Protocol for a systematic review of prognostic models for recurrent events in chronic conditions

• Main Authors: Victoria Watson, Catrin Tudur Smith, Laura Bonnett
• Format: Article
• Language: English
• Published: BMC 2020-01-01
• Series: Diagnostic and Prognostic Research
• Subjects: Prognostic factors; Prognostic models; Recurrence; Chronic condition; Prediction; Validation
• Online Access: https://doi.org/10.1186/s41512-020-0070-9

Abstract Background Prognostic models for repeated events of the same type are highly useful in predicting when a patient may have a recurrence of a chronic disease or illness. Whilst methods are currently available for analysing recurrent event data in prognostic models, to our knowledge, most are not widely known or applied in a medical setting. As a result, often only the first recurrence is analysed meaning valuable information for multiple recurrences is discarded. Therefore, the aim of this review is to systemically review models for repeated medical events of the same type, to determine what modelling techniques are available and how they are applied. Methods MEDLINE will be used as the primary method to search sources. Various databases from the Cochrane Library and EMBASE will also be searched. Trial registries such as Clinicaltrials.gov.uk will be searched, as will registered trials that are ongoing and not yet published. Abstracts submitted to conferences will also be searched, and non-English sources will also be considered. Studies to be included in the review will be decided based on PICO guidelines, where the study population and outcomes correspond to this study’s aims and target population. The prognostic models used in each study chosen for inclusion in the review will be summarised qualitatively. Discussion As recurrent event data is not widely analysed in prognostic models, the results from this systematic review will identify which methods are available and which are commonly used. It is also unknown if certain methods which will be identified in the review perform better given certain conditions. Therefore, if included studies assess predictive performance, the results of this review could also provide evidence to determine if certain models are better fitting dependant on the event rate of the chronic condition. The results will be used to determine if model selection varies across disease area. The review will also provide an insight into the development of any new methods used for analysing recurrent events. Trial registration The review has been registered on PROSPERO (CRD42019116031).