The most consequential paper in this issue reports that curcumin is associated with a 49% higher incidence of type 2 diabetes. For an audience that recommends turmeric, that sentence is alarming, quotable, and almost entirely wrong — not because the statistic is incorrect, but because the word does not mean what a herbalist assumes it means.
The study examined curcumin as E100, an industrial food colorant, found in dyed dairy desserts, packaged sauces, and biscuits at a cohort-wide mean intake of 0.61 mg per day. That is roughly three orders of magnitude below a therapeutic curcumin dose. The same paper separately analyzed naturally occurring dietary curcumin from turmeric-containing foods and found no association at all — hazard ratio 0.95, with a p-value for trend of 0.60. One word, two exposures, opposite findings, in a single paper.
Once you notice that pattern, June is full of it. A systematic review reports that mangosteen improved periodontitis outcomes — as a 4% topical gel applied during scaling and root planing, not the oral antioxidant capsule sold under the same name. A meta-analysis reports rare sugars improving glycemic response — postprandially, with no effect on HbA1c for one of the two. A dementia analysis reports the MIND diet as significant and the Mediterranean and DASH diets as not, from point estimates of 0.78, 0.85, and 0.80 respectively, which is a statement about confidence intervals rather than about diets.
None of this is misconduct. Every one of these papers reports its own distinction accurately, usually in the methods section. The failure is downstream, in the compression that happens between a paper and a practice recommendation — and it is worth building a habit against, because the curcumin example shows how far a single ambiguous noun can travel.
Studies at a Glance
When a Familiar Name Means Something Else
Three papers whose findings attach to a specific compound, route, or timeframe that shares a name with something quite different — and where the substitution changes the clinical conclusion entirely.
Food additive exposure and metabolic disease is the subject of a sustained European research programme that has previously examined sweeteners, emulsifiers, and nitrites in the same cohort. This installment addresses colorants, a class that includes both synthetic dyes and natural pigments used industrially.
Prospective cohort within NutriNet-Santé (NCT03335644), 2009–2023. 108,723 participants, 79.2% female, mean age 42.5 ± 14.6 years, with 1,131 incident type 2 diabetes cases over a median 8.05 years and 841,296 person-years. Exposure was assessed by repeated 24-hour dietary records linked to composition databases and supplemented by laboratory assay of colorant content, with cumulative time-dependent exposure recalculated every two years.
Show full methodology
Of 34 identified colorants, 12 met a 10% consumption threshold for individual modeling. Most synthetic azo dyes were too rare in this population to test, so the significant findings are predominantly natural and caramel colorants. Cox models adjusted for age, sex, BMI, physical activity, smoking, education, family history, energy intake, alcohol, and diet quality, with ultra-processed food intake in the primary model — a meaningful check against the objection that this simply measures processed food. Results were reported stable after further adjustment for sweeteners, nitrites, emulsifiers, preservatives, and trans fats. False discovery rate correction was applied. Funded by the European Research Council, the French National Cancer Institute, and public bodies; no food-industry funding and no conflicts declared.
Caramel colorants showed the most consistent associations, with plain caramel at 1.46 and sulfite ammonia caramel at 1.30. Several natural pigments were also implicated: β-carotene 1.44, carotenoids 1.39, anthocyanins 1.40, paprika extract 1.26, lutein 1.20. The curcumin result carries the largest single point estimate and is the one most liable to catastrophic misreading. Additive curcumin represented only about a third of total curcumin exposure, at a cohort mean of 0.61 mg/day and just 2.43 mg/day among consumers — against therapeutic supplementation typically dosed at 500 to 2,000 mg of curcuminoids.
