FormalPara Key Points for Decision-Makers

This is the first patient-preference study conducted in patients with la/mUC to use an online preference survey and MDT exercise to elicit the effect of changes in treatment attributes on patient preferences for cancer-directed treatment.

Estimated marginal utilities suggested that, of the treatment attributes tested, unit changes in ORR were most important to patients, followed by cancer-related pain and treatment-related risks. Estimated acceptable tradeoffs indicated that patients would accept an 8.4% decrease in ORR to reduce their pain level by 10 points or a 7.8% decrease in ORR to reduce the risk of peripheral neuropathy by 10%.

Patient preferences were heterogeneous and dependent on individual valuation of different treatment aspects. These data underscore the importance of physician–patient dialogue in selecting treatments that best align with each patient’s treatment preferences.

1 Introduction

Urothelial carcinoma (UC) is a cancer that begins in urothelial cells lining the urethra, bladder, ureters, renal pelvis, and some other organs, and is the most common type of bladder cancer in the USA and Western Europe [1, 2]. The most advanced form of UC is locally advanced or metastatic urothelial carcinoma (la/mUC), which is an aggressive disease with a 5-year survival rate of 8.3% in the USA [3]. Patients with la/mUC frequently experience disease- and treatment-related symptoms, including fatigue, hematuria, and other urinary symptoms. In addition, patients experience substantial negative impacts on health-related quality of life, with significantly worse physical and mental health observed than in matched noncancer controls [4, 5]. Moreover, patients with la/mUC report significantly greater opioid use than matched noncancer controls, suggesting that the former may also experience significant cancer-related pain, which can interfere with their day-to-day activities and reduce their quality of life [6, 7].

Cisplatin-based chemotherapy has been the standard-of-care first-line therapy for patients with la/mUC; however, up to 50% of patients are considered cisplatin ineligible due to reasons including renal impairment, hearing impairment, neuropathy, congestive heart failure, and poor performance status [8, 9]. For patients who are ineligible for cisplatin-based treatment, treatment guidelines recommend carboplatin-based therapy [2, 10]. However, the efficacy of both treatments is limited, with a recent network meta-analysis reporting that the overall response rate (ORR) was < 50% across standard-of-care treatments for patients with la/mUC [11]. Additionally, cisplatin and carboplatin are associated with side effects, including potentially severe, treatment-limiting adverse events, such as nausea, neutropenia, and nephrotoxicity [12,13,14,15,16].

Novel therapies for patients with la/mUC are under development with side-effect profiles that differ from those of traditional chemotherapies [17, 18]. These include checkpoint inhibitors in combination with platinum-based chemotherapy, and enfortumab vedotin plus pembrolizumab. Enfortumab vedotin plus pembrolizumab was recently shown to result in greater patient overall survival (OS) and progression-free survival (PFS) than standard-of-care cisplatin- or carboplatin-based chemotherapy in the ongoing EV-302 trial [19]. As more treatments become available, the task of choosing the most appropriate treatment for each patient is becoming more complex. Differences in each treatment’s attributes necessitate increasingly careful consideration to determine which treatment is best suited for an individual’s treatment goals and expectations. In addition, other factors including an individual’s physical status and preexisting conditions need to be considered. Consequently, a more patient-centered care approach has been increasingly utilized in recent years [20, 21]. This approach emphasizes active collaboration and shared decision-making between patients, their caregivers, and treating clinicians rather than focusing solely on clinical outcomes when making treatment decisions. The patient-centered care approach can be employed to help understand the benefit–risk tradeoffs that patients are willing to accept and how these decisions vary across patient populations [22, 23]. Preference information can help inform future patient-centered treatment strategies and is also increasingly used for patient-focused drug development and evaluation [24, 25]. Consequently, the United States Food and Drug Administration (FDA) recently published guidance recommending patient-preference research to identify what is most important to patients regarding their experiences of disease burden and treatment [26].

To facilitate the use of a patient-centered care approach in patients with la/mUC, researchers are increasingly investigating patient preferences around la/mUC treatment attributes. For instance, Apolo et al. reported that treatment response was most important to patients, whereas Grivas et al. reported that patients prioritized a better treatment experience, including reduced incidence of adverse events [27, 28]. However, more information is needed regarding how patients value attributes associated with first-line treatments. These attributes may include improvements in clinical outcomes (ORR, PFS, and OS), effect on cancer-related pain, and treatment-related toxicities (neuropathy, diarrhea, skin reactions, nausea and vomiting, etc.). We performed a cross-sectional, quantitative preference study to better understand the perspectives of patients with la/mUC, the treatment attributes they consider important when making treatment decisions, how patients value these factors, and the extent to which they will tradeoff between these attributes.

