The AgeX Trial

The AgeX Trial

The cluster-randomised AgeX trial will assess reliably the risks and benefits of offering an extra screening invitation to women aged about 47-49 (who will all be offered routine screening anyway about three years later) and, separately, of offering up to 3 additional triennial invitations to women after age 70 (who will already have been offered routine triennial screening at ages about 50-70). Linkage with routinely collected government records will help assess the short- term and long-term effects of the additional invitations on breast cancer incidence, patterns of treatment, breast cancer mortality, and other outcomes.

Recruitment procedures

This trial is embedded within the routines of the BSP, which currently uses two- view digital mammography. Other than randomisation, all aspects of screening will be conducted exactly as normal in the BSP, following its routine procedures (delays in which can slightly affect the exact ages at which invitations arrive). No direct contact with participants will be made by the research team, and the statistical analyses and reports will be of anonymised data.

The breast screening units in the 2009-10 pilot study became part of the main AgeX trial, as did most other NHS breast screening units in England. All use the same system of generating screening invitations (described below); the units not participating were mainly those using somewhat different systems, or with staff limitations or other operational issues.

A national database is used to create screening invitation batches, perhaps every few weeks, for each local breast screening unit. An invitation batch typically lists several hundred women of appropriate age who are recorded as registered with the same general practitioner or living in the same geographical locality (eg, one village, or one part of a town) where the local breast screening unit will next be working. Once generated, this batch is used by the local breast screening unit to invite the women in it for mammography. As this particular locality will not be visited again for about 3 years, the first routine invitation may well be somewhat after age 50, and the last somewhat before age 70.

The program that generates these batches was updated during 2016-18 to use each woman’s exact age. (Before then it had estimated age by subtracting the year of birth from the current year¹). In addition to the 50-70 age group, which will be offered routine screening invitations regardless of whether they are already in the trial, the program also identifies the new entrants into the trial, who are the cluster of age 47-49 and the cluster of age 71-73 years.

Each batch is randomly allocated to invite for screening either the trial entrants aged 47-49 or those aged 71-73 years, as shown in the figure. (The women aged 50-70 are unaffected by the random allocation of the batch; they are invited as normal and are not new entrants into the trial). The batch could also include trial participants invited 3 years ago at ages 71-73 (but now aged 74-76) for a second invitation and eventually those invited 3 years ago at ages 74-76 (but now aged 77-79) for a third invitation.

PHE (NHS England is now responsible for the breast screening programme) does the random allocation of each batch by their own specially written computer program with equal (50/50) probability and no stratification. (Lists of those randomised are eventually forwarded to the trial investigators.) A few women are excluded before randomisation because, for example, they have asked to be withdrawn from the national breast screening programme, are recorded as having had a bilateral mastectomy, or had been screened recently.

Each participant enters the trial on the date when the screening batch she is in is created and randomised; invitations generally go out a few weeks later. New entrants in the batch who are randomly allocated not to be invited join the trial as controls. This is approved by the Confidentiality Advisory Group (CAG).

Results are examined regularly by the Data Monitoring and Ethics Committee, and when clear answers emerge they will be published. The total number of women entering the trial (half offered additional screening and half not) is not a fixed, pre-determined sample size. If substantial uncertainty still persists, entry into the study may well continue.

Trial participant information

Screening units inform local General Practitioners that the trial is taking place in their area and a poster about the trial is displayed in their surgery (Annex 2). The BSP informs women and their GPs of the outcome of screening. Women invited for screening in this trial will be treated in exactly the same way.

Women invited for screening under the BSP receive with their screening invitation the standard BSP patient brochure. Women of any age, whether or not in the trial, who are invited for screening in an area where the trial is in progress receive with their invitation whatever version of the standard BSP patient brochure is current, plus the trial participant information sheet (Annex 3). This explains that the data will be analysed by research workers at the University of Oxford, who are responsible for the organisation of the trial.

Physical security

AgeX records are stored in the Richard Doll Building, a secure building within the Nuffield Department of Population Health (NDPH) at The University of Oxford Old Road Campus. Swipe card access is present on all external doors. External doors are monitored by CCTV. Visitors and deliveries are required to report to the Reception for verification by Reception staff, and all NDPH employees are encouraged to challenge anyone they don’t recognise in order to confirm identity and authorisation. High security areas (e.g. server rooms) are physically and electrically separate from other NDPH facilities and have additional security locks in place. Access is restricted to relevant staff. Offices are secured by door locks out of normal working hours when not in use. Open plan work space is secured by internal doors with additional swipe card or proximity card access. Response procedures to environmental and external threats (such as fire, flood and explosion) are covered in the NDPH IT Disaster Recovery Plan. Server rooms have air conditioning units to ensure that the servers operate within operating limits specified by the equipment manufacturers. Offices are secured by door locks out of normal working hours and when not in use. Richard Doll Building server room power is supplied from multiple mains feeds, with equipment split between feeds. Main servers and other key hardware are protected by uninterruptible power supply units (UPS) in order to maintain service in the event of a power outage and prevent corruption of information. A back-up generator provides additional cover in the event of a prolonged power outage. The Big Data institute (BDI) server room has a single UPS system with two separate circuits. Each rack has two power supply units fed from each circuit, power is taken from the least loaded distribution strip in the rack. All NDPH staff have access to on-site security staff as well as public emergency services through the internal telephone system. 

