12 PICOs, five unanswered: what Itvisma teaches about EU JCA

The first national assessment to use an EU Joint Clinical Assessment is complete.

It concerned Itvisma, a one-time gene therapy for people aged 2 years and older with spinal muscular atrophy.

Its final JCA scope contained 12 PICO questions. Five had no submitted comparative analysis. Evidence was submitted for seven, but several questions within those PICOs remained unresolved.[1,2]

This is the first practical national example of how an EU JCA moves into decision-making within a Member State. It also shows what happens when a large comparative scope meets the patients, studies and data that actually exist.[6,7]

A JCA evidence package needs coverage, credibility and restraint. Teams must map the full population and comparator scope, establish which comparisons the available data can support, and keep every claim within the uncertainty of the analysis. More complex modelling cannot create missing patients, missing variables or population overlap.

One medicine created 12 comparative questions

PICO stands for population, intervention, comparator and outcomes. It defines which patients are being assessed, which treatment they receive, what it is compared with and which outcomes matter.

The final Itvisma scope contained 12 PICOs across six patient populations. The comparators included nusinersen, risdiplam, best supportive care, intravenous onasemnogene abeparvovec and individualised treatment.[1-3]

Each PICO also contained further questions about previous treatment, functional status, age, genotype, outcomes and subgroups. Two earlier PICOs were removed after the proposed indication changed during the assessment.[1,3]

We explained in our first EU JCA insight how one indication can multiply into many evidence questions. Itvisma shows what that looks like in a completed assessment.

Five PICOs had no credible comparison

No comparative analysis was submitted for five of the 12 final PICOs. The reasons were not all the same.[1-3]

Two PICOs asked how intrathecal Itvisma compared with intravenous onasemnogene abeparvovec. The developer examined trials, registries and published evidence. The available patient populations did not overlap sufficiently, and data were even more limited for the narrower population requested in one PICO.[1,3]

The other three PICOs focused on patients with at least four copies of the SMN2 gene. Only two patients in one Itvisma trial met this description. Two other trials contained none. There was therefore too little evidence for Itvisma itself to support a comparison.[1,3]

The assessors agreed that omitting all five analyses was justified.[1]

This is important. No credible comparison can be the right conclusion. The problem is discovering that after the evidence programme can no longer change.

Seven submitted PICOs did not mean seven complete answers

Evidence coverage cannot be judged by placing a tick beside each PICO.

Four of the seven PICOs with submitted evidence reused analyses from the first two PICOs. That reuse was reasonable because the contributing patients already met the narrower population definitions. But it did not create four new evidence bases.[1-3]

Gaps also remained within the PICOs marked as having evidence. There was no accepted risdiplam comparison for previously treated patients. There was no best supportive care comparison for previously treated patients. Parts of the claimed population were poorly represented, and several requested outcomes had no usable comparative estimate.[1,2]

Coverage therefore needs to be mapped by population, treatment history, comparator, outcome, subgroup and timepoint. A PICO-level traffic light is too crude.

Coverage starts with the whole evidence landscape

Building that map means searching the full published evidence landscape. Relevant trials, registries, HTA reports and real-world sources must be found. Their populations, outcomes, baseline characteristics and potential treatment effect modifiers must then be extracted in a consistent form.

That is difficult to do quickly by hand across a scope as large as this one. AI-augmented workflows such as Evidax TRACE can map the published evidence in days, with experts making the decisions and a complete audit trail retained.

This does not create evidence that does not exist. It shows where each comparison is supported, where the evidence bridge breaks and which gaps still have time to be addressed.

A placebo is not always the same placebo

When no head-to-head trial exists, an anchored indirect comparison can connect separate trials through a shared control group. That connection is credible only if the control groups are sufficiently comparable.

For the comparison with nusinersen, both trials used sham procedures. But STEER used one needle prick, while CHERISH used needle pricks on days 1, 29, 85 and 274. For the comparison with risdiplam, STEER’s one-time needle-prick sham was compared with an oral placebo given every day for 52 weeks.[1,3]

The developer treated these controls as the shared comparator needed to connect each evidence network. The assessors warned that the differences could introduce bias. For the risdiplam comparison, they concluded that the different modes of administration violated the assumption that the control arms were equivalent.[1]

A placebo is not a universal comparator. How it is delivered, how often it is given and whether it behaves similarly within each trial all matter.

