Investigating the utility of machine learning methods to predict prognosis and guide treatment decisions for people with lung cancer (Lung-ORACLE)

University College London (UCL) requires access to NHS England data for the purpose of the following research project:

Investigating the utility of machine learning methods to predict prognosis and guide treatment decisions for people with lung cancer (Lung-ORACLE)

The following is a summary of the aims of the research project:

Derive and validate multiple prognostic models for lung cancer patients, using a variety of statistical approaches to identify the most accurate modelling approach, with a view to improve treatment decisions for lung cancer patients

The following National Disease Registration Service (NDRS) National Cancer Registration and Analysis Service (NCRAS) datasets will be accessed:

• NDRS Cancer Registrations (inc. Route to Diagnosis Information)- necessary because one of the main objectives of this research is to develop an accurate prognostic tool for patients diagnosed with lung cancer, to aid treatment decision making. The study team therefore require identification of people diagnosed with non-small cell lung cancer (NSCLC); details of their diagnosis date to estimate survival time; details of their cancer to include as prognostic variables (e.g. stage of cancer, grade of cancer); patient demographics which will impact survival to include as prognostic variables (e.g. age at diagnosis, ethnicity, IMD); route to diagnosis as a prognostic variable (as emergency admissions are likely to present in latter stages compared to GP referrals); cause of death and date of death to attain vital status and vital status date for survival analysis models.

• NDRS Linked Cancer Waiting Times (Treatments Only)- necessary because the study team need to understand all treatments undertaken to estimate impact on survival. Fields such as whether a patients’ treatment may be useful as a prognostic variable to explore as this could potentially impact effectiveness of treatments and survival. Knowledge of whether treatment was undertaken in a clinical trial setting is important as this is likely to not reflect real world survival; whilst knowledge of treatment intent will allow analysis of survival following treatment dependent on intent.

• NDRS National Radiotherapy Dataset (RTDS) - necessary because the study team's prognostic model aims to estimate survival dependent on treatments undertaken. Hence, data in RTDS is essential to determine what treatments patients underwent and the impact on survival. The dates of treatments will also enable the study team to link to the other datasets to identify the ramifications of cancer therapy as outlined elsewhere.

• NDRS Systemic Anti-Cancer Therapy (SACT) Dataset - necessary because the prognostic model aims to estimate survival dependent on treatments undertaken. Hence, data in SACT is essential to determine what treatments patients underwent and the impact on survival. The dates of treatments will also enable the study team to link to the other datasets to identify the ramifications of cancer therapy as outlined elsewhere.

• NDRS Somatic Molecular Dataset - necessary because tumour genes impact patient treatment eligibility and prognosis. Hence the study team will be able to adjust the model to consider impact of treatment dependent on tumour gene on survival.

• NDRS Linked Hospital Episode Statistics (HES) Admitted Patient Care (APC) - necessary for the identification of patients' attendance, source of admission, length of stay, diagnoses during admission. This is because hospital admissions may be prognostic to survival, hence these variables can be included in the prognostic model to determine their impact on survival. In addition, the linked data set will enable the study team to identify hospital admissions following initiation of cancer treatment (or during treatment), which will allow them to consider outcomes additional to survival - such as predicted days spent in hospital in the year following treatment initiation or predicted number of hospital admissions in the year following treatment initiation - all of which will provide a proxy for quality of life following treatment initiation which was important to our PPI group.

• NDRS Lung Cancer Data Audit (LUCADA) - necessary because (1) it is a source of data not available elsewhere e.g. comorbidities, FEV1, performance status which are important prognostic factors for survival; and (2) if it turns out not to be complete enough, including the data at least allows the assessment of completeness and potential usefulness. It is acknowledged that this data is only available for some years (diagnoses up to 2014 and some variables partially within that).

• NDRS National Lung Cancer Audit (NLCA) - necessary because (1) it is a source of data not available elsewhere e.g. Charlson comorbidity score, and smoking status; and (2) if it turns out not to be complete enough, including the data at least allows the assessment of completeness and potential usefulness. It is acknowledged that this data is only available for some years (diagnoses up to 2014 and some variables partially within that).

The level of the Data will be Pseudonymised.

