In systemic lupus erythematosus (SLE) trials, the control arm is never an untreated arm. Patients randomized to placebo receive an established, potent background regimen: glucocorticoids, hydroxychloroquine, and immunosuppressants such as mycophenolate or azathioprine. These are not inert comparators. They are active medicines that meaningfully reduce disease activity on their own, and those improvements register as responses on composite responder indices such as the SLE Responder Index-4 (SRI-4).
It is worth naming the components precisely because they have different design implications. What we loosely call âhigh placebo responseâ in lupus is really four things stacked together:
1. The pharmacological effect of background standard of care (SOC)
2. Regression to the mean in patients enrolled at a disease-activity peak
3. The natural fluctuating course of lupus
4. The true placebo effect, including the attention and structure of trial participation. Only this is a placebo effect in the strict sense, and it is probably the smallest contributor.
Patients in both arms frequently enter trials with active disease and then receive prompt initiation or intensification of SOC. Even where protocols attempt to standardize background therapy, real-world variability persists across sites, geographies and investigators, adding noise. Polypharmacy inside a lupus trial is unavoidable: effective therapy cannot ethically be withdrawn from patients with an unpredictable, potentially organ-threatening disease.
From EXPLORER to pooled analysis: a persistently high control-arm response
EXPLORER remains the reference case. It randomized 257 patients with moderately-to-severely active extrarenal SLE 2:1 to rituximab 1,000 mg or placebo on days 1, 15, 168 and 182, over 52 weeks. Critically, background azathioprine, mycophenolate mofetil or methotrexate was continued, and prednisone at 0.5â1.0 mg/kg/day was added at baseline in both arms and subsequently tapered. In other words, every participant received a potent new intervention at randomization. No differences emerged between rituximab and placebo on the primary or secondary endpoints, in either area-under-the-curve or landmark analyses.
The natural objection is that the endpoint was at fault. It was not the whole story. A 2019 re-analysis of the raw EXPLORER data re-scored the trial using SRI-4, a modified SRI-4 incorporating glucocorticoid reduction, a modified BICLA and a modified LLDAS. Rituximab failed to separate from placebo on all four (p>0.05 for each). When a drug cannot separate under four different response definitions, the comparator arm itself deserves scrutiny.
EXPLORER was not an isolated case. A systematic review and pooled analysis of 24 phase II/III biologic RCTs in non-renal, non-neuropsychiatric SLE (11,128 patients, of whom 3,899 received placebo plus SOC) quantified the problem. Placebo-treated patients had a mean baseline SLEDAI-2K of 10.4; 85.9% were on glucocorticoids, 72.8% on antimalarials and 48.5% on immunosuppressants. Their response rates were:
- 36.2% met the trial-defined primary endpoint
- 39.8% met SRI-4
- 29.2% met SRI-5 and 28.4% met SRI-6
- 30.9% met BICLA
Two things follow. First, more than a third of patients receiving no investigational agent are scored as responders. That is the mountain every new drug must climb. Second, and more usefully, the response rate falls by roughly ten percentage points as endpoint stringency rises. The endpoint is not a neutral measuring instrument; it is a design lever.
How this derails drug development
The structural consequence is that a biologic with genuine clinical activity may fail to separate from placebo; a trial may be read as negative despite a real therapeutic effect; and investment decisions may be taken on misleading evidence. The investigational product ends up tested not against nothing, but against a multi-drug therapeutic platform that already delivers benefit.
The anifrolumab program is the cleanest illustration available, because it functions as a natural experiment. In TULIP-1, SRI-4 at week 52 was 36% for anifrolumab 300 mg versus 40% for placebo; the primary endpoint was missed. In TULIP-2, the same molecule at the same dose in a comparable population achieved a BICLA response of 47.8% versus 31.5% for placebo (difference 16.3 percentage points; 95% CI 6.3â26.3; p=0.001). Same drug. Opposite verdict. The variable that changed was the primary endpoint.
This does not mean the two indices are interchangeable or that one is simply better. Across the anifrolumab phase 2/3 program, 78â85% of patients had concordant BICLA and SRI-4 outcomes (Cohenâs Îș 0.6â0.7), leaving 15â22% discordant. In the EXPLORER placebo arm, discordance reached approximately 25% (BICLA+/SRI-4â 6.9%; BICLAâ/SRI-4+ 18.4%). Notably, in TULIP-1 the discordant BICLA-non-responder/SRI-4-responder subgroup contained more than twice as many placebo patients as anifrolumab patients, driven largely by resolution of arthritis. Discordance of that magnitude, asymmetrically distributed, is more than enough to flip a read-out.
