Evaluation Criteria and How to Score High

Horizon Europe / IHI Scoring Grid

Each criterion is scored 0–5 with half-points; weightings differ per stage.

| Criterion | Stage 1 Weight | Stage 2 Weight | Threshold |

|-----------|---------------|---------------|-----------|

| Excellence | 50 % | 33 % | ≥4/5 |

| Impact | 50 % | 33 % | ≥4/5 |

| Quality & Efficiency of the Implementation | — | 33 % | ≥3/5 |

| Overall threshold | n/a | n/a | ≥10/15 (Stage 2) |

1. Excellence (Scientific & Technical Quality)

Evaluators look for:

- Clear, credible objectives aligned with the *Expected Outcomes* list.

- Progress beyond state-of-the-art in AI algorithms, data integration, interpretability.

- Soundness of methodology, including appropriate performance metrics, validation design, ethics-by-design and risk mitigation.

How to maximise score:

- Provide literature matrix benchmarking existing signal-detection algorithms vs proposed.

- Detail explainability techniques (e.g. SHAP, GNN attribution) to meet transparency requirements.

- Show pilot datasets already secured with data-sharing agreements.

2. Impact

Evaluators will assess:

- Credibility & magnitude of clinical, regulatory, economic and societal impacts; linkage to EU policies (AI Act, EHDS, HERA).

- Quality of KPIs, pathways to market / regulatory uptake, and scalability across therapeutic areas.

- Appropriateness of dissemination, exploitation, communication, including open-science commitments.

How to maximise score:

- Quantify impacts (e.g. cost-savings to regulators, ADRs prevented/year).

- Include a Regulatory Advisory Board with EMA, FDA, national agencies.

- Present a post-grant sustainability plan (membership-based AI-PV hub, integration into DARWIN EU network).

3. Quality and Efficiency of the Implementation

Focus points:

- Work-plan coherence: SMART milestones, critical path, contingency.

- Consortium capacity: complementarity, balanced gender & geographical coverage, access to data and patients.

- Resources allocation: realistic person-months vs deliverables; clear budget split EU vs in-kind.

How to maximise score:

- Use a matrix mapping partners ↔ tasks ↔ data assets.

- Show agile governance: Steering Committee, Ethics & AI Oversight Board, Data Protection Officer.

- Provide Gantt & PERT charts with review gates aligned to EMA qualification milestones.

Transversal Aspects Monitored

- Ethics & security self-assessment – robust, complete, with mitigation actions.

- Gender Equality Plan compliance uploaded for relevant entities.

- Open Science practices integrated (data-management plan M6, preprints, Zenodo).

- Risk management table (technical, legal, IP, reputational) with mitigation & fallback options.

Application Strategy

1. Build a Balanced, Impact-Ready Consortium

- Map competencies to the topic’s five technical work-streams (algorithm review & optimisation, data source evaluation, predictive modelling, recommendations / regulatory pathway, training & dissemination).

- Secure ≥ 1 EFPIA / IHI industry member early; negotiate their in-kind contribution (clinical safety databases, cloud credits, staff time, validation sites).

- Engage regulatory-savvy partners (e.g. national competent authorities, academic centres involved in EMA’s Darwin EU) to co-lead the Qualification procedure.

- Include SMEs / start-ups specialising in AI explainability and patient organisations to strengthen impact and societal relevance.

2. Stage 1 – 10-Page Short Proposal (Excellence & Impact-Heavy)

1. Needs & Gap Analysis (½ page): Quantify current SRS/EHR signal-detection limitations.

2. Overall Concept (2 pages):

• Novel AI approach (e.g. transformer-based temporal phenotyping)

• Data federation architecture

• Outline of 3–4 representative use cases (rare ADR, biologicals, paediatrics, etc.)

3. Ambition & Innovation (1 page): Position versus FDA/EMA guidance, highlight interpretability layer, synthetic data generators, federated learning.

4. Expected Outcomes & Impacts (2 pages): Provide quantitative KPIs (signal time-to-detection ↓50 %, false-positive rate ↓30 %). Map to call’s bullet list verbatim.

5. High-Level Work Plan (2 pages): Work-package table, Gantt, critical path.

6. Consortium Description (1 page): Roles & complementarities; public/private balance.

7. Ethics & Open Science (½ page): Mention AI Act conformity assessment, FAIR data.

8. Budget Overview (½ page): EU vs in-kind, showing >45 % private contribution.

3. Stage 2 – Full Proposal (Make It Implementation-Heavy)

- Detailed methodology for algorithm benchmarking (gold-standard labels, cross-site validation, federated differential privacy).

- Data governance architecture: EHDS-ready, GAIA-X compliant, with data-provenance ledger.

- Regulatory interaction plan: timeline for EMA Scientific Advice + Qualification of Novel Methodologies.

- IPR strategy: open-source core framework (Apache-2.0), layered on proprietary modular plug-ins; Joint Ownership & Access Rights table.

- Exploitation & Sustainability: create a European AI-PV Hub; align with EHDEN/DARWIN EU, WHO-UMC.

4. Writing Tips

- Use the call’s wording – evaluators tick boxes; repeat “transparency, trustworthiness, explainability”.

- Incorporate policy hooks: EHDS, AI Act, HERA, Pharmaceutical Legislation revision.

- Provide risk matrix covering AI bias, data heterogeneity, legal/ethical hurdles.

- Show gender-dimension in research (sex-specific ADR signal detection).

- Plan a fast-start: ready-to-run data sandboxes in month 3.

5. Submission & Consortium Administration

- Register each partner with a validated PIC in the Funding & Tenders Portal.

- Upload a signed Consortium Declaration of Honour and GEPs where needed.

- Check that in-kind contributions are properly encoded as IKOP in the budget table.