Providing guidance for genomics-based cancer treatment decisions: insights from stakeholder engagement for post-prostatectomy radiation therapy

Abstract

Background

Despite the emergence of genomics-based risk prediction tools in oncology, there is not yet an established framework for communication of test results to cancer patients to support shared decision-making. We report findings from a stakeholder engagement program that aimed to develop a framework for using Markov models with individualized model inputs, including genomics-based estimates of cancer recurrence probability, to generate personalized decision aids for prostate cancer patients faced with radiation therapy treatment decisions after prostatectomy.

Methods

We engaged a total of 22 stakeholders, including: prostate cancer patients, urological surgeons, radiation oncologists, genomic testing industry representatives, and biomedical informatics faculty. Slides were at each meeting to provide background information regarding the analytical framework. Participants were invited to provide feedback during the meeting, including revising the overall project aims. Stakeholder meeting content was reviewed and summarized by stakeholder group and by theme.

Results

The majority of stakeholder suggestions focused on aspects of decision aid design and formatting. Stakeholders were enthusiastic about the potential value of using decision analysis modeling with personalized model inputs for cancer recurrence risk, as well as competing risks from age and comorbidities, to generate a patient-centered tool to assist decision-making. Stakeholders did not view privacy considerations as a major barrier to the proposed decision aid program. A common theme was that decision aids should be portable across multiple platforms (electronic and paper), should allow for interaction by the user to adjust model inputs iteratively, and available to patients both before and during consult appointments. Emphasis was placed on the challenge of explaining the model’s composite result of quality-adjusted life years.

Conclusions

A range of stakeholders provided valuable insights regarding the design of a personalized decision aid program, based upon Markov modeling with individualized model inputs, to provide a patient-centered framework to support for genomic-based treatment decisions for cancer patients. The guidance provided by our stakeholders may be broadly applicable to the communication of genomic test results to patients in a patient-centered fashion that supports effective shared decision-making that represents a spectrum of personal factors such as age, medical comorbidities, and individual priorities and values.

Background

Despite exciting advances toward the promise of genomics-driven cancer treatment [1, 2], there is not yet an established framework for optimal communication of test results in a clinical setting in a way that cancer best informs decision-making [3]. Research on the implementation of gene expression profiling for cancer treatment decisions has identified variable levels of understanding among patients, with misperceptions of test validity [4], and concerns among oncologists regarding patients’ understanding of test results [5]. This suggests a need for decision aids to support communication of genomic expression profiling test results and informed decision-making. Optimal implementation of genomic risk stratification tools, which have the potential to better match patients with the right treatments for them, should be performed within a patient-centered context that meets the National Academy of Medicine (formerly Institute of Medicine) goals for patient-centeredness by considering “individual patient preferences, needs, and values, and ensuring that patient values guide all clinical decisions” [6, 7]. As an example, the current study involves the clinical situation of prostate cancer patients faced with decisions regarding postoperative radiation therapy (RT) decisions, where the consideration of genomic test-defined individualized cancer recurrence estimates may be implemented within a patient-centered shared decision-making model.

Many of the nearly half of patients with localized prostate cancer who undergo radical prostatectomy (RP) will either develop a biochemical recurrence, in the form of a rising prostate specific antigen (PSA) blood level, or will be considered to be at high risk of recurrence based on adverse pathological features [8,9,10]. These men will face treatment decisions regarding the use of adjuvant RT (ART) after RP and/or close biochemical observation with salvage RT (SRT) for PSA recurrence. ART has been shown to improved outcomes compared to observation for high-risk patients [11,12,13], but close observation with serial PSA testing and selective use of SRT is also an effective treatment strategy [14] that has the potential advantage of limits the total number of patients exposed to the risks of radiation-related toxicities [15]. Currently, there is a relative lack of comparative data to evaluate the relative benefits of ART versus SRT, and joint guidelines from the American Urological Association (AUA) and the American Society for Radiation Oncology (ASTRO) recommend that clinicians presents both ART and SRT as reasonable treatment options and counsel patients appropriately for shared decision-making [16].

