Infants With Hypothermia: Are They Just Like Febrile Infants? Free

In 2021, the American Academy of Pediatrics published a clinical practice guideline (CPG) for the management of well-appearing febrile infants 8 to 60 days old.¹  This CPG is based on decades of research to identify the optimal approach to risk-stratify febrile infants on the basis of their risk of a serious bacterial infection (SBI), particularly bacteremia and bacterial meningitis (invasive bacterial infections [IBI]). Despite a nonnegligible risk of SBI of ∼2% to 3%, there is no corresponding CPG to guide the management of infants with hypothermia, resulting in wide variability in diagnosis, evaluation, and management.² 

In this issue of Hospital Pediatrics, 2 studies build on the limited evidence for risk stratification of young infants with hypothermia with 2 different approaches. In a 4-site case-control study of 171 infants with temperatures <36.5°C and blood cultures obtained, Money et al identified several risk factors for SBI, including age, fever, and several laboratory variables, most strongly the presence of an abnormal urinalysis. Combining these risk factors produced full and reduced (abnormal urinalysis only) prediction models with moderate sensitivity and high specificity for SBI, though several infants with bacteremia and bacterial meningitis were misclassified as low-risk by the models.³  Westphal et al applied existing (with modifications to some) febrile infant risk stratification algorithms to 314 infants with hypothermia and found them to have fairly high sensitivities but low specificities in identifying SBI and IBI, though with small numbers of IBI in particular, the 95% confidence intervals were wide.⁴ 

These studies highlight a need for more guidance in the workup of infants with hypothermia. Without a prediction model specific to this population, many physicians default to using risk-stratification algorithms developed for febrile infants, a population extensively studied.⁵  Of the 6 febrile infant algorithms applied in Westphal’s study, the sensitivities of most were high but many had positive likelihood ratios around 1, indicating that a positive test (ie, being identified as nonlow-risk does not alter the pretest probability of SBI or IBI). Although application of these algorithms may help clinicians detect most infants with hypothermia and SBI or IBI, it potentially exposes many more infants to unnecessary painful procedures and antibiotics. This study highlights the potential pitfalls in applying algorithms derived and validated in 1 population, febrile infants, to another population. The procalcitonin-based Pediatric Emergency Care Applied Research Network (PECARN) prediction rule, for example, was derived and validated in febrile infants using robust statistical methods, and it performs well in febrile infants, including a higher specificity than other algorithms.⁶  It is understandable why clinicians would reflexively apply this algorithm to infants with hypothermia, though it was not intended for use in this population. Although Westphal et al evaluated the performance of the PECARN prediction rule using the originally derived procalcitonin level of 1.71 ng/mL,⁴  and not the rounded cutoff value of 0.5 ng/mL, which is currently recommended,¹  it is necessary to conduct a similar derivation and validation study in infants with hypothermia to identify the optimal combination of predictors.

Money et al used logistic regression to do just this: Identification of predictors of SBI and IBI in infants with hypothermia. Although the models had moderate sensitivity and high specificity overall, similar to many febrile infant algorithms, abnormal urinalysis was the strongest predictor and the models were primarily derived on the basis of urinary tract infections, with limited numbers of IBIs and several infants with IBI misclassified as low risk. Therefore, although these models are an important step toward evidence-based risk stratification of infants with hypothermia, they are not ready for use in the clinical environment. Furthermore, both studies only included infants with hypothermia who received at-minimum blood cultures, indicating that the treating physicians already had some level of suspicion for infection. In the study by Money et al, 934 out of 3376 infants with hypothermia received blood cultures, meaning that, for another 2442 infants, some other information or clinical gestalt was guiding the treating physician to not evaluate for IBI. This finding highlights a major challenge with derivation of a prediction model for IBI in infants with hypothermia: Spectrum bias. Any prospective research on this population would need to account for this spectrum bias in its design, and most likely would result in a prediction model that can be applied to infants with hypothermia for whom there is a clinical suspicion of IBI, not all infants with hypothermia.

What are other challenges unique to deriving a prediction model for infants with hypothermia? First, there are a few ways in which infants with hypothermia are different from febrile infants. As highlighted in the current studies, infants with hypothermia are on average younger (median of 5 days old in Westphal’s study) and more likely to have temperature instability.⁷  In an otherwise well-appearing infant with hypothermia, the low temperature may be just that: A lack of maturity in thermoregulation. However, during a period of time in which the prevalence of SBI and IBI are higher overall, how do clinicians tell the difference between immature thermoregulation and sepsis? Although Money et al identified several clinical and laboratory predictors, and previous studies have identified lower median temperatures, ill-appearance, elevated bands, lower platelets, complex chronic conditions, and repeated temperature instability as risk factors,^8–13  more investigation is clearly indicated.

The test characteristics of risk stratification tools in febrile infants have improved because of the emergence of biomarkers such as C-reactive protein, and in particular, procalcitonin, a key component of the PECARN prediction rule.^1,6  However, a challenge to retrospective research on infants with hypothermia is that procalcitonin was not widely available until recent years and its use was variable at sites, thus making it difficult to include in prediction models from a statistical power perspective. Additionally, as Westphal et al identify in their discussion, 1 of the difficulties in assessing biomarkers in infants with hypothermia is our lack of understanding of the host immune response of hypothermia in sepsis. One theory states that hypothermia in sepsis is a result of an altered immune response and that, consequently, inflammatory biomarkers may not be an accurate predictor.¹⁴  Another hypothesis is that hypothermia occurs in sepsis to conserve energy and protect vital organs.¹⁵  As pediatricians, we also lack consensus on what is considered hypothermia; some studies use the World Health Organization’s cutoff of <36.5°C, whereas others use the International Pediatric Sepsis Consensus Conference cutoff of <36°C.¹⁶  One study even attempted to statistically derive a hypothermia threshold, but found that any attempt to do so lacked sensitivity and specificity.¹⁷ 

With these challenges, the current lack of a validated risk-stratification algorithm for infants with hypothermia presents clinicians with a difficult dichotomy in clinical practice. The potential danger in not completing a septic workup in an infant with hypothermia includes underdiagnosing SBI or IBI. Studies so far estimate that anywhere from 2.2% to 8% of infants with hypothermia presenting to the emergency department have an SBI.^9–13  There is also evidence to show that, of those infants with IBI who initially presented with hypothermia, there is a higher mortality rate of up to 13% and that hypothermia is an independent risk factor for mortality in sepsis.^18,19  This indicates that hypothermia may be a sign of a poor outcome in cases of SBI or IBI. This exists in contrast to the fact that most infants with hypothermia are healthy. One study in the outpatient setting evaluated infants with incidentally found hypothermia and 0 of 212 infants had SBI.²⁰  This presents a true conundrum: Hypothermia is usually a benign finding, but when in the setting of SBI, can be a sign of very severe infection and poor prognosis. These are 2 drastically different ends of the spectrum. It would be unethical to work up every infant with hypothermia because of the potential consequences of overtreatment, but we need to elucidate how to pick out those who are ill.

So where does this leave us? This Hospital Pediatrics issue gives us 2 studies that provide evidence to help clinicians faced with this conundrum. However, in the end, the main conclusion of both studies is that we need further investigation. A few points are clear: Infants with hypothermia are a unique population, with their own set of risk factors, for which there is currently tremendous variability in provider management. A large, multicenter study with an adequate number of infants with IBI, and with a large sample of infants in whom a procalcitonin was obtained, is necessary. Although the myriad challenges to deriving a prediction model in this population exist, particularly around spectrum bias, it is clearly time for a robust investigation to derive and validate a prediction model for infants with hypothermia.