Strengths
- 108,723 participants with 841,296 person-years
- Repeated 24-hour records plus laboratory assay, not a single food-frequency questionnaire
- Adjusted for ultra-processed food intake in the primary model
- Stable after adjustment for other additive classes
- False discovery rate correction applied
- Publicly funded with no industry involvement
- Analyzed additive and natural exposure separately
Limitations
- Observational — no causal inference
- Self-selected volunteer cohort, 79.2% female and higher-educated
- Type 2 diabetes ascertained by self-report plus registry linkage, not biomarker
- Synthetic azo dyes too rare in this population to assess
- BMI adjustment may over-adjust for a mediator
- Residual confounding by broader dietary pattern cannot be excluded
The actionable message concerns ultra-processed foods carrying colorant additives, especially caramel colors, and not turmeric in any form. If a client asks whether they should stop taking curcumin because of this study, the answer is no, and the paper itself supplies the evidence: it measured natural dietary curcumin separately and found nothing. Because BMI was adjusted throughout, and colorant-heavy processed food plausibly drives weight gain that drives diabetes, these hazard ratios may if anything understate the pathway.
If one finding from this issue reaches a client in distorted form, it will be this one. “Curcumin linked to diabetes” will be read as an indictment of turmeric supplementation. The study says the opposite where turmeric is concerned, and the distinction is between a trace industrial dye in a packaged dessert and a gram-scale therapeutic extract.
Mangosteen (Garcinia mangostana) is sold widely as an oral antioxidant supplement on the strength of its xanthone content. Its periodontal evidence, by contrast, concerns local application to the periodontal pocket, where the pharmacological logic is entirely different: direct contact with inflamed tissue rather than systemic absorption.
Systematic review of 10 studies — a mix of randomized trials, cohort studies, and other clinical trials rather than randomized evidence alone — comprising 422 patients with periodontitis. The intervention was a 4% mangosteen gel used as an adjunct to scaling and root planing, compared against scaling and root planing alone or with alternative adjuncts. Outcomes were probing pocket depth, gingival index, clinical attachment level, and bleeding on probing, with inflammatory and antioxidant biomarkers as secondary measures.
Show full methodology
The authors declare no conflicts of interest and no grant funding is listed. No PROSPERO registration appears in the accessible record, and no formal risk-of-bias tool is named. The authors explicitly flag heterogeneity in formulations across the included studies, meaning the 4% gel was not necessarily the only preparation tested. Trial-level detail — individual study sizes, designs, blinding, follow-up, extract standardization, and xanthone content — was not available in accessible text. Total pooled sample of 422 across 10 studies gives a mean of roughly 42 per study, which is small for this indication. No I², publication bias assessment, or GRADE certainty rating was reported in accessible text.
The direction of effect is consistent across clinical and biochemical measures, and the biomarker changes — lower interleukin-6 and matrix metalloproteinase-9, higher total antioxidant capacity — are mechanistically coherent with the clinical improvements rather than merely co-occurring. What the review cannot support is any inference about oral mangosteen supplementation, which was not studied, and for which the pharmacokinetics of xanthone absorption make the extrapolation particularly unsafe.
Strengths
- Consistent direction across four clinical measures
- Biomarker changes coherent with clinical findings
- Adjunctive design against an active standard of care
- Addresses a real gap in periodontal adjunct options
- No conflicts declared
Limitations
- Only 422 patients across 10 studies
- Mixed study designs, not randomized evidence only
- Extract standardization and xanthone content not reported
- No I², publication bias, or GRADE reported
- Says nothing about oral supplementation
- No registration found and no risk-of-bias tool named
For a practitioner working alongside a periodontist, this supports 4% mangosteen gel as a reasonable adjunct to mechanical debridement, with the caveat that the evidence base is small and its standardization undocumented. What it does not support — and this is where the confusion will occur — is recommending oral mangosteen capsules for gum health, periodontal inflammation, or anything else. The route of administration is not incidental to the finding; it is the finding.
“Mangosteen improves periodontitis” will be read by a client, and possibly by a supplier, as a reason to take mangosteen capsules. The studied intervention was a gel placed in the periodontal pocket by a clinician. Nothing here speaks to swallowing the same plant.
