Results
A total of 35 patients presenting with FN and culture confirmed bacteremia, as well as, 35 matched controls (sex, age, category of malignancy and degree of neutropenia) with negative blood cultures were included in the study (70 total presentations of FN). Controls did not demonstrate diagnostic tests concerning for either a non-hematogenous bacterial infection or fungal infection. Nine of the 35 controls (26%) had a positive respiratory viral panel which was considered to be the cause of their fever, while the remaining 26 (74%) were considered to have FN of an unknown cause. All controls experienced spontaneous resolution of symptoms with symptomatic treatment and were discharged home without complications (development of healthcare-associated infection, sepsis, need for ICU admission, or death).
Patient demographic data and the frequency of presenting signs/symptoms often used for FN risk stratification are included in Table 2. Overall age ranged between 1.4 and 18.8 years old. Average age was 10.7 (± 6.3) and 10.0 (± 5.9) years for the cohort and control groups respectively. Twenty patients were female (57%) and 34 (97%) had presenting ANC’s < 500 cells/µL. Absolute monocyte count (AMC) was < 100 cells/µL in 24 of the controls (69%) and 22 (63%) of the cohort patients. Twenty-three matched cohort-control patient pairs (66%) were categorized as having Leukemia/Lymphoma, 8 (23%) as having a solid tumor and 4 (11%) as being status-post HSCT.
There was no statistically significant difference in peak temperature (Tmax), pre-acetaminophen temperature, or the total number of days of fever between the two groups (p = 0.080, 0.71 and 0.20 respectively). Temperatures 1-hour after acetaminophen administration were significantly different between the cohort and controls (p = 0.040). Patients in the cohort were found to have a higher mean temperature 1-hr after acetaminophen administration (101.2°F/38.4°C vs. 100.0°F/37.8°C for the controls). Additionally, presence of a pre-acetaminophen temperature ≥ 39°C was significantly more common among culture positive patients (p = 0.044). Despite these distinctions, no significant relationship was seen between the presence of a fever (temperature ≥ 100.4°F/38.0°C) 1-hour after acetaminophen and bacteremia (p = 0.16). Temperatures 2- and 6-hours after acetaminophen administration were not significantly different between the groups (p = 0.35 and 0.15 respectively). Changes in patient temperatures between the pre-acetaminophen,1-, 2- and 6-hour time points were also not significantly different between the groups. Mean mg/kg dosing of acetaminophen was not significantly different between groups (p = 0.35). Details regarding average values and significance of different temperature endpoints is detailed in Table 3.
Data regarding the timing of antibiotic administration was collected for both groups to account for possible impacts of early antibiotic administration on temperature among patients with bacteremia. Analysis showed no difference between the groups in duration of antibiotic coverage (hours) at the 1-hr post-acetaminophen time point (p = 0.14). Additionally, linear regression analysis demonstrated that duration of antibiotic coverage (hours) at the 1-hr post-acetaminophen time point was not predictive of patient temperatures among those patients with positive blood cultures (p = 0.22). Two patients in the study cohort were on prophylactic levofloxacin at time of presentation; however, administration of prophylactic levofloxacin was not found to differ significantly between the groups (p = 0.49).
Given the significant relationship seen between patients with culture confirmed bacteremia and temperatures 1-hour after initial acetaminophen administration, pre-acetaminophen temperature ≥ 39°C, evidence of focal infection and hypotension, a stepwise binary logistic regression was performed to assess the ability of these variables to predict bacteremia in this patient population. Regression modeling demonstrated that temperature 1-hour after acetaminophen was the only variable with significant predictive value for bacteremia (p = 0.011). A receiver operating characteristic (ROC) curve was generated using this regression model (Figure 1) and demonstrated an area under the cure (AUC) of 0.70.
Characterization and regression tree (CART) analysis was also performed to construct an additional model and identify optimal breakpoints in temperature 1-hr after acetaminophen administration. Modeling was again performed with inclusion of pre-acetaminophen temperature ≥ 39°C, evidence of focal infection and hypotension variables. An optimal decision tree for determination of bacteremia was constructed and is detailed in Figure 2. Sensitivity and specificity for the CART model were 62.9% and 74.3% respectively. The AUC-ROC of the CART model for the test arm was 0.71 (Figure 3). Examination of the CART decision tree breakpoints demonstrated that 12 of 14 patients (85.7%) with temperature ≤ 99.2°F/37.3°C 1-hr after acetaminophen were blood culture negative. Likewise, 4 of 5 patients (80%) with temperatures > 102.5°F/39.2°C 1-hr after acetaminophen were culture positive. All patients with hypotension or evidence of focal infection displayed positive blood culture results. Relative variable importance (defined as % improvement with respect to the CART model’s top predictor) demonstrated temperature 1-hr after acetaminophen as the models most important variable for predicting bacteremia (Figure 4).