Written evidence from PCD Research (PMA0105)

About the author, institution and reason for submission

I am a clinician scientist (MBBS, MRCPCH, PhD) with over 20 years’ experience across the NHS, academia, and drug development, with a particular focus on rare disease, genomics, and early-phase clinical development. I have held nationally appointed clinical roles in paediatric oncology and completed a PhD at University College London focused on identifying genetic dependencies and therapeutic targets.

I currently work as a senior drug development clinician, advising biotech and pharmaceutical organisations on clinical strategy, early-phase trials, and translational development, including advanced and genetic therapies.

In parallel, I am the Founder and Executive Chair of PCD Research, a non-profit organisation established to accelerate therapeutic development in primary ciliary dyskinesia, where I have led the creation of international research collaborations, funding initiatives, and engagement with NHS England to address inequities in diagnosis and care.

I am also involved in national policy and innovation initiatives, including contributing to the development of new regulatory approaches for rare disease through engagement with MHRA expert groups, and other cross-sector programmes.

I am submitting evidence to highlight a critical gap between the UK’s scientific capability in personalised medicine and its ability to deliver patient and economic benefit at scale. While genomics, AI, and advanced therapies are sufficiently mature to transform outcomes, progress is constrained by system-level barriers, including inequitable access to genomic testing, fragmented data infrastructure, and misalignment in evaluation and reimbursement frameworks.

This submission focuses on the practical policy actions required to address these barriers, with the aim of improving patient outcomes, enabling data-driven innovation, and ensuring the UK captures the full long-term fiscal and societal value of personalised medicine.

Executive Summary: Key points and policy recommendations: Personalised medicine, AI, and fiscal value

• The UK has strong scientific capability in genomics, AI, and advanced therapies, but delivery is constrained by data infrastructure, access to genomic testing, and system-level barriers.
• This results in avoidable long-term public expenditure and lost productivity.
• Personalised medicine should be treated as a strategic public investment with cross-departmental fiscal returns, consistent with HM Treasury Green Book principles.

1. Earlier genomic diagnosis reduces lifetime cost

• Access to whole genome sequencing (WGS) is inconsistent across clinical pathways: in some conditions it is used early, while in others access is limited to targeted panels following conventional diagnostics.
• The National Genomic Test Directory can act as a constraint where access to WGS requires evidence of superiority over panel-based approaches, creating a structural barrier.

Impact:

• Inefficient diagnostic pathways, delayed intervention, higher NHS utilisation
• Increased downstream costs (education, social care, welfare) and productivity loss

Action:

• Enable earlier, broader access to WGS in cases of clinical uncertainty
• Reform evidence requirements and provide equitable access. 

2. Data infrastructure is the binding constraint

• AI capability exists but is limited by fragmented, siloed genomic and clinical data across NHS, academia, and industry.
• In practice, access to high-quality datasets is often restricted, with data held in institutional silos and governed by incentives that do not prioritise reuse or linkage at scale.

Impact:

• Underutilised data assets, slower diagnosis, inefficient care and trials, reduced life sciences competitiveness.

Action:

• Accelerate HDRS and unified genomic records
• Establish the principle that patient data generated within publicly funded systems (including universities), should be treated as a strategic national asset, with governance models that prioritise responsible access, reuse, and public benefit
• Align incentives across academia and institutions to support data sharing and interoperability, rather than exclusivity.

3. AI can deliver near-term system efficiency

• High-impact applications: variant interpretation (VUS), genomic pipelines, patient identification, digital biomarkers

Impact:

• Reduced diagnostic delay and clinician workload
• Improved diagnostic capability, particularly in heterogeneous and rare conditions
• Creation of a genomically characterised patient population within a single-payer system, enabling:
• More efficient patient identification and stratification
• Faster, lower-cost clinical trials
• Increased attractiveness of the UK for life sciences innovation and investment
• Direct patient benefit through earlier diagnosis, improved access to trials, and accelerated therapeutic development

Action:

• Establish national standards for genomic data, phenotyping, and reporting to enable interoperability and AI integration at scale.
• Commission NHS England and relevant bodies to integrate AI-enabled workflows (e.g. variant interpretation, patient identification) into routine care.
• Ensure MHRA regulatory pathways are proportionate and responsive, enabling rapid evaluation and adoption of AI-driven tools in clinical practice.

