Association of Chest Radiography With Outcomes in Pediatric Pneumonia: A Population-Based Study Free

We conducted a retrospective cohort study of children discharged from the ED with a diagnosis of CAP using the Healthcare Cost and Utilization Project State ED and Inpatient Databases of Arkansas, Florida, Georgia, Iowa, Maryland, Nebraska, New York, and Wisconsin. These databases include administrative data on all ED (both children’s hospital EDs and general EDs) and inpatient encounters within those states. Patients can be followed between visits and different facilities. The states included 20.9% of the US child population in 2019 and were chosen because they have sufficient data quality to allow longitudinal follow-up of patients.

We evaluated data from July 1, 2014 through November 30, 2019 in each state. We evaluated children 3 months to 17 years old discharged from the ED with a primary diagnosis of CAP. CAP was defined using a previously-validated International Classification of Diseases, 9th Revision diagnosis code set (test characteristics according to reference standard of provider-confirmed CAP: sensitivity of 64%, specificity of 96%, positive predictive value 89%, negative predictive value 84%, Supplemental Table 4).²⁶  By including only those patients with a primary diagnosis code of CAP we sought to maximize specificity and increase the likelihood that captured patients had a true CAP diagnosis. To account for widespread adoption of ICD-10 codes during the study period, we selected the corresponding ICD-10 codes as previously described (Supplemental Table 4).²²  EDs with poor data quality regarding ascertainment of imaging were excluded from analysis. Poor data quality was defined by the proportion of patients at a given ED who were diagnosed with an extremity fracture and did not have an associated code for radiographic imaging, indicating the potential for inaccurate or incomplete data coding. We excluded patients presenting to EDs in which <70% of children with a fracture were discharged with a billing code for imaging.

Patients without a unique longitudinal patient identifier in the databases, or whose care involved a transfer to another hospital outside the database (eg, a hospital in another state), were excluded. Patients with 2 same-day ED visits in which the order of visits was unable to be ascertained were excluded. We excluded patients with complex chronic conditions given their susceptibility to invasive infections including complicated CAP and different indications for diagnostic testing.²⁷  Patients with codiagnoses of aspiration pneumonia or complicated CAP, defined in previous studies, all-cause hospitalization within the previous 30 days, previous ED visit or hospitalization for CAP within 6 months, and those discharged against medical advice or to a home health care facility were also excluded.^3,28  Children with previous ED visit or hospitalization for CAP within 6 months were excluded such that clinician decision-making around CAP diagnosis at the index ED visit were unlikely to be influenced by previous CAP care or concern for complicated CAP on the basis of history alone. Visits meeting all inclusion and no exclusion criteria were designated as index visits.

The primary exposure was performance of a CXR at the index ED visit, identified using procedure codes (Supplemental Table 4). These procedure codes were selected a priori by reviewing all procedure codes occurring in CAP patients in at least 1 of 25 000 cases.

The primary outcome was all-cause hospitalization within 7 days of the index ED visit.³  Secondary outcomes included 7-day ED revisits and 7-day hospitalization for severe CAP. Severe CAP was defined by the presence of 1of the following: acute respiratory failure, complicated CAP, sepsis, ICU admission, and/or death.²⁹  Codes used to identify the outcomes of acute respiratory failure, complicated CAP, and sepsis are displayed in Supplemental Table 4.^22,30–34  ICU admission was identified on the basis of the presence of an ICU room charge.

Variables at the patient level included age (3 months to <1 year, 1–5 years, 6–10 years, 11–18 years), sex, primary payer, season (defined as quarter of the year), weekend versus weekday presentation, and either acute or chronic asthma codiagnosis (defined by International Classification of Diseases, Ninth Revision Clinical Modification and International Classification of Diseases, 10th Revision Clinical Modification codes for asthma). Age was categorized a priori by using previously used categories given age-based differences in CAP prevalence and etiology.^10,35,36  ED-level variables included ED location (urban vs nonurban on the basis of the urban-rural classification scheme developed by the National Center for Health Statistics), the proportion of CAP patients who received a CXR, and both overall and CAP-specific pediatric ED volume.

We hypothesized that the cohort was relatively low-severity since all patients were discharged from their index ED visit. To further measure the severity of CAP, we determined whether each child received a parenteral antibiotic or laboratory testing (i.e., complete blood count, blood culture, inflammatory markers, nasopharyngeal studies or blood viral studies) at the index ED visit defined by procedure codes (Supplemental Table 4).

