A Prediction Model to Identify Febrile Infants ≤60 Days at Low Risk of Invasive Bacterial Infection

Abstract

Objectives: To derive and internally validate a prediction model for the identification of febrile infants ≤60 days old at low probability of invasive bacterial infection (IBI).

Methods: We conducted a case-control study of febrile infants ≤60 days old who presented to the emergency departments of 11 hospitals between July 1, 2011 and June 30, 2016. Infants with IBI, defined by growth of a pathogen in blood (bacteremia) and/or cerebrospinal fluid (bacterial meningitis), were matched by hospital and date of visit to 2 control patients without IBI. Ill-appearing infants and those with complex chronic conditions were excluded. Predictors of IBI were identified with multiple logistic regression and internally validated with 10-fold cross-validation, and an IBI score was calculated.

Results: We included 181 infants with IBI (155 [85.6%] with bacteremia without meningitis and 26 [14.4%] with bacterial meningitis) and 362 control patients. Twenty-three infants with IBI (12.7%) and 138 control patients (38.1%) had fever by history only. Four predictors of IBI were identified (area under the curve 0.83 [95% confidence interval (CI): 0.79-0.86]) and incorporated into an IBI score: age <21 days (1 point), highest temperature recorded in the emergency department 38.0-38.4°C (2 points) or ≥38.5°C (4 points), absolute neutrophil count ≥5185 cells per μL (2 points), and abnormal urinalysis results (3 points). The sensitivity and specificity of a score ≥2 were 98.8% (95% CI: 95.7%-99.9%) and 31.3% (95% CI: 26.3%-36.6%), respectively. All 26 infants with meningitis had scores ≥2.

Conclusions: Infants ≤60 days old with fever by history only, a normal urinalysis result, and an absolute neutrophil count <5185 cells per μL have a low probability of IBI.

FIGURE 1

Study sample.

FIGURE 2

Calibration plot for the IBI prediction model with the estimated probability of IBI (x-axis) plotted against the outcome (presence of IBI). The calibration curve (solid line), with a 95% confidence band, overlaps with the diagonal reference line of perfect calibration (hashed line), indicating that the prediction model fit the observed data well.