Every month this review asks the same question of the literature: what changed, and what should a practitioner do differently on Monday morning? March 2026 answered that question in an unusual way. Ten studies made the final cut from 403 unique papers, and in eight of them the sentence a reader would quote is not the sentence the data support.
This is not a complaint about bad science. Most of these papers are methodologically sound, and several are genuinely excellent — a unified reanalysis of raw microbiome sequence data from eleven countries, a Bayesian network meta-analysis spanning 167 randomized trials, a prespecified ancillary study nested inside one of the largest supplement trials ever run. The problem is narrower and more interesting than sloppiness. It is that the structure of a modern abstract — background, methods, results, a single declarative conclusion — systematically rewards the most quotable finding over the most accurate one. A meta-analysis reports that 62% of inflammatory bowel disease patients achieved remission on a Mediterranean diet, and reports in the same abstract that the diet showed no advantage over control diets whatsoever. Both sentences are true. Only one will travel.
Two threads run through this month’s selections. The first is the sponsorship question in botanical research, which arrives in an unusually stark form: an ashwagandha trial whose named authors include employees of the company that owns the product, the company that manufactures the extract, and the contract research organization that appears to have run the study. The second is the widening gap between surrogate endpoints and clinical ones, most visible in a Nature Medicine paper reporting that a daily multivitamin slowed two of five epigenetic aging clocks — nested inside a parent trial that found no effect on either invasive cancer or cardiovascular disease. Neither thread is new. Both got sharper in March.
What follows treats each study at the level of detail required to see where the headline and the data part company. Where a number could not be verified against full text, it is marked as such rather than smoothed over. Several of this month’s papers sit behind paywalls that made methods-level verification impossible, and that limitation is stated where it applies rather than hidden.
Studies at a Glance
The Sponsorship Question in Botanical Research
Two of this month’s herbal papers are, on their face, good news for practitioners: a well-designed ashwagandha trial with a biomarker endpoint, and a network meta-analysis of twenty curcumin and boswellia trials. Read at the level of funding and formulation, both become more complicated — not less useful, but useful in a narrower and more specific way than their conclusions suggest.
Ashwagandha (Withania somnifera) has one of the larger randomized evidence bases in Western herbal practice for stress and anxiety. That base is fragmented across distinct proprietary extracts — KSM-66, Sensoril, Shoden — whose withanolide profiles and extraction methods differ enough that results do not transfer cleanly between them. This trial tests a further variant: a sustained-release formulation engineered for extended plasma presence rather than higher total dose.
Double-blind, placebo-controlled, parallel-group, 3-arm RCT at two sites in Bangalore, India. 135 healthy adults with Perceived Stress Scale scores of 14–26 randomized 1:1:1 to 150 mg AshwaSR, 300 mg AshwaSR, or placebo once daily for 60 days. Mean age 34.79 ± 8.16 years. 126 completed. Registered CTRI/2023/06/053662.
Show full methodology
Recruitment ran 15 June to 23 October 2023. The extract is standardized to ≥4% total withanolides per manufacturer literature. Endpoints were change from baseline in perceived stress, sleep quality, psychological well-being, eating behavior, and serum cortisol. Attrition was roughly 6.7% overall; per-arm attrition was not reported. The specific validated instruments used for the three behavioral secondary endpoints, the statistical model, the power calculation, and the handling of missing data were not reported in accessible text.
Perceived stress fell significantly from baseline at both doses (P < .001 within-group), with both separating from placebo at P < .05. Exact between-group effect sizes and confidence intervals were not reported — only threshold significance. Sleep quality, psychological well-being, and eating behavior all improved on the same pattern. The most interesting result is the one the abstract does not reconcile: serum cortisol fell significantly only in the 300 mg arm. The 150 mg arm produced essentially equivalent subjective improvement without a detectable cortisol change.
Strengths
- Prospectively registered on CTRI
- Double-blind, placebo-controlled
- Dose-ranging design (two doses vs. placebo)
- Objective biomarker alongside subjective scales
- Low attrition (~6.7%)
Limitations
- Two authors are employees of Nutriventia, which owns the product
- Two authors are employees of Laila Nutra, the extract manufacturer
- Two authors are employees of the contract research organization
- Study funded by Nutriventia and Laila Nutra
- 60 days cannot address durability or discontinuation
- Healthy population, not a clinical anxiety population
- No numeric between-group effect sizes reported
If you recommend ashwagandha for stress on the basis of trial evidence, be specific about which extract. This trial supports AshwaSR at 150–300 mg once daily for at least 60 days in healthy adults with moderate perceived stress. A client taking a different brand, a root powder, or an unstandardized preparation is not covered by these data. Where an HPA-axis effect is the clinical goal, 300 mg is the dose with biomarker support. Standard cautions remain untested here: thyroid stimulation, additive sedation with CNS depressants, contraindication in pregnancy, and rare hepatotoxicity case reports that a 60-day trial in healthy adults cannot exclude.