χ-square analyses were used to identify differences in baseline characteristics among children who did and did not received CXR at the index ED visit. We identified the proportion of patients who experienced a 7-day all-cause ED revisit, 7-day all-cause hospitalization, and 7-day severe CAP in the whole cohort and after stratification by CXR performance. Mixed-effects logistic regression with a random intercept for ED was used to evaluate the association between CXR utilization and each outcome. To account for potential cofounding by illness severity and patient characteristics, we included the following covariates: age, sex, payer, season, weekend presentation, asthma codiagnosis, parenteral antibiotic administration, and laboratory testing performance at the index visit. To assess for systematic bias, we performed a sensitivity analysis excluding data from Florida and Arkansas, the 2 states responsible for the majority of patients excluded for a missing longitudinal patient identifier.

To assess the association of hospitals’ CXR performance with outcomes, we examined rates of each outcome by quartile of ED-level CXR performance. This evaluation was performed at the ED-level to evaluate variation in outcomes independent of individual patient-level performance of CXR. The association of being a high versus low CXR utilizer with outcomes was determined using mixed-effect models with the same structure as the patient-level model, replacing the exposure of patient-level CXR with ED-level CXR quartile.

In an exploratory posthoc analysis, we evaluated how patient variables (sex, age, payer, and asthma codiagnosis) and ED variables (ED location, annual pediatric ED volume, and annual pediatric CAP volume) modify the association of CXR and 7-day all-cause hospitalization through a stratified analysis using interaction effects. We performed mixed-effects logistic regressions with 7-day hospitalization as the dependent variable and the following independent variables: CXR, modifier, and CXR-modifier interaction. The interaction term tested whether the association between CXR and outcome differed by levels of a given modifier. The 2 severity measures (parenteral antibiotics and laboratory testing) were also included as covariates. An additional exploratory analysis was performed to assess whether EDs with lower CXR rates had higher discharge rates among all patients evaluated with pneumonia, which might suggest that such EDs have a “sicker” cohort of discharged patients. To do this, we used a linear regression model weighted by hospital-level pediatric volume assessing the relationship between CXR and discharge rates at the ED level among EDs who evaluated at least 27 patients (the number needed to ensure that EDs evaluated enough patients to ensure that the widest possible confidence interval [CI] for each ED-level CXR rate would be 20%). Finally, to assess how revisit diagnoses may be different among patients who do and do not have a CXR as part of their index visit, we compared the proportion of revisits with a primary diagnosis of CAP between these 2 groups using a χ-square test. All analyses were performed using R version 4.1.2 (R Foundation, Vienna, Austria).

Out of 362 918 potential study encounters, we analyzed 213 959 (59.0%), with most (n = 91 238) exclusions occurring due to a missing longitudinal identifier (Fig 1). The majority of children in the study cohort were between the ages of 1 to 5 years, had government insurance, and were seen at a nonurban ED with an annual pediatric visit volume >10 000 (Table 1). Most (86.1%) children had CXR at their index ED visit. Children who received CXR as part of their diagnostic evaluation were more likely to have an asthma codiagnosis and administration of parenteral antibiotics at the index ED visit. Children evaluated at EDs with low pediatric volume were less likely to have a CXR as part of their initial diagnostic evaluation (Table 1).

The 7-day ED revisit rate in the cohort was 8.9% (Table 2). Among those with a revisit, 13 977 (73.6%) returned to the same ED and 5010 (26.4%) returned to a different ED. The overall rates of 7-day hospitalization and 7-day revisit with severe CAP were 1.6% and 0.4%, respectively. Severe outcomes were rare, and ICU admission was the most common reason (50% of severe revisits).

Unadjusted and adjusted odds of 7-day ED revisit, 7-day hospitalization, and 7-day severe CAP by CXR are presented in Table 2. After adjusting for illness severity, CXR was associated with a lower odds (1.6% vs 1.7%, adjusted odds ratio [aOR] 0.81, 95% CI: 0.73–0.90) of 7-day hospitalization. CXR was not associated with overall ED revisits (9.0% vs 8.4%, aOR 0.99, 95% CI: 0.94–1.04) or hospitalization because of severe CAP (0.4% vs 0.4%, aOR: 0.88, 95% CI: 0.70–1.10). In the sensitivity analysis excluding Arkansas and Florida, the adjusted odds ratios for 7-day hospitalization, ED revisit, and hospitalization for severe CAP did not change significantly.

