Disparities in Guideline Adherence for Febrile Infants in a National Quality Improvement Project

This was a planned secondary cross-sectional analysis of an international, multisite quality improvement (QI) collaborative implementing the AAP’s CPG for febrile infants (Reducing Excessive Variability In Infant Sepsis Evaluation [REVISE] II).²⁴  This study was approved by the AAP’s Institutional Review Board as exempt with a waiver of informed consent. REVISE II took place at 103 hospitals across the United States and Canada and aimed to improve adherence to several recommendations from the AAP CPG for febrile infants. Participating sites enrolled as multidisciplinary teams representing inpatient and emergency medicine team members (eg, physicians, nurses, pharmacists). Sites were provided an extensive intervention bundle to support implementation of the measures, including real-time data review, collaborative-level data support, education, and resources specific to the CPG, smart-phrases and verbal scripts for shared decision-making (SDM), and templates for discharge instructions in English and Spanish, as described previously.²⁴  Adherence to measures by race and ethnicity were tracked and presented at study-wide webinars.

There were 4 primary measures, 4 secondary measures, and 5 balancing measures (Supplemental Table 3), derived from a subset of the 21 key action statements and 40 recommendations within the CPG.²⁴  Primary measures focused on opportunities to “safely do less,” specifically to reduce unnecessary lumbar punctures, antibiotics, hospitalizations, and to shorten hospital length of stay. Secondary measures focused on significant practice changes including evidence-based opportunities for SDM with families, appropriate follow-up from the emergency department (ED), and oral antibiotic use for infants with presumptive urinary tract infections. Balancing measures sought to ensure that infants that required a more extensive work-up were still receiving the appropriate evaluation and management.

REVISE II project enrollment began concurrently with guideline release in August 2021. Enrolled sites collected 12 months of data in a baseline period (November 2020 to October 2021). The QI intervention period launched in November 2021 and ran prospectively through October 2022. Infants 8 to 60 days old were eligible for the QI intervention if they presented to a participating institution with a temperature ≥38°C. Exclusion criteria mirrored the AAP CPG exclusions.²³  Because of a high proportion of missing race and ethnicity data (684 of 691, 99%), infants from Canadian hospitals were excluded from this secondary analysis, resulting in a final sample of 99 US hospitals.

Our primary outcome was adherence to REVISE II measures. The independent variables were race and ethnicity, categorized as non-Hispanic white (hereafter white infants), non-Hispanic Black (hereafter Black infants), Hispanic/Latino, and other.^10,25,26  Because of smaller frequencies within each racial category, American Indian or Alaska Native and Asian or Native Hawaiian or other Pacific Islander were categorized as “other” and analyzed together.²⁷  Race and ethnicity were identified retrospectively using medical record review. The dependent variable was adherence to each measure, assessed dichotomously within the baseline and intervention period. At the end of the study period, we queried sites’ institutional policies for ascertaining race and ethnicity, including if that had changed throughout the study (Supplemental Table 4).

We compared racial and ethnic differences in adherence to measures using differences in the proportion of patients meeting the metric. Race and ethnicity were compared with the overall proportion meeting the metric; therefore, each racial and ethnic group was compared with the overall cohort and not to each other. This approach intentionally does not use the white population as the reference category to decenter whiteness.

Several of the dependent variables (3 of the 4 primary measures, 3 of 4 secondary measures, and the balancing measure of appropriate evaluation for infants 22–60 days) required obtaining at least 1 inflammatory marker (procalcitonin, c-reactive protein, or absolute neutrophil count) for inclusion in the denominator of the measure. Infants who did not have at least 1 inflammatory marker (IM) assessed could not be appropriately classified as having been adherent or not to downstream measures (eg, to qualify for the primary measure of appropriately not obtaining cerebrospinal fluid in infants 29 to 60 days, the infant had to have had at least 1 IM obtained that was normal). As such, infants without an IM were excluded from these measures. To assess for bias in measure inclusion, racial and ethnic differences in the proportion of infants who had IMs obtained were assessed. Proportion differences for the balancing measure of invasive bacterial infections (ie, bacteremia or bacterial meningitis) were not included because of low numbers of cases. Analyses were performed using R version 4.2.2 (Vienna, Austria).

Our sample included 16 961 infants. Overall, 1616 (10%) infants were Black, 4447 (26%) were Hispanic/Latino, 7055 (42%) were white, 2375 (14%) were categorized as other race and ethnicity, and 1469 (9%) had unknown race and ethnicity (Table 1). The proportions of qualifying infants by race and ethnicity for each measure are in Table 2. Raw proportions by race and ethnicity and differences in proportion for each measure are available in Supplemental Table 5.

For the process measures, at baseline, a lower proportion of Black infants had IMs obtained (–5.6%, 95% CI –10.0 to –1.1; Fig 1). During the intervention period, when compared with the overall cohort, a significantly lower proportion of Hispanic/Latino infants had IMs obtained (–2.3%, 95% CI –4.0 to –0.6), whereas a higher proportion of white infants had IMs obtained (2.0%, 95% CI 0.7% to 3.3%).