“Ashwagandha reduces stress” is not what this trial demonstrates. “This sponsor’s sustained-release extract, at these doses, over 60 days, in healthy moderately stressed adults” is. Treat the reported effect as a manufacturer-favorable upper bound pending independent replication.
Curcumin’s poor oral bioavailability has driven a generation of enhanced-delivery formulations — phytosome complexes, piperine co-administration, micellar and nanoparticle systems. Whether these translate into better clinical outcomes, and whether boswellia benefits similarly, is the question this network meta-analysis sets out to answer.
Systematic review and network meta-analysis of 20 RCTs, 1,633 participants with predominantly mild-to-moderate knee osteoarthritis. CENTRAL, PubMed, EMBASE, and EBSCO Open Dissertations searched through March 2025 with citation snowballing. Random-effects modeling; Cochrane risk-of-bias tool; PRISMA 2020 and PRISMA-NMA reporting. PROSPERO CRD42024517194.
Show full methodology
Trials were geographically concentrated: twelve from India, three from Iran, two from Australia, one each from Armenia, Belgium, and Japan. The network treats “modified” enhanced-bioavailability formulations as separate nodes from standard extracts. Outcomes were VAS pain and WOMAC pain, stiffness, and function, with adverse events tracked as a safety endpoint. GRADE certainty ratings, formal publication-bias assessment, and domain-level risk-of-bias results were not reported in accessible sources.
Modified Curcuma longa formulations reduced VAS pain versus placebo; modified Boswellia serrata improved WOMAC pain, stiffness, and function versus other comparators, though exact effect sizes for those comparisons were not reported. Curcuma was described as non-inferior to NSAIDs on WOMAC total, without a stated non-inferiority margin. The headline effect size deserves scrutiny rather than celebration. By convention an SMD above 0.8 is a large effect; −2.82 is more than three times that. The interval spans from −5.30 — among the largest effects ever recorded for any osteoarthritis intervention — to −0.33, which is small. That width around that point estimate is the signature of few trials, high between-study variance, or both.
Strengths
- Prospectively registered (PROSPERO)
- Appropriate NMA methodology for ranking formulations
- 1,633 participants across 20 RCTs
- Explicit safety endpoint tracked
- Distinguishes modified from standard formulations
Limitations
- Headline confidence interval spans an order of magnitude
- Review-level COI disclosure does not cover the 20 pooled trials
- Trial-level industry sponsorship not reported or adjusted for
- “Modified” pools different delivery technologies as one category
- GRADE and publication bias not reported
- Evidence base concentrated in India and Iran
The actionable message is about formulation, not herb. The evidence supports bioavailability-enhanced preparations; it does not support turmeric powder or unstandardized boswellia, and that distinction is the whole point of the analysis. Ask manufacturers what delivery technology their product uses and whether their clinical evidence comes from that specific system. The supported population is mild-to-moderate disease. Combination curcumin-boswellia products are commercially popular, but the authors identify combination evidence as the gap their review could not fill. Standard interaction cautions apply: curcumin can potentiate anticoagulants and antiplatelets, warrants care in biliary obstruction, and piperine-enhanced formulations alter CYP3A4 and P-glycoprotein handling.
Two ways to misread this. The first is quoting the effect size as though it were stable. The second, more consequential, is recommending “curcumin” or “boswellia” generically when the analysis specifically credits enhanced-delivery formulations.
Dietary Patterns: Equivalence, Not Superiority
Three papers tested popular dietary patterns against comparators. In each case the pattern performed respectably and failed to outperform its alternative. That is a meaningful result, and a hard one to communicate to clients invested in a particular framework.
The Mediterranean diet is widely recommended in inflammatory bowel disease on the strength of its anti-inflammatory nutrient profile. Whether it actually outperforms other structured dietary approaches for inducing remission is a separate question, and one this meta-analysis addresses directly — then reports two answers that point in opposite directions.
