Use of Procalcitonin in a Febrile Infant Clinical Pathway and Impact on Infants Aged 29 to 60 Days

This study was determined to be exempt by our state’s multiple institutional review board.

This study was conducted across a network of a single academic quaternary pediatric ED and 5 satellite pediatric EDs and urgent care clinics (further referred to as EDs), all associated with a single university-affiliated freestanding children’s hospital. All sites share a common EHR, clinical pathways, and emergency medicine faculty. The annual number of ED patient visits across all sites is 160 000.

Fever is a symptom; therefore, International Classification of Diseases, 10th Revision codes for fever, as well as common fever-related codes (Supplemental Information), were used to identify subjects from the EHR. We retrospectively collected study variables on febrile infants who presented to an ED from January 2017 to December 2019. Infants aged 29 to 60 days were eligible if they met pathway inclusion criteria, including having a measured temperature ≥38.0°C or <36.0°C at home, at a primary care clinic, or in the ED. Infants were excluded if they had a corrected gestational age ≤37 weeks, weighed ≤2000 g, had a chronic illness, were immunocompromised, or needed immediate critical care. Because the focus for this study was pathway impact on the initial ED management, we excluded infants with prenatally identified high-risk conditions and infants directly admitted to an inpatient unit from a referring facility. If an abnormal temperature was not measured in the ED, a chart review was completed to identify infants with a temperature meeting the criteria at home or in the primary care office.

High-risk infants were defined by any of the following: procalcitonin level >0.3 ng/mL, WBC <5000 or >15 000 cells per μL, or absolute band count ≥1500 band cells per μL. Low-risk infants needed at least the procalcitonin level or CBC measured, and all results needed to be normal (procalcitonin level ≤0.3 ng/mL, WBC 5000–15 000 cells per μL, and absolute band count <1500 band cells per μL). Infants were considered to have an incomplete laboratory evaluation if both the procalcitonin value and CBC were missing.

To assess for clinically relevant differences in patient characteristics, the following covariates were selected. Age was divided into clinically relevant subgroups of 29 to 44 and 45 to 60 days. Fever at presentation was also used because infants may appear more ill when febrile. Patient location was selected because of the differences in procalcitonin testing availability between locations. Finally, the season of presentation, defined as bronchiolitis season (December to April) or nonbronchiolitis season (May to November), was used as an additional covariate because of pathway recommendations specific to patients with bronchiolitis. Season was selected because the data did not capture all bronchiolitis diagnoses.

Outcome measures were defined as the number of patients with the measure of interest (LP, antibiotics, and admission) over the total number of patients seen in the ED over the same time period. ED LOS was evaluated in minutes from ED admission time to ED discharge time. Missed SBI diagnosis was defined as the number of patients who returned to an ED or inpatient location within 7 days of initial evaluation with a diagnosis of UTI, bacteremia, or bacterial meningitis over the total number of patients seen in the same time period.

The cohort was divided into 3 groups: pre–pathway implementation (PI) (January 2017 to January 2018), implementation phase (February 2018 to March 2018), and post-PI (April 2018 to December 2019).

A bivariate analysis, excluding the implementation phase, was done on patient characteristics, including sex, race, ethnicity, age group, financial class, presence of fever, patient location, season, patient risk category, SBI diagnosis rate, and missed SBI diagnosis rate, by using the Pearson χ² test or Fisher’s exact test for categorical variables. The proportion of patients with a procalcitonin measurement was calculated by month and displayed on a line graph. In addition, a bivariate analysis of the proportion of infants undergoing LP by risk category was completed.

Outcome and balancing measures were first evaluated by using a pre-post analysis. To reduce the potential of biases in the pre-post analysis,¹⁹  the differences in LP, antibiotics, admission, and ED LOS over the study period were evaluated by using an interrupted time series (ITS) analysis via generalized linear models and segmented regression with a Poisson (LP, antibiotics, and admission) or negative binomial (ED LOS) distribution. Multivariable ITS models were controlled for significant differences between groups, specifically season. Predicted values for each outcome measure were calculated from the unadjusted segmented regression models and displayed with the observed monthly rates for LP, antibiotics, and admissions and the observed monthly median ED LOS in minutes to illustrate the linear trends over time.

A multivariable logistic regression was used to evaluate the odds of receiving an LP with a normal versus abnormal procalcitonin level. The model was adjusted for age group, fever at presentation, patient location, and bronchiolitis season. Variables were chosen a priori.