2 Methods

2.1 Study Design

The preference survey design was informed by a targeted literature review of published evidence and qualitative interviews with physicians, patients with la/mUC, and their caregivers (n = 10 per respondent type), which included concept elicitation, sample choice tasks, and cognitive debriefing [25, 29]. The results of the qualitative interviews with patients informed the selection of six treatment attributes for inclusion in the preference elicitation instrument. Level ranges for each treatment attribute included in the multidimensional thresholding exercise (MDT) were derived from the evidence review. Included were two benefits (ORR and cancer-related pain rated on a scale of 0 [none] to 100 [worst pain possible]) and four treatment-related risks (peripheral neuropathy, severe side effects, mild to moderate nausea, and mild to moderate skin reactions; Supplementary Table 1).

The online preference survey consisted of three parts: part 1 asked patients about their experience with la/mUC, including the time since their diagnosis, symptoms, and treatment experiences (including previous treatments); part 2 was a MDT to assess patients’ willingness to make tradeoffs between different treatment attributes and to determine key drivers of treatment valuation [30]; and part 3 captured sociodemographic information, health literacy, and numeracy.

The survey was pretested with 10 patients in one-on-one cognitive interviews to assess survey usability (i.e., whether the treatment attributes and descriptions were readily understood by patients and if selected treatment attributes were relevant to treatment decision-making). Patients who participated in the pretesting interviews were not eligible to participate in the final MDT survey. Updates to the survey and preference instrument following pretesting are detailed in the Supplementary Materials. The treatment attributes and level ranges used in the main survey are presented in Table 1.

Table 1 Treatment attributes and level ranges used in the main multidimensional thresholding survey (patient description)
Full size table

2.2 Patient Recruitment

USA-based patients aged ≥ 18 years with a physician-confirmed diagnosis of la/mUC were eligible to participate in the study. All included patients were identified and informed about the study by participating USA-based oncology physicians. In total, 100 oncology physicians were contacted, and of these, 5 agreed to participate and informed eligible patients about the study. The target sample size was 100 patients, as this was assessed to be the largest feasible sample for this patient population. Eligible patients who were interested in participating were required to independently contact the study’s recruitment partner and provide informed consent. Patients also requested that their treating physician complete an electronic clinical report form (eCRF). The eCRF confirmed the patients’ diagnosis and provided details on their cancer stage, previous treatments received, and their eligibility for cisplatin-based treatment. Once patient eligibility was confirmed, patients were sent a link to complete the preference survey.

2.3 MDT Methodology

The MDT was implemented in a two-step process. In the first step, patients completed a stepwise ranking exercise where they were asked which attribute they would choose to improve first (Supplementary Fig. 1). Improvements were described by changes from the least desirable level to the most desirable level for each treatment attribute. For instance, the possible improvement in ORR was described as an increase from an ORR of 50% to an ORR of 80%. Ranking started with the full list of attributes. Each time an attribute was selected as the most important attribute to improve, it was removed from the list, and the question repeated until a full ranking was identified based on the order in which attributes were selected (see Supplementary Materials).

In the second step, the results of the ranking were used to generate five thresholding exercises (TEs) consisting of ranked pairs of attributes, each eliciting a tradeoff between two attributes. Each TE consisted of three choices between two hypothetical treatments: Treatment A and Treatment B. Treatment A was always described using the least desirable level of the higher-ranked attribute and the most desirable level of the lower-ranked attribute. Treatment B was always described using the least desirable level of the lower ranked attribute (Supplementary Fig. 2). The levels of the higher-ranked attribute in Treatment B were selected from a structured design that used a thresholding value tree (Supplementary Fig. 3). Each value tree included the initial value for Treatment B and the levels to be used for each subsequent choice task in the TE, which were varied adaptively on the basis of patient choice in each task. The variation in levels was structured to identify the improvement in a higher-ranked attribute that was considered equally desirable as the largest-possible improvement in a lower-ranked attribute, thereby identifying the interval in which this indifference point was likely to lie (e.g., reducing mild to moderate nausea from 80% to 20%). For instance, if a patient ranked improvements in pain as more important than improvements in ORR, the TE identified how much a treatment needed to reduce the level of pain from a score of 50 for the patient to be willing to accept a reduction in ORR from 80% to 50%. Thus, each TE elicited maximum acceptable tradeoffs between two attributes given their importance ranking.

