This critical narrative review examines whether studies distinguish gluten-specific effects from the consequences of food substitution. The initial 16-reference set was the working scoping corpus available at the start of the review rather than the product of a formal systematic search. It comprised core background sources and previously identified studies addressing gluten-free or low-gluten diets, healthy or non-celiac populations, microbiome outcomes, nutritional composition, and cardiometabolic questions. This corpus was then expanded through targeted searches updated on 15 October 2026 using web-based retrieval of PubMed-indexed records, publisher-hosted articles, and accessible full-text material. Searches combined gluten-free or low-gluten diets with healthy adults, randomized trials, microbiome, cardiovascular outcomes, inflammatory biomarkers, nutritional composition, and trace elements. Searches for 2024-2026 publications updated the evidence. Selected executed search strings appear in Supplementary material S1. The approach was selective, rather than an exhaustive search of multiple bibliographic databases.

Priority was given to original human interventions in adults without known gluten-related disorders and to prospective cohorts without known celiac disease. Selection required relevance to the review question, an identifiable dietary exposure or comparator, and interpretable human outcomes; higher cardiometabolic risk was distinguished from absence of disease or symptoms. Mixed-age cohorts, disease-population reviews, and product surveys were retained only as explicitly indirect context. Systematic reviews contextualized primary reports rather than being counted as independent replications. Disease-specific treatment studies, isolated recipe-development studies, and retracted intervention evidence did not support conclusions about healthy adults. Reports sharing a trial registration or cohort were linked to avoid duplicate counting. Ascertainment of gluten-related disorders varied across the included interventions. Most reports defined eligibility through absence of known or diagnosed celiac disease or gluten-related disorders, clinical history, self-report, or study-specific exclusion criteria rather than uniform diagnostic screening with celiac serology and/or biopsy. Accordingly, this review refers to adults without known gluten-related disorders and does not assume that every participant had diagnostically confirmed absence of such disorders.

The synthesis recorded population, diet definition, duration, comparator, sample size, principal outcomes, and design limitations. Numerical results were checked against retrieved article text, tables, or indexed abstracts. Inaccessible full texts were not treated as fully assessed reports. For Campagna and Henriques, interpretation remained limited to the indexed abstract and, for Campagna, the trial extraction in the Cochrane review. Detailed estimates unsupported by retrieved material were omitted. This access limitation is part of the review’s appraisal.

ChatGPT (OpenAI) assisted targeted source retrieval, synthesis drafting, and language editing. Numerical statements were checked against retrieved sources. This assistance did not provide independent human duplicate screening or constitute formal risk-of-bias assessment.

Evidence was organized by the question each design answers. Randomized dietary comparisons address an assigned dietary strategy. Uncontrolled before-after studies establish temporal changes but do not isolate intervention effects. Habitual-intake cohorts provide associations rather than tests of adopting a strict gluten-free diet. Product comparisons characterize availability and formulation. Differences across populations, exposures, and outcome methods precluded a meaningful pooled estimate in this narrative synthesis. No independent duplicate screening, formal risk-of-bias instrument, or GRADE assessment was undertaken. Appraisal below is design-based and qualitative. Table 1 summarizes selected interventions, Table 2 compares product surveys, and Table 3 defines the interpretive boundaries of each evidence stream.

Table 1.Selected human interventions and their applicability.

Selected reports are shown, not an exhaustive trial inventory. Adult, higher-risk, and mixed-age populations are distinguished. Sample sizes vary by assay. The Mannila reports describe one trial. BMI, body mass index. CRP, C-reactive protein. FMD, flow-mediated dilation. LDL, low-density lipoprotein cholesterol. qPCR, quantitative polymerase chain reaction.

Table 2.Selected comparisons of packaged gluten free and gluten containing products.

Values are reported in GF versus GC order. GF, gluten-free. GC, gluten-containing. Comparisons are within each study. Categories, summary measures, comparator selection, and countries differ, so values must not be pooled or interpreted as a global average. Historical product composition does not describe every current brand.

Table 3.Evidence streams and the conclusions they support.

This interpretive framework is developed for the present narrative synthesis. It is not a formal risk-of-bias instrument or certainty grading system.

