Article Text
Abstract
Objective We aimed to evaluate the association of height of fever with invasive bacterial infection (IBI) among febrile infants <=60 days of age.
Methods In a secondary analysis of a multicentre case–control study of non-ill-appearing febrile infants <=60 days of age, we compared the maximum temperature (at home or in the emergency department) for infants with and without IBI. We then computed interval likelihood ratios (iLRs) for the diagnosis of IBI at each half-degree Celsius interval.
Results The median temperature was higher for infants with IBI (38.8°C; IQR 38.4–39.2) compared with those without IBI (38.4°C; IQR 38.2–38.9) (p<0.001). Temperatures 39°C–39.4°C and 39.5°C–39.9°C were associated with a higher likelihood of IBI (iLR 2.49 and 3.40, respectively), although 30.4% of febrile infants with IBI had maximum temperatures <38.5°C.
Conclusions Although IBI is more likely with higher temperatures, height of fever alone should not be used for risk stratification of febrile infants.
- accident & emergency
- infectious diseases
Data availability statement
De-identified data are available upon reasonable request if a data use agreement is executed between the requesting institution and Yale.
Statistics from Altmetric.com
What is already known on this topic?
Debate exists over the importance of height of fever in determining the risk of invasive bacterial infection in febrile infants <60 days of age.
No studies have primarily examined the association between height of fever and risk of invasive bacterial infection in febrile infants.
What this study adds?
In the largest sample to date of febrile infants with invasive bacterial infection, invasive bacterial infection was more likely with higher temperatures.
In contrast to prior studies, 30% of infants with invasive bacterial infection had maximal temperatures <38.5°C and 65% had maximal temperatures <39°C.
Although the likelihood of invasive bacterial infection is greater with higher temperatures, height of fever alone should not be used for risk stratification.
Introduction
Debate exists over the importance of height of fever in determining the risk of invasive bacterial infection (IBI), specifically bacteraemia or bacterial meningitis, in young febrile infants. Although temperature is a component of several clinical prediction models for febrile infants,1 2 and higher temperatures are associated with occult bacteraemia in older febrile children,3 no studies to our knowledge have primarily examined the association between height of fever and risk of IBI in young infants. This information could prove useful to clinicians in determining the need for diagnostic testing, including lumbar puncture, among infants in the first 2 months of life. Our objective was to evaluate the association of height of fever with IBI among febrile infants <=60 days of age.
Methods
We performed a secondary analysis of a multicentre case–control study of non-ill-appearing infants <=60 days of age who presented to one of 11 tertiary-care emergency departments (EDs) between 1 July 2011 and 30 June 2016 with a documented rectal temperature ≥38.0°C at home, in an outpatient clinic or in the ED. Cases of IBI (ie, bacteraemia and/or bacterial meningitis) were identified by query of each participating hospital’s microbiology laboratory database or electronic medical record for blood or cerebrospinal fluid cultures that grew an a priori bacterial pathogen that was treated as a pathogen by the medical team. Each IBI case was matched by hospital and visit date to two febrile infant controls without IBI who had at least urine and blood cultures obtained. Ill-appearing infants and those with complex chronic conditions were excluded. Three controls from the parent study were excluded as a precise temperature ≥38.0°C was not documented in the medical record. Further details of the study methodology, including data collection, have been previously published.2
Height of fever was defined as the highest temperature recorded within 48 hours prior to the ED visit or in the ED, as parent-reported temperatures are generally reliable and are frequently used for clinical decision-making.4 We determined the median and interquartile range temperatures for those with and without IBI and compared medians (with 95% confidence interval (CI)) using median regression. We then calculated the proportion of IBI cases at each half-degree Celsius interval and computed the sensitivity, specificity and interval likelihood ratios (iLRs) for the diagnosis of IBI.5 The iLR is calculated as the proportion of infants with IBI who have a given height of fever divided by the proportion without IBI who have a given height of fever. Multiplying the iLR by the prior odds of a disease results in the post-test odds of a disease.5 We reported binomial exact 95% CIs with the analyses. A receiver operating characteristic (ROC) curve was created to show the sensitivity/specificity trade-off. To evaluate the test performance by age, we created ROC curves by age (≤28 days vs 29–60 days) and determined the area under the curve (AUC).
