Variation in bacterial pneumonia diagnoses and outcomes among children hospitalized with lower respiratory tract infections

Abstract

Background:

Current diagnostics do not permit reliable differentiation of bacterial from viral causes of lower respiratory tract infection (LRTI), which may lead to overtreatment with antibiotics for possible bacterial community-acquired pneumonia (CAP).

Objectives:

We sought to describe variation in the diagnosis and treatment of bacterial CAP among children hospitalized with LRTIs and determine the association between CAP diagnosis and outcomes.

Design, Setting and Participants:

This multicenter cross-sectional study included children hospitalized between 2017 and 2019 with LRTIs at 42 children’s hospitals.

Main Outcome and Methods:

We calculated the proportion of children with LRTIs who were diagnosed with and treated for bacterial CAP. After adjusting for confounders, hospitals were grouped into high, moderate, and low CAP diagnosis groups. Multivariable regression was used to examine the association between high and low CAP diagnosis groups and outcomes.

Results:

We identified 66,581 patients hospitalized with LRTIs and observed substantial variation across hospitals in the proportion diagnosed with and treated for bacterial CAP (median 27%, range 12%–42%). Compared with low CAP diagnosing hospitals, high diagnosing hospitals had higher rates of CAP-related revisits (0.6% [95% confidence interval: 0.5, 0.7] vs. 0.4% [0.4, 0.5], p = .04), chest radiographs (58% [53, 62] vs. 46% [41, 51], p = .02), and blood tests (43% [33, 53] vs. 26% [19, 35], p = .046). There were no significant differences in length of stay, all-cause revisits or readmissions, CAP-related readmissions, or costs.

Conclusion:

There was wide variation across hospitals in the proportion of children with LRTIs who were treated for bacterial CAP. The lack of meaningful differences in clinical outcomes among hospitals suggests that some institutions may overdiagnose and overtreat bacterial CAP.

INTRODUCTION

Lower respiratory tract infections (LRTIs) are among the most common and costly reasons for pediatric hospitalizations in the United States.^1,2 Distinguishing bacterial from viral causes of LRTI will ensure appropriate management as antibiotic use is only beneficial for those with bacterial disease (i.e., bacterial community-acquired pneumonia [CAP]). However, distinguishing bacterial CAP from other causes of LRTI can be challenging as there are no clinical, radiographic, or laboratory findings that can reliably differentiate between the two that are routinely used in clinical practice.^3–9 Due to this diagnostic uncertainty, there is wide acknowledgment that antibiotics are commonly overused in LRTIs.¹⁰ Antibiotic overuse exposes children to potential antibiotic-associated side effects without benefits and contributes to rising rates of antibiotic resistance, which is a major public health threat.^11–13

While prior studies have evaluated variation in antibiotic use for isolated infections (e.g., viral bronchiolitis and bacterial CAP), less is known about the variation in bacterial CAP diagnoses among children hospitalized with LRTIs.^14–17 In this study, we sought to (1) describe variation in the diagnosis and treatment of bacterial CAP among children hospitalized with LRTIs across tertiary care children’s hospitals; and (2) evaluate the association between bacterial CAP diagnosis and outcomes (clinical, financial, utilization).

METHODS

Study design and setting

This multicenter cross-sectional study used data from the Pediatric Health Information System (PHIS), an administrative and billing database containing encounter-level data from 49 tertiary care children’s hospitals across the United States that are affiliated with the Children’s Hospital Association (Lenexa, KS). Hospitals submit encounter-level data, including demographics, medications, and diagnoses based on International Classification of Disease, Tenth Revision, Clinical Modification (ICD-10-CM) codes. Data are deidentified at the time of submission and data quality and reliability are assured by joint efforts between the Children’s Hospital Association and participating hospitals.¹⁸