2.4 Statistical Analysis

Descriptive analyses were used to summarize physician-reported patient characteristics from the eCRF and self-reported items in the quantitative survey. Sample size number (n) and percentages (%) were used for categorical variables and mean and standard deviation for continuous variables.

2.5 MDT Analysis

The analysis of the MDT data followed deterministic multi-attribute choice theory. The combination of the information from the ranking and thresholding exercises allowed for the estimation of a continuous and rank-order preserving multi-attribute utility function \({U}_{j}\) for every individual in the sample. The overall utility function was assumed to be linearly additive and was generally defined as:

$${U}^{j}={\sum }_{k\in \left[1;K\right]}{\omega }_{k}u\left({x}_{k}^{j}\right)$$

where \({U}^{j\in [1;J]}\) is the utility of alternative \(j\), \(u\left({x}_{\text{k}}^{j}\right)\) are partial utility values associated with the performance of the treatment on attribute \(k\in \left[1;K\right]\), and \({\omega }_{k}\) are preference weights defined as:

$${\omega }_{k}=\left\{{\omega }_{k}\in {\mathbb{R}}^{K};{\omega }_{k}>0;{\sum }_{k\in [1;K]}{\omega }_{k}=1\right\}$$

As with the overall utility function, the partial utility functions are also defined as linear and take the value 0 for the worst possible performance level and 1 for the best possible performance level.

Preference weights \({\omega }_{k\in [1;K]}\) were assumed to follow a Dirichlet distribution [41]. In the absence of further information, each weight vector that fulfills the condition \({\sum }_{k\in [1;K]}{\omega }_{k}=1\) was considered as equally likely. The individual-level ranking data for each respondent were used to impose constraints on the domain of the Dirichlet distribution, forming an individual-specific feasible weight space \({\Omega }_{n}\). The centroid \({\omega }_{n}^{c}\in {\Omega }_{n}\) of the feasible weight space was considered the expected vector of preference weights for individual \(n\). The centroid \({\omega }_{n}^{c}\) was estimated by making 10,000 draws from \({\Omega }_{n}\) using a hit-and-run simulation and calculating the arithmetic mean [42].

To account for sampling uncertainty, a Dirichlet distribution was fitted through the individual-level preferences resulting in sample-level average weights (\({\upomega }_{k}\)) [43, 44]. This was achieved by maximizing the log-likelihood function:

$$LL(\alpha )={\sum }_{j=1}^{N}\text{log}\left(\frac{1}{m}\stackrel{m}{\sum_{i=i}}\frac{1}{B(\alpha )}\stackrel{K}{\prod_{k=1}}{{\widetilde{w}}_{ij}}^{{\alpha }_{k}}\right)$$

Here, \(\alpha\) is a vector of dispersion parameters, from which the mean and variance of the preference weight vector can be obtained. The estimated concentration parameters (\({\alpha }_{k}\)) of the underlying Dirichlet distribution were then used to compute the average (\({\mu }_{{\omega }_{k}}\)) and variance (\({\sigma }_{{\omega }_{k}}^{2}\)) of the weights:

$${\mu }_{{\omega }_{k}}=\frac{\text{exp}\left({\alpha }_{k}\right)}{\sum\limits_{k}\text{exp}\left({\alpha }_{k}\right)}$$
$${\sigma }_{{\omega }_{k}}^{2}=\frac{{\mu }_{{\omega }_{k}}\left(1-{\mu }_{{\omega }_{k}}\right)}{1+\sum\limits_{k}\text{exp}\left({\alpha }_{k}\right)}$$

Both the individual and average preference weights are interpreted over the underlying value range of each attribute. To identify the impact of a unit change in the value of each attribute, the preference weights must be transformed into marginal utilities.