Systematic review and meta-analysis. PubMed, Embase, and Scopus searched 10 February 2025. Eight studies included; seven, comprising 223 participants, contributed remission data. The diet was tested alongside standard medical therapy, not as monotherapy. Endpoints were clinical remission and response, with endoscopic and histological response planned.
Show full methodology
PROSPERO registration was not reported in any accessible source. The operational definition of “Mediterranean diet” — which food-group targets, which adherence instrument — was also not reported, which matters because the Mediterranean diet is not a standardized intervention across trials. Risk-of-bias tool, GRADE certainty, and publication-bias assessment were not reported. None of the eight studies reported endoscopic or histological outcomes.
Pooled clinical remission was 62%, with very high heterogeneity. Crohn’s disease and ulcerative colitis performed comparably (0.67, 95% CI 0.38–0.87 and 0.56, 95% CI 0.24–0.84). And then the finding that reframes the paper: compared with control diets, the Mediterranean diet showed no significant advantage in inducing remission. The 62% figure is a single-arm pooled proportion describing how many patients on the diet — while also receiving standard medical therapy — were in remission. It does not isolate the diet’s contribution, and at I² = 78.6% it is not a stable estimate. The odds ratio is the number that tests the hypothesis, and it sits on the null.
Strengths
- Multi-database search strategy
- Subgroup analysis by disease phenotype
- Published the null comparative result rather than leading with the single-arm rate
- All thirteen authors declare no competing interests
- No industry funding
Limitations
- Only 8 studies, 223 participants for the headline estimate
- Very high heterogeneity (I² = 78.6%)
- Inconsistent definitions of disease activity across trials
- Concurrent medical therapy throughout confounds attribution
- No endoscopic or histological outcomes reported
- Mediterranean diet not operationally defined
The Mediterranean diet remains defensible in IBD as a low-risk, sustainable adjunct with well-established cardiometabolic co-benefits. What it should not be called is disease-modifying, or superior to a structured alternative like the Specific Carbohydrate Diet. The honest framing for a client is that it works about as well as other reasonable dietary patterns and is generally easier to maintain — a real advantage, just not the one the 62% figure implies. With no endoscopic or histological data, no claim about mucosal healing is supportable. This aligns with DINE-CD (Lewis et al., Gastroenterology 2021), which found symptomatic remission of 43.5% versus 46.5% comparing the Mediterranean diet against the Specific Carbohydrate Diet — no significant difference.
Sixty-two percent is the number that will circulate and it is the wrong number to act on. It does not mean 62% of patients enter remission because of the diet.
Intermittent fasting has been promoted for a decade on the premise that meal timing confers metabolic benefit independent of calorie intake. Testing that claim requires comparing regimens matched for the actual energy deficit achieved — which is precisely what this network meta-analysis does across the largest evidence base assembled on the question.
Bayesian random-effects network meta-analysis of 167 RCTs, 11,998 participants with overweight, obesity, or metabolic abnormalities. Three degrees of continuous energy restriction compared against four categories of intermittent fasting. PubMed, Embase, and Cochrane CENTRAL searched from inception to December 2022. GRADE applied throughout. PROSPERO CRD42022379621.
Show full methodology
Funded entirely by Chinese public research bodies — the National Key R&D Program of China, the Shanghai Rising-Star Program, and Shanghai municipal sources — with no commercial sponsor apparent. Published online 14 May 2025 and issued into the March 2026 print volume, which explains the unusually low PMID for a 2026 paper. Exact caloric-deficit thresholds per restriction category, the full four-category fasting taxonomy, network geometry, and inconsistency diagnostics were not reported in accessible sources.
When regimens were matched for absolute energy restriction, most intermittent fasting patterns produced weight loss comparable to continuous restriction. Severe continuous restriction was most effective overall (moderate certainty), with alternate-day fasting the only fasting regimen rated high certainty. Body measurements, blood pressure, lipids, and glycemic markers followed the same pattern. In subgroup analysis, intermittent fasting regimens — but not continuous restriction — showed weight rebound after 12 weeks.
