The Association of the MeMed BV Test With Radiographic Pneumonia in Children Free

We performed a secondary analysis of AutoPilot-Dx, a multicenter observational prospective cohort study. Methods of this study are described elsewhere,¹⁹  with relevant aspects described here. This study was approved by the investigator’s Institutional Review Board and adhered to the Strengthening the Reporting of Observational studies in Epidemiology guidelines for reporting of observational data.²⁴ 

Children were recruited from 2 university hospitals’ EDs, 1 in Germany and 1 in Italy, irrespective of their ED disposition (admission or discharge). Consecutively enrolled eligible patients were between 90 days to 18 years of age, had an illness duration of ≤7 days, and had either a respiratory tract infection or a fever without source were approached for consent for participation, which included the collection of a blood sample (irrespective of whether blood testing was ordered by the treating clinician). Fever was defined as temperature ≥38.0°C at home or in the ED. Patients with a recent febrile episode (within 2 weeks), antibiotic use of more than 48 hours, HIV or hepatitis B or C virus infections, primary immunodeficiencies, active malignancies, severe developmental delay, and severe congenital metabolic disorders, or who were taking immunosuppressive or immunomodulatory treatment were excluded.¹⁹  For the present analysis, we limited the sample to include only those who had a chest radiograph performed as a part of their testing and who had imaging available for radiologist interpretation.

Our primary outcome was radiographic pneumonia. We used this definition as chest radiographs are frequently obtained in the acute care setting to determine if a child has pneumonia,²⁵  and because, although radiography alone cannot differentiate between viral and bacterial etiology, radiographic pneumonia is frequently used as starting point for the initiation of therapy. For this study, each image was reviewed by board-certified pediatric radiologists who independently reviewed images, blinded to the original chest radiograph interpretation and all clinical data, and input data into a standardized data entry form. Chest radiographs were ascertained for the presence of any of the following: interstitial opacities, hyperinflation, atelectasis, alveolar infiltrates, and pleural effusions. Alveolar infiltrates were further defined as patchy versus lobar infiltrates and as focal, multifocal, or diffuse.^26,27  We defined radiographic pneumonia as an interpretation of infiltrate, with or without concomitant atelectasis. Our secondary outcomes included hospitalization, length of stay, ICU admission, use of supplemental oxygen, performance of thoracocentesis, and use of antibiotics.

Our predictor of interest was the MeMed BV test. MeMed has classified the continuous BV score (which ranges from 0–100, based on the probability of bacterial infection) into 5 groups (hereafter referred to as BV test): high likelihood of viral infection or nonbacterial (BV score 0–10), moderate likelihood of viral infection or nonbacterial (11–34), equivocal (35–65), moderate likelihood of bacterial infection (66–89), and high likelihood of bacterial infection (90–100). Clinicians taking care of the patient did not have access to the BV results. We also considered other clinical and laboratory predictors of illness known to be associated with radiographic pneumonia that were present within the study dataset,^10,11,28,29  including age, oxygen saturation in room air, lowest oxygen saturation during hospitalization, maximum temperature, and fever duration.

We compared clinical, historical, laboratory, and radiographic factors among the sample stratified by the presence of radiographic pneumonia patients using the Fisher’s exact and Wilcoxon rank-sum tests, using the Benjamini-Hochberg method to correct for multiple comparisons. We evaluated the association of the MeMed BV category and other predictors for the presence of radiographic pneumonia using univariable logistic regression. We transformed continuous predictors with restricted cubic splines, with 5 knots selected using maximum likelihood estimation with the first and last knot set at the fifth and 95th percentiles. We elected to use a splined approach to our analysis as this would allow for the modeling of potentially nonlinear relationships between continuous predictors and our outcome of radiographic pneumonia.³⁰  Next, we constructed a multivariable model to evaluate if the MeMed BV category was associated with radiographic pneumonia after adjusting for known clinical risk factors, using a complete case approach. We expressed our results as odds ratios (OR) with 95% confidence intervals (CI) and as adjusted proportions. We described measures of diagnostic accuracy (sensitivity, specificity, positive and negative predictive value, positive and negative likelihood ratios) at each threshold of the BV test. We described differences in our clinical outcomes as numbers with percentages within each of the 5 BV test groups. As a sensitivity analysis, we repeated our multivariable analysis after multiple imputation by chained equations. Analysis was performed using the rms (version 6.3-0)³¹  package in R, version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria).

