Cost of Pediatric Pneumonia Episodes With or Without Chest Radiography Free

This was a retrospective cohort study of children discharged with a diagnosis of CAP from any ED within the Healthcare Cost and Utilization Project State ED and Inpatient Databases of Arkansas, Florida, Georgia, Iowa, Maryland, Nebraska, New York, and Wisconsin. The databases contain administrative data on all ED and inpatient encounters in each state. These specific states were chosen for their data quality and, together, included 20.9% of the national child population in 2019.

We included children aged 3 months to 18 years discharged with a diagnosis of CAP between July 1, 2014, and September 30, 2019, using a previously-validated International Classification of Diseases (ICD), Ninth Revision, Clinical Modification, 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%) and the corresponding ICD, 10th Revision, Clinical Modification, codes (Supplemental Table 3).^12,21 

We excluded EDs with poor data quality where there was potential for incomplete or inaccurate data coding. Poor data quality was defined by identifying the proportion of patients who were diagnosed with an extremity fracture and did not have an associated code for radiographic imaging.²²  EDs where <70% of children with an extremity fracture were discharged without a code for radiographic imaging were excluded from analysis. This was based on the authors’ clinical experience in both pediatric and community hospitals, in which patients with an extremity fracture diagnosis nearly always undergo radiography.

We excluded patients without a unique longitudinal patient identifier or those who were transferred without a record from the receiving hospital. Children with complex chronic conditions, codiagnosis of aspiration pneumonia or complicated CAP, any-cause hospitalization within the previous 30 days, or previous ED visit or hospitalization for CAP within 6 months were excluded.^1,23,24  Children who left against medical advice or were discharged to a home health care facility were also excluded.

CXR was the primary exposure and was identified using procedure codes (Supplemental Table 3). We evaluated total costs within the 28-day period after ED discharge for CAP, inclusive of the index visit and all subsequent ED and hospital care. Costs were calculated using ED-level charges and multiplied by cost-to-charge ratios specific to each ED for each year. Charges in the database reflect the amount billed for hospital services, but not specific costs. The cost-to-charge ratios were developed as part of the database to estimate the actual cost of services.

Patient-level variables included in analysis included age (categorized on the basis of previous literature into 3–11 months, 1–5 years, 6–10 years, and 11–18 years), sex, primary payer, weekend versus weekday presentation, and asthma codiagnosis (defined by ICD, Ninth Revision, and ICD, 10th Revision, codes) given the potential for these variables to be confounders.^14,25,26 

To adjust for underlying CAP severity, we identified whether parenteral antibiotics or laboratory testing (ie, complete blood count, blood culture, inflammatory markers, nasopharyngeal and blood viral studies) were performed in the ED using procedure codes (Supplemental Table 3).

Descriptive statistics, including median and interquartile range (IQR), were used to report the costs of the index ED visit, subsequent visits, and full 28-day CAP episodes, stratified by performance of CXR at the index ED visit. Costs were winsorized at the 0.5th and 99.5th percentiles. Differences in baseline patient characteristics among those with and without CXR at the index visit were evaluated using χ² tests.

To adjust for patient-level confounders and markers of illness severity (performance of laboratory testing and/or administration of parenteral antibiotics), we constructed a mixed-effects linear regression model with costs as the outcome variable, CXR as the exposure, and all patient-level variables. In post-hoc exploratory analyses, we repeated the adjusted regression model restricting:

1. only to those who did not require a subsequent care after initial discharge from the index ED visit; and

2. only those who did require subsequent care (any revisit or admission).

This exploratory analysis was performed to determine whether the association between CXR and 28-day costs was preserved among the majority of patients who did not require subsequent care.

Itemized charges for specific laboratories and medications were not available within the database. We also performed a post-hoc exploratory analysis to assess how laboratory testing and parenteral antibiotics contributed to overall 28-day CAP costs. To evaluate costs of laboratory testing and parenteral antibiotics, we performed Wilcoxon rank sum tests to compare costs among those who did and did not receive antibiotics or laboratories at the index ED visit. Finally, we plotted cumulative mean costs for each day using the index visit arrival as day 0. We stratified cumulative mean costs by CXR and laboratory testing, because laboratory tests generate costs and are also associated with the likelihood of CXR. We performed Wilcoxon rank sum tests evaluating overall 28-day costs among those with and without CXR who received laboratory testing at the index ED visit and among those with and without CXR who did not receive laboratory testing at the index ED visit.

