Payer-Related Sources of Variation in Febrile Infant Management Before and After a National Practice Standardization Initiative

We conducted a retrospective multicenter crosssectional study using the Pediatric Health Information System (PHIS) database. PHIS includes deidentified daily billing and administrative data from 52 free-standing pediatric hospitals affiliated with the Children’s Hospital Association (Lenexa, KS). We included data from 37 hospitals; 15 hospitals submitting only inpatient encounters to PHIS or with known data quality concerns (eg, improper coding, incomplete records, or inconsistent use of billing codes) were excluded. Data are deidentified at the time of entry into the database and are subject to rigorous quality checks before inclusion. Patients can be tracked across encounters using a consistently encrypted medical record number. The PHIS database contains demographic information, and diagnostic, procedural, and billing codes for laboratory, imaging, and pharmacy services. Vital signs and laboratory test results are not included. This study was deemed nonhuman subjects research by our hospital’s institutional review board.

We included infants ≤60 days of age with an ED visit or hospitalization (observation or inpatient) and a diagnosis of fever between December 1, 2015 and November 30, 2018. We identified infants with fever based on the presence of an admission or discharge diagnosis code per previously described methods.³  We crosswalked previously validated International Classification of Diseases, Ninth Revision (ICD-9) codes for infant fever to ICD-10 codes.^4,5  Therefore, patients were included if one of the following ICD-10 codes were present: R50.2 (drug induced fever), R50.9 (fever unspecified), P81.8 (other specified disturbance of temperature regulation of newborn), or R50.81 (fever presenting with conditions classified elsewhere).

We excluded infants at risk for complicated clinical courses or comorbidities that may warrant testing or treatment not related to neonatal fever, including infants with any complex chronic condition, as described by Feudtner et al,⁶  and those admitted to an ICU. Birth encounters were excluded because infants with fever in the first few days of life may receive unique testing and treatment related to the risk of early onset neonatal sepsis. Nonstandard discharges (such as those infants transferred to other facilities) and direct admissions (ie, infants transferred directly to the inpatient unit) at participating sites were excluded because of the risk of incomplete data.

Our primary exposure was time as we wished to understand changes in management related to sociodemographic characteristics pre- and post- REVISE. We divided our study time into 3 periods consistent with the REVISE project: (1) pre-REVISE from December 1, 2015 to November 30, 2016, (2) REVISE intervention from December 1, 2016 to November 30, 2017, and (3) post-REVISE from December 1, 2017 to November 30, 2018.

Our primary outcome was proportion of patients receiving laboratory testing, whereas our secondary outcomes were a proportion of infants presenting to a hospital with febrile illness who were hospitalized. These outcomes were chosen as they align with the goals of the REVISE initiative.²  We defined laboratory testing as obtainment of the generally accepted minimal laboratory evaluation for febrile infants at the time of our study, which was prior to the American Academy of Pediatrics published guideline on febrile infant management. We identified tests using billing codes, and receipt of laboratory testing was included only if all 3 components of testing were obtained: (1) urine studies, including either urinalysis or urine culture, (2) complete blood count, and (3) blood culture.^2,7,8  Lumbar puncture was not included, because the utility of diagnostic lumbar puncture has been controversial among well-appearing febrile infants.⁹ 

We identified patient sociodemographic factors and clinical characteristics, including gender, age, primary payer type, race and ethnicity, and illness severity. We defined age as days of life and grouped infants aged 0 to 28 days and infants aged 29 to 60 days together to account for inherent differences in risk of infections among these age groups that may influence clinical practice and patient management. Within the PHIS database, patient race and ethnicity is reported as non-Hispanic White, non-Hispanic Black, Hispanic, Asian, Native American, or other. Due to low numbers of patients identified as Asian or Native American, we combined these groups of patients with the other category. The primary payer was identified as either government, private, or self-pay. We adjusted for severity of illness using a case mix index available in PHIS based on the Hospitalization Resource Intensity Score for Kids (H-RISK).¹⁰  In addition to the above patient characteristics, we identified hospital location by region, grouped as Midwest, Northeast, South, and West, to account for regional differences in management that have been previously described.^4,5 

To better understand the relationship between our study period and the REVISE initiative, we performed a subanalysis of patients seen only at REVISE-participating hospitals. We compared changes in disparities over time at non-REVISE hospitals to changes over time at REVISE-participating hospitals to better delineate the interaction between time, disparities, and participation in REVISE.

We also performed a subanalysis of infants ≤28 days to assess sociodemographic differences in this younger group of infants, as one might expect less practice variation in management given younger infants’ inherent increased risk of invasive bacterial infections.

