As a junior doctor some years back, more than I can believe now, I remember the anxiety and anticipation I felt as I contacted a family who had taken their toddler home from accident and emergency after reassurance. A blood culture had then rapidly grown a meningococcus. The child was fine after readmission and treatment. It is a singular concern for any doctor dealing with a child presenting acutely with possible infection that they might miss severe bacterial sepsis. Those of us who have worked in paediatric intensive care will all have felt a shudder hearing stories from parents of children with serious illness that was not recognised. We may have looked at our own sick children at 3:00 am and remembered a parent's chilling phrase or description. Last winter H1N1 flu again reminded us just how hard it can be to get a diagnosis right.

The failure to recognise or a delay in recognising serious illness was one of the themes identified in ‘Why children die: a pilot study’, after the national confidential enquiry in 2008.1 The need to develop robust systems for identifying the seriously ill child was emphasised. We all understand that clinical experience has an important role to play.2,–,4 Experience, however, takes a long time and a good deal of work to acquire and is not infallible: ars longa, vita brevis (the art is long, life is brief). As new tools have arisen, there has been interest in integrating both clinical and laboratory findings into tools that can help the doctor recognise illness and treat or reassure families: in hoc signo vinces (with this sign you will conquer).

In this issue of the journal, Galetto-Lacour et al5 (see page 968) present a large study to validate a risk laboratory index score based on combining specific clinical findings and measurements of C reactive protein (CRP), white cell count and procalcitonin to help physicians identify severe bacterial infection in children. Members of the group have been admirably committed over several years to studying acute infectious diseases in children and how to identify them early. The scoring system suggested raises the sensitivity of detection further to 86%.

Editorial on the paper by Galetto-Lacour, et al (see p 968)

Procalcitonin was first described in 1975 as a precursor in calcitonin biosynthesis5 and later found to be quickly elevated in sepsis.6 Over the last 15 years there has been persistent interest in its possible utility in the early detection of serious infection. It appears to be produced in response to tumour necrosis factor, interleukin 6 (IL-6) and bacterial endotoxin. We still have a poor understanding of what its role in sepsis might be. The assay to detect procalcitonin is not widely available to clinicians at present. The interest in procalcitonin follows that in CRP. First reported in association with acute infection in 1941,7 CRP is usually absent from the blood and is produced by hepatocytes rising rapidly within a few hours in response to IL-6. The assay for CRP became more widely available to hospital clinicians in the UK in the 1980s and is now a regular on biochemical profiles. It is now clear that there is genetic variability in the rise in CRP8 and that as part of innate immunity CRP plays a role in susceptibility to infection.9 Even before we knew about CRP, at the end of the 19th century, Cabot reported that the white cell count was often raised in the peripheral blood in patients with infection.10 The white cell count is included in the scoring system presented by Galetto-Lacour et al, showing just how important the original observation turned out to be, but alone in this study population it had close to only heads or tails sensitivity. It is clear that to date no single test has sufficient sensitivity and specificity to satisfy the worried paediatrician or emergency physician.

Galetto-Lacour et al have carefully described the population of children who presented to the Children's Hospital in Padua (Padova) and were used to validate the system. Severe sepsis in children in the developed world is thankfully relatively rare. In the study population of about 45 000 children, 92 were found to have severe bacterial infection and 13 children with severe bacterial infection would not have been identified by the test. The authors found that if antibiotics were prescribed on the basis of the scoring system, only 33% would have been treated compared with 67% being treated according to the clinician's decisions, achieving cost savings and reducing overall antibiotic prescribing.

There are, of course, potential pitfalls for clinicians incorporating new scoring systems11 into their clinical practice. Variation in the prevalence of the disease in different populations can have a significant effect on the rate of false negatives and positives. For conditions with low population prevalence, even with high test specificity the positive predictive value of the result will be reduced and the use of combined laboratory scores can be undermined where the variables being simultaneously tested are not independent of one another and vary from individual to individual. Where immunological markers are tested in infection, it must also be remembered that there is variation in the immune response in the normal population affecting responses in innate immunity and mucosal immunity as well as cellular and humoral responses.12,–,14 Therefore, test thresholds can be tricky to decide: biological readouts such as the secretion of inflammatory markers are usually continuous variables and what cut-off determines positive and negative is difficult. If a value is selected to increase the sensitivity of the test, it will usually reduce the specificity and vice versa. In the clinical situation where it needs to be determined whether a child is at risk of severe bacterial infection and the result affects treatment, a high sensitivity might be desirable so that no child goes undetected. The effect of increasing the sensitivity will tend to mean the test performs well as a negative predictor. In the clinical setting where there is a need to detect rather than rule out severe sepsis, the specificity becomes much more important.

Complex clinical algorithms have been tested for more than 20 years and although the authors are to be congratulated on their work, a truly reliable method of quickly detecting the seriously ill febrile child has not been found. In children it is still frequently necessary to use the options of observation in hospital, an experienced team and, sometimes, of giving empirical antibiotics without hard evidence. The old school approach clashes with modern 4 h targets and the drive to reduce expensive inpatient admissions, but it is still a sound approach if people are to trust us with their children.