A new biomarker combination differentiates viral from bacterial infections and helps monitoring response to antibiotics in hospitalized children

Abstract

The development of new diagnostic techniques to support the initial clinical evaluation is one of the key strategies to mitigate the inappropriate antibiotic prescription in Pediatrics. We conducted a prospective observational study to test a novel chemiluminescence assay that helps differentiating bacterial and viral infections by combining three host-response proteins (C-reactive protein, TRAIL and IP-10). We enrolled 255 children (age 1-243 months) presenting to a Pediatric Infectious Diseases Unit with fever (60%), cough (22.4%), gastrointestinal symptoms (22%), proven infection (38%). The 22.7% of patients carried at least one comorbidity. The baseline test demonstrated 51% sensitivity and 91% specificity. The diagnostic accuracy was better than traditional biomarkers (C-reactive protein, white blood cell count). Stratifying the analysis by pre-admission antibiotic treatment, naïve patients showed higher sensitivity for bacterial infection (0.70 vs. 0.15), better negative predictive value (0.60 vs. 0.45) and lower error rate (0.24 vs. 0.51). A significant score reduction was observed shortly after starting antibiotic therapy in a subgroup with microbiologically confirmed bacterial diagnosis (mean − 45.12, mean interval = 4 days). We concluded the test could improve infection management in pediatric settings, support initial therapy decision and potentially reduce unnecessary antibiotics. However, its optimal implementation requires a precise definition of the target population.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-20208-1.

Background

The fight against antibiotic resistance is one of the main actions undertaken by the One Health initiative by the World Health Organization (WHO), which proposes an integrated approach encompassing the development of new diagnostic techniques to distinguish viral and bacterial infections¹. According to Messacar et al., the aim of Antibiotic Stewardship is to provide the right test for the right patient at the right time². The right interpretation is essential to promptly drive the prescription of appropriate antibiotic therapies and reduce the unnecessary costs.

Many antibiotic prescriptions for children with respiratory infections presenting with fever are inappropriate. This occurs both in primary care and hospital settings, where prescription is often empirical due to limitations in diagnostic tools for initial etiological definition^3,4.

Generally, the first laboratory assessment in Pediatric Infectious Diseases (PID) is based on white blood cell (WBC) count and inflammatory markers, including C-reactive protein (CRP) and procalcitonin (PCT). Alternative biomarkers such as presepsin and adrenomedullin have been proposed but their use is not widespread nor validated⁵. Most markers have major limitations in differentiating bacterial and viral etiology at an early phase of infection (e.g. low sensitivity or long time to positivity) and are not effective in limiting empirical antibiotic prescriptions. Indeed, current practice includes the revision of prescriptions as early as 48 h after the onset of symptoms or as soon as microbiological results are available.

Novel rapid tests could provide a reliable early decisional support, avoiding misinterpretations and reducing unnecessary antibiotics. Nonetheless, several prediction models have been proposed to rule out bacterial infections or rule in viral etiologies with high specificity⁶.

The LIAISON^® MeMed BV^® test is a novel automated chemiluminescence assay that differentiates bacterial and viral infections by measuring three blood proteins involved in the host immune response with a short run out time (less than 30 min)⁷. The ability of the test to differentiate viral from bacterial infections reflects the different dynamics of its three constituent immune proteins. Tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) participates in regulating programmed cell death, which is critical for the response to viral infections. It is rapidly induced in viral infections and reduced in bacterial infections. Interferon-inducible protein-10 (IP-10) is a small cytokine involved in chemotaxis and cell growth inhibition. It is increased in response to viral infections compared with bacterial infections. Finally, CRP is an established inflammatory marker induced in response to a wide pattern of inflammatory processes, including bacterial infections. The algorithm combining these three proteins can capture the immune response to infection mediated by multiple biological pathways, resulting in a robust and consistent readout of whether the host is responding to a bacterial versus a viral infection. The test showed good performance in differentiating bacterial from viral infections in multiple prospective validation studies and in different clinical settings^8,9, most of which -however- were settled in primary care. With a high negative predictive value, the score promises to be effective in reducing antibiotic overuse in viral infections^10,11. In addition, newer data suggest that TRAIL is correlated with clinical severity in children with respiratory tract infections¹². Furthermore, the test has been shown to be more accurate than PCR, PCT and routine laboratory parameters in discriminating between bacterial or viral infections in children with fever^13,14. The MeMed BV^® test showed a high sensitivity and specificity in distinguishing bacterial from viral LRTIs outperforming routine parameters, including WBC count, CRP, and chest X-ray¹⁵. Recently, a pilot study indicated that the novel score impacts patient management in real world settings, contributing to physician decision-making process in the management of patients with acute infection and suggesting its introduction in the workflow of febrile children¹⁶. It has been tested both in the adult and the pediatric age and has been demonstrated to be accurate and cost-effective in several studies settled in emergency care, potentially driving considerable savings in antibiotic consumption^17–19. However, the test has been used in primary care settings and limited data are available in pediatric populations hospitalized and in children admitted to units for infectious diseases.

