Evaluating the reproducibility and verifiability of nutrition research: a case study of studies assessing the relationship between potatoes and colorectal cancer Authors: Yasaman Jamshidi-Naeini, Colby J Vorland, Pranav Kapoor, Bailey Ortyl, Jocelyn Mineo, Luke Still, Korlu Sorsor, Shelby Rodney, Xander Tooze, Brent Flickinger, Beate Henschel, Stephanie L Dickinson, David B Allison Published: 2026 ABSTRACT AbstractThe credibility of nutritional research is dependent on the rigor with which studies are conducted and the ability for independent assessment to be performed. Despite the importance of these, more work is needed in the field of nutrition to buttress the trustworthiness of nutrition research. To develop and apply a process for evaluating the rigor, reproducibility, and verifiability of nutritional research, using the relationship between potato consumption and Colorectal Cancer (CRC) as a case study. We updated existing systematic reviews to include studies on potatoes and CRC, assessing their design, execution, and reporting quality. We attempted to reproduce and verify the results of included studies by requesting raw data from authors and following statistical methods as described in the publications. Rigor was evaluated using four different tools: ROBINS-E, STROBE-Nut, Newcastle-Ottawa scale, and additional criteria related to transparency. Eighteen studies were included, none of which publicly share data. We managed to access data for only two studies, successfully reproducing and verifying the results for one. The majority of studies exhibited a high risk of bias, with significant limitations in reporting quality and methodological rigor. Research on the relationship between potato consumption and CRC risk is insufficiently reproducible and verifiable, undermining the trustworthiness of its findings. This study highlights the need for improving transparency, data sharing, and methodological rigor in nutritional research. Our approach provides a model for assessing the credibility of research in other areas of nutrition.Subject terms: Cancer, Risk factors Introduction Trust in nutrition science appears to be lower than trust in other domains of science [1, 2]. Trust in science should be predicated on the trustworthiness of the science [3]. In turn, the trustworthiness of nutritional research relies in part on the independent assessment of the quality of the research. This depends on how well the research is designed, executed, analyzed, and reported (i.e., rigor1 (Definition of rigor: “the strict application of the scientific method to ensure unbiased and well-controlled experimental design, methodology, analysis, interpretation and reporting of results” (1))) [4]. Trustworthiness can be enhanced if raw data can be obtained and independently analyzed to assess reproducibility2 (Definition of reproducibility: “obtaining consistent results using the same input data, computational steps, methods, and code, and conditions of analysis. This definition is synonymous with “computational reproducibility”” (1)) and verifiability3 (Definition of verifiability: “Verifiability subsumes, but goes beyond, reproducibility. That is, a study is said to be verifiable, and to have been verified, when: (a) the study is reproducible, and the results have been reproduced (by the definition of reproducibility above); and (b) a determination is made that the methods used to generate the results reproduced are valid methods and that the interpretations validly and logically follow from the obtained results.” [2]) [5]. Other research has examined the reproducibility of research published in specific journals [6–12] or disciplines [13, 14]. To our knowledge, no study has evaluated research on a nutrition topic for reproducibility, nor evaluated the verifiability. Potatoes are a staple food for many populations worldwide, estimated to be the third most consumed crop [15]. As such, it is paramount to understand their relationship with health. The Scientific Report of the 2020 Dietary Guidelines Advisory Committee assigned a grade of “moderate evidence” that dietary patterns higher in foods including potatoes are associated with a greater colon and rectal cancer (CRC) risk [16]. However, it is not clear from the report whether potatoes in isolation are related to CRC risk. While some systematic reviews have been published on the topic [17, 18], traditional systematic review procedures do not evaluate reproducibility and verifiability of research on a particular research question. Herein, we develop a process to assess the rigor, reproducibility, and verifiability of nutritional research, using the relationship between potatoes and CRC as a case study. We evaluate the reproducibility and verifiability primarily at the level of individual primary studies, as systematic review conclusions depend fundamentally on the reproducibility and transparency of the