We summarized continuous variables using medians and interquartile ranges (IQRs), and categorical variables by frequencies and percentages. Differences in categorical variables between pre-REVISE, REVISE, and post-REVISE periods were assessed using a χ² test for independence, whereas differences in continuous variables were assessed using a Kruskal-Wallis test. We used generalized linear mixed modeling techniques assuming an underlying binomial distribution to identify sociodemographic characteristics associated with laboratory testing and hospital admission and to compare the adjusted odds ratios (aOR) of laboratory testing and hospital admission across the pre-REVISE, REVISE, and post-REVISE periods. Odds were adjusted for patient- and hospital-level characteristics. All models included a random hospital-level effect to account for any clustering of discharges at the same institution. We assessed interactions between patient-level factors and hospital location with each period to determine if changes in outcomes over time were varied by patient or hospital characteristics. Analyses were performed using SAS, Version 9.4 (SAS Institute, Cary, NC), and P values < .05 were considered statistically significant.

We identified a total of 32 572 febrile infants from 37 PHIS-reporting hospitals. Table 1 displays patient characteristics, hospital characteristics and patient outcomes across all 3 periods. Most infants were male, > 28 days old, non-Hispanic White, and had a government payer. Most (74%) infants received laboratory testing and approximately half (49%) were hospitalized. Over time, the proportion of infants who identified as non-Hispanic White, who had private insurance, and who had a higher severity of illness increased. The proportion of infants who had laboratories obtained increased, while the proportion of hospitalized infants decreased over time (Supplemental Fig 2).

Approximately one-third of our total cohort, 10 749 infants, presented in the pre-REVISE period. A description of patient and hospital characteristics in the pre-REVISE period is provided in Table 2. Most infants were aged 29 to 60 days, and most infants had government insurance (61%). The plurality of infants were reported as Non-Hispanic White (45%). A majority (64%) were seen at REVISE-participating hospitals. Compared with infants seen at non-REVISE hospitals, patients seen at REVISE-participating hospitals were more likely to be Hispanic and non-Hispanic Black. Additionally, infants seen at REVISE-participating hospitals were more likely to have a government payer or be self-pay. REVISE-participating hospitals were more likely to be in the South.

In the pre-REVISE period, after adjusting for patient and hospital-level characteristics, we found differences in laboratory testing and hospital admission were associated with both age and payer (Table 3). Infants ≤28 days of age had higher odds of undergoing laboratory testing compared with older infants (aOR 1.24, 95% confidence interval [95% CI] 1.12–1.37, P < .001). Patients with government insurance (aOR 0.67, 95% CI 0.6–0.75, P < .001) or self-pay (aOR 0.43, 95% CI 0.33–0.56, P < .001) had lower odds of laboratory testing compared with infants with private insurance.

Compared with older infants, febrile infants ≤28 days of age also had higher odds (aOR 7.59, 95% CI 6.84–8.42, P < .001) of hospital admission. Infants described as self-pay had lower odds of being admitted (aOR 0.38, 95% CI 0.28–0.51, P < .001) compared with infants with private insurance. Importantly, the association of age and with variation in laboratory testing and hospital admission were similar among infants seen at non-REVISE versus REVISE-participating hospitals.

In our sub analysis of infants ≤28 days, we found similar changes in laboratory testing and hospitalization by payer-type. Specifically, infants ≤28 days with government insurance (aOR 0.60, 95% CI 0.49–0.73, P < .001) or self-pay (aOR 0.30, 95% CI 0.19–0.48, P < .001) had lower odds of laboratory testing compared with infants with private insurance. Similarly, infants ≤28 days with government insurance (aOR 0.68, 95% CI 0.55–0.84, P < .001) or self-pay (aOR 0.27, 95% CI 0.17–0.44, P < .001) had lower odds of hospitalization compared with infants with private insurance. We also found no difference in payer-associated variation in care among infants ≤28 days at non-REVISE versus REVISE participating hospitals.

Both age and payer continued to be associated with differences in febrile infant management post-REVISE. Participation in REVISE was not associated with changes in febrile infant management overtime. While we did note significant reductions in disparities over time among self-pay infants at REVISE participating centers (aOR of laboratory obtainment among self-pay infants was 0.39 [IQR 0.29–0.53] pre-REVISE versus 0.71 [IQR 0.53–0.93] P < .001, compared with privately insured infants) we observed similar changes in disparities across when comparing non-REVISE versus REVISE- participating hospitals. Specifically, the interaction between time and payer was not statistically significant between non-REVISE and REVISE- participating hospitals for either laboratory obtainment (P = .09) or hospitalization (P = .67; [Fig 1]).

This study describes payer-related sources of variation in febrile infant management and the impact of a national standardization initiative on sociodemographic sources of variation. We noted infants with private insurance were more likely to have laboratories obtained and more likely to be hospitalized compared with infants with self-pay or government payers. Over time, payer-related disparities were similar among both non-REVISE and REVISE-participating hospitals. Overall, our findings suggest that disparities exist in febrile infant care, and that participation in standardization projects, such as REVISE, may not be associated with disparity mitigation.

