Decision-tree derivation and external validation of a new clinical decision rule (DISCERN-FN) to predict the risk of severe infection during febrile neutropenia in children treated for cancer

Background: In 2017, international guidelines proposed new management of febrile neutropenia in children with cancer, adapted to the risk of severe infection by clinical decision rules (CDRs). Until now, none of the proposed CDRs has performed well enough in high-income countries for use in clinical practice. Our study aimed to build and validate a new CDR (DISCERN-FN) to predict the risk of severe infection in children with febrile neutropenia. Methods: We did two prospective studies. First, a prospective derivation study included all episodes of febrile neutropenia in children (aged <18 years) with a cancer diagnosis and receiving treatment for it who were admitted for an episode of febrile neutropenia, excluding patients already treated with antibiotics for this episode, febrile neutropenia not induced by chemotherapy, those receiving palliative care, and those with a stem cell allograft for less than 1 year, from April 1, 2007, to Dec 31, 2011 from two paediatric cancer centres in France. We collected the children's medical history, and clinical and laboratory data, and analysed their associations with severe infection. Sipina software was used to derive the CDR as a decision tree. Second, a prospective, national, external validation study was done in 23 centres from Jan 1, 2012, to May 31, 2016. The primary outcome was severe infection, defined by bacteraemia, a positive bacterial culture from a usually sterile site, a local infection with a high potential for extension, or an invasive fungal infection. The CDR was applied a posteriori to all episodes to evaluate its sensitivity, specificity, and negative likelihood ratio. Findings: The derivation set included 539 febrile neutropenia episodes (270 episodes in patients with blood cancer [median age 7·5 years, IQR 3·7–11·2; 158 (59 %) boys and 112 (41%) girls] and 269 in patients with solid tumours [median age 6·6 years, IQR 2·9–14·2; 140 (52 %) boys and 129 (48%) girls]). Significant variables introduced into the decision tree were cancer type (solid tumour vs blood cancer), age, high-risk chemotherapy, level of fever, C-reactive protein concentration (at 24–48 h after admission), and leucocyte and platelet counts and procalcitonin (at admission and at 24–48 h after admission). For the derivation set, the CDR sensitivity was 98% (95% CI 93–100), its specificity 56% (51–61), and the negative likelihood ratio 0·04 (0·01–0·15). 1806 febrile neutropenia episodes were analysed in the validation set (mean age 8·1 years [SD 4·8], 1014 (56%) boys and 792 (44%) girls), of which 332 (18%, 95% CI 17–20) were linked with severe infection. For the validation set, the CDR had a sensitivity of 95% (95% CI 91–97), a specificity of 38% (36–41), and a negative likelihood ratio of 0·13 (0·08–0·21). Our CDR reduced the risk of severe infection to a post-test probability of 0·8% (95% CI 0·2–2·9) in the derivation set and 2·4% (1·5–3·9) in the validation set. The validation study is registered at ClinicalTrials.gov, NCT03434795. Interpretation: The use of our CDR substantially reduced the risk of severe infection after testing in both the derivation and validation groups, which suggests that this CDR would improve clinical practice enough to be introduced in appropriate settings. Funding: Ligue Nationale Contre le Cancer.