The Bone and Muscle Safety Gap That GLP-1 Megastudies Weren’t Designed to Fill

The clinical team had built two GLP-1 programs before. Rigorous metabolic endpoints, cardiovascular safety packages, well-powered efficacy readouts. They knew what they were doing.

What they had never done was run a bone safety program inside an obesity trial — because until January 2025, no one had formally required them to. Then the FDA’s Revision 2 guidance for weight-reduction drugs landed. Clear and unambiguous: a representative sample of subjects in every weight-loss trial must now have baseline and follow-up DXA body composition measurement to confirm that weight reduction is driven by fat loss, not lean mass loss. The requirement sits in the General Safety Assessment section. It applies to every GLP-1 and obesity program seeking FDA approval. The regulatory signal is clear.

So they added DXA. Forty-one sites. Twelve countries. Standard protocol. What they didn’t add — because they had never had to think about it before — was centralized imaging quality control.

At the 18-month interim, the bone safety data came back noisy. Not alarming. Just inconclusive. Inter-site variability was wide enough that the statisticians couldn’t distinguish drug effect from measurement artifact. The program wasn’t derailed. It was delayed. And in a field where every week of delay costs between $600,000 and $8 million, that is not an operational footnote. It is a strategic failure. One that was entirely preventable.

But here is the other side of that story — the one most GLP-1 sponsors haven’t seen yet. The same imaging infrastructure that prevents that failure is also the infrastructure that creates a competitive opening no amount of marketing spend can manufacture.

The Scientific Opening That Megastudies Leave Behind

In the first head-to-head trial, semaglutide produced 13.7 percent mean weight reduction at 72 weeks. Tirzepatide achieved 20.2 percent in the same trial — and up to 22.5 percent at the highest dose in SURMOUNT-1. Those numbers are extraordinary — and they are locked in. Novo and Lilly have the efficacy story. Mid-sized biotechs developing next-generation GLP-1s, dual agonists, and muscle-sparing obesity agents cannot win a head-to-head weight loss percentage battle against the incumbents.

But that is not the battle that matters anymore.

The next generation of obesity drug approvals will be decided on quality of weight loss. Which drug loses fat without destroying muscle? Which drug preserves bone density while delivering metabolic benefit? Which sponsor can hand the FDA a clean, centralized, scientifically defensible body composition dataset — not just a scale readout — that demonstrates their drug’s specific safety profile for the skeleton and the musculature?

Lilly and Novo have the efficacy story locked. They also have the resources, the infrastructure, and the regulatory relationships that come from running programs at scale for a decade. That is not a gap to exploit — it is a baseline to match.

The opening for mid-sized biotechs is different. It is not about beating the giants on volume of data. It is about building a more precise scientific argument on a narrower, more defensible endpoint. A focused program — a dual agonist with an explicit lean-mass-sparing mechanism, a next-generation obesity agent with bone preservation data, a muscle-protective formulation targeting sarcopenic obesity — can generate a body composition dataset specifically designed to support a differentiated label claim. That is not something a megastudy optimized for primary weight loss endpoints was designed to produce. And that label distinction — if the data is clean enough to support it — is the competitive moat no marketing budget can manufacture.

“Mid-size biotechs can’t outspend Big Pharma on marketing,” said Dr. Olga Kubassova, President and CEO of Image Analysis Group. “But they can out-science them. If you use standardized, centralized DXA to prove your drug preserves bone and muscle better than the competition — and if that data is clean enough to support a superiority claim — you earn a label that payers will reimburse preferentially. That’s a competitive moat no marketing budget can buy.“

The 2026 Data That Every GLP-1 Sponsor Needs to See

For years, the bone safety question in GLP-1 programs was theoretical. That changed in February 2026.

Liu et al., publishing in the Journal of Clinical Endocrinology & Metabolism, conducted the first matched cohort study directly comparing semaglutide and tirzepatide users to non-users in patients at high fracture risk. 255 GLP-1 users versus 255 matched controls over a median 17 months. The findings carry an important nuance that every sponsor should understand precisely.