A strict gluten-free diet, a low-gluten diet, and a low estimated intake within an otherwise unrestricted population are different exposures. For example, the Hansen trial compared approximately 2 with 18 g/day, rather than complete exclusion with usual intake [4]. It is inappropriate to relabel such a contrast as evidence for a strictly gluten-free diet.

Food replacement introduces a second distinction. Replacing wheat bread with refined rice-starch bread changes more than gluten. Replacing it with a combination of legumes, buckwheat, nuts, and other foods produces a different nutritional comparison again. Total fiber matching also does not ensure matching of fiber type or fermentability. The analytical question is therefore not simply whether gluten was removed, but which components changed alongside it and which remained comparable.

For this review, evidence is considered directly relevant when participants lack a known gluten-related disorder and the intervention tests dietary restriction. Evidence is considered indirect when it comes from disease populations, product labels, or habitual-intake categories. These categories describe applicability, not a numerical quality score. Even an internally valid result from a disease population does not automatically answer a prevention question in healthy adults.

Gluten free interventions

De Palma et al. [5] studied ten healthy adults before and after one month of a gluten-free diet. Quantitative polymerase chain reaction (qPCR) indicated lower Bifidobacterium, Lactobacillus, and B. longum counts, with higher Enterobacteriaceae and Escherichia coli counts. Polysaccharide intake also decreased. Fecal samples stimulated less cytokine production in an ex vivo peripheral-blood-mononuclear-cell assay. This assay measured the immune-stimulatory properties of samples, not circulating inflammation or immune failure in participants. The small, uncontrolled design prevents attribution of these changes specifically to gluten removal.

Bonder et al. [6] analyzed 21 volunteers aged 16 - 61 years during a four-week gluten-free period and a later habitual-diet period, with a five-week washout. This is mixed-age contextual evidence, not an adult-only estimate. Eight taxonomic groups differed, including lower Veillonellaceae abundance, but alpha-diversity indices and measured gut-health biomarkers did not differ between diets. Interindividual variation exceeded diet-associated variation. Functional pathways were inferred computationally from marker-gene data rather than measured as metabolic flux. The findings support a microbial response to dietary change without establishing clinical harm.

Low gluten interventions

Hansen et al. [4] randomized 60 Danish adults to two eight-week dietary periods separated by a washout of at least six weeks. Participants had no known chronic disorders but met criteria involving elevated body mass index (BMI) and/or waist circumference, which narrows applicability to lean young adults. Compared with the high-gluten period, the low-gluten period altered selected taxa, reduced hydrogen exhalation, and improved reported bloating. The authors proposed qualitative differences in dietary fiber as an explanation. Because wheat-containing foods also provide fermentable fructans, reducing wheat intake can lower fermentable substrate delivery to the colon; this offers a plausible explanation for lower hydrogen exhalation and bloating that is not specific to the gluten protein [4]. Because dietary components changed together, the study does not quantify the independent contribution of fructans, other fiber fractions, or gluten.

Delmas et al. [7] reported microbiome outcomes within a registered trial. Forty adults provided samples at baseline and after eight weeks of low-gluten intake, and the protocol prospectively assigned a subgroup of 20 participants to continue the low-gluten diet for a further eight weeks. The 20 participants assessed at week 16 therefore represent a planned intervention-duration subgroup rather than loss to follow-up from an intended 40-person week-16 assessment. The microbiome report principally compares time points after a dietary switch, rather than a sustained gluten-containing parallel control. It reported lower richness and changes in Bifidobacterium, Akkermansia, cultured Enterobacteriaceae, and fecal metabolites. These endpoints were not accompanied by evidence of incident disease. Sequencing analyses used a smaller subset than the qPCR analyses, so the total enrolled sample should not be assigned to every assay.

Clinical meaning and consistency

Across these studies, the most defensible common finding is alteration of selected microbial features after changing cereal intake. A consistent clinical syndrome has not been demonstrated. The decrease in Lactobacillus in the earliest study is not an established universal response. Similarly, a change in Enterobacteriaceae does not identify pathogenic strains or establish infection. Richness, relative abundance, absolute counts, and culture-based enumeration answer different questions and should remain distinct when comparing studies [5]-[7].