Poor overlap made each adjustment more costly

For treatment-naive patients, the main analyses used matching-adjusted indirect comparisons, or MAICs.

A MAIC takes patient-level data from one trial and gives different weights to its patients so that selected baseline characteristics resemble those reported in another trial. This can reduce some population differences, but only for characteristics measured and reported in both studies.

The weighting also reduces the effective sample size. This is not the number of patients enrolled. It shows how much independent information remains after some patients receive far more weight than others.

In the main nusinersen motor-function analysis, 118 Itvisma patients became an effective sample size of 71. Adding SMN2 copy number reduced it to 21. In the corresponding risdiplam analysis, 123 patients became an effective sample size of 41.[1,2]

This is the trade-off. Adjust for too little and important differences remain. Adjust for more and the answer may rest on a small number of heavily weighted patients.

A more advanced model could not create missing data

The developer also submitted multilevel network meta-regression, or ML-NMR, as supportive evidence. This method can combine patient-level data with published study results and adjust for differences across a connected treatment network.

It could include more variables than the main MAICs. But important variables were still missing, the populations still overlapped poorly and the model had to extrapolate beyond the patients observed in the Itvisma trial.[1,3]

The small network also required an assumption that patient characteristics changed the effects of different treatments in the same way. The assessors questioned whether that was clinically plausible for medicines with different mechanisms and routes of administration.[1]

Their conclusion was blunt: “a mathematically superior framework does not mitigate the underlying high risk of bias and uncertainty.”[1]

The lesson is not that ML-NMR failed. It is that a better model and better source data are different things.

A pooled comparator still had to match the question

For previously treated patients, the developer used a pooled comparator group. Around 80% had received nusinersen and 20% had received risdiplam.[1,2]

The assessors accepted this pragmatically for the question about nusinersen. They rejected the same analysis for the question about risdiplam.

The method had not changed. Its relevance to the comparator had.

Evidence from a mixed treatment group cannot automatically answer a treatment-specific question just because that treatment is represented somewhere within the group.

No significant difference did not prove comparable efficacy

Some indirect estimates did not show a statistically significant difference between treatments. The dossier described some results as showing comparable efficacy or safety.[3]

The developer later clarified that comparable was intended as a descriptive conclusion, not a formal claim that the treatments were equivalent.[4]

The assessors retained an important warning. If a confidence interval is wide and crosses no difference, the result may still be compatible with benefit, harm or little difference. The analysis has not separated those possibilities.[1,2]

Not finding a difference does not prove that two treatments work equally well.

The JCA evidence moved straight into national assessment

Ireland’s National Centre for Pharmacoeconomics completed its Rapid Review of Itvisma on 16 July 2026. It was the first national assessment completed for a medicine with an EU JCA.[6,7]

The NCPE says the JCA report, including relative-effectiveness estimates from indirect comparisons, was used to inform that review. It provided earlier comparative information and a broader evidence base.[6]

The NCPE recommended that Itvisma should not be considered for reimbursement at the submitted price. It did not recommend a full HTA, and the final reimbursement decision remained with the Health Service Executive.[7]

This does not show that the JCA evidence caused the price recommendation. It does show that the JCA influenced the evidence considered nationally. The NCPE used it to inform the Rapid Review, and both the JCA report and the NCPE recommendation will be considered by the HSE.[6,7]

The 100-day dossier window became a 311-day assessment

The assessment scope was shared with the developer on 8 September 2025. The first dossier was submitted exactly 100 days later, on 17 December.[1,3]

The process continued. Revised dossiers followed in February and March 2026. The proposed indication changed, and the scope was revised on 9 March.[1,3]

The report was finalised 297 days after the scope was shared and endorsed after 311 days. It was published on 7 September 2026, 364 days after the scope was first shared.[1,5]

The 100 days were the first delivery window, not the full assessment.

Build coverage, credibility and restraint before submission

For a manual evidence synthesis, the dossier window is a poor time to start discovering the evidence landscape. Searches, extraction and comparison feasibility can take months.

