Equity in Receipt of a Lumbar Puncture for Febrile Infants at an Academic Center Free

This is a retrospective cross-sectional study looking at disparities in the care of febrile infants ≤60 days of age. We performed a secondary analysis of a database collected through retrospective chart review. A team including 2 authors (M.T., N.S.) created the database for an epidemiologic review of the febrile infant population at our institution. We collected study data using REDCap (Research Electronic Data Capture, version 10.9.4).¹⁰  The institutional review board of our institution approved this study under expedited review (STUDY18100106).

We collected data on all infants ≤60 days of age who presented to our large academic children’s hospital through our emergency department with documented fever (≥100.4°F or ≥38°C) by any route using a thermometer at home, on presentation, or during admission between January 1, 2015, and December 31, 2017. We excluded infants if they did not have a fever within 24 hours of presentation, had never been discharged from the hospital after birth, or if laboratories and cultures were obtained at an outside institution. For the primary analysis, infants were excluded if they did not have either commercial or public insurance or if their race was not self-identified as Black or white, because of small sample size. To ensure that these exclusions did not affect our results, we also performed a sensitivity analysis with all infants in this study dichotomized to Black, Indigenous, and people of color, and white. We also excluded infants described as ill-appearing because most current guidelines focus on well-appearing infants. We defined ill appearance as documentation in the provider physical examination of ill-appearing, lethargic, toxic, inconsolable, or cyanotic.^11–13  We collected laboratory data, imaging results, initial vitals, historical information, and physical exam findings through manual chart review.

The primary outcome was receiving of a guideline-concordant lumbar puncture (GCLP), defined as obtaining the recommended LP or appropriately not obtaining an LP as defined by the hospital’s clinical practice guideline. The guideline recommended attempting an LP for all febrile infants ≤60 days of age, regardless of clinical appearance, unless an infant was both >28 days of age and had respiratory syncytial virus (RSV)-positive bronchiolitis. Because we only collected laboratory data and not diagnosis codes, we used RSV positivity as a surrogate marker to identify infants that did not require an LP. We used insurance type, public versus commercial, and area-based median household income as proxies of SES. We determined area-based median household income (MHI) using ZIP code and corresponding census data.¹⁴  MHI was then divided into 3 income brackets: lower income, less than $42 000 annually; lower-middle income between $42 000 and $85 000 annually; and upper-middle and higher income, greater than $85 000 annually, based on average middle class income for a 3-person household in 2015.¹⁵  We used cut points instead of terciles because previous studies have shown that cut points better correlate with individual SES than terciles.¹⁶  The median household income for the county surrounding the hospital in 2015 was estimated to be $54 412, though the catchment area of the hospital extends far beyond the surrounding county. We also collected gestational age on infants for whom it was documented. We separated infants into premature (<35 weeks’ gestation), late preterm (35 weeks’ to <37 weeks’ gestation), and full term (37 weeks’ or greater gestational age at delivery).

Differences between patients receiving and not receiving a GCLP were assessed using Fisher exact test for categorical data and Wilcoxon rank sum test for continuous variables. The following covariates were assessed from the local Febrile Infant Database: patient characteristics (age, gestational age, sex), exposures (maternal group B Streptococcus status, daycare), presentation (temperature on arrival to emergency department), and laboratory values (white blood cell count, absolute neutrophil count [ANC], and urine white blood cell count). We included laboratory values that were recommended in the clinical practice guideline, which were a complete blood cell count and a urinalysis. Because C-reactive protein and procalcitonin were not recommended in the clinical practice guideline during the study period, these laboratory values were not obtained on most infants and thus were not included in the univariate analysis. We included age in days, temperature in degrees Celsius, and laboratory values as continuous variables; all others were categorical variables. Variables with a P < .1 on the univariate analyses, in addition to race, MHI, and insurance type, were included in a final multivariate logistic regression model with receipt of a GCLP as the binary outcome of interest. Statistical significance was designated at P <  .05. SPSS software, version 28.0, was used for all statistical analysis (IBM Corp, Armonk, NY).

Between January 2015 and December 2017, a total of 1800 febrile infants presented to our tertiary children’s hospital through the emergency department (Fig 1). We included 965 infants in our analysis. The median age of the infants at the time of presentation was 40 days (interquartile range = 27, 50). Most of the infants’ families self-identified as white (n = 791, 82.0%) and had an area-based MHI between $42 000 and $85 000 per year (n = 629, 65%). More infants had commercial insurance (n = 530, 54.9%) than public insurance (Table 1).

