An A is possible
In 2011, The Lancetpublished Rothwell et al., an individual-patient-data analysis of randomised trials (effect of daily aspirin on cancer mortality). Publi-Score scores it 79/100 — an A.
Effect sizes reported with confidence intervals, individual patient data (the gold standard for meta-analyses), randomised trials against control, an interaction test on treatment duration. The “level of evidence” criterion gets the maximum. No red flags.
This matters for what follows: our scale does not penalise a journal's prestige, and it can recognise solid work. An A exists, in The Lancet, and it is demanding. Keep it in mind.
Monteiro 2025: what the method says
The article (Monteiro et al., The Lancet2025; 406:2667-2684) is the first of a three-part Series. It combines narrative reviews, systematic reviews and meta-analyses to support a thesis: that the displacement of traditional diets by ultra-processed foods is a key driver of chronic disease. Publi-Score: 43/100, tier D.
This score is not a judgement on the importance of the topic, nor on the authors' reputation. It describes what the method, read criterion by criterion, allows one to assert:
- Structural conflict of interest. Several lead authors developed the Nova classification — the very tool used to classify the studies included in the synthesis. This is not a classic financial COI; it is a potential bias in selection and interpretation, declared but structural.
- Observational evidence, causal claims.Most of the data is epidemiological (prospective cohorts). The article concludes that UPF are a “key driver” — a causal term that observational designs cannot formally establish.
- Over-interpretation (spin).The discussion states that “the need for further evidence should not delay public health action” — wording that goes beyond what the presented data establish.
- Real, acknowledged rigour. Conversely: PROSPERO pre-registration, declared PRISMA and MOOSE compliance, Newcastle-Ottawa quality assessment, dual independent extraction. These points raise the score — the grid is not one-sided.
The D does not mean “bad article”. It means: the strong causal conclusions are not supported at the claimed level of evidence, and a structural bias weighs on the whole. That is information the name “Lancet” alone does not provide.
And a randomised trial? Same fragility
One might expect a randomised controlled trial on UPF to score higher. Dicken et al. 2025 (Nature Medicine), a crossover RCT on the same topic, scores 50/100, tier C.
C is not “good”. Neither the meta-analysis's great prestige nor the RCT's controlled design reaches high reliability here. This is not a match with a winner: both illustrate the same thing — the scientific evidence on ultra-processed foods is, to date, structurally fragile.That is exactly what a methodological score is for: making it visible rather than deferring to the journal's prestige.
Our criteria are public. Let's discuss them.
Everything above is derived from explicit criteria, identical for Rothwell (79/A), Monteiro (43/D) and Dicken (50/C). The same grid, with no prestige exemption. The criterion-by-criterion detail is available on each article's score page.
We do not claim to hold the truth. We claim to make the method legible and contestable. If you are a researcher, a journalist, or an author of one of the evaluated articles and want to discuss the scoring methodology — we are available to talk with anyone interested.
A score is not a final verdict: it is a transparent methodological reading, open to discussion. It is that transparency — not the prestige of the venue — that makes reliability.
See the detailed score for Monteiro 2025 →
Discuss the methodology with us →
Understand the grid: publi-score.ai/methodology