Allulose and tagatose are rare sugars increasingly used as sweeteners, marketed on the promise of sweetness without glycemic cost. Whether that translates into anything beyond a blunted post-meal curve is the clinically relevant question.
Systematic review and meta-analysis of 20 controlled human intervention trials comprising 1,033 participants — 12 trials of allulose and 8 of tagatose. Outcomes spanned postprandial glucose and insulin responses, HbA1c, fasting glucose and insulin, blood lipids, and body composition. Certainty of evidence assessed by GRADE. Prospectively registered (PROSPERO CRD42024567433).
Show full methodology
Doses ranged from 2.5 to 25.5 g for allulose and 5 to 75 g for tagatose in acute trials, with longer-term trials running 4 weeks to 12 months. Risk of bias was low in 18 of 20 trials. Heterogeneity was substantial for the allulose insulin outcome at I² = 96%, weakening that estimate despite its moderate GRADE rating. Both tagatose longer-term findings rest on a single study each — the HbA1c result on one 12-month trial at 15 g/day, and the fasting insulin result on one 4-week trial at 45 g/day, the latter at very low certainty with an implausibly wide interval that should not be quoted. The review received no third-party funding; one author has extensive industry and data-monitoring relationships and sits on the journal’s editorial board, with recusal noted. Gastrointestinal tolerance was not assessed at all — confirmed absent from the review rather than merely unreported, which is a genuine gap given that it is the practical dose limiter for both sugars.
Both sugars reliably blunt the acute post-meal glucose and insulin response, at moderate certainty. The divergence matters clinically: tagatose reduced HbA1c by a quarter of a percentage point while allulose did not move it at all — though that tagatose result rests on a single 12-month trial, alongside no allulose effect on fasting markers, lipids, or body composition. A postprandial effect that does not register in HbA1c means the benefit is confined to individual meals rather than accumulating into changed glycemic control.
Strengths
- 20 trials and 1,033 participants
- Prospectively registered
- GRADE certainty applied per outcome; 18 of 20 trials at low risk of bias
- Allulose and tagatose analyzed separately rather than pooled
- Reports the null outcomes explicitly
Limitations
- I² = 96% for the allulose insulin estimate
- Both tagatose longer-term results rest on a single trial each
- Gastrointestinal tolerance not assessed at all — confirmed absent from the review
- One author has extensive industry relationships
- No third-party funding, but one author sits on the journal editorial board
These are reasonable sugar substitutes for clients who want sweetness with a blunted post-meal response, and tagatose has modest evidence of translating into slightly better HbA1c. Do not present either as a metabolic intervention: no effect on lipids or body composition, and for allulose no effect on HbA1c or fasting glucose. The practical limiter in real use is gastrointestinal tolerance, which these sugars share with other poorly absorbed carbohydrates and which this review did not assess at all — expect dose-dependent bloating and laxation, and titrate accordingly.
A blunted postprandial curve is not glycemic control. Allulose moved the post-meal response and left HbA1c, fasting glucose, lipids, and body composition untouched. The difference between “improves glycemic response” and “improves glycemic control” is the difference between a single meal and a client’s actual disease trajectory.
Significance, Superiority, and Silence
Two syntheses where the shape of the statistics invites conclusions the data do not license: three similar diets separated only by interval width, and thirteen dietary patterns whose lack of evidence is not evidence of lack.
The MIND diet was constructed specifically to target cognitive outcomes by combining Mediterranean and DASH principles with emphasis on berries and leafy greens. Whether it outperforms its parent patterns is the question practitioners actually face when advising clients.
Systematic review and meta-analysis of cohort studies, pooling hazard ratios for highest versus lowest adherence separately for Mediterranean, DASH, and MIND patterns. 12 studies contributing 14 cohorts, searched through April 2024 across four databases. Random-effects modeling with subgroup analysis and meta-regression. Prospectively registered (PROSPERO CRD42024532389).
