Trends in Chest Radiographs for Pneumonia in Emergency Departments

Data for this study were obtained from the Pediatric Health Information System (PHIS), an administrative database that contains inpatient, ED-level, ambulatory, surgery–level, and observation encounter–level data from 52 not-for-profit, tertiary care pediatric hospitals in the United States. These hospitals are affiliated with the Children’s Hospital Association (Lenexa, KS). Data quality and reliability are assured through a joint effort between the Children’s Hospital Association and participating hospitals. Portions of the data submission and data quality processes for the PHIS database are managed by Truven Health Analytics (Ann Arbor, MI). For the purposes of external benchmarking, participating hospitals provide discharge and encounter data, including demographics, diagnoses, and procedures. Nearly all of these hospitals also submit resource use data (eg, pharmaceuticals, imaging, and laboratory) into PHIS. Data are de-identified at the time of data submission, and data are subjected to a number of reliability and validity checks before being included in the database. For this study, data from 30 hospitals were included. Twenty-two hospitals were excluded from analysis because of data quality issues or for not having complete ED data for the entirety of the study period.

We reviewed encounter-level data from 30 hospitals in PHIS for children aged 3 months to 18 years who were discharged from the ED between July 1, 2008, and June 30, 2018, with a primary diagnosis code for CAP, fever, or respiratory illness. Children with fever or respiratory illness were included in the study to characterize the broader cohort of children for whom CXR may be considered. CAP was defined by a set of previously validated International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes.¹⁹  We selected corresponding International Classification of Diseases, 10th Revision (ICD-10) codes to account for the adoption of ICD-10 codes by PHIS on October 1, 2015. Fever and respiratory illness were defined by using codes identified by a previously validated ICD-9–based classification system for health services research in emergency medicine,²⁰  and these ICD-9 codes were subsequently matched to corresponding ICD-10 codes (Supplemental Table 5). Because we chose 2 distinct previously validated sets of codes to define pneumonia and fever or respiratory illness, there are several overlapping codes that were captured in both cohorts.^19,20  We excluded visits for aspiration pneumonia and complicated pneumonia using a previously defined classification system given the different presentation and diagnostic workup in comparison with CAP.^3,21  Patients with complex chronic conditions were also excluded given their diversity in presentation with respiratory illness and the exclusion of these patients from the guidelines.²²  Patients who were transferred to a study institution from another hospital were also excluded given the inability to ascertain whether a CXR had been obtained before transfer.

Descriptive patient statistics (frequencies with proportions and medians with interquartile ranges) were used to characterize categorical and continuous variables. Hospital-level proportions of patients with diagnoses of either CAP or fever or respiratory illness were calculated. Hospitals were characterized by geographic region, payer mix, annual number of CAP and total ED visits, and annual CXR numbers by quartile. To assess hospital-level variation in CAP diagnosis rates, we estimated a logistic regression model with CAP diagnosis as the dependent variable and hospital (modeled as a set of indicator variables) as the independent variable and tested the null hypothesis that the odds of a CAP diagnosis were equal across all hospitals using a Wald test. A significant result would indicate a difference between at least 2 hospitals in the CAP diagnosis outcome. A robust variance estimator was used to accommodate the correlation resulting from the clustering of patients within hospitals. To examine practice variability over time, we estimated a set of logistic regression models with time (modeled monthly) as the independent variable and the following as dependent variables: (1) the proportion of patients with fever or respiratory illness with CXR performed, (2) the proportion of patients with CAP with CXR performed, and (3) the proportion of patients with CAP. To account for seasonal variability in disease incidence, we included Fourier regression terms (ie, sine and cosine terms to capture monthly cyclical patterns in the outcome). Yearly variation in influenza diagnosis rates was not included in the model given the small number of primary influenza patient encounters in comparison with the entire cohort. A significant result would indicate a change in CXR use and CAP diagnoses over time during the study period (July 1, 2008 to June 30, 2018). Interrupted time series analyses were performed to assess for changes in the rates of decline in CXR use and CAP diagnosis before and after IDSA guideline publication. To assess whether changes in CXR usage were associated with changes in revisit rates, we calculated the annual rate of 3-day revisits after an ED visit for patients with fever and respiratory illness, including both ED revisits as well as ED revisits associated with hospitalization. Finally, we correlated hospital-level CXR use with CAP diagnosis rates using Spearman’s correlation weighted by hospital volume. All analyses were performed by using the software package Stata SE, version 15.1 (Stata Corp, College Station, TX). All statistical tests were 2-tailed, and α was set at .05. The study was approved by the institutional review board at the study institution.

