Variation in Early Inflammatory Marker Testing for Infection-Related Hospitalizations in Children

In this multicenter cross-sectional study of children with an infection-related hospitalization (inpatient and observation), we used the Pediatric Health Information System (PHIS). PHIS is an administrative and billing database of 51 freestanding tertiary care pediatric hospitals in the United States that are affiliated with the Children’s Hospital Association (Lenexa, KS). Patient data are deidentified in PHIS; encryption of patient identifiers, however, allows for tracking of individual patients across multiple hospital visits. The current study included data from a total of 48 hospitals, with 3 hospitals excluded for incomplete data. Because we used deidentified data, this study was not considered human subjects research by the policies of the local institutional review board.

Children 0 to 17 years of age with an index hospitalization for infection at a PHIS-participating hospital from January 1, 2016, to December 31, 2017, were eligible for inclusion. Hospitalizations for infection were identified by using All Patient Refined Diagnosis-Related Groups (APR-DRGs) version 32 (3M Corporation, St Paul, MN). APR-DRGs are a patient classification scheme that groups medical patients on the basis of the principal diagnosis and incorporates severity of illness and risk of mortality using demographics and comorbidities. Although originally designed for inpatient stays, the PHIS database uses the APR-DRG grouper for both inpatient and observation stays because dedicated observation units are infrequent in pediatrics and observation stays are defined retrospectively by payer. We examined the following APR-DRG categories: infections of the upper respiratory tract (APR-DRG 113); major respiratory infections and inflammations (APR-DRG 137); bronchiolitis and respiratory syncytial virus (RSV) pneumonia (APR-DRG 138); pneumonia (APR-DRG 139); major gastrointestinal and peritoneal infections (APR-DRG 248); nonbacterial gastroenteritis (APR-DRG 249); osteomyelitis, septic arthritis, and other musculoskeletal infections (APR-DRG 344); cellulitis and other bacterial skin infections (APR-DRG 383); kidney and urinary tract infections (APR-DRG 463); postoperative, posttraumatic, and other device infections (APR-DRG 721); viral illness (APR-DRG 723); and other infectious and parasitic diseases (APR-DRG 724).

We excluded transfers in because of the potential inability to capture diagnostic test use from the transferring hospital. Finally, we excluded patients with cancer and immunodeficiency using Feudtner’s complex chronic conditions (CCCs)¹⁹  as well as hospitalizations that included ICU stays because these children may have more complicated infections and because patterns of usage of inflammatory markers may differ from those seen on general inpatient teams. Children within other CCC categories were retained within analyses.

Inflammatory marker testing was defined as obtaining either a ESR or CRP test. We focused on early testing, defined as during the first 2 days of the hospitalization, because we sought to capture early (emergency department or inpatient) test use across a broad range of infections rather than examine the impact of late or repeat testing on hospitalization outcomes. We chose to examine early testing because a previous study of diagnostic testing revealed that overuse occurred more frequently during initial testing compared with repeat testing.⁴  Additionally, this mitigated the risk of identifying a misleading effect-cause relationship because it would be challenging to determine if children who stay longer have more opportunities to get testing versus testing having a causal role in prolonged LOS. We describe both overall (ie, all PHIS hospitals) and individual hospital testing rates. Concomitant testing was defined as receipt of ESR and CRP tests during the first 2 days of the hospitalization.

Outcomes included variation in rates of inflammatory marker testing across hospitals. We also examined hospital-level LOS measured in days, all-cause 30-day hospital readmissions, and cost in US dollars. The time frame of 30 days was chosen to measure subsequent visits associated with treatment failure, antibiotic-associated adverse effects, or invasive bacterial infection. Cost of the index hospitalization included use from the emergency department visit and hospitalization. In PHIS, costs are estimated from charges by using hospital year–specific cost-to-charge ratios.

We examined demographic characteristics, including age, sex, race and/or ethnicity, and primary payer. We examined patient characteristics, including the number and types of medical complexity using CCCs.

We calculated unadjusted hospital-level inflammatory marker testing rates for individual infection APR-DRGs. We then adjusted testing rates for age, presence of a CCC, and severity using generalized linear models, controlling for clustering of patients within a hospital using a random intercept for each hospital. For generation of the heat maps, we grouped each hospital by quartile on the basis of adjusted tested rates across each infection APR-DRG. To group hospitals with similar inflammatory marker testing rates, we used k-means clustering, assigning each hospital to 1 of 3 groups on the basis of similarities in adjusted diagnostic testing rates across infection APR-DRGs.²⁰  Canonical discriminant analysis was used to determine the number of groups. We used descriptive statistics to describe patient and hospital characteristics overall and for each of the 3 testing groups. Comparisons in patient and hospital characteristics across hospital testing groups were conducted by using the Kruskal-Wallis test. Generalized linear mixed models were used to examine the association of hospital testing group and outcomes, with adjustment for age, presence of a CCC, and severity. Severity was defined by using hospitalization resource intensity scores for kids (H-RISKs),²¹  which was developed to quantify severity of illness among hospitalized children and used to assign relative weights to each APR-DRG and severity-of-illness level, facilitating comparison across APR-DRG groups. All statistical analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC), and P values <.05 were considered statistically significant.