Study population

We included children 3 months to 18 years of age who were hospitalized (inpatient or observation status) at participating hospitals between January 1, 2017 and December 31, 2019 with a discharge diagnosis of an LRTI based on ICD-10-CM codes, which included bronchiolitis, viral and bacterial CAP, and other LRTIs as defined in prior studies (Supporting Information: Table 1).^19,20 Patients with a diagnosis of asthma were only included in the study if they also had an LRTI diagnosis. We excluded hospitals with incomplete data during the study period (n = 1) and those without length of stay (LOS) (primary outcome) in hours (n = 6); the remaining 42 hospitals were included. To define a cohort of generally healthy children, we excluded patients with complex chronic conditions²¹ and those not discharged to home. To focus on children who were not critically ill, we excluded children admitted directly from the emergency department (ED) to the intensive care unit (ICU) using billing data and those with complicated pneumonia based on ICD-10-CM codes.¹⁹ As the aim of this study was to identify patients with LRTIs who received antibiotics for suspected bacterial CAP, we excluded patients transferred to or from acute care in which initial antibiotic treatment was unknown and children with ICD-10-CM codes for other nonrespiratory bacterial infections that are commonly treated with antibiotics (e.g., urinary tract infection, sinusitis, acute otitis media; Supporting Information: Table 2).¹⁹ The study was deemed exempt from review by the author’s Institutional Review Board.

Explanatory variable

We evaluated the proportion of children hospitalized with an LRTI who were diagnosed with and treated for bacterial CAP. For brevity, we will refer to this proportion as the “CAP diagnosis rate” but note that it is specific to bacterial CAP. We defined children diagnosed with and treated for bacterial CAP as those who had an ICD-10-CM discharge diagnosis code for bacterial CAP¹⁹ (Supporting Information: Table 1) and received at least one dose of antibiotics commonly used for bacterial CAP¹⁶ on the first two calendar days of the encounter (Supporting Information: Table 3). We chose this strategy to focus on patients who initially presented with concern for bacterial CAP and who were ultimately diagnosed with bacterial CAP. This definition purposefully excluded patients who received only one dose of antibiotics and were later diagnosed with a nonbacterial LRTI.

Outcomes

Our primary outcome was LOS in hours, given its clinical significance to both families and hospitals.¹⁶ Additional clinical outcomes included rate of transfer to the ICU and 30-day readmission or ED revisits in three categories (all-cause, pneumonia-related, and adverse-drug reaction-related). We included these three categories because they each identify a different cause for a return visit: all-cause captures all visits, pneumonia-related captures potentially ongoing or initially missed pneumonia diagnoses, and adverse-drug reaction-related captures potential antibiotic-associated side effects or adverse-reactions, as defined and used in prior studies.^22,23 We additionally evaluated hospital costs and initial utilization of diagnostics, defined as the performance of chest radiographs (CXR) or blood tests (including complete blood count, c-reactive protein, procalcitonin, erythrocyte sedimentation rate, or blood culture) on the first 2 days of the encounter.

Statistical analysis

Demographic and clinical characteristics were summarized with descriptive statistics, including frequency with percentage, mean with standard deviation, and median with interquartile range (IQR). Each hospital’s proportion of children with LRTI who were diagnosed with and treated for bacterial CAP was adjusted using a generalized estimating equation model including a priori patient-level confounders including age, race, payor, and H-RISK,²⁴ which is a measure of severity of illness. Although race is a social and not biological construct, it was included given documented disparities in antibiotic prescribing practices in outpatient settings.²⁵ We then grouped hospitals based on their adjusted rates into high, moderate, and low CAP diagnosis hospital groups using an outlier status approach. Hospitals were identified as high (or low) outlier if the 95% confidence interval (CI) of their risk-adjusted rate did not contain the overall mean rate across hospitals.

Multivariable regression with generalized estimating equations was used to examine the association between low and high hospital diagnosis groups and outcomes. We evaluated differences in outcomes between low and high diagnosis groups because we wanted to compare outcomes between hospitals on both extremes of the diagnosing spectrum. All statistical analyses were performed using SAS v.9.4 (SAS Institute), and p < .05 were considered statistically significant.

RESULTS

Patient cohort

We identified 66,581 patient encounters with a discharge diagnosis of an LRTI that met study criteria (Figure 1). Most patients were less than 5 years old (87%), and diagnoses included viral bronchiolitis (62%), bacterial CAP (29%), viral CAP (9%), and other LRTI (1%; Table 1).

FIGURE 1.

Patient flow diagram.

TABLE 1.

