Data from: Using risk of bias domains to identify opportunities for improvement in food- and nutrition-related research: an evaluation of research type and design, year of publication, and source of funding

Purpose: This retrospective cross-sectional study aimed to identify
opportunities for improvement in food and nutrition research by examining
risk of bias (ROB) domains. Methods: Rating were extracted from critical
appraisal records for 5675 studies used in systematic reviews conducted by
three organizations. Variables were as follows: ROB domains defined by the
Cochrane Collaboration (Selection, Performance, Detection, Attrition, and
Reporting), publication year, research type (intervention or observation)
and specific design, funder, and overall quality rating (positive,
neutral, or negative). Appraisal instrument questions were mapped to ROB
domains. The kappa statistic was used to determine consistency when
multiple ROB ratings were available. Binary logistic regression and
multinomial logistic regression were used to predict overall quality and
ROB domains. Findings: Studies represented a wide variety of research
topics (clinical nutrition, food safety, dietary patterns, and dietary
supplements) among 15 different research designs with a balance of
intervention (49%) and observation (51%) types, published between 1930 and
2015 (64% between 2000-2009). Duplicate ratings (10%) were consistent
(k=0.86-0.94). Selection and Performance domain criteria were least likely
to be met (57.9% to 60.1%). Selection, Detection, and Performance ROB
ratings predicted neutral or negative quality compared to positive quality
(p<0.001). Funder, year, and research design were significant
predictors of ROB. Some sources of funding predicted increased ROB
(p<0.001) for Selection (Interventional: industry only and none/not
reported; Observational: other only and none/not reported) and Reporting
(Observational: university only and other only). Reduced ROB was predicted
by combined and other-only funding for intervention research
(p<0.005). Performance ROB domain ratings started significantly
improving in 2000; others improved after 1990 (p<0.001). Research
designs with higher ROB were nonrandomized intervention and time series
designs compared to RCT and prospective cohort designs respectively
(p<0.001). Conclusions: Opportunities for improvement in food and
nutrition research are in the Selection, Performance, and Detection ROB
domains.