What are the benefits and harms of risk stratified screening as part of the NHS breast screening Programme? Study protocol for a multi-site non-randomised comparison of BC-predict versus usual screening (NCT04359420)

The following nine core outcomes will be compared at 6-months post completion of recruitment for those offered BC-Predict and those offered usual care (NHSBSP):

1. Screening attendance at first offered screening episode.

2. Screening attendance within 180 days of episode opening.

3. Number of technical recalls.

4. Number of recalls for assessment.

5. Number of routine recalls.

6. Number of breast cancer diagnoses (and type/grade).

7. Subsequent consultation in FHRP clinics (and mode: telephone or face-to-face).

8. Subsequent enrolment for more frequent screening.

9. Subsequent prescription of chemoprevention. Data will be collected on each of the following aspects of this: (a) participant agrees/disagrees in clinic to take chemoprevention, (b) chemoprevention not appropriate, (c) chemoprevention appropriate but prescription not filled, (d) chemoprevention appropriate and prescription filled.

Data for the nine core outcomes will be collected for each consented BC-Predict participant by the research staff for the 6 months following participants’ mammography appointment. For these participants, information will be available directly from NHSBSP and FHRP clinic records. For those in the usual care arm of the study, anonymised data will be provided by NHSBSP and FHRP services, to provide overall numbers for each of the core outcomes. We will prospectively record any refinements to procedures, to allow examination of how these impact on uptake of services.

We will also assess uptake of BC-Predict, and examine variation by study site. Where changes to recruitment procedures are made, we will keep notes of this, and examine the effects of these changes on uptake of BC-Predict, to inform how risk stratified screening should be rolled out. We will also examine variations in uptake of services by Index of Multiple Deprivation deciles derived from postcode of women [28] invited for NHSBSP, to assess any potential exacerbation of health inequalities brought about by BC-Predict.

The self-reported measures of potential harms and benefits of BC-Predict to be completed by a sub-sample of participants are shown in Table 1.

In total, approximately n = 18,700 women will be offered BC-Predict and n = 18,700 will be offered usual care NHSBSP over the total 16-month period. Based on the recruitment rate in PROCAS, n = 18,700 women being offered BC-Predict should result in 8000 women taking it up. The attendance rate to usual NHSBSP in Greater Manchester is 69% [5, 10].

Core outcomes will be compared for cohorts of women who are invited to BC-Predict and those who are invited to usual care (NHSBSP). Thus we will have in excess of 8000 participants in BC-Predict and NHSBSP groups for both comparisons: (a) within-site and (b) between sites over the same time period. The primary outcomes are binary. Logistic regression will be the primary statistical analysis method. We shall assess heterogeneity effects using interaction tests in the logistic regression, to examine differences in outcomes by time or location or screening type (prevalent v incident). Even in the presence of geographic or temporal heterogeneity of the effect, or of carryover effects continuing beyond the crossover period, we will still have sufficient data for a valid and fully powered comparison.

For core outcomes 1 and 2, we are interested in equivalence, in that we anticipate that invitation to BC-Predict will not substantially affect screening attendance. With 18,700 women in each group we will have in excess of 90% power to establish equivalence, defined as a 95%CI on the difference which does not exceed ±5% on the attendance rate at first offered appointment, if the latter is around 50% [37]. Similarly, we will have more than 90% power for the same comparison for eventual attendance within 180 days if the latter is around 70%.

Arguably, the most difficult to collect core outcome is 9, the proportion taking up chemoprevention. On the basis of PROCAS results we would anticipate that 1169 of the 8000 who consent to BC-Predict would have sufficient risk to be considered for chemoprevention and that 10% of these would take it up [8, 38, 39]. Thus 117 of the 8000 women (1.5%) receiving the intervention might be expected to be prescribed chemoprevention. It is anticipated that very few in the 18,700 sent the standard screening invitation would be prescribed chemoprevention, but even if as many as 0.9% did so, we would have 90% power to detect this as significant at 5% level with two-sided testing, and 80% power if 10% took up chemoprevention. The Greater Manchester Medicines Management group has agreed a shared care protocol stating that the initial prescription of tamoxifen and anastrozole should be made by a FHRP specialist. As such data from even those in the control arm should be available from prescriptions made in the FHRP clinics.

