An assessment of inter-rater agreement of the literature filtering process in the development of evidence-based dietary guidelines

Abstract

OBJECTIVE: To determine whether the literature filtering process, a vital initial component of a systematic literature review, could be successfully completed by nutrition professionals or non-professionals. DESIGN: Using a diet-disease relationship as the guideline topic, inter-rater agreement for the title and abstract filtering processes between and among professionals and non-professionals was assessed and compared with an expert reference standard. Predetermined eligibility criteria were applied by all raters to 185 titles and 90 abstracts. Filtering decisions were initially made independently and then revised after a within-pair consensus meeting. SUBJECTS: The raters were six dietitians (RD) and six nutrition graduate students (Grad). To assess inter-rater agreement (reliability), each group was divided into three pairs. RESULTS: Weighted and unweighted kappa statistics and percentage agreement were calculated to determine the inter-rater agreement within pairs. Sensitivity and specificity estimates were determined by comparing responses with those of an expert reference standard. Overall, Grad pairs demonstrated greater inter-rater agreement than RD pairs for title filtering (P<0.05); no differences were observed for abstract filtering. Compared with the expert reference standard, every rater and pair had false-negative responses for both title and abstract filtering. CONCLUSIONS: After consensus meetings, both RDs and Grads were comparable in their agreement on title and abstract filtering, although important differences remained compared with the expert reference standard. This study provides preliminary findings on the value of utilising a non-expert pair in developing guidelines, and suggests that the literature filtering process is complex and quite subjective.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 10 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.