We will perform a multicenter study of infants with hypothermia in the ED using the Pediatric Emergency Medicine Collaborative Research Committee (PEM CRC) to characterize infants with SBI and HSV and to compare infants with hypothermia with and without SBI. The PEM CRC offers a framework for multicenter investigations, supporting and facilitating the development of valuable and rigorous collaborative research in pediatric emergency medicine. PEM CRC plays a crucial role in developing high-quality protocols and identifying interested coinvestigators across its member sites. The PEM CRC has established a centralized data center at Texas Children’s Hospital, Baylor College of Medicine (BCM). This center offers data collection support services, including front-end logic checks for electronic data sheets, high-speed automated data sheet scanning, and database creation, encryption, and processing, and has conducted several important studies focused on topics relevant to the care of uncommon, but clinically important, conditions that occur in children presenting to the acute care setting [20-23]. We will include infants ≤90 days with a blood culture performed and who presented to a contributing hospital with a rectal temperature of <36.5 °C in the ED over a 10-year period. We will use a case-control approach. BCM will serve as the data coordinating center. A graphical summary of the study protocol is provided in Figure 1.

Infants ≤90 days of age presenting to the ED of a participating pediatric ED with hypothermia, defined as a minimum rectal temperature of <36.5 °C in the ED over a 10-year period (January 1, 2013, to December 31, 2022). The <36.5 °C criterion is determined based on preliminary work performed by the investigators [14] and uses the World Health Organization criteria to define hypothermia in neonates [25]. We will allow participating sites that are unable to provide data from 2011 to contribute data due to varying adoption of EHR systems across these hospital systems [23].

The exclusion criteria are as follows:

As our inclusion period includes the period of time prior to the adoption of ICD-10 in the United States, we will crosswalk the ICD-9 (International Classification of Diseases, Ninth Revision) diagnosis codes using generalized equivalence mappings, a resource developed by the US Centers for Medicare & Medicaid Services that provides forward and backward mappings between the diagnosis coding systems [32].

The EHR will be queried each participating ED to identify eligible age (≤90 days) encounters from 2013 to 2022 with at least 1 documented episode of hypothermia during the ED encounter. Due to the large volume of records anticipated to be identified at each site (varying between 1500 and 3000 encounters on initial efforts from 7 hospitals), data collection will proceed through 2 phases to maximize the efficiency of manual chart review and avoid the collection of unnecessary data. First, we will use a standardized data collection of all variables that can be ascertained from the EHR through computerized queries. This will include (1) demographic details such as age, race, ethnicity, and sex, and administrative data (payor status, year, and season of visitation); (2) medical history, ED chief complaints, multiple temperature measurements throughout the visit, and heart rate and blood pressure data; (3) laboratory test results, including complete blood counts (CBCs), metabolic panels, and tests for viral and bacterial pathogens (including the results of urine, blood, and CSF culture, results from urinalysis and urine dipstick testing, and results from HSV testing); (4) results from chest radiography, if performed; (5) medication administrations (antipyretics, antibiotics, and antiviral medications); (6) ED outcomes and details of hospitalization such as neonatal or pediatric intensive care unit admissions; (7) follow-up information on any return visits (>7 days); and (8) diagnosis and procedure codes for the index encounter and all prior encounters. Considering race and ethnicity and social constructs, we will collect these data acknowledging the limitations in how these data may be reported in a retrospective study and that the means of reporting may vary by institution and over time (eg, self-reported vs determined by registration staff). These data will be reported descriptively. We will not incorporate race or ethnicity in prediction models. To minimize duplication of efforts and maximize homogeneity with data collection procedures across sites, we will promote the sharing of Structured Query Language queries using Epic and Cerner platforms across participating sites. Data will be uploaded from the EHR via a secure, HIPAA (Health Insurance Portability and Accountability Act)-compliant REDCap (Research Electronic Data Capture; Vanderbilt University) database hosted at BCM using a data upload. Data from each site will undergo quality assurance by the study team to ensure that the uploaded data from each site is free from systemic errors.

