A Quarter as Likely to Be Offered It

A meta-analysis of 49 studies covering 5.5 million people with diabetes found that those who also had a mental health diagnosis received worse diabetes care across almost every measure — less monitoring, fewer tests, and a quarter of the odds of being prescribed a GLP-1.^[1] They were, however, half again as likely to be put on insulin.

What was measured

The review pooled cohort and case-control studies across Asia, North America, Europe and Australasia, covering 5,503,712 people with diabetes of whom 838,366 — 15.2% — had a diagnosed mental disorder. The comparison was the quality of diabetes care each group received, using standard monitoring indicators.

The insulin finding carries the argument better. Ten studies, an odds ratio of 1.52, and it points the same way: this group gets the older drug and not the newer one.

Why this is worse than it first appears

Consider what else is true of this population. Many are taking antipsychotics, and the most effective ones are the hardest on metabolism — clozapine and olanzapine drive weight gain and insulin resistance directly. Insulin itself causes weight gain. So the group being disproportionately prescribed insulin is the group already gaining weight from their psychiatric treatment.

More likely to be given the drug that adds weight, less likely to be offered the one that removes it.

And it is not that the treatment fails in them. A randomized trial of semaglutide added to clozapine or olanzapine improved blood sugar, weight and waist circumference with no worsening of psychiatric symptoms — antipsychotics and metabolic harm. The evidence that it works in exactly these patients exists.

People with serious mental illness die substantially earlier than the general population, largely from cardiometabolic disease. This analysis describes the care that gap is built from, one missed test and one unoffered prescription at a time.

The other thing this measures

That matters for reading almost every observational result on this register. When a claims-database study reports that GLP-1 users had far lower mortality than users of another diabetes drug, part of what it is measuring is documented here: the two groups were selected differently, along an axis that independently predicts dying.

The general problem is in when a hazard ratio is too good, and what a study can do to earn trust despite it is in how to trust an observational study. This paper supplies the numbers that make the concern concrete rather than theoretical.

What it does not establish

• It is association, not mechanism. Whether the gap comes from clinician assumptions, from shorter appointments, from fragmented care between psychiatry and primary care, or from patients declining, is not separated here.
• “Any mental disorder” covers a great deal — from anxiety to schizophrenia to dementia — and the review reports subgroups separately for a reason.
• Prescribing patterns change quickly, and studies spanning years of a shifting market may already understate or overstate current practice.
• Two studies is two studies. The headline figure needs replication before anyone builds policy on its size.

The authors’ own conclusion is that addressing these disparities has the potential to address the excess mortality associated with mental disorders. That is a reasonable inference from an association, stated as a possibility rather than a finding.