfDNA showed a pooled sensitivity of 0.81 (95% CI 0.75–0.86) and specificity of 0.72 (95% CI 0.65–0.78) with pooled diagnostic odd ratio (DOR) at 25.03 (95% CI 5.48–114.43) for the identification of sepsis in critically ill conditions. The cfDNA levels were significantly higher in patients with sepsis and being a helpful indicator for the critically ill conditions of sepsis. Nevertheless, results of the test must be interpreted carefully with the context of all clinical situations.


Eligibility criteria
The inclusion criteria were cohort human studies investigating diagnostic or prognostic accuracy of cfDNA in plasma or serum in sepsis of adults (aged ≥ 18 years old).Sepsis was defined as the standardized criteria, including Sepsis-1, Sepsis-2, or Sepsis-3, depending on the periods of publication 12 .Sepsis meant "life-threatening organ dysfunction caused by a dysregulated host response to infection, " according to the Sepsis-3 definition and severe sepsis form the Sepsis-1 and Sepsis-2.Positive blood culture alone was also identified as sepsis because of the requirement for urgent antimicrobial therapy in these patients to prevent a life-threatening complication.A quality determination of the DNA samples and the processes of cfDNA detections, including quantified polymerase chain reaction (qPCR), fluorescent-based method, or spectroscopy, were thoroughly assessed in all enrolled studies.The exclusion criteria were case reports, reviews, conference abstracts, and preprint articles.The potentially eligible articles were reviewed independently by 2 groups of reviewers to confirm their eligibility.Citation screening and selections were documented and summarized in a PRISMA-compliant flow chart.

Data extraction and analysis
The following data were extracted from the published articles as following: (i) general study information: author, year, country, study design, clinical setting; (ii) patient characteristics: sample size, age, gender proportion, sepsis definition, sepsis severity; (iii) biomarker measurement: time point of measurement and test method; (iv) mortality: follow-up duration and rate of mortality; (v) outcome measurement: biomarker concentration, the area under the receiver operating characteristic curve (AUC) for diagnosis and prediction of mortality with cut-off point together with sensitivity, and specificity.The data were recorded independently by two groups of authors on separate copies of the records, any discrepancies were determined by a consensus with the third group of reviewers.The Quality assessment was conducted using the Quality in Prognosis Studies (QUIPS) tool for prognostic studies 18 , while the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool for diagnostic studies 19 .
For pooling of the results, mean with standard deviations (SD) values were used for the calculations and the standard errors (SE) were calculated into SD using the Cochrane Collaboration formula (SD = SE x √ N ).For the values presented with median and range or interquartile range (IQR), the mean values and SD were estimated by statistical formula from Wan et al. 20 .Then, the results of the analysis were presented as the forest plots of pooled mean differences (MD) with 95%confidence interval.Statistical significance was defined at the p-value < 0.05.Heterogeneity was measured using the among-study variance (τ 2 ), χ 2 test, and I 2 statistical analyses.For the www.nature.com/scientificreports/measurement with an I 2 < 50%, the results were pooled using a fixed effects model, otherwise a random effects model was used.All statistical analyses were performed using Review Manager (Revman) 5.4.1 package software.

Study selection and characteristics
The systematic literature search retrieved 2742 articles.After initial screening by the title and abstract, 2688 articles were excluded.The full texts of the remaining 49 articles were examined and 21 studies were further excluded.Hence, 18 studies (17 prospective studies and 1 retrospective observation) with cfDNA analysis in adult blood samples at admission or at enrollment (published between 2006 and 2022) were included in the current metaanalysis (Fig. 1).There were nine and six studies from Europe and Eastern Asia, respectively, with two studies from North America and one study from Middle East Asia.The cfDNA detection methods, include ELISA (2 studies), spectroscopy (2 studies), fluorescent-based assay (6 studies), and qPCR (8 studies).The main characteristics of these studies are summarized in Table 1.A total of 1850 participants (healthy volunteers, patients with non-sepsis, and patients with sepsis) were enrolled with the number of participants across studies ranging from 3 to 221 cases.There were eight and four studies that examined the prognosis and diagnostic outcomes, respectively, and the remaining four studies simultaneously analyzed for both prognosis and diagnostic outcomes (Table 1).