Decisions regarding ART and SRT after RP are complex for patients and clinicians, since choices can be influenced by an individual’s recurrence risk, medical comorbidities and personal preferences and values. A recent Markov decision analysis model considering length and quality of life for ART versus SRT showed that the preferred treatment strategy for an individual patient is dependent upon the specific probabilities for cancer recurrence, as defined based on a genomic risk classifier assay, as well as death from causes other than prostate cancer [17]. Genomic predictors of prostate cancer outcomes are increasingly available for clinical use [18, 19], including the Decipher® genomic classifier (GC) test (GenomeDx Biosciences, San Diego, CA) that provides individual estimates of the risk of metastasis after RP and is intended to be used for decision-making regarding ART and SRT [20, 21]. Although the information provided by the GC test has been shown to influence the recommendations of clinicians for ART after prostatectomy [22, 23], GC test results can be complex and it is not yet clear how patients and their families will incorporate the genomic-based risk estimates into treatment decisions.

In the current report, we describe a stakeholder engagement program that we performed in order to develop a framework for individualized decision aids to assist in the shared decision making process for patients making ART and/or SRT decisions with individualized GC-based recurrence risk estimates. Stakeholder engagement is a critical process for the development and implementation of genomic tests for clinical decision-making, because various stakeholders may each have different perspectives on genomic information [24]. We evaluate findings from the stakeholder engagement process completed prior to development of a decision aid tool that would consider an individual patient’s age, genomic-based risk of cancer recurrence, and other inputs to provide personalized guidance for post-RP ART and SRT decisions. Although our long-term objective is to develop a personalized decision aid program based upon a decision analysis modeling approach that incorporates individualized model inputs, we engaged a diverse group of stakeholders at an early point in the conceptual development of the proposed decision intervention with the goal of influencing the direction and format of the decision aid program to create a practical, patient-centered tool.

Methods

Our stakeholder meeting interview team was comprised of two faculty investigators (TNS and JML) with prior research experience and training in stakeholder engagement. Stakeholders were identified and recruited through direct contact with the research team. Patient stakeholders were identified in a single radiation oncology clinic by their physician (TNS), while other stakeholder groups were identified through informal networking by email and at medical conferences. The team conducted a series of meetings with stakeholders to inform the process of developing a personalized decision aid program.

Each stakeholder meeting with investigators included only one class of stakeholders. For example, either prostate cancer patients or urologists, but not both, would be invited to each meeting. This approach was used in order to encourage participation and avoid imbalance in power or educational level between groups. Although this choice may have improved convenience and facilitated conversation, the lack of larger focus group may have limited the range of information obtained from the process. A prepared set of slides was used to kick-off each meeting and provide background information (see Additional file 1). Much of the background information was based upon recent research using Markov decision analysis modeling that demonstrated that treatment choices that maximized length and quality of life for individual patients are dependent upon GC-based recurrence risk and risk of mortality from causes other than prostate cancer [17]. Patient stakeholder meetings occurred with individual patients and the investigators, mostly to preserve patient privacy and to maximize convenience for participants by scheduling interviews on the same day as clinical appointments, but also to stay on topic and to avoid the risk of unfocused or tangential discussions that can be observed with larger focus groups [25].

Each participant was invited to provide feedback during the meeting, including opportunities to revise the overall project aims. No specific time limit was applied to the stakeholder meetings, but the discussions lasted 25–45 min based on the amount of feedback and discussion offered by participants. Semi-structured interview structure was followed with prompts from a standardized data collection form that provided interviewers potential questions. For example: “What are the most important issues that you would consider for this decision?”; “What evidence do you need to make a treatment decision?”; “What format would you prefer for information?”; “Where would you prefer to receive information to help with a treatment decision?”; and, “Who would you like to have with you during the process of receiving information?”. The interview also included open-ended questions such as: “How would you design this differently?”; “Do you have additional ideas regarding this topic?”; and “Do you have additional ideas regarding how you would feel about using genomic tests for decisions?”. No quantitative data were collected. Audio and/or video recording of the meeting was not performed. The interviewers transcribed content from the meetings immediately upon completion of the meetings. The study team informed participants regarding the intention to preserve anonymity and to avoid specific attribution of quotes to individual participants in order to facilitate candid discussion.

