A study of 405,023 UK Biobank participants asked how endocrine, nutritional and metabolic conditions relate to cancer over time. The authors compared a first cancer diagnosis, a further primary cancer and death as separate outcomes. For the protein analysis highlighted in the same paper, 31,866 people contributed to the mediation models. Those numbers belong to different parts of the research.

THE SHORT READ
  • The main cohort analysis considered 25 endocrine, nutritional and metabolic conditions. Eight helped define four profiles for further comparisons.
  • The protein mediation analysis used 31,866 people. Another analysis, searching for metabolite candidates, used 171,240.
  • Associations varied across cancer transitions. The validation dataset came from the UK Biobank cohort.
THE STUDY AT A GLANCEMetabolic conditions and cancer transitions · Wu et al.
Paper date
20 November 2026
Main sample
405,023 UK Biobank participants
Protein mediation analysis
31,866 participants
Design
Prospective observational cohort

Original publication: 20 Nov 2026 · The date above refers to this brief.

THE NUMBERS, IN CONTEXT

Sample sizes for three analyses in the same paper

people contributing to each analysis

Cancer transitions405023
Metabolite candidates171240
Protein mediation31866
0420000
The smaller analysis samples are separate counts within the study. Each bar has its own denominator.
THE EXPERIMENT, EXPLAINED

Read each transition from its starting state

  1. Follow participants without a prior diagnosis

    The first transition of interest is a first primary cancer.

  2. Start again from the first primary cancer

    The model can track a later primary cancer or death from this state.

  3. Read death from the relevant state

    Check whether the comparison begins after one or multiple primary cancers.

Press-News reading diagram of the transitions. Death is possible from multiple states in the study model.

A new diagnosis and a later outcome have different starting points

The authors used a model that tracks movement between states. A first cancer diagnosis starts with people free of cancer in the model. The analysis of another primary cancer starts with people who already had a first primary cancer. Death can be compared from more than one starting state. The label on each transition tells you who could experience the event being counted.

Source 1 ↗

The profiles summarized combinations of conditions

The authors examined 25 conditions and selected eight to build four statistical profiles. Each profile reflected a pattern across multiple diagnoses, with one pattern dominated by type 2 diabetes and another by osteoporosis. The study checked whether the profiles and their associations were reproducible in a held-out dataset from the same biobank.

Source 1 ↗

The result can change when the outcome changes

In the validation set, estimated hazard ratios for a first primary cancer ranged from 1.151 to 1.412 across the four profiles compared with the reference group. For death after multiple primary cancers, the profile dominated by lipid and lipoprotein disorders had an estimate of 0.972 (95% CI 0.728 to 1.296). That interval includes 1. A single risk label would conceal the differences between these transitions.

Source 1 ↗

The molecular analyses have their own denominators

The protein mediation work used 31,866 people, numerically equivalent to 7.87% of the main cohort count. The metabolite investigation used 171,240 people to look for candidates. A mediation model uses assumptions about how diagnoses, molecules and outcomes are related. Observed data can support a proposed pathway. Establishing a biological mechanism requires additional evidence from other designs.

Source 1 ↗

What a useful validation would add

An external cohort could test whether the groups recur with a different mix of people and whether the transition estimates hold with a different health record system. If such profiles are used to make predictions, readers would also need to see how predicted probabilities compare with observed outcomes in new populations.

Source 1 ↗
TRACE THE EVIDENCE

Check the population and outcome behind each estimate

Publisher’s version of record. Checked participant selection, the multistate design, Figure 4 results and the reported protein and metabolite analyses. The underlying data and code were outside this check.

01Does the main cohort size apply to the protein results?

What was observed
The main sample included 405,023 people. The protein mediation analysis included 31,866.
Where the conclusion stops
The quality and generalizability of the molecular result depend on the subset with the relevant data.

Source 1 · Methods, participant inclusion and Results, protein mediation

02Does a profile have the same association at every stage?

What was observed
In the validation set all four profiles were linked to a first primary cancer. The lipid profile estimate for death after multiple primary cancers included 1 in its interval.
Where the conclusion stops
A different starting state and outcome change who is compared and what the estimate means.

