The impact of adding cost information to a conversation aid to support shared decision making about low‐risk prostate cancer treatment: Results of a stepped‐wedge cluster randomised trial

Abstract Background Decision aids help patients consider the benefits and drawbacks of care options but rarely include cost information. We assessed the impact of a conversation‐based decision aid containing information about low‐risk prostate cancer management options and their relative costs. Methods We conducted a stepped‐wedge cluster randomised trial in outpatient urology practices within a US‐based academic medical center. We randomised five clinicians to four intervention sequences and enroled patients newly diagnosed with low‐risk prostate cancer. Primary patient‐reported outcomes collected postvisit included the frequency of cost conversations and referrals to address costs. Other patient‐reported outcomes included: decisional conflict postvisit and at 3 months, decision regret at 3 months, shared decision‐making postvisit, financial toxicity postvisit and at 3 months. Clinicians reported their attitudes about shared decision‐making pre‐ and poststudy, and the intervention's feasibility and acceptability. We used hierarchical regression analysis to assess patient outcomes. The clinician was included as a random effect; fixed effects included education, employment, telehealth versus in‐person visit, visit date, and enrolment period. Results Between April 2020 and March 2022, we screened 513 patients, contacted 217 eligible patients, and enroled 117/217 (54%) (51 in usual care, 66 in the intervention group). In adjusted analyses, the intervention was not associated with cost conversations (β = .82, p = .27), referrals to cost‐related resources (β = −0.36, p = .81), shared decision‐making (β = −0.79, p = .32), decisional conflict postvisit (β = −0.34, p= .70), or at follow‐up (β = −2.19, p = .16), decision regret at follow‐up (β = −9.76, p = .11), or financial toxicity postvisit (β = −1.32, p = .63) or at follow‐up (β = −2.41, p = .23). Most clinicians and patients had positive attitudes about the intervention and shared decision‐making. In exploratory unadjusted analyses, patients in the intervention group experienced more transient indecision (p < .02) suggesting increased deliberation between visit and follow‐up. Discussion Despite enthusiasm from clinicians, the intervention was not significantly associated with hypothesised outcomes, though we were unable to robustly test outcomes due to recruitment challenges. Recruitment at the start of the COVID‐19 pandemic impacted eligibility, sample size/power, study procedures, and increased telehealth visits and financial worry, independent of the intervention. Future work should explore ways to support shared decision‐making, cost conversations, and choice deliberation with a larger sample. Such work could involve additional members of the care team, and consider the detail, quality, and timing of addressing these issues. Patient or Public Contribution Patients and clinicians were engaged as stakeholder advisors meeting monthly throughout the duration of the project to advise on the study design, measures selected, data interpretation, and dissemination of study findings.

suggesting increased deliberation between visit and follow-up.
Discussion: Despite enthusiasm from clinicians, the intervention was not significantly associated with hypothesised outcomes, though we were unable to robustly test outcomes due to recruitment challenges. Recruitment at the start of the COVID-19 pandemic impacted eligibility, sample size/power, study procedures, and increased telehealth visits and financial worry, independent of the intervention.
Future work should explore ways to support shared decision-making, cost conversations, and choice deliberation with a larger sample. Such work could involve additional members of the care team, and consider the detail, quality, and timing of addressing these issues.
Patient or Public Contribution: Patients and clinicians were engaged as stakeholder advisors meeting monthly throughout the duration of the project to advise on the study design, measures selected, data interpretation, and dissemination of study findings.

K E Y W O R D S
conversation aid, costs of care, decision aid, financial toxicity, stepped-wedge cluster randomised design