We included 55 771 hospitalizations for infection from 48 hospitals (Table 1). In 7945 (14.2%), inflammatory marker testing was obtained in the first 2 days of hospitalization. The median age was 1 year (interquartile range [IQR] 0–6 years). A majority of patients were male and non-Hispanic white and had government insurance. Approximately 22% of patients had a comorbid CCC. CRP testing rates varied from 2.4% for bronchiolitis and RSV pneumonia to 57.7% for osteomyelitis, septic arthritis, and other musculoskeletal infections. ESR testing rates varied from 0.4% for bronchiolitis and RSV pneumonia to 50.4% for osteomyelitis, septic arthritis, and other musculoskeletal infections. Among patients with a CRP test, ∼45.9% also had an ESR test obtained (ie, concomitant testing of ESR and CRP), with the greatest rates of concomitant testing occurring for osteomyelitis, septic arthritis, and other musculoskeletal infections (84.4%), followed by major gastrointestinal and peritoneal infections (56.1%) (Supplemental Table 3). The median unadjusted LOS across hospitals was 2.0 (IQR 1.7–2.2) days, the median unadjusted 30-day all-cause readmission rate across hospitals was 7.3% (IQR 5.8%–8.8%), and the median unadjusted cost across hospitals was $2822 (IQR $1486–$5530).

We observed substantial variation in diagnostic testing across hospitals (Fig 1). Hospitals that obtained inflammatory marker testing frequently for one infection appear to test more frequently for other infections. For example, 6 hospitals tested above the median in all 12 diagnosis groups, and 12 hospitals tested below the median in all 12 diagnosis groups. Hospitals that tested more often for CRP also tested more often for ESR across infection subtypes. We observed similar patterns when examining concomitant testing (ie, hospitals that obtained concomitant testing for one infection appeared to obtain concomitant testing for other infections) (Supplemental Fig 3).

Using k-means clustering, we grouped hospitals into 1 of 3 groups (labeled A for the highest-testing hospitals, B for the moderate-testing hospitals, and C for low-testing hospitals) on the basis of similarities in adjusted diagnostic testing rates. The groups were composed of 6, 13, and 29 hospitals, respectively. Group A hospitals tended to perform more inflammatory marker (CRP or ESR) tests than hospitals in groups B and C (37.4% vs 18.0% vs 10.4%; P < .001) (Table 1, Fig 2). Although infants 0 to 1 month of age are frequently considered a distinct population regarding testing practices, differences in the proportions of inflammatory marker tests by group were similarly observed for these infants (Groups A–C: 19.4% vs 10.8% vs 5.9%; P < .001). We observed statistically significant but small differences across cluster groups in the distribution of patient demographic and clinical characteristics, such as age, race and/or ethnicity, payer, number and type of chronic conditions, and mean H-RISK (Table 1). Within our adjusted models, we observed a significant difference in mean LOS across testing groups. The hospitals in the highest-testing group (A) had longer adjusted LOS compared with hospitals in groups B and C (3.24 vs 2.85 vs 2.82 days; P < .01) (Table 2). There was no significant difference in adjusted mean 30-day readmission rates or costs.

In this multicenter cross-sectional study of early inflammatory marker testing in children, we observed wide variation in inflammatory marker use in children hospitalized with infections. We identified 2 major findings: (1) hospitals that obtained early inflammatory marker testing frequently for one infection obtained early inflammatory marker testing more frequently for other infections, and (2) hospitals with greater early inflammatory marker testing had longer LOS without a significant effect in readmission rates. These findings suggest opportunities to promote diagnostic stewardship.

As expected in our current study, inflammatory marker testing varied widely across infections. Although inflammatory markers were obtained infrequently for bronchiolitis (consistent with guidelines²² ), they were obtained with greater frequency for other infections, such as osteomyelitis, septic arthritis, and other musculoskeletal infections, in which studies have clearly defined evidence-based recommendations for use of inflammatory markers in diagnosing (eg, Kocher criteria²³ ) and guiding antibiotic therapy, including transitions to oral antibiotics. Inflammatory markers are used to assist in the diagnosis and management of infections and inflammatory conditions as well as to promote antimicrobial stewardship and reduce unnecessary exposure to antibiotics.^14,24,25  However, because CRP and ESR testing lack sensitivity and specificity, their utility in assisting with diagnosis lies primarily in cases in which there is either high or low pretest probability of a bacterial infection. Although diagnostic testing may help clinicians to exclude bacterial illnesses (ie, reduce antibiotic prescriptions) or transition to oral antibiotics sooner (ie, reduce exposure to parenteral antibiotics), overreliance on routine inflammatory marker testing may contribute to increased health care use and costs. For example, Kainth and Gigliotti¹⁷  previously reported that concomitant ESR and CRP testing within a large academic center resulted in increased expenditures without substantial clinical benefit. In other studies, authors describe a cascade effect, with practitioners obtaining increased rates of consultations, tests, and referrals in the setting of false-positive inflammatory marker testing.²⁶  Consequently, in future studies, researchers should seek to define best practices for inflammatory marker testing and investigate strategies that balance diagnostic stewardship with antimicrobial stewardship to tackle health care spending.