Trained study-site PIs will abstract variables from the EHR and enter them into a secure, HIPAA-compliant REDCap database hosted at BCM. The initial upload of patient data will include data for all encounters meeting minimum inclusion criteria (age ≤90 days presenting to the ED of a participating pediatric ED with at least 1 episode of a temperature <36.5 °C during the ED encounter). Exclusions will then be applied using the methodology described earlier. Outcome data with respect to SBI and HSV will be defined by the author team using a priori defined criteria (described in the Outcome Definitions section). Cases will be identified by querying each hospital’s EHR system or microbiology laboratory in the ED during the study period to identify infants ≤90 days with SBI. Controls will include infants with a minimum rectal temperature of <36.5 °C in the ED who had a blood culture performed and with no SBI identified. We will perform 3:1 nested matching of controls to cases based on site and proximity of the date of the encounter. Among encounters where blood cultures were obtained after antibiotic administration (which may be ascertained during manual chart review), cases or controls deemed ineligible will be replaced, and case-control matching will be reperformed within that subgroup. As prior work suggests that only some infants with hypothermia have a blood culture performed, the inclusion of only those infants with a culture performed will allow for the construction of models for infants for whom a clinician may have a suspicion for SBI.

After matching cases and controls, we will provide the record numbers of these selected cases to each of the site PIs. Site PIs will manually review each encounter to ensure whether it fulfills the inclusion criteria and does not meet any exclusion criteria. If cases or controls are found to meet exclusion criteria based on manual review, new assignments for these will be provided based on the next eligible matched control based on the proximity of the encounter. Site PIs will manually abstract the following variables: (1) birth history and relevant past medical history, including gestational age, birthweight, and maternal history (including history of group B Streptococcus positivity, maternal fever, and use of peripartum antibiotics), admission to the neonatal intensive care unit, presence of significant medical conditions diagnosed previous to the index ED encounter, immunization status, prior treatment for hyperbilirubinemia, and sick contacts in the home; (2) presenting history (history of seizure, apnea, abnormal behavioral or lethargy, cough, congestion, vomiting, diarrhea, alteration in feeding, decreased feeding, concern for environmental exposure, presence of reported hypothermia prior to ED arrival, lowest temperature prior to ED arrival, and mode of temperature assessment); and (3) physical examination data including (a) general ill-appearance (sick, toxic, inconsolable, or meningismus), (b) poor perfusion (shock, decreased pulses, cyanotic, and pale or ashen colored), (c) mental status changes (somnolence, lethargy, and altered), (d) sunken or full fontanelle, and (e) presence of respiratory distress, apnea, cyanosis, congested, seizure, clinical bronchiolitis, jaundice, or rash on examination. Historical and physical examination findings will be categorized as present or received or absent or not received based on documentation in the EHR. Variables categorized as either undocumented or missing based on the criteria in the manual of operations.

All data abstractors will undergo web-based training via teleconferences and have access to a detailed manual of operation at each site. Regular updates to monitor study progress and address questions throughout the chart review process during each phase of data collection will be conducted. The REDCap database will be established according to best practices, including data validation and strategies to minimize missing data, ensuring precise data capture. Additionally, we will conduct routine monitoring and compliance reviews of the database to maintain data accuracy during the study.