Statistical significance was set at P < .05. The data were analyzed by using SAS (version 9.4; SAS Institute, Inc, Cary, NC).

A total of 740 infants aged 29 to 60 days were included in the study. There were 283 infants pre-PI, 58 infants during implementation, and 399 infants post-PI. There were no significant differences in patient age, sex, race, ethnicity, or insurance pre- and post-PI. There was no difference in the SBI diagnosis rate between groups. There were significantly fewer patients in the bronchiolitis season post-PI (54.1% vs 38.6%; P < .001), likely a result of implementation occurring during bronchiolitis season. The patient risk category was also statistically different, with more patients being high risk (16.9% pre-PI to 26.8% post-PI) and fewer patients having incomplete laboratory test results in the post-PI group (32.9% pre-PI to 20.6% post-PI) (P < .001) (Table 1).

Before PI, 3.2% of febrile infants aged 29 to 60 days had a procalcitonin test result. Procalcitonin testing was rapidly adopted by providers, with 67.4% of patients having a procalcitonin test post-PI (Fig 1).

There was no change in the proportion of high-risk infants undergoing LP before or after PI (60.4% pre-PI versus 53.3% post-PI; P = .41); however, there was a significant decrease in the number of low-risk infants undergoing LP post-PI (31.7% pre-PI to 16.7% post-PI; P = .001) (Table 2).

The unadjusted ITS analyses are displayed in Fig 2. With and without controlling for bronchiolitis season in the ITS model, we found no significant immediate level change (shift) between the pre-PI and post-PI periods for all infants for the outcomes of LP (0.98 [95% confidence interval [CI]: 0.49–1.97]), antibiotics (1.17 [95% CI: 0.56–2.43]), admission (1.07 [95% CI: 0.59–1.96]) or ED LOS (1.08 [95% CI: 0.92–1.28]). In addition, despite increased use of procalcitonin testing overtime, there was no slope change post-PI compared with pre-PI for any outcome measure. Although the risk of admission and ED LOS appear to increase in the post-PI period, as shown in Fig 2 C and D, the effect estimate in Table 3 reveals that this increase was not significant, nor was there a significant slope change for LP or antibiotics (Table 3).

There was no difference in missed SBI diagnoses (0.7% vs 0%; P = .17). Pre-PI, there were 2 patients with missed SBI diagnoses. The first was high risk on the basis of the WBC but was discharged from the ED. The patient returned within 12 hours, was treated for group B Streptococcus bacteremia and presumptive meningitis, and recovered. The second infant was low risk, was called back for a positive urine culture result, was treated for UTI, and did not require admission. There were 0 cases of missed SBI diagnosis post-PI. There was 1 death pre-PI unrelated to the febrile illness presentation.

The adjusted logistic regression analysis revealed that an infant with a normal procalcitonin level (odds ratio [OR] = 0.11 [95% CI: 0.05–0.23]; P < .001) and infants in bronchiolitis season (OR = 0.56 [95% CI: 0.31–0.98]; P = .04) were less likely to undergo LP. Age group (OR = 1.59 [95% CI: 0.91–2.77]; P = .10), presence of fever (OR = 0.87 [95% CI: 0.48–1.57]; P = .63), and patient location (OR = 1.05 [95% CI: 0.60–1.85]; P = .86) did not significantly change the odds of LP.

There is a growing body of literature supporting the use of procalcitonin testing in the evaluation of febrile infants and suggestion that providers may be able to safely avoid unnecessary interventions in older low-risk febrile infants. Institutional clinical pathways are effective tools for translating evidence into practice and decreasing variation.^20,21  The goal of our pathway was to implement additional testing to risk stratify infants, ensuring the right infants receive interventions. We hypothesized that performing interventions on the right patients would translate to overall de-implementation of interventions in older febrile infants. We further hypothesized that de-implementation in this population would not be associated with a change in the rate of missed SBI diagnoses. Our PI accomplished some, but not all, of the intended goals. Our data align with current literature revealing that procalcitonin testing helps identify low-risk infants and safely allows for fewer interventions. As intended, we implemented procalcitonin testing and de-implemented LPs in low-risk infants. We estimate this eliminated 31 unnecessary LPs in low-risk infants post-PI. However, we did not see a decrease in resource use in the overall population post-PI, despite stable SBI and missed SBI diagnosis rates pre- and post-PI. There are several potential reasons for this finding.