4. Data-driven trials improve efficiency and investment

AI and digital biomarkers enable faster, decentralised trials; together with digital identification of patients.

Impact:

• Reduced recruitment timelines and trial costs
• Increased UK attractiveness for life sciences investment

Action:

• Digital identification of patients
• Develop trial-ready, molecularly characterised populations

5. Reimbursement misalignment creates fiscal inefficiency

NICE HTA frameworks (STA, HST) were not originally designed for one-off, high-cost, curative therapies and focus on NHS/PSS costs.

Impact:

• Undervaluation of therapies that reduce lifetime cross-government costs
• Misalignment between upfront NHS spend and long-term fiscal savings

Action:

• Incorporate societal and fiscal perspectives alongside standard HTA
• Expand managed access and outcome-based payment models
• Explore funding approaches aligned to long-term value

Strategic conclusion

• Personalised medicine shifts spending from reactive care to preventative, data-driven investment.
• Targeted action on data, access, and evaluation frameworks will deliver long-term fiscal savings, improved outcomes, and increased economic productivity.
• Without reform, the UK risks system inefficiency and loss of global competitiveness.

Answers to Questions 1 and 2

1        Personalised medicine and AI: the scientific background

Current state of the science and near-term opportunities

The scientific foundations of personalised medicine, particularly genomics, AI-enabled diagnostics, and advanced genomic therapies, are now sufficiently mature to deliver substantial patient benefit. Genomics is already embedded in clinical care in areas such as rare disease and oncology, while AI is increasingly supporting variant interpretation, imaging, and patient stratification. Advanced therapies, including gene editing and cell therapies, have demonstrated proof-of-concept in multiple indications.

However, the principal constraint is no longer scientific feasibility, but equitable access, evidence generation, and capacity for system-level deployment within the NHS.

The most significant near-term opportunity for patients lies in earlier and more equitable access to genomic testing, particularly whole genome sequencing (WGS). Currently, access to WGS is largely governed by the National Genomic Test Directory, which requires evidence that broader genomic approaches outperform targeted panels for specific indications. In practice, generating this evidence often falls to academic groups or patient organisations, where funding is limited. This creates a structural barrier to access and risks entrenching inequities.

Earlier use of genomic testing would deliver multiple benefits:

• Shortening the diagnostic odyssey, reducing repeated investigations and associated costs;
• Improving diagnostic yield, particularly in genetically heterogeneous conditions;
• Supporting pharmacogenomics, enabling safer and more effective prescribing;
• Reducing inequities, as broader genomic approaches are less biased by prior assumptions and panel design;
• Creating a larger pool of molecularly characterised patients, which improves variant interpretation and supports future innovation.

At present, genomic tests are often positioned late in the diagnostic pathway, after conventional tests. This approach is inefficient: many conventional tests have only positive predictive value but are effectively used to exclude patients from genomic testing. Moving genomic diagnosis earlier in the pathway would be both clinically beneficial and cost-effective, reducing downstream healthcare utilisation and societal costs such as missed work or education.

Importantly, expanding genomic characterisation across the population would also strengthen the UK’s position in innovation. Larger datasets of molecularly characterised patients accelerate the resolution of variants of uncertain significance (VUS), improve disease understanding, and create a trial-ready population for emerging therapies.

2        The role of AI in personalised medicine

Artificial intelligence has a significant but often overstated role in personalised medicine. In the near to medium term, its greatest impact will not come from fully autonomous discovery, but from improving efficiency, scale, and accuracy across existing clinical and research workflows.

AI can realistically accelerate personalised medicine in four key areas:

• Patient identification and diagnosis: Machine learning can identify patients earlier and more accurately by analysing clinical records, imaging, and genomic data. This includes identifying undiagnosed patients and shortening the diagnostic odyssey.
• Variant interpretation: AI tools can prioritise and interpret genetic variants, particularly variants of uncertain significance (VUS), which remain a major bottleneck in genomics.
• Therapeutic design and optimisation: AI can support in silico development of therapies, including optimisation of gene-editing approaches such as CRISPR-based interventions.
• Digital biomarkers and disease modelling: Machine learning can identify measurable indicators of disease progression, which are critical for clinical trials and scaling personalised therapies.

