Performance of Febrile Infant Algorithms by Duration of Fever Free

This is a secondary analysis of a prospective registry that includes consecutively all infants <90 days old with fever without a source (FWS) attended in 1 pediatric ED between 2008 and 2021. The database used in this registry has been described elsewhere.^11 –13  Data on febrile infants are prospectively recorded in the medical record by the attending clinician. On a monthly basis, 2 resident physicians review the data of all children <3 months of age and enter the patients in the database. This database is managed and reviewed by 2 emergency pediatricians. This pediatric ED is in a tertiary hospital and each year receives ∼55 000 children ≤14 years of age, with about ∼2300 of them being infants ≤90 days. We monitor the evolution of the patients by reviewing the medical records of those admitted to ward, and the primary care medical reports and any unscheduled return visits to the ED for those who managed as outpatients. A pediatric emergency physician reviews the data after entry.

• FWS: axillary or rectal temperature ≥38 °C (100.4 °F) measured either at home or at the ED (in the ED, temperature was measured rectally in all patients), in an infant in whom, after taking a medical history and conducting a physical examination, it is not possible to identify the source of the fever.

• Well-appearing: normal appearance, work of breathing and circulation to skin according to the Pediatric Assessment Triangle,¹⁴  as assessed by a physician within an hour of the infant arriving in the ED. Specifically, infants were classified as well appearing if the 3 components of the Pediatric Assessment Triangle were defined as normal in their medical report.

• Urinary tract infection: growth of ≥10 000 colony-forming unit/mL in patients with positive findings on urine dipstick (either leukocyte esterase or nitrite test) in a sample obtained by urethral catheterization.

• IBI: bacterial meningitis or bacteremia.

  • Bacteremia: isolation of a bacterial pathogen in a blood culture or a polymerase chain reaction test. Isolation of Staphylococcus epidermidis, Propionibacterium acnes, Viridians group Streptococci, Diphtheroides, or any other bacterium commonly considered a contaminant in an immunocompetent patient without cardiac disease, ventricular-peritoneal shunt, central catheters, or another indwelling device was not classified as a bacteremia.

  • Bacterial meningitis: detection of a bacterial pathogen in the cerebrospinal fluid, with or without associated pleocytosis.

• Low-risk patients: patients considered as low risk by each rule were:

  • PECARN: those who are well-appearing, >28 days, with normal urinalysis, PCT <0.5 ng/mL, and ANC ≤4000 cells/µL;

  • AAP: those who are well-appearing, >21 days, with normal urinalysis, PCT <0.5 ng/mL, and ANC ≤4000 cells/µL. If PCT was not available, those with maximum temperature <38.6 °C, CRP ≤20 mg/L, and ANC ≤5200 cells/µL;

  • Step-by-Step: those who are well-appearing, >21 days, with normal urine dipstick, PCT <0.5 ng/mL, CRP ≤20 mg/L, and ANC ≤10 000 cells/µL.

The PECARN and AAP rules are only applicable in patients between 29 and 60 and 8 and 60 days of life, respectively.

For the purpose of this subanalysis, we used the following exclusion criteria:

• lacking data on duration of the fever

• none of the studied biomarkers obtained

• blood culture not obtained

Duration of the fever was determined according to the time elapsed since the moment the family reported that they registered a temperature ≥38 °C and the moment the patient is evaluated by the pediatrician. For those patients whose chief complaint was other than fever and in which the fever was first registered on the arrival at the ED, we considered an evolution time of “0". Given that the time between the time the patient is attended to and the blood tests are drawn is not recorded, we believe it is important to point out that in our ED, the blood tests in young febrile infants are obtained, by protocol, in the first hour after the patient’s arrival to the ED, regardless of the duration of the fever reported by the family.