After modeling we observed differences in the pre-REVISE group in laboratory testing and hospital admissions for children based on payer type, but not their race or ethnicity. Although we initially hypothesized that there would be differences in management related to an infant’s race or ethnicity, we found that when socioeconomic status (as related to insurance status) was controlled for, race and ethnicity did not significantly contribute to variation in management.

Compared with infants with private insurance, self-pay and publicly insured infants with fever were 57% and 33% less likely to receive the typically recommended laboratory testing, respectively. Similarly, self-pay infants were 62% less likely to be admitted to the hospital. While literature on payer-associated disparities in febrile infant management is sparse, our findings mirror payer-associated care differences, which have been seen in other pediatric and adult conditions.^11–13 

We speculate our findings are because of the many complex and interwoven factors. Specifically, among self-pay patients, our findings may reflect parental wishes to do less testing given high out-of-pocket costs. Although we did not directly study provider implicit bias, this bias is known to influence patient care and is another important consideration.^14,15  Factors, such as perceived access to a primary care provider, limited English proficiency, and race and ethnicity are associated with variation in care in other pediatric conditions,^16,17  and the amalgam of these and other factors may best be represented by an infant’s insurance status. Additionally, provider and family shared-decision making may be affected by insurance status whereby caregivers of children with private insurance advocate for their infant to have more testing done, or providers anticipate that these caregivers would want more testing done compared with caregivers of self-pay or government insured infants. Whereas data on outcomes of shared decision-making and associated insurance status is sparse, increased resource use among non-Hispanic White children, compared with minority children, is well documented in the literature.^18,19 

It is also possible that our results reflect payer-associated coding bias (ie, infants with certain insurance types were more likely to be miscoded for fevers).

Based on prior work by Aronson et al,⁵  we know that 15% of infants ≤28 days with febrile infant codes don’t receive any testing, however differences in coding of fever based on payer status have not been previously studied. Although our findings were upheld in the subanalysis of infants ≤28 days, our current study is unable to further clarify to what degree these findings are related to differences in coding practices (ie, payer-associated miscoding) versus undertesting of febrile self-pay or government insured infants. Our results highlight the need for further studies, including qualitative provider and parent studies as well as more granular investigations into potential payer-associated bias in febrile infant coding, to better explain our results and understand socioeconomic disparities in febrile infant management.

Over time, changes in disparities in laboratory testing and hospitalization were not directly associated with participation in REVISE. Idea dissemination, dispersion, and a general awareness of best-practices may have spread from REVISE-centers to non-REVISE centers and may account for the changes we saw over time. In addition, local care guidelines unrelated to REVISE, even among REVISE-participating centers, may have influenced our results. Finally, participation in other national efforts aimed at delivering high-value medicine may have been associated with the change in outcomes and payer-related disparities seen at non-REVISE hospitals.

Although there is data to suggest quality improvement (QI) projects alone may decrease health disparities,^20,21  other research has shown that that QI interventions alone may not always be effective at closing disparity gaps.^22–24  It is likely that a more focused and integrated approach between QI and health disparities research is needed to ameliorate health disparities.^23,25  The goal of REVISE was to standardize practice for febrile infant management, and Biondi et al’s work supports that participation in REVISE was indeed associated with improvements in febrile infant care.²  We speculate that perhaps payer-related disparities would be reduced to a greater extent among REVISE-hospitals if this initiative had highlighted socioeconomic differences in management. Further studies are needed to better understand the potential of standardization-related QI projects in mitigating care disparities.

Our findings should be considered in the context of several limitations. First, utilizing the PHIS database allowed us to examine a large population of children across the nation that we could control for hospital-, regional-, and many patient-level factors. The retrospective administrative data within the PHIS likely omits relevant confounding variables, including patient-level characteristics (such as clinical appearance including patient vital signs). Second, although we controlled for severity of illness, it is possible that prehospital differences in unaccounted for covariates (eg, access to care) may have impacted the treatment they received in the ED and hospital setting. Third, like any database study, there is the potential for coding errors. For example, our study may have included some patients who did not have a febrile illness but were coded as such. These concerns were mitigated as much as possible by excluding hospitals with known coding or data quality concerns. It is also possible that reductions in disparities were associated with REVISE implementation but were not captured in our study because they occurred at excluded hospitals or at nonchildren’s hospitals, which are not included within the PHIS database. Finally, race and ethnicity may be prone to misclassification, as there is no known standard reporting method for reporting race and ethnicity among PHIS-reporting hospitals. Future studies are needed to investigate best practices regarding reporting of patient race and ethnicity.

Hospital payer contributes to disparities in febrile infant care. It is unclear if participation with the REVISE initiative mitigated care disparities, as we saw similar patterns in disparities over time among both participating and nonparticipating centers.

Our results suggests that a more deliberate approach to addressing sociodemographic disparities in future QI projects, including exploring reasons for these disparities, is needed to improve health equity.