Strengths
- 167 RCTs and 11,998 participants — the largest such synthesis
- Bayesian NMA is the appropriate method for ranking rarely-compared regimens
- GRADE certainty applied transparently
- Prospectively registered on PROSPERO
- Entirely publicly funded, no industry sponsor
Limitations
- Literature search stopped at December 2022
- Predates the 2025 BMJ network meta-analysis of 99 RCTs
- Predates the 2025 Annals 12-month adherence trial
- Equivalence holds only conditional on matched achieved restriction
- Network geometry and heterogeneity statistics not accessible
Stop treating meal-timing protocols as metabolically privileged. When the achieved deficit is equal, the approaches perform equivalently, and the clinical lever is adherence rather than schedule. Match the pattern to what the client can actually sustain. Two practical specifics: build relapse-prevention planning around the twelve-week mark for fasting-based plans, where this analysis found rebound; and recognize that for some clients the structure of a fasting protocol is itself the adherence mechanism. The 2025 Annals of Internal Medicine trial makes that point directly — 4:3 fasting produced greater weight loss than daily restriction at twelve months (−7.7 kg vs. −4.8 kg, P = 0.04) with lower attrition (19% vs. 30%).
“Timing is irrelevant” is a defensible reading of this paper and an incomplete one. The equivalence holds conditional on matched achieved restriction — and in practice, timing often determines whether that restriction gets achieved at all.
Dietary fat and cancer risk has produced decades of conflicting literature, much of it addressing one fat subtype or one cancer site at a time. An umbrella review pools across that fragmentation — and in doing so exposes that the answer differs not just by fat type but by cancer site, occasionally reversing direction between them.
Umbrella review of 23 systematic reviews and meta-analyses. PubMed, Embase, Web of Science, and Cochrane searched from inception to September 2025. Exposures were total fat, saturated, monounsaturated, and polyunsaturated fat, contrasted lowest versus highest intake. AMSTAR-2 for methodological quality, GRADE for certainty. PROSPERO CRD420251236506.
Show full methodology
Per-association GRADE ratings, I² values, and publication-bias statistics were not reported in accessible text — the abstract states the methods were applied but does not give per-association results. Funding was not reported; the authors declare no conflicts of interest. Umbrella reviews inherit the confounding and publication bias of their constituent reviews, including the possibility of the same cohort being counted across several included meta-analyses, which this review does not appear to address.
Higher total fat was associated with increased risk of bladder (RR 1.28, 95% CI 1.04–1.58), breast (RR 1.10, 95% CI 1.05–1.16), gastric (RR 1.18, 95% CI 1.00–1.39), and esophageal cancer (RR 1.31, 95% CI 1.13–1.49), and non-Hodgkin lymphoma (RR 1.26, 95% CI 1.12–1.42). Saturated fat carried the strongest signals. Monounsaturated fat was associated with increased esophageal (RR 1.70, 95% CI 1.01–2.84) and breast cancer risk (RR 1.08, 95% CI 1.01–1.16) — and decreased skin cancer risk (RR 0.90, 95% CI 0.85–0.96). That MUFA finding sits in real tension with randomized evidence: PREDIMED found extra-virgin olive oil associated with reduced breast cancer incidence.
Strengths
- Prospectively registered on PROSPERO
- Broad synthesis across 23 reviews
- AMSTAR-2 and GRADE frameworks applied
- Reports fat subtypes separately rather than pooling
- Authors declare no conflicts
Limitations
- All associations rest on observational evidence
- Inherits confounding and publication bias from constituent reviews
- Possible double-counting of cohorts across included meta-analyses
- Gastric total-fat interval touches 1.00
- MUFA finding conflicts with randomized PREDIMED evidence
- Per-association GRADE ratings not reported
- Funding not reported
The strongest and most actionable signal is saturated fat reduction for clients at elevated esophageal, liver, or gastric cancer risk, where intervals sit clear of the null. Shifting saturated toward polyunsaturated sources is a no-regrets recommendation that aligns with cardiometabolic guidance anyway. Do not revise olive oil advice on the strength of this paper — when an observational umbrella review and a randomized trial disagree on direction, the trial should carry more weight.
“Dietary fat is linked to cancer” flattens a picture that is specific to fat subtype and cancer site, and sometimes reverses direction between them. A blanket low-fat recommendation is not what these data support.
The Gut: Signatures and Interventions
One paper maps the gut at the moment disease begins; another tests what actually moves it. Together they mark the distance between a compelling mechanistic picture and an actionable clinical protocol.
Microbiome literature has a comparability problem. Different studies use different sequencing pipelines, different reference databases, and different bioinformatic choices, which means apparent disagreements between cohorts may be artifacts of processing rather than biology. This paper’s contribution is to remove that variable entirely by reprocessing raw sequence data through a single pipeline.