Overall, both groups showed significant BMD declines — GLP-1 treatment did not uniformly cause greater bone loss than controls across the whole cohort. But in the subgroup of patients without diabetes, GLP-1 users showed significantly greater total hip BMD loss than matched controls (-1.0% vs. -0.6%; p=0.04). And across the full cohort, the amount of weight lost directly predicted the magnitude of bone loss: r=0.32 at total hip, r=0.17 at femoral neck.

That dose-response relationship is the regulatory concern. The bone loss is not a side effect that can be avoided by patient selection — it scales with the drug’s primary mechanism of action. The more weight the drug removes, the more bone is lost. A sponsor without systematic DXA monitoring is running an uncontrolled safety risk inside their core efficacy data.

Supporting preclinical evidence adds mechanistic context: a 2025 study found that tirzepatide reduces Lachnospiraceae — a pro-osteogenic gut microbiome component — potentially driving bone loss through that pathway. And a secondary analysis of a randomized clinical trial found that while GLP-1 treatment combined with exercise preserved hip and spine BMD, GLP-1 treatment alone reduced BMD at those same clinically relevant sites.

The regulatory distinction between ‘acceptable physiological consequence of weight loss’ and ‘drug safety signal requiring label language’ is made with centralized, calibrated DXA data. Without it, you cannot make that argument to an FDA reviewer.

The Number That Determines Whether Your Data Tells the Truth: 8.18%

Dual-energy X-ray absorptiometry is the gold standard for bone mineral density assessment. It has been for three decades. Regulators accept it. Clinicians trust it. And because of that familiarity, it is one of the most underestimated sources of data risk in multi-center trials.

Within a single imaging center, DXA delivers reliable precision — a coefficient of variation for BMD of approximately 1.40%. That is the number sponsors mentally rely on when they design a study powered to detect a 2 to 3 percent treatment effect. Published research has documented inter-center DXA coefficient of variation exceeding 8.18% for bone mineral density across matched scanner models — versus 1.40% within a single center. For lean body mass, the deterioration is even starker: from 0.76% intra-center to 7.89% across centers.

The arithmetic is unforgiving. A trial powered to detect a 3% treatment effect, running against 8% inter-site noise, cannot produce a clean efficacy or safety signal. The drug may be protecting bone. The data will not be able to say so. No statistical adjustment after database lock can recover what was lost at the measurement level.

The sources of that variability are not mysterious. Different DXA scanner models use different hardware architectures, calibration algorithms, and software stacks — producing systematically different values for the same patient. A site that upgrades its DXA software mid-study without notifying the sponsor changes the measurement parameters for every scan taken afterward. Phantom calibration drift, technician positioning inconsistencies, cross-manufacturer scanner mismatches — each one invisible until it has already contaminated months of data.

“This is where most sponsors get surprised,” said Michael Clark, Chief Operating Officer at Image Analysis Group, who will represent IAG at ESCEO The 26th Edition of the WCO-IOF-ESCEO Congress  in April. “They assume DXA is DXA. It isn’t. The variability problem is real, it’s documented in the peer-reviewed literature, and it’s entirely preventable — but only if the quality control infrastructure is built before the first patient is scanned, not reconstructed after the interim read comes back inconclusive.“

December 2025: The FDA Changed the Rules — and the SABRE Story Behind It

In December 2025, the FDA qualified total hip bone mineral density — specifically percentage change from baseline at 24 months — as a validated surrogate endpoint for postmenopausal osteoporosis drug development. The first surrogate endpoint ever qualified through the FDA Biomarker Qualification Program. For an industry that had spent decades running fracture endpoint trials enrolling thousands of patients over five years, this is transformative.

The scientific foundation behind this qualification is extraordinary. The SABRE project assembled data from over 160,000 participants across 52 randomized controlled fracture trials. The R² for the relationship between total hip BMD change and all clinical fracture reduction was 0.71; for vertebral fractures, R²=0.73. This is one of the strongest biomarker-endpoint relationships in medicine.