Our aim was to test this new diagnostic instrument in a pediatric tertiary care center, assess its diagnostic performance according to the patients’ underlying characteristics and compare the results with the Standard of Care (SOC). The study was conducted at a University Children’s Hospital, thereby providing the opportunity for a thorough clinical and diagnostic evaluation of children at risk of infectious diseases.

Materials and methods

Study design and population

We conducted a prospective cohort study enrolling all patients consecutively hospitalized in the PID Unit of the Maternal and Child Department of Federico II University Hospital in Naples (Italy) for suspected acute infection from October 2022 to August 2024. Our cohort included patients with comorbidities (e.g. diabetes mellitus, autoinflammatory syndromes, rheumatological diseases in the active phase, CVC carriers, cystic fibrosis, hematologic malignancies etc.) as well as children with no comorbidities or risk factors. This was an observational study, as all patients were treated according to the SOC, in line with GCP principles and national and international clinical guidelines.

Data collection

The baseline assessment included demographic data, clinical characteristics, the presence of comorbidities and pre-admission and post-admission antibiotic treatment. Laboratory assessment at admission included routine inflammatory markers (WBC count, CRP and PCT) and microbiological tests aimed at etiological definition. LIAISON^® MeMed BV^® test, hereafter referred to as the “host-response score”, was performed as an additional test at admission in all patients, for whom informed consent had been obtained. The results of the tests were made available to clinicians retrospectively at the end of the study. The host-response score did not interfere with diagnostic and management algorithm, as results were provided in a double blinded fashion after the discharge and were evaluated in a post-hoc analysis with other clinical results and antimicrobial prescriptions.

Discharge assessment included the formulation of clinical or etiological diagnosis and possible prescription of antimicrobial therapy accordingly. The term “reference diagnosis” reports the diagnosis at discharge from hospital based on global assessment including clinical biochemical and microbiological data as well as the response to treatment. It reflects the International Classification of Diseases, 11th edition (ICD-11). Labeling was discussed and agreed upon by the expert panel process described later. This is routinely done as part of medical work and teaching process at the University hospital. The host-response score results were not yet known and hence were not considered in this phase. The original dataset was created and managed using Microsoft Excel (released 16.91).

Laboratory methods

All serum intended for the host-response test execution were analyzed using the LIAISON^® MeMed BV^® test (DiaSorin).

The test quantifies the levels of TRAIL, IP-10 and CRP relying on an automated chemiluminescent immunoassay (CLIA). The algorithm provided with the kit assigns a score ranging from 1 to 100 by combining the results of the three individual analytes. The test includes five possible outcomes, which are based on the score:

• High likelihood of viral infection or other non-bacterial etiology (0 ≤ score ≤ 10).

• Moderate likelihood of viral infection or other non-bacterial etiology (10 < score < 35).

• Indeterminate (35 ≤ score ≤ 65).

• Moderate likelihood of bacterial infection or co-infection (65 < score < 90).

• High likelihood of bacterial infection or co-infection (90 ≤ score ≤ 100).

The analysis was performed on LIAISON^® XL platforms, ensuring a high level of automation throughout the process, from sample handling to result generation⁷.