Our primary outcome of interest is culture-confirmed SBI, defined as positive blood, urine, or CSF cultures [33-35]. Classification for pathogenic versus contaminant bacteria (from false positive cultures) will be determined based on established criteria for the classification of suspected pathogens published by the Pediatric Emergency Care Applied Research Network [36]. Indeterminate cases will be reviewed by the investigator team (which includes a specialist in pediatric infectious disease; ATC). UTI will be defined as the growth of ≥10,000 colony forming units/mL of a pathogen in urine culture plus an abnormal urinalysis (defined as the presence of leukocyte esterase, nitrites, or >5 white blood cells per high-power field on urine microscopy) [37]. We will use SBI as our outcome of interest instead of invasive bacterial infection (IBI; defined as bacteremia or bacterial meningitis and which is a preferred definition in the literature pertaining to infants with fever [26]), given the relative rarity of IBIs in previous publications and our pilot work (~2%). Cases that remain ambiguous will be reviewed by the investigator team (which includes a specialist in pediatric infectious disease; ATC), blinded to the context of relevant study predictors. We will plan an a priori sensitivity analysis limited only to infants with IBI, wherein we assess the performance of the SBI-based predictive model when using this alternative outcome. This 2-tiered approach has been previously applied to research on young infants with fever [16]. We will evaluate return visits within 7 days. For any return visits, we will review testing performed to identify potential missed cases of SBI. Our secondary outcome of interest will be HSV infection. HSV will be identified by the presence of a positive polymerase chain reaction or culture from any source [23].

We have selected candidate predictors based on a review of prior literature [4,5,9,12,15,16,38] and following discussion among members of the study team. For each predictor, we will seek to ensure that data elements can be reliably and accurately extracted from the EHR. A list of candidate predictors that we will study is provided in Table 1.

As with all retrospective chart reviews, there is a risk of encountering missing or inaccurately documented data. We will implement clear definitions and provide detailed instructions for data extraction and interpretation, with a particular focus on subjective data points. Additionally, we will analyze the factors associated with such missingness. These measures will help reinforce the assumption that the missing data are missing at random, thus enhancing the robustness of the study findings. We will evaluate complete case versus multiple imputation-based approaches for missing data for data elements that are missing in <20% of cases. Data missing in excess of this proportion will be excluded from multivariable modeling.

We estimate identifying 41,250 infants with hypothermia in the proposed study from 25 sites over 10 years. We expect that at least 825 (2%) will have an SBI, resulting in a 95% CI width for the prevalence estimate of ±0.14%. Of these, approximately 25% (n=205) will have an IBI [5,6,14]. Greater than 80% of patients with hypothermia who have a blood culture also have a CBC performed [44]. From our sample of 41,250 infants anticipated from objective 1, we estimate that 25% (approximately n=10,310) will have a blood culture performed; and of these, 8250 will additionally have a CBC. As an exploratory component of this aim, we will evaluate inflammatory markers (specifically, CRP and procalcitonin). Prior work suggests that the use of these biomarkers is increasing over time, with 30% of patients with procalcitonin and 39.6% with CRP in 2019 [19]. Projected from our pilot data, we expect to have 650 patients with procalcitonin from our study, of whom 65 will have an SBI.

For our third objective, we anticipate having a sufficiently large sample size for the development of robust multivariable models. Under the 3:1 case-control design, the proportion of individuals with SBI in the dataset will be fixed at 0.25. Assuming 825 patients with SBI from objective 1, we estimate that the concordance index for the model will have a maximal CI half-width of 2.3%. At the expected model concordance index of roughly 0.80, the CI half-width will be 1.8%, which represents substantial precision in estimating the model’s quality of fit [45].

This study involves the review of medical records for patients. Institutional review board (IRB) approval as exempt from the requirements of informed consent has been obtained from the Ann & Robert H. Lurie Children’s Hospital of Chicago (IRB #2023-5779) and the 26 other participating hospitals (Table 2). Data use agreements have been established between all hospitals and BCM. All data uploaded into the REDCap repository are deidentified. No compensation is provided to participants. Adverse events, including breaches of confidentiality, will be reported to the local IRBs of Ann & Robert H. Lurie Children’s Hospital of Chicago.

This study aims to identify risk factors for SBIs among young infants with hypothermia and to develop a predictive model to inform clinical decision-making. We hypothesize that a subset of infants with hypothermia, particularly those with specific clinical and laboratory features, are at increased risk for SBI and that a multivariable model can reliably stratify this risk. By leveraging a large multicenter dataset, we anticipate that our findings will support more targeted evaluation and management strategies for this vulnerable population.