After review of all eligible patients discharged from the ED with a diagnosis of CAP or fever or respiratory illness and after subsequent removal of patients meeting exclusion criteria, there were 2 787 929 patients included in the fever or respiratory illness cohort and 190 825 patients included in the CAP cohort (Fig 1). The majority of children in both the fever or respiratory illness and CAP cohorts were between 1 and 5 years of age. Children with CAP, and those with fever or respiratory illness, were younger than children seeking care for other reasons during the study period (Table 1). There were no material differences in race, ethnicity, or payer between children with fever or respiratory illness or CAP and the overall ED population. In encounters in which the etiology for CAP was specified as part of the diagnosis code, bacterial pneumonia was more common than viral pneumonia (Supplemental Table 6).

ED visits for fever or respiratory illness accounted for 15.8% of all ED visits, and ED visits for CAP accounted for 1.1% of visits. There were no major differences in the hospital characteristics between patients with fever or respiratory illness and those with CAP (Table 2). Children diagnosed with CAP were more likely to be cared for at a hospital in the highest quartile of ED visits for pneumonia. The rate of CAP diagnosis among all ED visits varied across hospitals (range 0.5%–2.2%; Wald test of CAP diagnosis rates across all hospitals: P < .001) (Fig 2). Patients with fever or respiratory illness represented 15.8% of visits across all hospitals, with evidence of variation at the hospital level (range 11.1%–26.3%; P < .001).

Overall CXR usage decreased over the 10-year study period among all patients presenting to the ED (21.3% decrease; odds ratio [OR] 0.953; confidence interval [CI] 0.938–0.969) (Table 3). CXR use declined from 30.4% to 18.6% (48.8% change; P < .001; OR 0.994; CI 0.992–0.996) for children with fever or respiratory illness and from 86.6% to 80.4% (7.2% change; P < .001; OR 0.996; CI 0.995–0.998) for children with CAP (Fig 3 A and B). There was a statistically significant decrease in patients diagnosed with CAP from 7.8% to 5.9% during the study period (32.3% change; P < .001; OR 0.997; CI 0.995–0.999) (Fig 3C). For both cohorts, using interrupted time series analyses, we observed that the rates of CXR decline did not change significantly after guideline publication. Additionally, rates of decline in CAP diagnoses were also not impacted by the guideline (Supplemental Fig 5). With the decline in CXR usage, we did not observe an increase in 3-day ED revisit rates or ED revisits associated with hospitalization for patents with fever or respiratory illness (Table 4). We observed a direct correlation between hospital-level CXR use and CAP diagnosis rates when weighted by hospital volume (Fig 4). Hospitals that performed more CXRs for patients with fever and respiratory illness observed a higher rate of CAP diagnosis among children with fever or respiratory illness (correlation coefficient 0.58; P < .001).

In this large cross-sectional study of pediatric patients, we observed a decreasing trend in CXR usage for children with respiratory illnesses overall, and for children with CAP specifically. With these declines, there has been a corresponding decrease in the rate of CAP diagnosis over the study period. Over the course of the study period, we did not observe an increase in revisit rates or subsequent admissions in patients presenting with fever or respiratory illness. We also have identified hospital-level variation in CXR use for patients presenting to the ED with complaints of fever or respiratory illness and have observed a correlation between a hospital’s use of CXR and pneumonia diagnosis.