studies they include. Systematic reviews were used as a method to identify relevant studies, rather than as the primary unit of reproducibility assessment. By evaluating whether individual studies can be independently reproduced and verified, we provide insight into the trustworthiness of the broader evidence base. The contribution of our work is the development and application of a framework. This framework includes systematic identification of studies, direct requests for raw data, independent reproduction of reported statistical analyses, verification of analytical procedures, and evaluation of methodological rigor and transparency using multiple established tools and criteria. This structured approach can be applied to other research questions in nutrition to systematically assess the credibility and reproducibility of the evidence base. Methods Search strategy and screening We identified three recent systematic reviews by Schwingshackl et al. [17], Mofrad et al. [18], and Mofrad et al. [19] that drew conclusions about the association between potatoes and CRC risk, and used them as starting point to identify relevant studies. The primary focus of reproducibility and verifiability assessment was at the level of individual studies, including access to raw data and reproducibility of reported analyses. We also updated the systematic review of [17] up to June 2022 to identify new studies since their last search. We first modified their PubMed search string based on the purpose of the current work, and then with the help of a librarian, constructed our search strings for Web of Science and Scopus (Appendix 1). The search was run on June 13, 2022. Two analysts independently screened all studies for relevance and included all that met predefined inclusion and exclusion criteria using Covidence (Fig. 1). We included studies that met the following criteria: Study design (self-identified by the authors): cohort, case–control, nested case–control, or randomized controlled trial; human participants aged 18 years or older on average; participants free of CRC at the onset of the study unless specified otherwise; studies on CRC recurrence or in CRC survivors were included if participants were not currently diagnosed with cancer, and for case-control designs, studies were included if there was a control group without CRC, even if the “cases” had CRC; studies must have included an analysis for the independent relationship between potatoes (in any form) and CRC. We excluded studies that met the following criteria: study type: reviews, animal studies, and in vitro studies; participants: studies including participants younger than 18 years of age on average or exclusively pregnant women; studies where all participants had CRC at the onset, except in case-control studies where this criterion did not apply; studies without an independent analysis for the relationship between potatoes (in any form) and CRC. Characteristics were extracted from each study, including the design, cohort used, follow-up length, statistical methods and adjustment factors, sample size, country, gender(s) studied, dietary and cancer assessment methods, stage of cancer for the participants, potato type studied, inclusion and exclusion criteria, results, funding, data sharing statements, and the current journal data sharing policy. Flowchart of study identification and screening process. Assessing reproducibility and verifiability We first attempted to contact authors and request raw data by emailing, mailing letters, and calling them. We first emailed the corresponding author, and, if upon not receiving a reply, emailed other authors whereby we could locate them based on name and affiliation listed in the publication. Where we could obtain access to raw data, we first assessed whether we could obtain the same results for the potato-CRC association(s) by following the methods as published (Fig. 2). We first checked whether, following the reported methods in the article, we could obtain the same number of participants included in the potato-CRC association(s) and whether the reported summary statistics could be reproduced. Each check was conducted by a statistician member of our team who documented their process in statistical code, and the process was checked by a senior statistician. For any articles that could not be reproduced, we emailed the author(s) and shared our report and asked for clarification to identify potential discrepancies or misunderstandings on our part. Flowchart of reproducibility and verifiability assessment. If, after those procedures, we could not obtain the same results, or if the authors did not respond, the results were considered irreproducible and thus unverifiable. If we could reproduce the results, we verified the results by assessing whether the statistical methods were appropriate for the study design and whether the statistical conclusions and statements