Kaliciak et al. [8] found that a gluten-free diet did not fully normalize microbial patterns in celiac disease. This provides disease-management context, not a direct harm estimate for healthy adults. Disease activity, mucosal injury, treatment response, and dietary restriction coexist in that setting. Microbial findings require linkage to validated measures of intestinal function or symptoms before taxonomic direction is treated as a health outcome.

Replacement cereals and the OAT GUT BRAIN trial

Mannila et al. [9] compared oat-rich with rice-rich low-gluten diets for six weeks in adults with cardiometabolic risk. The full report describes 72 randomized participants and 69 analyzed participants, despite the abstract’s shorthand description of n = 69. The study was powered using serum total cholesterol as the endpoint, while multiple dietary, clinical, gastrointestinal, and biochemical variables were analyzed. LDL cholesterol fell by 0.41 versus 0.17 mmol/L in the oat and rice groups, respectively (group-by-time p = 0.047). The report used a two-sided p < 0.05 threshold for these analyses and did not report a multiplicity adjustment across the broad set of clinical and dietary outcomes; the LDL result should therefore be interpreted as a nominal secondary/exploratory finding rather than as a single confirmatory endpoint. Constipation favored oats, whereas waist reduction favored rice. Both groups restricted gluten, so the contrast tests replacement strategies, not the effect of gluten removal. The per-protocol analysis and higher-risk population further constrain generalization.

The 2026 secondary report from the same registered trial (NCT05526092) examined prespecified study domains including fecal microbiome, plasma short-chain fatty acids (SCFAs), and inflammatory markers [10]. A greater rise in plasma butyrate was reported with oats (interaction p = 0.033); SCFA analyses used the nominal p < 0.05 threshold. For metagenomic and inflammatory-marker analyses, the report applied Benjamini-Hochberg false-discovery-rate correction. The Shannon-diversity diet-by-time interaction remained significant after this correction (q = 0.025), whereas no significant inflammatory-marker changes remained after correction. These linked reports arise from one experiment and contain multiple endpoints, so nominal lipid and SCFA findings should not be interpreted as equivalent in evidential status to multiplicity-corrected microbiome or inflammatory results. They show why microbial, metabolic, inflammatory, and clinical endpoints require separate interpretation.

Nutritional quality and affordability of replacement foods

Product surveys show recurring disadvantages in some gluten-free categories, particularly protein and fortification, but also demonstrate category-specific exceptions. In the Canadian survey of 398 gluten-free and 445 gluten-containing products, staple foods contained median protein values of 7.0 versus 10.9 g/100g. Energy and sugar differences in the staple category were not statistically significant [11]. A 2025 Canadian update compared 581 products in each group and reported iron fortification in 5% of gluten-free versus 46% of comparator staples in 2024 [12]. These findings concern food availability, not measured deficiency in consumers.

The Norwegian survey included 423 gluten-free and 337 comparator products. Overall protein and fiber were lower, but bread fiber was similar between groups. Importantly, investigators preferentially selected gluten-containing comparators meeting dietary guidance or carrying a Nordic nutrition label, and often selected whole-grain breads. This choice informs the nutritional comparison but limits a simple attribution to gluten status. The comparator was not an unrestricted sample of all conventional foods [13].

The Austrian study found lower protein in several categories and substantial price premiums [14]. In Slovenia, gluten-free-labelled foods often contained less protein and sugar, with differences depending on category [15]. These results challenge blanket statements that gluten-free foods always contain more sugar or less fiber. Median differences across mixed product groups also depend on the composition of the sampled market.

An Australian survey of 3213 products adds an important qualification. Gluten-free breads and breakfast cereals did not have significantly different mean Health Star Ratings from comparators, while gluten-free pasta scored lower. Protein was lower across the core categories [16]. Similar composite scores alongside nutrient-specific differences show why the selected outcome changes the nutritional interpretation. A survey does not establish clinical benefit or harm from eating those products.

The practical inference is to evaluate replacement foods by their complete composition, portion size, cost, and contribution to the total diet. Label-based surveys cannot establish habitual intake, bioavailability, or nutritional status. Their strongest role is to identify potential substitution problems and priorities for reformulation. Affordability is also relevant to adherence when the diet is medically necessary, rather than an argument against treatment [12].