Systematic review and unified bioinformatic synthesis. 36 eligible studies; 18 contributed raw 16S rRNA sequence data for reanalysis, 24 contributed to a supplementary meta-analysis. 1,743 patient samples across eleven countries — 678 Crohn’s disease, 399 ulcerative colitis, 130 healthy controls, 405 symptomatic controls — all treatment-naïve at new onset. 990 were mucosal biopsies.
Show full methodology
All raw data were reprocessed through a single QIIME2 pipeline, eliminating cross-study bioinformatic heterogeneity. Differential abundance was modeled in MaAsLin2 adjusting for methodological differences; the full adjustment variable list was not accessible. Risk of bias was assessed with ROBINS-E, though the overall judgment distribution was not reported. Cohorts spanned both pediatric and adult populations.
Alpha diversity was reduced across multiple comparisons, including adult Crohn’s and ulcerative colitis versus healthy controls. Beta diversity separated fecal from mucosal communities clearly — least so in ulcerative colitis — with additional geographic clustering. Multivariate modeling showed depletion of anaerobic taxa and enrichment of aerobic and facultatively anaerobic, orally-associated genera in both disease phenotypes, with Granulicatella and Haemophilus among those enriched. Exact effect sizes and q-values for individual taxa were not accessible.
Strengths
- Reprocessing raw data through one pipeline is a genuine methodological advance
- 1,743 samples across 11 countries
- Treatment-naïve, new-onset patients only
- Includes both pediatric and adult cohorts
- Symptomatic controls as well as healthy controls
- ROBINS-E risk-of-bias assessment applied
Limitations
- Only 18 of 36 eligible studies had retrievable raw data
- Data-sharing studies are unlikely to be a random sample
- Cross-sectional design cannot establish causal direction
- PubMed structured COI field is empty despite documented author industry ties
- Several authors report relationships with IBD drug manufacturers
- Exact taxon-level effect sizes not accessible
None yet, and that is the honest answer. The paper tests no dietary or herbal intervention. The shift toward oxygen-tolerant organisms is consistent with a broader hypothesis about luminal oxygen availability, and one could construct a rationale for supporting butyrate-producing anaerobes through fermentable fiber — but that is the practitioner’s extrapolation, not this study’s finding. Its value is as a cleaner map of what the gut looks like at disease onset, which is a precondition for eventually testing whether anything moves it.
Do not treat an onset signature as a validated diagnostic or a demonstrated therapeutic target. The institutional press framing about paving the way for new treatments is aspiration, not result.
Fiber supplementation is the standard first-line dietary recommendation for functional constipation across most gastroenterology and dietetic guidelines. This network meta-analysis tests that hierarchy against whole-food approaches and finds it inverted — though the certainty behind the reordering varies considerably by outcome.
Bayesian network meta-analysis of 19 RCTs in functional constipation. Four databases searched to 20 August 2025. Interventions grouped as fruit-based foods, multicomponent foods, vegetables with whole grains, fiber supplements, mineral water, pharmacologic comparators, and placebo. Outcomes were defecation frequency, stool consistency, and constipation severity, ranked by SUCRA. PROSPERO CRD420251134853.
Show full methodology
73.7% of included studies were rated low risk of bias. Network regression, sensitivity, and subgroup analyses were conducted to explore heterogeneity sources. Exact SUCRA values, effect estimates with credible intervals, I² statistics, and publication-bias assessment were not reported in accessible text — only ranked comparisons with GRADE-style certainty labels. Whether the underlying trials were pediatric, adult, or geographically concentrated was not reported.
For defecation frequency, fruit-based foods outperformed fiber supplements (moderate certainty), mineral water (low certainty), and placebo (moderate certainty). Multicomponent foods outperformed fiber supplements, pharmacologic comparators, and placebo at low-to-moderate certainty. For stool consistency, multicomponent foods beat pharmacologic comparators and placebo at high certainty. For severity, fruit-based foods ranked first. Mineral water performed relatively better in longer trials, suggesting a slower-onset strategy.
Strengths
- Prospectively registered on PROSPERO
- Bayesian NMA handles multi-arm comparison networks appropriately
- 73.7% of trials at low risk of bias
- Subgroup and sensitivity analyses conducted
- Directly tests the standard fiber-first hierarchy
Limitations
- 19 trials spread across seven intervention nodes
- Several comparisons rest on thin indirect evidence
- Food categories pool heterogeneous, non-standardized interventions
- Exact effect sizes behind rankings not accessible
- Population age and geography not reported
- Most certainty ratings are low or moderate
This supports a real reordering of the usual hierarchy: whole-food approaches ahead of isolated fiber supplements as first-line dietary management. Confidence should be graded by outcome — high for stool consistency with multicomponent foods, only low-to-moderate for defecation frequency, which is often what patients care about most. Because the food categories are not standardized, this cannot yet be translated into an exact prescribable protocol without the full intervention table. Mineral water appears to be a slower-onset strategy, useful for setting client expectations.