Reference diagnosis assessment

Reference diagnosis, which represented the gold standard for the calculation of diagnostic performance of the new test, was determined with a panel diagnosis procedure²⁰. The expert consensus process involved three pediatricians with expertise in PID, who independently assigned etiologic diagnoses (bacterial, co-infection or viral infections) to each patient relying on clinical, laboratory, microbiological and radiological data and response to therapy. Co-infections were defined as combined viral and bacterial infections and did not include infections from only ≥ 2 distinct viral agents. The experts were blinded to their peers’ classifications. In case of discordances, an internal revision was made by a researcher with PID expertise, not involved with clinical management neither with data analysis for the present study. The host-response score result was not considered for the diagnosis as it was disclosed only after the discharge from hospital.

Statistical analysis

Primary outcome

The primary outcome was the diagnostic accuracy of the host-response score, which was evaluated by likelihood ratio (RV = sensitivity/1-specificity), negative likelihood ratio (-RV = 1-sensitivity/specificity), sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for bacterial infection. Diagnostic performance outcomes were estimated from the comparison between the reference diagnosis and the host-response score. Patients with indeterminate host-response score were described in the baseline population and then were excluded from the statistical analysis. However, a narrative description of clinical symptoms and diagnosis of these patients was provided in Fig. 1, illustrating the enrollment process.

Fig. 1.

Flowchart illustrating the process of enrollment and evaluation. Patients with suspected infection (either symptoms with acute onset or microbiological evidence) were enrolled and received the baseline assessment. Patients with an indeterminate host-response score were excluded from the primary outcome analysis. A subgroup of patients with confirmed bacterial infections was further evaluated in the antimicrobial response analysis, which involved repeated clinical and laboratory assessments and a follow-up host-response score. Abbreviations: TB, tuberculosis; MIS-C, Multisystem Inflammatory Syndrome in Children (MIS-C); EBV, Epstein-Barr virus; CMV, Cytomegalovirus; MPV, Metapneumovirus; AV, Adenovirus.

Secondary outcomes

Secondary outcomes were the variation of the score before and after the onset of antibiotic therapy and changes of diagnostic performance related to pre-admission antimicrobial therapy.

Power analysis

Sample size was estimated according to Akoglu et al.²¹. We assumed an expected sensitivity of 94%¹⁷, a type I error rate of 5%, and a desired margin of error of ± 10%. Based on these parameters, a minimum of 72 diseased subjects and 31 non-diseased subjects were required. Given an expected bacterial disease prevalence of 30%, which reflects the type of population hospitalized in the PID Unit where the study was settled, the total required sample size was approximately 240 participants.

Statistical methods

Statistical analysis and data visualization were performed using R with the RStudio interface (released 2024.04.1 + 748). Differences of categorical variables between groups were evaluated by the Chi-square test Comparisons of non-parametric continuous variables, such as the score and biomarkers values, were performed with Mann-Whitney test or with Kruskal–Wallis test (when considering between two or three groups, respectively). Modifications of the host-response score during antimicrobial therapy were analyzed with paired T-test. Changes in diagnostic performance related to different durations of pre-admission treatment were tested with linear regression model. The statistical significance level was set at p-value < 0.05.

Ethical statement

Data were collected in an anonymous dataset using Microsoft Excel (released 16.91). Personal and sensitive data were processed pursuant to European Regulation no. 679/2016, so-called GDPR and the Italian legislation currently in force on privacy. All biological samples intended for the host-response test execution were collected and stored until the time of analysis. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee for Biomedical Activities Campania 3, Naples, Italy (#113/2024 and #226/2021). Parents and children involved in the study provided written informed consent/assent.

Results

Study population

We enrolled 255 patients aged 1 month to 20 years (median = 35 months, interquartile range = 6-101.5 months) [Fig. 1; Table 1].

Table 1.

Baseline characteristics of the study population.

	N = 255
Sex, n (%)
Female	112 (43.9%)
Age in months, median (IQR)	35.0 (6.00–101.5)
Comorbidities, n (%)	58 (22.7%)
Hematological/oncological	12 (20.7%)
Hematological, non oncological	4 (6.9%)
Inherited disorder of metabolism	11 (19%)
Inflammatory/reumathologic	5 (8.6%)
Diabetes, other metabolic disorders Neurological	4 (6.9%) 8 (13.8%)
Other	14 (24.1%)
Clinical features, n (%)
Fever	154 (60.4%)
Proven infection	97 (38.0%)
Cough	57 (22.4%)
Diarrhea/vomiting	56 (22.0%)
Upper respiratory	12 (4.7%)
Systemic	20 (7.8%)
Lymphadenopathy	15 (5.9%)
Other	87 (34.1%)
Initial laboratory assessment, mean (SD)
CRP (mg/L)	33.9 (54.8)
PCT (ng/dL)	1.59 (5.99)
WBC count (n/mm3)	10,500 (8190)
Neutrophils (n/mm3)	5330 (5220)
Elevated inflammatory markers, n (%)
CRP > 20 mg/L	92 (36.0%)
PCT > 0.5 ng/dL	36 (24.0%)

