Osteoarthritis Endotypes, Phenotypes, and Theratypes

An IAG Curated Review of: Mobasheri A, Bay-Jensen AC, Thudium CS, Karsdal MA. Recent Advances in Understanding the Endophenotypes of Osteoarthritis. EMJ. 2026;11(2):84–93. DOI: 10.33590/emj/R62I0997.

Author: Image Analysis Group (IAG) — Imaging Clinical Research Organization (iCRO)

Topics: osteoarthritis clinical trials · OA endotypes · DMOAD trial design · MRI biomarkers · knee OA imaging · AI imaging analytics

• Osteoarthritis (OA) is a heterogeneous, whole-joint disease defined by distinct clinical phenotypes, molecular endotypes, and emerging theratypes — not a single condition.
• Most disease-modifying OA drug (DMOAD) trials have failed because heterogeneous patient populations diluted treatment effects. Phenotypic enrichment is the fix.
• Molecular endotypes (inflammatory, cartilage-driven, bone-driven, metabolic) become trial-actionable only when paired with imaging biomarkers — MRI, X-ray, DXA, and CT.
• The FDA, FNIH PROGRESS OA, IMI-APPROACH, and IHI-PROBE consortia are converging on MRI-based prognostic enrichment for knee OA trials.
• IAG’s DYNAMIKA® platform delivers centralized, AI-assisted imaging analytics aligned with this next generation of OA drug development.

Mobasheri and colleagues have produced a timely conceptual reset for osteoarthritis. Their thesis is simple and, in our view, correct: OA is not one disease but many. The historical failure of disease-modifying osteoarthritis drug (DMOAD) programs reflects a generation of clinical trials that pooled biologically dissimilar patients into a single “knee OA” cohort and then expected a single mechanism of action to produce a signal.

The authors define a precise vocabulary – clinical phenotypes (observable traits), molecular endotypes (mechanistic drivers), and theratypes (therapy-responsive subtypes) and argue that phenotypic and endotypic enrichment is essential for the next wave of OA trials (Mobasheri et al., EMJ 2026).

At IAG, we read this paper as a direct call to action for the imaging community. The authors acknowledge that their strict definition of a clinical phenotype excludes imaging, but they also concede that “identifying the specific molecular drivers within these phenotypic clusters” requires structural information “visible on imaging (i.e., X-ray radiography and MRI).” They explicitly endorse the recent call for screening MRI as a tool for enhancing OA clinical trials (Guermazi et al., Skeletal Radiol 2023).

The molecular taxonomy they advocate cannot be operationalized in a trial setting without imaging at screening, at follow-up, and increasingly as the primary structural endpoint. This is the position IAG has held for nearly two decades, and it is now converging with regulatory thinking, with public–private consortia like IMI-APPROACH and IHI-PROBE, and with the most recent late-stage DMOAD readouts

How big is the OA burden?

• ~600 million people affected worldwide
• >10% of the global burden of disease
• ~1 billion projected cases by 2050
• No approved disease-modifying therapy to date

Osteoarthritis is the most common form of arthritis and a leading cause of pain, disability, and healthcare cost (Hunter and Bierma-Zeinstra, Lancet 2019; GBD 2021 Osteoarthritis Collaborators, Lancet Rheumatol 2023). Current management is restricted to symptomatic relief with NSAIDs and intra-articular corticosteroids; nothing alters the underlying pathophysiology (Hawker, Clin Exp Rheumatol 2019).

The review distinguishes three concepts the field has used interchangeably for too long:

Clinical phenotype

Externally observable traits – malalignment, joint instability, obesity-driven metabolic OA, pain sensitization (Dell’Isola and Steultjens, PLoS One 2018; Courties et al., Curr Opin Rheumatol 2017).

Molecular endotype

Mechanistic clusters defined by biochemical, cellular, and “omics” signatures – for example, the inflammatory, low-repair, and bone/cartilage-driven endotypes identified in the IMI-APPROACH cohort (Angelini et al., Ann Rheum Dis 2022; Karsdal et al., Osteoarthr Cartil Open 2025).

Theratype

A subtype defined by predictable response to a specific therapeutic mechanism – a concept borrowed from CFTR modulator development in cystic fibrosis (Clancy et al., J Cyst Fibros 2019).

