This was a prospective study conducted at the Pediatric Inpatient Unit of the Hospital das Clínicas of the Federal University of Minas Gerais (UFMG), Belo Horizonte, Minas Gerais, Brazil. The Ethical Committee approved the study, which was developed from June 2016 to June 2018.

We enrolled children and adolescents under 18 years old diagnosed with cancer and FN. FN was defined as an axillary temperature > 38°C in a single measurement or > 37.8°C for at least one hour or two measurements within 24 hours and neutrophil count < 500 cells/mm³ or between 500 - 1000 cells/mm³, with a predicted decline. Patients were monitored from the time of the FN diagnosis until the termination of the antibiotics use. Eligible patients were identified by a daily blood count test and the presence of fever in medical records. Patients with hematopoietic cell transplants were excluded because they had been admitted in a different section of the hospital. All patients who agreed to participate in the study previously signed an Informed Consent Form (ICF).

Demographic, clinical and laboratory data of the patients at the onset of FN were collected: age, sex, type of neoplasia, absolute neutrophil count, absolute monocyte count, platelet count, CRP, hemoglobin, first episode of FN, place where first fever occurred (at home or in the hospital), first positive blood culture test and presence of central catheter. The first blood culture was collected for screening of infection at the beginning of the episode of febrile neutropenia.

The outcome variable (dependent) was defined as the documented infection.

All patients were diagnosed and treated according to the established protocol at the Pediatric Inpatient Unit, which includes, anamnesis and comprehensive physical exams, propaedeutic investigation with blood count, CRP, blood culture, routine urine analysis and urine culture. Empiric antimicrobial therapy was conducted with fourth generation cephalosporin.

We used the standard infection criteria of the National Healthcare Safety Network (NHSN) of the Centers for Disease Control and Prevention (CDC)14 for bacterial infections and the European Organization for Research and Treatment of Cancer/Mycoses Study Group (EORTC / MSG)15 for invasive fungal infections. Thus, infections were divided into 8 groups: fever with no determined infectious focus; bloodstream infection (laboratory-confirmed bloodstream infection (LCBI) and mucosal barrier injury laboratory-confirmed bloodstream infection (MBI-LCBI)); Clostridioides difficile infection; neutropenic enterocolitis; invasive fungal infection (proved, probable a possible invasive fungal infection); skin and soft tissue infection; lower respiratory tract infections, and; genitourinary tract infections. The signs and symptoms of upper respiratory infection were considered as other infections.

The microbiological isolation was performed by an automated method (VITEK2) and the resistance profile was determined by the disk diffusion test (Kirby Bauer). The sensitivity profile of microorganisms was established by the Hospital Infection Control Commission (CCIH), based on the Clinical and Laboratory Standards Institute (CLSI).16

The mean, median, quartiles and standard deviation were obtained through descriptive analysis. The variables of interest were described by a 95% confidence interval (95%CI). The association between two categorical variables was performed using the Pearson's chi-square test and Fisher's exact test. For significant differences (p < 0.05) between groups, the degree of association between two categorical variables was calculated by the Odds Ratio. The Student's t-test compared the means between children with and without infection. Homogeneity of variance was verified by the Levene test. The Logistic Regression model identified variables that jointly influenced the occurrence of documented infections and the Hosmer and Lemeshow fit evaluated the fit of the model to the survey data. The Cox & Snell and Nagelkerke pseudo-R² measurements verified if the variables were sufficient to explain the outcome. The categorical variable with more than two categories (degree of neutropenia > 500, between 500 and 100 and < 100) was dichotomized and transformed into a Dummy variable. All results were considered significant for a significance probability of less than 5% (p < 0.05).

The 172 episodes of FN in 77 patients were evaluated, with an average of 1.64 episodes per patient. The mean age of the children at the onset of an episode was 7.3 years and 54.7% of the FN episodes were in male patients. The distribution of the underlying diseases was as follows: 68.6% of the population was diagnosed with leukemia and lymphoma and 31.4% had a solid tumor. Patients with tumor relapse corresponded to 21.5% of the study population.