A total of 804 infants (83.3%) of the 965 included infants received a GCLP. There were 161 infants who either received a guideline-discordant LP (n = 19, 11.8%), or did not receive an LP when it was indicated by the clinical practice guidelines (n = 142, 88.2%). The 19 infants who had a guideline-discordant LP were all older than 28 days of age and positive for RSV. There was no difference in gestational age, sex, MHI, race, group B Streptococcus status, daycare exposure, and blood cell count between infants who did and did not receive a GCLP (Table 2). Age at presentation, temperature on arrival, ANC, and urine white blood cell count had P values less than .1 and thus were included in the multivariate analysis.

As the age of the infant increased (adjusted odds ratio, 0.95; 95% confidence interval [CI], 0.94–0.97), the infant had lower odds of receiving a GCLP, whereas infants with higher temperature on arrival to the emergency department had increased odds of receiving a GCLP (adjusted odds ratio, 1.36; 95% CI, 1.04–1.78) (Fig 2). Race, insurance type, and MHI were not significantly associated with receiving a GCLP (Fig 2).

The sensitivity analysis, including all infants who had a self-identified race listed were dichotomized to Black, Indigenous, and people of color versus white, continued to show no statistical association with GCLP for race or either socioeconomic factor.

We found that self-identified race and SES were not significantly associated with physicians’ decisions to follow guideline recommendations regarding completing LPs in the evaluation of febrile infants. The factors associated with receiving a GCLP included younger age and a higher temperature on arrival to the emergency department, which have been shown to be characteristics of infants at higher risk for sepsis.^8,17–21  Previous studies have similarly shown variations in care based on the age of febrile infants.²² 

This study is the first of its kind to look specifically at the evaluation of febrile infants. However, our results are contrary to previous studies that have shown racial disparities in laboratory ordering practices during the evaluation of pediatric patients.⁵  In 2013, Payne and Puumala found a statistically significant decreased odds of obtaining both laboratory and radiologic testing for children that identified as African-American compared with non-Hispanic white children.⁵  This difference in radiologic testing was again present in a more recent study from 2016 through 2019.⁴  Neither study could determine if their findings represent health care overuse for non-Hispanic white children or underuse for non-Hispanic Black children. Regardless, both studies demonstrate nonequitable care. There have been active efforts to address these disparities across the health care system, including the national level, however, even in 2019, Black patients overall continue to experience health care disparities, which are present in approximately 40% of the quality measures identified by the Agency for Healthcare Research and Quality.²³ 

One possible explanation for the lack of identified racial or socioeconomic disparity in our study is protocolization. Payne and Puumala in 2013 were able to identify that the evaluation of head injury, which had an established protocol, did not have racial disparities in the care provided, unlike other presenting complaints.⁵  Similarly, Raman et al showed a decrease in the variability of sepsis identification after introducing a standardized electronic sepsis alert.²⁴  These studies identify a possible explanation for our results, which is that the establishment of a protocol can be beneficial in combating racial and socioeconomic disparities. It is likely not only the protocol itself that helps ensure universal evidence-based care, but if a physician decides to deviate from standard guidelines, as many physicians did in our study, they had to carefully evaluate and document their reason for deviation, forcing confrontation of potential implicit biases.

Other possible reasons for the lack of identified disparities include limited diversity in our population as well as a limited number of infants who did not receive a GCLP, both of which might hinder our ability to detect disparities. Our hospital system commonly uses clinical practice guidelines to assist with management decisions and thus our findings might be explained by our culture of guideline adherence.

At the time of the study period, from 2015 to 2017, many protocols across the country recommended a full evaluation including blood, urine, and cerebral spinal fluid studies, for febrile infants ≤60 days of age.^25,26  Over the past 5 years, there have been changes in the evaluation of the febrile infant population.^27,28  Recently, in July 2021, the American Academy of Pediatrics published updated national recommendations for the evaluation of febrile neonates.²⁸  The new guidelines recommend performing LPs in select groups of infants based on the findings of studies, such as PECARN, which was a large multicenter prospective observational study that identified procalcitonin and ANC as a less invasive, highly sensitive way to better identify an infant’s risk of serious bacterial infection.²⁷  Decreasing the need to subject febrile infants to the invasive procedure of an LP is important because it eliminates any pain caused by the procedure and the small infection risk, as well as decreasing the length of stay from unsuccessful LP attempts.^29,30  The analysis of the implementation of previous guidelines, and whether they are associated with disparities, can inform future protocol implementation. Thus, the results of this work will enable hospitals to prioritize equity during implementation efforts of their new practice guidelines. We plan to assess the impact of the new American Academy of Pediatrics recommendations to evaluate if racial or socioeconomic disparities develop.