Analyses will focus on comparisons between the responses of the BC-Predict and NHSBSP groups at 6 months follow up, controlling for baseline responses and baseline patient characteristics. We will use ANCOVA, first with baseline responses to the same questionnaires as covariates, secondly treating both baseline and 6-month responses as related endpoints, using hierarchical linear models. Out of the available self-reported outcomes, we have selected the primary outcome to be anxiety (State Trait Anxiety Inventory) at 6-months, but we will also examine effects on all measures included in Table 1, as well as effects at 3 months. We will use the variables concerning knowledge, and attitudes to screening, as well as screening attendance, to assess the extent to which decisions to attend screening are informed, in line with a standard approach to assessing this [17]. The measures of health status (EQ. 5D-5 level) and capability (ICECAP-A) will be converted to preference weights using published algorithms [40] and population tariffs [41], as appropriate.

In addition to providing information about potential harms of BC-Predict, secondary analyses will examine whether women who are randomised to receive questionnaires differ in terms of uptake of screening or BC-Predict. This will inform about the likelihood of biases being introduced by comparisons of questionnaire responses in a possible subsequent definitive trial.

The sample size calculation is based on the six-item short-form of the state scale of the State Trait Anxiety Inventory [29], which measures general anxiety currently experienced on a scale of 20 to 80. Previous research in England with women invited to breast cancer screening found a mean state anxiety score of 37 [42]. A score of 49 has been found in patients with a diagnosis of anxiety disorder [43].

Assuming a two-tailed independent samples t-test, then n = 1054 (n = 527 women per experimental group) will be required to have 90% power (with α = 0.05) to detect a small standardised difference of d = 0.2. This equates to a difference between adjacent response categories (e.g. “not at all” and “somewhat”) on 2.5 of the 20 items on the full form of the scale. We anticipate that asking 1054 women per group will result in responses from n = 527 women per group being obtained at both baseline and 6 months, assuming a 70% response rate on both rounds.

An early economic analysis [44] will aim to identify the indicative estimates of the incremental costs and consequences and key drivers of the cost-effectiveness of a risk-stratified NHSBSP compared with the usual NHSBSP. A decision-analytic model-based cost-effectiveness analysis will capture the incremental NHS costs and consequences for a cohort of women eligible for NHSBSP in the UK over a life-time horizon. A decision-analytic model (a decision-tree combined with a published model) [11] will be structured to represent the care pathways of current NHSBSP practice (no risk feedback) and the proposed BC-Predict intervention in a sample of women eligible for the NHSBSP. The cost of the risk-stratified NHSBSP will be identified using a micro-costing study [45] and take account of the cost of the addition of SNPs to the risk estimation algorithm. The decision-tree will recognise the uptake of appropriate healthcare services (General Practice contact; FHRP Clinic referral and proportion of women starting chemopreventive medication e.g. anastrozole/tamoxifen/raloxifene). A published model [9] will be used to understand the lifetime impact on NHS costs and patient consequences of using different screening intervals based on risk-prediction, or usual NHSBSP. Using an economic model allows data assimilation from various sources (BC-Predict and systematic reviews; structured expert elicitation methods [46]) in a structured framework [47]. The model-base case analysis will focus on changes in health status (using EQ-5D-5 L) but explore the impact on capability (ICECAP-A) in a scenario analysis. These data will be obtained from the self-reported outcomes (health status (EQ-5D5L) [34]; capability (ICECAP-A) [35]) collected in the prospective study (see Table 1) and supplemented with published data to allow estimation of the impact on a life-time horizon. The EQ-5D5L [34] and ICECAP-A [35] have published preference weights that will allow calculation of quality adjusted life years for health and capability with and without the intervention. Parameter uncertainty in the decision-tree component of the model will be quantified using probabilistic sensitivity analysis for the base case analysis and scenario (capability) analysis. These two outcomes (health and capability) will then be used in two distinct value of information analyses (Expected Value of Perfect Information (EVPI) and expected value of partial perfect information (EVPPI)). The EVPI represents the maximum amount that should be spent on future research to gain perfect information to eliminate the possibility of a wrong (funding) decision. Further steps are then necessary to understand key parameters driving the uncertainty. This involves estimating the EVPPI that tells a decision maker which parameters are contributing to the uncertainty in the model and help to guide what type of additional evidence is most valuable.

