Evidence about diet and lifestyle exposures
Diet and lifestyle research examines interacting exposures over long periods. Appraisal considers whether a design fits the exposure, how bias and confounding were addressed, and whether findings agree across populations and methods.
Exposure and study design
A dietary pattern is a complex system of interacting components rather than a single nutrient that can be put into a pill. Single-nutrient experiments remain useful for deficiency diseases and questions about additives, using in vitro work, animal studies and randomised controlled trials. Research on long-term dietary patterns moved from case–control studies and international comparisons towards prospective cohorts, alongside some randomised trials.
Applying the drug-trial paradigm to dietary patterns can undervalue prospective cohort evidence. Cohort appraisal should consider measures to minimise reverse causation, recall bias and selection bias, adjustment for confounding, design quality, statistical analysis and careful interpretation. Replication across populations, including multinational studies, helps establish which findings hold in different settings.
Converging evidence and causality
Bradford Hill criteria can help assess causality in epidemiological evidence. Dietary patterns, individual nutrients, whole foods and biomarkers provide different ways of examining diet-related effects. Consistent findings across randomised trials, prospective cohorts, ecological and migrant studies, cross-sectional studies, animal research and in vitro work strengthen causal inference. These designs provide complementary evidence rather than a case built on randomised trials alone.
Appraisal should consider multiple pathophysiological pathways rather than a single outcome, because dietary components act together. An isolated finding that conflicts with the broader nutrition evidence should be interpreted in the context of that evidence base.
Disagreement between designs
A 2012 review of cancer prevention described why early case–control findings about fruit and vegetables could look more protective than prospective cohort findings: cases and controls could recall past diets differently, and healthier, more health-conscious people could be more likely to participate as controls. Prospective cohorts avoided those particular retrospective recall and control-selection problems; their associations between total fruit and vegetable intake and cancer risk were weak at best in the evidence reviewed. Smoking and alcohol also confound these comparisons, because high fruit and vegetable intake is associated with lower smoking and drinking.
A weak or undetectable association for total fruit and vegetable intake does not exclude a protective effect of an individual food, a group of foods or a particular substance. Observational studies and randomised trials may disagree because of differences in population, exposure dose, timing, adherence, follow-up length or primary endpoint. A null trial tests a particular intervention in a particular population over its follow-up period; it does not establish absence of an effect in every setting. Combinations of nutrients and other bioactive food components are difficult to test in their natural food context, and cancer-prevention trials may need older participants to achieve sufficient statistical power.
The mix of lifestyle study designs
A preliminary 2025 survey, rather than a formal review, examined lifestyle studies in ClinicalTrials.gov and title-indexed publications in PubMed. The registry search used the exact intervention term lifestyle, removed suspended, withdrawn and unknown-status records, and filtered titles and interventions for lifestyle, coaching, health coaching, counselling, motivational, habits and behaviour. The PubMed title searches covered 1995–2025 and lifestyle, nutrition, physical activity, stress and sleep, with article-type filters.
The registry search found 6,179 studies, comprising 701 observational and 5,478 interventional records; after exclusions, 271 observational and 2,846 interventional studies remained. Among the interventional studies, 2,342 were randomised and 1,308 were masked. The included studies targeted obesity in 660 cases, diabetes in 426, cardiovascular disease in 323, cancer in 230, neurological conditions in 183, mental health and stress in 100, female conditions in 71 and sleep in 62. Nearly half of the interventional studies tested one lifestyle intervention; the remainder combined it with other lifestyle interventions, drugs, devices, diagnostics or procedures. Fewer than one in ten records, 268 of 3,117, posted results.
