Evidence mapA paper, made visible 2023 · ASPREEFull paper record

Aspirin · Metformin · Cancer

Two medicines.
Two different kinds of evidence.

Could these medicines change cancer outcomes in older adults with diabetes?

2,045 people with diabetes About 4½ years A secondary analysis of a larger trial
THE FOUR FINDINGSChoose one to see how it works

AspirinAssigned by chance · in metformin users

MetforminObserved use · all participants with diabetes

ASPIRIN + METFORMIN VS PLACEBO + METFORMIN

What happened when people received aspirin?

Outcome: death from cancer
1

A medicine assigned by chance

965older adults with diabetes, using metformin
Aspirin or a look-alike pill

Neither people nor researchers knew which pill.

Aspirin516 people · 100 mg daily
Placebo449 people · no active aspirin
About 4½ years of follow-upMedian 4.6 years in metformin users

Chance assignment makes the aspirin comparison more informative about cause. This paper examines a small subgroup of a larger trial.

2

Who died from cancer?

Two equal-sized pictures. Each represents 100 people.

Aspirin

4.8in 100

25 of 516 in the actual group

Placebo

2.0in 100

9 of 449 in the actual group

Died from cancerNo recorded cancer death
+2.8
more cancer deaths per 100Observed group difference during follow-up

Pictures rescale the recorded counts to the same group size. They are not a prediction for you, and do not adjust for follow-up time or other differences.

3

How much can we trust the direction?

← Fewer cancer deathsMore →

The range points to more, but the size is uncertain.

Estimate95% uncertainty range2.53× (1.18–5.43×)
What does this ruler mean?

The paper compares how quickly cancer deaths occurred during follow-up, using a hazard ratio. A value of 1 means the same hazard. The line shows the reported 95% confidence interval; wider means less precision. It describes uncertainty under the statistical model, not every possible bias. This is a different measure from the simple percentages in the people pictures.

4

What can we actually say?

A warning signal worth investigating.

More cancer deaths occurred with aspirin in this subgroup. Only 34 cancer deaths inform this comparison, so the size of the effect is uncertain.

Check the numbers in Table 3
1 of 5 views
See the whole study, including the other aspirin subgroup+
19,114people in the parent ASPREE trial2,045with diabetes in this analysis

Metformin use was already established at enrollment. Aspirin or placebo was assigned by chance across the trial.

Already using metformin 965 people

Aspirin516 people

56 cancer diagnoses
25 cancer deaths

Placebo449 people

45 cancer diagnoses
9 cancer deaths

Not using metformin 1,080 people

Aspirin508 people

70 cancer diagnoses
22 cancer deaths

Placebo572 people

74 cancer diagnoses
22 cancer deaths

The aspirin hazard ratios in non-users of metformin were 1.10 (0.79–1.52) for cancer diagnoses and 1.16 (0.64–2.09) for cancer deaths. These ranges include no difference. Comparing one subgroup’s “significant” result with another’s “not significant” result does not establish that the effects differ.

Source: Tables 1 and 3 ↗
See what makes the conclusions less certain+

Existing differences

Metformin users differed in age, other medicines and health measures. Statistical adjustment cannot guarantee that every relevant difference was removed.

Medicine use changed

Groups were defined at enrollment. During follow-up, 11% of metformin users stopped for at least a year; 22% of initial non-users began taking it.

A smaller, secondary analysis

This paper reanalyzes a subgroup of the trial. Few cancer deaths and several comparisons make chance findings a concern. The whole trial’s enrollment is not this comparison’s sample size.

Not all outcomes are here

This paper focuses on cancer diagnoses and deaths. It does not by itself decide whether either medicine’s total benefits outweigh its harms for a particular person.

Paper, method and source trail+

Association of metformin, aspirin, and cancer incidence with mortality risk in adults with diabetes.

Orchard and colleagues · JNCI Cancer Spectrum · March 2023 · PMID 36857596

This visual reading is curated for this paper from its full text. Group sizes come from Table 1; metformin comparisons from Table 2; aspirin comparisons from Table 3. Percentages in the people pictures are calculated as recorded events ÷ participants at enrollment × 100. They are crude observed proportions, not time-adjusted risks. Hazard ratios and confidence intervals are reproduced separately from the paper.

About 19% of participants had a personal cancer history at enrollment. Cancer diagnoses refer to each person’s first recorded in-trial cancer, not necessarily their first-ever cancer. Cancer deaths are a separate outcome; the two counts should not be added together. The analysis used Cox regression, with measured confounder adjustment for the metformin comparison.

Numbers trace back to the paperCurated from the full text · Group sizes, recorded events and reported estimates