Systematic reviewindependent cross-checkfull text checked★ key study I-DRY-12 in-depth review
Overview of 41 reviews: milk more often linked to benefit than harm
Milk consumption and multiple health outcomes: umbrella review of systematic reviews and meta-analyses in humans.
Overview
Zhang X et al.. Nutr Metab (Lond) 2021 · PMID 33413488 · DOI 10.1186/s12986-020-00527-y
Milk was linked more often to benefits (heart, stroke, bowel cancer, diabetes, bones) than to harms (prostate cancer, Parkinson's, acne); most evidence was rated low certainty.
Does the study support the specific claim? Partly · Methodological quality: low
Plain-language summary
Study explained simply
- What did the researchers want to find out?
Summarise milk and health.
- How was the study done?
Umbrella review.
- How many people, animals or samples were studied?
41 meta-analyses.
- What was actually found?
Mostly beneficial associations; harmful associations for prostate cancer, Parkinson's, acne, infant iron deficiency and, per extra 200 g/day, hip fracture.
- How large was the observed effect?
Not pooled.
- What do the results mean in everyday life?
Milk is not shown to be generally harmful.
- What are the study's limitations?
Umbrella review without protocol; mostly observational, low/very-low certainty; overlap not handled; milk only (no yogurt/cheese).
- What does it say about Elawady's specific claim?
Contradicts blanket harm claims; confirms specific signals.
- Do the results apply to Germany and Europe?
Moderate.
- What can explicitly not be concluded from this study?
Causation; that milk is harmless for everyone; anything specific about yogurt or cheese.
Scientific in-depth review
1. Full reference
- Authors
- Zhang X et al.
- Journal
- Nutr Metab (Lond)
- Year
- 2021
- Identifier
- PMID 33413488 DOI 10.1186/s12986-020-00527-y
- Verification
- verified – PubMed efetch 2026-10-25
- Retracted
- no
- Corrections
- none
- Extent checked
- full text
- Conflicts of interest and funding
Competing interests: authors declare none. Funding: Chinese Medical Board Grant on Evidence-Based Medicine (No. 98-680), National Natural Science Foundation of China (No. 30901427, No. 71974135), mainly used for literature search and language.
2. Research question
Milk consumption and multiple health outcomes.
3. Study design and population
- Study design
- Systematic review – Umbrella review of meta-analyses (PubMed, Embase, Web of Science to 16 April 2019); 1,857 records, 85 eligible meta-analyses, 41 most recent/largest included covering 45 unique outcomes; AMSTAR 2 and GRADE applied; no re-analysis of primary data.
- Population
Humans in included meta-analyses
- Country
Multinational
- Period
Literature search from inception to 16 April 2019
- Sample size
41 meta-analyses, 45 unique outcomes (number of primary studies per meta-analysis 2-22)
- Inclusion and exclusion criteria
Included: meta-analyses (with or without systematic review) of interventional and/or observational studies of milk consumption and health outcomes reporting OR/RR/HR or MD/SMD, English language; newest/larger one kept when duplicated. Excluded: systematic reviews without meta-analysis, animal/in vitro data, and studies on dairy products other than milk (cheese, yogurt, butter etc. not covered) and milk from other species.
4. Methods and endpoints
- Exposure / intervention
Cow's milk (total, low-fat/skim, high-fat/whole) as categorised in the included meta-analyses.
- Comparator
Highest vs lowest intake, or dose-response per stated increment (mostly 200 g/day or 244 g/day).
- Endpoints
45 outcomes.
- Statistical methods
Extraction of summary estimates, I2 and Egger's test from each meta-analysis (Egger recalculated only when >10 studies and data available); AMSTAR 2 for methodological quality, GRADE for certainty; no pooling across meta-analyses.
