Structured clinical question

In children under three with a temperature >38.5°C and no clinical focus of infection [subject], is measurement of procalcitonin [intervention] a good screening test to exclude serious bacterial illness [outcome]?

Search strategy and outcome

Details of papers examined are given in table 1⇓. Please see http://www.bestbets.org/cgi-bin/bets.pl?record=00597 for further information.

The primary source was Medline using PubMed: ((“procalcitonin”[Substance Name] OR procalcitonin[Text Word]) AND (“infant”[MeSH Terms:noexp] OR “child, preschool”[MeSH Terms])) OR ((“procalcitonin”[Substance Name] OR procalcitonin[Text Word]) AND (“Child*”)). Outcome: 127 articles (last check 13th May 2006) were found. Five studies which addressed the clinical scenario were selected. One narrative review article was found which identified four of these five papers (one was published after the review).¹ Several other papers using procalcitonin for specific presentations with a fever (rash, diarrhoea) were not included.

Secondary search methods: Search replicated on Cochrane library, EMBASE, CINAHL, BestBETs, CatCrawler. Article reference lists were reviewed, and manufacturer’s website and clinical trial registers searched. No additional relevant articles were found. One systematic review on procalcitonin as a marker for bacterial illness was found but it did not focus on children under age three.²

Commentary

Estimating pretest probability and then applying the likelihood ratios to determine post-test risk is the optimal method to utilise a test. Treatment decisions are then based on post-test risk and how serious the outcome would be if the illness were left untreated. In the post Haemophilis influenzae b era, Streptococcus pneumoniae and Neiserria meningiditis (in its epidemic forms) are the main pathogens with rapid and sometimes serious effects. The incidence of serious bacterial illness depends on the definition used, the height of the fever, the source population, local organisms and immunisation practices, and most importantly the age of the child. In the five studies reviewed, the prevalence of serious bacterial illness ranged from 11% all the way up to 43% during a meningococcal epidemic. There are hundreds of studies demonstrating a much lower incidence in many settings.

The commonest occult source in these studies was the urinary tract and two of these studies used dimercaptosuccinic acid (DMSA) to differentiate pyelonephritis from simple lower urinary tract infection (UTI). Screening urinalysis to exclude UTI in young children is therefore recommended. Pneumonia is rarely present without clinical symptoms or signs but these may be vague, for example abdominal pain and vomiting. One study defined consolidation as a local infection,³ while another classified it as an invasive infection.⁴ Meningitis will be present in a few and, if suspected, cerebrospinal fluid (CSF) examination is required and often recommended in the first 6 weeks of life where the signs of meningitis can be difficult to detect. Skin, bone and other infections may sometimes be present and a careful examination is required. A small number of remaining children will have occult bacteraemia. In the five studies examined, occult bacteraemia was diagnosed in 9.4%, 3.2%, 12.5%, 4% and 5.8% of patients. These are the cases likely to be missed after clinical examination. In the post Hib vaccination era, the risk of bacteraemia may be less than 2%⁵ and may be reduced further by pneumococcal vaccination⁶ and targeted meningococcal vaccine programs.⁷ Our local incidence of serious bacterial infection is low and in the clinical scenario above in a fully vaccinated child with a clear urine and no respiratory symptoms, I would estimate the pretest probability to be around 1% in our population.

The fear of missing a case has lead to widespread use of antibiotics and admissions to hospital, although many children with occult bacteraemia will recover without treatment.⁸ A diagnostic test to separate the serious bacterial illnesses from the more common virus has long been sought (whether it be a clinical scoring system, a single test or a group of tests). C-reactive protein (CRP) had been the best test to date but may not be elevated in the first 24 h of a serious bacterial illness but is raised in many viral illnesses. Many of these studies included CRP, but this was not the focus of this critical appraisal. Procalcitonin is the latest candidate that has shown the most promise in these papers. No one previous screening tool has reached the sensitivity of procalcitonin, although combinations of clinical scoring tests and laboratory tests may.⁹ Unfortunately, the lack of a true gold-standard for ”serious bacterial illness” has created heterogeneity in these studies which makes meta-analysis difficult, but attempts have been made in other settings.²

Traditionally, when taking a continuous measure (such as a serum concentration) physicians have preferred a single cut-off value to determine the presence or absence of disease. Sophisticated techniques have been developed to select that cut-off point (receiver operator characteristic curves). There is no reason more than one level cannot be chosen. For example, procalcitonin levels could be divided into less than 0.5, 0.5–2 and more than 2. A procalcitonin level greater than 2 has a positive likelihood ratio of at least 3.5 and increases pretest probability more than 3.5-fold. Higher procalcitonin levels (>5 or >10) would have even higher positive likelihood ratios, but these are rarely calculated.

As a “rule-out” test, procalcitonin is currently limited by the laboratory tests available.¹⁰ Normal procalcitonin is likely to be around 0.05 ng/ml in healthy individuals. Current assays have a functional detection range down to 0.5 ng/ml. Having the ability to set lower cut-off points in the future could mean procalcitonin may have even better sensitivity.

A procalcitonin level of around 0.5–1.0 had a negative likelihood ratio (95% CI) in the five studies reviewed being 0.05 (0.02 to 0.20), 0.09 (0.02 to 0.35), 0.09 (0.06 to 0.15), 0.09 (0.02 to 0.36) and 0.25 (0.04 to 1.59). This allows a practitioner to reduce the pretest probability by between 4- and 20-fold. If your pretest probability were 10%, then with a procalcitonin level of less than 0.5 the post-test risk of serious bacterial illness would be between 0.5% and 2.5%. Somewhere between 40 and 200 children would need to be treated to ensure you were treating the one child with a serious bacterial illness.

Dr Herd is supported by the Joan Mary Reynolds fellowship.