Quality of enrolled studies
The QUIPS and QUADAS-2 were used to evaluate prognostic and diagnostic studies, respectively 18,19 , as illustrated in Fig. 2. In the QUIPS measurement, 12 included prognostic studies [21][22][23][25][26][27][28][31][32][33]36,38 with sepsis-related mortality were evaluated. Study participatin and outcome measurement bias was identified as a concern in > 50% of the included studies.Many studies failed to specify whether they followed consecutive or random enrollment and reported no exclusion criteria.For outcome measurement, the risk of bias occurred due to the failure to clarify the time of measurement for the outcomes.There was a high-risk concern in three studies due to records of the hospital or global mortality which possibly did not relate to sepsis (Fig. 2A).
Eight studies with diagnostic outcomes 24,26,30,31,33,34,36,37 were evaluated using QUADAS-2 measurement.For the index test domain of risk of bias, greater than 50% of studies were scored as unclear as these studies reported mean biomarker levels rather than AUC data, but cut-offs were not calculated.As well, it is also unclear if the index test was interpreted while blinded to the patient outcome.One study assigned a high risk for reference standard due to critically ill patients being included regardless of bacteremia, but all of the patients were not positive for blood culture.Four studies demonstrated an unclear risk of flow and timing bias due to the inappropriate timing of sample collection.Studies stated that biomarker was taken upon admission of study enrollment, but this specific time interval may vary for each patient and affect cfDNA levels studies.In this study, none of the studies had high concerns for applicability with respect to the reference standard (Fig. 2B).