The registry counts by intervention type were:
| Intervention | Interventional | Observational |
|---|---|---|
| Behavioural | 1,917 | 82 |
| Other, mostly lifestyle | 481 | 75 |
| Dietary supplement | 150 | 4 |
| Drug | 140 | 8 |
| Device | 94 | 8 |
| Procedure | 42 | 6 |
| Other categories | 22 | 9 |
| Total | 2,846 | 271 |
The counts by lifestyle pillar were:
| Pillar | Interventional | Observational |
|---|---|---|
| Lifestyle, coaching, habits, behaviour, counselling or motivational | 1,756 | 204 |
| Nutrition, diet | 436 | 40 |
| Exercise, physical activity | 487 | 17 |
| Sleep | 67 | 6 |
| Stress, mindfulness, meditation | 100 | 4 |
The countries with more than ten studies were tabulated as follows:
| Country | Interventional | Observational |
|---|---|---|
| United States | 1,132 | 56 |
| Canada | 123 | 12 |
| United Kingdom | 80 | 14 |
| Spain | 136 | 11 |
| Italy | 80 | 19 |
| Netherlands | 66 | 7 |
| Germany | 64 | 5 |
| Sweden | 63 | 1 |
| Norway | 58 | 4 |
| France | 44 | 17 |
The registry studies were grouped by start year:
| Start year | Interventional | Observational |
|---|---|---|
| 2020–2025 | 1,187 | 124 |
| 2015–2019 | 688 | 72 |
| 2010–2014 | 539 | 37 |
| 2005–2009 | 299 | 20 |
| 2000–2004 | 102 | 9 |
| Before 1999 | 21 | 9 |
The PubMed searches produced the following counts by article type:
| Article type | Lifestyle | Nutrition | Physical activity | Stress | Sleep |
|---|---|---|---|---|---|
| Clinical trial or randomised controlled trial | 2,520 | 17,423 | 24,997 | 11,745 | 7,394 |
| Observational study | 316 | 2,103 | 1,750 | 1,163 | 1,598 |
| Case report | 125 | 3,635 | 2,275 | 3,258 | 3,029 |
| Review | 3,053 | 45,627 | 22,812 | 31,460 | 15,031 |
| Systematic review or meta-analysis | 862 | 7,021 | 9,541 | 4,073 | 3,807 |
| Editorial | 586 | 6,359 | 3,256 | 3,232 | 3,116 |
| Letter | 487 | 63,640 | 4,041 | 3,646 | 3,739 |
| Guideline | 27 | 750 | 242 | 82 | 167 |
| Total | 7,949 | 145,808 | 68,672 | 58,577 | 37,714 |
The predominance of registered interventional studies may partly reflect journal requirements for trial registration before publication. Few registry records contained results, and a fuller review would need other registries and checks for overlap with PubMed publications. The counts challenge an assumption that lifestyle medicine rests only on observational evidence; randomised trials such as PREDIMED also study lifestyle interventions.
Observational-study counts ranged from 7.0% to 21.6% of clinical-trial counts across the PubMed searches. In the lifestyle search, reviews accounted for 38.4% and systematic reviews and meta-analyses for 10.8%; 27 of 7,949 records, approximately 0.34%, were guidelines.
Appraising lifestyle real-world data
Long-term, multifactorial effects of daily habits make real-world data useful for evaluating outcomes and effectiveness across diverse populations over time. Recommended checks for reporting lifestyle real-world data cover four areas:
- Ethics: confirm that consent covers collection and sharing with third parties or care teams; identify data and sources transparently, including storage links where needed; check applicable data-protection and research-ethics requirements; and ensure protection, de-identification and encryption of digital data.
- Diversity: consider the source, cultural context, socioeconomic status and access before generalising. Real-world data can include people excluded from trials, but wearable and app data may favour populations in middle- and high-income countries.
- Long-term effects: examine timeframes, population criteria, missing-data exclusions and dropout. Data entry, adherence and continued participation affect reliability; a workflow can show how data were selected.
- Quality: examine inconsistent or inaccurate activity logging, self-report bias, study design and implementation across sites, unmeasured confounding, reporting guidelines and analyses such as propensity score matching and sensitivity analyses. STROBE and HEALM can assist appraisal.