Demographic, clinical, and baseline laboratory characteristics of the enrolled patients. Age is reported as median with interquartile range (IQR). Continuous variables (CRP, PCT, WBC count, neutrophils) are expressed as mean ± standard deviation (SD). Categorical variables are presented as absolute numbers with percentages in parentheses. Elevated inflammatory markers are defined as CRP > 20 mg/L and PCT > 0.5 ng/dL.

Most of the patients were hospitalized due to fever (n = 154, 60.4%). Other common symptoms at presentation were cough (n = 57, 22.4%) and gastrointestinal symptoms including diarrhea or vomiting (n = 56, 22%). Ninety-seven patients (38%) were hospitalized due to already proven infections (mainly COVID-19 diagnosed through positive antigenic swabs).

A total of 58 (22.7%) patients had chronic underlying medical conditions. Among comorbidities, the most common were hematological oncological conditions (n = 12, 20%), inherited disorders of metabolism (n = 11, 19%) and neurodevelopmental disturbances (n = 8, 14%).

The baseline laboratory investigations included CRP (mean = 33.9 mg/L, SD = 54.8 mg/L) and WBC count (mean = 10500 cells/mm³, SD = 8190 cells/mm³) in all patients. Procalcitonin was obtained in 145 patients, ranging from 0 to 52.7 ng/L (mean = 1.59 ng/L, SD 5.99 ng/L).

Overall diagnostic performance of the host-response score

The reference etiological diagnosis was consistent with bacterial infection in 97 cases (38%) and bacterial/viral co-infection in 71 cases (27.8%), whereas the presence of any bacterial agent was excluded in 87 cases which were classified as viral (34.1%). The diagnostic work-up, according to the SOC, included serology, microscopy and culture, single and multiplex PCR, and allowed to reach microbiologically confirmed diagnosis in 184 cases (72%).

The combination of host-response biomarkers produced a score consistent with a bacterial infection in 82 cases (32%) and viral in 141 cases (55%), significantly correlating with the etiology defined by the panelists (p < 0.001). In 32 patients the test had an indeterminate outcome, and according to the manufacturer’s guidelines these cases were excluded from further evaluations⁷. A descriptive analysis of these patients is provided [Figure 1].

The median score correlated with the reference diagnosis, resulting 52.03 and 48.81 for bacterial infections and co-infections, respectively, and 2.00 for viral infections (p < 0.001). There were 12 outliers for the score among viral infections: in 8 cases SARS-CoV-2 was detected, 5 of which had a strong inflammatory component due to underlying inflammatory diseases (n = 2) or associated diabetic ketoacidosis (n = 3).

The score distribution was consistent with bacterial infections and co-infections, revealing a “prevalence” of bacterial over viral infection when several pathogens coexisted [Figure 2].

Fig. 2.

Host-response score and single biomarkers according to the aetiology. The box plots illustrate the distribution of host-response score score (upper panel) and its components (lower panels) according to reference diagnosis. Abbreviations: ****, p-value < 0.00001; ns, p-value ≥ 0.05.

When comparing the group who received a viral reference diagnosis to those with a bacterial or co-infection reference diagnosis, both the host-response score and the single biomarkers showed a significant difference (Score: 16.65 ± 26.70 vs. 49.73 ± 37.3, p < 0.001; CRP (mg/l): 6.90 ± 9.02 vs. 48.34 ± 50.1, p < 0.001; TRAIL (pg/ml): 144.77 ± 97.30 vs. 69.60 ± 55.70, p < 0.001; IP-10 (pg/ml): 711.66 ± 695.1 vs. 426.40 ± 537.68, p < 0.001).