Source 1 · Results, Figure 4D

Numbers you can inspect

MeasureValue & unitOrigin & method
Main cohort405023 peopleReported Participants in the main UK Biobank analysis.
Source 1 · Methods, participant inclusion and exclusion
Conditions studied25 conditionsReported Endocrine, nutritional and metabolic conditions examined.
Source 1 · Abstract and Results
Conditions used for profile definition8 conditionsReported Selected indicators for the four profiles.
Source 1 · Results, profile definition
Statistical profiles4 profilesReported Diagnosis profiles C1 through C4.
Source 1 · Results, Figure 4
Protein mediation sample31866 peopleReported Participants in the protein mediation analysis.
Source 1 · Results, Figure 6D–E
Protein mediation count divided by main sample count7.87 percentCalculated 31,866 divided by 405,023, times 100, rounded to two decimals.
Source 1 · Methods and Results, protein mediation
Metabolite candidate sample171240 peopleReported Participants in the metabolite analysis described in Results.
Source 1 · Results, identification of metabolic mediators
Smallest first cancer hazard ratio in validation set1.151 hazard ratioReported Lower end of the reported range across four profile estimates.
Source 1 · Results, Figure 4D
Largest first cancer hazard ratio in validation set1.412 hazard ratioReported Upper end of the reported range across four profile estimates.
Source 1 · Results, Figure 4D
Lipid profile, death after multiple primary cancers0.972 hazard ratioReported Validation estimate. 95% CI 0.728 to 1.296.
Source 1 · Results, Figure 4D

Compare the actual experiments

These studies answer different questions. Read the unit and endpoint before comparing results.

StudyUnit & settingReadoutInterpretation boundary
Counting a cohort and its molecular subsets

Source 1 · Methods and Results, protein mediation

Participants with different measurement setsDisease outcomes and molecular assaysBiomarker analyses have their own denominators even in large biobanks.
Reading time-to-event comparisons

Source 1 · Results, Figure 4D

Biobank participants and participants in a review of imaging age studiesTransition-specific hazards and all-cause mortality hazardsEach hazard ratio belongs to an outcome and a defined starting group.
Download evidence table (CSV)

The export includes claims, available numbers, methods and source locations. It contains our reading notes and published summaries; it is not raw participant data or an independent reanalysis.

Evidence update · 21 Nov 2026
First publication. Any calculations in this brief use reported figures. The reading notes identify possible follow-up questions. AI source check; no human editorial or clinical review.

CONNECT THE EVIDENCE

Which participants underpin each result?

AnalysisParticipantsQuestion addressed
Cancer transitions405,023Track recorded diagnoses and deaths across follow-up.
Protein mediation31,866Evaluate protein candidates in statistical mediation models.
Metabolite candidate analysis171,240Look for metabolite patterns linked to the profiles.

Counts come from participant selection and from the protein mediation and metabolite candidate sections.

READER QUESTIONS

Your questions, answered

Was protein mediation tested in all 405,023 people?

The main cohort covered 405,023 people. The protein mediation analysis used 31,866 participants.

What does a transition measure?

It describes a change from one defined state to another, such as moving from no recorded cancer to a first primary diagnosis.

Did a new profile mean the same association at every transition?

The estimates differed across outcomes. For one profile, the estimate for death after multiple cancers had a confidence interval that included a hazard ratio of 1.

What does the validation set establish?

It tests whether analyses built in one subset give consistent results in another subset of the biobank. An independent cohort would test the results in a different setting.

LIMITATIONS

Limits of this interpretation

  • The main data come from an observational UK Biobank sample. Held-out participants come from the same underlying cohort.
  • The molecular analyses depend on who had the relevant measurements and on model assumptions.
  • Hazard ratios compare modeled rates for specific transitions. An absolute probability would need follow-up timing and a starting risk.
SOURCE NOTES

Sources & transparency

  1. Wu, Yao, Chu et al. (2026) — Endocrine, nutritional and metabolic diseases drive cancer risk through subtype-specific and shared proteomic mediators

    Publisher’s version of record. Checked participant selection, the multistate design, Figure 4 results and the reported protein and metabolite analyses. The underlying data and code were outside this check. · Accessed 21 Nov 2026

    DOI: 10.1038/s41467-026-78791-4

Prepared and source-checked with AI for Press-news Team. The source notes identify the material available during preparation. The article received no human editorial or clinical review. The source list supports the reported results and the calculations shown.

Source check: AI source check — primary research and stated source limits

Clinical review: Not applicable to this educational guide

Suggest a correction