Our current study highlights how several hospitals with increased inflammatory marker testing for one infectious diagnosis tend to have increased inflammatory marker testing for other infectious diagnoses, including increased rates of concomitant testing. Although variation in testing across conditions is desirable (ie, variation based on differential evidence), substantial variation in adjusted testing rates across hospitals is perhaps more worrisome and highlights an opportunity to standardize aspects of care across institutions, especially where these differences are associated with differences in hospitalization outcomes (eg, increased LOS). The reasons behind increased intensity of testing at some hospitals are likely numerous and include both individual provider factors, such as provider experience, practicing defensive medicine, and low tolerance of diagnostic ambiguity, as well as systems factors.²⁷  Systems factors, such as local testing culture, may impact diagnostic test use at an organizational level and contribute to patterns in overuse similar to that observed in our study. For example, local protocols and/or policies, driven by factors such as the training environment or the intensity of services hospital wide (eg, more oncology patients or patients with immunodeficiency with limited ability to mount a fever response), may influence testing patterns at individual institutions. These results highlight that in addition to addressing individual testing behaviors, future interventions to impact diagnostic test use should also address testing culture within health care systems.²⁸ 

In our study, increased inflammatory marker testing was associated with increased LOS, without concomitant reductions in 30-day readmissions or association with costs. This finding is consistent with a growing body of literature revealing that variation in testing is associated with increased health care use.^29–31  Although differences in LOS in adjusted analyses were modest in our study (ie, <12 hours), one cannot underestimate the impact that improving LOS can have on patients, families, and health care systems. For example, reducing LOS may help alleviate some of the emotional and financial stress that patients and their families experience, even for hospitalizations for transient illnesses, while simultaneously reducing an individual patient’s risk of obtaining false-positive results that require additional follow-up.^32–38  Additionally, reducing LOS can improve hospital efficiency and may potentially lead to cost savings.³⁹  Our observation of no statistical differences in readmissions and cost is not wholly unexpected. Readmissions are overall uncommon in children, and although costs did decrease in a relatively linear manner across high- to low-testing institutions, these differences were not statistically significant. Our findings for costs likely reflect the fact that LOS is the predominant driver of pediatric inpatient costs and that institutional intensity of testing and services is more likely to influence LOS compared with cost. Taken together, our findings further suggest that increased resource use is not synonymous with improved health care delivery and that there are opportunities to reduce inflammatory marker testing, including reducing rates of concomitant testing.

Efforts such as the Choosing Wisely campaign have led to enhanced recognition of diagnostic overuse and initiation of local and national quality improvement initiatives aimed at reducing overuse. Recent studies reveal that implementation of quality improvement methodology and clinical practice guidelines may encourage reductions in unnecessary diagnostic testing. For example, use of quality improvement initiatives, such as education and standardized communication, has led to reductions in electrolyte and complete blood cell count testing as well as reductions in chest radiography for asthma.^40–42  Similarly, use of clinical practice guidelines has effectively lead to reductions in unnecessary bronchodilator, antibiotic, and chest radiography use in pediatric bronchiolitis.^43,44  Although future studies are needed to define best practices for inflammatory marker testing and to outline achievable benchmarks for testing, use of quality improvement methodology and clinical practice guidelines may be effective strategies to reduce unnecessary inflammatory marker testing.

Our study should be interpreted in the context of several limitations. First, the PHIS database is an administrative database and does not contain data pertaining to clinical decision-making surrounding diagnostic testing. Consequently, we are limited in our ability to evaluate the appropriateness of diagnostic testing (ie, to determine if diagnostic test obtainment truly impacted the decision to start an antibiotic or the decision of when to transition to oral therapy). Additionally, we are unable to examine how local protocols and/or policies influence testing patterns at individual institutions because these data are not collected by PHIS. We sought to broadly describe variation in inflammatory marker use; however, procalcitonin testing was obtained infrequently across PHIS-participating hospitals during our study period and was not included in the analyses despite its demonstrated improved sensitivity and specificity compared with CRP testing for identifying febrile infants.^45,46  Although we accounted for factors such as age, presence of a complex chronic condition, and APR-DRG severity of illness in our analyses, unaccounted-for differences in patient characteristics could certainly have contributed to variation and observed differences in the relationship between test variation and hospitalization outcomes across groups. Our focus on variation in initial testing during the first 2 days of hospitalization was only one aspect of pediatric diagnostic test overuse, and in future evaluations, researchers should focus on examining the impact of repeat diagnostic test obtainment on hospitalization outcomes.

For children hospitalized with infection, inflammatory marker testing varies widely across infection types and hospitals. Hospitals with higher inflammatory testing for one infection tend to test more frequently across other infection diagnoses and have longer LOS, suggesting a culture of overuse at some hospitals. Our results highlight the need to define best practices for diagnostic test use and the need for future quality improvement initiatives centered on optimizing diagnostic stewardship processes to improve health care value.