Why this matters in plain language

This taxonomy is not built for the bedside. It is built for trial design – selecting the right patients so that an anti–IL-1β agent is tested in inflammatory progressors rather than mechanically driven progressors, and a senotherapeutic is tested in a senescence-driven cluster rather than a post-traumatic one (Mobasheri and Loeser, Nat Rev Rheumatol 2024). Phenotypic enrichment increases the probability of detecting efficacy, reduces outcome variability, and shrinks the sample size required to demonstrate a treatment effect.

The authors acknowledge a tension in their own framework. They define clinical phenotypes as “purely clinically observable,” excluding imaging and biochemical properties. Yet the OA subsets they tabulate include categories such as “minimal joint disease” that require Kellgren–Lawrence (KL) radiographic confirmation, and they cite the call for screening MRI in DMOAD trials.

From IAG’s vantage point and from every DMOAD program that has progressed to late-stage development in the past five years, this tension resolves in one direction. The molecular endotypes the authors describe map onto, and become measurable through, structural imaging phenotypes. Three lines of evidence make the case.

The Foundation for the NIH (FNIH) Biomarkers Consortium PROGRESS OA has externally validated a panel of MRI biomarkers – quantitative cartilage thickness, semi-quantitative cartilage and meniscal scores, effusion-synovitis, Hoffa-synovitis, and bone marrow lesion (BML) burden as prognostic biomarkers for radiographic and symptomatic progression in knee OA, and is now pursuing formal regulatory qualification (Hunter et al., Skeletal Radiol 2023; ACR Open Rheumatology 2025).

The FDA has explicitly endorsed a “prognostic enrichment imaging biomarker panel” as a context of use for knee OA trials (FDA Biomarker Qualification Letter, 2019).

The three endotype clusters identified by Angelini, Karsdal and colleagues in IMI-APPROACH -inflammatory, low-repair, and subchondral bone/cartilage-driven were derived from biochemical markers, but their validation, prognostic utility, and trial-enrichment value have been demonstrated against MRI- and radiograph-based progression endpoints, including minimum joint space width (minJSW), osteophyte size, BML burden, and bone shape (Angelini et al., Ann Rheum Dis 2022; van Helvoort et al., BMJ Open 2020; Widera et al., Osteoarthr Cartil Open 2023).

The Guermazi et al. Skeletal Radiology 2023 paper that Mobasheri cites argues that conventional radiography alone is too noisy, too slow, and too weakly correlated with pain and function to support modern DMOAD trials. The authors recommend MRI as the modality of choice for both eligibility screening and outcome assessment (Guermazi et al., Ther Adv Musculoskelet Dis 2023).

This is no longer theoretical:

Bottom line: the endotypes the review describes are most actionable when paired with imaging phenotypes derived from MRI, CT, DXA, and X-ray. Biochemical markers tell us what is driving disease; imaging tells us where, how much, and whether the drug is working.

The authors note that current multiplex cytokine measurements may not reflect dynamic disease activity over time, and that no single biomarker or panel has yet been qualified for clinical decision-making in OA. Imaging fills several of these gaps in ways the review under-emphasizes.

Spatial specificity

Biochemical markers integrate signal across the whole body and cannot localize disease activity to a target compartment. MRI resolves:

• Synovitis to the suprapatellar, infrapatellar (Hoffa), and parameniscal recesses
• BMLs to specific subchondral regions
• Cartilage thickness loss compartment-by-compartment

(Hunter et al., Osteoarthr Cartil 2011; Roemer et al., Skeletal Radiol 2025).

Phenotyping at screening

ROAMES (Rapid OsteoArthritis MRI Eligibility Score) and similar MRI-based eligibility tools allow trialists to match the patient’s structural phenotype to the drug’s mechanism of action — synovitis-driven knees for anti-inflammatories, BML-burdened knees for bone-acting agents, cartilage-loss progressors for cartilage-protective agents (Guermazi et al., Ther Adv Musculoskelet Dis 2023).

Dynamic disease activity

Dynamic contrast-enhanced MRI (DCE-MRI), analyzed on platforms such as DYNAMIKA, has been shown to capture perfusion-based signatures of BMLs that correlate with pain in the LOSEIT cohort and to provide a more sensitive measure of dose response to anti-inflammatory therapy than conventional semi-quantitative scoring (Daugaard et al., ACR 2020; Boesen et al., Eur J Radiol 2021).

Predictive validity for hard outcomes

MRI cartilage thickness loss over two years has been independently associated with progression to total knee replacement (TKR) — the clinical outcome that matters most to patients and payers (Eckstein et al., Arthritis Rheum; Hunter and Bierma-Zeinstra, Lancet 2019).