A contemporary data point sharpens the picture further. In the phase 3 obinutuzumab trial published in March 2026, the placebo-plus-standard-therapy arm achieved an SRI-4 response of 53.5% at week 52 (versus 76.7% for obinutuzumab). In a prespecified additional analysis in which response status was unaffected by non-fatal intercurrent events, the placebo figure rose to 68.5%. A control arm approaching 70% is the most vivid available evidence that the comparator in a lupus trial is a therapeutic platform, not a control condition, and the 15-point gap between the two analyses is produced entirely by how intercurrent events are handled.
The bar keeps rising: SOC is a moving target
Standard of care is not static, and each improvement raises the control-arm ceiling for the next entrant. In lupus nephritis, AURORA 1 randomized 357 patients to voclosporin or placebo on a background of mycophenolate mofetil 1 g twice daily plus rapidly tapered low-dose oral steroids. Complete renal response at week 52 was 41% versus 23% (OR 2.65; 95% CI 1.64â4.27; p<0.0001). This is excellent news for patients and a genuine problem for trial designers: the 23% control response of 2017â2019 becomes the new floor, and the next molecule must beat a regimen that now includes a calcineurin inhibitor. Sponsors modelling control-arm performance on trials that predate current practice will systematically under-power their studies.
Nine strategies to answer the real question
The question a lupus trial actually asks is not âdoes this drug work?â It is: âon top of an already effective regimen, does this drug add enough to produce a detectable incremental signal?â Nine design and trial conduct strategies help answer it.
1. Stabilize the background. Standardize and fix concomitant medication before and during the treatment period. Prespecify permitted doses, require a stabilization window before randomization, and prohibit elective changes. Once steroid and immunosuppressant doses are allowed to drift, the control curve drifts upward with them.
2. Prespecify the estimand, not just the endpoint. Under ICH E9(R1), define in advance how rescue therapy, treatment discontinuation, protocol-restricted medication and loss to follow-up are handled. As the obinutuzumab data show, the same trial can yield a 53.5% or a 68.5% control response depending on this choice alone. Sponsors who leave it to the statistical analysis plan are ceding control of their headline number.
3. Control rescue and lock the steroid policy. Structured, protocol-mandated tapering schedules with rescue permitted only at protocol-defined events. A high-dose steroid pulse given to a flaring control patient can erase a drugâplacebo gap overnight. Where steroid-sparing is itself the claim, the taper schedule must be built into the primary or key secondary endpoint.
4. Choose endpoints for stringency and sustainability. The pooled data show SRI-5, SRI-6 and BICLA all run roughly ten points below SRI-4 in control arms. Sustained response, time-to-flare, steroid-free survival, LLDAS and DORIS remission are more placebo-resistant than a single week-52 binary. Because organ systems respond differently, pair a composite primary with prespecified domain-specific secondaries (CLASI for skin, joint counts for arthritis, proteinuria and eGFR for nephritis) so that a real effect in one domain is not averaged away by stability in another. Failure to meet a primary endpoint may reflect the designâs failure to isolate the drugâs contribution rather than the moleculeâs failure.
5. Enrich the population, but do not over-rely on it. Requiring higher disease activity (e.g. SLEDAI-2K â„10), serological activity (anti-dsDNA positivity, low complement) or a defined biomarker phenotype reduces spontaneous improvement and concentrates treatable disease. Enrichment is necessary but not sufficient: in TULIP-1, patients with a high type I interferon signature showed SRI-4 of 35.9% on anifrolumab versus 39.3% on placebo. Enrichment sharpens a signal that exists; it does not create one.
6. Maintain expert central review at entry and throughout. At the entry gate, confirm each participant genuinely crosses the activity threshold. Thereafter, centrally review every BILAG domain, SLEDAI item, CLASI score and joint count; cross-check EDC entries against laboratory and eCOA data; and confirm that lesions recorded as active are genuinely reversible. Irreversible damage â old peripheral nerve injury, scarring alopecia, residual proteinuria â must be coded as damage, not activity. Continuous expert monitoring also catches protocol deviations such as missed tapers or unreported rescue medication (e.g. unrecorded steroid bursts) before they inflate the control arm.
7. Power realistically against a contemporary control assumption. If the expected absolute treatment effect is around 10â15 percentage points, large samples are unavoidable; recent pivotal programmes have ranged from several hundred to roughly 800 patients. Base the control-arm assumption on trials reflecting current SOC, not on historical rates, and stress-test the sample size against a control response 5â10 points above the point estimate. Robust site backup plans protect the enrolment assumption on which the power calculation rests.
8. Evaluate adaptive features deliberately. Bayesian adaptive designs, dose-finding with response-adaptive randomization and prespecified futility rules have moved from theory into lupus practice; rozibafusp alfa and efavaleukin alfa phase 2b studies both used this framework, and the FDA accepted efavaleukin alfaâs protocol into its Complex Innovative Trial Design Meeting Program. Adaptive features are a tool for allocating patients efficiently under uncertainty. They require extensive simulation and strict type I error control, and they are not a remedy for an inflated control arm on their own.
