Nutrition research provides evidence to support public health policies, advances in clinical practice, and improved quality of life for patients. Reporting diet- and nutrition-related randomized controlled trials (RCTs) requires particular attention to field-specific methodological aspects, such as baseline dietary patterns, food intake assessment instruments, and statistical analysis methods [1-3]. Strategies to deal with methodological limitations and increase transparency in nutrition research include prioritizing large and long-term RCTs, pre-registering protocols and statistical analyses, clearly defining and standardizing methods for data collection and outcomes assessment, and promoting data sharing [4-7]. Guidelines on how to handle sensitive issues, such as public-private partnerships and working with industry, are available [8,9]. All of these recommendations, however, require complete and transparent reporting, which is also crucial when these nutrition- or diet-related RCTs are pooled in a systematic review or meta-analysis. Clearly specified methods for identifying, selecting, and appraising primary studies included in systematic reviews are essential to a systematic review being of high enough quality to inform clinical practice [10,11]. Systematic reviews are the cornerstones of therapeutic evaluation as they assess and grade the available body of evidence on interventions.

Inappropriate reporting of research often means an important and avoidable source of that research wasted [12], with consequences for research consumers such as scientists, clinicians, policy makers, the media, and patients [13,14]. Researchers can minimize this source of waste by adhering to reporting guidelines, which define a minimum set of information required when reporting medical research to ensure its usefulness [15]. One of the first reporting guidelines was the Consolidated Standards of Reporting Trials (CONSORT) for writing up reports of RCTs, developed in 1996 and updated in 2001 and 2010. CONSORT consists of a 25-item checklist and flow diagram to help authors report their RCTs [16]. Item 5 of CONSORT 2010 concerns an RCT’s intervention. To improve reporting completeness and replicability of interventions, CONSORT item 5 was extended into a 12-item reporting guideline, “the Template for Intervention Description and Replication” (TIDieR) [17]. Other extensions are also available for specific designs (eg, crossover trials) [18], interventions (eg, nonpharmacologic treatment interventions) [19], and data (eg, harms) [20]. Similar guidance to improve reporting completeness for systematic reviews and meta-analyses—the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) Statement—was developed in 2009 and updated in 2020 [21].

Journal endorsement of these reporting guidelines is associated with better reporting [22-24]. For example, a Cochrane review including meta-research studies assessing the reporting completeness of RCTs demonstrated a standardized mean difference of 0.68 (99% CI 0.38 to 0.98) on the aggregate score of all CONSORT items in favor of CONSORT-endorsing journals over nonendorsers. Allocation concealment, scientific rationale in the introduction, sample size, and sequence generation were also more adequately described in RCTs published in CONSORT-endorsing journals [22]. A study evaluating journal instructions for authors found that 27 out of 33 high-impact nutrition and dietetic journals with an impact factor over 2.88 mentioned CONSORT [23]. In a cross-sectional study in the field of nursing, the median rates of adherence to the PRISMA statement were 73% (IQR 59.5%-94.6%) for reviews published in journals that endorsed PRISMA and 64.9% (IQR 17.6%-92.3%) for reviews published in journals that did not endorse PRISMA [24]. As far as we know, the correlation between reporting completeness and journal endorsement of PRISMA has not been explored in the nutrition field.

Few studies assessed reporting completeness of nutrition research. Although previous studies have found poor reporting in nutrition- and diet-related studies [25-31], most focused on the association of particular funding sources with methodological and reporting quality [25-27,30,31] or risk of bias [28,29]. We identified only 1 systematic review of RCTs assessing the quality of reporting in RCTs of dietetic interventions in primary care. The authors found that none of the 27 included RCTs reported all CONSORT items, with a median of 11.5 of the 28 items completely met, 8 partially met, and 10 not met at all. The items most often not reported were “a description of where the full protocol can be accessed (n=26, 96%), ”registry number and name of trial registry“ (n=23, 85%), ”mechanism used to implement the random allocation sequence“ (n=22, 81%), and ”who generated the random allocation sequence, enrolled participants and assigned participants'' (n=22, 81%) [32]. We did not find any study assessing the completeness of reporting of systematic reviews in nutrition. One scoping review evaluated 57 studies that assessed adherence to individual PRISMA items, where 79% (n=45) of them focused on therapeutic interventions and related to surgery clinical areas. It identified 11 items that are fewer than 67% of systematic reviews adhered to, including structured summary, protocol and registration, search, data items, risk of bias in individual and across studies (in both methods and results sections), additional analyses, and funding [11].

CONSORT and TIDieR do not cover all possible practices to enhance research reproducibility, such as the inclusion of data-sharing and code-sharing statements. A survey of a random sample of 149 biomedical articles published between 2015 and 2017 found that 18.3% discussed publicly available data [33]. Similarly, an automated assessment of all PubMed Central articles estimated that in 2020, overall 15% of articles had a data-sharing statement and 3% had a code-sharing statement [34].

