Racial Differences in Sepsis Recognition in the Emergency Department

This was a retrospective cohort study in a tertiary-care, urban, academic children’s hospital ED with an existing electronic health record–based sepsis alert from June 1, 2016, to May 31, 2017.

Subjects were included if they visited the ED and self-identified or were identified by the parent at the ED registration as either non-Hispanic white (NHW) or non-Hispanic black (NHB), the 2 most common racial groups during the study period. A more detailed subanalysis was conducted on ED patients with either a positive first sepsis alert result or those who were treated for sepsis by using the sepsis pathway and/or order set.

Since 2014, the ED has had an electronic sepsis alert system in place designed to reduce the number of missed sepsis patients (Fig 1). The sepsis alert consists of 2 stages followed by a huddle process and has been described in detail elsewhere.⁵  Briefly, the first stage of the alert has a positive result if the patient has either tachycardia or hypotension for their age. The patient must have a positive first alert result to prompt the second stage of the alert: a nursing assessment performed by the triage or bedside nurse. The second stage of the alert has a positive result if the patient has an underlying high-risk condition, an abnormal mental status, or delayed capillary refill. A high-risk condition was defined as <56 days old, asplenia, sickle cell disease, bone marrow or solid organ transplant, central line, malignancy, immunodeficiency, immunocompromise, functional technology dependence (such as need for mechanical ventilation), or significant developmental delay that affects the patient’s ability to walk, talk, or eat.

A positive second alert result prompts a sepsis huddle, in which the ED team assesses the patient at the bedside and determines the need for sepsis treatment using bundled care in the sepsis pathway. Sepsis pathway activation is defined as using an associated order set in the electronic health record. The sepsis pathway can also be activated by clinicians, on the basis of their bedside judgment, for patients who did not trigger the sepsis alert. In this study, we aimed to evaluate each phase of the sepsis alert for racial differences.

We used an existing QI data set extracted from the electronic health record. Data elements included race, ethnicity, sex, age, payer, first alert result, second alert result, reason for positive second alert result, sepsis pathway activation, and ICU admission within 24 hours of the ED visit.

We compared the frequency of positive first alert results, positive second alert results, second alert components, and sepsis pathway activation between NHB and NHW patients. Frequencies, percentages, and odds ratios (ORs) were reported for descriptive elements. Unadjusted comparisons were made by using χ² testing. Multivariable logistic regression was performed by using predictors of sepsis pathway activation, including race (primary predictor), sepsis alert components (comorbidities, decreased capillary refill, and altered mental status), sex, age, and payer.

We evaluated independent predictors of sepsis pathway activation, including race (primary predictor), sepsis alert components, sex, age, and payer, using multivariable logistic regression. We tracked ICU admission frequency as an outcome measure. All analyses were performed with NHB as the reference group. Analyses were conducted by using Stata 15.0 (Stata Corp, College Station, TX).

Human Subjects: This study was determined to be exempt from review by the Children’s Hospital of Philadelphia Institutional Review Board.

During the study period, there were 97 338 ED visits, of which 56 883 (58.4%) were by NHB patients and 23 008 (23.6%) were by NHW patients (Table 2). There were 12 650 (13.0%) positive first alert results, defined as tachycardia or hypotension for age (Table 3). There were 1298 (1.3%) positive second alert results, defined as an underlying high- risk condition, a change in mental status, or delayed capillary refill. A total of 242 patients had a sepsis pathway activation when using the alert system (0.25% of all visits, 1.9% of patients with a positive first alert result, and 18.6% of patients with a positive second alert result). An additional 84 patients (0.086%) had a sepsis pathway activation outside of the alert system based on clinical judgment.

Subsequent analysis was limited to patients who had a positive first alert result (Table 4). Demographic characteristics of this subset are listed in Table 2. We next compared the frequency of each stage of the sepsis alert between NHB and NHW patients, using both unadjusted and adjusted analyses (Table 4). NHW patients were slightly more likely than NHB patients to have a positive first alert result (OR 1.1; 95% confidence interval [CI] 1.1–1.2), a difference that disappeared after adjusting for payer, age, and sex (adjusted odds ratio [aOR] 0.66; 95% CI 0.024–18.7). NHW patients were more likely than NHB patients to have a positive second alert result (OR 3.0; 95% CI 2.5–3.3), a difference that was attenuated but persisted after adjusting for payer, age, and sex (aOR 2.4; 95% CI 2.1–2.8). The odds of sepsis pathway activation were increased in NHW patients compared with NHB patients (aOR 1.4; 95% CI 1.02–2.1). Stepwise multivariable regression was also performed with high-risk condition as the biggest contributor to the racial difference in sepsis pathway activation, followed by age, delayed capillary refill, abnormal mental status, and payer (Table 5).