9. Include PROs and refine outcome instruments. LupusPRO, treatment response measures for SLE and fatigue and quality-of-life instruments capture clinically meaningful change that composite indices miss. Time-to-flare curves and sustained steroid-free survival frequently reveal benefit invisible to a binary week-52 responder status, and they speak directly to the outcomes payers and patients weigh.
A tenth measure is worth raising explicitly as a hypothesis rather than a recommendation: aligning visit intensity with everyday practice. Monthly study visits deliver assessment, education and contact that plausibly contribute to the expectation component of control-arm response. Some recent protocols retain essential laboratory monitoring while extending routine face-to-face visits to every two to three months, with interim symptom checks by tele-visit or secure application. The rationale is coherent, but I am not aware of randomized evidence in SLE that this reduces control-arm response, and any such design must be reconciled with safety-monitoring obligations and timely flare detection in a potentially organ-threatening disease. It should be prespecified, justified to the ethics committee and monitored closely, not adopted casually.
Modest, consistent, additive…that is the realistic target
Even successful lupus therapies do not produce dramatic transformations. Belimumab improved SRI-4 by approximately 14 percentage points in BLISS-52 and approximately 10 points in BLISS-76 over placebo, with benefit consistent across high-disease-activity subgroups and a reduced risk of severe flare; anifrolumab raised BICLA response by 16.3 points in TULIP-2. In clinical terms that is roughly one additional responder for every six to ten patients treated. These are meaningful, reproducible, additive gains layered onto standard care, and they should not be presented as ‘miracle cures’ to patients, payers or investors.
Practical advice to sponsors is to lock the steroid policy in the protocol, link mechanism to phenotype, define the estimand before the endpoint, and prespecify an outcome that is both regulator-acceptable and resistant to control-arm inflation.
Rather than trying to outperform placebo in a vacuum, the goal should be to demonstrate reliable, additive benefit in a heavily treated population. Every lupus investigational product deserves a fair chance to show its true added value, without SOC noise hiding the signal.
Continue the placebo response in rheumatology trials conversation at Indero’s breakfast event at the ACR Convergence 2026.
About the author
Dr. Juan Ovalles, MD, PhD, is Senior Director of Rheumatology at Indero, providing medical leadership and strategic guidance in rheumatology research. He collaborates with business development, supports operational and regulatory teams, and builds relationships with key stakeholders. With more than 14 years of clinical development experience, Dr. Ovalles began his career as an investigator in 2012 and transitioned to the CRO industry in 2021. A board-certified rheumatologist with a PhD in biomedical sciences, he specializes in innovative therapies for immune-mediated, inflammatory diseases.
References
Merrill JT, et al. Efficacy and safety of rituximab in moderately-to-severely active SLE: the EXPLORER trial. Arthritis Rheum. 2010;62(1):222â233. (PMID 20039413)
Scherlinger M, et al. Rituximab in moderate to severe non-renal SLE: a reanalysis of the EXPLORER study. Ann Rheum Dis. 2019;78(7):1007.
Tselios K, Wakani L, Gladman DD, Su J, Urowitz MB. Response to placebo in non-renal, non-neuropsychiatric SLE: a systematic review and pooled analysis. Rheumatology. 2021;60(1):73â80.
Morand EF, et al. Trial of anifrolumab in active SLE (TULIP-2). N Engl J Med. 2020;382(3):211â221.
Furie RA, et al. TULIP-1: anifrolumab in active SLE (SRI-4 primary endpoint not met; 36% vs 40% placebo).
Bruce IN, et al. Concordance and discordance in SLE clinical trial outcome measures: analysis of three anifrolumab phase 2/3 trials. Ann Rheum Dis. 2022;81(7):962â969.
Askanase AD, et al. Evaluating the concordance between BICLA and SRI4 in the placebo arms of the EXPLORER and ATHOS trials. Lupus Sci Med. 2025;12(1):e001483.
Furie RA, et al. Phase 3 trial of obinutuzumab in active SLE. N Engl J Med. March 2026. (SRI-4 76.7% vs 53.5% placebo; 85.4% vs 68.5% in the intercurrent-event-unaffected analysis)
Rovin BH, et al. Efficacy and safety of voclosporin versus placebo for lupus nephritis (AURORA 1). Lancet. 2021;397(10289):2070â2080.
Parodis I, et al. Outcomes of patients with SLE treated with belimumab: post hoc BICLA analysis of five phase III trials. RMD Open. 2025;11(2):e005444.
ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials.
FDA Complex Innovative Trial Design Meeting Program.
