Summary

• AstraZeneca and Oracle are exploring the use of AI, clinical information and real-world evidence to improve trial recruitment and healthcare workflows.
• Their first initiative, with M42 in Abu Dhabi, will focus on identifying uncontrolled hypertension and supporting clinical research.
• The non-binding collaboration includes plans for shared governance, technical components and outcome measures; no completed deployment or clinical benefit has been established.

AstraZeneca and Oracle have entered a non-binding collaboration to investigate how artificial intelligence and healthcare data could improve clinical trial recruitment, safety monitoring and the movement of research evidence into patient care.

The agreement, announced on 20 October, will combine AstraZeneca’s scientific, clinical development and regulatory expertise with Oracle’s cloud infrastructure, health data technology and AI-enabled clinical applications.

Its first planned programme will focus on hypertension care and clinical research in Abu Dhabi, working with healthcare company M42. The partners intend to explore whether digital workflows can help clinicians identify people with uncontrolled hypertension, apply relevant clinical guidance and find potentially suitable trial participants.

The initiative remains exploratory. The companies have described areas for collaboration and methods for assessing them, rather than announcing a completed clinical system or independently demonstrated improvements in patient outcomes.

Trial recruitment depends on clinical information

Clinical studies require participants who meet defined eligibility criteria covering diagnosis, previous treatment, laboratory findings and other medical characteristics.

Identifying those patients can involve examining several sources of information. Relevant details may appear in electronic health records, clinical notes, test results or documentation held by different healthcare organisations.

The information is not necessarily arranged in the same way as the eligibility requirements contained in a trial protocol. Clinicians may need to interpret records manually before deciding whether a patient should be considered for further screening.

AstraZeneca and Oracle want to examine whether AI can help with that process by identifying potentially eligible patients earlier and bringing the relevant information into clinical workflows.

Such systems could assist with locating records and comparing patient characteristics against research criteria. However, eligibility remains a clinical and regulatory determination, and participation requires appropriate consent and assessment.

The distinction is important because an automated system may identify similarities between a patient’s record and a trial’s requirements without having all the information needed for a final decision.

Incomplete records, outdated diagnoses and differences in how healthcare providers record treatment could produce incorrect matches or overlook suitable candidates.

Any useful deployment will therefore need evaluation against established clinical procedures, including the accuracy of candidate identification and the time required for staff to review findings.

Abu Dhabi provides the first proposed application

The collaboration with M42 establishes a defined starting point in hypertension, a condition that can require continuing monitoring and decisions about treatment over time.

According to the partners, the programme will explore AI-enabled workflows to identify patients whose hypertension is not adequately controlled and support the application of guideline-directed care.

It also aims to improve the identification of potential research participants, connecting information collected during ordinary healthcare delivery with opportunities to participate in clinical studies.

This is a different proposition from developing an AI system that independently diagnoses hypertension or prescribes treatment. Clinical professionals would remain responsible for interpreting information and deciding which interventions are appropriate.

Successful identification depends on the reliability of blood pressure readings, the completeness of treatment records and the ability to account for other conditions affecting a patient’s care.

Healthcare organisations must also determine how any suggested actions are presented to clinicians. An alert may be useful if it identifies a genuine gap in care, but excessive or inaccurate alerts can increase workloads and reduce confidence in the system.

M42’s participation provides an opportunity to examine these issues in a practical healthcare environment. The partners have not disclosed a confirmed number of patients enrolled, completed clinical outcomes or a timetable for broad deployment.

Safety monitoring and evidence generation form part of the agreement

Beyond the hypertension initiative, AstraZeneca and Oracle intend to explore applications involving earlier diagnosis, clinical trial optimisation and drug safety monitoring.

The companies have identified oncology, cardiovascular, renal and metabolic diseases, respiratory and immunology, and rare diseases among the therapeutic areas that could be considered.

Safety monitoring involves examining information that may emerge from clinical trials, healthcare records and reports of adverse events. Analysts must distinguish possible signals that warrant investigation from events that occur for unrelated reasons.

AI could support the organisation and examination of that information, allowing relevant records or unusual patterns to be identified for further assessment.

However, a statistical association does not establish that a medicine caused a particular event. Clinical evaluation, pharmacovigilance procedures and applicable regulatory requirements remain necessary.

Using healthcare data for research also creates methodological questions. Records gathered during treatment may be incomplete or reflect differences in clinical practice between participating institutions.

Real-world evidence can support understanding of medicines outside the controlled conditions of a trial, but its reliability depends on data quality, study design and suitable methods for addressing bias.

The collaboration’s research applications will therefore require evidence that they produce reliable information and improve established processes without introducing unacceptable errors.

Governance must connect healthcare and research systems

Oracle and AstraZeneca plan shared governance arrangements, reusable technical components and agreed measures of performance for selected projects.

These elements will be important because clinical information is sensitive and may be held under different legal, contractual and organisational arrangements.

Health data can include diagnoses, medications, appointments and other details that require particularly careful handling. Systems accessing that information need appropriate permissions, secure processing arrangements and clear responsibilities for organisations using the results.

Technical interoperability presents a separate challenge. Two healthcare systems may record similar clinical information using different terminology, formats or conventions, complicating attempts to compare records accurately.

A shared technical component may help connect data sources, but it cannot guarantee that information collected for different purposes is complete or directly comparable.

The collaborators will also need procedures for assessing changes to AI systems over time. Updates affecting how a model interprets information can alter its performance, requiring further validation before a revised version is used in a clinical workflow.

Clear governance can establish which organisation is responsible for testing, data access, incident management and decisions about deploying or withdrawing a tool.

Evidence from initial projects will shape expansion

AstraZeneca and Oracle envisage projects across the United States, Middle East and Europe, working with local healthcare organisations and governments where appropriate.

Those ambitions would involve different health systems, information standards and legal requirements. An application tested in one clinical setting may need substantial adaptation before use elsewhere.

The initial agreement gives the companies a framework for examining potential applications rather than a commitment to implement an identical product across multiple countries.

Commercial and clinical success will depend on what the first projects demonstrate. Relevant measures could include recruitment accuracy, the time required to assess potential trial participants, staff workload and the reliability of information used in clinical decisions.

The hypertension programme in Abu Dhabi provides the first opportunity to generate that evidence. Subsequent deployment will depend on the results, the governance arrangements agreed with healthcare partners and the ability of the technology to meet the requirements of particular clinical services.