We used a qualitative descriptive approach with directed content analysis to structure the study. The goal of qualitative descriptive studies is to describe the experiences of those participating in the health-related behavior or process under consideration [20]. When used for health services research, this approach can identify and differentiate common and unique factors experienced by patients and clinicians when engaging with health services.

The TDF is a widely used behavior change framework, consisting of 14 domains that are used in the identification of determinants of change [21]. TDF categories are then mapped to the Behavior Change Wheel (BCW) components of Capability, Opportunity, and Motivation. The BCW is an evidence-based framework for understanding drivers of behavior and translating that understanding into actionable change [22]. Both TDF and BCW were chosen for this analysis over other frameworks, such as the Consolidated Framework for Implementation Research. The TDF can be more useful for the design of qualitative research interview guides by specifying discrete domains, which are applicable at the individual level as well as at the systems level. Our focus was predominantly on the behaviors and experiences of the patients and providers, which is best suited to TDF and BCW over the more process-oriented Consolidated Framework for Implementation Research.

The research team comprised 4 researchers: one senior qualitative PhD researcher who is a nurse (AS) and three MD researchers board-certified in internal medicine (KC, DB, and AB). Of the MD researchers, one was a practicing oncologist, and two were oncology fellows in the third year of their training. The nurse researcher had no oncology background and brought methodological expertise. All identify as having White European heritage.

Adult patients seen in oncology clinics between January and June 2022 at participating VA medical centers (VAMCs) in Pittsburgh and New York with a histologically confirmed prostate, gastrointestinal, or lung cancer were included in the study. We chose these 3 cancers as they together represent more than 60% of all patients with cancer within the VA, and PO is well established in all 3 cancers. We chose common cancers to understand practice patterns in patients with relatively common oncologic diseases. Patients with other cancers were excluded. The US Department of Veterans Affairs includes both VAMCs with inpatient and outpatient care as well as VA Community–Based Outpatient Care Centers with outpatient care only. Our study was conducted only at VAMCs. The VA is a single-payer health care system that serves 9 million American veterans and veteran family members.

Prospective patient participants were recruited by either letter or flyer. A letter was mailed to patients at the Veterans Affairs–New York Harbor Health Care System (VA-NYHHS), whereas a flyer was used at the Pittsburgh VAMC. At the Pittsburgh VAMC, flyers were placed in highly visible locations in the oncology clinics, such as the front desk at registration and in patient examination rooms. Providers were encouraged to inform patients that the study was available. The recruitment of provider participants was email-based. Emails were sent to oncology physicians at the VA-NYHHS, Greater Los Angeles Area (GLA), Pittsburgh, Michigan, and Seattle VAMCs. Informed consent was obtained from all participants, with all study methods approved by the local VA institutional review board (IRB). Patients were eligible for this study if they were receiving medical care at one of the participating VAMCs and were receiving treatment for prostate, lung, or gastrointestinal cancer. Only English-speaking patients were eligible, as US military enlistment requires English proficiency. Providers were eligible if they provided care for patients with cancer in one of the participating VAMCs.

The semistructured interview guide was constructed through an expert review process by a multidisciplinary team of oncologists, qualitative health researchers, and patient representatives. All 14 TDF domains were represented by at least 1 question in the interview tool. Three patient and three provider interviews, not included in the sample population, were used to refine the interview guide and construct a draft code list. This was done to provide test samples for assessing the feasibility of the interview guide.

Two interviewers then conducted semistructured interviews with patients and providers that lasted between 30 and 60 minutes. All interviews were conducted via audio using patients’ telephone numbers through Microsoft Teams. The interviews were transcribed using Microsoft Teams’ automated transcription feature, with verification by study personnel. Interviews continued until data saturation was achieved.

Directed content analysis using the TDF categories guided the analytical process. Directed content analysis allows for codes dictated by a preanalysis identified list, as well as purely descriptive codes, to emerge during the process [23].