Demographic and clinical characteristics of patients by CAP diagnosis group

	Overall	Low CAP diagnosis hospitals	Mod CAP diagnosis hospitals	High CAP diagnosis hospitals	p Value
N, hospitals	42	17	11	14
N, encounters	66,581	28,561	15,917	22,103
Age (years)
<2	44,798 (67.3)	19,368 (67.8)	11,053 (69.4)	14,377 (65)	<.001
2–4	12,308 (18.5)	5587 (19.6)	2583 (16.2)	4138 (18.7)
5–11	7379 (11.1)	2855 (10)	1748 (11)	2776 (12.6)
12–18	2096 (3.1)	751 (2.6)	533 (3.3)	812 (3.7)
Female sex	28,554 (42.9)	12,151 (42.6)	6817 (42.8)	9586 (43.4)	.20
Race/ethnicity
Non‐Hisp White	28,645 (43)	13,464 (47.1)	5928 (37.2)	9253 (41.9)	<.001
Non‐Hisp Black	13,917 (20.9)	5549 (19.4)	4765 (29.9)	3603 (16.3)
Hispanic	16,354 (24.6)	6468 (22.6)	3428 (21.5)	6458 (29.2)
Asian	2202 (3.3)	739 (2.6)	640 (4)	823 (3.7)
Other	5463 (8.2)	2341 (8.2)	1156 (7.3)	1966 (8.9)
Payor
Government	39,024 (58.6)	16,170 (56.6)	9655 (60.7)	13,199 (59.7)	<.001
Private	23,969 (36)	10,635 (37.2)	5629 (35.4)	7705 (34.9)
Other	3588 (5.4)	1756 (6.1)	633 (4)	1199 (5.4)
Encounter diagnosesa
Bronchiolitis	41,103 (61.7)	18,679 (65.4)	10,051 (63.1)	12,373 (56)	<.001
Bacterial CAP	19,089 (28.7)	6353 (22.2)	4589 (28.8)	8147 (36.9)
Viral CAP	5857 (8.8)	3312 (11.6)	1115 (7)	1430 (6.5)
Other LRTI	532 (0.8)	217 (0.8)	162 (1)	153 (0.7)
H-RISKb, mean (SD)	0.72 (0.46)	0.72 (0.49)	0.69 (0.41)	0.73 (0.46)	<.001

Note: Values represent N(%) unless otherwise specified.

Abbreviations: CAP, community-acquired pneumonia; LRTI, lower respiratory tract infection.

^aNot mutually exclusive.

^bMeasure of severity of illness.

Bacterial CAP diagnosis rate

Across all hospitals, the median adjusted proportion of children with an LRTI who were diagnosed with and treated for bacterial CAP was 27% (IQR: 25%–30%), with a range of 12%–42% (Figure 1). Hospitals were grouped into low (n = 14), moderate (n = 11), and high (n = 17) CAP diagnosis groups with median adjusted CAP diagnosis rates of 24% (IQR: 22%–25%), 27% (IQR: 26%–28%), and 30% (IQR: 30%–33%), respectively (Supporting Information: Figure 1). Patient demographics and clinical characteristics across hospital groups are displayed in Table 1.

Outcomes between low versus high bacterial cap diagnosis hospital groups

There was no difference in LOS, rates of ICU transfer, or cost between low and high CAP diagnosis groups (Table 2). High CAP diagnosis hospitals had a statistically higher rate of 30-day CAP-related ED revisits compared to low CAP diagnosis hospitals (0.6% [95% CI: 0.5, 0.7] vs. 0.4% [0.4, 0.5], p = .043), but no difference in CAP-related readmissions or all-cause or adverse drug reaction-related readmissions or ED revisits. Additionally, high CAP diagnosis hospitals had a higher rate of utilization of CXRs (58% [95% CI: 53, 62] vs. 46% [41, 51], p = .019) and blood tests (43% [95% CI: 33, 53] vs. 26% [19, 35], p = .046) compared to low CAP diagnosis hospitals.

TABLE 2.