Discussion
Sepsis remains an important global health-care problem that still needs better diagnostic and prognostic tests.Although the Surviving Sepsis Campaign acknowledges the possible value of the new biomarkers for sepsis management, the updated guidelines still have no recommendation on the use of any biomarkers for the prognosis prediction or sepsis diagnosis 12 .However, the delay in sepsis management (diagnosis and treatment) increases mortality, prolongs the length of hospital stay, and increases the costs of treatment, highlighting the need for reliable biomarkers for diagnosis of early sepsis and prognosis prediction 13,14 .For cfDNA, our study is the first comprehensive systematic review and meta-analysis which assess the performances of cfDNA as a potential diagnostic and prognostic biomarker for sepsis.Only the longitudinal studies were included to test their quality using QUIPS and QUADAS-2 tools 19 and our meta-analysis demonstrated the potential of cfDNA for mortality prediction and diagnosis of sepsis.Overall, these results indicate that cfDNA should be further investigated as a measurement to guide clinical evaluations in identifying sepsis and sepsis survival outcomes.The increased cfDNA in blood during sepsis is possibly released from various types of cell death (apoptosis and necrosis) or cell damage 40,41 , which are a pivotal role in the sepsis pathogenesis 42 .Then, the abundance of cfDNA might be a good indicator for sepsis-induced cell damage that theoretically be correlated with sepsis severity.Indeed, our meta-analysis identified a moderate certainty due to a moderate effect size for the differences in cfDNA levels within 24 h of sepsis.The cfDNA did not only increase in patients with sepsis compared with non-sepsis controls or SIRS (ICU cases), but cfDNA was also elevated in sepsis non-survivors when compared with sepsis survivors.Interestingly, cfDNA levels, measured even at the earliest stages in ICU or at admission (the possible closest time-point to the onset of sepsis), were able to predict mortality rate, as indicated by the pooled AUC for prediction at 0.76 (95%CI 0.64-0.87);an acceptable value for the clinic use 43 .Additionally, patients with initially high cfDNA at admission were also significantly associated with higher mortality rates than the patients with lower cfDNA 27,31 .For the discrimination between sepsis and ICU control with the combined AUC (0.80), pooled sensitivity (0.81), pooled specificity (0.72), and calculating DOR (25.03) indicated cfDNA as a good diagnostic biomarker for sepsis for the practical use 44,45 .However, in subgroup analysis between sepsis versus SIRS, the capacity of cfDNA for sepsis discrimination was decreased as represented by the reduced pooled AUC from 0.80 (sepsis vs. non-sepsis ICU) into 0.75 (sepsis vs. SIRS in ICU), supporting SIRS as an overlapping spectrum of sepsis with significant cell damage 46 .High cfDNA (compared with control) in patients with SIRS, despite an  undetectable pathogen, might be an early sign of rapid progression into sepsis after a short follow-up period 22,47 .
Similarly, low cfDNA levels in some patients with sepsis might be related to the transition from sepsis to the recovery phase.Considering an indicator of disease severity, cfDNA levels in severe sepsis were higher than the uncomplicated sepsis in a few publications 36 , while other studies could not observe a significant difference 30,36 .Some of this variability may be explained by biological differences between patients and/ or different sensitivity of the used assays among publications 48,49 .For the major difference between real-time qPCR and fluorescent-based assay for measurement, cfDNA was quantified using qPCR for the β-actin housekeeping gene, thus detecting a subset of cfDNA in nuclear DNA, but not mitochondrial DNA (mt-DNA) or microbial DNA.Meanwhile, the fluorescent-based assay detects all of cfDNA (nuclear DNA, mt-DNA, and microbial DNA) 49 showing a high certainly due to a low effect size (I 2 = 24%) for differences in cfDNA levels to detect severe subjects and demonstrating the highest SMD (subgroup analysis in Table 2).Although the tests for cfDNA are relatively simple and inexpensive, especially in fluorescent-based assay, the cfDNA measurement in routine clinical practice is still uncommon.The combination of cfDNA with the current sepsis scoring system may yield even stronger predictive power.Indeed, the current clinical scoring of sepsis, such as Acute Physiology and Chronic Health Evaluation (APACHE) II and Sequential Organ Failure Assessment (SOFA), exhibit only a moderate discriminative power for sepsis mortality prediction with that AUC at 0.6 to 0.7 50,51 .
The cfDNA in sepsis might be produced from cell damage from inadequate oxygenation (hypoxia) and cell death [52][53][54] in both parenchymal and immune cells, especially granulocytes and cells from several organs, as determined by the cfDNA methylation profile 55 , that might be different from patients with trauma 56 .While the sepsis-associated cfDNA is possibly primarily released by activated neutrophils due to the prominent production of polymorphonuclear cells (PMN) in response to pathogens, trauma-associated cfDNA mostly originate from the injured cells because of the less PMN responses against the damaged host molecules 56 .More profound neutrophil extracellular traps (NETs), the extracellular decondensed chromatin with antimicrobial molecules, in early sepsis compared with the trauma-induced SIRS 57 might be one of the main sources of sepsis-associated cfDNA 58 , supporting NETs dysregulation in several diseases [57][58][59] .Also, NETs are one of the mechanisms responsible for acute respiratory distress syndrome (ARDS), vascular damage, and microthrombi leading to multiorgan failure and death 59 that are common in sepsis.Indeed, the roles of cfDNA as a major crosslink between inflammation and coagulation, referred to as "immunothrombosis" 60,61 , partly through the induction of Toll like receptor-9 (TLR-9) 62,63 , and coagulation system by activation of primary and secondary hemostasis 64,65 are well-known.In animal studies, cfDNA-mediated severe sepsis is demonstrated as scavengers of cfDNA attenuate cfDNA-induced inflammation 39 and sepsis severity 66 supporting cfDNA as a key contribution in sepsis and the use of cfDNA as the biomarker or therapeutic target.Indeed, compared to other potential biomarkers for sepsis diagnosis such as procalcitonin (PCT), our findings exhibited that calculating the DOR of cfDNA was higher than PCT showed calculating DOR of 12.5 from meta-analysis, whereas pooled sensitivity and specificity were similar 67,68 .However, there are preanalytical considerations in sample collecting for cfDNA analyses as preservation and process of preparation might interfere with cfDNA levels 49 .Additionally, the kinetic of cfDNA in sepsis is still uncertain.The fetal cfDNA is rapidly cleared from maternal plasma after delivery 69 and increased cfDNA from hemodialysis returns to baseline within 30 min after stopping the session 70 , suggesting that cfDNA is eliminated through the reticuloendothelial system in the livers and spleens or clearance through glomerular infiltration in the kidneys 69 .Because alteration in blood levels of biomarkers with a long half-life is too slow to represent a real-time patient's situation 12 , the short half-life of cfDNA together with the capacity to induce inflammation results in the use of cfDNA as an indicator of cell damage in several conditions with immediate emergencies 71 .In our systemic review, the included studies 29,31,35,39 demonstrated that cfDNA measurement in sepsis within 24 h of admission was the highest value observed and also showed the highest property of mortality prediction.More studies on this topic are warranted.
Nevertheless, several limitations of the current study should be discussed.First, there are risks of bias regarding patient misclassification, partly due to variations in the reference standard and sepsis criteria in the included studies such as severe sepsis defined in the Sepsis-1 and 2, but not classified in the Sepsis-3.Some studies included patients who were clinically diagnosed with sepsis without microbiological evidence 12 .Second, there was a combination of the studies with different cfDNA measurement assays that could not suggest the difference among these assays.Third, our study was unable to conduct the ideal cut-off point of circulating cfDNA levels due to the limited of reporting on the raw data to map out the ROC curve.Fourth, there was no comparison between cfDNA and other sepsis biomarkers.Finally, only studies published in English were included.Despite these limitations, cfDNA was demonstrated as a good indicator for sepsis diagnosis and mortality prediction.Then, we proposed cfDNA as a part of an interesting multi-biomarker panel for early sepsis diagnosis and sepsis outcomes.Future studies of cfDNA in sepsis are interesting.