Here is the pointed observation for anyone in the GLP-1 space: the companies whose data helped build this qualification include prominent GLP-1 and bone drug manufacturers — among them Amgen, Eli Lilly, and Merck. The same companies whose GLP-1 programs are generating bone loss signals helped create the regulatory standard those programs will now be evaluated against.

The qualification came with a direct technical requirement: evidence of standardized, centralized imaging review and longitudinal calibration of all DXA systems used in the trial. Total hip BMD has moved from a supportive safety measure to a critical efficacy endpoint — which means a single uncalibrated scanner can now disqualify a meaningful portion of a trial’s primary endpoint data.

The precision threshold the FDA expects is a CV below 1.5%. The inter-site CV that most uncentralized global trials produce is above 8%. That gap is not a documentation problem. It is an infrastructure problem. And it exists right now, inside programs designed before this surrogate qualification changed the rules.

The Hidden Signal: Myosteatosis and What Standard DXA Cannot See

The FDA DXA mandate addresses the lean mass question at the whole-body level. But there is a more granular signal that standard DXA alone cannot capture — and it is the one that will define the next generation of obesity drug differentiation.

Myosteatosis — fat infiltration inside the muscle tissue itself — is emerging as the key biomarker separating drugs that preserve functional muscle from drugs that merely preserve mass on a scale. A patient can maintain lean body weight while their muscle quality deteriorates. Standard DXA measures what’s there. It cannot tell you whether the muscle it’s measuring is functional or fat-infiltrated.

A 2025 scoping review co-authored by Dr. Carrino, Prof. Boesen, Dr Kubassova, and researchers from Johns Hopkins and Merck Serono — published in Frontiers in Immunology (PMID 39931069) — demonstrates precisely this: quantitative MRI captures fatty replacement and muscle atrophy that conventional imaging cannot see. The same principle applies directly to sarcopenic obesity in GLP-1 programs. A drug claiming to be ‘lean-mass sparing’ needs more than a scale and a DXA scan to prove it.

A landmark 2026 study published in Communications Medicine puts hard numbers behind this. Basty et al. compared DXA and MRI across 32,961 UK Biobank participants and found that while DXA reliably captures fat mass, it overestimates lean mass in the abdominal region by nearly double compared to MRI — and critically, failed to detect a 4–5% longitudinal muscle loss that MRI clearly identified over just 2.5 years. DXA actually showed lean mass increasing in women during the same period when MRI was recording muscle decline. For a drug developer making a muscle-preservation claim, the measurement tool is not a technical footnote. It is the scientific argument.

This matters particularly for next-generation assets — oral GLP-1s, dual and triple agonists, muscle-sparing formulations — explicitly positioning against the Lilly and Novo first-generation story.

The osteoarthritis (OA) intersection adds a further dimension — and it is directly relevant to any sponsor whose obesity program enrolls patients with comorbid joint disease, an increasingly common profile given the metabolic overlap between obesity and OA. Boesen’s own prospective cohort work demonstrated something counter-intuitive: in 117 obese OA patients, mean weight loss of 12 kg produced a 13-point improvement in knee pain — yet no significant change in imaging markers of synovitis on DCE-MRI or static MRI. Pain relief was driven by mechanical unloading, not suppression of inflammation. It is important to note that this study used first-generation liraglutide over an accelerated 8-week weight-loss protocol — earlier and less potent than current semaglutide or tirzepatide programs at full therapeutic dose. The mechanistic principles hold, but the magnitude of effects may differ with newer, more potent agents. The same research group’s systematic review found that dietary weight loss causes hip DXA bone loss in OA patients — reinforcing that any weight-loss intervention, GLP-1 or otherwise, requires DXA bone monitoring built in.

For sponsors with osteoarthritis pipeline assets or obesity programs enrolling patients with comorbid osteoarthritis: pain is not a reliable endpoint for disease modification, and bone monitoring cannot be optional. GLP-1 drugs reduce both mechanical joint load and systemic metabolic inflammation simultaneously — which means imaging that captures both structural and compositional changes is the only way to understand what is actually happening in the joint.

