Individual participant data validation of the PICNICC prediction model for febrile neutropenia

What is it about?

Risk-stratified approaches to treating cancer, giving more therapy to more advanced stages and less to more localised disease is common; this is also possible with complications of therapy, such as febrile neutropenia. Febrile neutropenia is a common complication of cancer treatments and children and young people which may be the first hint of a life threatening infection. To manage this best, allowing the shortest possible hospitalisation times, we need a good way of predicting the chance of poor outcomes. There have been many proposed models to do this, but few of them have been tested widely and shown to be effective (validated). In this study, an individual participant data (IPD) meta-analytic validation of the “Predicting Infectious ComplicatioNs In Children with Cancer” (PICNICC) prediction model for microbiologically documented infection was undertaken. Pooled estimates of effectiveness were produced. The included the area under the receiver operator characteristic curve (AUCROC), which is interpreted as ineffective at 0.5, and perfect at 1.0. The PICNICC model was poorly predictive, with AUCROC 0.59, 95% CI 0.41 to 0.78, compared with the value from the original paper of 0.72, 95% CI 0.71 to 0.76. Three different simple methods to improve these numbers (recalibration) failed to improve performance meaningfully. This meta-analysis shows the PICNICC model should not be used at admission of children with FN and further work should focus on validating alternative prediction models. This type of validation assessment across collections of data from diverse locations is essential before these type of rules are clinically used to avoid treatment decisions being wrong.

Featured Image