CXR performance was frequent but varied between EDs (median 91.5%, IQR: 85.3%–95.0%). The median rate of CXR among hospitals in the lowest quartile of CXR utilization was 72.5% as compared to 97.0% among hospitals in the highest quartile of CXR utilization. Unadjusted and adjusted odds of 7-day ED revisit, 7-day hospitalization, and 7-day severe CAP for patients seen at hospitals with the highest CXR utilization (quartile 4) as compared to those with the lowest CXR utilization (quartile 1) are shown in Table 3. In the ED-level analysis, higher CXR utilization among index visits was associated with reduced odds of 7-day ED revisit (quartile 4: 8.5% vs quartile 1: 9.4%, aOR: 0.87, 95% CI: 0.80–0.96), 7-day hospitalization (quartile 4: 1.4% vs quartile 1: 1.9%, aOR 0.78, 95% CI: 0.65–0.94), and 7-day severe CAP (quartile 4: 0.3% vs quartile 1: 0.1%, aOR: 0.69, 95% CI (0.51–0.95).

The association between CXR and 7-day hospitalization did not materially differ when stratified by patient sex, age, payer, and codiagnosis of asthma. When stratified by ED-level variables, the association between CXR and 7-day hospitalization was significant among nonurban EDs (aOR: 0.64, 95% CI: 0.57–0.72) as compared to urban EDs (aOR: 1.13, 95% CI: 0.94–1.36, P value nonurban versus urban <.001). The association between CXR and 7-day hospitalizations varied significantly by annual pediatric ED volume (P < .001). The association was stronger at EDs with annual pediatric ED volumes between 1800 to 4999 (aOR: 0.61, 95% CI: 0.45–0.82) as compared to EDs with annual pediatric ED volumes <1800 (aOR: 1.10, 95% CI: 0.69–1.76), between 5000 to 9999 (aOR 0.67, 95% CI: 0.52–0.86), and >10 000 (aOR 0.84, 95% CI: 0.75–0.95).

In the linear regression analysis comparing the association between EDs’ CXR and discharge rates, CXR utilization was associated with increased discharge rates (CXR rate 0.39 percentage points higher per point increase in discharge rate, CI: 0.35–0.43). The primary diagnosis upon revisit was CAP after 53.7% of index visits with a CXR and 51.6% of index visits without a CXR (P = .053).

To our knowledge, this is the first study to evaluate the association between chest radiography and clinically important outcomes among children with CAP who were discharged from a broad cross-section of EDs. Our findings represent an important next step in the investigation needed to determine the value of CXR in the diagnostic evaluation of CAP. Our study evaluates children diagnosed with CAP across a variety of ED settings, including tertiary care center and those with low-pediatric volume and, thus, is broadly generalizable. Our study excluded children with a history of recent CAP diagnosis and, therefore, our findings may not be generalizable to those recently treated for CAP.

We found that performance of CXR at the index ED visit was associated with a small but significant reduction in 7-day hospitalizations at both the patient and the ED levels. Additionally, hospitals with higher rates of CXR utilization had lower rates of ED revisits and progression to severe CAP. The inverse relationship between CXR performance and 7-day hospitalization was more pronounced among nonurban hospitals and hospitals with moderate annual pediatric ED volumes. Although significant benefits were associated with performance of CXR, the modest effect sizes preclude a recommendation to routinely perform CXR for the diagnostic evaluation of CAP in children. With that said, our findings do support the need to evaluate whether CXR should be standard in CAP evaluation.