The underlying technologies required for these applications largely already exist. The key constraint is not technical capability, but the availability, quality, and accessibility of data, and the ability to integrate AI into clinical systems.

Primary barriers include:

• Fragmented health and genomic data infrastructure;
• Limited access to large, diverse, and well-annotated datasets;
• Siloed data ownership, particularly where academic or institutional incentives restrict sharing;
• Lack of standardisation in genomic testing and reporting;
• Challenges integrating AI tools into NHS workflows and IT systems.

Government initiatives such as the Health Data Research Service (HDRS) and the development of a unified genomic record are important steps, but further work is needed to ensure data is both accessible and usable at scale.

A key structural issue is that genomic data, particularly in rare disease, is often held in fragmented academic cohorts and treated as intellectual property. This significantly limits the ability of AI systems to learn from sufficiently large datasets. Greater open access to anonymised genomic and clinical data, with appropriate safeguards, will be essential to unlock the full potential of AI.

a) Where existing AI tools could be most effective

Existing AI tools could deliver meaningful impact now in several areas:

1. Variant interpretation and VUS resolution
This is one of the most immediate and high-impact applications. AI can prioritise variants, predict functional impact, and integrate genomic with phenotypic data. Accelerating VUS resolution would directly improve diagnostic rates and clinical decision-making.

2. End-to-end genomic analysis pipelines
AI can streamline the pathway from sequencing to clinical report, reducing turnaround times and improving consistency. This could include:

• Automated variant prioritisation;
• Integration of clinical data;
• Digital dashboards for tracking diagnostic progress.

3. Patient identification and trial recruitment
Machine learning could identify eligible patients for clinical trials both pre- and post-diagnosis, improving recruitment efficiency and enabling trials in smaller, more specific populations.

4. Therapeutic design and optimisation
AI can support the accelerated design of gene-editing tools and other targeted therapies, including prediction of off-target effects and optimisation of efficacy.

5. Digital biomarkers and disease progression models
These are essential for clinical trials in personalised medicine, particularly for rare diseases where traditional endpoints may be difficult to measure. Machine learning to develop digital biomarkers also enables decentralised clinical trials, improving access for rural or underserved populations; inclusion of rare disease patients who are geographically dispersed; diversity, which improves the generalisability of results; improved recruitment and retention.

Barriers to adoption

Despite the availability of these tools, adoption is limited by:

• Siloed and inaccessible data;
• Lack of interoperability across NHS systems;
• Regulatory and validation challenges – currently being addressed through MHRA rare disease pathway.
• Limited integration into clinical workflows.
• There are significant structural challenges and residual uncertainties in reimbursement pathways for genetic therapies within the UK.
• Existing health technology assessment (HTA) frameworks, including NICE’s Highly Specialised Technologies (HST) and Single Technology Appraisal (STA) processes, were not originally designed to accommodate the evidentiary and economic characteristics of one-off, potentially curative interventions. These therapies are typically associated with high upfront acquisition costs, limited trial populations, immature long-term outcomes data, and substantial uncertainty in durability of effect, factors which complicate conventional cost-effectiveness modelling.
• Current NICE evaluations are conducted from an NHS and Personal Social Services (PSS) perspective and rely on cost–utility analysis using QALYs within established willingness-to-pay thresholds. While appropriate for many chronic interventions, this framework may not fully capture the value of genetic therapies that deliver functional restoration or disease modification over a lifetime horizon, despite the application of modifiers for severity and rarity. In particular, non-health and cross-sectoral impacts are not routinely incorporated within the reference case.
• A more comprehensive value assessment framework would incorporate a broader societal perspective, accounting for the wider economic burden of inherited disease. This includes downstream costs and benefits related to education, social care utilisation, disability and welfare support, informal caregiving, and productivity effects (including labour market participation of both patients and caregivers). These factors are especially salient in paediatric-onset and rare genetic conditions, where lifetime burden is substantial and multi-sectoral.
• Recent analyses (e.g. ABPI) have highlighted that exclusion of these wider societal effects may lead to systematic undervaluation of transformative therapies in HTA decision-making. Addressing this gap may require further methodological development within NICE and across the UK HTA landscape, including greater use of scenario analyses adopting a societal perspective, consideration of alternative discounting approaches where appropriate, managed access agreements to address uncertainty, and outcome-based payment models to better align cost with realised long-term benefit.
• These effects are particularly pronounced in paediatric-onset and rare genetic conditions, where untreated disease imposes substantial, lifelong fiscal and societal costs across multiple departments. Failure to incorporate these wider costs and benefits risks systematic undervaluation of interventions that generate net public value over the long term, and may lead to suboptimal investment decisions at a whole-of-government level.