The majority of children diagnosed with CAP had a CXR as part of their initial workup. This finding is consistent with 1 previous study that revealed that >75% of children with a CAP diagnosis receive a CXR in the ED.⁵  During the study period, we observed a decline in CXR usage, as measured by trend analysis. This trend was consistent among patients for all ED visits and when stratified by diagnoses specific for CAP and fever or respiratory illness. The steady decline in CXR usage among patients with CAP and among patients presenting with all diagnoses may reflect an increasing emphasis on value-based care and more judicious use of diagnostic imaging over time.^23,24  These findings are also consistent with 1 previous study that documented modest changes in diagnostic testing for patients with a CAP diagnosis, including CXR, after implementation of the 2011 pneumonia guidelines.⁷  Given that a decline in CXR use was not associated with increased ED revisit rates or hospitalizations for patients with fever or respiratory illness, we believe that the observed reduction in CXR use did not result in delays in pneumonia diagnosis or return visits for progressively more severe pneumonia.

Our study illustrates significant hospital-level variability in CXR usage for patients with a CAP diagnosis, which is consistent with previous studies revealing hospital-level variation in CAP evaluation.^5,25  When analyzed at the hospital level, increased CXR usage was associated with increased CAP diagnosis rates. This contrasts with 1 previous study in which authors reviewing PHIS data before national guideline implementation observed no association between CXR usage and CAP diagnosis.⁶  The observed association between CXR usage and CAP diagnosis may reflect differences in the rates of CAP in baseline patient populations presenting to different EDs. It is possible that the use of CXR may be associated with a reduction in pneumonia diagnoses if CXR is being used to confirm a suspected diagnosis of pneumonia. However, it is also possible that the liberal use of CXR for children with respiratory illnesses may lead to an increase in the overall diagnosis of CAP. Although this study cannot distinguish between these 2 possibilities, we are reassured that the reduction in the use of CXR over time was not associated with an increase in pneumonia diagnoses. Further study is needed to evaluate clinical outcomes in patients discharged from the ED with clinically diagnosed pneumonia (without CXR) to assess for missed CAP diagnoses and to assess provider-level pretest probability for CAP to determine if CXR is being used appropriately. Our findings revealing a decline in both CXR use and CAP diagnoses suggest against the latter, but further study is needed to assess whether physicians are appropriately using CXR for CAP evaluation at the hospital level.

Our study has several important limitations. By extracting data from the PHIS database, we evaluated only freestanding children’s hospitals, and as a result, our findings may not be broadly generalizable to other health care settings. Patients presenting to PHIS hospitals may be sicker than the general population of children in whom CAP may be considered and therefore may introduce a selection bias to this study. The PHIS database does not include information regarding medications prescribed on discharge; thus, we were unable to directly evaluate trends in antibiotic prescribing associated with changes in CXR rates. In classifying diagnoses by ICD-9 and ICD-10 codes, we chose a set of previously validated codes to define CAP and fever or respiratory illness. There may be inherent differences in the coding practices at both the hospital and provider levels among PHIS-participating hospitals that contribute to selection bias. This bias may be minimized by the use of previously described and, in some cases, validated ICD-9 and ICD-10 codes. To capture the patient population in whom a CAP diagnosis may be considered, we intentionally chose a broad definition of fever and respiratory illness. In doing so, we recognize that some of the codes in this classification scheme may not be relevant for a patient undergoing CAP evaluation. Although we excluded patients transferred in from outside hospitals, we were unable to assess in our included patient population testing performed before the ED visit, such as in the outpatient primary care setting.

Our study reveals hospital-level variation in CXR use and illustrates a decline over the 10-year study period in CXR use for workup of patients presenting to the ED with CAP and respiratory illness. In addition, we found no evidence that decreased CXR use and increased reliance on the history and physical examination for CAP diagnosis were associated with an increase in CAP diagnoses over time. These findings provide reassurance that decreased resource use for CAP diagnosis may not be associated with overdiagnosis of CAP. Future studies are needed to better define the role for CXR in CAP evaluation in the ED and to assess whether variation in CXR usage at the hospital level leads to overdiagnosis of CAP and inappropriate use of antibiotics.

CAP

community-acquired pneumonia

CI

confidence interval

CXR

chest radiograph/radiography

ED

emergency department

ICD-9

International Classification of Diseases, Ninth Revision

ICD-10

International Classification of Diseases, 10th Revision

IDSA

Infectious Diseases Society of America

OR

odds ratio

PHIS

Pediatric Health Information System