Counts describe this library, not the strength of evidence.
How to read this evidence
Comparators explain many apparent contradictions. The 2022 trial tested adding a window to calorie restriction; the 2026 synthesis compared timing and duration across trials. Neither supports assuming the shortest window is best. The corrected PREDIMED report also illustrates why publication history matters.
In a review of 9 randomized trials (6 in the meta-analysis), low-glycaemic-index (low-GI) diets lowered HbA1c and fasting blood glucose in type 2 diabetes versus higher-GI or control diets (both P < 0.001 for HbA1c). Individual trials were mixed, with no significant HbA1c difference in four studies, and the abstract gives no pooled effect size.
Across 260 milk-feeding studies and 173 studies of other dietary exposures, probiotics in late pregnancy and lactation may reduce eczema (RR 0.78, 95% CI 0.68-0.90) and fish oil may reduce egg sensitisation (RR 0.69, 95% CI 0.53-0.90), both with moderate certainty. Many other diet exposures were inconclusive or showed no effect, and many studies had a high risk of bias.
In 654 people with type 2 diabetes followed for a median 9.3 years, larger visit-to-visit variability in HbA1c or fasting glucose over 24 months predicted cardiovascular events, kidney outcomes and death, and often did so better than mean HbA1c. Variability did not predict retinopathy overall, and the abstract gives no effect sizes or sample details beyond counts.
In 609 adults without diabetes, 12-month weight change was -5.3 kg on a healthy low-fat diet and -6.0 kg on a healthy low-carbohydrate diet, a 0.7 kg gap (95% CI -0.2 to 1.6 kg) that was not statistically significant. A tested genotype pattern and baseline insulin secretion did not predict who did better on which diet. Both arms received 22 coaching sessions and 79% finished, so this describes coached dieting in adults aged 18 to 50.
In a randomized trial, gestational diabetes occurred in 74 of 434 women (17.1%) on a Mediterranean diet with extra virgin olive oil and pistachios versus 103 of 440 (23.4%) on a low-fat standard diet (adjusted RR 0.75, 95% CI 0.57 to 0.98). Results cover the 874 women who completed the study, not everyone assigned.
In a secondary analysis of the DEVOTE insulin trial, higher day-to-day variability in fasting self-measured glucose was associated with severe hypoglycaemia (HR 4.11, 95% CI 3.15 to 5.35) and all-cause mortality (HR 1.58, 95% CI 1.23 to 2.03) before adjustment. Both links held after adjustment, but the link with major cardiovascular events did not, and the design shows association only.
Among 1053 people with self-reported Parkinson's disease, eating more fresh vegetables, fresh fruit, nuts, nonfried fish and olive oil was associated with a slower rate of progression, while fried foods, beef and soda were linked to faster progression (all P < 0.05). This cross-sectional survey relies on self-reported diet and symptoms, so it cannot show that food changes the disease.
Among 11,396 UK adults aged 29 to 64, those eating home-cooked meals more than five times a week ate 62.3 g more fruit and 97.8 g more vegetables daily than those eating them fewer than three times. This is a one-time snapshot with self-reported meals, so it cannot show that cooking at home causes better health.
The international journal of behavioral nutrition and physical activity
Across 15 studies, digital behaviour change programs were linked to 41 more minutes per week of moderate-to-vigorous activity (95% CI 12 to 71) in cancer survivors, plus a small drop in BMI or weight. The evidence is thin: few studies, high risk of bias and high heterogeneity.
Journal of cancer survivorship : research and practice
Among 1,865 Greek adults (mean age 73), each one-point rise in the Mediterranean diet score was associated with 10% lower odds of dementia, and higher scores went with better memory. These are early results from an ongoing study, and an association in older people cannot show that the diet protects the brain.
Of 25 highly rated apps for children and adolescents, overall quality was moderate (total Mobile App Rating Scale score 3.6 on a 5-point scale), with information quality lowest (mean 2.8, SD 0.8). Apps were selected for popularity (user rating of at least 4), so this describes well-liked apps, not whether any of them changes behavior.
The international journal of behavioral nutrition and physical activity
This systematic review found probable or convincing evidence of causal effects on heart disease, stroke or diabetes for 10 foods and 7 nutrients, with fruits, vegetables, nuts, whole grains, fish and fiber protective, and processed meats, sugar-sweetened beverages, trans-fats and sodium harmful. The abstract gives no effect sizes, and most evidence comes from observational cohorts, so the causal label depends on the authors' judgement.
In 65 adults in New Zealand with overweight or obesity plus a cardiometabolic condition, a 12-week whole food plant-based programme cut BMI 3.9 kg/m2 more than normal care at 6 months (P<0.0001), but the cholesterol difference was not significant (P=0.1). Only 49 completed 6 months, and the study was not blinded to diet.
Among US adults, metabolic syndrome prevalence rose from 25.3% in 1988-1994 to 34.2% in 2007-2012, and the increase appeared in every sociodemographic group. These are repeated cross-sectional surveys, so they show how common the condition was in each period, not why it rose or who developed it.
Pooling 12 randomized trials with 974 adults with diabetes, app-based interventions lowered HbA1c by 0.48% (95% CI 0.19% to 0.77%) compared with usual care alone, without excess adverse events. The feature comparisons were exploratory subgroup analyses across few trials, so they point to hypotheses rather than proven app designs.
In a pilot trial of 25 overweight adults with type 2 diabetes, a 32-week online ketogenic diet and lifestyle program changed HbA1c by -0.8% versus -0.3% on a plate-method program (P=.002), and weight by -12.7 kg versus -3.0 kg (P<.001). The trial was very small, and more people dropped out of the plate-method arm (6/13) than the keto arm (1/12), which can skew the comparison.