Results
We analysed 540 febrile infants, including 181 (33.5%) with IBI and 359 (66.5%) without IBI. Of the 181 infants with IBI, 155 (85.6%) had bacteraemia without meningitis and 26 (14.4%) had bacterial meningitis (19 of 26 had concomitant bacteraemia). Ninety-one infants with IBI (50.3%) were ≤28 days. Temperatures by day of life and IBI status are shown in figure 1.
The median temperature was higher among infants with IBI (38.8°C; IQR 38.4–39.2) compared with those without IBI (38.4°C; IQR 38.2–38.9) (median difference: 0.36°C; 95% CI 0.02 to 0.53). Among infants with IBI, the median temperature was similar for infants with bacteraemia without meningitis and those with bacterial meningitis (38.8°C (IQR 38.4–39.2) vs 38.8°C (IQR 38.4–39.0)). Temperature distribution by presence and type of IBI is shown in figure 2A. The iLRs for IBI were higher as temperature increased, except among the few infants with temperatures ≥40.0°C (figure 2C). Combining all infants, the AUC was 0.65 (95% CI 0.60 to 0.70) (figure 2B). The AUC was 0.58 (95% CI 0.50 to 0.66) for infants ≤28 days and 0.71 (95% CI 0.65 to 0.77) for infants 29–60 days of age. Sensitivity for IBI at a temperature cut-off of 38.5°C was 69.6% (95% CI 62.4 to 76.2), for 39°C was 34.8% (95% CI 27.9 to 42.2), for 39.5°C was 10.5% (95% CI 6.4 to 15.9) and for 40°C was 3.9% (95% CI 1.6 to 7.8). Specificity for IBI using a temperature cut-off of 38.5°C was 50.7% (95% CI 45.4 to 56.0), for 39°C was 85.5% (95% CI 81.4 to 89.0), for 39.5°C was 95.3% (95% CI 92.3 to 97.2) and for 40°C was 97.2% (95% CI 94.9 to 98.7).
Discussion
In this multicentre case–control study, we evaluated the association of height of fever with IBI in the largest sample to date of infants <=60 days of age. We observed that IBI was more likely with higher temperatures, particularly ≥39°C. However, almost two-thirds of infants with IBI had maximal temperatures <39°C.
In a multicentre prospective study that derived and validated a prediction rule for serious bacterial infection (SBI), the mean temperature was slightly higher for febrile infants with SBIs (which included both IBI and urinary tract infections) compared with infants without SBIs. On recursive partitioning analysis, however, temperature was not significantly associated with the presence of SBI and was not incorporated into the prediction rule.6 This cohort only analysed 30 infants with IBI,6 so our study builds on this prior work by isolating the association of maximal temperature specifically with bacteraemia and bacterial meningitis. Given our findings, risk stratification strategies that use height of fever combined with results of urine and blood testing1 2 might be employed in clinics or EDs that do not have access to procalcitonin, which is a component of newer clinical prediction rules.6 7 Additionally, our reported iLRs may be used to determine post-test probabilities of IBI at any height of fever. For instance, if the baseline pretest probability of IBI is 2%, an infant with a temperature of 39.0°C has a ‘post-test’ probability of 4.8% (2%×iLR 2.49, after conversion to and from the odds scale). The same infant with a temperature of 38.0°C would have an IBI probability of 1.2% (2%×iLR 0.60, after conversion to and from the odds scale). However, our study demonstrates that no single temperature threshold is sufficiently sensitive to exclude the diagnosis of IBI and forego diagnostic evaluation, and therefore height of fever should not be used in isolation to determine whether an infant is at low risk of IBI.