We then examined racial differences in sepsis pathway activation in the subset of patients treated through the alert and/or huddle process and in the subset of patients treated outside of that process on the basis of clinician judgment, as listed in Table 3. There was no difference in sepsis pathway activation between NHW and NHB patients treated through the alert and/or huddle process (OR 1.2 [95% CI 0.9–1.7] and aOR 0.93 [95% CI 0.62–1.4], respectively). However, for patients treated outside of the alert process by using clinical judgment alone, NHB patients were more likely to be have sepsis pathway activation (OR 3.6 [95% CI 2.2–6.1] versus aOR 3.4 [95% CI 1.8–6.4]).

We went on to analyze the 3 components of the second alert: high-risk condition, abnormal mental status, and delayed capillary refill. There was no difference between NHW and NHB children in the presence of a high-risk condition (OR 1.2; 95% CI 0.90–1.6) or abnormal mental status (OR 0.86; 95% CI 0.66–1.1). Comorbidities seen in NHW and NHB children are shown in Table 7. Of note, only 1 high-risk condition can be selected through the existing electronic sepsis alert. Consequently, it is possible for a patient to have multiple comorbidities, but the data set would only reflect the single condition selected by the triage nurse. NHW patients were more likely than NHB patients to have delayed capillary refill time recorded by the provider (OR 1.8; 95% CI 1.2–2.6). There was no racial difference in ICU admission frequency among the patients treated for sepsis (aOR 1.3; 95% CI 0.68–2.5).

The median time from ED triage to initial antibiotic treatment for all patients treated on the sepsis pathway was 78 minutes (interquartile range [IQR] 44–172) for NHB patients and 64 minutes (IQR 50–119) for NHW patients, although this difference was not statistically significant (Table 6). For patients treated on the sepsis pathway through the alert system, the median time to antibiotics was 77 minutes (IQR 47–160) for NHB patients and 61 minutes (IQR 49–102) for NHW patients. For patients treated on the sepsis pathway outside of the alert system, the median time to antibiotics was 106 minutes (IQR 37–246) for NHB patients and 74 minutes (IQR 53–200) for NHW patients. Neither subset of patients had racial differences in antibiotic administration times that were statistically significant.

We demonstrated that in a large, academic, pediatric ED with an established electronic sepsis alert, NHW patients were slightly more likely to be treated with the institutional sepsis pathway than NHB patients. Although there were no differences in sepsis pathway activation for patients identified through the alert, NHW patients were more likely than NHB patients to be identified and treated for sepsis outside the alert. These differences persisted after adjusting for high-risk condition, sex, age, and insurance status. These findings suggest that sepsis treatment when using an electronic alert, compared with clinician judgment alone, may reduce racial differences in care.

A previous study found that black adults were at an increased risk of sepsis and associated the increased risk with having chronic comorbidities that compromise immune function.¹⁴  Our research supports the finding that chronic comorbidities are associated with increased risk of sepsis, with high-risk condition being the highest contributor to the unadjusted racial difference in sepsis treatment. Another study reported a higher incidence of and mortality from sepsis in adult black patients compared with Hispanic patients and white patients after adjusting for socioeconomic status.¹⁵ 

The implementation of an electronic sepsis alert introduces objective elements, aiming to reduce variability across practitioners by implementing a standardized process.¹³  Sepsis detection also has subjective components because the decision to initiate sepsis treatment ultimately depends on clinician judgment and signs of infection. Previous studies have identified racial disparities in pediatric EDs, especially in the context of medical decision-making that involves subjectivity and clinical uncertainty. One study identified disparities in ordering laboratory studies and imaging for pediatric ED patients with subjective conditions that lacked a standardized treatment protocol, although no disparities were identified when an established evaluation protocol was in place.¹⁶  Pediatric EDs have shown that although there were no racial differences in cranial computed tomography order use for children at low or high risk of brain injury, racial differences were noted for those with intermediate risk of brain injury, with NHB and/or Hispanic patients being less likely than NHW patients to have computed tomography ordered.¹⁷  A study on the management of abdominal pain, a subjective complaint, showed that NHB children were less likely to receive pain medicine in the ED compared with NHW children.¹⁸ 

There are multiple proposed explanations for the observed increased frequency in sepsis pathway activation in NHW patients treated outside the alert compared with NHB patients, although there was no racial difference in sepsis pathway activation in patients treated through the alert. One possible explanation is that the racial differences were a result of provider implicit bias. Unlike patients with the alert, which includes a screening and huddle process, patients treated outside the alert are subject to clinician judgment alone, wherein they may be more susceptible to implicit provider bias. A growing body of research shows that physicians in multiple settings, including the pediatric ED, have implicit bias favoring white patients over black patients.^19–21  The literature on the impact of bias on medical decision-making is mixed and is an important area for future research.¹⁹ 