To determine uptake and acceptability of a DNA based risk estimate as part of routine NHSBSP appointments, and to quantify the higher proportions of women at high/moderate and lower risk obtained by adding SNP information.

A subset of women will have the option to provide a sample of saliva from which DNA can be extracted. DNA will be extracted with standard techniques and currently known breast cancer SNPs associated with breast cancer typed. The results of this testing will be incorporated into the BC-Predict risk algorithm. Adding a genetic SNP score from a saliva sample to other risk factors not only potentially increases the accuracy of risk estimation, but also increases the discrimination of risk estimation, so that more women are identified as being at higher or lower risk, and fewer identified as being at population-average risk. It thereby increases the proportion of women identified at high-risk who can benefit from being offered NICE approved additional screening and drug prevention from 4 to 6%.

In total, it is expected that 1000 women will provide a DNA sample and receive a personalised breast cancer risk estimate incorporating their Polygenic Risk Score. All women invited for screening at Withington Community Hospital and Oldham Integrated Care Centre will be potentially eligible for the SNP sub-study, however, this will only be offered to women on a pragmatic basis depending on whether a member of staff is on site to assist with taking consent. A separate paper consent form will be completed by the participant in addition to the online consent form for the main study. Women giving their consent will be provided with an Oragene kit to place their salivary sample. They will be guided on site by a member of staff as to how to complete the sample.

The proportion of women in the 1000 providing saliva DNA who are classified as NICE actionable high and moderate risk as well as below average risk will be compared to their classification without a SNP Polygenic Risk Score. Chi square statistics will compare the difference between risk categories with and without the addition of the SNP Polygenic Risk Score. These data will also be used in the proposed economic analysis.

The main objectives are to explore service users’ views on acceptability of BC-Predict (interviews) and to assess the perceived impact of BC-Predict on the NHSBSP, FHRP Clinics and General Practice (focus groups).

In addition to the quantitative measures of impact of BC-Predict, we will also carry out qualitative work as part of a process evaluation to understand the key issues behind successful implementation of the BC-Predict system [48]. The qualitative work will comprise one-to-one interviews with NHSBSP service users to explore acceptability of BC-Predict amongst women at varying levels of risk, where there is currently a dearth of evidence [49]. It will also employ focus groups with healthcare professionals to investigate the implementation, delivery and impact of BC-Predict on the current NHSBSP. This qualitative work will give insight into capacity issues, indication of the training and support required to deliver BC-Predict on a larger scale, as well as communication challenges and pathways for both service users and healthcare professionals. This will enable us to build an evidence base to inform practice and policy should BC-Predict be rolled out to the wider NHSBSP.

For both interviews and focus groups, data will be analysed using a manifest level approach to thematic analysis as the themes are likely to be predominately deductive. Thematic analysis seeks and reports the patterns inherent within the data collected. It is a common qualitative analysis method that results in a rich, complex, yet accessible account of the data [51]. Themes will be coded at the manifest (or explicit) level [52]. It will do so taking an essentialist approach, which means that we aim to report the experiences, meanings and the reality of the participants [53].

Coding will be conducted systematically and iteratively. Negative cases will be sought to test the emerging coding framework. Regular coding meetings will be held to refine the coding structure. Data will be coded by independent researchers to check reliability in qualitative methods and ensuring that the fit between data and analysis is maximised. Percentage agreement on presence will be calculated. Coding will continue until the team are satisfied that codes and themes adequately describe and capture the data. Data will be stored and organised within Nvivo software.

To evaluate the impact of providing materials for women at low risk explaining that less frequent screening may provide a better balance of benefits and harms for them.