Diet quality and circulatingbiomarkers

Henriques et al. [17] studied 23 women in a single-blind, non-randomized crossover experiment, using gluten-containing or gluten-free muffins against a gluten-free dietary background over six weeks. Body composition, resting energy expenditure, and gastrointestinal symptoms did not differ between the three-week periods. The gluten-free period, compared with the habitual baseline diet, involved lower fiber and several B-vitamin intakes and a higher Dietary Inflammatory Index. That index summarizes dietary composition. It is not a direct measure of systemic inflammation, and the baseline comparison differs from the within-experiment gluten comparison.

Lange et al. [18] followed 27 healthy adults, predominantly women, through four weeks of gluten avoidance without a parallel control. Leukocyte counts decreased from 5.24 to 5.00/nL (p = 0.0135). C-reactive protein (CRP) fell numerically from 0.89 to 0.62 mg/L, but the reported p value was 0.0689. Flow-mediated dilation also did not reach statistical significance (p = 0.0897), and quality of life did not improve. Despite the abstract’s anti-inflammatory framing, these data do not establish a CRP benefit. The absence of a control and multiple biomarker measurements further constrain inference.

These studies are not direct contradictions. One evaluates dietary composition and a calculated index, while another measures blood variables after an uncontrolled intervention. They support neither a general pro-inflammatory claim nor a general anti-inflammatory claim for gluten restriction in healthy adults. A controlled effect on a prespecified circulating endpoint would be more informative than an isolated within-group change.

Body weight and cardiometabolic risk factors

Campagna et al. [19] enrolled 60 healthy adults in a six-month dietary comparison. The abstract reported favorable changes in selected hematological, lipid, and psychological measures. In contrast, the Cochrane synthesis found no clear between-group differences for the trial’s blood pressure, low-density lipoprotein (LDL) cholesterol, or BMI outcomes, with considerable imprecision [20]. The reports concern different endpoints and comparisons. Within-group improvement should not be substituted for a treatment effect relative to the control group.

Xin et al. [21] pooled 27 articles involving participants with and without celiac disease. The overall weight estimate was 1.20 kg (95% confidence interval [CI] −1.16 to 3.55), which does not establish either a weight-loss benefit or equivalence. Because populations were mixed, it is not a healthy-adult effect estimate. Barone et al. [22] examined weight outcomes in celiac disease specifically. Changes after treatment in that population occur in a setting of altered disease activity and absorption, and should not be used to predict weight change after elective restriction in healthy adults.

Cardiovascular disease and diabetes

Lebwohl et al. [23] followed 110,017 adults without known celiac disease in two US cohorts. The reported hazard ratio of 0.95 (95% CI 0.88 - 1.02) compared participants in the highest fifth of estimated gluten intake with those in the lowest fifth. Cox models were conditioned on age and follow-up cycle and, in the pooled analysis, stratified by cohort. The multivariable model adjusted for demographic, anthropometric, clinical, medication, lifestyle, and reproductive covariates, including race, BMI, height, diabetes, aspirin/non-steroidal anti-inflammatory drug use, statin and multivitamin use, alcohol, smoking, parental coronary-heart-disease history, hypertension, hypercholesterolemia, physical activity, and menopausal status/hormone use, with additional dietary covariates in the fully adjusted estimate. The association remained null. Further adjustment for whole grains produced a hazard ratio of 1.00 (95% CI 0.92 - 1.09), whereas adjustment for refined grains produced 0.85 (95% CI 0.77 - 0.93), illustrating the dependence of interpretation on grain-source covariation. This was a study of habitual intake, not random assignment to a strict gluten-free diet.

Zong et al. [24] analyzed three US cohorts and reported an inverse association between gluten intake and type 2 diabetes. After additional cereal-fiber adjustment, the highest versus lowest quintile hazard ratio was 0.87 (95% CI 0.81 - 0.93). Gluten intake was strongly correlated with other carbohydrate components. The result does not establish a protective effect of gluten, and these cohorts overlap with those used in the coronary analysis. Their participant totals should therefore not be added as independent populations.