were appropriate given the analyses and results obtained. Assessment of rigor To assess the rigor of the studies included, we used four methods. For each method, two investigators independently rated each article, and discrepancies were resolved through discussion. ROBINS-E tool, which assesses risk of bias in non-randomized studies of exposures [20]. ROBINS-E tool, which assesses risk of bias in non-randomized studies of exposures [20]. STROBE-Nut checklist, for reporting quality with an extension for nutritional epidemiology [21]. STROBE-Nut checklist, for reporting quality with an extension for nutritional epidemiology [21]. Newcastle-Ottawa scale, a quality checklist for non-randomized studies [22]. Newcastle-Ottawa scale, a quality checklist for non-randomized studies [22]. Additional criteria that we specified related to indicators of rigor (registration of outcomes or analysis plan, data and code availability), including common issues in scientific articles identified in our previous work [23]. Additional criteria that we specified related to indicators of rigor (registration of outcomes or analysis plan, data and code availability), including common issues in scientific articles identified in our previous work [23]. The IRB of Indiana University determined that this study did not require human subjects review. Results Eighteen studies were included [24–41]; their characteristics are described in Table 1. All studies were non-randomized, which limits the ability to draw causal conclusions. Of the 18 studies, 11 were prospective cohorts and 7 were case controls. The studies were done in 12 different countries. Six studied included only women, the others included both women and men. Seventeen studies used a food-frequency questionnaire to assess potato consumption, and one used 24-h dietary recalls. CRC was rigorously ascertained in almost all studies using registry linkage or review of medical records. None of the studies publicly shared data, and two studies indicated that data were available on request. Table 1Characteristics of included studies.StudyDesignCountryGender(s) studiedDietary assessment methodPotato description as the exposureOutcome ascertainmentAsli [32]Prospective cohortNorwayWomenFFQPotatoes (details unspecified)Registry linkageFlood [33]Prospective cohortUSAWomenFFQPotatoes excluding French friesRegistry linkageKato [34]Prospective cohortUSAWomenFFQPotatoes (details unspecified)Self-report and medical recordsMichels [35]Prospective cohortUSAWomen, MenFFQPotatoes (mashed or baked), French fries, Potato chipsSelf-report and medical recordsSteinmetz [37]Prospective cohortUSAWomenFFQPotatoes (details unspecified)Registry linkageMucci [36]Prospective cohortSwedenWomenFFQPan-fried potatoes, Potato chips and French friesRegistry linkageYeh [38]Prospective cohortTaiwanWomen, MenFFQSweet potatoesRegistry linkage and medical recordsLiu [39]Case-controlChinaWomen, MenFFQPotatoes (details unspecified)Medical recordsPartula [40]Prospective cohortFranceWomen, Men24-h recallPotato and tuber fiberSelf-report and medical recordsTayyem [41]Case-controlJordanWomen, MenFFQFried potato, potato and corn chipsMedical recordsAgnoli [24]Prospective cohortItalyWomen, MenFFQPotatoes (details unspecified)Medical recordsAnnema [25]Case-controlAustraliaWomen, MenFFQPotatoes (details unspecified)Registry linkageDeneo-Pellegrini [26]Case-controlUruguayWomen, MenFFQPotatoes, sweet potatoesMedical recordsKhan [27]Prospective cohortJapanWomen, MenFFQPotatoes (details unspecified)Self-report or family contactLevi [28]Case-controlSwitzerlandWomen, MenFFQPotatoes (details unspecified)Medical recordsLin [29]Prospective cohortUSAWomenFFQPotatoes, French fries, Potato chipsSelf-report and medical recordsPing [30]Case-controlJapanWomen, MenFFQPotato products (details unspecified)Medical recordsTayyem [31]Case-controlJordanWomen, MenFFQSweet potatoes, mashed potatoesMedical records Characteristics of included studies. Reproducibility and verifiability assessment We received no response from 10 of 18 research groups (56%) (Table 2). Two contacts indicated that they were retired and declined to further work with us. We received one cost estimate to access data that was above our available resources ($51,644.73). One group shared a data file that was insufficient to reproduce analyses and declined to share additional data. One group was willing to share data conditional on IRB approval and a data use agreement, but the timeline from contacting them to initiating these was too long (>8 months). One group permitted (required as a condition of granting access to the data) us to visit their group in another country (France) and analyze their data on their computers, and one group sent us all data needed to reproduce and verify their analyses. Table 2Summary of contacting authors for raw data.Result of contactNumber of studiesaNo response10Responded but declined