AS provided coding training to the physician team and reviewed findings as they emerged. A team-based coding approach structured the consensus-based process [24]. Based on guidance from Cascio et al [24], we drew from principles of team-based coding to generate intercoder consensus throughout the analytic process. The consensus process was documented throughout the study to record the process of data review, idea generation, and analytical technique. The consensus process was conducted in person through meetings with the entire coding group.

Atlas TI.web (version 9.24.3; Lumivero) was used to code interviews. Each interview, whether from a patient or clinician, was coded independently by 3 research staff members (DB, KC, and AB). Initial codes came from the theoretical framework to start the process, and others emerged iteratively during the coding process. All team members met after 3 interviews were coded to harmonize code names and conceptual definitions. The codes were then compared and reviewed to achieve consensus on any coded data about which study staff disagreed. Disagreements regarding code application were discussed among the study team until a consensus was reached. A common codebook was then generated, and the team members checked in periodically to continue coding harmonization throughout the process.

The team then met to finalize categories and themes that emerged independently from the data as well as to evaluate how coding aligned with the TDF theoretical framework. Similar code groups were aggregated and mapped to related BCW sources of behavior (Capability, Opportunity, and Motivation) based on the consensus of the study team.

A determination of minimal risk and approval was granted by the VA-NYHHS IRB for our protocol (1605219‐1). We received subsequent approval from the Research and Development Committee and the Research Safety and Security Subcommittee. Consent forms were reviewed and approved by each committee and subcommittee. Informed consent was obtained from each participant in accordance with IRB-approved documents. Privacy and confidentiality were protected in accordance with data security specifications outlined in the IRB-approved protocol and were further reviewed by the institutional data security officer. Participant compensation consisted of US $25 gift cards for each participant.

At the VA-NYHHS, letters were sent to 136 patients, of whom 10 (7.4%) enrolled. At the Pittsburgh VAMC, flyers were posted for patients attending the oncology clinic. Of these patients, 7 (1.8%) enrolled, from approximately 400 patients who passed through the clinic. A total of 17 patients and 16 providers (n=33) were interviewed.

Patient participants were predominantly aged 70 to 80 years, entirely male, and mostly White. Twenty-four providers received the email, of whom 16 (67%) participated. Of the 16 providers who were interviewed, 1 provider requested to reply to demographic questions by email but did not subsequently respond. Providers were mostly New York–based, predominantly aged 40 to 50 or 50 to 70 years, and predominantly White. Providers were mainly MDs. Patient and provider characteristics are further outlined in Tables 1 and 2 and include self-reported age, gender, race or ethnicity, as well as the state of the site from which the participant was recruited. For providers, additional information was collected regarding the number of years in position, type of professional degree, and leadership role status.

Our findings suggest that there is a cyclical feedback loop related to the delivery of PO. The feedback loop is driven by the opportunity for this kind of care, as dictated by the resources of the system where care is delivered. Satisfaction from the clinician and patient side is the result of technical process outcomes associated with the PO testing process. Motivation to engage with PO services is a product of clinician learning and success with the tool and the resulting success with patient treatments, which also generates happiness for the patient.

The following sections illustrate how the feedback loop occurs, with the TDF providing structure for the categories that shape it. Figure 1 illustrates the feedback loop of Capability, Opportunity, and Motivation on the part of both patients and providers, all of which contribute to patient and provider engagement leading to PO. Sections are organized by the framework’s components.

In the TDF, behavioral regulation is defined as the components aimed at managing or changing objectively observed or measured actions, while motivation aligned with the BCW is defined as brain processes that energize and direct behavior [21,22].

In our analysis, statements that referred to self-moderation of actions on the part of patients and providers fell into this category. Patients reflected on the importance of a sense of motivation and autonomy in navigating their care, while providers identified motivation and adaptability as components of effective action. Furthermore, positive feedback emerged as providers sensed that patients were motivated to pursue testing. As one clinician described:

This category captures the way in which respondents felt communication and coordination between various individuals affected PO and cancer care. Three key motifs emerged, centered on the intersections of multilevel aspects of communication.