First, we did not anticipate an increase in the identification of high-risk infants. Providers followed recommendations to obtain risk stratification laboratory tests more frequently post-PI, as evidenced by significantly fewer patients having incomplete laboratory evaluations post-PI. The increase in laboratory tests resulted in a significant and unexpected increase in the proportion of high-risk infants but no change in the proportion of low-risk infants. This raises the possibility that the chosen procalcitonin cutoff value of <0.3 ng/mL was too restrictive, making it harder for an infant to be considered low risk. Despite pathway recommendations that high-risk infants undergo LP, only 53.3% of high-risk infants had an LP post-PI, which is statistically unchanged from the 60.4% of high-risk infants undergoing an LP pre-PI. This suggests that post-PI, providers changed behavior regarding low-risk infants but did not change behavior regarding high-risk infants. The reasons for the differential adoption of pathway recommendations is unclear. It is possible that clinicians did not agree with the pathway definition of high risk. Alternatively, unmeasured elements of the history or physical examination may have influenced providers. One study revealed that despite providers reporting to follow a published guideline for the management of well-appearing febrile infants, adherence to the self-selected guideline was poor.²²  Interviews with providers at our institution are needed to better understand the knowledge, attitudes, and behaviors that influence the local management of febrile infants.

The right amount of intervention for the febrile infant population is also unknown. Given the rarity of missed SBI diagnoses, it is difficult to use missed SBI diagnosis as a measure of appropriate evaluation of the febrile infant population. In our study, there were no cases of missed SBI diagnosis in the post-PI period, but this was not statistically different from the pre-PI missed SBI diagnosis rate. Even pre-PI, our organization had low rates of interventions compared to previously published rates at other hospitals. In a 2015 study, Chua et al²³  examined the association between hospital-based guidelines for LP in febrile infants aged 29 to 56 days and clinical outcomes at 32 Pediatric Health Information System hospitals. The authors reported rates for LPs (47.8%), antibiotics (52.7%), and hospitalizations (52.5%) among 25 hospitals with either no recommendation for LP or recommendations for selective LP.²³  Although the populations are not identical in age (29–56 vs 29–60 days), our institution achieved low pre-PI rates for LPs (27.2%), antibiotics (23.0%), and admissions (35.0%) compared with the report by Chua et al²³  of other Pediatric Health Information System hospitals, suggesting that there may not have had been opportunity to deimplement interventions. In addition to challenges in determining the right amount of intervention, it is difficult to know what proportion of infants should be considered high risk. Many studies include urinalysis results in risk stratification,¹⁰  making it difficult to compare our proportion of high-risk infants to reports in the literature.

Our results bring into question the utility of procalcitonin testing for risk stratification of febrile infants at our institution because we did not see an overall change to resource use; however, we are unable to draw conclusions about the utility of procalcitonin testing in the evaluation of febrile infants for populations outside our organization. Our work highlights that for organizations considering the addition of procalcitonin testing for risk stratification of febrile infants to potentially decrease or increase interventions, it is important to understand drivers of provider behavior and baseline resource use. The introduction of more testing may not result in the intended change, depending on the chosen cutoff value, existing provider behaviors, and baseline use. This may be true for other pathway recommendations. As part of pathway development and implementation, teams should consider upfront the cost/benefit ratio of additional tests and treatments and the potential need for future de-implementation of initially recommended practices that do not result in a benefit for patients or that become outdated.

There are limitations to our study. Because fever is a symptom, not a diagnosis, it is difficult to know if we identified all febrile infants during the study period. Limited chart review may have missed subtle findings that impacted clinical decisions. It is possible that other elements of the pathway contributed to our results, including the stronger recommendation in the new pathway to use bloodwork to risk stratify patients. In addition, we have not captured cost data, which are useful when assessing the impact of clinical pathways. Finally, this pathway was implemented and evaluated at a single pediatric academic institution where baseline rates of intervention were lower than those reported in other studies, which limits generalizability.

Our institution implemented a clinical pathway for febrile infants aged ≤60 days recommending procalcitonin testing for risk stratification. Among infants aged 29 to 60 days, procalcitonin testing was rapidly adopted, and resource use for low-risk febrile infants decreased. Along with an increase in laboratory evaluation, the proportion of infants categorized as high risk increased post-PI; therefore, we did not see a decrease in overall resource use. Further exploration is required to determine what drove these results, including the chosen procalcitonin cutoff value and drivers of provider behavior.

We thank all members of the pathway development team, especially Lalit Bajaj, MD, MPH, medical director of clinical effectiveness, for his leadership and guidance during pathway development. In addition, we thank Lilliam Ambroggio, PhD, MPH, for her assistance with the statistical analysis.