At the ED-level, EDs with high CXR utilization had fewer revisits, hospitalizations, and CAP within 7 days. We also observed a positive association between CXR utilization and discharge rate at the ED-level. Therefore, EDs with high CXR utilization are both more likely to discharge patients with CAP from the index ED visit and fewer of those discharged patients are likely to revisit and require subsequent hospitalization. These findings suggest that clinicians may be using CXR as part of their decision-making around disposition from the ED and may be identifying patients safe for discharge after CXR. The observation that hospitals with high CXR utilization have lower rates of 7-day severe CAP supports this conclusion. CXR may be providing clinicians with information that guides decision-making around hospitalization, antibiotic use, anticipatory guidance, and follow-up. In addition, clinicians at EDs with low CXR utilization may be more likely to discharge patients with a presumed diagnosis of CAP when they may actually have an alternative diagnosis such as asthma. These children may therefore be discharged without appropriate treatment at the index ED visit and may be more likely to return to the ED for definitive care. Without appropriate treatment up front, they may also have a worsening clinical trajectory and require subsequent hospitalization. In support of this hypothesis, we found that children without CXR are no more likely to have a CAP diagnosis at revisit than those who had CXR as part of their initial workup.

Our findings diverge from previous research including only tertiary care centers. A previous such study found no association between CXR and 3-day hospitalization after discharge from the ED with CAP diagnosis (aOR: 1.03, 95% CI: 0.88–1.22).¹⁹  The reasons for this difference are likely multifactorial but could include differences in illness severity and clinician prognostic accuracy between pediatric and general EDs. We found that the overall ED revisit rate was 8.9% and that a quarter of these patients revisited to another facility. This is consistent with previous literature demonstrating that revisit to another facility is common and underscores the importance of following patients between hospitals.²³ 

Given that we observed an association between CXR and reduced 7-day hospitalization, increased CXR use in settings with low CXR utilization may improve patient outcomes. However, the effect size observed in this study is small, particularly when considering the absolute risk and risk reduction observed in the study outcomes. It is unclear whether this effect outweighs the limitations associated with CXR, such as costs, availability, and radiation exposure. Additionally, this study did not evaluate the use of CXR in children who presented with respiratory illness and were ultimately discharged with non-CAP diagnoses. It is important to consider the risks of increased CXR utilization in patients discharged without CAP as compared with the observed benefits among those discharged with CAP. Future research that incorporates patient families as part of a shared decision-making process is needed to determine the effect size that is important and clinically meaningful. A randomized controlled trial evaluating the impact of CXR on patient-centered outcomes, such as antibiotic use among those at low risk of bacterial disease, ED revisits, and revisits with admission after ED discharge is needed to better define the role of CXR. In addition, the rise of new diagnostics such as point of care lung ultrasound for CAP may allow for comparative effectiveness studies of different CAP diagnostic modalities.

There are several important limitations to this study. The diagnosis of CAP was defined by diagnostic billing codes which are imperfect in their capture of patients with clinician-confirmed CAP. However, we chose a previously validated set of diagnosis codes with high specificity at identifying clinician-confirmed CAP. The databases did not include clinical information, so we were unable to evaluate the use of outpatient antibiotics. However, most patients with CAP diagnosis are prescribed antibiotics, as we believe was likely the case in this cohort.^10,37  In addition, without clinical information, it is difficult to account for differences in illness severity among patients, which likely drive clinicians’ decisions about diagnostic workup. For example, patients with CXR had higher rates of asthma codiagnosis and parenteral antibiotic administration suggesting higher rates of illness severity. However, all patients were discharged from the ED initially, suggesting they were mildly ill, which would tend to limit heterogeneity in severity. The quartiles of CXR utilization among EDs in this study were relatively clustered at high rates of CXR utilization, and thus we were unable to evaluate the association of CXR and disposition among hospitals in which CXR use is infrequent. In addition, we were able to adjust for certain indicators of illness severity including administration of parenteral antibiotics in the ED and use of laboratory testing. We were unable to identify revisits to EDs outside the database. However, because the database incorporates all ED and inpatient encounters within a given state, few revisits were likely missed. In addition, we were unable to account for clinician and caregiver comfort with discharge from the index ED visit which may have impacted revisits. Finally, a substantial portion (25%) of the eligible patients were excluded given lack of longitudinal data, which likely contributed to selection bias in the cohort. However, in a sensitivity analysis excluding the two states accounting for a majority of the patients without longitudinal data we found no difference in our primary outcome.

CXR was associated with a small but significant reduction in 7-day hospitalization among children discharged from an ED with CAP. Similarly, EDs that perform more CXRs have reduced rates of 7-day ED revisits, hospitalizations, and hospitalizations with severe CAP. However, the modest effect sizes observed preclude specific recommendations regarding the utility of CXR in CAP diagnosis and management. A controlled trial of CXR should be conducted to further clarify the value of CXR for children with suspected CAP.