Using a large, multicenter sample of infants presenting to the ED with hypothermia, this study will characterize the epidemiology of SBI and HSV and assess the diagnostic value of biomarkers in the prediction of SBI. This work will culminate in a statistical model to determine the risk of SBI among this patient population using clinical and laboratory factors and facilitate the development of a parsimonious model.

First, preliminary research suggests that not all infants with hypothermia receive testing for SBI in EDs and that the proportion who do receive testing varies based on the criteria used to define hypothermia [15,19,44,46]. However, this is of limited impact, as our model will allow physicians to risk-stratify children for SBI on the subset of patients for whom diagnostic testing may be considered due to a suspicion of SBI. To further investigate this potential bias, we will compare demographics among children who are tested and who are not tested for SBI. A second limitation pertains to missing data; some infants may not have documentation of relevant risk factors or data required to calculate outcomes. We will evaluate both complete case versus selective multiple imputation-based approaches for missing data, acknowledging that missingness may be related to clinical decisions, such as ordering blood tests. Third, we plan to construct a parsimonious model that is generalizable (ie, both reproducible and transportable) to best serve the needs of clinicians who manage infants with hypothermia in the ED. We will assess the internal validity of the model using bootstrap resampling techniques, from which we will derive overoptimism indexes for various measures of model fit, including the concordance index [47].

Although this will be the largest sample of infants with hypothermia in the ED, our findings may not be completely representative of the prevalence of SBI among these patients because most participating institutions are academic institutions and pediatric hospitals; therefore, the findings may not be generalizable to different care settings. However, the PEM CRC represents geographic and socioeconomically diverse areas, and we would not expect that a systematic bias would be introduced.

The majority of prior work has been limited to single-center data and therefore is of limited generalizability [6,9,12]. Limitations of prior multicenter work include small sample size, incorporation of infants with fever within the study sample, inclusion of infants with hypothermia during their hospitalization (and not only the ED stay, where management options differ) [4,5], and lack of internal validation. We will build upon these prior studies by providing a larger, analytically robust approach to identify risk factors for SBI in this population. The proposed study will be significantly larger, focused on the initial ED presentation and decision-making, and represent a more diverse population than prior studies. It will address a continued knowledge gap in the care of young infants.

The findings from this study allow for additional research that will potentially allow for the implementation of real-time risk calculation for the presence of serious infections among infants with hypothermia. Our preliminary results (which corroborate with other work [14]) suggest that SBIs occur infrequently but are associated with high mortality, heightening challenges in risk stratification in this population. Future steps may therefore be intended to safely facilitate a reduction in low-value testing. Following the completion of this study, we will seek to externally validate and implement the model via clinical decision support systems and guidelines, reducing both overtreatment in low-risk infants and health care variability, thus improving care standardization and reducing the incidence of low-value care. This will potentially lower health care costs, decrease antibiotic exposure and unnecessary hospitalization, and alleviate the emotional and financial burden on families. At the study’s conclusion, dissemination efforts will involve broad participation from key partners in pediatrics, emergency medicine, hospital medicine, primary care pediatrics, infectious disease, and dissemination or implementation science.

We aim to address the critical gap in evidence-based guidance for managing hypothermia in young infants presenting to the ED. By leveraging a large, multicenter dataset, we will delineate the epidemiology of SBI and HSV in this vulnerable population, evaluate the predictive value of biomarkers, and develop a clinically useful prediction model to stratify the risk of SBI. Our findings are expected to enhance patient care by reducing unnecessary invasive procedures and antibiotic exposure. If a robust model is developed and validated externally, these results will contribute to more standardized and evidence-based care, potentially reducing health care costs and the burden on families. Despite limitations such as variability in testing and the generalizability of the findings to nonacademic settings, this research marks a significant advancement in understanding and managing hypothermia in young infants.