For adherence to the primary measures (appropriate lumbar punctures, appropriate disposition from the ED, appropriate receipt of antibiotics, appropriate length of inpatient stay), there were no differences seen by race or ethnicity at baseline or during the intervention period (Fig 2).

Within the secondary measures, at baseline, a lower proportion of Black infants had documented SDM for ED discharge for infants 22 to 28 (–2.4%, 95% CI –4.3 to –0.5). There were no other differences by race or ethnicity in the secondary measures (appropriate recommended follow-up from the ED, appropriate use of SDM for obtaining cerebrospinal fluid (CSF) for infants 22 to 28 days, and appropriate oral antibiotic use for infants 29 to 60 days with positive urinalyses). During intervention, a lower proportion of Hispanic/Latino infants had appropriate follow-up recommendations from the ED documented (–3.6%, 95% CI –6.4 to –0.8), whereas a higher proportion of white infants had follow-up documented (2.5%, 95% CI 0.3 to 4.8) (Fig 3). Black (–12.5%, 95% CI –23.1 to –1.9) and Hispanic/Latino infants (–6.9%, 95% CI –13.8 to –0.03) were less likely to have SDM documented regarding obtaining CSF.

For the balancing measures (appropriate age-based evaluations, ED revisits, readmissions), at baseline, there were no differences for infants 8 to 21 days and aged 22 to 60 days in receipt of guideline-concordant evaluation (Fig 4). At baseline and during the intervention period, a higher proportion of Hispanic/Latino infants had return visits to the ED within 7 days (Baseline: 3.3%, 95% CI 0.2 to 6.3; Intervention: 2.8%, 95% CI 1.0 to 4.6).

In this secondary analysis of a multicenter QI collaborative, there were no differences in primary measures by race and ethnicity in the baseline period, although there were process, secondary, and balancing measure differences. During the intervention period, differences emerged after AAP CPG publication and implementation of the QI bundle. Hispanic/Latino infants had a lower proportion of having IMs obtained, leading to fewer Hispanic/Latino infants qualifying for most of the primary and secondary measures, as well as lower proportions of infants having appropriate follow-up documented and higher return visits to the ED. Finally, clinicians were less likely to document engaging in SDM for obtaining CSF with the parents of Black and Hispanic/Latino infants during the intervention period. These findings demonstrate disparities in care by race and ethnicity after the publication of an evidence-based CPG intended to minimize variation.

The etiology behind the emergence of disparities after intervention remains unclear. Standardization of health care, and specifically in febrile infant management, has been proposed as a mechanism by which to reduce health disparities; however, with the introduction of SDM in various portions of this CPG, the impact of implicit bias and racism on clinician behavior has been raised as potential sources of disparity.^17,28,29  Although best practices for implementing SDM and mitigating the role of implicit bias are necessary to ensure equitable outcomes in implementation where care is not standardized,^30,31  the disparities that emerged in this study are not solely explained by the introduction of SDM as they emerged in other measures where SDM was not a factor.³²  Previous literature has identified potential outcomes from implementation initiatives on health disparities to include: (1) uniform improvement in all groups and maintenance of the health disparity, (2) disproportionate improvement for the group experiencing the disparity and reduction of the health disparity, or (3) disproportionate improvement for the group experiencing better care at baseline and a widening of the disparity.^33–35  Some of our results represent a fourth category, an intervention-generation inequity where there previously was none.

Postintervention, white infants had a higher proportion of IMs obtained and of documentation of appropriate recommendations for follow-up after ED discharge when compared with the overall cohort. This suggests that white infants experienced a disproportionate benefit from the intervention. Previous studies have found no difference based on race and ethnicity in time to antibiotics,¹⁸  receipt of guideline concordant lumbar puncture,¹⁷  or additional unnecessary interventions (lumbar puncture, empirical antibiotics, and hospitalizations).¹⁹  Our results, in contrast, represent variation in care delivery based on infant race and ethnicity after intervention and offer important insight into potentially modifiable behavior to improve care more equitably.

In the intervention period, clinicians were less likely to document SDM about obtaining CSF with the parents of Black and Hispanic/Latino infants. The potential impact of racism on SDM has been inadequately investigated^36,30,31  but is likely mediated by cultural discordance and stereotyping, exacerbating the inherent power imbalance that exists between physicians and patients.³⁵  Previous studies that have shown that Black and Hispanic/Latino patients and caregivers experience fewer opportunities for SDM, less emotional responsiveness from clinicians, and more challenges to their preferences when compared with white patients.^37–40  Although our study did not analyze the quality or result of SDM around cerebrospinal fluid testing, clinicians less frequent engagement in SDM with the families of Black and Hispanic/Latino infants highlight an important intervention-generated inequity as a result of differential provision of SDM. Future national guidelines, particularly when incorporating SDM, should incorporate antiracist principles, such as implicit bias training, cultural humility, effective communication with patients, and ongoing audit and feedback of disparities when identified.⁴¹  Additionally, poor communication with Black families can be mitigated by promoting a diverse work force and increasing the likelihood of racial concordance.^42–45 

