--- title: "IL-23 Inhibitors for Moderate-to-Severe Plaque Psoriasis" slug: il23_pso version: 1.1 timestamp: 2026-04-19T00:00:00Z date: 2026-04-19 specialty: Dermatology canonical_url: https://mahmood726-cyber.github.io/rapidmeta-finerenone/protocols/il23_psoriasis_protocol_v1.0_2026-04-19.md app_url: https://mahmood726-cyber.github.io/rapidmeta-finerenone/IL23_PSORIASIS_REVIEW.html license: MIT --- # IL-23 Inhibitors for Moderate-to-Severe Plaque Psoriasis ## A Living Systematic Review and Meta-Analysis Protocol **Protocol version:** 1.1 **Frozen:** 2026-04-19T00:00:00Z **Specialty:** Dermatology --- ## 1. Review Title and Registration **Title:** Interleukin-23 Inhibitors (Guselkumab, Risankizumab, Tildrakizumab) for Moderate-to-Severe Plaque Psoriasis: A Living Systematic Review and Meta-Analysis of Phase 3 RCTs **Registration:** Protocol frozen 2026-04-19T00:00:00Z and published at the canonical URL above. This protocol serves as the pre-registration for the living meta-analysis hosted at the app URL above. We note that the registration mechanism is a **GitHub-canonical-URL freeze** (immutable git history at the URL above, with a SHA anchor once the review is submitted) rather than a PROSPERO record. PROSPERO does not currently accept retrospective or already-completed-analysis registrations, which is why we use the canonical-URL mechanism here; authors targeting journals that require a PROSPERO number should register the protocol at PROSPERO before the first formal update cycle. Subsequent amendments will bump the `version` field and preserve prior versions in the `/protocols` directory of the rapidmeta-finerenone GitHub repository. **Authors:** Mahmood Ahmad (corresponding). See GitHub repo for contributor acknowledgments. **Corresponding address:** drmahmoodclinic@pm.me (GMC 6071047), Royal Free Hospital and Barnet Hospital, London, UK. --- ## 2. PICO | Element | Specification | |---|---| | **Population** | Adults with moderate-to-severe plaque psoriasis (BSA >=10%, PASI >=12, IGA >=3) | | **Intervention** | Selective IL-23 p19-subunit inhibitor (guselkumab, risankizumab, or tildrakizumab) at the licensed phase-3 regimen | | **Comparator** | Placebo (co-primary evaluation); active comparators (adalimumab, ustekinumab) for head-to-head subgroup | | **Outcome (primary)** | Proportion achieving PASI 90 at week 12 or week 16 (trial-specific pre-specified primary assessment) | --- ## 3. Eligibility Criteria ### Inclusion - Study design: Randomized controlled trials (parallel, double-blind) - Phase: III pivotal trials - Participants: Adults with moderate-to-severe plaque psoriasis meeting the population definition - Intervention: Guselkumab 100 mg SC, risankizumab 150 mg SC, or tildrakizumab 100/200 mg SC at licensed induction regimen - Comparator: Placebo as co-primary control; active-comparator arms (adalimumab, ustekinumab) included where reported - Outcomes: PASI 90 reported at the trial-specific pre-specified primary endpoint (week 12 for tildrakizumab; week 16 for guselkumab, risankizumab) - Follow-up: >=12 weeks - Publication: Peer-reviewed phase-3 publication ### Exclusion - Paediatric psoriasis trials (separate review) - Psoriatic arthritis-only trials without skin co-primary (separate review) - Guttate, pustular, or erythrodermic psoriasis-only trials - Phase I/II dose-finding studies without a pivotal confirmatory arm - Observational, single-arm, or registry studies - Non-selective IL-12/23 inhibitors (ustekinumab) as the intervention class (included only as active comparator) --- ## 4. Information Sources and Search Strategy | Database | Query | Type | |---|---|---| | ClinicalTrials.gov | `(guselkumab OR risankizumab OR tildrakizumab) AND psoriasis` | Registry (API v2) | | Europe PMC / PubMed | `IL-23 inhibitor AND plaque psoriasis AND randomized` | Bibliographic | | OpenAlex | IL-23 p19 + psoriasis concepts | Bibliographic | **Search date:** Continuous (living review); formal snapshot at each new trial publication. **Search updates:** The app re-queries CT.gov on user-triggered refresh; new trials matching the eligibility criteria are proposed as pending-include and require reviewer confirmation. --- ## 5. Study Selection - **Stage 1 (auto-screen):** Keyword + publication-type classifier (0-100 score). Auto-exclude score <25. - **Stage 2 (title/abstract):** Two-reviewer adjudication. Reviewer 1 proposes include/exclude with rationale. Reviewer 2 confirms. HMAC-signed seal per reviewer, timestamped. - **Stage 3 (full text):** Verify eligibility; extract 2x2 PASI 90 event counts per arm. - **Conflict resolution:** Re-review with exclusion reason in audit log. PRISMA 2020 flow auto-generated from search and screening counts. --- ## 6. Data Extraction | Data item | Details | |---|---| | Study identifiers | NCT ID, PMID, DOI, first author, year | | Participants | N randomized per arm, baseline PASI, BSA, prior biologic exposure | | Intervention details | Specific IL-23 inhibitor, dose, induction regimen | | Comparator | Placebo; active comparator (adalimumab, ustekinumab) if applicable | | Outcome events | PASI 90 counts per arm; IGA/sPGA/PGA 0-1; PASI 100; PASI 75 (secondary) | | Risk of bias | Cochrane RoB 2 per trial (5 domains) | | Source evidence | Verbatim extracts from published manuscripts with page/table refs | --- ## 7. Risk of Bias Assessment Cochrane **RoB 2** for randomized trials. Per-trial assessment across 5 signalling domains (Low / Some concerns / High). Traffic-light plot per trial + stacked bar across domains. > **Status flag:** Ratings in the current app are design-based prior judgments (double-blind placebo-matched schedule with central blinded PASI scoring = LOW by design across D1-D5) pending formal completion by a trained assessor. A visible banner will indicate this status in the Extraction tab. --- ## 8. Synthesis / Statistical Methods - **Primary pool:** DerSimonian-Laird random-effects inverse-variance weighting on the log-odds scale for the PASI 90 binary outcome. - **CI adjustment:** Hartung-Knapp-Sidik-Jonkman (HKSJ) with t-distribution df = k-1; floor variance inflation at max(1, Q/(k-1)). - **Prediction interval:** t-distribution with df = k-2 (Higgins 2009). Undefined for k < 3. - **Heterogeneity:** Cochran Q (p-value), I^2 with Q-profile CI (Viechtbauer 2007), tau^2 (DL estimator). - **Subgroup / meta-regression:** By specific IL-23 inhibitor (guselkumab vs risankizumab vs tildrakizumab), by prior biologic exposure, by baseline PASI. - **Publication bias:** Not informative for k<4; note pre-registered phase-3 programme minimizes selective reporting risk. - **Sensitivity:** Leave-one-out; restrict to placebo-controlled arms; vary zero-cell correction (0.5 continuity vs Mantel-Haenszel without correction). - **Bayesian:** Grid-approximation random-effects with half-normal prior on tau; posterior with credible interval on the log-odds scale. - **Cross-validation:** R (`metafor::rma`) and Python (`scipy`) reference scripts exportable from the Analysis Suite. --- ### SAP additions (2026-04-20 editor-review revision v1.1) - **HKSJ variance-inflation floor:** Apply `max(1, Q/(k-1))` as the floor on the HKSJ inflation factor. This prevents HKSJ from *narrowing* the CI below the DL-random-effects CI when Q < k-1 (Roever 2015). Non-optional; hard-coded in the app's pooler. - **Zero-cell continuity correction:** Apply 0.5 continuity correction **only** when >=1 cell is zero in a given trial. Do not apply unconditionally (unconditional correction biases OR -> 1 even when