Metformin in STAMPEDE's Nonmetastatic Cohort

Why Study Metformin in Nonmetastatic Prostate Cancer?

Androgen-deprivation therapy (ADT) is an important part of treatment for many men with high-risk nonmetastatic prostate cancer. It can also alter body composition and glucose metabolism, increasing the clinical relevance of metabolic health during treatment. Metformin, a widely used glucose-lowering medicine, therefore had a plausible rationale for investigation alongside standard prostate cancer care. That rationale is a reason to test the intervention, not evidence that it improves cancer outcomes.

STAMPEDE evaluated metformin as an addition to standard care in men with prostate cancer, including a nonmetastatic group. The case study question is narrower than whether metformin is useful for diabetes or metabolic health: did adding metformin improve cancer-related outcomes for the nonmetastatic patients represented in this comparison? Answering that requires keeping the enrolled population, the randomized treatment contrast, and the outcome definitions distinct.

Clinical Question and Trial Context

The nonmetastatic patients in STAMPEDE were men with clinically significant, locally advanced or otherwise high-risk prostate cancer for whom long-term ADT-based treatment was appropriate. This was not a sample of all men with newly diagnosed prostate cancer: men with low-risk disease, and men with distant metastases, do not represent this particular cohort. The metformin comparison also excluded patients with diabetes, so its results should not be taken as a trial of metformin for diabetes management.

Participants were assigned to standard care or standard care plus metformin. Standard care was based on ADT, with additional local treatment such as radiotherapy used as appropriate in this setting. Metformin was given at a target dose of 850 mg twice daily, with treatment intended for up to two years. The clinically relevant contrast is therefore addition of metformin to the treatment strategy used in the trial, not metformin versus no cancer treatment.

The biological rationale included the possibility that modifying insulin-related and other metabolic pathways could influence tumour growth, as well as the practical interest in countering metabolic changes associated with ADT. These mechanisms are plausible but do not establish a clinical effect. A randomized comparison is needed because metabolic markers, tumour progression, and survival are different outcomes.

Outcomes and Reported Efficacy

The efficacy findings must be read by outcome and disease setting. Time-to-event measures such as failure-free survival (FFS) summarize the time until a protocol-defined treatment failure event. Overall survival (OS) is the time until death from any cause. FFS can register disease progression or other defined failure earlier than OS, while OS is clinically direct but may require longer follow-up and can be affected by subsequent treatments and competing causes of death.

In the reported STAMPEDE metformin comparison, the overall randomized population did not show a statistically significant OS improvement with metformin. The nonmetastatic analyses showed a more favorable efficacy signal than the metastatic setting, including favorable time-to-event findings. This pattern is clinically interesting, but it should not be simplified to “metformin improves survival for all patients.” The strength of a subgroup conclusion depends on the estimate’s precision, the analysis plan, and evidence that treatment effects differ between disease settings.

A favorable estimate within the nonmetastatic group is not, by itself, proof that the effect differs from the metastatic group. The relevant comparison for effect modification is an interaction between treatment and metastatic status, not whether one subgroup’s p-value is below a threshold and another subgroup’s is not. Nor does a statistically significant time-to-event result imply a large absolute benefit: the baseline event rate and length of follow-up are needed to judge clinical magnitude.

Why the Result Needs Careful Reading

The nonmetastatic result matters because it concerns a group for whom long-term disease control is a central goal and ADT-associated metabolic effects are clinically relevant. Still, several checks are essential before using the result to guide interpretation or decisions:

• Confirm the denominator. Check that the estimate refers to the nonmetastatic cohort, not the whole randomized population or the metastatic subgroup.
• Check the endpoint. FFS, progression-free measures, and OS answer different questions. Do not substitute one for another in a summary.
• Read the confidence interval. The point estimate alone hides how compatible the data are with smaller, larger, or absent effects.
• Check follow-up and event definitions. A time-to-event endpoint depends on when events are counted and how censoring is handled.
• Separate efficacy from tolerability and adherence. An assigned treatment can have a different effect in practice if patients stop it, miss doses, or experience adverse effects.

These checks are particularly important in a platform trial. The platform can evaluate several treatment comparisons over time, but each comparison has its own eligibility, randomization, control experience, and follow-up. A result from one comparison should not automatically be transferred to another arm or to all participants in the platform.

Justifying the Subgroup Interpretation

A concise way to reason about subgroup findings is to separate two questions. First, is there evidence of a treatment effect within the nonmetastatic group? Second, is the treatment effect different between the nonmetastatic and metastatic groups? The first concerns an estimate within one group; the second concerns a contrast between estimates.

For example, let \(\theta_0\) be the log hazard ratio for metformin versus control in nonmetastatic disease and \(\theta_1\) the corresponding log hazard ratio in metastatic disease. The interaction contrast is \(\theta_0-\theta_1\). A test of this contrast addresses whether the estimated relative effects differ across settings. It is not equivalent to testing \(\theta_0=0\) and \(\theta_1=0\) separately: one subgroup can have a small p-value and the other a large one even when their treatment effects are not demonstrably different.

This distinction explains why the nonmetastatic signal should be described as evidence in that cohort, while claims that metformin works only in nonmetastatic disease require stronger interaction evidence. A subgroup estimate can be useful even when the interaction is uncertain, but its uncertainty and exploratory status should remain visible.

Limitations and Common Mistakes

The trial informs a specific treatment decision in a selected clinical setting. It does not prove that metformin should be prescribed as an anticancer drug to all men with prostate cancer, nor does it replace its established role in managing diabetes. Applicability can be limited by differences in disease risk, comorbidity, concomitant treatment, adherence, and contemporary standards of care. In addition, trial-level average effects do not guarantee benefit for an individual patient.

• Wrong: “The nonmetastatic finding was favorable, so metformin reduces mortality for every man with prostate cancer.” Correction: State the disease setting and endpoint, and distinguish the subgroup result from the overall population and other settings.
• Wrong: “The relative hazard reduction is the same as the absolute risk reduction.” Correction: A hazard ratio is a relative time-to-event measure; calculate absolute benefit from time-specific risks.
• Wrong: “A significant result in one subgroup and a nonsignificant result in another proves different effects.” Correction: Examine the treatment-by-subgroup interaction and its uncertainty.
• Wrong: “A plausible metabolic mechanism confirms the clinical benefit.” Correction: Mechanistic rationale motivates a trial; randomized outcome data determine whether efficacy was demonstrated.