We evaluated 225 781 (64.2%) of the 351 884 children who were discharged from the ED with a diagnosis of CAP (Fig 1). Most (86.2%) children in the cohort had CXR at their index ED visit (Table 1). The majority of children in the cohort were between 1 to 5 years of age, had public insurance, and were evaluated in an urban ED. Children who had a CXR at initial evaluation were more likely to receive intravenous antibiotics (20.2% vs 11.9%, P < .001) and have laboratory testing (47.1% vs 15.6%, P < .001).

Median costs of the 28-day episodes, index ED visits, and subsequent visits were $314 (IQR 208–497), $288 (IQR 195–433), and $255 (IQR 133–637), respectively (Table 2). Median unadjusted total 28-day costs were $324 (IQR 218–505) and $244 ($146–432) among patients with and without CXR, respectively. However, after adjusting for patient-level variables and illness severity, CXR at the index ED visit was associated with a $33 (95% confidence interval [CI] 22–44) 28-day savings per patient who received a CXR. The predicted annual cost savings if CXR had been performed in all patients in the cohort would be $204 624 (95% CI 136 833–272 416).

Of the 34 529 (15.3%) children who required subsequent care after discharge from the index ED visit, the median cost of subsequent visits was $254 (IQR 133–620) among children with CXR and $261 (IQR 135–727) among children without CXR. Of the 5683 (2.5%) children who required admission after discharge from the index ED visit, the median cost of subsequent visits was $3832 (IQR 2396–6636) among children with CXR and $3862 (IQR 2317–7067) among children without CXR. Of the 1002 (0.4%) children who required ICU admission median costs of subsequent care were $8226 (IQR 5527–8226) among children with CXR and $8226 (IQR 5742–8226) among children without CXR.

In the 34 529 children who required any subsequent care, CXR at the index visit was associated with a 28-day cost savings after adjusting for patient-level variables and illness severity (decrease of $232, 95% CI −294 to −170, P < .001). Of the 191 252 (84.7%) children who did not require subsequent care, CXR at the index visit was associated with a small 28-day cost savings after adjusting for patient-level variables and illness severity (decrease of $8, 95% CI −10 to −2, P < .003).

In the overall cohort, 42.7% of children received laboratory testing and 19.0% received parenteral antibiotics at the index ED visit. Median unadjusted index costs among those who did and did not have laboratory testing were $385 (IQR 260–587) and $239 (IQR 166–338, P < .001) and among those who did and did not receive parenteral antibiotics were $430 (IQR 276–984) vs $266 (IQR 184–389, P < .001).

Among children without laboratory testing at the index ED visit, children who had a CXR at time of initial diagnosis had lower mean cumulative 28-days costs as compared with children who did not have a CXR at time of initial diagnosis (Fig 2, mean [SD] $400 [712] vs $402 [833], P < .001). Similarly, among children who did have laboratory testing at the index ED visit, performance of a CXR was associated with overall 28-day cost savings (mean [SD] $694 [1053] vs $908 [1290], P < .001).

This study aimed to determine the potential impact of clinical versus radiographic CAP diagnosis on costs of CAP-associated care. Understanding this association is important given the challenges associated with CAP diagnosis. Not only is the CAP clinical exam unreliable, but there is also conflicting evidence supporting the use of CXR for CAP diagnosis and guidelines which specifically recommend against its routine use.^14,27  We have evaluated this association across both tertiary care and low pediatric volume EDs, thus making our findings widely generalizable. We found that performance of CXR at the time of CAP diagnosis in the ED is associated with a small, but significant, 28-day cost-savings overall. However, when stratified by receipt of subsequent care, CXR was predominantly cost-saving only among patients who required subsequent care, which was a minority of the overall cohort. This finding suggests that CXR performance for CAP diagnosis may be associated with lower costs for children who require subsequent health care after initial discharge from the ED.