5. Actual results
- Results
Per 200 ml/day or high vs low: stroke RR 0.93 (0.88-0.98), CVD 0.94 (0.89-0.99), hypertension 0.96 (0.94-0.98), colorectal cancer 0.82 (0.76-0.88), metabolic syndrome 0.79 (0.64-0.97), obesity OR 0.81 (0.75-0.88), osteoporosis 0.61 (0.50-0.75), T2DM 0.87 (0.78-0.96), Alzheimer's OR 0.63 (0.44-0.90). Harmful associations: prostate cancer high vs low 1.11 (1.03-1.21), per 490 g/day 1.03 (1.00-1.06); prostate cancer mortality 1.50 (1.03-2.17); CHD mortality 1.04 (1.02-1.06); high-fat milk and stroke 1.04 (1.02-1.06); Parkinson's 1.45 (1.23-1.73), per 200 g/day 1.17 (1.06-1.30); acne OR 1.48 (1.31-1.66); infant iron-deficiency anaemia RR 3.67 (2.73-5.19); gastric cancer 1.44 (1.15-1.81); DLBCL 1.49 (1.08-2.06); ovarian cancer OR 1.23 (1.03-1.46). Hip fracture: high vs low 0.93 (0.75-1.15), but per 200 g/day 1.09 (1.07-1.11) (Malmir 2019, publication bias P = 0.015). All-cause mortality 0.99 (0.98-1.00); total mortality per 244 g/day 1.01 (0.96-1.06, I2 94%). AMSTAR 2 of included meta-analyses: 11.1% high, 30.6% moderate, 38.9% low, 19.4% critically low; GRADE: none high, 41.6% moderate, 40% low, 18.4% very low.
- Absolute risks
not reported in full text
- Authors' conclusion
Milk consumption does more good than harm for human health; results support milk as part of a healthy diet; more well-designed RCTs are warranted. Allergy and lactose intolerance need caution.
6. Strengths
- Broad search of three databases with duplicate independent data extraction
- AMSTAR 2 and GRADE applied to each included meta-analysis
- Reports both beneficial and harmful associations in full tables
- No competing interests declared; non-industry funding
7. Weaknesses
- No registered protocol reported; English-language only; screening process by two reviewers not clearly described; list of excluded studies not given
- Only one (newest/largest) meta-analysis per outcome selected; overlap of primary studies between meta-analyses not assessed
- Most included meta-analyses rated low or critically low by AMSTAR 2 and no outcome graded high certainty; evidence almost entirely observational
- Milk only: cheese, yogurt and other dairy products explicitly excluded
- Text says milk was 'not related to hip fracture', although Table 3 shows a significant dose-response RR 1.09 (1.07-1.11) per 200 g/day; the conclusion 'more good than harm' is a qualitative tally without weighting
- Mixed metrics (highest vs lowest and various dose increments) reduce comparability; no absolute risks
8. Risk of bias
- Orienting tool
- AMSTAR 2 (applied to the umbrella review itself)
- Type of assessment
- structured assessment based on the full text by a single reviewer (no formal double review)
- Full formal assessment
- no
- Overall risk of bias
- high
- Rationale
Simplified single-reviewer AMSTAR 2 from the full text (2 critical weaknesses: no protocol, no list of excluded reviews; several critical items partially met) which corresponds to 'low' to 'critically low' confidence. A formal assessment needs two independent reviewers. Prior abstract-based entry did not contain a structured judgement.