The Data will be minimised as follows:

• Limited to a study cohort identified by the NDRS as meeting the following criteria: Adult patients (aged 18 and over) who received a diagnosis of non-small-cell lung cancer (as defined by specific ICD codes) in the UK between 2010-2021

UCL is the controller as the organisation responsible for ensuring that the Data will only be processed for the purpose described above.

The lawful basis for processing personal data under the UK GDPR is:

Article 6(1)(e) - processing is necessary for the performance of a task carried out in the public interest or in the exercise of official authority vested in the controller.

The lawful basis for processing special category data under the UK GDPR is:

Article 9(2)(j) - processing is necessary for archiving purposes in the public interest, scientific or historical research purposes or statistical purposes in accordance with Article 89(1) based on Union or Member State law which shall be proportionate to the aim pursued, respect the essence of the right to data protection and provide for suitable and specific measures to safeguard the fundamental rights and the interests of the data subject.

The funding is provided by the National Institute of Health and Care Research. The funding is specifically for the study described. The funder will have no ability to suppress or otherwise limit the publication of findings.

Amazon Web Services provides IT hosting services that support the UCL Data Safe Haven and will store the Data as contracted by UCL.

Data will be accessed by a PhD student affiliated with UCL, substantive employees of UCL and a individual on an honorary contract with UCL. Any individual working with the Data held under this Data Sharing Agreement (DSA) must have completed relevant data protection and confidentiality training and are subject to UCL’s policies on data protection and confidentiality. Individuals accessing the Data will do so under the supervision of the Principle Investigator. UCL will be responsible and liable for any work carried out by individuals accessing the data who will only work on the Data for the purpose described in this DSA.

To inform this research, a Public and Patient Involvement and Engagement (PPI) group completed an online survey and participated in an online meeting. The group supported the collection of the data for the purposes described above.

Further, following consultation with healthcare professionals (HCPs) to inform this research, they stressed that any model deployed in healthcare must be appropriately designed for use to support patients and ensure it is helpful. Members of the PPI group upon initial consultation additionally had many thoughts and ideas around prognostic model design/use and were keen the patient/public voice was heard. Consequently, throughout the research, and in addressing the objectives, it will adopt an approach/methodology with commitment to involving HCPs, those who support patients, and members of the public in model design and development.

Approaches to this will include Multidisciplinary expert panel meetings and PPI group sessions. The PPI group was convened by advertising involvement via UCL PPI networks, charities (BME Cancer Communities, Roy Castle lung cancer foundation, Maggies, Macmillan cancer support) and patient advocate groups (Independent Cancer Patients’ Voice). The group comprises individuals with personal experience of lung cancer or other cancers, those impacted by cancer of a relative, and those interested in this research. Via professional connections and reaching out to charity staff, the Multidisciplinary expert panel was formed to consist of professionals across the country. To offer a broad range of views on model development and use, job roles include medical oncology, clinical oncology, thoracic surgery, nurse specialist, respiratory physician, palliative care physician, Macmillan GP, Maggies centre head. Both PPI and Multidisciplinary expert panel meetings provide a platform for the groups to offer their views on prognostic model design and approach; for example what is important to patients when making cancer treatment decisions, and how user friendly are current prognostic models in clinical use for other cancers. In recent meetings for instance, the groups have provided feedback on an initial prototype for the lung cancer prognostic model - highlighting the need for appropriate language, simplicity and security.

In line with the national data opt-out policy, opt-outs are not applied because the data is not Confidential Patient Information as defined in section 251(10) and section 251(11) of the National Health Service Act 2006.

Where individuals have opted out of disease registration by the National Disease Registration Service (NDRS), their data has been permanently removed from the registry and therefore will not be disseminated under this Data Sharing Agreement (DSA). https://digital.nhs.uk/ndrs/patients/opting-out

No data will flow to NHS England for the purposes of this Data Sharing Agreement (DSA).

The NHS England NDRS will provide the relevant records from the above-listed datasets. The Data will contain no direct identifying data items. The Data will be pseudonymised and individuals cannot be reidentified through linkage with other data in the possession of the recipient.