Show full methodology
The number of cohorts contributing to each diet-specific estimate was not reported in accessible text — a consequential gap, since it determines whether MIND’s tighter interval reflects genuine consistency or simply fewer, more homogeneous contributing studies. I² per diet, publication bias testing, quality assessment scores, and GRADE ratings were likewise unavailable. Mediterranean subgroup analyses did reach significance under specific conditions: HR 0.72 (95% CI 0.55–0.93) when follow-up was ten years or less, and HR 0.63 (0.48–0.82) when NINCDS-ADRDA diagnostic criteria were used.
All three point estimates fall between 0.78 and 0.85 and all three point the same protective direction. MIND has the smallest sample and the narrowest interval, an unusual combination that in the absence of the per-diet cohort counts cannot be distinguished from fewer and more homogeneous contributing studies. The decisive external context is that the only randomized trial of the MIND diet — Barnes et al., New England Journal of Medicine, 2023, randomizing 604 participants over three years — found no significant cognitive difference against an active control diet, with both arms improving similarly.
Strengths
- Prospectively registered
- Three dietary patterns compared within one analysis
- Subgroup and meta-regression conducted
- Mediterranean subgroup findings reported transparently
Limitations
- Per-diet cohort counts not reported
- I², publication bias, and GRADE unavailable
- MIND’s narrow interval on the smallest sample is unexplained
- The randomized trial of MIND found no advantage over active control
- Observational adherence carries healthy-adherer bias
- Diagnostic criteria varied across cohorts
Present all three patterns as sensible for cardiometabolic health with a possible, unproven, dementia-specific increment — not as a hierarchy in which MIND is validated and the others are not. The statistical distinction being drawn here is about interval width, and the point estimates for Mediterranean and DASH are numerically comparable to MIND’s. A client who finds Mediterranean eating sustainable should not be moved to MIND on the strength of this analysis.
Reading “MIND significant, Mediterranean and DASH not” as “MIND works and the others don’t” is the classic confidence-interval fallacy. Their point estimates differ by 0.07. And the one randomized trial of MIND specifically found it no better than an active comparator, which is the kind of observational-versus-randomized discordance that usually signals healthy-adherer confounding.
“Anti-inflammatory diet” is among the most-used phrases in this field and among the least precisely defined. This umbrella review grades what the evidence actually supports across fifteen named dietary patterns.
Umbrella review of 30 systematic reviews and meta-analyses covering 225 primary studies and 15 dietary patterns, searched 1990 to March 2025 across five databases. Methodological quality assessed with AMSTAR-2 and certainty with GRADE. Primary-study overlap across reviews was quantified using corrected covered area. Prospectively registered (PROSPERO CRD42023472469).
Show full methodology
AMSTAR-2 ratings were sobering: 30% of source reviews high quality, 33% moderate, 27% low, and 10% critically low — meaning over a third of the underlying evidence base has major deficiencies by the authors’ own instrument. Overlap was slight overall at 1.4% but moderate-to-high specifically for vegan, Nordic, and vegetarian patterns, meaning pooled impressions for those risk double-counting the same trials. Inflammatory marker definitions and units were inconsistently reported across included meta-analyses.
The Mediterranean diet, appearing in 16 of the 30 reviews, showed the most consistent signal — significant CRP and IL-6 reductions and adiponectin increases, with certainty ranging from high to low depending on the analysis. Vegetarian patterns showed CRP reductions in about half of analyses at low to very low certainty, plus a fibrinogen reduction. The remaining thirteen patterns — including DASH, vegan, ketogenic, low-glycemic, Paleolithic, and low-FODMAP — were explicitly inconclusive or limited by a paucity of studies. Only the Western pattern was flagged as proinflammatory, on narrative-synthesis and low-certainty grounds.