| Domain | Judgement | Rationale |
|---|---|---|
| Item 1 PICO components | low | Exposure (milk) and outcomes (any health outcome) defined. |
| Item 2 protocol registered before review (critical) | high | No protocol or registration reported. |
| Item 3 explanation of study-design selection | low | Meta-analyses of any design included, rationale given. |
| Item 4 comprehensive literature search (critical) | some concerns | Three databases plus reference lists, but English only and no grey literature. |
| Item 5 duplicate study selection | some concerns | Disagreement resolution by third researcher mentioned; duplicate screening not explicitly stated. |
| Item 6 duplicate data extraction | low | Two authors extracted independently. |
| Item 7 list of excluded studies with reasons (critical) | high | Only flowchart; 44 of 85 eligible meta-analyses dropped by 'newest/larger' rule without listing. |
| Item 8 description of included studies | low | Tables give author, year, milk type, number/design of studies, cases, model, I2. |
| Item 9 risk-of-bias assessment of included reviews (critical) | low | AMSTAR 2 applied to each included meta-analysis (details in supplement). |
| Item 10 funding of included studies | high | Not reported. |
| Item 11 appropriate synthesis methods (critical) | some concerns | No pooling, which is appropriate; but selection of one estimate per outcome and no handling of primary-study overlap. |
| Item 12 impact of risk of bias on synthesis | some concerns | GRADE given, but conclusions not stratified by AMSTAR/GRADE level. |
| Item 13 risk of bias considered in interpretation (critical) | some concerns | Limitations discussed, but overall 'more good than harm' conclusion does not reflect that most evidence is low/very low certainty. |
| Item 14 heterogeneity discussed | low | I2 distribution reported and discussed. |
| Item 15 publication bias (critical) | some concerns | Egger results extracted where available; mostly not available; authors note unreported publication bias likely. |
| Item 16 conflicts of interest | low | COI and funding of the review reported. |
Full-text review (25 October 2026): Added search dates, eligibility (milk only; cheese/yogurt excluded), full effect sizes incl. harm signals (CHD mortality 1.04, prostate cancer mortality 1.50, hip fracture 1.09 per 200 g/day, Parkinson's, acne, infant iron deficiency), AMSTAR 2/GRADE distribution of included reviews, funding. Structured AMSTAR 2 of the umbrella review itself: overall changed to high risk of bias (low confidence); bias_risk some_concerns -> high; methodological_quality moderate -> low.
9. Assessment of the statistical interpretation
A collection of mostly observational relative associations of modest size (typically 4-20%), most rated low or very low certainty by the authors' own GRADE; not independent confirmations because outcomes come from single selected meta-analyses. Harm signals (prostate cancer, Parkinson's, acne, hip fracture per 200 g/day, infant iron deficiency with cow's milk instead of formula) coexist with benefit signals; no absolute risks.
10. Relevance to the specific claim
Overall balance for C-DRY-03, C-DRY-04, C-DRY-11; confirms some harm signals.
11. Independent contrary findings
n/a
12. Transferability to Germany and Europe
Moderate.
13. Overall methodological judgement
Useful map of the milk evidence that shows milk is not consistently harmful and is associated with several benefits, but methodological confidence in the umbrella review is low (no protocol, no overlap handling) and its 'more good than harm' conclusion is qualitative. It covers milk only, not yogurt or cheese.
| Multidimensional assessment | |
|---|---|
| Authenticity of the source | verified |
| Methodological quality | low |
| Risk of bias | high |
| Precision of results | not applicable |
| Directness | direct |
| Reproducibility | partly replicated |
| Consistency with the overall evidence | mixed |
| Regional transferability | moderate |
| Adequacy of the interpretation (by the archive) | not applicable |
14. Open questions
Overlap-corrected re-analysis; certainty-weighted synthesis; effects of fermented dairy (outside scope); dose-response for hip fracture and prostate cancer; transferability to Middle-East populations.
15. Linked fact checks
- Milk is a 'substitute for blood' and carries every hormone, pesticide and antibiotic the animal received; therefore milk must be avoided. – Studies contradicting or limiting the claim
- High milk consumption raises mortality (women drinking 3+ glasses/day had 93% higher mortality) and increases rather than prevents hip fractures. – Studies with mixed or unclear results
- Dairy consumption increases prostate cancer risk (7% per 400 g/day; 34% for more than 2.5 servings/day). – Studies with mixed or unclear results
- Fresh milk, yogurt (incl. laban rayeb, ayran, labneh) and white/cottage cheese are harmful and must be avoided, whereas aged cheeses (cheddar, mozzarella, romi), processed cheese triangles, cream (qeshta), butter and ghee are allowed. – Studies contradicting or limiting the claim