Metformin: Effects & Evidence Base — what’s proven and what isn’t

Metformin is often discussed in everyday life as a “drug with many side ideas.” In research, however, the situation is much more structured: For type 2 diabetes, the evidence is particularly robust, while additional benefits in other indications (e.g., osteoarthritis, central arterial stiffness, colorectal cancer survival) do appear in meta-analyses, but the strength can vary substantially depending on the endpoint and study design. For pregnancy and chronic kidney disease, systematic reviews on safety are the key reference point.

First lifestyle levers: where metformin truly makes sense in daily life

Metformin is not a substitute for lifestyle. When it comes to blood glucose, insulin resistance, and “inflammation reality,” movement, weight management, and diet are usually the strongest levers—and the medication dose can only be meaningfully interpreted (depending on the person) in that context. Metformin complements these steps, especially in type 2 diabetes, not the other way around.

Practically, that means: if you are considering metformin (or any other antidiabetic therapy), it’s worth checking the baseline levers first. In many cases, weight loss and consistent physical activity reduce insulin requirements—so metformin may start later, need a lower dose, or its effect may “unfold” more clearly. Studies on lifestyle interventions are a topic of their own—however, methodologically the rule is: the effect of lifestyle on metabolic markers is often clinically relevant, while added benefits of metformin outside diabetes can be more inconsistent across meta-analyses.

Even for “indirect” effects through sleep, light, and stress: there are plausible biological pathways, but the strength and consistency for specific clinical endpoints depend heavily on what the studies actually measured. Therefore, the most important connection for you is this: metformin influences metabolic pathways, but your starting point (body weight, activity, eating pattern, sleep quality) largely determines how large the additional benefit of a medication is in real life.

If metformin is being discussed because of osteoarthritis, the methodological separation is crucial: observational data may show associations, but to evaluate progression, pain, and functional capability, you need meta-analyses that aggregate carefully defined endpoints. This is exactly where the evidence base below can help you interpret the data—while lifestyle remains the baseline. For comparison, it can also be useful to look at other evidence-based “supplement/drug-adjacent” topics, e.g., Glucosamine: Effects & Evidence Base — what’s proven and what isn’t.

Effects in type 2 diabetes: what’s best supported based on evidence

In type 2 diabetes, metformin is the most thoroughly studied medication in your list of outcomes. However, the meta-analyses relevant here often focus on secondary effects (e.g., arterial stiffness, osteoarthritis, cancer survival) rather than primarily on the HbA1c core effect. For interpretation, it is decisive whether the included studies were controlled and what the baseline risk was.

The key point upfront: the works you provided (e.g., those related to osteoarthritis, vascular function, colorectal cancer) do not automatically constitute “proof of a diabetes-primary effect.” Instead, systematic reviews and meta-analyses show that metformin may be associated with certain downstream or proxy endpoints. This is scientifically legitimate, but it changes the question: is the focus on hard clinical events, or more on measurable functional markers? And: are the effects large enough to be considered clinically relevant?

In such meta-analyses, you should unpack the study details:

• Which endpoints were measured? For “central arterial stiffness,” this is more a physiological function than “death/heart attack.”
• How was metformin used? Dose, treatment duration, and co-medication often vary.
• How strong were the confounding controls? If many data come from observational studies, “metformin use” can act as a marker of access to care, disease course, or treatment adherence.
• What was the baseline risk? An apparent effect can differ across groups with different underlying risks.

It also matters to distinguish “surrogate” from “clinical event.” Even if a meta-analysis finds an effect on a proxy measure, that does not necessarily mean the same level of clinical relevance (e.g., fewer events or better survival rates) applies. That is the alignment you should make in the next sections when you evaluate specific outcomes such as osteoarthritis or colorectal cancer survival (the framework is provided by the relevant meta-analyses themselves: e.g., (Yao et al., 2026, PMID 41067453), (Kao et al., 2026, PMID 41506068), (Rahmanian et al., 2026, PMID 41496261)).

