Review question
To systemically review and critically appraise existing risk stratification models for the prediction of the outcome of febrile neutropenic episodes in children/young adults with cancer.

Search strategy
Relevant electronic sources will be searched for clinically relevant studies which develop or validate a clinical prediction model. Reference lists of relevant systematic reviews and included articles will also be reviewed. Published and unpublished studies will be sought and no language restrictions applied. Non-English language studies will be translated if time permits. Two reviewers will independently screen the title and abstract of studies for inclusion. Disagreements will be resolved by consensus, or if this proves impossible, by recourse to an independent adjudicator.

Types of study to be included
Studies will be included in the review if they meet the following criteria.

Population: Children or young people (aged 0 – 18y) who are receiving treatment for cancer or leukaemia (including extra-cranial and intra-cranial tumours) presenting with febrile neutropenia. Studies with adults which report data for patients <18y will be included if outcome data are reported separately.

Predictor variables: Clinical decision rules (CDR) using clinical and haematological or biochemical variables used to predict outcome for the particular episode of febrile neutropenia (e.g. age, sex, type of cancer, type or intensity of treatment, duration since treatment, presence or absence of a central venous catheter, clinical appearance, peak temperature, blood pressure, heart rate, admission blood count).
Alternative variables may also be found and included.

Outcomes (At least one of): Survival, need for intensive care, need for high-dependency care, single organ impairment (oxygen requirement, renal impairment, hepatic impairment), invasive bacterial or fungal infection, any documented infection (including radiologically confirmed pneumonia), duration of admission.

Study design: Both prospective and retrospective studies which derive or validate a clinical decision model to predict any of the above outcomes.

Strategy for data synthesis
Key study characteristics, the clinical decision rules, methods use to derive the rules, predictive ability of the models studied and study quality will be summarised in narrative and tabular forms.

If it is considered appropriate (based on clinical and statistical homogeneity and the necessary data being available) appropriate meta-analytical methods will be used to estimate the accuracy of the studied CDR, although this is unlikely. These may include the use of hierarchical summary receiver operator characteristic (HSROC) curve modelling with binary data. Other forms of data are unlikely to be present.

Heterogeneity will be explored through consideration of study populations (both the geographical location and nature of the diseases under study), study design, predictor variables assessed and outcomes chosen. If data pooling is considered appropriate then heterogeneity will be quantified using the I2 statistic.