Once received, the requested data will be uploaded to the UCL Data Safe Haven. Amazon Web Services (AWS) is a processor acting under the instructions of UCL. AWS’ role is limited to secure backup of data stored in UCL’s Data Safe Haven. UCL uses offsite data centre services provided by VIRTUS data centre. VIRTUS does not have access to the data.

The Data will be accessed by authorised personnel via remote access.

The Controller(s) must confirm and provide evidence upon audit by NHS England that access via any remote device complies with the data security obligations within this DSA and the Data Sharing Framework Contract.

For remote access:

- Remote access will only be from secure locations situated within the territory of use (as further restricted elsewhere within the DSA if so done) stated within this DSA;

- Access controls granting users the minimum level of access required are in place;

- Remote access is only via secure connections (e.g., VPNs or secure protocols) to protect data;

- Multifactor authentication (MFA) is required for remote access;

- Device security, including up-to-date software and operating systems, antivirus software, and enabled firewalls are utilised for the remote access;

- All remote access is undertaken within the scope of the organisation’s DSPT (or other security arrangements as per this DSA) and complies with the organisation’s remote access policy.

The above applies in addition to any condition set out elsewhere within the DSA (e.g. who may carry out processing, and for what purpose).

Remote processing will be from secure locations within England. The data will not leave or be accessed outside of England at any time.

Data will be accessed by an individual with an honorary contract with UCL. Aside from this individual, access is restricted to substantive employees and students of UCL who have authorisation from the Principal Investigator.

All personnel accessing the Data have been appropriately trained in data protection and confidentiality.

The Data will not be linked with any other data not already referenced in this Data Sharing Agreement (DSA).

There will be no requirement and no attempt to reidentify individuals when using the Data.

Analysts from UCL will process the Data for the purposes described above.

The expected outputs of the processing will be:

• Submissions to peer reviewed journals expected from 2026 onwards

• Report to NIHR

• A PhD thesis due for submission

• Presentations at appropriate conferences: British Medical Journal Future Health,

American Association for Cancer annual meeting, British Thoracic Society.

• Production of a prognostic tool to aid treatment decisions for patients diagnosed with lung cancer. The tool will incorporate (and test user thoughts on) aspects of quality of life never before considered for such prognostic tools used in clinical practice. It is not intended that the real-world model implementation of the prognostic tool will be covered during this research. This will involve future work to undertake a large multi-centre trial to assess model effectiveness and implementation strategy development and testing of this.

The outputs will not contain NHS England Data and will only contain aggregated information with small numbers suppressed as appropriate in line with the relevant disclosure rules for the dataset(s) from which the information was derived.

The outputs will be communicated to relevant recipients through the following dissemination channels:

• Journals

• Workshops

• Webinars

• Social media

Outputs will be published from March 2026 onwards.

• Public reports

• Press/media engagement

The findings of this research study are expected to contribute to evidence-based decision-making support for policy-makers, local decision-makers such as doctors, and patients, in the following ways:

1) By working collaboratively with clinicians and patients throughout the study team will ensure the research is interpreted and presented appropriately so as to increase confidence in prognostic models for cancer

2) The work will inform how such prognostic models (including lung and other cancer types) should be designed for use in clinical practice in the future.

3) The study team will review how clinician behaviour in therapeutic decision-making may change due to the availability of model predictions, and the prognostic accuracy afforded by the models versus clinician judgement alone. This will provide an estimate for the potential impact such models may have on lung cancer patient outcomes if implemented widely into clinical care.

4) Following on from this work, future implementation of the lung cancer prognostic tool (following a future multi-centre trial to assess model effectiveness) has the potential to improve clinical outcomes for lung cancer patients and aid clinicians in recommending treatments and providing accurate prognostic predictions. From a patient perspective the uncertainty around outcomes and treatment decisions will be reduced.

Hence, this work provides evidence paving the way for designing effective, implementable prognostic models in the future; informing best practice to improve the care, treatment and experience of health care users relevant to the subject matter of the study.

It is hoped that through publication of findings in appropriate media, the findings of this research will add to the body of evidence that is considered by the bodies, organisations and individual care practitioners charged with making policy decisions for or within the NHS or treatment decisions in relation to specific patients.

The project team plan to engage with a variety of charities and use connections with NIHR to ensure that findings reach a wider audience as possible.

Not stated in the register.