Evidence hierarchy: RCTs, observational and animal data — why it matters for metformin

Meta-analyses are only as strong as the studies they include. If the synthesis is dominated by observational data, residual confounding (e.g., due to differences in health behavior) remains an issue. RCTs are generally considered more causally robust, whereas systematic reviews on safety often bundle heterogeneous study designs.

Why this is especially important for metformin: metformin is not used for a single indication. Public discussion quickly turns into “metformin can also do X,” while the background evidence often comes from different settings—and endpoints range from lab markers to major clinical events. For you, that means every claim about effects should be checked against which level of evidence it actually draws on.

• RCTs (randomized controlled trials): When available, they provide the most causally reliable evidence. For many added benefits outside diabetes, however, RCTs are less common or not yet consistent enough.
• Observational studies: They are valuable for discovering signals or capturing rare situations. But selection bias and residual confounding can distort results.
• Animal and mechanistic studies: They support biological plausibility, but they do not replace clinical safety or outcome data in humans. Particularly for pregnancy and kidney-risk contexts, mechanistic ideas are not a safety proof.

Meta-analyses cannot “magically” close this gap—they aggregate what exists. That’s why the right question is not only “Is there a meta-analysis?” but also:

1. Which endpoints?
2. Which study designs?
3. How large and how consistent are the effects?
4. How strong was the methodological quality or risk control in the original studies?

If you also consider safety aspects, you should focus especially on systematic reviews that examine humans directly. For pregnancy, for example, (Brinkmann et al., 2026, PMID 41354637) provides a systematic, meta-analytic evaluation of maternal, neonatal, and long-term outcomes in gestational diabetes. And for chronic kidney disease, (Patiño-Cardona et al., 2026, PMID 41548682) assesses safety in this specific risk context. Such works are methodologically more designed to answer “could it cause harm?” more cleanly than RCT or mechanistic extrapolations.

Evidence on concrete outcomes: osteoarthritis, vessels, and lifespan

Metformin shows associations in multiple systematic reviews and meta-analyses with osteoarthritis outcomes, vascular function (central arterial stiffness), and survival in colorectal cancer. At the same time, effect strength varies by endpoint and study design; clinical relevance is not automatically the same as statistical significance.

Let’s start with osteoarthritis: For knee and hip osteoarthritis in people with type 2 diabetes, there is a systematic review and meta-analysis on incidence, progression, and the risk for joint replacement (Yao et al., 2026, PMID 41067453). In addition, there is a meta-analysis specifically on the effectiveness of metformin on pain, function, and quality of life in knee osteoarthritis (Kao et al., 2026, PMID 41506068). A further meta-analysis exists on effectiveness in managing osteoarthritis in type 2 diabetes (Chen et al., 2026, PMID 41649758). This suggests: it’s not just “a single analysis”—there are multiple syntheses.

Methodologically, for osteoarthritis you should separate two things:

• Course/progression and intervention risk (harder endpoints, but often dependent on data quality and definitions),
• Patient-reported outcomes such as pain/function (clinically relevant, but sensitive to study design, measurement instruments, and study duration).

For vascular function, there is a meta-analysis on central arterial stiffness in people with type 2 diabetes (Weningtyas et al., 2026, PMID 41401897). “Central arterial stiffness” is a physiological measurement and can be understood as a consequence of metabolic processes and vascular aging. Whether and how strongly this translates into hard clinical endpoints is not automatically readable. This is exactly where an evidence-strength assessment is needed: an effect on a functional marker is not equivalent to “fewer strokes.”

For lifespan/survival in colorectal cancer, (Rahmanian et al., 2026, PMID 41496261) provides a systematic review and meta-analysis on the influence of metformin on survival outcomes. The decisive interpretation question is: how well were tumor stage, treatment context, and ongoing therapy co-management controlled? In cancer survival, differences in disease stage or treatment pathways can strongly determine the results.

Here is a compact overview of the outcomes you provided:

Important: even if meta-analyses find “an effect,” you must compare effect size, heterogeneity, and whether the endpoint is a surrogate vs. clinically meaningful. That’s the methodological obligation when metformin is discussed beyond diabetes.

