Table 2

A practical way forward. Recommendations for researchers, product developers, policy makers and funders to accelerate the development and implementation of prognostic tools for the management of febrile illnesses in resource-limited settings, informed by a recent stakeholder consultation exercise.

Practical steps to improve the design and reporting of studies aiming to accelerate the development and implementation of prognostic tools for the management of febrile illnesses in resource-limited settings
ResearchersProduct developersPolicy makers and funders
1. Describe and respect the clinical use-case that the prognostic test or algorithm aims to fulfil
The study population must reflect the clinical problem that the novel test or algorithm aims to address, for example, the inclusion of outpatients for studies aiming to develop tools for community-based use. Technology must be developed in partnership with users to ensure it meets their needs. Integrated care models must be advocated for and adopted rather than vertical disease-specific programmes, and training of health workers must be prioritised to support the sustained uptake of new tools.
2. Measure candidate predictors using common frameworks for data collection
Candidate predictors should be measured using comparable methodologies to encourage data sharing,44 and predictors already identified as promising must be included to allow evaluation of external validity.47 52 53
3. Define relevant outcomes against which candidate predictor(s) will be assessed
Comprehensive outcome sets that include surrogate endpoints must be defined, particularly for use-cases where mortality may not be a relevant or feasible outcome. Ideally these should be prospectively agreed on by all members of the research community.54
4. Use standardised tools to assess human and material resources available in the targeted settings
Study settings must be described using standardised tools to contextualise findings and encourage pooling of data from similar environments.45
5. Report findings in accordance with existing guidelines
Study design must be adequately reported (eg, the proportion of participants who had met the endpoint at the time candidate predictors were measured)27 and results should be summarised using metrics that reflect clinical decision making (eg, positive and negative predictive values, likelihood ratios and net-benefit analyses). Simple technology that can provide quantitative outputs should be invested in to allow cut-offs to be tailored to different risk-benefit scenarios.
  • Number of checkmarks indicate the relative importance of each recommendation for each group.