Strengths
- 30 reviews spanning 225 primary studies and 15 patterns
- AMSTAR-2 and GRADE both applied
- Primary-study overlap formally quantified — rarely done
- Prospectively registered
- Reports the quality of its own inputs honestly
Limitations
- A third of source reviews rated low or critically low quality
- Moderate-to-high study overlap for three named patterns
- Inflammatory marker units inconsistently reported
- Generates no new pooled estimates of its own
- Cannot exceed the rigor of its inputs
When a client asks what an anti-inflammatory diet actually is, the defensible answer is Mediterranean, with vegetarian as a less certain second. That is a narrower claim than the phrase usually carries. The thirteen patterns without adequate evidence should be described as understudied rather than ineffective — the distinction matters for a client already succeeding on, say, a low-FODMAP protocol for other reasons. Any effect size taken from this review should be traced to its originating meta-analysis before being quoted, since this paper pooled nothing itself.
“Only the Mediterranean diet is anti-inflammatory” overreads this. Thirteen patterns were inconclusive because they have not been adequately studied, which is not the same as having been tested and failed. Equally, the Mediterranean numbers are not this review’s own findings — they are inherited from meta-analyses of varying and sometimes critically low quality.
Associations Pointing in Unknown Directions
Four studies where the finding is real and the direction of the arrow is not established — by design, by confounding, by surrogate endpoint, or by who funded the work.
Micronutrient status in ADHD is a persistent question in integrative practice, and the association between low ferritin and ADHD symptoms has been reported for two decades. What has never been resolved is the direction of the arrow.
Systematic review and meta-analysis of 46 case-control studies comprising 5,515 children and adolescents with ADHD and 8,166 controls, searched through March 2026 across three databases including CNKI. Quality assessed by Newcastle-Ottawa Scale with a mean of 6.9 of 9. Zinc contributed 36 studies, iron 23, ferritin 9, copper 24.
Show full methodology
Roughly 41% of included studies originate from China, reflecting the CNKI search. Primary studies span 1996 to 2025, crossing multiple DSM revisions without harmonized diagnostic criteria — a limitation the authors acknowledge. Specimen types were pooled across serum, plasma, whole blood, hair, and urine, which differ in reference ranges and assay characteristics. Most included studies did not control for dietary intake or stimulant medication use. Funding came from regional Chinese science and education bodies; no industry funding.
Ferritin was also lower at SMD −0.54 (95% CI −1.00 to −0.27). Heterogeneity was essentially maximal across every element, between 91.6% and 99.1%, meaning the pooled estimates summarize wildly disparate individual studies rather than a uniform signal. An SMD near −1.01 for zinc is also very large for a trace element outside frank malnutrition, a magnitude more consistent with heterogeneous small studies, mixed specimen types, and deficiency-endemic sampling than with a uniform biological effect. Differences were more pronounced in younger children and in samples from developing countries.
Strengths
- 46 studies with 5,515 cases and 8,166 controls
- Four elements assessed separately
- Newcastle-Ottawa quality appraisal applied
- Leave-one-out sensitivity analysis conducted
- Reports the copper null result
- No industry funding
Limitations
- Case-control design cannot establish direction of causality
- Heterogeneity between 91.6% and 99.1%
- Most studies did not control for diet or stimulant medication
- Diagnostic criteria unharmonized across three decades
- Specimen types pooled across serum, hair, and urine
- Developing-country subgroup likely reflects broader deprivation
Checking ferritin and possibly zinc as part of a general nutritional assessment in a child with ADHD is defensible, particularly given the established link between low ferritin and sleep or restless-legs symptoms. That is opportunistic screening, not an ADHD-specific test. Supplement to correct a documented deficiency, evaluate against actual symptoms, and do not frame it as ADHD treatment — the randomized evidence for zinc or iron supplementation improving global ADHD severity is weak and inconsistent, and NICE guidance advises against routine nutritional supplementation for this indication.
