Development of an EHR-based Alert to Improve Sepsis Recognition in Hospitalized Children with Cancer

Introduction:

Sepsis carries a high risk of morbidity and mortality in hospitalized patients with malignancy or immunocompromise, yet can be challenging to recognize due to high prevalence of abnormal vital signs and laboratory values in this population. Additionally, systemic and individual factors may contribute to delays in identification of sepsis in these patients (Fig. 1). The objective of this initiative was to develop an Electronic Health Record (EHR)-embedded sepsis screen and automated alert specific to pediatric cancer patients admitted to the cancer/transplant unit to facilitate early sepsis recognition.

Fig. 1.

Fishbone diagram of multiple factors leading to delay in recognition of sepsis in pediatric cancer and transplant patients.

Methods:

Our quaternary, freestanding children’s hospital initiated continuous, automated EHR-based sepsis screening for children admitted to general inpatient units in July 2019 based on a combination of vital signs, nursing documentation, and laboratory values (Table 1). When sepsis screening criteria are met, an EHR-based interruptive alert fires to the patient’s nurse, and the clinical team is subsequently notified. In addition, we maintain a database of confirmed sepsis cases with inclusion based on modified criteria from the Improving Pediatric Sepsis Outcomes collaborative, followed by chart review for verification. To evaluate the feasibility of implementing the EHR-based alert for children admitted to the cancer/transplant unit, the screening algorithm ran for 14 weeks in the background of the EHR. Health care providers were not notified in real time of patients meeting criteria for an alert, but all alert data were captured. We compared positive sepsis screenings to our hospital’s database of confirmed sepsis cases to evaluate the performance of the algorithm and reviewed the charts of patients triggering a silent alert to determine appropriateness and timeliness of the alerts.

Table 1.

Patient Criteria for Positive Sepsis Screen and EHR Alert

Criteria	Alert Logic
Temperature <35 °C or >38 °C	Abnormal temperature AND hypotension OR Abnormal temperature AND tachycardia AND at least one other criterion
Systolic hypotension for age
Tachycardia for age
Capillary refill (central or peripheral) >3 s or <1 s
	Alert fires no more than once every 12 h per patient
Skin color mottled
Skin temperature cool or cold
Mental status lethargic or inconsolable
WBC count high or low for age
Lactic acid >2.0

Results:

During the 14-week study period, the screening algorithm identified 87 potential sepsis episodes in 1,855 patient days. There were 6 confirmed cases of sepsis on the cancer/transplant unit during this time, and 5 of 6 were identified by the algorithm. Importantly, all 5 of these positive screens would have initiated an alert via the EHR before the time at which the sepsis diagnosis was made by the clinical team. Using confirmed sepsis cases in the database as the gold standard, the screening algorithm yielded a sensitivity of 83.3% and specificity of 95.7%. However, the positive predictive value was low (5.7%), with number needed to screen to identify 1 true case of sepsis of 18. Given the relatively low positive predictive value and high number needed to screen, a nurse-driven patient assessment was added as a second step to the screening process, allowing for real-time clinical data to augment the precision of the alert. The nursing assessment includes an evaluation of the patient’s perfusion, mental status, and urine output.

Conclusions:

An automated screening algorithm with an EHR-based alert can appropriately identify immunocompromised children with sepsis, though with a relatively high false positive rate, placing frontline teams at risk for alert fatigue. A multimodal screening process, combining an EHR alert with secondary nursing assessment, is a feasible approach to apprise clinicians of immunocompromised patients at risk of sepsis before their decompensation and to promote earlier intervention.