Lessons Learned from Setting Up a Prospective, Longitudinal, Multicenter Study with Women at High Risk for Breast Cancer

Lesson 1. Study Phases and Cohort Size

The first and predominant challenge is that, to obtain a longitudinal series of prediagnostic samples from a sufficient number of breast cancer cases, a very large, long-running biobank needs to be constructed. Even in cohorts of women at high risk for breast cancer, only about 1%–6% of the study participants develop an event over a period of 4–8 years (9–11). Therefore, much effort is needed to set up the biobank and to acquire and store the great number of samples, whereas final analysis will only be performed in a nested case–control design on a limited number of cases and controls to answer the research question. Since the start, the focus of the study should be on setting up a large, longitudinal cohort of healthy, high-risk women. The required time to reach a large number of inclusions and a sufficient follow-up period makes this “inclusion and biobanking phase” the most time-consuming part of the study phases. When enough cases have been developed to provide sufficient statistical power to test the predictive value of changes in biomarker levels in relation to cancer occurrence, the next phase can be initiated. This “sample and data analysis phase” includes selecting matching controls based on the characteristics of the cases, retrieving samples of the selected study subjects from the biobank, analyzing biological samples in the laboratory, and performing subsequent statistical analyses. In the meantime, the “inclusion and biobanking phase” becomes a “follow-up and biobanking phase,” allowing for more women to develop events for the subsequent validation studies.

A few potential pitfalls should be taken into consideration. The first pitfall is to only engage experts at specific phases during the study. Early commitment and involvement of several experts lets the study run smoothly from the start and avoids making incorrect assumptions or decisions. We recommend involving all necessary experts who need to be engaged from the start of the study. A list of suggested experts for the presented study design and cohort is presented in Table 1.

Another pitfall is to mainly focus on the number of inclusions in the cohort, while it is also relevant to regularly monitor more aspects of the cohort, such as study drop-outs. Specifically for biomarker studies, issues such as the number of successfully, serially acquired samples for analysis should be monitored closely. Some samples may be absent because of missed visits, lost by unexpected flaws, processing errors, and/or laboratory technical failures. These issues and recommendations on how to avoid them will be further elucidated in lessons 3, 4, 7, and 9. Finally, predicting the time frame of the study is challenging. From our experience, more than 5 years are needed for the initial “inclusion and biobanking phase.” This depends on the extent of the abovementioned hurdles, which cannot be anticipated beforehand. We recommend to monitor the participation rate and to be prepared for a prolonged time frame of the study.

Lesson 2. Funding

The lack of fast results in prolonged longitudinal studies makes funding less appealing, both to governmental and private sponsors. Preliminary data and achieving milestones are usually the foundations to apply for the next grant. However, given the long “inclusion and biobanking phase,” which was described in lesson 1, limited data will be generated for a long period of time. This is a relevant and delicate challenge, because progress of the study is highly dependent on getting continuous, long-term funding for one project. Discontinuation of funding is prejudicial at any phase of the project. Essential budgetary costs also refer to the salaries of the team members involved in running the study (Table 1); intermittently downsizing the team due to financial shortcomings endangers, among others, adequate data acquisition and, in the end, its interpretation.

To ensure continuous funding, principal investigators (PI) should apply for several grants throughout the course of the study. Each grant application should focus on one of the study aims. Infrastructural grants are ideal to achieve biobanking aims in the “inclusion and biobanking phase,” whereas research grants, including high-risk pilot grants and proof-of-concept grants, are more suited for aims focused on acquiring results as part of the “sample and data analysis phase.” Private sponsors that share the vision that a prolonged study is required to obtain translational results, may be instrumental.

Lesson 3. The Inclusion and Exclusion Criteria are Dynamic

Lesson 4. Delayed Inclusion Moment and Discontinuation of Participation

Both studies aimed to pose as little as possible extra burden on the study participants with regard to hospital visits. This means, in theory, that the first opportunity to inform potential candidates about the study is during their regular scheduled screening appointment. By Dutch law, it is required that individuals who are asked to participate in a scientific study should have a reasonable amount of time to consider this request, in principle prohibiting start of the study on the day that information about the study is provided. Consequently, by holding on to having sample acquisition coincide with a regular screening appointment, the actual moment of study inclusion could be delayed for 6–12 months (i.e., when the next screening appointment is scheduled). This observation led to an adapted protocol in the TESTBREAST study, where research nurses now attempt to inform potential participants about the study in advance by post or e-mail, so that inclusion in the study can start at the moment of the already scheduled screening visit.