We evaluated models for radiographic pneumonia using the splined MeMed BV score, each of the 3 components used to calculate the BV score (CRP, TRAIL, and IP-10) and procalcitonin. We compared the OR and 95% CI for each, comparing the odds of having radiographic pneumonia among those with the first quartile relative to the third quartile of the continuous laboratory predictor in both univariable and multivariable models. As in the primary analysis, the multivariable models were fit separately to the biomarker with age, oxygen saturation on room air, maximum temperature, and fever duration. For each of these multivariable models, we calculated the area under the receiver operator characteristic curve (AUROC). We compared each model to the one constructed using the BV score by the paired DeLong test. As a posthoc analysis, we additionally constructed multivariable model including CRP and IP-10 (and the aforementioned clinical variables) and compared this to the model made using the BV score.

A total of 1008 children were enrolled in the parent study. A chest radiograph was obtained in 206 (20.4%) of these patients. Chest radiograph images were unavailable for 23 patients. Among the remainder, 1 was uninterpretable because of overexposure, leaving 182 children in the final study sample. The median age was 2.5 years (interquartile range [IQR] 1.2–4.9 years) and 58% were boys. The median BV score was 58 (IQR 10–94); this was lower among children without radiographic pneumonia (median 39, IQR 8–86) compared with those with radiographic pneumonia (median 85, IQR 34–99; P < .01).

Overall, 74 (41%) of children had infiltrates, meeting our outcome of radiographic pneumonia. One of these patients had features of both atelectasis and infiltrate. Among those with infiltrate, 55 (74.3%) had patchy infiltrates and 19 (25.7%) had lobar infiltrates. Of 74 patients with infiltrates, 42 (56.8%) were focal, 29 (39.2%) were multifocal, and 3 (4.1%) were diffuse. Among other chest radiograph features within the 182 analyzed patients, 121 (66.5%) had interstitial opacities, 13 (7.1%) had hyperinflation, and 9 (4.9%) had pleural effusions. Radiographs were normal in 27 (14.8%) patients.

Study sample characteristics among patients with and without radiographic pneumonia are provided in Table 1. Besides the BV score, variables that were different between children with and without radiographic pneumonia included oxygen saturation on room air, lowest oxygen saturation during hospitalization, duration of fever, CRP, TRAIL, and the presence of atelectasis and pleural effusions on radiography. There was no difference in IP-10 concentrations between children with and without pneumonia. Twenty-three children were classified as having a high likelihood of bacterial disease by the BV test but did not have radiographic pneumonia. Common diagnoses for these children included pneumonia (n = 7), upper respiratory tract infections (n = 3), and acute tonsillitis (n = 3).

The most common radiographic finding among children classified as having a high likelihood of viral infection on the BV test was interstitial opacities, whereas the most common finding among children with a high likelihood of bacterial infection on the BV test was infiltrate (Supplemental Table 6). Among those with an infiltrate, children classified as having a higher likelihood of viral infections on the BV test more frequently had patchy findings (92%) with lobar findings occurring infrequently (8%), whereas those with a high likelihood of bacterial infection on the BV test had a lower frequency of patchy infiltrate (57%) and a higher frequency of lobar infiltrates (43%). Among children with an infiltrate, radiographic findings were similar among febrile patients with a moderate or high likelihood of viral infection, equivocal, or moderate likelihood of bacterial infection on the BV test (Supplemental Table 7). A similar finding was noted when stratifying children based on CRP into tertiles (Supplemental Table 8).