2.6 Marginal Utilities

Sample-level average weights obtained from the Dirichlet regression model were transformed into marginal utility estimates by dividing each attribute weight by the corresponding level range (where level range is equal to the difference between the best/most desirable level [\({B}_{k}\)] and worst/least desirable level \([{W}_{k}\)]). The marginal utility for each attribute (\({u}_{k}\)) indicates the effect of a 1-unit change in the attribute on the treatment utility. Because this consisted of a linear transformation of the attribute weights, the mean (\({\mu }_{{u}_{k}}\)) and variance (\({\sigma }_{{u}_{k}}^{2}\)) of the marginal utilities were computed as follows:

$${\mu }_{{u}_{k}}=\frac{{\mu }_{{\omega }_{k}}}{{B}_{k}-{W}_{k}}$$
$${\sigma }_{{u}_{k}}^{2}=\frac{{\sigma }_{{\omega }_{k}}^{2}}{{\left({B}_{k}-{W}_{k}\right)}^{2}}$$

2.7 Marginal Rates of Substitution

Marginal rates of substitution (MRS) were calculated using marginal utility estimates. For each attribute \(k\), MRS is calculated as the ratio of attribute \(k\) and a numéraire \(\uprho\):

$${\text{MRS}}_{k}=\frac{\partial \text{v}({\mathbf{x}}_{ntj})/\partial {x}_{k}}{\partial \text{v}({\mathbf{x}}_{ntj})/\partial\uprho }$$

Thus, \({\text{MRS}}_{k}\) is the change in attribute \(k\) that patients are willing to accept for a 1-unit decrease (increase) in \(\uprho\), or the amount of 1 treatment attribute that patients would require to accept a change in the numéraire. Alternatively, \({\text{MRS}}_{k}\) can be interpreted as the amount of attribute \(k\) that is equivalent to 1 unit of \(\uprho\). MRS were calculated using treatment benefit (ORR) as the numéraire.

The Delta method was used to obtain the standard error (SE) for marginal utilities and MRS estimates. [45] The 95% confidence interval (CI) estimate was based on the standard normal distribution (μ_θ ± 1.96 SE θ), where θ denotes the parameter/quantity of interest.

2.8 Subgroup Analyses

Subgroup analyses were used to explore the effects of patient sociodemographic and clinical characteristics (patient age, time since diagnosis, cancer type, current cancer stage, treatment experience, disease-symptom severity, cisplatin eligibility, current cancer treatment, cancer treatment goal, and other health conditions; Supplementary Table 2) on weights and marginal utilities in separate interacted Dirichlet regression models (i.e., one model for each characteristic of interest). The characteristics were included in the regression model as additional covariates (\({Z}_{nl}\)):

$${\alpha }_{k}={\beta }_{0k}+\sum_{l\in [2;L]}{\beta }_{lk}{Z}_{nl}$$

where \({Z}_{nl}\) is the lth level of the covariate for patients \(n\). Given that covariates entered the models as dummy-coded variables (0/1), the \({\beta }_{lk}\) parameters captured the effects of the different levels relative to the first, which served as reference category. The effect for the reference category was captured by the model constant (\({\beta }_{0k}\)).

Model estimates were used to compute the mean preference weights and marginal utilities for each subgroup. The mean estimates were then compared across subgroups with independent two-tailed z-tests.

All analyses were conducted in R software (version 4.0.5) using the hitandrun package for the hit-and-run simulation and the DirichletReg package for the Dirichlet regression [42, 46].

2.9 Data Quality

The quality of the MDT data was assessed on the basis of patient tradeoff behavior and survey completion time. Patient decision-making was considered “dominated” (i.e., driven by only one treatment attribute) if choices in all MDT tasks were recorded as A or B. Patients were classified as “low engagement” if they completed the full survey in < 5 min.

3 Results

3.1 Patient Sociodemographic and Clinical Characteristics

Of 120 patients invited to participate, 100 (83%) contacted their physician to request completion of the eCRF to confirm their eligibility. All 100 of these patients fully completed the preference instrument. Mean patient age was 64.9 years (standard deviation, 7.6), 54% were female, and 38% identified as white (Table 2). Additional patient demographic information is presented in Supplementary Table 3.

Table 2 Patient sociodemographic characteristics
Full size table

Overall, 62% of patients reported being diagnosed with la/mUC < 1 year prior to participating in the survey. When questioned about their treatment goals, 57% of patients indicated that their overall treatment goal was to extend their length of survival, and 43% indicated that their main treatment goal was to maximize their quality of life for their remaining time alive. Additional patient-reported clinical characteristics are presented in Supplementary Table 4.