GLP-1 drugs can change the joint environment through two overlapping mechanisms at once: reducing mechanical load via weight loss and altering systemic metabolic inflammation. That makes imaging especially valuable, because sponsors need to see whether improvements are reflected in cartilage, bone marrow lesions, synovitis, osteophytes, and overall joint structure rather than relying on pain scores alone.

Over 20 years, IAG developed deep expertise in osteoarthritis-focused phenotyping. We offer sponsors longitudinal imaging strategies that identify who has inflammatory-dominant, structurally progressive, or obesity-driven disease’, said Dr Olga Kubassova, President and CEO of Image Analysis Group.

Using X-ray, Ultrasound, and MRI, including quantitative and dynamic contrast-enhanced protocols, IAG can help define baseline disease burden to enrich for the right patients. The next step would be selection of the right imaging strategy to track structural change over time with standardized central reads and using more quantitative novel imaging biomarkers.

What Building a Differentiated Case Actually Requires

The difference between a GLP-1 bone safety program that generates a clean interim read and one that produces an inconclusive dataset is not the drug. It is the operational infrastructure that was — or was not — established before the study opened.

Centralized imaging operations for bone health and body composition trials require four things that most sponsors treat as optional. Site pre-qualification with verified scanner compatibility, documented software versions, and phantom baseline assessments before patient enrollment begins — catching Hologic/GE Lunar mismatches before they contaminate the dataset. Cross-site harmonization: standardized phantom calibration schedules, positioning protocols, and technician training that make longitudinal comparison valid across a global site network. The European Spine Phantom with linear regression cross-calibration achieves prediction accuracies with standard errors below 0.03 g/cm² when implemented correctly. Real-time quality control that flags problems within hours of data upload — not weeks, not at the interim analysis. And central independent read by radiologists and DXA specialists with deep musculoskeletal expertise, eliminating interpretive variability across dozens of sites.

Image Analysis Group builds this infrastructure and brings it into every trial that they help designing and managing. Their DYNAMIKA™ platform provides real-time QC with 12-hour turnaround on uploaded scans, centralized reader workflows, and sponsor-facing dashboards tracking site compliance, imaging queries, and reader status across the full study network.

“I’ve seen programs lose six months of clean data because a site upgraded their scanner software mid-study and no one caught it until the interim,” said Michael Clark, Chief Operating Officer of IAG. “The question isn’t whether the variability problem exists — the peer-reviewed literature documents it definitively. The question is whether your imaging CRO has the infrastructure to catch it before it becomes your problem.”

IAG’s therapeutic area leadership spans Dr. Carrino, who brings FDA Radiology Devices Panel experience and over 300 peer-reviewed publications; Prof. Boesen, whose expertise spans quantification of bone and soft tissue using DXA, MRI, DCE-MRI, CT, and ultrasound — including co-authoring landmark work validating DCE-MRI against synovial biomarkers across RA, PsA, OA, and seronegative arthritis (ACR Open Rheumatology, 2026; PMID 41510581); and Michael Clark, who leads operational delivery across IAG’s global site network of 700+ completed trials.

The imaging infrastructure decisions that determine whether a bone health or body composition program succeeds or stalls at interim analysis are made at protocol design — not at database lock. For mid-sized biotechs building their case against Lilly and Novo, that decision is the most important scientific choice in the program. And it has to be made before the first patient is scanned.

ABOUT IMAGE ANALYSIS GROUP

Founded in 2007, Image Analysis Group (IAG) is a specialist imaging CRO combining medical imaging expertise, a global pre-qualified site network, and the DYNAMIKA cloud platform to power drug development across oncology, musculoskeletal, and metabolic disease programs. 700+ trials delivered. Partnerships include Lilly, Takeda, Ferring, and Amgen. Headquartered in London with operations in the US, EU, India, China, and Australia.

This article is sponsored by IAG.