by John Gever

Up to one-third of children with febrile bacterial infections failed to receive antibiotics in a large emergency department, while 20% of youngsters without such infections received antibiotics unnecessarily, Australian researchers said.

A computer algorithm based on standard test results could do a better job of assigning feverish children to the correct treatment sooner, Jonathan C. Craig, MBChB, PhD, of the University of Sydney, and colleagues reported online in BMJ.

Although 85% to 95% of emergency room physicians estimated that 5% of febrile children would have serious bacterial infections, case data indicated that 7.2% had pneumonia, bacteremia, or a urinary tract infection, the researchers found.

“Emergency department physicians tend to underestimate the likelihood of serious bacterial infection in young children with fever, leading to undertreatment with antibiotics,” the authors concluded on the basis of their two-year prospective study.

The researchers analyzed all cases of children younger than 5 presenting with fever from 2004 to 2006 in the emergency department at the Children’s Hospital of Westmead, in suburban Sydney. That meant a study population of nearly 16,000 patients, for whom complete follow-up was available on about 14,400.

Craig and colleagues wrote that 40 clinical signs and symptoms were entered into the hospital’s electronic records database for all patients. The system also required physicians to estimate the likelihood that the patient had any of 10 potential diagnoses. They were free to use their clinical judgment in making diagnoses and setting treatment.

Some 1,054 children were determined to have serious bacterial infections on the basis of imaging and/or culture results, with some children having more than one. All but 26 infections were in the urinary tract or bloodstream or were pneumonia.

“Almost all had the relevant reference standard test performed during their emergency department assessment,” the researchers wrote.

But antibiotics were not prescribed for significant numbers of patients with the serious bacterial infections: 33% of urinary tract infection and pneumonia cases and 19% of bacteremia cases.

Follow-up data indicated that most of these children did eventually receive antibiotics. Of 363 cases of infection that were initially untreated, only eight were recorded as still unwell after a mean of 10.2 days of follow-up, and none still had fever.

Meanwhile, 20% of children without an identified bacterial infection — 2,686 out of 13,557 — received antibiotics, the researchers found.

Statistical work to correlate the initial clinical and laboratory findings with ultimate diagnoses yielded a model based on 26 parameters that predicted diagnoses with better accuracy than the Westmead emergency physicians.

The parameters ranged from patients’ general appearance at presentation to capillary refill time.

Plotting the model’s specificity versus sensitivity yielded receiver-operating characteristic curves for the three major types of infection, with areas of 0.74 to 0.84. (A value of 1 means that both specificity and sensitivity are 100%.)

For example, the model’s diagnosis of pneumonia had a sensitivity of 80% when the specificity was 70%.

Craig and colleagues said the physicians’ early estimates of likely diagnoses tended to emphasize specificity over sensitivity. Hence, the specificities of their estimates were in the range of 90% to 100%, but the sensitivities were only 10% to 50%.

The researchers emphasized that there are often sound clinical reasons for withholding immediate antibiotic treatment when bacterial infections are suspected.

For example, diagnosis of urinary tract infection requires an uncontaminated specimen and clinicians often delay antibiotics until they see culture results. “Within an emergency department setting, this approach is not unreasonable,” the researchers wrote.

Yet the delays seen in the study may have extended the duration of febrile illness and contributed to return hospital visits, Craig and colleagues argued.

They recommended two interventions to improve clinical decision-making for feverish children: using a computer algorithm like the one they devised to aid diagnosis, and more standardized, evidence-based interpretation of urinalysis and chest radiograph results.

An accompanying editorial by two researchers from the University of Oxford in England, however, suggested it would be premature to adopt the recommendations without more research.

“Before widespread implementation, we will need to have evidence showing the effect of using such a model on patient management and outcomes,” wrote Matthew J. Thompson, MBChB, and Ann Van den Bruel, MD, PhD.

The editorialists also said the diagnostic algorithm would be more useful if it also included infections such as meningitis, osteomyelitis, and septic arthritis, which were seen in small numbers in the study.

“These are rare but crucial to identify, so excluding them may limit the model’s usefulness,” Thompson and Van den Bruel wrote.

Craig and colleagues cited other limitations to their analysis, including lack of microbiological and/or radiological verification in some children, and the uncertain validity of the initial physician estimates of disease in the emergency department.

John Gever is a MedPage Today Senior Editor.

Originally published in MedPage Today. Visit MedPageToday.com for more infectious disease news.