The tempting chain of reasoning — low zinc is found in ADHD, therefore deficiency contributes to ADHD, therefore supplementation should help — inverts a cross-sectional association into a causal treatment claim. ADHD is associated with restrictive and sensory-driven eating, and stimulant medication suppresses appetite. Either could lower measured nutrient status without deficiency having caused anything.
The EAT-Lancet planetary health diet was designed for environmental sustainability, and its metabolic effects are being characterized retroactively. The obvious objection to any such analysis is that people who adhere to a dietary index also adhere to calorie targets — so the useful question is whether this paper tested that.
Secondary analysis of two Spanish controlled dietary intervention trials, Obekit (NCT02737267) and FLiO (NCT03183193), comprising 403 adults with overweight or obesity, all assigned a 30% energy deficit. Adherence to the Planetary Health Diet Index was related to BMI, waist circumference, and body fat percentage at 3 and 6 months using linear mixed-effects models with random intercepts, plus mediation analysis with 1,000 bootstrap simulations.
Show full methodology
The energy-intake objection was tested directly and answered. The authors report that adjusting for total energy intake did not alter results, because the adherence index is energy-normalized by construction — intakes are expressed per 1,000 kcal. Effect sizes were stable across four nested models adding sex, age, and intervention group. That does not eliminate residual confounding by general adherence behaviour, but it does rule out the simplest version of the objection. Attrition was substantial: 403 enrolled and 259 completed, a 35.7% dropout the authors concede may bias associations upward. Funded by the Government of Navarra, CIBERobn, Fundació la Marató de TV3, and a Novo Nordisk Foundation grant, with in-kind support from two Spanish food companies; authors declare no conflicts.
The quartile relationship was graded rather than threshold-like, with the third quartile also separating from the first (−0.65 kg/m², p = 0.031). Mediation analysis identified reduced red and processed meat as the strongest single contributor, followed by whole grains (ACME −0.23, p = 0.020); fruit, vegetables, fish, nuts, eggs, and fats were not significant mediators. One counterintuitive finding deserves attention: higher baseline adherence was inversely associated with achieving 10% weight loss (OR 0.96 per point, p = 0.011), while improvement in adherence predicted it. Those already eating well had less room to move.
Strengths
- Energy intake explicitly tested as a confounder and reported as non-altering
- Index is energy-normalized by construction
- Nested within two controlled trials rather than free-living observation
- Effect sizes stable across four nested models
- Mediation analysis identifies specific food groups
- Authors declare no conflicts
Limitations
- 35.7% attrition, which the authors concede may inflate associations
- All participants were in an assigned 30% deficit — says nothing about unrestricted eating
- Residual confounding by general adherence behaviour remains possible
- Secondary analysis of trials designed for other questions
- Two Spanish cohorts — generalizability untested
This is more useful than a typical adherence-index analysis because it anticipated the obvious objection and tested it. For a client already committed to an energy deficit, structuring that deficit around less red and processed meat and more whole grains tracks with better body composition, independently of how many calories they end up eating. The baseline-adherence finding is worth carrying into practice too: clients already eating well should expect smaller changes, which is a matter of expectation-setting rather than failure.
The finding still lives entirely inside a 30% energy deficit. Nothing here speaks to what the pattern does for someone not restricting, and the 36% dropout means the completers may be systematically better adherers than the enrolled sample.
Weight regain after successful diet-induced loss is among the most clinically frustrating problems in practice, and almost nothing reliably prevents it. Pasteurized Akkermansia muciniphila has a mechanistic rationale and a 2019 pilot trial behind it, making this the first adequately sized test of the maintenance question.
Double-blind, placebo-controlled randomized trial at Maastricht UMC+, registered NCT05417360. An 8-week low-energy diet requiring at least 8% weight loss to qualify, followed by 24 weeks of ad libitum maintenance with daily supplementation. 90 participants randomized against a planned enrollment of 108. Eligibility was adults 20–70 with BMI 28 to under 40, excluding type 2 diabetes, gastrointestinal disease, cardiovascular disease, cancer, and recent antibiotic or probiotic use.