Adjusted outcomes across CAP diagnosis hospital groups

	Low CAP diagnosis hospitals	Mod CAP diagnosis hospitals	High CAP diagnosis hospitals	p Value (low vs. high)
Clinical outcomes
LOS (hours)—mean	37.6 (35.7, 39.5)	39.8 (36.1, 43.8)	37.1 (34.1, 40.5)	.80
Transfer to ICU (%)	0.6 (0.4, 0.8)	0.7 (0.5, 1)	0.4 (0.3, 0.6)	.31
30‐day readmission/revisits (%)
All‐cause readmissions	4.8 (4.4, 5.2)	4.4 (3.8, 5.1)	4.3 (3.7, 5)	.15
All‐cause ED revisits	8.5 (7.2, 10.1)	6.3 (5.6, 7.2)	7 (6.4, 7.6)	.054
CAP‐related readmissions	0.8 (0.6, 1)	0.6 (0.5, 0.8)	0.7 (0.6, 0.9)	.66
CAP‐related ED revisits	0.4 (0.4, 0.5)	0.3 (0.2, 0.4)	0.6 (0.5, 0.7)	.04
Adverse‐drug reaction readmissions	0.4 (0.2, 0.8)	0.2 (0.1, 0.3)	0.3 (0.2, 0.4)	.65
Adverse‐drug reaction ED revisits	1.3 (1, 1.6)	1.2 (0.8, 1.6)	1.3 (1, 1.6)	.90
Financial outcomes
Costs ($)—mean	4479 (3795, 5287)	3590.3 (2725, 4730)	3874 (3181, 4720)	.31
Diagnostic utilization
Chest radiograph (%)	46.1 (41, 51.2)	48 (43.4, 52.7)	57.5 (52.7, 62.1)	.02
Blood testinga (%)	26 (18.7, 34.8)	30.1 (21.6, 40.3)	42.7 (33, 52.9)	.046

Note: Values represent mean or % (95% CI). Rates were adjusted for age, race, payor, and H‐RISK (measure of severity of illness).

Abbreviations: CAP, community‐acquired pneumonia; ED, emergency department; ICU, intensive care unit; LOS, length of stay.

^aincluding complete blood cell count, c‐reactive protein, erythrocyte sedimentation rate, procalcitonin, or blood culture.

DISCUSSION

In this multicenter study of children with LRTIs, there was substantial variation across hospitals in the proportion of patients who were diagnosed with and treated for bacterial CAP. High bacterial CAP diagnosis hospitals had similar clinical outcomes to low CAP hospitals and increased utilization of CXRs and blood testing. Our findings suggest potential opportunities for diagnostic and antibiotic stewardship in pediatric CAP.

Few studies have evaluated the proportion of children with LRTIs who are diagnosed with and treated for bacterial CAP in the hospital setting. Those studies used differing diagnostic criteria for LRTIs, which limit generalizability. One prospective study of children who presented to an ED in Switzerland with an LRTI (defined as fever, respiratory symptom, and respiratory sign) identified that 49% of children were diagnosed with bacterial CAP and received an antibiotic prescription.²⁶ In the Netherlands, a prospective study of children in the ED found that 25% of children with LRTI (defined as fever and cough or dyspnea) received antibiotics.²⁷ Our study builds upon this work by evaluating the bacterial CAP diagnosis rate among a hospitalized cohort of US children with LRTI, highlighting variability across hospitals, and exploring the association between bacterial CAP diagnosis rate and clinical outcomes as well as diagnostic utilization.

We found that more than one in four hospitalized children with an LRTI were diagnosed with and treated for bacterial CAP. Although it is difficult to determine what proportion of children truly have a bacterial CAP that warrants treatment, this proportion is higher than what the evidence supports. In a large prospective study among children hospitalized with clinical and radiographic-confirmed CAP that used the most robust diagnostics available, only 15% had a bacterial infection identified; this finding suggests that few children with clinical and radiographic CAP likely require antibiotics.¹⁰ Thus, we can extrapolate that even a lower proportion of children with LRTIs truly need antibiotics for bacterial CAP.