AI-driven scalability

Deep learning applied to baseline radiographs and MRIs now predicts joint space narrowing progression, pain progression, and TKR with AUCs in the 0.80–0.90 range, opening the door to centralized, blinded, AI-assisted reads at the scale modern multi-site DMOAD trials require (Tiulpin et al., Sci Rep 2019; Chaudhari et al., Skeletal Radiol 2023; Pedoia et al., Sci Rep 2021).

The review notes, almost in passing that ongoing trials of GLP-1 and GIP agonists for obesity-related OA “do not include any imaging modalities.” This is a critical observation that deserves more emphasis.

Obesity-driven, metabolic-syndrome–associated OA is one of the fastest-growing OA phenotypes globally, and the GLP-1/GIP class is producing weight-loss effects of a magnitude that will almost certainly translate into mechanical unloading and metabolic-inflammatory shifts in the joint (Binvignat et al., Nat Rev Rheumatol 2024; Courties et al., Joint Bone Spine 2019).

Running these trials without longitudinal MRI or quantitative X-ray endpoints means the field will miss the chance to:

• Characterize how a non–joint-targeted therapy modifies joint structure
• Track BML burden, synovitis, and cartilage thickness over time
• Build the imaging-anchored evidence base needed for a structural-modification claim

From IAG’s experience supporting musculoskeletal and obesity trials, embedding even a lightweight imaging substudy (bilateral knee MRI plus standing X-ray at baseline and at the primary endpoint) is operationally feasible and scientifically transformative.

• Establishes a clean, defensible vocabulary the field has needed for years
• Correctly frames the OA trial-failure problem as patient heterogeneity, not weak biology
• Honest about limitations – no qualified biomarker panel, no validated translation from omics clusters to clinical decisions
• Positions IMI-APPROACH and the new IHI-PROBE consortium as the right vehicles for federated, consortium-led progress

The authors’ strict definition excludes the very tools — MRI, X-ray, DXA, CT, ultrasound — that operationalize endotypes in trials. In practice, the field has already moved to integrated phenotypes that combine clinical traits, biochemical markers, and imaging features. The FNIH PROGRESS OA panel and the IMI-APPROACH endotype clusters both depend on imaging for validation and trial use.

This mapping is what allows phenotypic enrichment to function as a trial-design tool — and it is what regulators increasingly look for in DMOAD submissions.

Federated learning and standardized core outcome sets only deliver their value when paired with regulatory-grade, centrally read, AI-assisted imaging pipelines. This is where the operational reality of running a DMOAD trial meets the conceptual ambition of endophenotyping (Kubassova, LinkedIn 2026).

Recent work in the IMI-APPROACH neuropathic pain subgroup shows that patients with a likely neuropathic component have less structural damage on imaging — a critical insight that only emerges when pain phenotyping and quantitative imaging are run together (van Helvoort et al., RMD Open 2021).

For biotech and pharma teams designing OA programs in the next 24 months, the review’s framework combined with the imaging evidence base above points to a concrete checklist.

• Screen with MRI, not just X-ray. Use MRI-based eligibility tools (ROAMES-style) to enrich for the phenotype your drug is designed to modify. Retain KL grading where regulators still expect it.
• Match endpoint to mechanism. Anti-inflammatory programs should anchor on synovitis and BML readouts; cartilage-protective programs on quantitative cartilage thickness; bone-acting programs on bone shape and BML; weight-loss programs on composite whole-joint scores with DXA body composition.
• Pre-specify imaging-based phenotype subgroups in the SAP. Endotype-stratified analyses, even as secondary endpoints, generate the evidence needed for theratype claims and label-supporting biomarker work.
• Centralize and automate reads. Site-level reads cannot deliver the reproducibility regulators expect for synovitis area, BML volume, cartilage thickness, or B-score. AI-assisted central reading with auditable workflows is now the standard of care.
• Plan for biomarker qualification, not just trial readout. FNIH PROGRESS OA, OARSI Clinical Trials Symposium recommendations, and the FDA’s Biomarker Qualification Program offer real pathways to durable regulatory acceptance (Karsdal et al., Osteoarthr Cartil Open 2022).

Mobasheri and colleagues have written the conceptual case for personalized OA drug development. The next step… the one IAG has been building toward since 2007 – is operationalizing that case in trials, and that requires imaging at every stage: screening, stratification, longitudinal disease activity, and structural endpoint adjudication.