to share data (reason: retired)2Shared data file but was insufficient to reproduce analyses1Cost to access too high1Language barrier on application for data access1Willing to share conditional on local IRB approval and data use agreement, but timeline to initiate these processes was too long (>8 months)1Allowed data access from institutional servers1Shared data1aRepresents number of studies; for some studies, multiple authors were contacted (see “Methods”). Summary of contacting authors for raw data. aRepresents number of studies; for some studies, multiple authors were contacted (see “Methods”). Of the two studies for which we had access to data [33, 40], we could reproduce and verify the results of one study, and we could not reproduce and therefore not verify the results of the other study. Reproducibility analysis of Flood et al Details and numerical values of our reproducibility results are provided in Appendix 2. In summary, we first aimed to reproduce the quintiles of fruit and vegetable intake as reported in Table 2 of Flood et al. [33]. While there were some discrepancies within the original paper regarding the number of participants included in the analyses (e.g., inconsistencies between the text and table footnote), we obtained a sample size nearly identical to that in the table footnote (45,490 in the paper vs. 45,494 in our analysis). The paper reported hazard ratios (HRs) and 95% confidence intervals (95% CIs) for quintiles of fruit and vegetable intake in servings per day per 1000 kJ. However, the dataset we received did not include a variable for servings per day. Without a variable for servings, we tested whether we could use frequency of consumption by dividing the weekly frequencies by seven to estimate daily servings. We then converted kilocalories to kilojoules to calculate servings per day per 1000 kJ. We obtained reasonably comparable values for median and cutoff points across the quintiles of fruit intake. However, for vegetable intake, we were unable to reproduce the median and cutoff values reported in the paper (Table 2 of the original paper). A similar pattern emerged for hazard ratios. The HRs we calculated based on the quintile cutoffs provided in the original paper were largely consistent with those reported by Flood et al. [33] for fruit intake, but not for vegetable intake (which includes potatoes). For example, for quintile 5 of vegetable intake, we obtained an HR of 0.567, while the paper reported an HR of 0.92. We also calculated HRs using the cutoffs derived from our analysis (which differed from those reported in the paper), as well as through additional exploratory analyses in an attempt to reproduce the original HRs. We were unable to reproduce the exact results. However, these differences did not alter the overall conclusions of the study. That is, despite likely using a different method to define servings, our results closely approximated the published results, leading to conclusions similar to those of the original paper. Given our inability to reproduce the broader category results for total vegetable intake and quintile cutoffs, we did not proceed with calculating HRs specifically for potato consumption and CRC. The authors of the original study were collegial throughout the process, and this was the only study for which the complete dataset required for reproducing analyses was sent to us. However, with over twenty years having passed since the original study, the lead author had since moved institutions and no longer had access to all the data as they did at the time of publication. While the authors were willing to assist, the time elapsed was a limiting factor for providing further clarification. This experience highlights the importance of robust documentation of statistical procedures and data sharing to ensure the reproducibility of scientific research—an aspect of scientific research that is more widely recognized and emphasized today than it was when the original study was conducted. Reproducibility analysis of Partula et al To reproduce the analyses in Partula et al., a condition of access to the data required members of our team to travel from the United States to France to work on the data on their servers. Prior to our visit, we were provided with the analytical code, which we reviewed to familiarize ourselves with the approach. Once in France, we checked and re-ran their code on the final analytic dataset for the association between potato fiber and CRC. However, we were not allowed to take the analysis output off their servers. Instead, we documented the output on-site and cross-checked the numbers against those reported in the paper. We obtained the same numerical values, with minor discrepancies in the second or third decimal places for a few values. These differences did not affect the overall conclusions of the analysis. Thus, we reproduced the models related to potato fiber