“Spin,” as defined by Boutron et al [35], is the use of specific reporting strategies, from whatever motive, to highlight that the experimental treatment is beneficial, despite a statistically nonsignificant difference for the primary outcome, or to distract the reader from statistically nonsignificant results.

Spin is frequently found in the medical literature, regardless of study design, and its presence can affect the interpretation of results. A systematic review of 35 reports which investigated spin found that more than 56% of trials included spin in their abstract (n=13) and main text (n=16) and that 26.3% of systematic reviews (n=2) included spin in their main text [36]. We did not identify any studies assessing reproducible research practices or the presence of spin in nutrition systematic reviews or meta-analyses.

This protocol describes 2 independent meta-research studies that aim to (1) assess the reporting completeness of diet- and nutrition-related RCTs, identify their main reporting limitations, and estimate the frequency of reproducible research practices among them; and (2) assess the reporting completeness of systematic reviews with meta-analysis of diet- and nutrition-related RCTs, identify their main reporting limitations, evaluate their risk of bias, and estimate the frequency of spin among them.

Two meta-research studies will be conducted to answer the research questions presented in Textbox 1. Both protocols are registered in the Open Science Framework (doi: 10.17605/OSF.IO/BF47G and 10.17605/OSF.IO/ZTFYX).

For meta-research study 1, we will include diet- and nutrition-related RCTs published in peer-reviewed journals. We will consider self-identification as an RCT as an inclusion criterion. All diet- and nutrition-related RCTs, regardless of the population or outcomes evaluated or the framework design will be included (eg, noninferiority, superiority, or exploratory). There will be an exclusion of non-RCTs, pilot and feasibility trials, studies that assess pharmaceutical or herbal medicines as an intervention, and trials without isolated nutrition- or diet-related interventions (meaning that trials combining more than 1 intervention will not be eligible). RCTs that do not include any term related to the categories of intervention described in Textbox 2 in their titles or abstracts will be excluded.

For meta-research study 2, systematic reviews will be included with meta-analysis of diet- and nutrition-related RCTs published in peer-reviewed journals. All Cochrane publications will be excluded as their moderate to high methodological quality is likely to ensure a high level of reporting completeness [37].

We adapted the categories proposed by Naude et al [37] to include “shared foods of complete diet and dietary patterns” and “enteral or parenteral formulas” and remove “nutrition-related policies.”

For both meta-research studies, the language of publications will not be an eligibility criterion aiming to avoid language bias. Our team has researchers proficient in English, Portuguese, and Spanish. If necessary, Google Translate or private translation services will be used to extract data from publications in other languages. The nutrition intervention categories of interest are listed in Textbox 2 [37].

For meta-research study 1, we will use the search strategy developed by Durão et al [38] to identify RCTs of nutritional interventions [32]. For meta-research study 2, a search strategy is constructed using nutrition- and diet-related terms [38] combined with a validated search strategy to identify systematic reviews or meta-analyses [39]. No limitation will be set for language during the literature search. PubMed search is conducted (via the National Library of Medicine) using the search strategies presented in Multimedia Appendices 1 and 2. There will be a search for RCTs (June 1, 2021, to June 1, 2022) and systematic reviews with meta-analysis (July 01, 2021, to July 01, 2022) indexed in 1 calendar year.

For both meta-research studies, we will organize the retrieved reports in EndNote (Clarivate). We will remove duplicates using EndNote’s automated deduplication, then manually check for and exclude any remaining duplicates. The final library without duplicates to Rayyan will be imported [40], where 2 reviewers will independently determine the eligibility of each report in a 2-stage process. They will screen titles and abstracts and select all RCTs and systematic reviews with meta-analysis assessing any diet- or nutrition-related intervention. Liberal acceleration will be applied, in which both reviewers must exclude a record for it to be excluded, while only 1 reviewer must include a record for it to be included. The reviewers will then read the full texts to confirm study eligibility. In both screening rounds, any discrepancies will be resolved through discussion between the reviewers (FS and JL, acknowledged), with adjudication by a third reviewer (MSM). All reviewers have experience in conducting systematic reviews.

The eligible RCTs and systematic reviews with meta-analysis will be grouped according to the intervention categories described in Textbox 2, and the proportion of RCTs and systematic reviews in each category is calculated. A convenience sample of 100 RCTs will be randomly selected for inclusion in meta-research study 1 and 100 random systematic reviews with meta-analysis for meta-research study 2 that meet the above criteria, with no formal sample size calculation. We aim to maintain the proportion of each intervention category (Textbox 2) in both meta-research studies, for example, for a given intervention category that comprises, say, 50% of the total sample of potentially eligible RCTs, we will randomly select 50 of those for our sample of 100 RCTs. A web-based random number generator will be used to create the list of publications to include from an enumerated sequence of all studies in each intervention category [41].

A flowchart will be presented to demonstrate each step of the study selection process. A list of the studies excluded at the full-text screening stage will be presented with reasons for their exclusion for both meta-research studies.