A large outpatient study reported a prevalence of IBI of 0.4% among febrile infants aged ≥25 days with temperatures <38.6°C.1 Although we were not able to assess the prevalence of IBI at different temperature thresholds, 30% of infants with IBI in our study had maximum temperatures <38.5°C. Therefore, even though the likelihood of IBI increases with higher temperatures, clinicians should use caution in deferring IBI testing for febrile infants with temperatures in the lower range.
Our study has several limitations. First, due to the case–control design we were unable to evaluate IBI prevalence, and we were only able to evaluate height of fever in matched controls, rather than the entire population of febrile infants without IBI. Second, we obtained data through medical record review, and documentation of clinical appearance, among other elements, may not be accurate. Third, we studied infants with rectal temperatures ≥38.0°C; our results should not be applied to infants with fever measured through other methods (eg, axillary, temporal). Fourth, because the infants in this study were all evaluated in tertiary-care centres, our results may not be generalisable to general EDs or outpatient settings.
Conclusions
While higher fevers, particularly ≥39°C, are more likely to represent IBI than lower-grade fevers (38.0°C–38.9°C) in febrile infants <=60 days of age, 30% of infants with IBI had a maximum temperatures <38.5°C. Height of fever alone should not be used for risk stratification of febrile infants.
Data availability statement
De-identified data are available upon reasonable request if a data use agreement is executed between the requesting institution and Yale.
Ethics statements
Ethics approval
The parent study was approved by each hospital’s institutional review board.
Footnotes
Correction notice This paper has been corrected since it was published online. A degree symbol has been added before the ‘C’.
Collaborators The following collaborators in the Febrile Young Infant Research Collaborative acquired data for this study and/or contributed to the parent study: Elizabeth R Alpern, MD, MSCE (Ann and Robert H Lurie Children’s Hospital of Chicago, Chicago, Illinois), Whitney L Browning, MD (Vanderbilt University School of Medicine, Nashville, Tennessee), Elana A Feldman, MD (Lucile Packard Children’s Hospital Stanford, Palo Alto, California), Catherine E Lumb, MD (University of Alabama at Birmingham, Birmingham, Alabama), Russell J McCulloh, MD (Children’s Mercy Hospital, Kansas City, Missouri), Christine E Mitchell, BSN (Children's Hospital of Philadelphia, Philadelphia, Pennsylvania), Lise E Nigrovic, MD, MPH (Boston Children’s Hospital, Boston, Massachusetts), Nipam Shah, MBBS, MPH (University of Alabama at Birmingham, Birmingham, Alabama), Samir S Shah, MD, MSCE (University of Cincinnati College of Medicine, Cincinnati, Ohio), Sarah J Shin, BSN (Children's Hospital of Philadelphia, Philadelphia, Pennsylvania), and Derek J Williams, MD, MPH (Vanderbilt University School of Medicine, Nashville, Tennessee).
Contributors Study concept and design: KAM, MIN, PLA. Data acquisition: MIN, CMP, SD, MEW, AGD, RCL, LFS, RDM, SNR, CW, FB, PLA. Data analysis: KAM, PLA. Interpretation of data: all authors. Drafting of the initial manuscript: KAM, PLA. Critical revision of the manuscript for important intellectual content: all authors. Statistical analysis: KAM. Study supervision: PLA. Approval of the final manuscript for submission and agreement to be accountable for all aspects of the work: all authors.
Funding This project was supported by grant numbers K08HS026503 (KM) and K08HS026006 (PA) from the Agency for Healthcare Research and Quality (AHRQ) and by CTSA grant number KL2 TR001862 (PA) from the National Center for Advancing Translational Sciences (NCATS), a component of the National Institutes of Health (NIH). The content is solely the responsibility of the authors and does not represent the official views of AHRQ or NIH.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.