CXR was associated with a 28-day cost savings across patients who did or did not receive laboratory testing. However, the savings were larger among children who received laboratory testing than those who did not. These findings suggest that CXR may be most beneficial among children with higher illness severity. Given the lack of clinical data in the database, we are limited in our ability to determine why CXR is cost-saving. We hypothesize that clinicians may be using CXR in their clinical decision-making regarding disposition from the ED visit, identifying children with CAP who are at lower risk for revisit. Alternatively, CXR may provide a more reliable CAP diagnosis than the clinical exam. Children with clinically diagnosed CAP may have other non-CAP diagnoses which require return to care for definitive treatment.

Most children in the cohort did not require subsequent visits after initial discharge from their index ED visit. In the children who did not require subsequent care, CXR was only minimally cost-saving and likely not clinically significant. This finding suggests that the cost savings of CXR are entirely because of the prevention of downstream care.

The modest cost savings in this study preclude a blanket recommendation for or against the routine use of CXR in the diagnosis of CAP, on the basis of cost alone. However, our findings support that CXR is noninferior to clinical diagnosis from a cost perspective among children with CAP discharged from the ED who do not require subsequent care. Previous studies have shown that CXR performance offers improved diagnostic accuracy over the clinical exam and likely allows for earlier detection of complicated pneumonias.²⁷  When considering these benefits and our findings that CXRs do not add to overall costs of care, CXR performance, on balance, is probably worthwhile. Still, CXR may be most beneficial among children at higher risk for severe CAP and who may require subsequent care. In light of these findings, we suggest that clinicians consider performance of CXR at time of CAP diagnosis for children in whom there may be uncertainty around discharge, risk of subsequent care after ED discharge, or risk of severe CAP. Although large-scale studies are in progress to develop prediction models, there is currently no standardized approach to accurately identify children at risk for severe CAP.^28–30  With the development of these models, clinicians may ultimately be able to use CXR at time of diagnosis as a cost-saving strategy that may minimize the need for downstream provision of care.

There are several important limitations to this study. Children with CAP were identified using a set of previously validated diagnostic code classifications with imperfect test characteristics.³¹  Although we chose a classification system with high specificity to maximize capture of true CAP diagnosis, coding likely missed some children with CAP diagnosis and some may have been misclassified. The claims data set used in this study did not include clinical data. Thus, we were limited in our ability to assess underlying patient illness severity, which likely impacted decision-making around CXR performance and overall 28-day costs. However, we were able to ascertain performance of laboratory testing and administration of parenteral antibiotics, which we included as markers of illness severity in our primary analysis. Although there is considerable variation in laboratory performance for children with CAP, we felt that laboratory testing was a reasonable marker for illness severity given that it is recommended in national guidelines for children with more serious disease.^14,25  In addition, although we were unable to capture the timing of antibiotics and laboratory testing in relation to the CXR performance, on the basis of clinical experience, it was felt to be unlikely that CXR would drive the decision to perform laboratories or give parenteral antibiotics among children well enough to be discharged from the ED. Ambulatory centers were not captured in this database, and thus we were unable to ascertain costs of CAP episodes in this setting. Importantly, this study was not able to address costs associated with CXR among those children who received CXR for suspicion for CAP but were ultimately discharged from the ED without a diagnosis of CAP. The costs of CXR in these patients are important when considering the overall costs of CXR among children who are undergoing CAP workup. In addition, we were unable to identify costs associated with provision of care outside of the primary database. However, the database in this study incorporates all ED and inpatient encounters within a given state. Approximately one-quarter of the cohort was excluded given lack of a longitudinal patient identifier, which may have contributed to underlying selection bias.

In this large retrospective cohort study of tertiary and community EDs, we found that CXR was associated with a 28-day cost savings among children discharged from the ED with a diagnosis of CAP. Subsequent costs among children who required subsequent care after initial ED discharge accounted for the majority of the observed cost savings. Therefore, CXR may be a cost-saving strategy among children with CAP by averting subsequent health care after initial discharge from the ED. Future studies, incorporating decision analysis or a prospective cohort, are needed to determine whether the observed cost benefit among children with CAP outweighs the costs of negative CXRs performed among children undergoing CAP workup who receive an alternative diagnosis.