The Cochrane review identified one randomized trial and three observational studies for primary cardiovascular prevention and judged important results to have low or very low certainty [20]. This supports restraint in clinical recommendations. It does not justify asserting either proven cardiovascular benefit or proven cardiovascular harm from elective gluten avoidance.

Trace element exposure

Bulka et al. [25] analyzed National Health and Nutrition Examination Survey data from 7471 participants, including 73 reporting a gluten-free diet. Adjusted geometric mean ratios were 1.5 (95% CI 1.2 - 2.0) for total urinary arsenic and 1.7 (1.1 - 2.4) for total blood mercury. The mixed-age sample was not restricted to healthy adults, although an adult-only sensitivity analysis was reported as similar. The paper also examined arsenic estimates excluding selected organic species, so describing it as measuring total arsenic alone would be inaccurate. Cross-sectional exposure biomarkers did not establish dietary causation or clinical toxicity.

Punshon and Jackson [26] measured essential and potentially toxic elements in gluten-containing and gluten-free foods and identified differences related in part to rice-based ingredients. Together, these studies provide a rationale for examining replacement grain sources. They do not demonstrate that removing gluten directly increases metal exposure. Arsenic speciation, serving size, consumption frequency, seafood intake, and other sources are necessary for interpreting an individual’s exposure. Product concentrations and biomarker measurements should not be translated directly into clinical risk without an exposure assessment.

The evidence is more coherent when three linked processes are separated: restriction of gluten-containing foods, substitution with other foods, and downstream biological or clinical responses. Current whole-diet experiments often modify the first two processes together. An effect of the assigned diet therefore does not automatically identify an effect of the gluten protein. This distinction applies equally to apparently favorable and unfavorable findings.

For example, microbial substrate availability and reported bloating may change together while the clinical meaning of the microbial shift remains uncertain. A nutrient-poor substitute may lower dietary quality, while an alternative pattern built around nutrient-dense foods produces a different comparison. These possibilities are not evidence that outcomes depend entirely on replacement. They explain why isolating gluten requires controls beyond a gluten-free label.

The proposed framework organizes what each comparison identifies. The oat-versus-rice trial supplies an empirical example of outcome differences between two low-gluten strategies [9][10]. It does not establish that replacement explains every reported effect or quantify an independent gluten effect. A complementary design would reintroduce purified gluten against a matched dietary background. Comparing that estimate with a whole-food substitution trial would help distinguish the protein-specific question from the practical consequences of changing a diet.

Routine gluten restriction for general health improvement is not supported by the selected evidence in adults without a medical indication. This conclusion does not apply to patients requiring treatment for celiac disease, and it does not exclude individualized dietary assessment for symptoms. Diagnostic evaluation should precede self-directed elimination when a gluten-related disorder is suspected, because assessment and management depend on the clinical context [2].

For people already following the diet, nutritional assessment should focus on the replacement pattern: fiber sources, protein, fortification, food variety, and affordability. A gluten-free pattern does not inherently require dependence on refined starch substitutes. Recommendations should address demonstrated dietary gaps rather than presume deficiency or toxicity from the label alone [3][11].

A useful next generation of trials would compare at least two questions separately: a matched gluten challenge assessing a protein-specific effect, and a whole-diet substitution strategy assessing real-world effectiveness. Energy, macronutrients, total fiber, fiber type, whole-grain exposure, and processing should be measured rather than assumed equivalent. A factorial design would help separate gluten from replacement quality when feasible.

Primary outcomes should be prespecified and selected for the hypothesis, with sample-size calculations based on the between-group contrast. Inflammation studies should distinguish dietary indices from blood markers. Microbiome studies should report assay-specific sample sizes, repeated-measures analyses, multiple-testing control, and absolute abundance where feasible, alongside functional or clinical outcomes. Demonstrating a taxonomic change alone is insufficient.

Short-term mechanistic trials and longer-term clinical studies serve complementary purposes. The former require adherence monitoring and dietary characterization. The latter require sufficient duration, retention, and endpoint assessment to evaluate sustained benefit or harm. Future studies should also report nutrient adequacy, cost, symptoms, and adverse events, while specifying whether effects differ by baseline diet, sex, age, or metabolic status. These are research priorities, not findings established by the present review.