The association of univariable splined predictors for clinical and laboratory variables are presented in Figs 1 and 2, respectively. The univariable odds of radiographic pneumonia increased with rising BV likelihood of bacterial infection, though not uniformly (Table 2). When using the high likelihood of viral infection BV test as the referent, the OR of radiographic pneumonia was similar among those with a moderate likelihood of viral infection, equivocal risk, and moderate likelihood of bacterial infection, with point estimates ranging from 1.04 to 2.02. Those with a high likelihood of bacterial infection had higher odds of radiographic pneumonia (OR 4.31, 95% CI 1.86–10.01).

Complete data were available for 166 patients (91.2% of the total sample). In a multivariable model combining the BV category and the splined clinical predictors of age, oxygen saturation in room air, maximum temperature, and fever duration, the BV test group corresponding to a moderate likelihood of viral infection had similar performance to those classified as having a high likelihood of a viral infection. The odds of pneumonia in the equivocal and moderate likelihood of bacterial infection were similar (OR 9.62 [95% CI 1.94–47.81] and 7.81 [95% CI 1.84–33.23], respectively). The odds of radiographic pneumonia among children with of a high likelihood of bacterial infection were substantially higher (OR 23.61 [95% CI 6.30–88.6]). Confidence intervals overlapped across all BV test groups. Our findings were similar when using a multiple imputation approach. When evaluating measures of diagnostic accuracy, the sensitivity when using a moderate or high likelihood of bacterial infection was 60.8% and the specificity was 62.0% (Table 3). Sensitivity increased at lower thresholds, with corresponding declines in specificity.

Clinical outcomes among children within the study sample were similar across BV scores with respect to hospitalization, length of stay, and use of the ICU. Use of antibiotics was lower among those with a high likelihood of viral infection (37%) and increased progressively among children with higher likelihood of bacterial infections by BV score (Table 4). Use of oxygen was lower among those with a higher likelihood of bacterial infection by BV score. A subgroup analysis evaluating these clinical outcomes stratified by the presence of radiographic pneumonia demonstrated similar findings (Supplemental Tables 9 and 10).

The BV score was significantly associated with radiographic pneumonia in univariable and multivariable analyses (Table 5). CRP, TRAIL, IP-10, and procalcitonin did not have significant univariable associations with radiographic pneumonia but were associated with this outcome in multivariable analyses. The highest AUROC for the multivariable models was found with the BV score (0.87), followed by the TRAIL (0.86) and CRP (0.85), though 95% CI of the AUROC for each of these models overlapped. The performance of models using CRP, TRAIL, and procalcitonin were similar compared with the performance of the model using the BV score. A model which used IP-10 and CRP (but not TRAIL) had similar performance to the BV score (AUROC 0.85, 95% CI 0.79–0.91; P = .38).

We performed a secondary analysis of data from a prospective cohort study to evaluate the association between the BV assay and radiographic pneumonia in children. Children with equivocal, moderate, and high likelihoods of bacterial infection had higher odds of having radiographic pneumonia. Besides antibiotic use, the BV test was not meaningfully associated with clinical outcomes examined in this study.

Our findings corroborate prior work evaluating the performance of the BV test in children, which have focused on expert adjudication as the outcome measure. Prior work evaluating the MeMed BV test used expert adjudication as the reference etiologic standard. In the parent study by Papan et al, a direct comparison between children determined (by expert consensus) to have a bacterial or viral infection (after removing children with indeterminate etiology), the assay demonstrated a sensitivity of 93.7% and a specificity of 94.0%.¹⁹  We expand upon this work by performing a granular analysis of children with suspected pneumonia using an outcome of radiographic pneumonia instead of expert adjudication.