intake and CRC. We also verified that the statistical procedures used were appropriate to answer the study question. Assessment of rigor ROBINS-E The results of the assessment of the seven domains are shown in Table 3. Sixteen (89%) of the studies were judged at high risk of bias, mostly due to measurement weaknesses and lack of specified or state-of-the-art missing data handling procedures. Table 3Risk of bias assessment using ROBINS-E.StudyConfoundingMeasurement of the exposureSelection of participants into the study (or analysis)Post-exposure interventionsMissing dataMeasurement of the outcomeSelection of the reported resultOverall Risk of BiasAsli [32]LowSome concernsLowLowHighLowSome concernsHighFlood [33]Some concernsSome concernsLowLowHighLowSome concernsHighKato [34]HighSome concernsLowLowHighLowSome concernsHighMichels [35]LowSome concernsLowLowHighLowSome concernsHighSteinmetz [37]Some concernsHighLowLowSome concernsLowSome concernsHighMucci [36]Some concernsHighLowLowSome concernsLowSome concernsHighYeh [38]LowHighLowLowLowLowSome concernsHighLiu [39]LowHighSome concernsLowHighLowSome concernsHighPartula [40]Some concernsLowLowLowSome concernsLowSome concernsSome concernsTayyem [41]LowHighSome concernsLowLowLowSome concernsHighAgnoli [24]LowHighLowLowSome concernsLowSome concernsHighAnnema [25]LowHighSome concernsLowLowLowSome concernsHighDeneo-Pellegrini [26]Some concernsHighLowLowLowLowSome concernsHighKhan [27]HighHighLowLowLowLowSome concernsHighLevi [28]Some concernsHighLowLowHighLowSome concernsHighLin [29]LowHighLowLowHighLowSome concernsHighPing [30]Some concernsHighLowLowHighLowSome concernsHighTayyem [31]LowHighSome concernsLowHighLowSome concernsHigh Risk of bias assessment using ROBINS-E. STROBE-Nut The results of the assessment of the six sections are shown in Table 4. Most items in the “title and abstract”, “introduction”, and “discussion” sections were reported, but some items tended to be missing from “methods”, “results”, and “other information” sections. Table 4Reporting quality assessed using STROBE-Nut.StudyTitle and abstractIntroductionMethodsResultsDiscussionOther informationAsli [32]67%100%83%100%100%33%Flood [33]100%100%74%90%100%0%Kato [34]33%100%58%60%83%0%Michels [35]67%100%65%91%100%0%Steinmetz [37]67%100%72%70%100%0%Mucci [36]67%100%65%92%100%0%Yeh [38]67%100%56%70%67%0%Liu [39]100%100%68%50%100%33%Partula [40]100%100%85%92%100%67%Tayyem [41]100%100%75%44%83%33%Agnoli [24]100%100%72%70%100%0%Annema [25]100%100%80%73%100%33%Deneo-Pellegrini [26]100%100%70%44%100%0%Khan [27]67%100%39%60%100%0%Levi [28]67%100%74%25%67%0%Lin [29]100%100%74%90%100%0%Ping [30]67%50%70%44%50%0%Tayyem [31]67%100%75%44%83%33% Reporting quality assessed using STROBE-Nut. Newcastle-Ottawa scale The number of stars for each section, as well as the total stars, are shown in Table 5. Table 5Quality assessment using Newcastle-Ottawa Scalea.StudySelection (# stars)Comparability (# stars)Exposure/outcome (# Stars)Total starsAsli [32]3238Flood [33]3227Kato [34]3238Michels [35]2215Steinmetz [37]3227Mucci [36]3238Yeh [38]4228Liu [39]2237Partula [40]2237Tayyem [41]1214Agnoli [24]2237Annema [25]3216Deneo-Pellegrini [26]2226Khan [27]3216Levi [28]2215Lin [29]2226Ping [30]3216Tayyem [31]1214a “A study is judged on three broad perspectives: the selection of the study groups; the comparability of the groups; and the ascertainment of either the exposure or outcome of interest for case-control or cohort studies, respectively.” Stars are awarded such that the highest quality studies are awarded up to nine stars. https://www.ohri.ca/programs/clinical_epidemiology/oxford.asp. Quality assessment using Newcastle-Ottawa Scalea. a “A study is judged on three broad perspectives: the selection of the study groups; the comparability of the groups; and the ascertainment of either the exposure or outcome of interest for case-control or cohort studies, respectively.” Stars are awarded such that the highest quality studies are awarded up to nine stars. https://www.ohri.ca/programs/clinical_epidemiology/oxford.asp. Additional criteria The results of the assessment are reported in Table 6. No studies were preregistered, had publicly available data or code, or corrected for multiple testing. Six (33%) used inappropriate causal language, and three (17%) exaggerated non-significant effects. Seven (39%), based on the methods and statistical analysis described, appear to contain errors in failing to statistically account for sources of clustering that would require raw data to definitively confirm. Table 6Additional criteriaa.StudyPre-registered?Data publicly available?Code publicly available?Inappropriate causal language?Exaggerating significant effectbExaggerating non-significant effectcClustering errorsdDINS errorseControlled for multiple testingAsli [32]NoNoNoNoNoNoNoNoNoFlood [33]NoNoNoNoNoNoNoNoNoKato [34]NoNoNoNoNoNoNoNoNoMichels [35]NoNoNoYesNoNoNoNoNoSteinmetz [37]NoNoNoNoNoYesNoNoNoMucci [36]NoNoNoNoNoNoLikelyNoNoYeh [38]NoNoNoNoNoNoLikelyNoNoLiu [39]NoNoNoYesNoNoLikelyNoNoPartula [40]NoNoNoNoNoNoNoNoNoTayyem [41]NoNoNoYesNoYesLikelyNoNoAgnoli [24]NoNoNoNoNoNoLikelyNoNoAnnema [25]NoNoNoYesNoNoNoNoNoDeneo-Pellegrini [26]NoNoNoNoNoNoLikelyNoNoKhan [27]NoNoNoYesNoNoNoNoNoLevi [28]NoNoNoNoNoNoNoNoNoLin [29]NoNoNoNoNoNoNoNoNoPing [30]NoNoNoYesNoNoNoNoNoTayyem [31]NoNoNoNoNoNoLikelyNoNoaFor language-related claims and design/analysis questions, we assessed only the potato-CRC questions.bTouting the importance of a statistically significant difference despite the estimated effect size being scientifically insignificant.cTouting the importance of an estimated effect size despite no statistically significant difference.dOur assessments are based on summary data and the descriptions of the methods in all but two cases (Flood et al. and Partula et al.). We would need the raw data to definitively confirm.eDifferences in Nominal Significance (DINS) errors are when p-values from within-group tests are compared to reach statistical conclusions, which is not a valid approach. Additional criteriaa. aFor language-related claims and design/analysis questions, we assessed only the potato-CRC questions. bTouting the importance of a statistically significant difference despite the estimated effect size being scientifically insignificant. cTouting the importance of an estimated effect size despite no statistically significant difference. dOur assessments are based on summary data and the descriptions of the methods in all but two cases (Flood et al. and Partula et al.). We would need the raw data to definitively confirm. eDifferences in Nominal Significance (DINS) errors are when p-values from within-group tests are compared to reach statistical conclusions, which is not a valid approach. Conclusion Research on the relationship between potato consumption and CRC is insufficiently reproducible and verifiable. Of the eighteen studies included in our sample, none of them publicly shared data, and we were only able to access data to reproduce studies in 2 of 18 cases (11%), and for one of those we had to leave our country to do so. These results are in line with existing observations. A meta-analysis of meta-research studies of medical and health research found that 8% (95% CI, 5–11%) of studies declared public data availability, and 2% (95% CI, 1–3%) actually publicly shared data [42]. Gabelica et al. requested data from authors of nearly 1800 publications in which they declared data available on reasonable request, and received data in 6.8% of cases [43]. One of the publications in our set (6%) was reproducible and verifiable. This is lower than other estimates; however, other studies use varying sampling methods and definitions of reproducibility. In samples from journals Psychological Science and Cognition, with author involvement, 24% and 62% of outcomes were reproducible [6, 7]. The earliest publication date for included studies was 2014, and the authors defined a major numerical error as a 10% or greater error. In another analysis of the reproducibility of primary outcomes from randomized controlled trials published in The BMJ and PLOS Medicine between 2013 and 2016, data were obtained and reproduced fully for 46% of trials [11]. In a sample of papers published in Science in 2011–2012, an estimated 26% were reproducible [8]. In 2015, publications in so-called “political behaviorist” journals, 22% could be fully reproduced [12]. Of a survey of registered reports in psychology literature published between 2014 and 2018, the main results were reproduced for 34%, though authors were not contacted if data or code were not publicly available [13]. In social learning literature published between 1955 and 2018, 23% were reproducible [14], and among 13 economics journals published between 2008 and 2013, key results were reproducible with assistance from authors in 43% of cases [9]. One reason for the lower rate of reproducibility in our case is the low rate of data sharing. In our sample, 28% of articles were published in the 1990s. Minocher et al. showed that in social learning research, there is a decline in data recovery by half every 6 years [14]. Whereas we attempted to contact authors using several methods, it is likely that some never received our communications because they are no longer active in research or have moved institutions. Nonetheless, this highlights structural barriers to data access that arise when responsibility for data sharing rests solely with individual authors, and the importance of public data sharing practices, or at minimum, long-term plans to maintain access to datasets. Additionally, insufficient documentation of data processing steps and statistical procedures can limit the ability to fully reproduce the analyses, even when data are shared. Reproducibility requires not only access