Radiographic Uncertainty and Antibiotic Prescribing in Pediatric Pneumonia

Over one million children are diagnosed with community-acquired pneumonia (CAP) across ambulatory settings annually nationwide.1 There is no consensus regarding whether a chest radiograph (CXR) should be performed. Many clinicians perform CXR, while others presumptively prescribe antibiotics.2 This decision varies given challenges in the history and physical examination,3 mixed evidence for CXR,4 variable radiologist CXR interpretation,5 and CXR availability. Uncertainty in CXR interpretation may drive antibiotic overuse.6 We evaluated how radiographic uncertainty influences antibiotic prescribing in children undergoing CAP evaluation.We prospectively evaluated children 3 months to 18 years old with CXR for possible CAP between March 2015 and April 2025 in a large pediatric emergency department (ED).7 Children with complex chronic conditions, current antibiotics, and concurrent bacterial infection were excluded. CXR reports, written by board-certified radiologists, were independently interpreted by 3 pediatric ED attendings blinded to clinical data. Interpretations were characterized as: no pneumonia (normal CXR, bronchiolitis/atelectasis), equivocal for pneumonia (favor bronchiolitis/atelectasis, equal probability of bronchiolitis/atelectasis or pneumonia, favor pneumonia), and pneumonia. Treating attendings indicated whether they would have prescribed antibiotics without the CXR (ie, pre-CXR plan). We dichotomized the data by pre-CXR antibiotic plan and performed unadjusted and adjusted logistic regression models with CXR interpretation as the exposure and antibiotic prescribing as the outcome. We adjusted for age, fever (≥38°C), lowest oxygen saturation, wheeze, decreased breath sounds, retractions, respiratory distress, age-adjusted tachypnea, presence of a positive viral test, and performance of blood work (complete blood count, procalcitonin, or blood culture). In sensitivity analyses, we explored the effects of (1) different equivocal results on decision-making and (2) excluding patients where attendings reviewed the CXR before indicating their pre-CXR plans. All analyses were performed in R version 4.3.2 (R Foundation, Vienna, Austria) using P < .05 for significance. The study institution's institutional review board approved this study.We evaluated 2142 children, 64% of whom were under 5 years (Supplemental Table 1). Most (73%) had no pneumonia on CXR, 13% had pneumonia on CXR, and 14% had equivocal CXRs. Attendings planned to treat 304 (14%) children with antibiotics pre-CXR, and 604 (28%) received treatment (Figure 1). Among children with no pre-CXR antibiotic plan, equivocal (adjusted OR [aOR]: 38.5, 95% CI 26.2–57.6) and pneumonia CXRs (aOR 374.0, 95% CI 191.0–826.0) were associated with antibiotic prescribing compared with a no pneumonia CXR (Table 1). Among children with a pre-CXR antibiotic plan, equivocal (aOR: 37.6, 95% CI 12.4–166.0) and pneumonia CXRs (aOR 43.9, 95% CI 16.1–158.0) were associated with antibiotic prescribing compared with a no pneumonia CXR. All equivocal categories (favor bronchiolitis/atelectasis, equal probability, and favor pneumonia) were associated with increased antibiotic prescribing (Supplemental Tables 2 and 3). Equivocal and pneumonia CXRs were associated with increased antibiotic prescribing after exclusion of the 577 (24.3%) patients for whom the attending reviewed the CXR before indicating their pre-CXR plan (Supplemental Table 4).In this large prospective cohort of children undergoing CXR for suspected CAP, we observed that radiographic uncertainty was common and associated with increased antibiotic prescribing. Our findings underscore the concern that CXRs may inadvertently drive antibiotic use.This investigation addresses an important issue around the value of CXR in pediatric CAP. CXRs are reassuring when negative,7 associated with reduced health care visits,8 and valued by caregivers.9 Alternatively, CXRs are associated with radiation exposure, cannot reliably distinguish viral from bacterial disease,10 and may promote antibiotic overuse in uncertain cases. These tradeoffs highlight important areas for future research, including (1) standardizing CXR interpretation, potentially through artificial intelligence, (2) identifying which children benefit from CXR through risk modeling, and (3) developing decision support tools that effectively communicate diagnostic uncertainty.This study has limitations. Given the single-center design, generalizability may be limited if thresholds for CXR differ by practice setting. Radiographic uncertainty was determined by radiology reports; clinicians may have interpreted the image differently. Viral testing was infrequent but was included given its potential influence on prescribing. We were unable to identify drivers of antibiotic use with equivocal CXR, including whether caregiver preferences were incorporated.In conclusion, radiographic uncertainty is associated with increased antibiotic prescribing in suspected CAP, highlighting how CXRs may promote antibiotic overuse. Future work should explore drivers of antibiotic treatment in uncertainty and assess which children are appropriate for watchful waiting.