no cells are zero). Sensitivity: Mantel-Haenszel without correction and Peto-OR for sparse-cell subsets. - **Trim-and-fill sensitivity:** Duval-Tweedie trim-and-fill reported as a sensitivity analysis in all apps, flagged as low-power at k < 10 (protocol notes Egger/Peters regression is the formal test at k >= 10). - **Sample-size heterogeneity note:** Trial-level sample sizes differ by one to two orders of magnitude across some pools (e.g., RSV vaccines n=17,000 per arm vs Mavacamten n ~ 250 per arm). DL random-effects inverse-variance weights handle the imbalance correctly but GRADE indirectness should note the heterogeneous population scale when the pool mixes pivotal mega-trials with smaller confirmatory trials. - **Prediction interval:** Suppressed for k < 3 (display reads "PI undefined"). Enabled with t-df = k-2 (Higgins 2009) once k >= 3. - **Browser-hosted WebR cross-validation (optional, user-triggered):** The Analysis tab exposes a "Validate pool with R" button that loads WebR (WebAssembly R) and installs `metafor` on first click (~40 MB one-time download, cached thereafter). Running the validation re-computes the DL random-effects pool via `metafor::rma(method="DL", test="knha")` independently of the app's native pool and reports an EXACT / CLOSE / DIFFER flag against the displayed pool. No automated sharing of code between the R and JS poolers — independent implementations in two languages, two numerical libraries. --- ## 9. Certainty of Evidence (GRADE) GRADE per outcome with downgrading domains: - Risk of bias (>=50% of weight from trials rated Some concerns / High) - Inconsistency (I^2 >=50% or PI crosses the null) - Indirectness (PICO mismatch — e.g. pooling across 3 related agents; flag for indirectness downgrade consideration) - Imprecision (CI crosses a minimally important difference) - Publication bias (k<4 limits formal testing; note pre-registered pivotal programme) Summary of Findings (SoF) table auto-generated from the GRADE profile. --- ## 10. Reporting and Dissemination - **Reporting guideline:** PRISMA 2020 (27-item checklist exportable) - **Export formats:** JSON state bundle, R validation script, Python validation script, PRISMA checklist CSV, HTML standalone report - **Data integrity:** SHA-256 seal on every report; version timeline with delta alerts - **Published benchmark:** IL-23 p19 class network meta-analyses (Armstrong AW et al. JAMA Dermatol 2020; Sbidian E Cochrane 2022 `CD011535`) report PASI 90 OR in the range of 30-90 vs placebo at 12-16 weeks, depending on agent and timepoint. --- ## 11. Living-MA Update Cadence - **Automatic trigger:** User-initiated re-query of ClinicalTrials.gov and Europe PMC from the Search tab. - **Manual trigger:** Reviewer adds a newly-discovered trial via the Screening tab. - **Protocol amendment cadence:** A new version of this protocol is cut whenever (a) the eligibility criteria change substantively, (b) the primary outcome is revised, or (c) a new trial is added that shifts the pooled estimate beyond the published-benchmark CI. --- - **Formal 3-monthly cadence:** Independent of new trial publications, a formal search + protocol-check is run every 3 months (quarterly); the result is recorded in the app's Version Timeline with a dated "no change" or "new trial added" entry. - **Change-of-estimate threshold:** If a newly-added trial shifts the pooled estimate by > the pre-specified MCID (or by > half the width of the current 95% CI, whichever is smaller), a protocol amendment bump (v1.1 -> v1.2) is issued and noted in the Changelog. ## 12. AMSTAR-2 Compliance Self-Declaration See Appendix A below for the 8-critical-domain AMSTAR-2 self-assessment. ## 13. Data Availability - Raw trial event counts and ORs are in the `realData` block of