Given the long time period until publication of study results, discontinuation of participation among enrolled women is prone to occur. Therefore, efficient strategies to keep compliance should be applied. An essential starting point is having dedicated research nurses who clearly describe the aim of the study to participants, and the importance of collecting serial samples. Also, a movie portraying a study visit and/or explaining the set-up of the study provides further clarification to participating women. To keep study subjects informed about the course of the study, one should consider sending regular study newsletters, frequently update the study website, and organize annual meetings where the study team is present. The latter provides study participants ample opportunity to ask questions, which in our studies has proven to be very well appreciated. A combination of abovementioned strategies leads to a clear realization of the importance of their study participation. This is not only relevant for the participating women, but also for team members in participating hospitals.

Despite these various strategies, drop-outs are almost an implicit component of long-term clinical studies. The main reasons for women to stop participating in the NAF study, in rank order (from highest to lowest), were: (i) reaching the end of the study according to protocol (established as 10 years in the NAF study; 18%), (ii) repeated unsuccessful aspiration (17%), (iii) preventive bilateral mastectomy (8.6%), (iv) loss to follow-up (7.4%), (v) no screening appointments at the hospital (6.7%), and (vi) development of breast cancer (6%), among other reasons. Planned prophylactic mastectomies, specifically for BRCA1 or BRCA2 mutation carriers, are unavoidable drop-outs that can be expected. For the NAF study, a specific drop-out reason was repeated unsuccessful NAF acquisition (i.e., 0 μL), which can be related to parameters such as menopausal status, spontaneous nipple discharge, breast size, bilateral oophorectomy, and previous use of hormone replacement therapy or antihormonal treatment (34). Other reasons were logistic issues, such as relocation of participants to another area and hospital, no longer having time to come to the outpatient clinic, or the end of hospital screening (i.e., return to regular out-of-hospital nationwide screening). These problems could theoretically be surpassed if blood draws could be taken at any location (e.g., at a blood draw clinic or at the general practitioner's office). However, in practice, women may overlook or postpone having their blood drawn. To avoid long waiting times as a potential barrier, the TESTBREAST study team established a rule of priority for blood sampling at the laboratory. This resulted in an increased willingness of up to 86% of the study population to have their blood samples taken. All of these factors should be considered when estimating the number of required inclusions.

To summarize, to safeguard the inclusion rate, the study team should ideally inform potential candidates about the study before the next scheduled clinical screening visit. For subsequent study visit invitations, the research team should create and maintain a recall system to invite study subjects timely and sample acquisition should be facilitated by implementing a fast track for obtaining blood samples. On top of that, the core study team should keep study subjects and other participating team members informed about the course of the study. These strategies diminish the number of unnecessary drop-outs and missed study visits.

Lesson 5. Logistics

Given the high number of participants and an even higher number of study appointments in this trial design, it can be challenging to maintain continuous registration of the incoming data. For example, in the NAF study, a two-center study in which every study visit takes about 1 hour, up to 12 study visits are planned every week. Study visit registration aspects include digital processing of a questionnaire and collection, processing, registration, and biobanking of NAF and blood samples which are aliquoted into 22 vials (three vials per breast for NAF, 10 vials for serum, and six vials for plasma). Questionnaire data comprise lifestyle and hormonal factors. Sample registration data comprise NAF sample color, including bloody appearance, consistency, and volume per aliquot. NAF color can vary between study subjects, between breasts from the same woman, and even within a breast. NAF color registration provides insight into the color variation and permits exploration of its association with breast cancer risk. NAF samples should be labeled as bloody or nonbloody before analysis, as bloody NAF samples could lead to alterations in biomarker analyses compared with nonbloody NAF. TESTBREAST study visits comprise a questionnaire and a blood draw, which, in contrast to the NAF study, does not need to be performed by trained study team members, but can be done by regularly trained hospital personnel. Because the TESTBREAST is a nine-center study with one to four visits per participant per year (as established by hospital-specific screening guidelines and commitment by study participants), the number of incoming questionnaires and blood samples to be processed and stored is high. Therefore, it is key to set up a well-established administrative process. To guarantee a good cohesive administration that is appropriately maintained according to data-monitoring rules, annual monitoring by an independent clinical research associate (CRA) throughout the complete course of the study is advisable. In addition, both a data manager and a(n) (online) data management system (e.g., ProMISe for the TESTBREAST study) are essential to keep the incoming information up to date and facilitate the traceability of samples (37). Recent developments of online databases also allow integration of online questionnaires. This ensures completeness of the questionnaires and automated integration of the information directly into the database.

