Beyond the Scale: Blood Pressure, Lipids and What Else Changes

The scale is the thing everyone watches, and it is not the only thing that moves. Pooling the individual records of 3,136 people across three semaglutide trials, systolic blood pressure fell 4.95 mmHg further than on placebo.^[1] The surprise is in the subgroups: that reduction was roughly the same whether or not someone had hypertension to begin with, which is not what the researchers expected to find.

Why this analysis is worth more than most

Most meta-analyses pool published summaries — the averages each trial reported. An individual patient data meta-analysis goes back to the per-person records and re-analyses them together, which allows questions the original papers never asked. That is what happened here, across three randomized trials of semaglutide 2.4 mg run over 68 weeks.^[1]

The question they wanted to answer was a sensible one. Blood pressure drugs generally work harder in people whose pressure is higher, so if semaglutide lowers pressure you would expect a bigger effect in people with hypertension. They split the participants four ways to look.

Four groups, one answer: about five points, more or less regardless. A drug that lowers blood pressure by the same amount whether or not yours is high is behaving unlike an antihypertensive and more like something changing the underlying state.

Cholesterol, blood sugar and the rest

The broadest comparison available is a network meta-analysis of 76 trials covering 15 different GLP-1 drugs and 39,246 participants with type 2 diabetes, looking at glycemic control, weight and lipid profile together.^[2] All fifteen lowered HbA1c against placebo. Tirzepatide came out strongest on glycemic control.

A network meta-analysis is how you compare drugs that were never tested against each other, by chaining through their common comparators. It is a legitimate and useful method, and it is weaker than a head-to-head trial, because the chain inherits every difference between the trial populations it links.

The list is longer than it used to be

A 2026 review in the Lancet Diabetes & Endocrinology takes the widest view yet, framing obesity as a gateway condition and gathering the trial evidence across the systems it drives.^[3] The conditions it works through include type 2 diabetes, fatty liver disease, chronic kidney disease, heart failure, cardiovascular disease, sleep apnea, polycystic ovary syndrome, osteoarthritis, muscle mass, depression, quality of life, food cravings, binge eating, substance use and neurodegenerative disease.

That breadth is genuinely striking and it is also where care is required. A review covering fifteen conditions cannot give each the scrutiny a single-condition article can, and the strength of evidence differs enormously between them — a randomized trial in sleep apnea is not the same thing as an association in neurodegeneration.

• Sleep apnea — an FDA-approved indication, with its own randomized trial.
• PCOS — eleven trials, and two meta-analyses that disagree.
• Substance use — an enormous cohort signal, and a randomized trial that missed.
• Mood and suicidality — where the warning came off the label in February 2026.
• Food cravings — measurable now, and contested as a construct.
• Muscle mass — and why the lifestyle comparison reframes it.

What to do with any of this

If you take blood pressure medication and start a GLP-1, the pressure change is a reason for your prescriber to keep watching rather than a reason to adjust anything yourself. What the trials show about people coming off those medicines, and why dizziness is the signal to raise, is in dizziness and blood-pressure pills on a GLP-1. The blood-pressure analysis specifically tracked whether participants’ antihypertensive medications were changed during the trials, which is the kind of thing that has to be accounted for before a number like 4.95 means anything.^[1]

And treat “beyond weight loss” claims with the skepticism the evidence deserves. Some of these effects are established by randomized trials with hard endpoints. Others are associations in observational data, of the kind that have often failed to survive testing. The list being long is not the same as the list being proven.