Article type
Abstract
Background: The results of mathematical modelling (MM) are used in different ways when formulating clinical or public health guidelines. There is no standardised approach to the incorporation of MM into guideline development.
Objectives: To describe the uses of MM in World Health Organization (WHO) guidelines and to provide guidance on if, when and how to use MM optimally in public health guidelines.
Methods: We reviewed all guidelines approved by the WHO Guidelines Review Committee 2007-2016 and recorded all instances that mentioned MM. We extracted the following data from each guideline: the questions that MM addressed; whether and how MM influenced the recommendation; if a de novo MM was developed and, if so, the model details. We used descriptive statistics to synthesise the data.
Results: There were 188 guidelines, of which 42 referenced MM. Of these, MM directly impacted the recommendations in 17 and GRADE profiles included MM in 11 guidelines. Preliminary analyses show that MM was used for a variety of types of questions, including risk, prognosis, intervention effectiveness and effect of diagnostic tests on health outcomes, particularly when primary data were sparse or nonexistent, e.g. for emerging diseases; long-term health outcomes; and, where contextual factors such as baseline disease prevalence varied. Guidelines rarely reported an assessment of model quality or why specific models were selected. Models developed de novo did not provide sufficient detail to assess assumptions or parameters and thus model outputs.
Conclusions: MM are frequently used to inform recommendations in WHO guidelines, but reporting of both existing and de novo models is poor. This review contributes to ongoing work at WHO that will provide guidance on when to consider using MM to inform guidelines; how to assess the quality of models; and, how to incorporate the results of MM into a body of evidence.
Objectives: To describe the uses of MM in World Health Organization (WHO) guidelines and to provide guidance on if, when and how to use MM optimally in public health guidelines.
Methods: We reviewed all guidelines approved by the WHO Guidelines Review Committee 2007-2016 and recorded all instances that mentioned MM. We extracted the following data from each guideline: the questions that MM addressed; whether and how MM influenced the recommendation; if a de novo MM was developed and, if so, the model details. We used descriptive statistics to synthesise the data.
Results: There were 188 guidelines, of which 42 referenced MM. Of these, MM directly impacted the recommendations in 17 and GRADE profiles included MM in 11 guidelines. Preliminary analyses show that MM was used for a variety of types of questions, including risk, prognosis, intervention effectiveness and effect of diagnostic tests on health outcomes, particularly when primary data were sparse or nonexistent, e.g. for emerging diseases; long-term health outcomes; and, where contextual factors such as baseline disease prevalence varied. Guidelines rarely reported an assessment of model quality or why specific models were selected. Models developed de novo did not provide sufficient detail to assess assumptions or parameters and thus model outputs.
Conclusions: MM are frequently used to inform recommendations in WHO guidelines, but reporting of both existing and de novo models is poor. This review contributes to ongoing work at WHO that will provide guidance on when to consider using MM to inform guidelines; how to assess the quality of models; and, how to incorporate the results of MM into a body of evidence.