A turnover of study team members is expected during a long-term study. Such a turnover can occur in team members that are daily and routinely involved in the study, like research nurses and the coordinator of the clinical study. In the Netherlands, coordination of a clinical study is usually performed by an MD-PhD student. As they ought to finish their research project within 4 years, several PhD students are usually involved in such a project before it reaches its end. A way to maintain consistent working procedures is to establish and frequently update standardized operating procedures (SOP), which are supervised by a CRA. Also, contact information of team members should be updated and departing team members should train new incoming members about the SOPs. In multicenter studies, a fixed contact person on-site is responsible for the correct local execution of the study. A site training of a specific study technique (such as the nipple fluid aspiration) might be necessary in some studies. Structured meetings with the local study coordinators are recommended to secure proper collection of data.

Lesson 6. Sample Processing

A difficulty of long-term, prospective studies is to guarantee optimal quality of biological material for multiple years until the time of analysis. For instance, a prolonged time in the freezer is accompanied with the risk of sample quality loss and loss of sample by evaporation (38). Therefore, correct sample handling during collection and storage is essential. Sample collection and processing aspects to be considered include consistent use of the same type of collection tubes and buffers, a predefined time until sample processing and freezing, and a defined monitored temperature at which samples are stored. In addition, aliquoting samples is a good strategy to later avoid freeze–thaw cycles, which possibly influences sample quality. Finally, pilot sample analysis should not be postponed until the end of the study, as it allows timely sample quality monitoring and assessment of natural temporal biomarker fluctuations (Fig. 1). In this context, pilot testing to ensure that the collected samples are adequate for biomarker analyses should be performed. As an example, in the NAF study, pilot testing has shown that, regardless of the RNA concentration at the start, miRNA qPCR can always be measured given the high sensitivity of this technique (39).

Working with a number of hospitals over a long period of time comes together with the potential pitfall that different types of collection tubes and a variety of buffers are used across centers or even within centers, which could hamper a comparable sample analysis. It is important to maintain consistency among and within participating hospitals. Recommended approaches include providing sets of collection tubes, labels, laboratory forms, and questionnaires to all centers, as has been part of the standard procedure in the NAF and TESTBREAST studies.

Finally, prolonged processing time of samples can lead to the alteration of biomarker characteristics and cause the final results to deflect (40). Therefore, continuous presence of a team member in the laboratory is required to process new samples within the defined time limit. The standards used for sample processing in the NAF and TESTBREAST studies are depicted in Table 3.

Lesson 7. Sampling at Events

For both longitudinal cohorts of the NAF and the TESTBREAST studies, multiple samples from study subjects are prospectively acquired; the last sample is collected if and when an event occurs. As the primary aim of the studies is to compare data from liquid biopsies obtained at the time of an event with those acquired before diagnosis, sampling at the time of an event is valuable. This sample is relevant in the investigational setting, because biomarker levels at the time of the event best reflect the carcinogenetic signature and, as such, function as the reference levels in the paired analysis (Fig. 1).

As participating women follow an intensive regular screening program, an event may be detected during such an appointment, which is usually combined with a study visit. In those cases, sampling at the time of an event is guaranteed. However, for interval cancers, chances are that liquid biopsy collection is missed, because neither the treating physician nor the study participant is sufficiently aware of the importance to inform the study team: the physician because he/she might not be informed about study participation and the participant may neglect to inform the study team about her diagnosis, because of her overwhelming current situation. An active, regular search by the study team to rule out or confirm new breast-related pathology is not efficient, is labor-intensive, and has a high chance of still leading to discovering the events too late (i.e., after they started treatment). A practical, automated solution is to generate a “study alert notification” in electronic platforms such as the electronic health records or in the nationwide online registry of pathology reports (in the Netherlands, this has the acronym PALGA; ref. 41), so that new events will be reported to the study team before the start of treatment. Such interconnection between the cohort and tissue repositories and cancer registries is also essential to ensure a complete follow-up, provided that this interconnection is included in the informed consent form. Probably the most practical approach is instructing study participants to inform the research team every time they have additional hospital visits for breast-related issues. Therefore, participants must be well-informed, well-instructed, and perhaps, regularly reminded that when they develop an event they should contact the study team for final sampling. Still, understandably, this is a delicate time for a woman to consider an additional hospital visit, especially in the context of a clinical study.

Finally, there is a chance that women develop an event after having completed the study period. According to the Medical Research Ethics Committee guidelines, the protocol of a clinical study is required to have a specified delimited study period. Absence of such a limit could lead to endless clinical studies with the threat of losing aim and perspective. Still, we recommend to add a prolonged time margin to the study protocol, which would allow women to participate for a longer period and to have an additional sample taken at the time of an event, provided that this sample is acquired within a reasonable amount of time after the end of the study period.

By combining the abovementioned approaches, chances of acquiring a sample at the time of an event are increased, leading to a more complete intrasubject sample series in the biobank and, thereby, achieving the main aim of the study.