All patient-choice data were considered high quality, as all patients (n = 100; 100%) demonstrated tradeoff behaviors by selecting each treatment alternative (A or B) in at least one choice task, and all patients took ≥ 15 min to complete the preference survey.

3.2 Marginal Utilities

Marginal utilities for all attributes were found to be statistically significant (P < 0.001), indicating that all included treatment attributes were preference relevant (Fig. 1).

Fig. 1
figure 1

Marginal utilities (a) and minimum acceptable reduction in overall response rate (marginal rates of substitution) that patients would accept for a decrease in other treatment attributes in the multidimensional thresholding analysis (b); analysis characteristics: participating patients, N = 100; number of study parameters, n = 6; log-likelihood, 681.9; Bayesian information criteria, 1336.1; Akaike information criterion, 1351.8. *The marginal utility for each attribute indicates the effect of a 1-unit change in that attribute on overall treatment utility, with a larger marginal utility indicating a greater impact on patient preferences and that a given attribute is more important to patients. †The minimum acceptable reduction in ORR was calculated using marginal utility values and assessed the scale of change in the numéraire (ORR) that patients would accept in exchange for a 10-unit/10% decrease in another treatment attribute. CI confidence interval, ORR overall response rate, SE standard error.

Full size image

On average, of the attributes assessed, unit changes in ORR had the greatest influence on patient preferences, with a mean marginal utility of 0.65 (95% CI 0.55–0.75), followed by cancer-related pain [0.55 (0.47–0.62)], risk of peripheral neuropathy [0.51 (0.43–0.58)], risk of severe side effects [0.36 (0.29–0.43)], risk of mild to moderate nausea [0.24 (0.20–0.28)], and risk of mild to moderate skin reactions [0.22 (0.18–0.26)].

3.3 Tradeoffs

On average, patients were willing to tradeoff an 8.4% decrease in ORR to reduce their pain level by 10 points (pain scored from 0 to 100) or a 7.8% decrease in ORR to reduce their risk of peripheral neuropathy by 10%. Additionally, for a 10% decrease in severe side effects, mild to moderate nausea, or skin reaction, patients would accept decreases in ORR of 5.5%, 3.7%, or 3.4%, respectively (Fig. 1). Additional data on the model outputs are presented in Supplementary Table 5.

3.4 Subgroup Analysis Results

Subgroup analyses found no statistically significant differences in patient-preference weights or marginal utilities due to patient age, time since diagnosis, cancer type, current cancer stage, treatment experience, cisplatin eligibility, current cancer treatment, cancer treatment goal, or other health conditions. The only statistically significant difference was seen among patients who considered the severity of their bladder cancer symptoms to be moderate at their worst. A pairwise comparison found that they placed significantly greater importance on ORR than patients who rated their symptoms as severe (P < 0.05) (Supplementary Table 6).

4 Discussion

Understanding the attributes that matter to patients with la/mUC is critical in selecting the best treatment for each individual patient, especially now that multiple treatment options are becoming available. Although improved clinical outcomes as measured by ORR, PFS, and ultimately OS are key features in selecting treatments, reducing cancer-related pain and minimizing treatment-related toxicities are also very important to patients and play an important role in their treatment preferences. We used a MDT exercise to elicit the effect of changes in treatment attributes on patient preferences for cancer-directed treatment in patients with la/mUC. MDT was selected over other preference methods, including discrete choice experiments (DCEs), because a large sample size is typically required to achieve robust preference estimates in a DCE, particularly where subgroup analyses are required. For example, a recent systematic review found a mean sample size of over 500 patients across included DCE studies. [47] Conversely, the Dirichlet regression approach used in this study to aggregate MDT response data can produce reliable preference estimates on the basis of a sample of 100 respondents, enabling the generation of stable preference estimates while eliciting individual-level patient preferences with a smaller sample size [41, 48]. This was an important consideration given the challenges associated with recruiting patients with late-stage cancer in non-clinical research. To identify patients with locally advanced or metastatic UC with standard-of-care first-line therapy treatment experience, physician-confirmed diagnosis and reported treatment experience were required to accurately identify this specific patient group [23].