Women at lower risk of breast cancer are likely to receive less benefit from the NHSBSP but are more likely to experience overdiagnosis and treatment for cancers that will not cause them harm if left untreated. In this sub-study, we have chosen the risk threshold of 1.5% or below in newly screened women over 10 years, as this is the average risk level for a 40-year old woman, who currently would not be screened for a further 10 years. (Note that in the rest of the PROCAS study, a threshold of 2% or below was used to indicate below-average risk [10], so we use the term “low” to distinguish this distinct threshold in the present research). PROCAS indicated that 13.5% of women screened have a 10-year breast cancer risk of less than 1.5% when assessed by Tyrer-Cuzick and mammographic density [10]. This group of women are at a lower risk of developing breast cancer and the tumours they develop are much more likely to be early stage and slow-growing [10, 12]. Existing data suggest that a risk stratified NHSBSP may not only be potentially cost-effective [11], but also that it may be potentially more cost-effective to delay screening in low-risk women by optimising the screen interval [54].

Nearly 90% of the population have indicated that “screening is almost always a good idea” [55] and many women would feel aggrieved if they felt they were being denied a service inequitably. Relatedly, attendance at screening provides reassurance and peace of mind [56], so a lack of screening may result in increased worry about breast cancer. However, this view may be partly due to a lack of general awareness of issues such as overdiagnosis [57]. Many national screening figures believe that less frequent screening for women at low risk may be an important component of risk-stratified screening. Further, our ongoing developmental work suggests that this idea is acceptable to many women, including women who have received a below-average risk estimates.

In the final 4 months of the 16-month period of implementation of risk estimation, we will extend the offer of risk provision to include information about how, for women at low risk of breast cancer delaying further NHSBSP for a further period of 5 years may provide a better balance of benefits and harms for them. This information will be presented to all women in East Lancashire and Oldham as part of the invitation process, and repeated for women at low risk (< 1.5% over 10 years) as part of their risk feedback letter and accompanying leaflet. Every woman identified as being at low risk will be asked to complete the online questionnaire assessing harms and benefits of BC-Predict at 6 months, in contrast to the main BC-Predict study, where only a sub-sample will be asked to complete this questionnaire.

The primary outcome for this sub-study will be intentions to take up screening in 3 years in women assessed 6 months after being told that they are at low risk and receive the recommendation to delay attending screening. Our main focus will be on estimating the proportion of women intending to take up screening, but we will also formally compare this proportion with women who receive provision of “below average risk” breast cancer risk estimates (< 2%) but no screening recommendations, recruited over the previous 12 months of BC-Predict. Intention to attend screening is a consistent predictor of subsequent screening attendance (r = + 0.42 in a systematic review that identified k = 19 such tests) [58].

A qualitative process analysis will be conducted in line with MRC guidance [48]. A sample of low risk women (up to n = 12) who had received low risk estimates will be interviewed 1 month after receiving the feedback letter. They will be sample purposively to provide variation in the three screening sites (Oldham Integrated Care Centre, Burnley General Hospital and East Lancashire mobile breast screening van). Interviews will focus on the extent to which the possibility of receiving an estimate of low risk was considered before consenting to risk estimation, the acceptability of the information communicated and particularly the recommendation to delay screening, and any deliberations about delaying screening. Data will be analysed using a manifest, inductive thematic analysis, and will involve comparison of interviews with those women told they are at below average risk, but given no particular recommendation.

A decision analytic-model based economic analysis will be used to understand the potential relative cost-effectiveness of using a modified screening interval for women identified to be at low-risk of breast cancer as part of a stratified-BSP compared with the current NHSBSP. This analysis will build on an early economic analysis we previously conducted [11]. The model will be populated with data from the published literature and the current study to understand the relative costs and benefits of the modified screening interval for low-risk women assuming the perspective of the NHS and the impact on QALYs over the lifetime horizon for the defined population of women eligible for NHSBSP. Extensive sensitivity analysis will be used to understand the key drivers of relative cost-effectiveness when implementing a modified screening interval for low-risk women.