There are multiple limitations to this study. The first limitation is the lack of ethnicity data. Unfortunately, the institutional Febrile Infant Database did not collect ethnicity data, so we were unable to perform a subanalysis based on infants identifying as Hispanic. Despite this limitation, the population that our hospital serves has a very small Hispanic population, approximately 3%, potentially mitigating the impact of this significant limitation.¹⁴  The second limitation is the limited availability of information about individual SES. Given the retrospective nature of the study, we relied on information collected by the Febrile Infant Database. However, the use of ZIP code and corresponding census data has been shown to be a strong measure of area-based SES, which is associated with area-based risk, privilege, and supports, as well as self-reported education level and individual SES.¹⁶  The third limitation is that we were unable to differentiate children who were febrile on presentation to the ED and those who became febrile on the floor within 24 hours of admission. Despite this, the same clinical practice guideline applies to all febrile infants <60 days of age, regardless of location, and thus we do not think this detracts from the validity of our conclusions. In addition, we were unable to confirm whether all children who were RSV-positive had clinical bronchiolitis but used RSV positivity as a surrogate marker. An additional limitation is that this study was completed at a single center. We are a large academic center with many protocols in place, which may limit the applicability of the results to smaller community hospitals. Another important limitation that leaves room for future study is that we were limited in our ability to differentiate clinical outcomes. The database we used only collected the laboratory and imaging evaluation completed in the emergency department with results and not the decision to admit a patient or administer antibiotics.

Given that post hoc power analyses have limited utility,³¹  and the study used data collected retrospectively, we reviewed previously published studies to address sample size needs relating to type II error. In previous studies looking at disparities in care, most have a smaller total sample size than our study, although they all have more diverse populations than our site serves.^32–37  For post hoc analyses, it is recommended to instead focus on effect sizes and their 95% CIs. A 95% CI demonstrates a rational range of effect sizes that are consistent with the observed data and that the true effect size is likely to lie within these values and provides readers with information to assess whether a false negative finding may have occurred.³⁸  Although all researchers must be aware of the possibility of type II error, it is critical in disparities research. We have taken measures to ensure accurate representation of our results including reporting our results using odds ratios, rather than relying on an arbitrary P value.³⁹  We feel that the high rate of guideline concordance, despite the obvious challenges it causes with limiting type II error, is a sign that high compliance with evidence-based guidelines could be a step in improving health equity overall.

Studying health disparities is critical but challenging. There are many factors that contribute to this challenge, many of which are enumerated previously. During this research process, we began to appreciate the complex nature of health disparities. Given the importance of identifying disparities to appropriately address inequities, not dismissing small variations in care, data collection, and careful statistical analysis is imperative.³⁹  Performing a priori power calculation would ensure that the study is adequately powered to provide accurate representation of results while limiting type I and II error. We know that high cognitive load and time pressure lead to increased disparities,⁴⁰  which the field of pediatrics, especially pediatric hospital medicine, has been experiencing nationwide for the past 2 years, so there are many areas to investigate for disparities, but focusing on areas of impactful decisions, such as invasive testing, duration of admission, pain management, medication delivery, and follow-up rates are areas for future research. We also plan to expand to evaluate difference in language preference as another population at risk for being exposed to inequitable care.

Overall receipt of a GCLP was not associated with race or SES. We believe that the lack of observed disparities could be due to protocolization, as has been seen with previous studies, though limited diversity and a culture of high guideline compliance may also contribute. It will be important to continue to assess the impact of race and SES in the evaluation and management of febrile infants as the new national consensus guidelines are implemented.

We thank Drs Maria Widmann, Ag Nuwan Perera, Jacob White, Molly Marshall, Melissa Lutz, and Allison Williams for the collection of data and development of the CHP Febrile Infant Database used for this analysis.