Our study demonstrated large variation across hospitals in the proportion of children with LRTIs who were diagnosed with and treated for bacterial CAP, with some hospitals diagnosing and treating CAP in more than 40% of patients with LRTIs. While there may be true differences in rates of bacterial CAP across sites, it is unlikely that this explains all the variability that we observed. Importantly, this variation was not associated with disparate outcomes. Specifically, hospitals that diagnosed and treated bacterial CAP less frequently did not have a longer LOS or higher all-cause or pneumonia-related readmission or revisit rate compared to hospitals that diagnosed bacterial CAP more. Thus, we did not find evidence that low CAP diagnosis hospitals are undertreating or missing diagnoses of bacterial CAP. In our study, the high CAP diagnosis group had a statistically higher rate of CAP-related ED revisits, which is likely not clinically significant (absolute difference only 0.2%) but may be related to billing practices or diagnostic momentum given no differences in all-cause or adverse drug reaction-related revisits.^28,29 Variability in care has been associated with decreased quality of care and increased resource utilization and costs.^30,31 Thus, opportunities exist across all hospitals, particularly those that diagnose bacterial CAP frequently, to improve the diagnosis and subsequent treatment of bacterial CAP. Further studies are needed to determine how clinicians at high bacterial CAP diagnosis hospitals evaluate, diagnose, and treat patients with LRTIs and how that differs from those in the low bacterial CAP diagnosis group. It is also important to evaluate how hospital culture may drive some of these practice differences. Given rising rates of antibiotic resistance, it is critical that we develop and implement evidenced-based practices for the diagnosis and management of pediatric CAP.^29,32,33

We found that hospitals that diagnosed bacterial CAP more frequently also obtained CXRs and blood tests more commonly, with approximately half of children with LRTIs receiving CXRs and blood tests in some hospitals. This finding aligns with a prior PHIS study of children discharged from the ED with a respiratory illness or fever from 2008 to 2018, which found a direct correlation between hospital-level CXR use and CAP diagnosis rates,³⁴ but contrasts an older PHIS study of children with respiratory illness (including upper respiratory infections) or fever in the ED which demonstrated no association.³⁵ Current national guidelines for pediatric CAP recommend obtaining a CXR for hospitalized children but recommend against routine blood tests for children without severe CAP, although there is no explicit definition of severe CAP.³⁶ Increased CXRs and blood tests in some hospitals may be appropriate if there are truly more cases of CAP (and severe CAP). In line with prior literature, it is also possible that liberal use of CXRs and blood tests for children with LRTIs led to an increase in the diagnosis of bacterial CAP, potentially from borderline, incidental, or nonspecific findings.^37,38 Our findings along with recent studies demonstrating the low utility of CXRs and blood tests in LRTIs including CAP, and the potential harms associated with these tests (e.g., radiation, pain, and sleep disturbance of blood draws), suggest opportunities for diagnostic stewardship.^39–44

Our study should be interpreted in the context of several limitations. First, our administrative data source does not contain clinical data (e.g., clinical history, physical examination findings, and diagnostic results such as CXR, respiratory viral and blood tests) that may be important covariates and relevant to clinical decision-making surrounding the diagnosis, use of diagnostics, and treatment. Thus, we are limited in our ability to evaluate the appropriateness of bacterial CAP diagnoses or use of diagnostic testing among children hospitalized with LRTIs. Second, we cannot be sure that antibiotics prescribed were specifically used for CAP. However, we minimized the impact of this potential limitation by focusing on antibiotics commonly prescribed for CAP and excluding patients with a discharge diagnosis for other common bacterial infections. Third, there is potential for misclassification bias for patients with an LRTI or bacterial CAP. However, we used definitions derived from prior literature to minimize this limitation.^19,20 Fourth, although the study sample size was large, because our unit of analysis was hospital groups rather than individual patients, we may be underpowered to detect clinically significant differences between the hospital groups. Fifth, while we estimated return visits related to CAP based on CAP-related ED revisits and readmission rates, overall rates are likely underestimated as we were not able to include visits to primary care, or urgent care or hospital systems that are not within PHIS. Finally, because these data are from tertiary care children’s hospitals, our findings may not be generalizable to other settings.

CONCLUSION

For children hospitalized with LRTIs, the rates of diagnosing and treating bacterial CAP varied widely across hospitals. Hospitals with higher rates of bacterial CAP diagnoses more frequently obtained CXRs and blood tests. Variability in bacterial CAP diagnosis rates with similar outcomes between low and high CAP diagnosis hospitals highlights potential opportunities for diagnostic and antibiotic stewardship in pediatric CAP.