Learn more
Blog

Obesity Drugs Work. How Do They Compare?

By
Jon Walsh

November 17, 2026

Comparing obesity drugs is increasingly about where each treatment fits across efficacy, tolerability, and longer-term patient benefit. A head-to-head trial can answer a specific comparison, but sponsors have to decide which comparisons are worth running. I think digital twins can play a key role in that process, helping teams identify advantages in their treatments without having to run expensive, risky head-to-head trials.

Suppose you’re developing a dual agonist with monthly dosing.  Your drug produces weight loss similar to other dual agonists, less than some drugs and more than others with different mechanisms of action (MoAs).  There are also established drugs (e.g., another dual agonist like tirzepatide) with strong commercial presence and substantial resources behind them.  Clinically, you may have a useful drug. Commercially, why would someone choose it? That is a much harder question than whether it beats placebo.

The answer depends on factors like safety, tolerability, dosing, and implications for comorbidities, and how these compare to other drugs.  It’s a complex picture; how do we understand it?

Which comparison is worth running?

The direct way to compare drugs is to randomize participants to receive one or the other. Lilly did this with tirzepatide and semaglutide in SURMOUNT-5. There will be good reasons for other sponsors to run head-to-head trials, too. But each trial commits resources to a particular comparator, population, and set of outcomes. These are huge, expensive trials – sponsors want to be sure of the outcome, and practically can’t run many.

During development, in silico methods to simulate head-to-head studies can be used to understand outcomes and contrast treatments.  This can be done across treatments and outcomes, allowing teams to have a deep and broad understanding how their drug fits into the landscape.  When the treatment is approved, real world data may be used to understand efficacy, but it can take time to accumulate data and build a commercial story.

Real world data is our conduit to approved treatments, including ones available within a few months.  Published trial results are the source of data on more recent drugs, and can be used to provide population-level constraints to model efficacy of treatments. Network meta-analysis and population-adjusted indirect comparisons already help teams compare treatments across studies. These methods have assumptions about how the studies and populations relate, and the available data determine which adjustments are possible.  Digital twins can extend these types of population-level approaches to models that can simulate individual patient data, cohorts, and allow for counterfactuals across treatments.

What would make your drug worth choosing?

In my last post, I described a frontier of treatments across efficacy, tolerability, quality of weight loss, and maintenance. Let's take the case of an oral obesity drug. Perhaps its most promising feature is tolerability: patients may be able to stay on a useful dose with fewer adverse effects. That could matter a great deal, even if the average weight loss is similar to an established treatment.

Before investing in that claim, though, I would want to understand what we are seeing. Did our trial enroll patients who were more likely to tolerate treatment? Were dose escalation and the handling of discontinuation comparable? As we discussed in the first post, what patients do after discontinuing can change the apparent treatment effect. Putting the headline results next to each other won't resolve those questions.

Then there is the population in which the advantage matters. A subgroup doing well on your drug is interesting, but differentiation requires understanding how that subgroup would do on the alternatives. For our sponsor, the useful output would be an estimate of how the treatments compare in the intended population, with the uncertainty made explicit. The team could then assess whether a tolerability advantage looks credible enough to follow-up with using more robust evidence after the study ends, and to make program strategy decisions with greater confidence.

The questions get bigger than a single trial

At Unlearn, we are working on the obesity trial problems described in the first post, including differential dropout, and testing approaches against held-out trial data. I see comparative effectiveness as an important direction for that work. It asks more of the model than predicting outcomes under placebo, and starts to move beyond problems where alternative approaches exist or can be defined to ones that lean on models more and more heavily. 

This is the appeal of digital twins – being able to build on model capabilities as the questions expand. The trial team might be concerned with discontinuation. The program lead wants to understand whether the tolerability finding justifies changes to designs to build evidence around. Later, the asset team may want to know what evidence would distinguish the drug after approval. A model supported across the relevant treatments and outcomes could provide forecasts those teams use in their own analyses.  Naturally, each application would still need its own assessment of whether the model is suitable.  

For a sponsor developing an obesity drug now, understanding where the drug might fit should help determine which evidence to build next. These are the questions sponsors are bringing to us, and we would welcome the conversation.

Heading 1

Heading 2

Heading 3

Heading 4

Heading 5
Heading 6

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

Example of a caption
Block quote that is a longer piece of text and wraps lines.

Ordered list

  1. Item 1
  2. Item 2
  3. Item 3

Unordered list

  • Item A
  • Item B
  • Item C

Text link

Bold text

Emphasis

Superscript

Subscript