Show full methodology
Four authors, including the senior and corresponding author who serves as a paid scientific advisor, have financial, employment, or patent ties to The Akkermansia Company. The formal funding source could not be confirmed — company materials describe it as company-funded while the registry lists the university as sponsor. The exact CFU dose, placebo composition, and intention-to-treat handling could not be verified. No confidence intervals appear anywhere in the accessible reporting, only standard deviations. The authors concede two limitations: the short 24-week window, and the absence of a heat-inactivated comparator arm that would isolate an active-ingredient effect.
The primary outcome favours the supplement by roughly two kilograms of avoided regain, which is clinically meaningful if it holds. The reporting is thinner than the venue would suggest: no confidence intervals, an unexplained 17% enrollment shortfall, and an unverifiable dose. Independent commentary characterized the difference as modest and the sample as small, cautioning against practice change. A separate larger trial from the same sponsor group had a null primary endpoint with positive findings only in an exploratory low-baseline-Akkermansia subgroup — a pattern that should raise rather than lower the bar for the cardiometabolic-responder claim reported here.
Strengths
- Addresses weight regain, a genuine and underserved clinical gap
- Double-blind and placebo-controlled
- Qualifying weight loss required before randomization
- 24-week maintenance period
- No treatment-related serious adverse events
- Authors concede the missing inactivated-strain arm
Limitations
- Four authors have financial ties to the manufacturer
- No confidence intervals reported anywhere
- 90 randomized against a planned 108, unexplained
- Exact dose and placebo composition unverifiable
- No heat-inactivated comparator to isolate the active ingredient
- A larger trial from the same group had a null primary endpoint
- Single site, 24 weeks
This is a promising signal in a problem area with few options, and it is not yet a recommendation. The trial used a powder; commercial capsule products cannot be assumed equivalent when the trial dose itself is unverifiable. If a client raises it, the honest framing is that one small manufacturer-connected trial found about two kilograms less regain over six months, that a larger trial from the same group missed its primary endpoint, and that independent replication has not happened.
A Nature Medicine masthead invites more confidence than the reporting supports. Ninety participants against a planned 108, no confidence intervals, an unverifiable dose, no inactivated-strain control, and a third of the authorship financially connected to the product is a combination that belongs in the promising-but-unreplicated category.
Vitamin D receptors are expressed throughout the intestinal epithelium, giving a plausible mechanism for supplementation to influence gut microbial ecology. Whether that influence matters clinically is a separate question this review was not designed to answer.
Systematic review of 14 randomized controlled trials comprising 1,458 participants, examining the effect of vitamin D supplementation on human gut microbiota composition and diversity. Prospectively registered (PROSPERO CRD42024628080). Outcomes included taxonomic abundance, alpha and beta diversity, and selected biomarkers including calprotectin and TMAO.
Show full methodology
Searched January 2015 to November 2024 across four databases by a team based principally at the University of Sharjah. Dose ranges, supplementation durations, baseline vitamin D status distributions, sequencing methodology, risk-of-bias results, and funding and conflict disclosures were all unavailable in accessible text. Diversity findings were described as variable rather than consistent. Worth noting for context: this review’s positive framing sits in some tension with larger primary trials, including a 2019 randomized trial in overweight adults and the large Australian D-Health microbiome sub-study, both of which reported largely null effects on community-level composition. Whether the review engages that tension could not be determined.
The taxonomic direction was reasonably consistent: increases in Bifidobacterium and Lactobacillus, both conventionally regarded as beneficial, and a shifted Bacteroidetes to Firmicutes ratio. Diversity effects varied by dose, duration, and baseline vitamin D status. Every one of these outcomes is a surrogate. The review reports no clinical endpoint — no symptom score, no disease incidence, no functional measure — and the relationship between a shifted abundance ratio and any patient-relevant outcome remains unestablished.