For meta-research study 1, a total of 2 reviewers will independently extract data from the selected RCTs to evaluate their reporting completeness, identify their main reporting limitations, and assess their use of predefined reproducible research practices. We have developed a data extraction form (Multimedia Appendix 3) that includes (1) publication data, such as PMID, first author’s name, journal in which the RCT was published, and the journal’s field according to Web of Science. (2) Basic description of RCT data, categorized into population, intervention, comparator, outcome, and study design (PICOS) domains, adapted from Naude et al [37]. (3) Use of practices that promote research transparency and reproducibility, such as the use of CRediT [42] to acknowledge authorship; availability of a full protocol publication, statistical analysis plan, and materials/data/code; open access; disclosure of funding source; and disclosure of conflicts of interest. All these practices will be categorized as “yes or no.”

An additional data extraction form to evaluate adherence to items from CONSORT [16], its extensions for abstracts [43], harms [20], crossover [18], nonpharmacological [19], cluster trials [44], and TIDieR [15] statements will be developed. Each item will be evaluated as fully reported, not reported, partially reported, or not applicable. As there is an inclusion of RCTs published after the COVID-19 pandemic began, items from the CONSERVE reporting guideline will also be included. CONSERVE was developed to improve the reporting completeness of trials modified due to the COVID-19 pandemic or other extenuating circumstances [45] (Multimedia Appendix 4).

For meta-research study 2, a total of 2 reviewers will independently extract data from the selected systematic reviews with meta-analysis of nutrition- or diet-related RCTs. We have developed a data extraction form (Multimedia Appendix 5) to capture (1) Publication data, such as PMID, first author’s name, journal in which the meta-analysis was published, and the journal’s field according to Web of Science. (2) Meta-analysis research questions, categorized into PICOS domains, adapted from those used by Naude et al [37]. (3) The presence of spin in the abstract and the main text, classified into four main categories: (A) misleading reporting, including 10 items in the main text and 8 items in the abstract; (B) misleading interpretation, including 13 items in the main text and 10 items in the abstract; (C) inappropriate extrapolation, including 5 items in the main text and 3 items in the abstract, as proposed by Yavchitz et al [46]; and (D) methodological quality according to the ROBIS tool. Phases 2 and 3 of the tool will be followed to make an overall judgment of risk of bias (which will be rated as low, high, or unclear) [47].

An additional data extraction form to evaluate adherence to PRISMA [21], its extensions [48,49], and TIDieR [15] items was also developed (Multimedia Appendix 6).

For both studies, we will pilot-test the data extraction forms on 5 randomly selected full-texts before proceeding with full data extraction to ensure consistency in the interpretation of data items and refine the data extraction form, if necessary. All extracted information will be entered directly into the study databases using REDCap [50]. Any disagreement between the pairs of reviewers will be resolved by a third reviewer.

For both meta-research studies, descriptive aspects of included publications (ie, data on bibliometrics, PICOS, research transparency practices, presence of spin, and risk of bias) will be presented both as absolute and relative frequencies, either in diagrams or tables.

For meta-research 1, calculation of an aggregate score of adherence to all CONSORT, its extensions, CONSERVE, and TIDieR items will be presented. The items of relevant reporting guidelines’ checklists in 170 questions with categorical answers (Multimedia Appendix 3) were fragmented. Most questions presented 2 options as answers (yes or no), whereas some questions presented 3 or 4 options as answers (yes, no, partially, or not applicable). Answers “yes” will be scored as 1 point, while answers “no” will be scored as 0 points and answers “partially” will be scored as 0.5 points. Answers such as “not applicable” will be discounted for that particular study. Therefore, the final score can range from 0 to 170.

For meta-research 2, we will similarly calculate an aggregate score of adherence to all PRISMA, its extensions, and TIDieR items. For meta-research 2, all items of the corresponding reporting guidelines were fragmented into 114 questions with the same possible categorical answers (and respective attributable scores) described above (Multimedia Appendix 6). Thus, the final score for the meta-research of meta-analyses can range from 0 to 114.

In the final report, the results will be presented as frequencies (and respective percentages in relation to the total number of relevant studies) for individual items and proportion of the total achievable score for individual studies. No data aggregation or cutoff points are anticipated for the descriptive analyses.

The general features between RCTs and meta-analyses with above-average and below-average reporting score proportions were also compared, which are stratified by the mean or median value (depending on the distribution) using the Student t test or Mann-Whitney test for quantitative variables and the chi-square test for categorical variables. Logistic regression models will be constructed to explore the determinants of reporting completeness.

Any changes to the protocol will be detailed and justified in the publications describing the results of the study.

As both meta-research studies will involve reviewing and collecting evidence from previously published papers, ethical approval is not required. Both meta-research studies will be reported according to the PRISMA adaptation for meta-epidemiological methodology research [51]. Any updates and amendments to this protocol will be described and justified in the final publications. The results will be published in peer-reviewed journals.