to raw data, but also access to the exact data processing workflow, variable definitions, and detailed statistical code or procedures. Shared datasets should adhere to the FAIR (Findable, Accessible, Interoperable, Reusable) principles [44], accompanied by comprehensive data dictionaries. Furthermore, statistical syntax should clearly link data management and processing steps, including the management of missing data and data checking procedures, to the final analytical models with annotated scripts or notebooks [45]. To address these issues, data, data processing documentation, and statistical code—structured according to these standards - should be deposited in trusted repositories or maintained by institutions under clear and predefined access conditions. This would ensure that access does not depend on the availability or discretion of individual authors and would reduce the risk of selective data sharing. Access procedures should be transparent, consistent, and based on predefined criteria to ensure fairness and allow independent researchers to verify published findings while maintaining appropriate privacy protections. Journal, funder, and institutional policies also play an important role. Requiring data and code availability, along with complete documentation of data processing and statistical procedures, as a condition of publication or funding would substantially improve reproducibility. Another reason for low data sharing may be the observational study designs in our sample. Indeed, randomized controlled trials assigning participants to different levels of potato consumption over the long time periods required to observe CRC incidence would be potentially impractical. As a result, most evidence on this topic comes from observational studies. Observational studies historically have fewer requirements for data sharing compared to clinical trials [46], which are often subject to stricter funder and journal policies. This difference likely contributes to the lower rates of data availability and reproducibility observed in our sample. Our findings emphasize the lag in data sharing by observational studies behind clinical trials [47]. Implications for rigor and reporting tools Tools such as ROBINS-E, STROBE-Nut, and the Newcastle-Ottawa Scale provide structured approaches to evaluating methodological quality and reporting completeness. However, they do not include several essential elements required to ensure reproducibility and verifiability, such as public availability of raw data, analytical code, detailed documentation of data processing steps, or sufficient descriptions of statistical procedures necessary to independently reproduce results. The absence of these elements has substantial implications. In our sample, even studies that scored favorably on existing reporting and quality tools could not be reproduced or independently verified due to lack of access to data or incomplete documentation of analytic procedures. This demonstrates that compliance with existing rigor and reporting tools does not guarantee reproducibility or verifiability. That is, current guidelines for observational studies, as were the studies included in our sample, are incomplete with respect to reproducibility assessment and other aspects of transparency. To address some of these gaps, guidelines explicitly require reporting of public data and code availability statements, preregistration of analysis plans, and sufficiently detailed documentation of data processing and statistical methods to permit independent reproduction. Our findings may inform future updates to existing reporting guidelines and rigor assessment tools, particularly in nutrition research where reproducibility depends heavily on access to observational datasets and detailed analytical procedures. There are limitations to our work. We did not preregister our search or analytic plans due to the exploratory nature of our procedure development, but this is an element of rigor that should be done to increase transparency. We assessed rigor in four different ways, as each tool describes something different (risk of bias from methodological aspects of each study, a minimal set of what should be reported to interpret each study, a measure of quality, and our own additional criteria (preregistration, data and code sharing, etc.)). Each tool has strengths and limitations and inherent subjectivity, and there may be disagreement about how to interpret the results. Even more tools, such as GRADE [48], could be applied. Future work could standardize the use of these tools for the goals herein. Finally, within each publication, we only attempted to reproduce analyses for potato-CRC associations and thus cannot rule out errors in other aspects of the work. In conclusion, for research on the relationship between potatoes and CRC, improvements can be made to increase transparency through preregistration, data