A descriptive statistical analysis was carried out for all collected variables. Normally distributed data were expressed as mean and SD; non-normally distributed data were expressed as median and interquartile range; categorical variables were reported as percentages. For nonnormally distributed data, comparison was performed using Wilcoxon test; comparison of normally distributed data were performed using the t test for independent samples. If the comparison was between more than 2 groups, it was performed using the analysis of variance test. For categorical data, the χ² test was used, and Fisher exact test, if any of the variables had <5 subjects in any of the categories. To analyze whether over the years studied there was any variation in the time of evolution of fever with which families consulted, a Poisson regression was performed, adjusting for possible confounding factors such as the general condition or age of the patients. For the objective of the study, we classified the patients into 3 groups based on caregiver-reported hours of fever (<2, 2–12, and >12) and analyzed the area under the curve of the biomarkers (WBC, ANC, CRP, and PCT) and the diagnostic performance of 3 commonly used clinical decision rules (PECARN, AAP, and Step-by-Step). To analyze the accuracy of the clinical rules, values of sensitivity (Sn) and specificity (Sp) were calculated, as were their 95% confidence intervals (95% CI). As a note, for the evaluation of the PECARN and the AAP rules, the age groups of 29 to 60 and 8 to 60 days of life were respectively used, as advised by the authors after publication. Data were analyzed with Stata 17 (Stata Corp, College Station, TX).

We obtained the approval of the Ethical Committee of our hospital. Given that all the data were extracted from a database in which patients were anonymous and inclusion in the registry did not imply any additional interventions, informed consent was not required.

During the period of the study, we included 2993 febrile infants <90 days old in the registry. Of them, we excluded 308 (10.3%) because the duration of the fever was not registered, 112 (3.7%) because a blood culture was not obtained, and 8 (0.3%) because none of the studied biomarkers was obtained. After applying the exclusion criteria established for this study, we finally included 2565 (85.7%). The characteristics of the patients are shown in Table 1. The median duration of the fever was 4 (interquartile range, 2–12) hours, with 633 (26.3%) patients evaluated with <2 hours since the fever was first noticed, 1228 (47.9%) between 2 and 12 hours, and 704 (27.5%) at least 12 hours after the beginning of the fever. In Table 2, we show the characteristics of the 3 groups.

The rate of infants attended to in the first 2 hours after the fever was increased from 17.0% in the first year of the study to 30.0% in the last year (P = .02). This shorter time to presentation of patients each year of the study was observed even after adjusting for age and general condition of the patients, with a β coefficient of –0.04 (P < .01), meaning that for each year elapsed in the sample, the duration of fever on arrival at the hospital was 0.96 times that of the previous year.

Seventy-six patients (3.0%) were diagnosed with an IBI (69 bacteremia, 3 meningitis, and 4 meningitis with associated bacteremia). Of the 19 infants diagnosed with IBI resulting from Streptococcus agalactiae, 10 (52.6%; 95% CI, 31.7–72.2) presented to the ED with less than 2 hours of fever duration. In Fig 1, we show the area under the curve for each biomarker for identifying IBIs by hours of evolution of the fever. The performance of WBC, ANC, and CRP decreased significantly in patients with less than 2 hours of duration of the fever and only PCT remained with a similar performance. A secondary analysis of biomarker performance in the age groups included in the AAP (6–60 days of life) and PECARN (29–60) rules is shown in Supplementary Table 6. Because the performance of the biomarkers seems to change with the duration of fever, we show in Table 3 the Sn and Sp values of the cutoff points for each 1 used in the clinical decision guidelines. Because Step-by-Step and the PECARN and AAP clinical guidelines consider some patients to be high risk based on their general condition, in this table, we have included only well-appearing patients.

Table 4 summarizes the diagnostic values of the Step-by-Step, the PECARN, and the AAP rules. The lowest Sn was identified in 3 rules among patients with <2 hours of fever: 83.3% (95% CI, 64.1–93.3) for the Step-by-Step, 90.9% (95% CI, 62.3–98.4) for the PECARN rule, and 88.2% (95% CI, 65.7–96.7) for the AAP rule. However, the group with the highest Sn varied, being those with fever of 2 to 12 hours for the Step-by-Step (100%; 95% CI, 89.8–100) and those with ≥12 hours of fever for the PECARN (100%; 95% CI, 51.0–100), and the AAP (100%; 95% CI, 74.1–100) rules. Among patients with <2 hours of fever, the PECARN and the AAP rules respectively misdiagnosed 1/11 (9.1%) and 1/16 (6.3%) of the IBIs in the group of patients in which these rules are recommended to be applied. On the other hand, the Step-by-Step rule misdiagnosed 4/24 (16.7%) of the IBIs in the entire population (Table 5). Of the 3 infants with fever <2 hours of evolution and S. agalactiae bacteremia with normal biomarkers (according to the Step-by-Step rule), 2 had the blood test repeated at 6 and 15 hours of fever, respectively. Both patients had elevated PCT and 1 also had elevated CRP. Both were admitted with antibiotics. The third patient was discharged, and the family was contacted by telephone after the blood culture results were received. The patient continued with fever, but in good general condition. Laboratory tests were repeated 30 hours after the onset of fever, with elevated PCT and CRP, and the patient was admitted with intravenous antibiotics. All 3 patients did well.