Strengths
- Restricted to randomized trials rather than observational data
- 1,458 participants across 14 trials
- Prospectively registered
- Reports dose and baseline-status dependence rather than a single pooled claim
- Includes biomarker outcomes alongside taxonomy
Limitations
- All outcomes are surrogate measures
- No clinical endpoint assessed
- Dose ranges and durations not available
- Diversity findings inconsistent across trials
- Sequencing methodology heterogeneity likely
- Sits in tension with larger null primary trials
If a client is supplementing vitamin D for an established indication, this offers a plausible ancillary benefit to gut ecology. It is not a reason to start vitamin D for gut health, and it should not be presented as one. The dose and baseline-status dependence is the practically useful finding: effects appear contingent on starting status, which fits the general pattern that correcting deficiency does more than supplementing sufficiency.
Increased Bifidobacterium is a laboratory result, not a health outcome. No trial in this review measured whether anything happened to the participants as a consequence. Treating a favourable taxonomic shift as a demonstrated benefit skips the step where the benefit has to be shown.
Synthesis & Emerging Themes
One Word Can Carry Three Orders of Magnitude
The curcumin finding deserves to become a teaching case. A single paper reported a hazard ratio of 1.49 for curcumin as a food colorant and 0.95 for curcumin from turmeric-containing foods, and the difference between those exposures is roughly a thousand-fold in dose and entirely different in matrix. The paper handled this correctly. The risk lives downstream, in the compression to a headline, and it is not hypothetical — a practitioner who reads “curcumin linked to diabetes” and de-prescribes turmeric has acted against the paper’s own evidence. The same structure appears twice more this month: mangosteen as a gum-applied gel rather than a capsule, and rare sugars improving a post-meal curve rather than glycemic control.
Absence of Evidence Keeps Getting Reported as Evidence of Absence
Two of this month’s syntheses turn on the same statistical misreading. The dementia meta-analysis found hazard ratios of 0.78, 0.85, and 0.80 for MIND, Mediterranean, and DASH, called one significant and two not, and thereby invited a hierarchy that a 0.07 spread in point estimates cannot support. The anti-inflammatory umbrella review found thirteen of fifteen dietary patterns inconclusive — which reflects how little they have been studied, not that they were tested and failed. In both cases the honest summary is less satisfying and considerably more accurate than the one that will circulate.
Where the Arrow Points
The trace-element and ADHD meta-analysis is the clearest instance this month of an association whose direction is simply unknown. Children with ADHD had markedly lower zinc, iron, and ferritin. They also, as a population, eat more restrictively and are more often on appetite-suppressing stimulants. A case-control design cannot distinguish a nutritional contribution to the disorder from a nutritional consequence of it, and heterogeneity near 99% means the pooled estimate is not summarizing a stable phenomenon in any case. The same directional problem, in gentler form, runs through the Planetary Health Diet analysis, where every participant was already in a 30% energy deficit.
What Remains Unresolved
Whether pasteurized Akkermansia genuinely prevents weight regain is a live and important question that one 90-person manufacturer-connected trial without reported confidence intervals cannot settle, particularly given that a larger trial from the same group missed its primary endpoint. Whether vitamin D’s consistent effect on gut taxonomy translates into anything a patient would notice is unaddressed by any trial in this review. And whether the MIND diet has cognitive properties beyond its Mediterranean and DASH components remains, after this analysis and a null randomized trial, genuinely open.
“Before repeating a finding, check that the noun in your sentence means what it meant in theirs.”
This closes a four-month sequence of these reviews. The habits that recur across all four — read past the conclusion sentence, find the interval, ask who the study was about, check what the words denote — are less a critique of the literature than a description of what reading it carefully requires.