The primary issue is therefore not a lack of promising tools, but system-level barriers to deployment.

b) Significance of recent AI advances (including genomics models)

Recent advances in AI, including large-scale models for genomics, represent a significant step forward in what is technically possible, particularly in predicting the functional impact of genetic variation. However, in the context of personalised medicine, especially rare disease, the most immediate challenge remains more fundamental - understanding and resolving variants of uncertain significance at scale. At current rates, resolving all VUS could take decades.

Rare disease is fundamentally a genomic and data problem, but one that is currently constrained less by sequencing capacity and more by our limited ability to interpret variation, particularly at the level of individual, often unique (“private”) variants.

Many rare diseases are genetically heterogeneous, with hundreds of possible causal genes and a long tail of ultra-rare or even single-patient variants. In this context, the central challenge is not simply identifying variants, but determining whether a specific variant, often never previously observed, is pathogenic. This leads to a high burden of variants of uncertain significance (VUS), leaving many patients without a definitive diagnosis.

AI models that predict the functional impact of genetic variation are therefore highly relevant. However, their effectiveness is fundamentally constrained by the lack of sufficiently large, diverse, and integrated datasets. For heterogeneous rare diseases:

• Individual variants may only be observed once or a handful of times globally, making statistical or model-based inference extremely difficult;
• Genotype–phenotype correlations are weak or poorly characterised, particularly for non-coding or structural variants;
• Datasets are fragmented across institutions, with genomic and clinical data held in separate silos across the NHS, academia, and industry;
• Academic cohorts are often treated as intellectual property, limiting access and reuse;
• Population diversity is limited, reducing the generalisability and equity of AI predictions.

As a result, even the most advanced AI models cannot reliably classify many variants because they lack the scale and depth of data needed to learn meaningful biological relationships. Addressing this requires a shift toward large-scale, integrated, and longitudinal datasets that link genomic data with detailed clinical phenotypes and outcomes across the population. For rare disease, this is particularly critical - aggregating data is often the only way to observe enough instances of a given gene or pathway to draw meaningful conclusions.

The importance of this is twofold. First, improving variant interpretation will directly increase diagnostic rates and reduce the diagnostic odyssey for patients. Second, and more strategically, building large, well-characterised cohorts enables patient identification, trial recruitment, and therapeutic development, particularly for targeted and genetic therapies.

Without this data infrastructure, AI will remain limited to incremental improvements. With it, rare disease represents the most immediate and tractable setting in which to realise the full potential of AI-enabled personalised medicine.

c) Realism of Government goals and investment priorities

Government ambitions to use AI to accelerate drug discovery and personalised medicine are realistic in principle, but will not be achieved without addressing foundational infrastructure and data challenges.

The most important priority is to build:

• A large, genomically characterised population, particularly in areas of clinical uncertainty;
• Accessible, linked health and genomic datasets;
• Standardised approaches to genomic testing, including broader use of unbiased WGS.

Rare disease provides a clear proof of concept. In these conditions, there is often a direct link between genetic variation and protein function, making them well-suited to AI-driven approaches. If the UK can build the infrastructure to: identify patients early; generate and link genomic and clinical data; enable secure but open data access;

then this model can be extended to more complex, common diseases where the relationships are less well understood.

Earlier digital identification of patients and matching with inclusion / exclusion criteria would make the UK a more attractive destination for life sciences investment. Crucially, this also creates a molecularly characterised, trial-ready population, which is essential for both rare disease and personalised therapies.

Key investment priorities should therefore include:

• Data infrastructure and interoperability, enabling linkage of genomic and clinical data at scale;
• Standardisation of genomic testing and reporting, including broader and earlier use of WGS in cases of clinical uncertainty;
• Integration of AI into clinical workflows, particularly for diagnosis, variant interpretation, and patient identification;
• Support for translational research linking genomic variation to functional and clinical outcomes;
• HDRS;
• MHRA regulation in rare disease.