Lesson 8. Biobanking

As a consequence of the high number of samples in these specific studies, storage room might become an issue. As explained in lesson 5, around one to 22 aliquots per study visit per study subject were acquired throughout the TESTBREAST and the NAF studies, respectively. To give an impression, in the NAF study, more than 15,000 vials were stored in 11 years' time for 555 study subjects. In the TESTBREAST study, more than 3,000 samples were acquired between 2011 and 2019 from more than 930 participants (some blood sample series for the same women surpass the 20 visits). Therefore, it is essential to timely ensure substantial biobank storage, which will enable holding the samples safely for many years. In parallel, a well-defined biobanking system that allows for continuous sample registration in a database and consensual labeling throughout the study should be operational. Labels should be cold and moisture resistant, printed, and include standardized information, ideally the name of the team member who acquired and handled the sample, study subject number, visit number, sample type, volume, aliquot number, sample date, and a QR code. Specifically for the NAF study, breast side of the NAF sample should be added to the label.

Lesson 9. Sample Handling at the Time of Analysis

As highlighted in lesson 1, only a relatively small part of the cohort is expected to develop an event. Depending on the ratio of sample volume attained and technical volume required, samples could perhaps only be used once. For instance, in the NAF study, given the small volume acquired (10–50 μL), the majority or the complete volume may be needed for a single experiment. Thus, application of a trustworthy, familiar technique that has already been optimized to analyze the presence of biomarkers provides reassurance.

The exploration of improved, cheaper, new, faster, and potentially better techniques developed during the course of the study should be tested. This is inherent to technological developments and cannot be anticipated upon. Therefore, sufficient sample volume has to be acquired to allow for such technical exploration and comparisons. We recommend considering including technical testing in the list of secondary aims of the study protocol. Obtaining more volume is understandably less of a limitation for blood samples than for the low-volume NAF samples. As highlighted in lesson 8, a downside of obtaining higher volumes in general is that even more samples will have to be stored in the biobank, which demands for more storage.

Lesson 10. Adjustment of the Biomarker of Choice

The biomarker of choice to investigate in the high-risk cohort often derives from case–control studies. Specifically for the TESTBREAST study, the focus of research is based on previous promising results in proteomic expression profiling, showing very high sensitivity and specificity for the detection of breast cancer (42–44). However, preliminary case–control studies may also lead to alteration of the initially chosen biomarker due to insufficient diagnostic accuracy in interim analyses or in other studies. For instance, in the NAF study, gene methylation was the subject of investigation at first, but when interim analyses did not reveal a sufficient AUC to justify further translation into a clinical test (45), the biomarker of choice changed to miRNA. Furthermore, alterations may occur due to the discovery of new classes of biomarkers. Therefore, it is important to create the ideal conditions to be able to switch to another biomarker during the study by, among others, obtaining broad informed consent. Nevertheless, it is difficult to define what the ideal conditions are to be able to switch to new biomarkers, since this anticipates future developments. The use of buffers that allow different types of isolations, of, for example, DNA, RNA, and proteins, is advisable. Storing sufficient amounts of biofluids in different aliquots allows even more types of analyses than initially anticipated for the project. Moreover, storing different blood (half) products, such as whole blood, plasma, and serum is recommended, but, as a downside, leads to increased preprocessing costs. When switching to another biomarker or technical platform, one has to be aware that new markers and platforms will have to be technically validated, and the validity of the “old” data has to be established and might even have to be discarded.

Lesson 11. Nested Case–Control Analysis

There are a few issues that hamper the nested case–control analysis. First, the envisioned time frames of sampling (Fig. 1) are seldom exactly reached. The main reasons include delayed screening appointments at the hospital, which are due to hospital logistics, a woman's preference, pregnancy (a contraindication for imaging), or lactation (leads to reduced imaging sensitivity; ref. 12). Because study visits are preferentially combined with hospital screening visits, a change of the latter affects the timepoints of the study. Consequently, in practice, there will be variation in intervals of sample acquisition within the same subject and between study subjects. Second, another naturally occurring issue is incompleteness of sample series. This is because of missed study visits or unsuccessful sample collection. The statistical approach that allows data analysis with individual repeated measurements in case of missing data points and flexible time schedules, includes the use of linear mixed models (46). A solution could be to cluster data sampled at several timepoints into categorized subgroups (e.g., <6, 6–12, and >12 months before the event, as clustered by Weber and colleagues; ref. 47). Still, to optimally allow paired comparison analysis, it is of importance to continuously keep a real-time overview of the number of events and also to keep the number of successfully acquired samples and timing of sample acquisition up to date. If incomplete series are timely noticed, adjusting the sample sizes accordingly is still possible.

Authors' Disclosures

No disclosures were reported.