The ranking is clear; the certainty behind it mostly is not. Only the stool-consistency comparisons reach high certainty, and defecation frequency — arguably the outcome patients care most about — rests on low-to-moderate evidence.
Surrogates, Subgroups, and Small Trials
Three studies that are statistically significant and clinically ambiguous — each for a different reason. A surrogate endpoint inside a null trial, a subgroup finding that fails a continuity test, and a p-value drawn from nine participants per arm.
COSMOS randomized 21,442 older US adults to daily cocoa extract and/or a multivitamin-multimineral in a 2×2 factorial design, with median follow-up of 3.6 years. Its prespecified primary outcomes were null: total invasive cancer for the multivitamin (HR 0.97, 95% CI 0.86–1.09, P = 0.57) and total cardiovascular events for cocoa extract (HR 0.90, 95% CI 0.78–1.02, P = 0.11). This prespecified ancillary study asks whether either intervention moved epigenetic aging.
Prespecified ancillary study in 958 COSMOS participants (482 women, 476 men) with paired blood samples at baseline and two years. Five DNA-methylation clocks assessed: PCHannum and PCHorvath (first-generation, chronological-age trained), PCPhenoAge and PCGrimAge (second-generation, physiology and mortality trained), and DunedinPACE (pace-of-aging). Parent trial NCT02422745.
Show full methodology
Principal-component clock versions were chosen specifically to reduce technical noise in repeated-measures designs. The intervention was daily multivitamin (Centrum Silver) and/or cocoa extract providing 500 mg flavanols including 80 mg epicatechin. The adjustment covariate list, the selection process from 21,442 participants down to 958, and any multiple-comparison correction across the five clocks were not accessible. Funding included investigator-initiated grants from Mars Edge and Pfizer Consumer Healthcare (now Haleon), including donation of the study pills. FOXO Technologies provided in-kind DNA methylation data generation and preprocessing, and one author is a former FOXO Technologies employee.
The multivitamin slowed two of five clocks. The other three showed no significant effect, and cocoa extract affected none of the five. The subgroup result is the most clinically suggestive: among participants with accelerated biological aging at baseline, the multivitamin effect on PCGrimAge was −0.236 years, versus −0.013 years (95% CI −0.130 to 0.104) in those aging normally, with P = 0.018 for interaction. Effects are measured in fractions of a year against instruments whose test-retest variability independent commentators put substantially higher.
Strengths
- Nested in a large, well-conducted, placebo-controlled RCT
- Prespecified analysis plan
- Reliability-optimized PC clock versions appropriate for longitudinal designs
- Tested a biologically plausible effect-modification subgroup
- Reported the cocoa null result clearly
Limitations
- Five clocks tested, two significant, correction method not confirmed
- Effect sizes are fractions of a year against noisy instruments
- Epigenetic clocks are surrogate endpoints, not clinical outcomes
- Parent trial was null on invasive cancer and total CVD
- Study pills donated by Mars Edge and Pfizer Consumer Healthcare
- Methylation data generated in kind by FOXO Technologies
- One author is a former FOXO Technologies employee
- 958 of 21,442 selected by an unreported process
Epigenetic clocks are statistical constructs correlated with, but not demonstrated to cause, morbidity and mortality. This analysis shows a small effect on two of five such constructs, inside a trial that found no effect on cancer or cardiovascular incidence. That combination does not support recommending a multivitamin for disease prevention. The effect-modification finding is the genuinely interesting thread — supplementation may matter more for people already showing accelerated aging than for those aging normally — and it is hypothesis-generating rather than actionable. Cocoa flavanols have no support here whatsoever.
Telling a client their multivitamin is measurably slowing their aging overstates what a few months of movement on a secondary methylation composite can bear — particularly when the same trial’s hard endpoints came back null. A companion Nature Medicine editorial framed healthspan benefit as an open question, which is the right posture.
Ultra-processed food consumption has been linked to a widening range of chronic disease outcomes, but evidence in pregnancy remains scarce. This secondary analysis of a randomized trial brings an unusual strength to the question — repeated within-person dietary measurement across pregnancy rather than a single snapshot.