The classification by host-response score was compared to the reference diagnoses, showing a consistent distribution (p < 0.001) [Supplementary Table 1]. In our cohort, an antibiotic prescription coincided with a reference diagnosis of bacterial infection or co-infection except for one patient who had Clostridioides difficile infection with positive toxins and was not treated with antibiotics as not having clinically relevant bowel function alterations.

The results of diagnostic performance analysis is summarized in Fig. 3. The overall accuracy in our sample was 0.659 (95% CI 0.593–0.721), with 51% sensitivity, 91% specificity, LR + 5.98, LR- 0.53 [Figure 3a]. In comparison with the routine biomarkers, the ROC curve demonstrated higher AUC with respect to isolated CRP and neutrophil count. Procalcitonin was not considered in this comparison because of missing data [Figure 3b].

Fig. 3.

Diagnostic performance of the host-response score in the entire population. Diagnostic performance indicators calculated by concordance with reference diagnosis and their 95% confidence intervals (Fig. 3a). Comparison of Receiver Operating Characteristic curve for the host-response score and the routine biomarkers (Fig. 3b).

Effect of antibiotics on the host-reponse score

In our cohort, 83 patients had already received antimicrobial therapies before admission to the PID Unit.

In most cases, antibiotics were started 24 to 72 h (mean pre-admission therapy length 3.25 days, SD 2.28) before admission. The frequency of pre-admission antibiotic therapy length according to the reference diagnosis and the host-response score is summarized in Table 2.

Table 2.

Frequency of antibiotic before hospital admission according to reference diagnosis at discharge and score result.

Pre-admission antibiotics	Bacterial Etiology		Viral Etiology		Total
	V score (n = 69)	B score (n = 75)	V score (n = 72)	B score (n = 7)	V score (n = 141)	B score (n = 82)
No	28 (40.6%)	41 (54.7%)	65 (90.3%)	6 (85.7%)	93 (66.0%)	47 (57.3%)
Yes	41 (59.4%)	34 (45.3%)	7 (9.7%)	1 (14.3%)	48 (34.0%)	35 (42.7%)
≤ 24 h	7 (10.1%)	10 (13.3%)	3 (4.2%)	0 (0%)	10 (7.1%)	10 (12.2%)
24–72 h	18 (26.1%)	13 (17.3%)	3 (4.2%)	0 (0%)	21 (14.9%)	13 (15.9%)
> 72 h	16 (23.2%)	11 (14.7%)	1 (1.4%)	1 (14.3%)	17 (12.1%)	12 (14.6%)

Data are shown as number of patients with percentage in parentheses. Aetiology was classified as bacterial or viral, and further stratified based on diagnostic scores (V score = viral score; B score = bacterial score).

The host-response score demonstrated a different distribution relating with the reference diagnosis when comparing the overall population with patients who had not received antimicrobial therapies before admission and baseline assessment. In fact, the naïve subgroup exhibited a reduction in “false negatives” scores in patients classified as co-infections, increased median scores for both bacterial infections and co-infections, and a generally decreased proportion of indeterminate results [Fig. 4a and b].

Fig. 4.

Impact of pre-admission antimicrobial therapy on the host-response score. The density plots depicted in the upper panels show the distribution of the score according to the reference diagnosis in the overall population (4a) and in naïve patients (4b). The density reported on the y axis quantifies the frequency of single scores corresponding to the score value reported of the x axis. The colored dots represent the score of individual patients; the colored vertical lines represent the median score of patients grouped by reference diagnosis. The lower panel represents the variation of diagnostic performance indicators by the administration of pre-admission antibiotic treatment (4c). The colored error bar represents 95% confidence interval, while black squares represent the punctual estimates. Abbreviations: NPV, negative predictive value; PPV, positive predictive value.

Significant variations were found in diagnostic performance according to prior antimicrobial treatment. When comparing the group of patients who were naïve to antimicrobials with those who received antibiotic treatment before the baseline assessment, the test showed a higher sensitivity (0.699 vs. 0.146), as well as better negative predictive value (0.594 vs. 0.453) and lower error rate (0.243 vs. 0.506) compared to the whole population. However, specificity was lower in the naïve group (0.872 vs. 0.971) [Figure 4c].

Further analysis using linear regression modeling showed a progressive decline in sensitivity in groups with increasing durations of pre-admission antimicrobial therapy, with a significant negative angular coefficient (β = −0.22, p = 0.022) [Table 3]. Comprehensive data on diagnostic performance stratified by antimicrobial pre-admission treatment length is reported in Supplementary Table 2.