Data collection and handling

AgeX records are processed securely within the Nuffield Department of Population Health (NDPH). The University of Oxford acts as Sponsor and Data Controller. Data security complies with the Data Protection Act (University of Oxford: registration number Z575783X) and with Unit, Departmental and University data security policies. The NDPH has current NHS Data Security Protection Toolkit accreditation for storage of linked NHS data (ref: EE133863- MSD-NDOPH-NDPH); relevant data security and governance policies are available on request from the study team. All recruitment data and linked follow- up data are stored separately with restricted access within the study team, and used to prepare de-identified datasets for analysis. Access to identifying information provided by participants (name, address, date of birth,) is restricted within the study team. Files are stripped of identifiers such as name or NHS number before any data analyses. All those working on data are bound by legal agreement not to identify participants, and where possible data are provided in a form which minimises this risk (eg. tabulations, or suppressing part of birth date, or small numbers in data cells).

The Breast Screening Select system provides information on trial entrants, including patient identifiers such as name, NHS number and date of birth, for record linkage. Participants’ records are linked securely and electronically to:

•  NHS screening records (for screening history and information about procedures done and diagnoses) at NHS England;
• Death and cancer registry data, including the Cancer Outcomes dataset, held by NHS England (previously Public Health England for information on cause- specific mortality and details of incident cancers, including tumour histology, size, stage, grade, nodal involvement and receptor status, as well as on treatments such as chemotherapy and radiotherapy;
• NHS Hospital Episode Statistics, held by NHS England (formerly NHS Digital and Health and Social Care Information Centre)) for information on cause- specific hospital admissions and procedures, including surgical treatments such as mastectomy, lumpectomy, axillary clearance, etc; and
• Data approved for anonymised transfer from a nationwide prospective study with information on quality of life.

Such routine records cease if women are notified as emigrating, on which date their trial follow-up is censored, but this will probably affect fewer than 1% of participants per decade.

Deaths of women with any history of breast cancer will be reviewed, blind to the random allocation, by an Endpoint Committee to determine whether this was a breast cancer death (defined, because of the difficulty of determining the exact cause of death, to include all deaths with uncontrolled life-threatening breast cancer thought to have been present). These breast cancer deaths provide the principal endpoint of the trial, which is breast cancer mortality.

For each probable breast cancer death, trial organisers will seek a narrative of the diagnosis and treatment of that cancer. This will include the date on which a relevant breast abnormality was first found, how it was found and investigated, the date of diagnosis of breast cancer and the characteristics of the cancer at the first diagnosis (including histology, size, nodal spread, distant spread and receptor status).

Analysis plan

Analysis as two separate trials, one in younger women and one in older women The findings will be monitored, analysed and reported as two entirely separate trials. One is a trial among younger women (randomly allocated at age 47-49 to additional screening invitation or control) of the effects of an extra screening invitation 3 years before routine screening would normally have begun. The other is a trial among older women (randomly allocated at age 71-73 to additional screening invitation or control) of the effects of up to 3 extra screening invitations among those who have had their final routine invitation.

Primary analyses

The primary analyses among older women will be of breast cancer mortality

1.  Up to but not including age 80, and, eventually,
2. Subdivided by separate time periods (0-4, 5-9, 10-14, etc years after the exact date of randomisation) and by receptor status (ER+, other).

1.  Up to but not including age 60, and, eventually,
2. Subdivided by separate time periods (0-4, 5-9, 10-14, etc years after the exact date of randomisation) and by receptor status (ER+, other).

In younger women deaths from breast cancer diagnosed after the first routine screen at age 50-52 would not be expected to be affected by the random allocation and hence will be uninformative. To achieve greater sensitivity, the primary analyses will consider separately these uninformative breast cancer deaths and all other breast cancer deaths, if this can be done reliably without introducing any material bias between the two arms of the trial.

In both age ranges most deaths will be from causes other than breast cancer. Although results on mortality from other causes (and from all causes) will be reported, there is expected to be insufficient power for crude analyses of all-cause mortality to assess reliably the effect of additional breast screening on all-cause mortality.

For reasons of statistical power, therefore, the most reliable estimate of the effect of additional screening invitations on all-cause mortality may well come from combining the effects on breast cancer mortality (and any procedural mortality) estimated from this trial with the small long-term effects of medical radiation estimated from other studies, assuming no other effects on mortality.

It is expected at present that first results of the primary analyses will be released for peer review in the mid-2020s and that thereafter observations will continue and more definitive findings released periodically. If at any stage, however, the Data Monitoring and Ethics Committee should advise that there is proof beyond reasonable doubt that additional screening at age 47-49 years or throughout the 70s is appropriate without any material adverse effect on other mortality, the results would be submitted promptly for peer review.