Secondary analysis of IMPACT BCN, which randomized 1,221 singleton pregnancies at high risk for small-for-gestational-age newborns in Barcelona 2017–2020. 812 participants with complete dietary data at both weeks 19–23 and weeks 34–36 were included, assessed by a validated 151-item food frequency questionnaire and classified by NOVA. Participants were grouped into tertiles of change in ultra-processed food consumption. Registered NCT03166332.
Show full methodology
Preeclampsia was defined as hypertension plus target organ involvement. Logistic regression adjusted for maternal age, socioeconomic status, baseline energy intake, ethnicity, pre-pregnancy BMI, and intervention arm, with a second model adding nulliparity and smoking. The parent trial randomized participants to a Mediterranean diet intervention, mindfulness-based stress reduction, or usual care, with reduced SGA prevalence as its primary outcome. Reasons for exclusion of the ~409 participants without paired dietary data, and comparison of included versus excluded characteristics, were not reported. No multiple-comparison correction across the tertile, continuous, and subclass models was reported.
Across tertiles of increasing ultra-processed food consumption, the odds of preeclampsia rose more than twofold, with a significant trend. Among subclasses, pre-prepared dishes carried the strongest association. Modeled continuously, the same exposure showed no association at all. That discordance is the paper’s central interpretive problem. A genuine monotonic dose-response relationship generally appears in both specifications. When it appears only in tertiles, the candidates are threshold effects, sensitivity to cutpoint placement, residual confounding concentrated at the extremes, or chance in a modest sample. The paper reports no formal nonlinearity test that would distinguish genuine threshold behavior from artifact.
Strengths
- Repeated within-person dietary measurement rather than a single snapshot
- Validated 151-item food frequency questionnaire
- NOVA is the standard framework for UPF research
- Adjusted for the parent trial’s own intervention arm
- Nested within a registered randomized trial
Limitations
- Tertile and continuous models disagree
- Secondary analysis of a trial powered for a different outcome
- Preeclampsia is uncommon, so effective case count is small
- Confidence interval spans 1.06 to 4.97 — very imprecise
- High-risk SGA population, not general obstetric
- Food-frequency data carry reporting bias
- No formal nonlinearity test reported
Counseling pregnant clients — especially those already flagged as high-risk — against increasing ultra-processed intake as pregnancy progresses is sensible, low-risk advice consistent with the Mediterranean-pattern guidance the parent trial already supports. Pre-prepared dishes carried the strongest subclass signal and are a reasonable specific target. What this should not become is a quantified risk claim. No client should be told that each additional serving of processed food doubles their risk; the continuous model directly contradicts that framing.
Lead with the contradiction rather than burying it. The distinction is between “increasing ultra-processed intake in pregnancy raises preeclampsia risk dose-dependently” — not supported — and “there is a statistically fragile signal in a small secondary analysis that does not survive a linear dose-response test,” which is what the data show.
Creatine and cognition has a growing evidence base in younger adults, sleep-deprived populations, and vegetarians, generally at doses of 3–20 g/day of monohydrate. Creatine specifically in perimenopausal and menopausal women is close to an empty literature. This trial is described as the first RCT in that population — which is exactly why it merits attention and exactly why it should not be over-read.
Randomized, double-blind, placebo-controlled trial over 8 weeks. 36 apparently healthy peri- and postmenopausal women, mean age 50.1 ± 5.7 years, randomized to four arms: creatine HCl 750 mg/day, creatine HCl 1,500 mg/day, creatine HCl plus creatine ethyl ester at 800 mg/day combined, or placebo. Registered NCT06660004.
Show full methodology
Four arms across 36 participants is roughly nine per arm. Brain creatine was measured by frontal magnetic resonance spectroscopy in a subset of sixteen participants — approximately four per arm. The specific cognitive battery was not named in any accessible source, nor were the randomization method, allocation concealment, per-arm attrition, or statistical model. The trial was partially funded by Vireo Systems, a commercial creatine HCl manufacturer, which also supplied the products tested; the senior author has reported creatine-industry advisory relationships and co-owned creatine patents.
Medium-dose creatine HCl was reported superior to placebo for reaction time and frontal brain creatine, with favorable lipid modulation at P < 0.05. Reduction in mood-swing severity did not reach significance. No severe adverse events. A word on those percentages: they appear to represent within-group change from baseline rather than a between-group effect size, but the underlying metric is not specified in any accessible source, and secondary press accounts contradict both each other and the abstract. No confidence intervals are reported for any outcome. An independent appraisal characterized the trial as largely negative overall, noting no significant effects on most menopausal symptoms including fatigue and exercise tolerance.