Table 3.

Linear regression analysis of sensitivity of the host-response score according to pre-admission antibiotic therapy length.

	Sensitivity	95% CI	β	95% CI	p-value
No	0.699	0.595–0.790	−0.22	−0.37−0.08	0.022*
Yes
≤ 24 h	0.300	0.067–0.652
24–72 h	0.143	0.030–0.363
> 72 h	0.059	0.001–0.287

Regression coefficient (β) reflects the association between pre-admission antibiotic use and diagnostic sensitivity. Sensitivity and β coefficient are reported with 95% confidence intervals (CI).

Patients who didn’t receive antibiotics before admission serve as the reference category.

* = statistical significance at p < 0.05.

In a subgroup of 16 patients who had not previously received antimicrobials, but for whom bacterial diagnosis was eventually established with bacterial isolation, a paired comparison of the host-response score was performed before and at least 24 h after antimicrobial therapy onset, revealing a sharp and rapid reduction (score = 75.125 vs. 42.45; mean − 45.12, 95% CI −62.85, −27.4; p < 0.0001) [Figure 5]. The mean reduction per day of treatment was 16.10% for the host-response score and 13.20% CRP levels (mean T0-T1 interval = 4 days).

Fig. 5.

Paired comparison of the host-response score before and after the onset of antimicrobial therapy. Abbreviations: T0, baseline score determination; T1, score determination during antimicrobial therapy; ****, p-value < 0.00001.

Discussion

In the latest years many new diagnostic approaches including new biomarkers of infection, molecular syndromic testing and fast antimicrobial susceptibility testing are being proposed to early distinguish viral from bacterial etiologies of infections and optimize antimicrobial therapies. Among biomarkers, combinations of proteins deriving from the host immune response have the advantage of relying on affordable laboratory techniques and being versatile as they measure nonspecific immunologic and inflammatory responses. By exploiting the combination of different biomarkers, it is possible to achieve a higher diagnostic power than single inflammatory markers (e.g. CRP, procalcitonin)²².

The host-response score generated by combined quantification of CRP, IP-10 and TRAIL (LIAISON^® MeMed BV^®) has been tested in pediatric cohorts showing adequate diagnostic performance in differentiating bacterial and viral causes of acute inflammation and has considerable potential in modifying clinicians’ intention to prescribe antibiotics. However, available evidence has mainly been produced in primary care settings mainly in emergency departments, in children with respiratory conditions^11,23.

Our study was conducted in a pediatric tertiary care ward (specifically in a PID Unit), in a variety of acute clinical conditions at admission and also included children with underlying chronic diseases. In this setting the clinical work up is far more accurate than that in primary setting or emergency and in fact, many novel aspects of the test have emerged in comparison with prior studies. Above all, we reported the high proportion of patients who had already received antimicrobial therapies before admission, mainly because of national health system organization, and its impact on the host-response score results. Indeed, as the hospitalization in tertiary care facilities usually follows the evaluation and medical decisions either by primary care physicians, emergency departments or secondary care hospitals, antimicrobial therapies were often undertaken before enrollment in the present study. This, along with the possibility to repeat the test during the disease, enabled us to explore the effect of antibiotics on the test sensitivity and its progressive change in response to therapy.

In a first analysis of diagnostic performance in the overall sample, we encountered suboptimal sensitivity along with high specificity and positive predictive value. Likelihood ratios were quite reliable taking into account the cut-off values reported in reference literature²⁴, and the overall performance was higher than single routine biomarkers. On the other hand, negative predictive value, being equal to 0.51, was not adequately high to rule out a bacterial infection with sufficient reliability. This, along with a low overall sensitivity, is one of the main limitations of our sample of population which had a high degree of heterogeneity in terms of acute conditions, individual risk factors and exposure to pharmacological treatments. For instance, layering the analysis by length of pre-admission antimicrobial treatment, we showed an increase in sensitivity to 0.70 and NPV to 0.60 when considering only naive patients, and a progressive decrease with longer time of antibiotic administration. Considering these results, our study could serve as a starter for exploring new applications of these diagnostic tests depending on the setting, the site and the specific etiology of the infection. In the emergency care they could help in the initial diagnostic approach ruling out bacterial infections due to its high negative predictive value, as previously described¹¹. In pediatric wards, it could be implemented as a driver for therapeutic management, being a marker of appropriateness of antimicrobial therapies for its positive predictive value and proving their effectiveness due to rapid response to treatment.