Main subsidiary analyses

The plausibility of the assumption of no material effects on other mortality will be checked by subsidiary analyses of cause-specific mortality, interpreted with due allowance for the effects of chance when multiple endpoints are analysed. Although subsidiary analyses of all-cause mortality will also be reported, they will not contribute to the primary analysis of breast cancer mortality.

The main subsidiary analyses will be of the details of breast cancer incidence and of the patterns of breast cancer investigation and treatment. Information on screening outcomes, such as recall and biopsy rates, will be collected not only for the women randomised to extra screening invitations but also for the first routine screening invitations at ages 50-52. In addition, many other outcomes available from linkage with routine NHS records will be assessed.

Exclusions from primary analyses and main subsidiary analyses

The primary and main subsidiary analyses will be restricted to women who can be identified and followed up and for whom unbiased evidence (which cannot itself be altered by the random allocation) shows that allocation of whether or not to send a screening invitation is likely to determine whether or not the woman is actually screened.

These analyses will therefore exclude: duplicate randomisations; women whose NHS records could not be flagged; women who before randomisation had already withdrawn from the BSP; women who had already died or were known from unbiased records to have moved away from the address held by the Breast Screening Select (formerly National Breast Screening) System; women known from unbiased records to have had cancer (except non-melanoma skin cancer), breast disease, or breast surgery; and women entered since mid-2016 who were not quite 71 at the time (Annex 1).

Of the remaining women, a proportion would not take up a screening invitation even if they were sent one, thereby diluting any effects of inviting women for screening. To reduce this dilution and increase statistical power, women who did not take up their previous cancer screening invitation will be excluded from the primary and main subsidiary analyses. (For, in older women a strong predictor of acceptance of an invitation for breast screening is previous attendance for breast screening at their last routine invitation. Likewise, in younger women previous attendance for routine cervical screening is a strong predictor of acceptance of a first invitation for breast screening.)²

Statistical power

Women aged 47-49: The median age of the women randomised in their late 40s will be 48 years. Among one million women during the 2010s with no breast cancer before age 48 who are randomly allocated not to have a screening invitation before age 51, about 1500 might be expected to die before age 60 from a breast cancer diagnosed at or before their first routine screening invitation 3 years later.

If an additional screen at age 48 would reduce this by 15% to 1275 expected deaths, then an evenly randomised trial among 2 million such women with perfect compliance (100% uptake) with all invitations would reliably detect this expected difference of 225 relevant deaths.

With 4 million evenly randomised but with a more realistic uptake rate of only two-thirds (and negligible screening by the BSP or other providers among those not invited), the expected difference would be 300 relevant deaths (2700 vs 3000, with standard error 75 and hence a 92% chance of achieving 2p<0.01). The exclusions described above will somewhat increase this statistical power, as will longer follow-up (13).

Intra-cluster correlation has little impact on the power calculations, as the screening batches are so small (median about 100 women aged 47-49, in the pilot study) that only a small proportion will have more than 1 woman with a breast cancer diagnosed at or before routine screening began that causes death.

Women aged 71-73: Among one million women during the 2010s who had been screened at age 69 with no relevant abnormality detected, and who still had not been diagnosed with breast cancer by age 72, about 4000 might be expected to die of a subsequent breast cancer before age 80.

If an additional screen at age 72 would reduce this by 15% to 3400 deaths, then in an evenly randomised trial of 2 million such women with perfect compliance (100% uptake by those invited and no self-referral among other women) the expected difference would be 600 relevant deaths, ensuring a highly significant result.

But, women over 70 are already able to request screening every 3 years, and a minority already do this (4). With a more realistic uptake rate of 70% of those invited accepting the invitation and a realistic self-referral rate of about 10%, in a trial of 2 million such women the expected difference would be only about 360 relevant deaths (3580 vs 3940, with standard error 87 and hence a 94% chance of achieving 2p<0.01).

Again, however, the exclusions described above will somewhat increase this statistical power, as will longer follow-up and additional screening invitations at ages 74-79.

Consent, confidentiality and trial supervision

Section 251 approval for including women in the trial without their consent and for the use of patient-identifiable data without consent was obtained initially from the National Information Governance Board for Health and Social Care, then annually from it and any successor, which is currently the Health Research Authority Confidentiality Advisory Board (CAG). With respect to consent for screening, the standard procedures of the NHS BSP apply, whereby attending screening is taken as implied consent.

Individual records will be linked to NHS England datasets, but will be anonymised once data linkage has been completed. The trial will be conducted in accordance with all relevant aspects of the requirements of CAG and the Data Protection Act. The data will be treated with appropriate confidentiality, and used only for medical research.

Datasets will be analysed only in anonymised form, and publications will not identify individuals. Results will be disseminated in peer-reviewed open-access publications, at medical conferences and on the web.