Stakeholder groups

There were a total of 5 stakeholder groups included in the study: patients with prostate cancer (5 participants), urological surgeons (5 participants), radiation oncologists (5 participants), genomic testing industry representatives (5 participants), and biomedical informatics faculty members (2 participants). The stakeholder participants were chosen for their experience within their respective group. The industry representatives were all employees of GenomeDx, the company that markets the Decipher GC test. The IRB approval of the stakeholder engagement protocol permitted participation of up to 15 prostate cancer patients, but this was limited to 5 patients after investigators noted that response content had been exhausted with no new themes identified after the initial 3 patient stakeholder sessions.

Meeting preparation

Prior to each meeting, a common slide deck was shared with participants (see Additional file 1). At the start of each meeting, the study team first presented the slides, allowing time for questions and discussion, to provide background information. The materials provided included an overview of stakeholder engagement, general clinical context regarding ART and SRT decision-making for prostate cancer, background and results for our group’s Markov modeling approach using individualized model inputs for ART versus SRT after RP [17], and a general proposed framework for a patient-centered decision aid based that incorporated individual model inputs to include at least GC test results and risk of death from causes other than prostate cancer. The materials were intended to provide the above information at a high school reading level. The investigators based the proposed decision aid heuristic model on the Markov modeling methods in order to provide an analytic framework that could synthesize a range of clinical outcomes into summary model outputs.

Analysis

Transcribed data was collected from each stakeholder meeting. In the event of the interview team members having different content transcribed, an inclusive approach was taken with inclusion of all transcribed content and statements. Stakeholder meeting content was reviewed and the content was summarized by stakeholder group according to themes. Since the semi-structured interviews permitted open-ended discussion, providing a broad range of content, our synthesis of content into themes focused on presenting the broadest spectrum of content possible. Quantitative analysis was not performed.

Results

A total of 22 stakeholders participated in the meetings. Among the 5 patient stakeholders, all participants held college degrees and 3 participants held graduate degrees, reflecting the patient population at our institution. Most physician stakeholders were male (9 of 10), with one female urologist included. Stakeholders from the genomic testing industry were all employees of GenomeDx.

Stakeholders offered comments along a range of topics and perspectives. Stakeholder content was categorized according to 4 subjects: empathy, privacy concerns, design and formatting, and location or context of decision aid delivery to patient (Table 1). The majority of suggestions from all stakeholder groups focused on aspects of decision aid design and formatting, as well as delivery location/context and access to the decision aid, with the least time in discussion focusing privacy concerns. In general, stakeholders indicated that privacy considerations did not present significant barriers to the proposed decision aid framework, partly due to an understood distinction between tumor genomics versus an individual patient’s germline genetic information. Regarding location and context of decision aid delivery, a common theme was that stakeholders felt information should be disseminated across multiple platforms and should be available both before and during consult appointments. Regarding design, stakeholders consistently emphasized the need for simple graphs and figures and to remove any jargon and acronyms that might confuse patients. Stakeholders consistently recommended that only individualized data should be featured in the primary report, with the goal of showing patients and providers personalized information without the added complexity of a side-by-side comparison with average numbers. Emphasis was placed on the need to address the challenge of explaining quality-adjusted life years (QALYs) – a numerical measure of disease burden that incorporates length and quality of life – in decision aid materials, since the analytic framework of the personalized reports is based upon a Markov modeling approach that using individualized inputs [17]. It was recommended that the decision aid format be interactive, with patients being able to iteratively adjust model inputs to see the impact on model outcomes.

Table 1 Summary of content from stakeholder meetings regarding development of personalized decision aids to guide genomics-informed treatment decisions for radiation therapy after prostatectomy

Stakeholders were enthusiastic about the potential value of using decision analysis modeling with personalized model inputs for risk of cancer recurrence and other patient-specific factors such as age, other medical problems and lifestyle. Table 2 displays key decision aid elements that were identified based on stakeholder engagement that will be used to design a more effective patient-centered tool.