Emotions and motivation statements reflected the interactions between how patients and clinicians highlighted the role of emotions in influencing PO practice. Unsurprisingly, most of the negative patient-reported emotions reflected fear of their cancer, its progression, and death. Providers cited concerns about the emotional and practical implications of detecting germline mutations.

Nonetheless, finding an actionable mutation was in itself a significant provider motivator:

Thus, the balance of emotions and motivations encapsulates the inherent rewards and challenges of the PO treatment process for both clinicians and patients.

Providers had clear goals for PO care that motivated them when working with patients throughout their cancer treatment. They formed into 2 categories focused on patient outcomes and quality of care. Both providers and patients shared a mutual outcome goal of an increase in the length of life or a cure to their disease, as well as their desire to maintain a good quality of life. As one provider stated: “The goals of patient care would be to help people be on the best therapy with the lowest side effects that keeps them alive the longest” (Provider 010). Many provider respondents expressed the importance of the highest quality care, including safety, personalized care, and the ability to treat actionable mutations. They viewed this as inherently intertwined with a good patient outcome.

Among both patient and provider participants, external distractors emerged as barriers to PO, with responses generally categorized as prioritization either within the clinic or outside the clinic.

For the second least-frequently encountered domain, the reported “skills” involved in PO include the abilities to obtain tumor tissue, select the appropriate kit, and properly complete the ordering process:

The majority of difficulties reported were in the latter category, with convoluted and unstandardized ordering processes sometimes creating difficulties with PO.

As in the Goals domain, providers and patients noted an anticipated increase in the length or quality of life as a perceived positive consequence of PO, and providers feared missing this opportunity.

Sources of motivation for patients and providers came from similar and distinct sources. This relationship was always dyadic in nature.

How providers were able to operationalize PO treatments was the result of internal processes. These could facilitate or inhibit timely care delivery.

To illustrate the intersections of our results, we offer a conceptual model in Figure 1. Under the auspices of our overarching theme of “The Precision Oncology Feedback Loop,” the conceptual model illustrates the contribution of each of the areas of the BCW to the patient and provider interaction that ultimately influences PO implementation. Capability, opportunity, and motivation all influence and variously reinforce or negate each other, contributing to the implementation of PO.

Our study aimed to characterize the barriers and facilitators to PO. We found significant determinants of PO implementation at the patient, provider, and systems levels. In our analysis, the TDF provides a useful framework to conceptualize PO service implementation and, combined with the BCW, informs the design of interventions to improve PO implementation.

Our findings associated with system and operational issues are consistent with other studies. Prior research reported substantial structural or logistical issues in PO implementation, including coordinating or awaiting sample collection [25]. Our results add to previous work by presenting a more detailed analysis of specific factors associated with these problems, including interdepartmental communication–related barriers and the obstruction of interdepartmental communication via EHR ordering processes. Our data on the domains of Cognitive and Interpersonal Skills reiterated these challenges in communication. Furthermore, our finding about the role that staffing plays in PO care delivery, with a specific nod to the added benefit of patient navigators to help streamline test ordering, sample procurement, and results communication from pathology to providers, is novel.

From the patients’ perspective, our findings are associated with the social determinants of health around patient financial, logistics, and education specific to the VA population. The existing literature reports concerns around insurance coverage, high copays, and overall costs associated with treatment, which contrasts with our findings because VA participants do not bear those financial burdens [26]. This discrepancy is likely related to the VA single-payer model, which minimizes patient costs and third-party payer interactions, thereby eliminating a barrier to PO. Notably, existing literature is concordant with our findings that logistical challenges outside of cost (eg, transportation) represent barriers [26,27].

Patient education was perceived as a facilitator in our population, but only some patients noted receiving materials from their physicians, with many others turning to external resources. Several prior studies note that education can facilitate precision medicine when available and appropriately disseminated because it promotes patient engagement with care [28]. Yet, information sources sought by patients outside of their clinical team also suggest there is a risk for patients around misinformation in understanding precision medicine [26]. Our findings are also consistent with previous work identifying limited patient knowledge as a barrier to PO clinical trial accrual [28].