and code sharing, and to increase the reporting of key methodological aspects of the study designs. Existing study designs have common weaknesses in their ability to provide causality, in the measurement of potato consumption, and in their handling of missing data. Additionally, the credibility of the systematic reviews that have examined the relationship between potato consumption and CRC risk depends on the reproducibility and transparency of the underlying primary studies. Our findings indicate that the reproducibility and verifiability of this research is not adequate and therefore reduces the trustworthiness of its findings. This lack of trustworthiness reduces confidence in both individual study findings and pooled estimates derived from systematic reviews. Consequently, the current body of evidence on potato consumption and CRC risk should be interpreted with caution, and improvements in data sharing and reporting practices are needed to strengthen the reliability of nutrition research. To our knowledge, ours is the first survey of the reproducibility and verifiability of a specific research question, and the first of the same in nutrition research. Additional work should explore this in other areas of nutrition. Supplementary information Supplementary Information Acknowledgements We are grateful to librarian Thea Atwood at Indiana University for her assistance with search string translation. We thank Lilian Golzarri Arroyo for reviewing our decisions regarding the potential clustering issues in the papers. Author contributions CJV, YJ-N, BF, and DBA designed research; CJV, YJ-N, PK, BO, JM, LS, KS, SR, XT, BH, and SLD conducted research; CJV, YJ-N, PK, BO, BH, and SLD analyzed data or performed statistical analysis; CJV, YJ-N, and DBA wrote paper; DBA had primary responsibility for final content. All authors critically reviewed, edited, and approved the final manuscript. Funding This research was funded by the Alliance for Potato Research and Education. Supported in part by NIH grants R25DK099080 and R25HL124208. The assertions expressed are those of the authors and not necessarily those of the NIH or any other organization. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Data availability Data and code are publicly available here: https://osf.io/6jkwy/. The two datasets we received can be requested from the corresponding authors of the respective papers. We did not have any special access privileges that others would not have. Competing interests In the last 36 months prior to initial submission of this manuscript, DBA has received personal payments or promises for same from: Amin Talati Wasserman for KSF Acquisition Corp (Glanbia); General Mills; Law Offices of Ronald Marron; Novo Nordisk Foundation; VitaSoft; and Zero Longevity Science (as stock options). Donations to a foundation have been made on his behalf by the Northarvest Bean Growers Association. The institution of DBA (and BH), Indiana University, and the Indiana University Foundation have received funds or donations to support his research or educational activities from: Alliance for Potato Research and Education; American Egg Board; Eli Lilly and Company; Linus Technology; Mars, Inc.; National Cattlemen’s Beef Association; National Pork Board; Pfizer, Inc.; Ro; Takeda Global Research & Development Center, Inc; USDA; WW (formerly Weight Watchers); and numerous other for-profit and non-profit organizations to support the work of the School of Public Health and the university more broadly. BF is Partner & Technical Expert for The Science Angle which offers regulatory and scientific consulting services to the food industry and trade associations. In the 36 months prior to the initial submission, CJV has received honoraria from The Obesity Society, The Alliance for Potato Research and Education, National Cattlemen’s Beef Association, and American Pistachio Growers. The institution of CJV, Indiana University, has received funds to support his research from: National Cattlemen’s Beef Association; Alliance for Potato Research and Education; the Gordon and Betty Moore Foundation; and NIH. In the 36 months prior to the initial submission, YJ-N has received honoraria from The Alliance for Potato Research and Education and The University at Buffalo. At the time of submission of this manuscript, YJ-N is employed by Abbott Diabetes Care (ADC); the conduct of the study and the writing of this manuscript were completed prior to the start of her employment with ADC. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41387-026-00444-8. References Associated Data Supplementary Materials Supplementary Information Data Availability Statement Data and code are publicly available here: https://osf.io/6jkwy/. The two datasets we received can be requested from the corresponding authors of the respective papers. We did not have any special access privileges that others would not have.