Conclusions
The comprehensive meta-analysis includes current studies to demonstrate the use of cfDNA levels as a biomarker in critically ill conditions of sepsis.In this study, cfDNA concentrations were significantly higher in sepsis than the control with the presentation of prognostic and diagnostic capacities.We encourage further studies in this area with the aforementioned guidelines together with the use of cfDNA.

Figure 1 .
Figure 1.Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) 2020 flow chart.

Figure 2 .
Figure 2. Quality assessment of included diagnostic studies, according to the QUIPS tool for prognostic studies and the QUADAS-2 tool for diagnostic studies.

Figure 3 .
Figure 3. Forest plots of Standardized mean difference (SMD) in cfDNA measurements.(A) SMD of cfDNA levels in patients with sepsis compared to controls, (B) SMD of cfDNA in severe sepsis compared to uncomplicated sepsis.

Figure 4 .
Figure 4. Forest plots of SMD in cfDNA measurements of sepsis survivors compared to non-survivors (A) controls and the pooled area under the receiver operating characteristic curve (AUC) for predicting sepsis related mortality (B).

Figure 5 .
Figure 5. Forest plots of sensitivity and specificity of cfDNA measurements (A) and summary receiver operating characteristic curve (SROC; black line = sepsis vs. total ICU control, blue line = sepsis vs. non-sepsis ICU control, red line = sepsis vs. ICU control with SIRS) (B) for sepsis diagnosis.Forest plots of the AUC for sepsis diagnosis (C).

Table 1 .
Characteristics of included studies.UK United Kingdom, US United State, ICU intensive care unit, ED emergency department, EICU emergency intensive care unit, ESICM European Society of Intensive Care Medicine, qPCR quantitative real-time polymerase chain reaction, ELISA enzyme-linked immunosorbent assay, SIRS systemic inflammatory response syndrome, NA not applicable.

Table 2 .
Subgroup analysis of cfDNA measurements of sepsis survivors compared to non-survivors.ICU intensive care unit, qPCR quantitative real-time polymerase chain reaction, NA not applicable.