Strengths
- First RCT of creatine in this specific population
- Double-blind and placebo-controlled
- Multi-domain outcomes including brain imaging
- Low water-soluble doses chosen to avoid GI and water-retention adherence barriers
- Prospectively registered
Limitations
- Roughly nine participants per arm
- Imaging endpoint at roughly four per arm is exploratory at best
- Partially funded by a creatine HCl manufacturer
- Manufacturer supplied the tested products
- Senior author holds creatine patents and industry advisory roles
- Most prespecified menopause outcomes were null
- No confidence intervals reported for any outcome
- Underlying metric of the headline percentages unverifiable
This does not support recommending a specific creatine form or dose for menopausal cognitive symptoms. It supports two narrower claims: low-dose creatine HCl was well tolerated over eight weeks in this population, and there is a reaction-time and brain-creatine signal at 1,500 mg/day worth replicating by an independent, adequately powered group. Practitioners already using creatine monohydrate for other indications have no basis here to switch clients to HCl or ethyl ester, and none to promise relief from hot flashes, fatigue, mood symptoms, or exercise tolerance — none of which improved.
A P < 0.01 from a nine-person arm, in a trial funded and supplied by the manufacturer of the tested product, with most prespecified menopause outcomes null, is a preliminary signal. It is not evidence that creatine treats menopausal brain fog.
Synthesis & Emerging Themes
The Framework Matters Less Than We Would Like
Three of this month’s studies converge on the same uncomfortable message. The Mediterranean diet did not beat control diets in IBD. Intermittent fasting did not beat calorie-matched continuous restriction. Whole foods did beat fiber supplements for constipation — the one clear hierarchy reversal of the month — but mostly at low-to-moderate certainty. For a profession organized substantially around dietary frameworks, this is worth sitting with. The evidence increasingly suggests the framework is a delivery vehicle for adherence, and adherence is the active ingredient. The 2025 Annals fasting trial makes the point precisely: 4:3 fasting won at twelve months not through metabolic privilege but through lower attrition.
The Surrogate Endpoint Problem Is Getting Worse as Measurement Improves
Epigenetic clocks are a real scientific achievement, and the COSMOS ancillary study is competent work. But it sits inside a trial whose hard endpoints were null, tests five constructs and reports two, and was measured with technology supplied in kind by a company with a stake in the result. As biomarkers proliferate, the gap between “we can measure a change” and “the change matters to a patient” widens. The discipline of asking which endpoint was prespecified, and what the parent trial found on outcomes patients can actually feel, becomes more important rather than less.
Botanical Research Has a Structural Sponsorship Problem
The ashwagandha trial is well designed. It is also authored by the product owner’s staff, the extract manufacturer’s staff, and the contract research organization’s staff. The curcumin network meta-analysis has clean review-level disclosures and pools twenty trials whose own sponsorship it does not assess. The creatine trial was supplied by a manufacturer whose product it validates. None of this makes the findings false. It does mean that for the botanical and supplement literature specifically, effect sizes should be read as upper bounds until independent replication exists — and that practitioners should be able to name which extract, at which standardization, a recommendation rests on.
What Remains Unresolved
Whether the IBD onset signature is cause or consequence is the question the field most needs answered, and cross-sectional data cannot answer it. Whether enhanced-bioavailability curcumin’s effect is real at anything like the reported magnitude awaits trials large enough to narrow that interval. Whether the multivitamin effect on epigenetic aging in already-accelerated individuals reflects something clinically real is genuinely open, and genuinely interesting.
The through-line this month is a reading skill rather than a finding. Eight of these ten papers contain, somewhere below the conclusion sentence, a number that complicates the headline: an odds ratio sitting on the null beneath a 62% remission rate, a continuous model contradicting a tertile model, three clocks that did not move, a confidence interval spanning an order of magnitude, a literature search that stopped three years before publication. None of those complications were hidden. Every one was reported by the authors, in the abstract, in plain language. They simply were not the sentence that got written last — and the sentence written last is the one that becomes the practice recommendation, the podcast segment, the client handout.
“The abstract is written to be quoted. The results are written to be checked. When they disagree, the results are the paper.”
The question worth carrying into next month is not whether to trust the literature. It is whether we are reading far enough into it. The most consequential number in a paper is rarely the one in the conclusion.