Addressing challenges in the conduct of a systematic review of clinical prediction models in rheumatoid arthritis

Article type
Authors
Archer R1, Clowes M1, Hamilton J1, Pandor A1, Stevenson M1, Stevens J1, Hock E1, Essat M1, Poku E1
1School of Health and Related Research (ScHARR), University of Sheffield
Abstract
Background: Health professionals need to be able to judge which patients newly diagnosed with rheumatoid arthritis (RA) may experience a worse prognosis to inform appropriate care.
Objective: Systematically review evidence on selected tests and assessment tools in the evaluation of prognosis in patients with early RA
Key challenges: Two key challenges were encountered. Firstly, the commissioned topic was broad in scope with an extremely large volume of potential evidence. Secondly, consideration of which methods were most appropriate for use was required.
Addressing challenges: In order to ensure feasibility, it was essential to focus on the key prognostic variables, outcomes and study designs that would best address the review question. Scoping searches using sensitive validated filters were used to assess the size of the available literature. Prognostic variables were selected (following agreement with clinical advisors) based on ready availability in UK clinical practice; clinical experience; and literature scoping. Selected outcomes have clinical relevance, importance to patients and are widely reported in RA research. Following discussion with the commissioner we focused the review to primary studies describing the development, external validation and impact of eligible clinical prediction models including at least one eligible prognostic variable in adult patients (aged 18 years and above) with early RA.
Good practice in prognostic reviews (http://methods.cochrane.org/prognosis/) was consulted. Studies were categorised using the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) classification. Data extraction and quality assessment were guided by the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) and criteria in the draft Prediction model study Risk Of Bias Assessment Tool (PROBAST). The lead reviewer and statistician collaborated on data synthesis, presented as a narrative synthesis and meta-analysis.
Conclusion: Discussions between the review team, clinical advisors and commissioner were crucial in focusing a scope that ensured feasibility and delivery of the review.
Patient or healthcare consumer involvement: Patients provided comments on the Plain English Summary and draft report.