When analyzing the subgroup of well-appearing patients older than 21 days of age and with normal urine dipstick, in patients with <2 hours of fever, the usual cutoff point of 0.5 ng/mL for PCT showed an Sn of 37.5% and an Sp of 95.7%. In these infants, PCT = 0.14 ng/mL was the highest cutoff point with an Sn of 100% (Sp, 53.8%). The performance of this cutoff point was better in this subgroup of patients than the combination of biomarkers of Step-by-Step (ANC >10 000 cells/mcL, CRP >20 mg/L, and PCT ≥0.5 ng/mL; Sn = 50.0% and Sp = 82.2%), and of the AAP and PECARN rules (ANC >4000 cells/mcL and PCT ≥0.5 ng/mL; Sn = 83.3% and Sp = 58.3%). We only applied the biomarkers of the PECARN rule to patients aged between 29 and 60 days of life.

Our study shows that commonly used biomarkers and clinical decision rules in young febrile infants have a poorer performance if they have a very short duration of fever. Although this is not the first study to evaluate the performance of biomarkers in febrile infants according to the duration of fever,^8 ,9  it is, to the best of our knowledge, the 1 with the largest sample size in patients with such a short evolution time.

During the past 3 decades several changes have caused the classic clinical rules, which were very useful some decades ago, to show a poorer performance today.¹⁵  Changes in the epidemiology of invasive infections, improvements in the management of intrapartum antibiotherapy, or the introduction of different vaccines have made it necessary to develop new clinical decision rules.^4 ,16 

However, a variable that has not been considered is that multiple social factors have led patients and their families to seek for acute care earlier. In the study by Dagan et al published in 1988,¹⁷  half of the patients who consulted for fever of more than 12 hours of evolution, whereas in more recent studies, the median number of hours of fever evolution was 2.⁷  In our study, the younger the patient is, the sooner they are brought to the ED. This poses a problem, first of all, because the usual biomarkers reach their release peaks at longer times.^18 ,19  In addition, because the current clinical rules were derived from patient samples with longer evolution times, the cutoff points of the biomarkers used may not be sensitive enough.^6 ,20 

Regarding the latter, it is striking that the sensitivity of the Step-by-Step rule is maximal in infants with a duration of fever between 2 and 12 hours, whereas that of the PECARN rules is a maximum of 12 hours’ evolution. Precisely when we observe the duration of fever in the studies from which both rules were derived, we see that the Step-by-Step sample presented a median of 5 hours of fever, with an interquartile range of 2 to 12. In the PECARN rule, however, there were almost 40% of patients with fever of more than 12 hours of evolution.^6 ,20  This may explain, at least partially, that the optimal performance of each rule is in a different time range.

Biomarker kinetics has been the main explanation given for the differences in biomarker performance as a function of length of fever. However, when analyzing the bacteria isolated in the different groups, we see how the etiology of IBI is different according to the hours of fever. In the group with a shorter evolution time, S. agalactiae isolations are more frequent, while Escherichia coli is the most commonly isolated bacterium in patients with more than 2 hours of evolution, in addition to other gram-negative bacteria. Gangoiti et al demonstrated that the biomarker response was different depending on the isolated bacteria,²¹  so the possibility that not only are the biomarkers kinetic, but also that the etiology of the IBIs is the cause of these differences should not be ruled out. It is concerning that most potentially severe IBI, those caused by S. agalactiae, are identified in those infants carried very promptly to the ED, when the accuracy of clinical predictor rules is limited.