The choices made by patients in the MDT exercise demonstrated that of the tested attributes, improving the ORR (i.e., reducing the tumor burden) was on average the biggest driver of treatment preferences, followed by decreasing cancer-related pain related to la/mUC. Changes to the risk of severe side effects, although important, were less influential than other tested attributes. Subgroup analyses did not detect any statistically significant differences between patient subgroups in patient-preference weights or marginal utilities, with one exception. The only significant difference was found between patients who self-reported their disease symptoms as moderate versus those who reported their symptoms as severe, which may indicate that the severity of a patient’s la/mUC—and therefore of their symptoms—may impact their treatment preferences. Although preferences were similar overall, patients reported different treatment goals, with nearly equal proportions of patients reporting maximizing their duration of survival and maximizing their quality of life as their main treatment goals, suggesting that there may be differences in underlying patient priorities when selecting treatments.

This study was novel in its use of ORR as the primary treatment benefit, and to our knowledge only one other preference study to date has used ORR as a measure of treatment benefit in la/mUC [35]. However, that study assessed treatment preferences among medical oncologists only and utilized a DCE methodology, whereas this study focused on patient preferences and used a MDT approach. ORR was selected over OS as a treatment attribute for this study, as drugs are more frequently receiving accelerated approval from the FDA on the basis of ORR alone, while larger confirmatory trials addressing changes in OS are pending. In these situations, without clear OS data, physicians, patients, and their caregivers may be left with only ORR and adverse event information to inform their decisions. The finding that changes in cancer-related pain had the next greatest impact on patient treatment preferences after ORR suggests that patients with la/mUC may benefit from a greater focus on pain management as part of their treatment plans. This is in keeping with previous qualitative studies that highlighted the importance of pain among patients with la/mUC and another retrospective study that found that this patient population requires opioid analgesia more frequently than matched controls, indicating that poor pain control is a frequent issue for patients with la/mUC [4,5,6, 29].

This study highlights that treatment decisions are highly personal and dependent on individual valuation of different treatment aspects. These data underscore the importance of physician–patient dialogue, as patients may benefit from in-depth communication and tailored discussions with treating clinicians to identify the treatment option that best aligns with patients’ preferences on clinical outcomes, side effects, health-related quality of life, and cancer-related pain control when making treatment decisions.

4.1 Limitations

Selection bias due to non-probabilistic physician recruitment may have limited the diversity of the sample; however, this recruitment approach meant that all included patients had a physician-confirmed diagnosis of la/mUC, whereas most preference studies rely on a self-confirmed diagnosis and patient-access panels for recruitment. The sample contained a broader representation of a range of racial groups and more patients with stage III disease than seen in a recent USA-based chart-review study, which reported a higher mean age and a greater proportion of male patients [49]. Therefore, despite attempts to mitigate this bias through the use of patient-selection guidelines and sample quotas, it cannot be guaranteed that participating patients had perspectives representative of the overall patient population.

Information bias due to recall, reporting, or observer bias during the qualitative interviews or quantitative survey may have occurred, despite mitigation approaches that included recruiting patients with recent experiences relevant to la/mUC and ensuring that patients were not asked questions with a long recall period (i.e., longer than 4 weeks).

Although data are presented in the context of first-line treatment preferences, it was not appropriate to ask patients to make choices as though they were selecting/considering their own first-line treatment. However, the majority of patients (62%) reported being diagnosed with la/mUC < 1 year prior to the survey, suggesting their response would be broadly reflective of first-line treatment choices. At the same time, patient preferences did not vary by time since diagnosis and thus were consistent with those of patients diagnosed < 1 year previously, supporting the application of all responses in the context of first-line treatment preferences.

5 Conclusions

These results demonstrate that treatment preferences for patients with la/mUC are dependent on individual valuation of different aspects of treatment, particularly attributes of clinical response, pain control, and treatment-related side effects.

For patients with la/mUC, changes in ORR were most important among the treatment attributes tested. Patients were willing to make tradeoffs between treatment attributes, indicating that a lower ORR may be acceptable in exchange for a reduction in cancer-related pain or decreased risk of treatment-related adverse events. These tradeoffs between possible improvements in treatment efficacy and either greater pain symptoms, or possible negative consequences of treatment, underscore the importance of physician–patient dialogue as part of treatment decisions to identify which treatment option best aligns with each patient’s treatment preferences.