Dashboard

Disease indication

Rheumatoid Arthritis

The models were never very good. It worked anyway — and the reasons why are the useful part

Therapeutic goal

Disease-modifying: Stop the immune process eroding cartilage and bone, judged on radiographic progression rather than on how the patient feels. Rheumatology coined the term DMARD, and it remains the only field on this site with a structural endpoint everyone accepts — which is a large part of why it succeeded.

Approved disease-modifying drugs
20+

Conventional synthetic DMARDs, five TNF inhibitors, IL-6 receptor blockers, T-cell costimulation blockade, B-cell depletion and several JAK inhibitors.

high confidence
Distinct mechanisms that worked
7

Antimetabolite, TNF, IL-6, IL-1, T-cell costimulation, CD20 and JAK. Not one bet that paid off — seven independent ones, which is what a tractable disease looks like.

moderate confidence
Radiographic progression
Arrestable

Modern treat-to-target regimens can halt structural joint damage outright — the endpoint Parkinson's has never had and DM1 is still trying to establish.

high confidence
Typical pivotal design
~800 · 52 wk × 2

Two pivotals, a year long, with radiographic follow-up. The structural endpoint is what makes it expensive and what makes the claim believable.

moderate confidence

Preclinical model validity

Read the construct and predictive columns together, then compare them with Parkinson's. Here the models are largely wrong about how the disease starts — and predicted the clinic anyway. Construct validity is not what decided the outcome.

Validity of Rheumatoid Arthritis preclinical models across face, construct, and predictive validity. Predictive validity is shown for the Prevent joint damage goal.
ModelFaceDoes it look like the disease?ConstructDoes it arise the same way?PredictivePrevent joint damage
ValidityNoneLowModerateHighNo evidencenever tested — not a low score

Translation ledger

More approved mechanisms than the rest of this site combined, out of models with poor construct validity. The failures in this list are as instructive as the successes.

Interventions that showed benefit in a Rheumatoid Arthritis preclinical model, and what happened when they reached the clinic.
InterventionRested onReachedOutcome
IL-17 blockade in rheumatoid arthritis
Novartis and others · Neutralisation of interleukin-17A
CIA
Phase 2/3
2017
Failed
Fostamatinib
Rigel / AstraZeneca · Spleen tyrosine kinase (SYK) inhibition
CIACAIA
Phase 3
OSKIRA · 2013
Failed
Tofacitinib
Pfizer · Janus kinase inhibition
CIA
Approved
2012
Approved
Tocilizumab
Chugai / Roche · Interleukin-6 receptor blockade
CIA
Approved
2010
Approved
Rituximab
Genentech / Roche · Anti-CD20 B-cell depletion
Human synovium
Approved
2006
Approved
Abatacept
Bristol-Myers Squibb · CTLA4-Ig blockade of T-cell costimulation
CIA
Approved
2005
Approved
Anakinra
Amgen · Interleukin-1 receptor antagonist
CIAK/BxN
Approved
2001
Approved
TNF inhibitors (infliximab, etanercept, adalimumab)
Centocor, Immunex, Abbott · Neutralisation of tumour necrosis factor
Human synoviumCIAhTNF-Tg
Approved
1998
Approved
Methotrexate
Multiple · Folate antimetabolite with adenosine-mediated anti-inflammatory effects
Approved
1988
Approved
induced autoimmune model · Mouse / rat

Collagen-induced arthritis (CIA)

Susceptible strains immunised with type II collagen in adjuvant develop an erosive polyarthritis. The workhorse of rheumatoid arthritis drug discovery, and the model behind most of the mechanisms now approved.

Validity profile
HighFace
LowConstruct
HighPredictive · symptoms
HighPredictive · progression
Strengths
  • Genuine synovitis, pannus formation and erosion of cartilage and bone
  • Predicted TNF blockade, IL-6 blockade, costimulation blockade and JAK inhibition
  • Well characterised, reproducible, and cheap enough to screen against
Limitations
  • Driven by deliberate immunisation against a cartilage antigen; human disease is not anti-collagen driven
  • Acute and largely monophasic, where human disease is chronic and relapsing
  • No citrullinated-peptide autoimmunity, the hallmark of the human disease
  • Restricted to particular MHC haplotypes and inbred strains
In plain language

Mice are deliberately immunised against a protein found in their own cartilage, and their joints become swollen and eroded. It is not how people develop rheumatoid arthritis at all — and yet nearly every drug that now works was tested here first.

What it means commercially

The single most commercially productive model on this site. Note carefully what it did and did not do: it was excellent at confirming that blocking an inflammatory mechanism protects joints, and useless at telling anyone which mechanism to pick. The picking came from human tissue.

Technical detail

Type II collagen in complete Freund's adjuvant in DBA/1 or susceptible H-2q/H-2r strains produces anti-CII antibody, complement activation and erosive synovitis. Low construct validity — the initiating autoantigen and the ACPA biology of human RA are absent — with high predictive validity for anti-inflammatory disease modification.