Regarding provider Knowledge, previous survey research reported that approximately 1 in 5 providers felt “low confidence” in their knowledge of clinical genomics and believed the tests would be used for less than 10% of patients [29]. Additional studies report that a lack of provider knowledge in genomic medicine may lead to delayed use [30]. Our findings reflect similar insecurities in providers regarding PO knowledge. Our data, however, build on previous literature by noting enthusiasm for systems-level educational interventions (ie, tumor boards noted as “immensely helpful”) and increasing comfort in interpreting NGS results (ie, providers feel “better in terms of outlining what’s pathogenic and what’s not”). Our data suggest a role for educational initiatives at the provider and patient levels to mitigate perceived knowledge gaps. Taken together, our findings along with the existing literature suggest that a deliberately implemented educational initiative may help facilitate PO uptake while also ensuring that patients do not fall victim to misinformation.

We noted that findings regarding the domains of Memory and Attention have substantial interplay with the findings in physician Environment, insofar as both relate to the structural/logistical burden that physicians face when implementing PO. Our findings were concordant with previous work suggesting barriers to precision medicine in time investment by physicians reviewing records, coordinating or awaiting sample collection, determining patient eligibility, and resolving insurance-related issues [26].

With regard to the domains of Goals and Reinforcement, previous studies similarly found that patients sought the benefits of PO clinical trials over previous less-targeted approaches to cancer care [31]. Our results support similar findings not only among patients considering clinical trials but also among all patients considering PO as standard care.

The limitations of our study include the following: (1) a small sample size comprising entirely of men, (2) VA-only study sites, (3) a risk of social desirability bias on the part of participants, and (4) bias in interpretation. Regarding sample size, our number of 33 (17 patients and 16 providers), although a relatively small cohort, was determined using theoretical saturation in which recruitment stopped when few additional new themes emerged. This approach increases efficiency while covering the majority of relevant themes. Of note, this is the largest sample of both patients and providers in any published work on the topic.

Regarding VA-only study sites, our study was conducted within the VA and represents viewpoints from a single, uniform health system. As such, external factors including insurance coverage, access to care, and access to trials are different from those in the private health care sector. Nonetheless, the VA setting does allow for geographic diversity and frequently samples patients with limited economic means, a population with well-reported underrepresentation in research [32,33].

A further consequence of VA-only sites is that our sample includes only male patients. The VA patient population is approximately 90% male, but the number of female patients is growing [34]. They may have unique PO treatment needs; thus, their perceptions may differ from those of their male counterparts. While we considered oversampling female patients, the cancers they are treated for are different from those of male patients and may change the PO oncology experience. Future research should focus specifically on female veterans to understand their needs with regard to PO.

We also chose to focus on the cancers that are most frequently treated in VA oncology clinics and to incorporate PO, that is, prostate, lung, and colorectal cancers, which yielded a predominantly male patient base. We made these decisions to focus on high-volume cancers, which can illuminate systemic challenges in a way that infrequently treated cancers cannot.

Finally, we recognize the constraints of social desirability bias along with biases in the interpretation of results. To mitigate social desirability bias, we maintained participant anonymity and confidentiality and used verbal and nonverbal cues to increase patient comfort [32]. To mitigate bias in interpretation, all interviews were coded independently by 3 reviewers, and disagreements on coding were resolved by discussion until consensus.

In conclusion, our interviews with patients and providers added essential data to an understudied and quickly growing area of cancer care. Patients and providers reported barriers to Knowledge, Environment, Goals, and Attention. When mapped to the TDF or BCW, these identified barriers facilitate the design of a multilevel intervention for patients, providers, and systems to educate and improve the implementation of PO. Future research should engage with broader populations of specific cancers and also examine geographic variation more closely.