Unlike other studies, children with comorbidities and underlying conditions, including possible causes of immunosuppression and dysregulated inflammatory response, were not excluded. Underlying high risk conditions could have led the panel’s clinicians to overestimate bacterial reference diagnosis and prescribe antibiotics more easily. Although showing lower inflammatory markers at the initial assessment, these group of patients had a higher antibiotic prescription rate both before and during hospitalization rate compared to patients without underlying chronic clinical conditions. These data are beyond the aim of this study, and therefore they are not detailed in the results section. Further research could investigate the role of modifications of the single proteins included in the score in different chronic diseases.

We found some similarities with other studies conducted on the same score system. The rate of indeterminate results was 12.5% in our cohort. In other reports it was between 8 and 13% and these cases were usually excluded from the analysis as in our report, carrying the risk of overestimating diagnostic performance^11,23. However, about half of these patients received antibiotics before admission, which possibly further affected the correct detection of a bacterial etiology.

Moreover, previous studies reported that infections sustained by intracellular pathogens, such as Mycoplasma pneumoniae, can be misclassified as viral infections^25,26. This could have impacted on our results considering that a cluster of Bordetella pertussis occurred during the enrollment and several pneumonia cases of Mycoplasma pneumoniae that were included in our cohort. The analysis of individual pathogens either viral or bacterial is beyond the aim of the present paper.

Finally, the potential economic benefits from the inclusion of this score in the SOC should be cited. Despite we did not conduct financial analysis in the present study, it is reasonable to expect that in most cases it could have allowed the sparing of antibiotic therapies and multiple microbiological diagnostic tests to investigate specific viral etiologies in order to rule out a bacterial diagnosis. Some published cost-impact studies settled in different health systems analyze these aspects, showing significant savings per patient when adopting this score in the diagnostic work-up compared to the SOC^27,28.

Conclusion

This study represents one of the first applications of a novel host-response score in a tertiary PID unit, encompassing a heterogeneous population with a wide spectrum of symptoms and underlying chronic conditions. The overall analysis showed better performance over single traditional biomarkers, while we observed a relevant reduction in sensitivity among patients who had already received antibiotics prior to admission. Our results are consistent with evidence produced in emergency departments, highlighting that the test is useful in identifying patients with viral infections for whom antibiotic therapy is not indicated. Furthermore, our study showed the impact of prior antibiotic exposure on the test performance, highlighting the importance of better defining the target population in advance to proceed with optimal implementation in real-world settings. The test yielded potential for guiding antimicrobial therapy in pediatric wards, especially when considering naïve patients. Finally, its faster decline compared to CRP during antibiotic treatment suggests it may serve not only in initial diagnosis but also as a potential marker of effectiveness.

These results need to be validated in larger populations, and the pathogenetic effects of chronic inflammatory conditions on biomarker expression should be studied in different phases of infections. In conclusion, the score could be a valuable tool in infection diagnosis, providing a reliable support to distinguish viral from bacterial etiologies and integrating the bundle of interventions for antibiotic stewardship in pediatric care. However, its optimal implementation in different real-world settings requires a precise definition of the target population.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, A.G., G.P. and E.B.; methodology, A.G., E.B., D.B. and F.P.; formal analysis, F.P., E.B.; investigation, M.P., V.C. and F.P.; data curation, F.P., S.B., G.A.; statistical analysis, D.B. and F.P.; writing—original draft preparation, F.P. and S.B.; writing—review and editing, A.G., G.P., E.B. and F.P.; data visualization, F.P.; supervision E.B. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by EU funding within the NextGeneration EU-MUR PNRR Extended Partnership initiative on Emerging Infectious Diseases (Project no. PE00000007, INF-ACT).

Data availability

The dataset used for the study is made available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

This observational study was approved by the Ethics Committee for Biomedical Activities Campania 3, Naples, Italy (#113/2024). The study was conducted in accordance with the Declaration of Helsinki. Parents and children involved in the study provided written informed consent/assent.