Although we did not analyze adherence to measures by language for care, the disparity in documented discussions regarding SDM, follow-up within 1 day from the ED, and revisits for Hispanic/Latino infants may be in part because of suboptimal communication with some caregivers whose primary language is not English. Additional interventions (lumbar puncture, antibiotics, admission to the hospital),¹  longer lengths of stay,⁴⁶  and readmissions⁴⁷  have been associated with speaking languages other than English, even when no differences were seen by race and ethnicity. Additionally, an increased risk of ED revisits has previously been documented for children with Spanish-speaking parents, elevating the need to target and enhance improved communication.^48,49  This is particularly important given the introduction of SDM in the CPG, which requires bidirectional communication. The rigorous study of effectively implementing SDM with a focus on language justice is critical.^31,50  Along with the impact of language barriers, disparities in SDM, ED revisits and hospital length of stay were likely impacted by the intersectionality of the social determinants of health, including insurance status, discrimination, access to transportation, and paid time off work.^3,51  Our findings are consistent with previous literature that has demonstrated that Hispanic/Latino children face significant barriers to optimal health and experience worse health care and worse health outcomes across the board.³  Further studies focused on the barriers to timely discharge and ED revisits may be helpful for elucidating potential targets of future QI interventions.

As we move from description of disparities to implementation of evidenced-based interventions to mitigate disparities,⁵²  future national guidelines and implementation interventions should reflect an enhanced equity focus from planning through measurement.^52–54  In our intervention, because each site had relatively low numbers overall by individual race and ethnicities, we were unable to investigate disparities at the hospital-level as most sites had fewer than 100 qualifying infants per measure. To navigate this in future work, equity measures for small numbers in QI need to be developed. Finally, we must look further “upstream” to target factors contributing to health outcomes, such as policy, community-level, and environmental projects to improve disparities and advance health equity.⁵⁰ 

Our data have several limitations. With data collection by medical record review, we did not standardize how race and ethnicity were assessed and recorded at individual sites. Our findings reflect the known wide variation in race, ethnicity, and language data collection, which is problematic for identification and intervention on health disparities.⁵⁵  For example, 7 participating hospitals used visual assessment by staff to document race and ethnicity, which is not best practice.²⁷  In addition, although the total number in the study was almost 17 000 infants, we were unable to divide the data into additional time points, as in the primary REVISE II analysis, because of the relatively small numbers per race and ethnicity by each measure. Also, assessment of our secondary measures of appropriate follow-up from the ED and SDM relied on clinician documentation in the medical record. We also were unable to adjust for confounding by site-effects because of the small qualifying numbers at any given site for the specific measures. Because the demographics of infants likely vary by site, practice variation across hospitals may explain some of the inequity detected. Further exploration at the individual hospital level is needed to evaluate whether measurable disparity exists locally at participating sites. Future quality improvement initiatives should also consider collecting additional socioeconomic factors, such as language for care, insurance status, and income. In addition to race and ethnicity, the intersectionality of an individual patient and their family contributes to how a clinician assessing a child may be influenced in their decision-making.⁵⁶  Factors, such as language for care, socioeconomic status, or preferred gender or family structure, underly the implicit and explicit biases that clinicians bring when caring for children.^56,57  Within REVISE II, we identified an association with race and ethnicity and adherence to guideline recommendations but did not explore the causation or root causes of the observed disparities. Investigating the role of implicit and explicit bias is an important area of further study. Lastly, as REVISE II occurred during coronavirus disease 2019 pandemic, there were substantial variations between sites in viral testing because of variable testing protocols^58,59 ; we are unable to parse out of the data how viral testing may have influenced obtaining IMs by race and ethnicity.

Our study also had many strengths. The size of the cohort, 16 961 infants, allowed us to identify differences based on race and ethnicity in a large, geographically diverse sample. We were also able to look at both a baseline and the intervention period to identify the emergence of disparities after implementation. The diversity of the measures enabled us to elucidate potential disparities on multiple levels. Finally, our study highlights the importance of stratification of outcomes based on race and ethnicity to ensure that interventions meant to improve care do not inadvertently generate or promote disparity.

In a multicenter QI collaborative of 99 hospitals, we found differences in the provision of guideline-concordant care for febrile infants by race and ethnicity after the publication of an evidence-based CPG intended to standardize practice. Given our findings, we recommend that future guidelines implement best practices for equity-focused QI and pursue continued rigorous analysis of implementation results by race and ethnicity. Future investigation is needed to develop implementation strategies to ensure equitable delivery of evidence-based care for febrile infants.

Please see Appendix for full list of AAP REVISE II QI Collaborative authors.

AAP

American Academy of Pediatrics

CI

confidence interval

CPG

Clinical Practice Guideline

CSF

cerebral spinal fluid

ED

emergency department

IM

inflammatory markers

REVISE II

Reducing Excessive Variability in Infant Sepsis Evaluation II

SDM

shared decision-making

UA

urinalysis