The mean absolute neutrophil count at the beginning of the episode was 342.3 cells/mm³ and 61% of the cases had severe neutropenia (< 100 cells/mm³). The mean absolute monocyte count at the beginning of the episode was 137.1 cells/mm³ and 74.4% presented a monocyte count less than < 100 cells/mm³. The mean CRP was 68 mg/L and 79.9% of the cases presented CRP > 90 mg/L in the first measurement.

Of all the cases, 19% were the first episode of FN induced by chemotherapy and 53.2% experienced the fever onset at home. The average duration of the fever, hospitalization and antibiotic therapy were 13.1, 41.7 and 19.7 days, respectively. A total of 31 (18%) patients had septic shock and 21 (12.2%) were transferred to the Intensive Care Unit. The mortality rate was 4.7% (8 patients). The clinical and demographic data are summarized in Table 1.

The microorganisms were isolated from 82 blood cultures; gram-positive bacteria were isolated from 41.2% of positive blood cultures, gram-negative were isolated in 49.6% and fungi, in 7.2% of the cultures. The most prevalent bacteria were Staphylococcus epidermidis (13.4% of the positive blood cultures), followed by Pseudomonas aeruginosa (12.1%) and Klebsiella pneumoniae (10.9%). The candida species was the only species of fungi isolated from the blood. The most common pathogen from the urine cultures was Klebsiella pneumoniae (4 urine samples). The pathogens isolated from the skin cultures were Acinetobacter baumannii (1 case) and Fusarium sp. (3 cases). Resistant pathogens corresponded to 30.9% of the analyzed sterile-site cultures.

Among the episodes of FN, 56 (32.5%) presented resistant bacteria colonization and/or infection. In 14 cases (25%), resistant enterobacteria were identified, 14 (25%) were extended-spectrum beta-lactamase-producing bacteria (ESBL), 11 (19.6%) were multiresistant Acinetobacter sp., 9 (16%) were Enterococcus sp. Vancomycin-resistant (VRE), 7 were (12.5%) multidrug-resistant Pseudomonas sp. and 1 case (1.8%) was Stenotrophomonas maltophilia.

Fever with unidentified infectious focus was the most frequent clinical diagnosis, corresponding to 58.3% of the febrile neutropenia episodes. A bloodstream infection occurred in 15.6% of the episodes; 22 (10.4%) of the bloodstream infections were related to a mucosal barrier breakdown. There were episodes with multiple confirmed infections.

The remaining 5 possible types of infections accounted for less than 10% of the cases and we considered the signs and symptoms of upper respiratory infections as other infections, accounting for 6.1% of the febrile neutropenia episodes (Table 2).

From univariate analysis, the risk factors presenting a statistically significant association (p < 0.05) with an infection outcome were: female, absolute monocyte count < 100 cell/mm³ at the onset of the episode, platelet count < 50,000 at the onset of the episode, CRP > 90 mg/dl at the onset of the episode, hemoglobin < 7 mg/dl at the onset of the episode, two or more episodes of febrile neutropenia, a fever onset during hospitalization and a positive first blood culture (Table 3).

The variables that remained in the multivariate model associated with infection were: a platelet count at the onset of episode ≤ 50,000 (2.32 odds ratio (OR), 95%CI 1.11; 4.76, p = 0.021), CRP > 90 mg/dL (3.18 OR, 95%CI 1.32; 7.68, p = 0.01), fever onset with hospitalized patient (5.88 OR, 95%CI 2.75; 12.57, p < 0.001), first episode of febrile neutropenia (0.34 OR, 95%CI 0.13; 0.92, p = 0.034), and first blood culture with a negative result (0.27 OR, 95%CI 0.07; 0.84, p = 0.025) (Table 4).

The model accuracy showed an 84.4% specificity, a 63.2% sensitivity, 25% false positives and 24.3% false negatives. The parameters that evaluated the fit of the logistic model show pseudo-R² values close to, or below, 0.30 (Cox & Snell pseudo-R² = 0.239 and Nagelkerke pseudo-R² = 0.321).