The BV test did not perfectly discriminate patients with and without radiographic pneumonia. Radiographic findings of infiltrate can be found in children with pneumonia regardless of the etiology of the infection (eg, viral or bacterial); this is important given the decline of suspected bacterial CAP after the introduction of pneumococcal vaccination in developed countries.³²  Prior work has suggested that radiographic findings poorly discriminate between the etiology of pneumonia,³³  though notably more children with lobar infiltrates were classified as having a high likelihood of bacterial infection in our study. Prior literature and national guidelines emphasize that chest radiography cannot be used to discriminate etiology of pneumonia.^34,35  As such, the assumption that all radiographic pneumonia is caused by bacterial infections, as described in some previous work, is lacking definitive evidence. In this study, approximately 28% of children with radiographic pneumonia had either a high or moderate likelihood of a viral infection based on the BV test.

The BV test in isolation may not be adequate to identify which children may be at the highest risk of radiographic disease, particularly as its performance was similar to CRP in isolation and procalcitonin. As such, more research is required to determine if the MeMed test offers greater potential in the prediction of radiographic pneumonia given its specificity to pathogen-based host response. In a study of adults and children presenting to an American ED where clinicians had access to the BV test results before clinical decision-making (30% had a diagnosis of pneumonia), 83% of patients with suspected bacterial disease by BV assay were given antibiotics, compared with 23% of patients with viral illness. These findings suggest that clinicians may be amenable to using the BV assay to assist with antimicrobial decision-making.³⁶  Additional research is required to evaluate how the timing of biomarker acquisition may impact decision-making (eg, collected during the initial part of the encounter, before review radiography results, or after, to determine a need for antibiotic). This requires buy-in among key partners, such as front-line clinicians and patient caregivers.

Finally, we observed differences in radiographic findings based on BV test results. Patients with a higher likelihood of bacterial infection on the BV test more frequently had lobar infiltrates compared with those with a higher likelihood of viral infection. These findings suggest that the BV test may have a role in limiting antimicrobial use among children with radiographic pneumonia, which is beneficial given the ambiguity in national guidelines on antibiotic use in pediatric pneumonia.³⁵  However, a similar difference in radiographic infiltrates was observed when stratifying patients by CRP. Since etiology cannot be definitively determined from radiographic findings, further research is needed to better evaluate the role of the MeMed test in safely reducing antibiotic use in children with suspected pneumonia.

Our findings are subject to limitations. This was a secondary analysis of a previously performed prospective study. As such, we were unable to include some variables that have been demonstrated to be associated with pneumonia in our clinical models (such as auscultatory findings).^28,29  As with the parent study, the study sample may be subject to ascertainment bias: as enrolled children were required to provide blood samples, those with lower acuity disease may have more frequently declined participation. Some variables (such as oxygen saturation and procalcitonin) were missing, though this occurred infrequently. Our findings when using splined predictors may be impacted by the smaller sample size, leading to broad confidence intervals. Although the MeMed BV results were not available to the treating clinician, other laboratory test results (including CRP, which is a component of the BV test) were available if ordered clinically; this in turn may have impacted decisions to acquire a chest radiograph and impacted secondary outcomes, such as decisions to treat with antibiotics. Despite these limitations, the findings from this study represent an important evaluation of a novel assay for the identification of radiographic pneumonia in children, which has not been reported in the literature.

We evaluated the association of the MeMed BV assay with pediatric radiographic pneumonia. When used in isolation, the BV score is associated with pneumonia, although this did not occur in a uniform manner with rising predicted risk of bacterial infection. This effect was magnified in multivariable models that incorporated clinical variables. Besides antibiotic use, the BV test was not associated with clinical outcomes. Further research is needed in larger samples to better understand its role in the management of pediatric pneumonia, including possible development of multivariable models that incorporate this assay.

We thank Anne Lakes (Ann & Robert H. Lurie Children’s Hospital of Chicago) for assistance with data acquisition for this study.