But the dossier window is not a point of no return. Evidax TRACE can establish the initial evidence landscape in days, giving teams time to prioritise analyses and manage gaps while the assessment is live. Earlier is still better because no search or model can add a missing patient population or an unmeasured variable to a completed trial.

The Itvisma case suggests three practical tests.

1. Build coverage

Anticipate the full population and comparator matrix across Member States. Search the complete evidence landscape, then map every PICO by treatment history, outcome, subgroup and timepoint.

2. Test credibility

Before choosing a method, test whether the patients, control groups, outcomes and follow-up are sufficiently comparable. Check which prognostic factors and treatment effect modifiers are available in both datasets. Assess population overlap and the likely effective sample size.

This is the logic behind combining Evidax TRACE with an Evidax Comparison Blueprint. First find the evidence. Then establish which comparisons are feasible, which method is defensible and what each analysis could credibly support.

3. Show restraint

Do not force an analysis where no credible evidence bridge exists. Treat it as an evidence gap to address, or to state openly if it cannot be closed.

Keep every conclusion within the uncertainty of the analysis. A complex method does not justify a stronger claim.

No Joint Scientific Consultation under the HTA Regulation was conducted for Itvisma. That does not show that a consultation would have prevented these gaps. For eligible developers, it is now an opportunity to test planned populations, comparators and outcomes while the evidence programme can still change.[1,3]

Take home

• Itvisma was the first medicine with an EU JCA to complete a national assessment.
• Its final JCA contained 12 PICOs. Five had no comparative analysis, and the assessors agreed that the available evidence could not support them.
• Evidence submitted for seven PICOs did not mean every population, comparator and outcome had been answered.
• A connected evidence network is credible only when the control groups, populations and outcomes are sufficiently comparable.
• MAIC and ML-NMR could not remove uncertainty caused by missing variables, poor overlap and unsuitable comparator evidence.
• Map the full evidence landscape early, test each comparison before choosing the method and keep every claim within what the analysis can support. Coverage. Credibility. Restraint.

References

1. Member State Coordination Group on Health Technology Assessment. Joint Clinical Assessment Report of Onasemnogene abeparvovec. Version 1.0. European Union; 2026. Endorsed 16 July 2026. https://health.ec.europa.eu/publications/joint-clinical-assessment-report-onasemnogene-abeparvovec-itvisma_en

2. Member State Coordination Group on Health Technology Assessment. Joint Clinical Assessment Summary Report of Onasemnogene abeparvovec. Version 1.0. European Union; 2026. Endorsed 16 July 2026. https://health.ec.europa.eu/publications/joint-clinical-assessment-report-onasemnogene-abeparvovec-itvisma_en

3. Novartis Pharma AG. Dossier of the Joint Clinical Assessment of a Medicinal Product: Intrathecal onasemnogene abeparvovec (Itvisma) in the treatment of spinal muscular atrophy. Version 1.0. 3 September 2026. https://health.ec.europa.eu/publications/joint-clinical-assessment-report-onasemnogene-abeparvovec-itvisma_en

4. Member State Coordination Group on Health Technology Assessment. Annex to the Joint Clinical Assessment Report on Onasemnogene abeparvovec: factual accuracy check by the health technology developer and response by the assessors. European Union; 2026. https://health.ec.europa.eu/publications/joint-clinical-assessment-report-onasemnogene-abeparvovec-itvisma_en

5. European Commission, Directorate-General for Health and Food Safety. Commission publishes joint clinical assessment report on an advanced therapy medicine for spinal muscular atrophy. 7 September 2026. https://ec.europa.eu/newsroom/sante/newsletter-archives/78849

6. National Centre for Pharmacoeconomics. First national assessment of a drug with an EU Joint Clinical Assessment. 17 July 2026. https://www.ncpe.ie/first-national-assessment-of-a-drug-with-an-eu-joint-clinical-assessment/

7. National Centre for Pharmacoeconomics. Onasemnogene abeparvovec intrathecal (Itvisma). HTA ID 26025. 16 July 2026. https://www.ncpe.ie/onasemnogene-abeparvovec-intrathecal-itvisma-hta-id-26025/