Trentham et al., J Exp Med 1977; Brand et al., Nat Protoc 2007high confidence

What the clinic costs

A common disease with a hard structural endpoint: large trials, two pivotals and radiographic follow-up. Expensive per programme — against a likelihood of approval and a market that both dwarf the other indications here.

The structural endpoint is what costs the money. Proving that joints stop eroding means a year on drug with radiographic follow-up in both pivotals — and against seven approved mechanisms, an active comparator rather than placebo alone.

3.7×
the exposure
2.7×
the cost

1,795 patient-years against 480 for a symptomatic programme. Cost grows more slowly than exposure because the fixed enrolment cost of each participant — screening, baseline imaging — is paid once however long they are then followed. The goal, not the molecule, still sets the budget.

Citedenrolment, duration and study counts, from named trials.Assumptionthe two cost parameters below.Computedeverything else.

Clinical programme by phase: enrolment, duration, exposure in patient-years, and modelled cost.
PhaseDesignN × studiesDurationPatient-yearsModelled cost
Phase 1
SAD / MAD, healthy volunteers
Immunosuppressive mechanisms carry infection risk, so first-in-human work is not abbreviated.
Conventional first-in-human designmoderate confidence
60
9 mo
45
$1.6M–$2.7M
Phase 2
Dose-ranging on ACR response
Validated composite response criteria give a clean, fast efficacy signal — the equivalent of which Parkinson's has never had for disease modification.
Biologic and JAK inhibitor Phase 2 programmesmoderate confidence
300
6 mo
150
$6.6M–$11M
Phase 3
Two pivotals, 52 weeks, with radiographic progression
Large and long because the claim is structural: joint damage scored on serial radiographs, usually against methotrexate or an approved biologic rather than placebo.
Registrational biologic and JAK inhibitor programmesmoderate confidence
800 × 2
1 yr
1,600
$51M–$85M
Total2.3 yr of clinical development, treating phases as sequential1,795$59M–$98M
AssumptionThe two cost parameters
Fixed, per participant
$12,000–$18,000
Running, per year on study
$20,000–$35,000

Model assumptions, not citations — the only free parameters here. The fixed component covers screening, baseline imaging and randomisation, incurred once per participant; the running component covers visits, monitoring and site fees for each year on study. Splitting them matters: a single rate per patient-year would price short symptomatic trials far below what pivotal trials actually cost.

Calibration: priced this way a single pivotal trial in this programme comes to $8.8M–$14M, against published pivotal trial costs of $1233M (median $19M) across all indications. A mid-sized 24-week trial in a common disease belongs near the lower quartile of that spread, and lands there. Moore et al., JAMA Internal Medicine 2018.

ComputedRisk-adjusted cost
$396M$654M

Programme cost divided by a 15% likelihood of approval from Phase 1 in autoimmune disease — what one success costs once the failures are paid for. BIO / Informa / QLS, Clinical Development Success Rates — autoimmune programmes sit well above the all-indication average and roughly two and a half times neurology.

Read this one carefully: This is the only indication here where the area-level rate is arguably too pessimistic rather than too optimistic. It pools first-in-class attempts with the many follow-on entrants into mechanisms already proven in rheumatoid arthritis, and a seventh TNF inhibitor is not a coin flip. Read it as a portfolio average across very unequal bets.

moderate confidence
For scale — industry-wide, per approved drug
$985M
Median R&D cost per approval
Wouters et al., JAMA 2020 (capitalised; estimates range $314M–$2.8B)
$2.6B
Capitalised cost per approval
DiMasi et al., J Health Econ 2016

These are portfolio figures spanning all of R&D and already carry the cost of failure, so they are not comparable with the per-programme clinical costs above — they are the order of magnitude those costs roll up into.

What is changing

The remaining problems are no longer about finding a mechanism that works. They are about choosing among the ones that do, and about the patients none of them help.

Choosing between mechanisms, not finding them

With seven working mechanisms, the question has become which one suits which patient. Synovial biopsy can now classify tissue by its dominant cell type, and trials have shown that this classification predicts whether B-cell depletion or IL-6 blockade will work better — the accessible tissue paying off a second time.

Synovial pathotype stratification trials

Difficult-to-treat disease is the remaining need

Roughly one patient in ten cycles through multiple mechanisms without adequate response, a group now formally defined as difficult-to-treat rheumatoid arthritis. The unmet need has moved from the disease as a whole to a refractory minority that none of the seven mechanisms reaches.

EULAR definition of difficult-to-treat rheumatoid arthritis

Intercepting the disease before it starts

Autoantibodies against citrullinated proteins appear years before the first swollen joint, making it possible to identify people at high risk and to treat pre-emptively. Trials in at-risk individuals have shown that progression to clinical arthritis can be delayed — moving the field from treatment toward prevention.

Trials of intervention in ACPA-positive at-risk individuals

Educational. Validity ratings are a coded reading of the published literature, not a consensus standard. Approval years refer to the first rheumatoid arthritis indication in the US and may differ elsewhere or for other diseases. Every figure carries an explicit confidence flag. Not medical, regulatory, or investment advice.