AI-Powered JCA Readiness: What Three Published Assessments Reveal

AI-Powered JCA Readiness: An Overview

AI-Powered JCA readiness has moved from theoretical discussion to measurable practice. Validation against the first three published EU Joint Clinical Assessment reports shows that structured AI workflows can deliver a complete, source-traceable anticipated scope (comparators, outcomes, and PICO sets) in under one hour when given only population and intervention as input. This article examines the evidence, the technical approach, the practical implications for evidence teams, and the boundaries that still require human expertise.

The EU Joint Clinical Assessment process under Regulation (EU) 2021/2282 creates a hard 100-day window once the official scope is issued. Manufacturers who wait for that scope before starting systematic literature review, indirect comparisons, and dossier assembly face an almost impossible timeline. Those who begin earlier must anticipate the comparators, outcomes, and population structures that assessors across 27 Member States are likely to define. Until recently, this anticipation work required weeks of specialist cross-referencing of national HTA reports, clinical guidelines, drug registries, and published literature. The first three published JCA reports (tovorafenib, lurbinectedin, and tarlatamab) now provide a concrete ground-truth test of whether AI can compress that work while remaining complete and auditable.

Why Anticipated Scoping Matters More Than the 100-Day Clock

The formal clock starts only after the Health Technology Assessment Coordination Group issues the consolidated PICO list. By then, the opportunity to shape evidence generation has largely passed. Teams that treat the anticipation window as optional routinely discover missing comparators or unanticipated sub-populations once the official scope arrives, forcing compressed and higher-risk evidence work. The evidence challenge for JCA is therefore front-loaded: the quality of the anticipated scope determines whether the subsequent 100 days can be spent generating and synthesising evidence rather than scrambling to fill gaps.

Manual anticipation across 27 Member States is slow because the relevant signals are scattered. National HTA decisions, ESMO and national guidelines, reimbursement lists, and trial protocols must be read, compared, and reconciled. The resulting list of comparators and outcomes is rarely identical to the final HTACG scope, yet it is the best available starting point for evidence planning. The validation study examined whether an AI-aided platform could produce a reviewer-ready version of that starting point from population and intervention alone.

How the Validation Works

Three published JCA reports served as ground truth: tovorafenib (Ojemda) in relapsed or refractory paediatric low-grade glioma with BRAF alteration (JCA 202406), lurbinectedin (Zepzelca) in extensive-stage small-cell lung cancer first-line maintenance (JCA 202416), and tarlatamab (Imdylltra) in ES-SCLC second-line and beyond (JCA 202417). These products span different indications, lines of therapy, and degrees of population complexity, providing a useful early test set.

The platform received only the product name, licensed indication from the EMA EPAR, and mechanism of action. No trial data, prior HTA decisions, or information from the published JCA reports themselves were supplied. Outputs were compared on three dimensions: comparator identification, outcome coverage, and PICO-set structure. Performance was judged by completeness for dossier preparation rather than classical precision or recall. False positives (extra comparators or outcomes) carry low cost because a trained reviewer can discard them quickly. False negatives (missing a required comparator or core outcome) carry high cost because they can leave evidence gaps inside the 100-day window.

Technical Workflow: From Limited Inputs to Reviewer-Ready Scope

AI-Powered JCA Scoping Workflow

The validated process follows a clear, repeatable sequence designed for transparency and human oversight:

1. Input restriction – Only the product name (INN), licensed indication as stated in the EMA EPAR, and mechanism of action are supplied. This replicates the information available to a manufacturer before the official scope is issued.
2. Multi-source synthesis – The platform draws from HTA reports, clinical practice guidelines, drug registries, and related literature across all 27 EU Member States.
3. Structured generation – Comparators, outcomes, and candidate PICO sets are produced with supporting rationale for each item.
4. Source linking – Every output element is tied to its exact source document, creating a full audit trail.
5. Human finalisation – An HEOR analyst or subject-matter expert reviews the draft, confirms or adjusts population labels, discards non-relevant items, and owns the final anticipated scope.

This workflow keeps the human expert in control at every decision point while removing the weeks of manual assembly that previously dominated the anticipation phase. The same source-linked evidence base can later feed systematic literature review and data extraction without restarting searches or losing provenance.

Comparator Results Across the Three Assessments

Across the three drugs, the platform captured 10 of 11 JCA-defined comparators (91 percent). No clinically critical comparator was missed.

• Tovorafenib: Seven JCA-defined comparators. Six were identified (vinblastine, carboplatin + vincristine, dabrafenib + trametinib, trametinib monotherapy, bevacizumab + chemotherapy, and a closely related multi-agent regimen mapping to TPCV). Everolimus was not surfaced; it is label-restricted to SEGA/TSC and functions as an individualised option outside the core BRAF-altered indication. Five additional clinically real options were also surfaced.
• Lurbinectedin: The platform correctly identified atezolizumab continuation maintenance as the JCA-defined comparator. It also identified durvalumab as an alternative maintenance checkpoint inhibitor used in several EU Member States.
• Tarlatamab: The platform successfully identified all three core JCA-defined comparators: topotecan, CAV, and platinum rechallenge. Also, it surfaced five additional treatment options, including agents used in relevant clinical trial control arms.

The single miss was an edge case rather than a material gap for evidence planning. Additional comparators provided useful breadth for reviewer consideration rather than noise.

