The instalment that keeps the others honest

Six parts of mostly good news. Now the other column.

This paper makes a strong claim in its conclusion. It calls for the combination to be used "early and simultaneously" and says it should "perhaps be considered for first-line treatment" in high risk patients. That is a big ask of doctors. So it is fair to ask whether the evidence underneath can carry that weight.

My honest answer: not yet. Here is why.

The authors' own grade is "low"

This is not me being harsh. It is the paper's own verdict.

The authors ran a GRADE assessment, the standard system for rating how confident we can be in a body of evidence. For all-cause mortality, HbA1c, BMI, blood pressure, eGFR and cholesterol, certainty was rated low. For fasting glucose, very low.

GRADE's own wording for "low" is that the true effect may be substantially different from the estimate. For "very low", the true effect is likely to be substantially different.

So the paper grades its own evidence as low certainty, and then concludes with a call to action. Those two things sit uncomfortably together. The data says "this might be right". The conclusion reads like "do this now".

The mortality result rests on two studies that disagree

Part 3 flagged this. It deserves more weight here.

The headline 0.49 figure comes from only two studies. The I² for that pooled result was 93%, which means the two studies disagree substantially on the size of the effect. When you pool two studies that disagree that much, the single number you get out the other end is more of an average of two different stories than a precise measurement of one.

On top of that, the authors state the mortality values are unadjusted. They did not, and could not, correct for differences in the patients who ended up on the combination versus a single drug.

Why "unadjusted" is such a big deal

Think about who gets prescribed two newer, pricier diabetes drugs at once.

Often it is people with good insurance or good public coverage, people attending specialist clinics, people whose doctors follow the latest guidelines closely, people well enough organised to fill two prescriptions and keep taking both. Every one of those traits is associated with living longer, regardless of which drugs they take.

This is called confounding by indication, and a close cousin, the healthy adherer effect. Observational data is always vulnerable to it. Adjustment tries to correct for it. Here, the mortality numbers were not adjusted.

The Jensen study itself hints at the complexity. The people on the combination in that registry had been diabetic longer (9 years on average versus 6 or 7 for the single-drug groups). That is a different population, and the direction of the bias is not obvious.

Most of the numbers are before and after, not versus

This one is easy to miss and it matters.

The mortality result compares the combination against single drugs. But the HbA1c, BMI, blood pressure, glucose and eGFR results mostly compare people on the combination at six months against themselves at the start. There is no parallel control group in that analysis.

Before and after comparisons always look flattering. People often start a new treatment when their numbers are at their worst, and numbers at their worst tend to drift back toward average on their own. That is regression to the mean. People also change their diet, exercise and other habits after a fresh diagnosis or a doctor's warning. A before and after design cannot separate any of that from the drugs.

Some of the evidence is thin

  • Three of the thirteen studies scored 5 or lower on the quality checklist, rated low quality by the authors.
  • Several studies did not state strategies to deal with confounding at all.
  • The smallest included study had 15 patients.
  • Only English language papers were included.
  • There were too few studies to test for publication bias, so we do not know whether negative findings went unpublished.
  • Cardiovascular death specifically could not be analysed. The paper's title emphasises cardioprotective effects, but the cardiovascular mortality meta-analysis was not possible.

A mismatch in the terms

The results section and abstract call the mortality figure an odds ratio of 0.49. The GRADE summary table lists it as a relative risk (RR 0.49). These are different measures. When an outcome is uncommon they come out close, so it probably does not change the conclusion much, but a careful reader should notice that the same number is described two different ways in the same paper.

One contradicting study

The discussion itself mentions an analysis of three US claims datasets (12,584 people) where adding an SGLT2 inhibitor to a GLP-1 drug reduced major cardiovascular events and heart failure hospitalisations but did not reduce all-cause mortality. That study was not part of the meta-analysis, but it is a reminder that not every dataset agrees with the headline.

Who wrote it, and who helped

The paper declares no funding and no competing interests, which is good. It also acknowledges medical writing support and editorial assistance from BioQuest Solutions. That is common and not sinister, but it is worth knowing that professional medical writers helped shape the text.

So where does that leave us?

Steelman the paper and you get this: two very large real-world datasets both show a big survival advantage, the direction matches strong randomised trial evidence for each drug alone, and there is no obvious safety signal. That case is real.

Steelman the critique and you get this: the survival advantage is unadjusted, drawn from two disagreeing studies, in populations that likely differ in ways that also predict survival, and the other outcomes are uncontrolled before and after comparisons rated low certainty by the authors themselves.

Both are true. The honest conclusion is "promising, plausible, not proven". The call for first-line use outruns that. What would settle it is the thing that does not yet exist: a randomised trial of the combination against each drug alone, with mortality as an outcome.

Until then, any time someone tells you the combination "halves your risk of death", you now know exactly how much weight that sentence can hold.

This series discusses a peer-reviewed meta-analysis of observational studies. Nothing here is medical advice. Do not start, stop or combine prescription medications without your doctor.