The following are additional possible explanations for our findings. It is possible that it is more difficult to detect increased capillary refill time or cyanosis in darker-skinned patients, as has been shown previously for identifying limb ischemia and recognizing pressure ulcers.²²  In fact, we found a 1.8-fold increase in the odds of delayed capillary refill reported in NHW patients compared with NHB patients. It is important to note, however, that capillary refill time in the context of this alert is operationalized as a given provider’s opinion of the patient’s perfusion status, and less as an objective test in and of itself because capillary refill time is known to change with ambient temperature,²³  and also has poor interrater reliability.²⁴  A second explanation may be access to care. Geographically, the children’s hospital ED in this study is near a low–socioeconomic-status community (demographically, 71.2% NHB and 17.4% NHW).²⁵  Patients in low–socioeconomic-status areas may frequent the ED instead of visiting the pediatrician’s office in nonurgent situations, or conversely, patients with poor access to care may wait until their illness is more severe before seeking care.^26–28  Only the former could result in the observed racial difference. A third explanation may be increased requests among white parents compared with those among black parents, resulting in a greater frequency of sepsis workup.¹⁷  Using the electronic sepsis alert may mediate parental request by incorporating more objective components into the medical decision-making process.

This study has several limitations. First, the study was performed at a single academic center that has had sepsis QI efforts in place since 2012. Consequently, the findings may not be generalizable to other institutions, especially institutions without a standardized sepsis protocol.¹⁶  Second, additional confounders could be considered that are not available in our current data set. For example, it is possible that white children had a greater percentage of high-acuity visits than black children. Acuity was measured by the presence of high-risk conditions, decreased capillary refill, and abnormal mental status (positive second alert result). The increased frequency of positive second alert results in NHWs compared with NHBs (OR 3.0; 95% CI 2.5–3.3) suggests that white children had higher-acuity visits than did black children. However, unlike positive second alert results, which are specific for sepsis risk factors, triage acuity is a measurement of acuity that factors in the chief complaint and the patient’s stability on ED presentation. Triage acuity would be interesting to consider because it can influence clinician judgment when assessing the patient. Even if there was higher acuity in white children compared with in black children, this does not explain the absence of racial differences in the patients treated through the alert. With the subset of patients treated outside the alert, both NHWs and NHBs would likely have lower acuity because they did not trigger the sepsis alert with normal vital signs and without high-risk factors for infection. Because there may be additional unmeasured confounders, it is difficult to determine if the identified racial differences were clinically appropriate or representative of a racial disparity due to provider implicit bias, parent preferences and/or expectations, or institutional practices. An alternative explanation may be that the observed racial differences were clinically appropriate. Third, it is important to underscore that the study outcome was sepsis pathway activation (meaning, sepsis treatment), not the presence of sepsis itself. This data set did not contain organ dysfunction–based outcomes, and thus, we cannot comment on differences in sepsis prevalence between races. This challenge is magnified by the following issue: it is possible that in our cohort of treated patients, there are false-positive results (ie, patients who were treated for sepsis who were not septic). One challenge of ED-based sepsis studies is the difficulty in determining who these patients are because if they improve with treatment and never develop the organ dysfunction that defines severe sepsis, it is unclear if this is because of or despite treatment. Finally, we did not evaluate sepsis alert performance in racial groups other than black and white, nor did we evaluate patients of Hispanic ethnicity. In addition, we were not able to evaluate the impact of racial concordance of provider and patient on sepsis alert performance. These topics will be addressed in future analyses.

Future studies are needed to explore sources of the observed racial differences, which may exist at the level of the patient, parent, provider, and health care system. The data set could be expanded to collect information on triage acuity, patient socioeconomic status, and access to a primary care provider. Future research is also needed to evaluate if these racial differences in sepsis detection are associated with differences in patient outcomes, such as timeliness of antibiotics and fluids, organ dysfunction, and mortality.

In summary, in a single-center academic ED with an electronic sepsis alert in place, we demonstrated that NHW patients treated outside of the alert were more likely than NHB patients to be identified and treated for sepsis. There was no difference in sepsis recognition in the subset of patients treated when using the alert. These findings may suggest that using the sepsis alert reduces racial differences in sepsis detection and treatment when compared with using clinician judgment alone, a hypothesis that will be examined in future prospective studies.

aOR

adjusted odds ratio

CI

confidence interval

ED

emergency department

IQR

interquartile range

NHB

non-Hispanic black

NHW

non-Hispanic white

OR

odds ratio

QI

quality improvement