Safety in special situations: pregnancy and chronic kidney disease

For pregnancy (especially gestational diabetes) and for chronic kidney disease, metformin safety depends strongly on the individual baseline situation. For gestational diabetes, there is a systematic review and meta-analysis on maternal, neonatal, and long-term outcomes (Brinkmann et al., 2026, PMID 41354637). For chronic kidney disease in type 2 diabetes, (Patiño-Cardona et al., 2026, PMID 41548682) evaluates safety in the corresponding risk bracket.

For pregnancy, the key question is not only “is metformin fundamentally bad?” but also: which endpoints were investigated (maternal complications, neonatal outcomes, long-term development), and how well were differences between groups accounted for. In the provided evidence, (Brinkmann et al., 2026, PMID 41354637) bundles exactly these dimensions within a systematic framework. Even so, individualized medical decision-making remains central: gestational diabetes is not a homogeneous condition, and differences in treatment intensity, glucose control, and gestational week can influence outcomes.

For chronic kidney disease, an additional safety factor applies: renal clearance and the risk of medication-associated side effects depend on kidney function. (Patiño-Cardona et al., 2026, PMID 41548682) systematically evaluates metformin safety in this context. For you, a practical principle follows: “safe” is not absolute—it depends on the degree of kidney function and co-factors (e.g., co-medication and acute risk situations). If you are in this situation, the decision to continue/change should always be made based on your kidney values, risk profile, and treatment goals.

Here is an important methodological add-on: safety data in meta-analyses can only be as good as the definitions in the original studies and the comparability of the included groups. Therefore, you should not read safety reviews as blanket clearance, but as an evidence-based framework for the clinician’s risk assessment.

If you want, I can formulate a checklist for the doctor visit next (pregnancy/kidney: which values and questions are useful).

Carcinoma and developmental data: colorectal cancer and congenital malformations

For cancer survival, there is systematic evidence, and for prenatal exposure there is a meta-analysis plus drug-target Mendelian randomization regarding congenital malformations. Still, interpretation differs: survival data are clinically direct, while MR approaches require additional assumptions and do not replace RCTs 1:1.

For colorectal cancer survival, the relevant synthesis is (Rahmanian et al., 2026, PMID 41496261). Again, the strength of the result depends on how well tumor stage, treatment pathways, and confounding were controlled. For oncology endpoints, differences in disease characteristics and treatment are especially common. That makes clinical interpretation demanding: a “survival signal” is not automatically equivalent to a causal benefit for all patients.

Additionally, there is a meta-analysis plus drug-target Mendelian randomization for developmental data on fetal congenital malformations across 11 organ systems (Ji et al., 2026, PMID 41628713). This is methodologically interesting because MR approaches try to support causality more strongly than pure observational data—but it relies on assumptions about the instrumental variable. Therefore, MR is not “automatic safety” and does not replace clinical evidence from randomized settings.

What you can practically take from this:

• If metformin is already used in pregnancy or discussed there, the systematic review for gestational diabetes (Brinkmann et al., 2026, PMID 41354637) supports risk assessment across multiple outcome categories.
• If the focus is prenatal exposure and malformations, (Ji et al., 2026, PMID 41628713) provides a broader evidence line across many organ systems, but causal interpretation depends on MR assumptions.

For counseling, that does not mean “ignore” or “overweigh,” but rather read differently: MR can support hypotheses, but safety for a specific individual remains a question of that person’s risk situation, timing aspects during pregnancy, and the totality of evidence.

What you should take away from this

• Lifestyle first: metformin complements type 2 diabetes—the biggest levers remain movement, weight management, and diet.
• Added benefits are inconsistent: for osteoarthritis, central arterial stiffness, and colorectal cancer survival, there are meta-analyses, but clinical meaning must be evaluated endpoint-specifically.
• Check the evidence hierarchy: a meta-analysis can exist, but what matters is whether the data are mostly from controlled studies or from observational research.
• Safety is context-dependent: for pregnancy (gestational diabetes) and chronic kidney disease, systematic reviews support the assessment—the concrete decision remains medical because kidney function and individual risk profile determine the outcome.
• MR is not a substitute for RCTs: development/malformations are investigated using MR approaches among others—this can provide useful hints, but it is not a blanket clearance.