The DMOAD programs most likely to read out positively in the next five years are those that combine the molecular endotyping the authors champion with the imaging phenotyping that regulators, payers, and patients ultimately need to see.

The field has the conceptual framework. It has the imaging modalities. It has the AI and the central-reading infrastructure. It has biomarker consortia pursuing formal qualification. What it needs now is sponsors who design OA trials as if every patient is a different disease because, as this review makes plain, they are.

contact@www.ia-grp.com
www.ia-grp.com

1. Mobasheri A, Bay-Jensen AC, Thudium CS, Karsdal MA. Recent advances in understanding the endophenotypes of osteoarthritis. EMJ. 2026;11(2):84–93. DOI
2. Mobasheri A, Loeser R. Clinical phenotypes, molecular endotypes and theratypes in OA therapeutic development. Nat Rev Rheumatol. 2024;20(9):525–6.
3. Guermazi A, Roemer FW, Crema MD, Jarraya M, Mobasheri A, Hayashi D. Strategic application of imaging in DMOAD clinical trials. Ther Adv Musculoskelet Dis. 2023;15.
4. Guermazi A, et al. A call for screening MRI as a tool for osteoarthritis clinical trials. Skeletal Radiol. 2023;52(11):2011–9.
5. Hunter DJ, Bierma-Zeinstra S. Osteoarthritis. Lancet. 2019;393(10182):1745–59.
6. GBD 2021 Osteoarthritis Collaborators. Global, regional, and national burden of osteoarthritis. Lancet Rheumatol. 2023;5(9):e508–22.
7. Angelini F, et al. Osteoarthritis endotype discovery via clustering of biochemical marker data. Ann Rheum Dis. 2022;81(5):666–75.
8. van Helvoort EM, et al. Cohort profile: IMI-APPROACH. BMJ Open. 2020;10(7):e035101.
9. Widera P, et al. Machine learning-supported patient selection for OA clinical trials: IMI-APPROACH. Osteoarthr Cartil Open. 2023;5(4):100406.
10. Hunter DJ, et al. FNIH Biomarkers Consortium PROGRESS OA study. Skeletal Radiol. 2023.
11. Roemer FW, et al. Multifaceted imaging strategies for clinical trials of knee osteoarthritis. Skeletal Radiol. 2025.
12. Karsdal MA, et al. Reflections from the 2021 OARSI Clinical Trial Symposium. Osteoarthr Cartil Open. 2022;4(3):100262.
13. Karsdal MA, et al. The inflammatory endotype in osteoarthritis: 2024 OARSI CTS reflections. Osteoarthr Cartil Open. 2025;7(2):100572.
14. Sanchez-Lopez E, et al. Synovial inflammation in osteoarthritis progression. Nat Rev Rheumatol. 2022;18(5):258–75.
15. Mobasheri A, et al. The role of metabolism in the pathogenesis of osteoarthritis. Nat Rev Rheumatol. 2017;13:302–11.
16. Binvignat M, et al. Obesity and adipose tissue dysfunction in osteoarthritis pain. Nat Rev Rheumatol. 2024;20(9):565–84.
17. Boesen M, et al. Novel quantitative digital image analysis methodology (DYNAMIKA) for assessment of inflammatory changes on MRI in RA. Eur J Radiol. 2021.
18. Daugaard C, et al. Perfusion in BMLs on DCE-MRI (DYNAMIKA) and pain in knee OA — LOSEIT. ACR Meeting Abstracts, 2020.
19. Tiulpin A, et al. Multimodal machine learning–based knee OA progression prediction. Sci Rep. 2019;9:20038.
20. Pedoia V, et al. Deep learning for large-scale MRI-based phenotyping of OA. Sci Rep. 2021;11:10915.
21. Chaudhari AS, et al. Deep learning applications in osteoarthritis imaging. Skeletal Radiol. 2023.
22. van Helvoort EM, et al. Neuropathic pain in the IMI-APPROACH knee OA cohort. RMD Open. 2021;7(3):e002025.
23. Kubassova O. Imaging as the unlock for osteoarthritis drug approvals. LinkedIn. May 2026.

This curated review was prepared by Image Analysis Group (IAG) for educational and commentary purposes. It is not a substitute for the original article and reflects IAG’s perspective on the role of imaging in osteoarthritis clinical trials. American English used throughout. All third-party sources are cited and linked; no content reproduced verbatim from the source article without attribution.