Outcome Coverage and Mechanism Awareness

Core clinical effectiveness outcomes (overall survival, progression-free survival, objective response rate) were identified for every drug. The standard safety bundle (grade ≥3 adverse events, serious adverse events, discontinuations) was captured in full. Health-related quality-of-life instruments appropriate to each population were named (PedsQL for the paediatric glioma setting; EORTC QLQ-C30/LC13 plus EQ-5D for the lung-cancer settings).

The platform also surfaced mechanism-specific signals that generic tools often miss: cytokine release syndrome and ICANS for the BiTE therapy tarlatamab, and immune-mediated adverse events for the checkpoint-inhibitor context of lurbinectedin. These additions improve the completeness of the safety section of an anticipated scope.

Certain JCA-specific composite endpoints (symptomatic disease control, PFS2, complete response rate) were only partially captured. These endpoints are defined by assessors during the scoping process and cannot be derived from published sources alone. Their partial capture correctly reflects a boundary that requires subject-matter expert input regardless of the tool used.

PICO Set Structure

Primary PICO content was correctly captured in all three cases. Lurbinectedin produced an exact structural match to the published single-PICO scope. For complex multi-population cases (tovorafenib stratified by BRAF alteration type; tarlatamab stratified by treatment-free or recurrence-free interval), the platform identified the same clinical distinctions using clinically equivalent but differently labelled categories. Population labelling therefore requires final expert confirmation to align with the precise interval thresholds or subgroup definitions that assessors ultimately adopt. The platform generated more granular splits in the tovorafenib case, a conservative posture that supports broader evidence planning.

Source Traceability and Regulatory Alignment

Every comparator, outcome, and PICO element is linked to its exact source document with rationale. This audit trail supports the transparency and reporting requirements set out in the HTACG general principles on the use of artificial intelligence in JCA dossier preparation (adopted July 2026). Accountability remains with the human expert team: the platform produces a structured starting point; the team reviews, finalises, and owns every decision before it enters the dossier.

The same source-linked architecture also supports later stages of the workflow. Once the anticipated scope is finalised, the identical evidence base can feed systematic literature review, data extraction, and evidence synthesis without restarting searches or losing provenance. Teams already using pharma evidence generation solutions and systematic literature review services can therefore move from anticipated scoping into full dossier preparation inside a single continuous environment. Related discussion of the end-to-end path appears in the companion piece on JCA scoping to evidence synthesis.

Benefits

Accelerating JCA Readiness with AI

The principal operational gain is time. Manual cross-referencing across 27 Member States typically requires three to five weeks of specialist HEOR effort. The validated platform produced equivalent reviewer-ready output in under one hour. That compression frees expert capacity for judgment rather than information gathering: confirming relevance, adding any missing items, and defining assessor-specific parameters such as exact treatment-free-interval thresholds.

A concrete illustration helps show the shift in practice. An evidence team preparing for a new oncology indication can now generate a full draft anticipated scope on a Monday morning, complete a structured expert review the same day, and begin targeted literature searches and protocol development by mid-week. Previously, the same team would still be assembling comparator lists from national HTA sites and guidelines at the end of the first week. The time saved does not remove the need for clinical expertise; it relocates that expertise to higher-value decisions.

Limitations

The approach is not a black-box replacement for expertise. Population labelling differences, JCA-specific composite endpoints, and final comparator selection still require human review. The validation sample remains small (three products). The ground truth is the published JCA scope, which incorporates manufacturer input and expert consultation after the manufacturer’s own evidence package was submitted; it is therefore a higher bar than pure anticipation from public sources alone. These limitations are inherent to the current state of the field and should be stated clearly in any dossier disclosure of AI assistance.

Best Practices for Teams Adopting AI-Powered JCA Readiness

1. Restrict initial inputs to the information a manufacturer actually possesses before the official scope (licensed indication and product details).
2. Treat platform output as a structured first draft, never as a final scope.
3. Require documented human review of every comparator, outcome, and population label.
4. Maintain the full audit trail for dossier disclosure under the HTACG AI principles.
5. Integrate the anticipated scope directly into the subsequent systematic literature review so that evidence generation begins from a common, traceable base.
6. Disclose the tool name, version, purpose, and human-oversight process in the JCA dossier.
7. Use the additional comparators and outcomes surfaced by the platform as a structured checklist during expert review rather than discarding them automatically.

These practices keep the process compliant, auditable, and focused on completeness for dossier preparation.

Looking Ahead

As more JCA reports are published, validation sets will grow, and performance characteristics will become clearer across therapeutic areas. The core architecture (source-linked multi-Member-State synthesis from limited inputs, reviewer-ready framing, and continuous evidence-generation pathway) is already usable. Teams that build disciplined processes around it will enter the 100-day window with a higher proportion of their evidence work already complete and fully attributable.

AI-Powered JCA readiness does not eliminate expert judgment. It relocates that judgment from information assembly to evidence evaluation and final decision-making. The three published assessments demonstrate that this relocation is both feasible and measurable today.

For organisations developing or refining their JCA preparation processes, the natural next step is to examine how an anticipated scope feeds into full evidence synthesis and dossier assembly. Further reading on the operational timeline and workflow is available in the related resources on the MadeAi platform for life sciences and market access solutions.

Author’s Note: This article was supported by AI-based research and writing, with Claude 5.1 assisting in the creation of text and images.