Table 2 Key elements and themes identified during stakeholder engagement meetings that will inform the development of a personalized decision aid program. QALYs = quality-adjusted life years

Discussion

We engaged a total of 22 stakeholders, including patients, physicians, biomedical informatics faculty members, and genomic testing industry representatives, to gain perspectives and insights regarding our stated goal of developing a framework to deliver personalized decision aids to assist in the shared decision making process for patients making ART and/or SRT decisions with individualized GC-based recurrence risk estimates. Stakeholders were enthusiastic about the general concept of personalized decision aids based upon a heuristic model of Markov decision analysis modeling using several individualized model inputs for event probabilities. Stakeholders provided valuable insights regarding optimal presentation of model results for the decision aid, which will directly influence the design of the decision aid. Stakeholder insights, summarized in Tables 1 and 2, provided valuable guidance regarding decision aid content and design as well as delivery format. Based on stakeholder input, our group plans to move forward with development of a personalized decision aid program that aims to support a patient-centered approach to communicating genomic classifier test results and guiding shared decision-making for cancer treatment decisions.

Physicians recognize genomic testing as a valuable tool for clinical care [26, 27], but substantial challenges remain regarding effectively communicating test results to patients and making clinical decisions that represent both genomic information and patient preferences for treatment [5, 27, 28]. Patients value genomic testing results [29] but, understandably, report substantial confusion regarding the value of gene expression profiling for cancer treatment decisions [4], suggesting that decision aids may be of value to better communicate GC results and help inform appropriate decision-making. Our stakeholder engagement study provides valuable insights into how to effectively communicate GC results in a patient-centered approach [7]. Stakeholders provided guidance on important features of a personalized decision aid program, with recommendations to: include a range of individualized inputs in the Markov model, present results in simple figure forms, allow for an interactive model design, keep genomic data anonymous, streamline results as much as possible, and make the decision aid available on mobile devices and in clinical situations. Integration of genomic information into the electronic health record for clinical use raises myriad challenges and opportunities, so stakeholder input is important when planning successful integration with electronic medical records in the clinical setting [30].

Although many of the insights from stakeholders shown in Table 1 may be generalizable to other clinical situations where genomic test results provide personalized estimates of cancer recurrence, the relevance of some stakeholder comments may be limited to proposed decision aid programs that employ an analytic approach based upon decision analysis modeling like the one developed by our group. It is important to note that all members of our patient stakeholder group reported at least a college degree, and most members had a graduate degree, so our study may fail to fully reflect the broader population. In particular, our finding that there were not privacy concerns among our stakeholders likely reflects the participants’ ability to readily distinguish genomic testing of tumor from genetic testing. It is likely that many members of the public would not have such clear understanding of tumor genomic tests versus genetic tests of disease risks, and that privacy concerns and fear of discrimination based on genetic testing likely remain of significant concern for other patients. The current study is also limited by the lack of inclusion of family members or primary care providers as stakeholders, since both groups have been shown to influence the treatment preferences of prostate cancer patients [31]. Furthermore, we did not engage representatives from insurance companies, so our study does not include the perspective of primary payers. Our findings do not address an emerging genomic test that provides direct predictions regarding response to postoperative radiation therapy [32], a development that could provide more clinically applicable information and reduce the degree of need for a personalized decision aid program. The relatively small sample sizes of stakeholders in the current study presents a substantial limitation of the current study, since the selective sample of participants limits the generalizability of our findings to other populations or clinical situations. Furthermore, the decision to interview some stakeholders, such as patients, individually and to interview other stakeholders, such as physicians and industry representatives, in a group setting may introduce problems in the comparability of findings across stakeholder groups. However, intergroup comparisons were not a major focus of the current study.

Conclusions

In conclusion, our stakeholder engagement process included perspectives from a range of stakeholders and provided valuable information that will inform the design of a personalized decision aid program for genomic-based treatment decisions regarding ART and SRT after prostatectomy. The guidance provided by our stakeholders may be broadly applicable to the communication of genomic test results to patients in a patient-centered fashion that supports effective shared decision-making.