Once the lower performance of biomarkers and clinical rules in patients with a shorter time of evolution has been observed, the question arises as to how these patients should be managed. First, it is clear that patients who do no dot appear well or are younger than 21 days of age should be considered at high risk of IBI regardless of the results in the ancillary tests.^13 ,22 ,23  However, in older well-appearing patients, several options are considered that correspond to the variability in the management of these patients.²⁴  A possible approach would be to repeat the determination of biomarkers after a few hours. Intuitively, this approach could make sense, given that, as mentioned previously, both CRP and PCT reach maximum levels after a few hours.^18 ,19  However, there is limited evidence regarding the use of biomarkers serially, and the appropriate time frame for repeating the determination has not been defined.²⁵  In addition, performing the same test several times increases the probability of obtaining a false-positive result, thus increasing iatrogenicity in healthy patients.²⁶  Also, we must also take into account that, when determining the appropriate cutoff points, the current ones were established with samples that included patients with short evolution times. It is likely that if a second determination is to be performed a few hours later and the same cutoff points are used, the specificity of that cutoff point will decrease because there will be more healthy patients in whom the biomarkers will have risen above the cutoff values. However, in our sample, this phenomenon only occurred in CRP because Sp values for the cutoff point of 0.5 ng/mL for PCT remain the same regardless of fever duration.

Another option could be to wait a few hours to perform the first analytical extraction, allowing time for the biomarkers to reach their maximum performance. This approach, similar to that taken in patients with suspected paracetamol poisoning,²⁷  may not be safe considering that 40% of IBIs in the group of patients with less than 2 hours of fever were caused by S. agalactiae. Although these patients were in good condition on arrival at the ED, and none of the 3 worsened clinically during observation, using this strategy means assuming the risk of developing sepsis or meningitis while waiting for a delayed blood test. Furthermore, as mentioned previously, the systematic determination of biomarker values after a certain number of hours of fever would make it necessary to reevaluate what the optimal cutoff points for these biomarkers might be.

Finally, the option that the authors consider most appropriate would be to establish specific clinical rules for these patients with a shorter evolution time. In our sample, as we have seen, a cutoff value for procalcitonin with an Sn of 100% (0.14 ng/mL) was very similar to that proposed by Maniaci et al.²⁸  The Sp of this cutoff point appeared to be even better in this subgroup of patients than that of the combination of biomarkers of the analyzed clinical rules. However, the sample size and the single-center nature of our study do not allow clear conclusions to be drawn in this regard. Larger multicenter studies are needed to confirm these results.

Our study has several limitations. Length of the fever was not registered in all the infants evaluated but was completed in 89.7% of cases. The prevalence of IBI was similar in those patients without fever duration being registered, so it would not significantly affect the results. In addition, and very importantly, the duration of the fever is recorded based on the information given by the family, which may lead to some inaccuracy about the exact time. However, these are the conditions in which we find ourselves in clinical practice, and therefore we consider that our results reflect those that we could find in real conditions. On the other hand, it must be taken into account that the patients included in the study are younger than 3 months, some of them even neonates. In these patients, the degree of vigilance shown by the parents is much greater, as well as the number of manipulations, including diaper changes and breastfeeding. In addition, the accuracy of parents in determining whether the child has a fever is quite good, especially in ruling out fever.^29 –31  The authors believe, therefore, that it is unlikely that an infant of this age would remain febrile for a significant period without the family noticing. Second, it is a single-center study, and therefore we should be cautious in extrapolating results to other centers, particularly because of the differences in the prevalence of IBI between populations. However, to our knowledge, this study has the largest sample of patients with an evolution time of <2 hours. In addition, our registry includes all patients evaluated, on a 24/7 basis, prospectively, so the results are quite robust. Third, the time cutoff points were selected ad hoc, not by statistical criteria. The cutoff point of 2 hours was chosen based on a previous study carried out by our research team on the application of the step-by-step approach.³²  This study showed that all misdiagnosed IBIs occurred in infants with <2 hours of fever. On the other hand, the 12-hour cutoff point was chosen based on CRP kinetics. Nevertheless, even if it was not by a mathematical method, we believe that our results adequately support the selection of the cutoff points. Finally, we include only infants with FWS. Hence, our results can be only applied to febrile infants with FWS, not being possible to extrapolate them to infants with fever with a source, such as upper respiratory tract infection symptoms.

Management of febrile young infants with a very short duration of fever should be more cautious. The performance of blood biomarkers, except for procalcitonin, in febrile young infants decreases in these infants, and this may alter the performance of different clinical decision rules.

ANC

absolute neutrophils count

CI

confidence interval

CRP

C-reactive protein

ED

emergency department

FWS

fever without source

IBI

invasive bacterial infection

PCT

procalcitonin

PECARN

Pediatric Care Applied Research Network

Sn

sensitivity

Sp

specificity

WBC

white blood cell