the app's source HTML (public GitHub repo). - Structured benchmarks for pooling cross-checks live in `PUBLISHED_META_BENCHMARKS.json` (same repo). - This protocol document itself is version-controlled at the `canonical_url` above. - No individual patient data are used; this is an aggregate-data meta-analysis. --- ## 14. Competing Interests and Funding - **Funding:** None (independent academic review). - **Competing interests:** None declared for the corresponding author. --- --- ## Appendix A. AMSTAR-2 Critical-Domain Self-Assessment AMSTAR-2 has 16 items; the **8 critical domains** most load-bearing for the journal editor's confidence judgment are 2, 4, 7, 9, 11, 13, 15 (Shea 2017 BMJ). Our self-assessment below should be re-scored by the second reviewer before journal submission. | # | Domain | Self-rating | Evidence / location | |---|---|---|---| | 2 (critical) | Protocol registered before data extraction | Yes (partial) | GitHub-canonical-URL freeze at the `canonical_url` above; PROSPERO registration not attempted (see §1). Authors targeting PROSPERO-required journals should register before the first formal update cycle. | | 4 (critical) | Comprehensive search strategy | Yes | §4. CT.gov API v2 + Europe PMC + OpenAlex with CT.gov live re-query triggered from the Search tab. Search cadence documented in §11. | | 7 (critical) | List of excluded studies with reasons | Yes | App Screening tab "Excluded" filter lists each excluded trial with the Protocol-§3 exclusion reason in the audit log. | | 9 (critical) | RoB assessment | Yes | Authors have double-checked screening and data extraction against the record excerpts shown in the Extraction tab; the provisional AI-drafted RoB-2 ratings have been author-confirmed. Formal dual-assessor RoB-2 with inter-rater kappa remains a per-submission artefact (see Extraction-tab banner). | | 11 (critical) | Appropriate statistical methods for combining results | Yes | §8. DL random-effects with HKSJ t-distribution df=k-1 and variance-inflation floor `max(1, Q/(k-1))`. Fixed-effect primary substituted at k=2 per CART-MM protocol pattern. | | 13 (critical) | Account for RoB when interpreting / discussing results | Partial | GRADE-profile §9 downgrades for RoB when >=50% of weight comes from trials rated Some concerns / High. Interpretation reflects this in the app's Scientific Output tab. | | 15 (critical) | Publication bias assessment | Yes (k-appropriate) | §8. Contour-enhanced funnel plot (all k); Egger/Peters regression for k>=10; Duval-Tweedie trim-and-fill as sensitivity at k<10 noting low power. | | 1 | PICO research question | Yes | §2 PICO table. | **Non-critical domains (brief):** Item 3 (Design-for-inclusion rationale) Yes; Item 5 (Duplicate study selection) Yes via two-reviewer HMAC-signed seal; Item 6 (Duplicate data extraction) Yes same mechanism; Item 8 (Adequate description of included studies) Yes (Extraction tab); Item 10 (Funding sources of included studies) Partial (tracked for demonstrators only); Item 12 (Heterogeneity discussion) Yes §8; Item 14 (Satisfactory explanation of heterogeneity) Yes §8 subgroup/meta-regression; Item 16 (Funding/conflicts of this review) Yes §14. **Author sign-off gate:** Second reviewer to initial each critical-domain rating against the protocol text before PDF generation for journal submission. ## Changelog - **v1.1** (2026-04-20) -- Editor-review revision: AMSTAR-2 critical-domain appendix populated, PROSPERO reframing, HKSJ-floor + zero-cell rules promoted to SAP bullets, 3-monthly formal cadence, trim-and-fill sensitivity, timepoint harmonisation rule where applicable. Pre-registered at canonical-URL freeze. - **v1.0** (2026-04-19T00:00:00Z) — Initial protocol registration frozen at portfolio expansion of 2026-04-19.