| INTRODUCTION
Healthcare costs are rising substantially in the United States and internationally, [1][2][3][4][5][6][7] leading to psychological, social, behavioural, and health-related challenges for patients. The associated cost-related hardship, often called financial toxicity, can result in delayed or forgone care 8,9 and is even associated with an increased risk of mortality. 10,11 In addition to the direct costs of healthcare, patients experience indirect cost burdens such as lost wages from missing work for appointments or illness-related disability. Communicating the relative or specific costs of options is an important part of patient-centred decision-making. 12 Patients want to know about cost information 13,14 and clinicians acknowledge its importance and impact on patients' choices and adherence to care. 15 Yet clinician communication and patient decision aids rarely include costs of treatment options to support decisions. 16 Although financial toxicity affects patients with cancer worldwide, [1][2][3][4][5] it is particularly a problem for patients in the United States.
The United States has a healthcare system that requires patients to use insurance to share the cost of care, and each insurance option in the private sector can vary in terms of the amount of cost-sharing provided. 6,7 In addition, more than 30 million people in the United States are uninsured and struggle to identify ways to pay for care through hospital billing options, the government, or social service agencies. [17][18][19] As many as 18% of patients in the United States have medical debt as a result, 19 and it is the most common form of debt in the United States. 20 Even patients who have adequate insurance coverage for cancer treatment receive out-of-pocket bills for copayments, medications, and support services. They also have indirect costs of care from lost wages or time off work, payment for transportation to/from health visits, and disruption to their daily routine.
In the context of early-stage, favourable risk prostate cancer, 21 there are several reasonable treatment options for patients to consider including active surveillance, radiation (external beam or brachytherapy) and prostatectomy (typically robotic). Each choice is similarly effective in preventing prostate cancer-related mortality but carries different tradeoffs and costs, 22 especially to patients with varying insurance coverage. 23 For example, prostatectomy and radiation therapy cost more than active surveillance both in terms of out-of-pocket health expenses and downstream indirect costs from time off work and recovery. [22][23][24][25] However, some patients prefer to intervene with surgery or radiation rather than actively monitor a known cancer, even if it is low-risk or favourable risk, because they worry about the cancer growing or spreading. 26 Others might choose active surveillance even with the increased risk of repeat biopsies and imaging because they want to avoid the possible side effects of surgery or radiation such as incontinence and erectile dysfunction. 27,28 The complexity of trade-offs, preferences, and difficulty estimating costs to patients can complicate shared decisionmaking.
We previously developed a conversation-based decision aid (called an Option Grid [OG]) containing information about low-risk prostate cancer management options. OGs are brief tabular comparisons of options that activate patients before clinical visits and facilitate efficient conversations during visits. 29 They can increase shared decision-making across diverse contexts. 30-32 by promoting deliberation and dialogue, while providing evidence-based information. 33 We added a prompt to consider the relative costs of prostate cancer management options. In past work, this approach increased the frequency of cost conversations about early-stage breast cancer decisions. 34 No such intervention has been developed and evaluated for prostate cancer.
In this study, we aimed to assess the impact of a conversationbased decision aid (OG) containing cost information about low-risk prostate cancer management options, combined with a brief training session for urologic surgeons, on the frequency and quality of patient-urologic surgeon cost conversations. We hypothesised that: 1.1: Urologic surgeons assigned to training and use of the decision aid would engage in more frequent cost conversations than urologic surgeons in usual care.
1.2: Urologic surgeons assigned to training and use of the decision aid would be more likely to make a referral (e.g., to social service organisations, billing representatives, social workers or financial navigators) to address specific cost details than urologic surgeons in usual care. We also aimed to examine the impact of the conversation-based decision aid and surgeon training on decision quality, including measures of decisional conflict, decision regret, and shared decisionmaking. We hypothesised that: 2.1: Patients of urologic surgeons assigned to training and use of the decision aid would report less decisional conflict, less decisional regret at 3 months follow-up, and more shared decision-making than patients in usual care.

| METHODS
Detailed methods are described in a published protocol. 35 Reporting follows the 2018 CONSORT extension for stepped wedge randomised controlled trials. 36

| Design
We conducted a stepped-wedge cluster randomised controlled trial with urologic surgeons as clusters, four sequences, and at least one cluster assigned to each sequence ( Figure 1). A stepped-wedge design involves delivering an intervention at regular intervals, or steps, following a baseline period with no intervention. In this type of design, studies often need fewer clusters to achieve the same statistical power as a larger cluster randomised trial, and each can act as their own control due to the baseline period. 37 In addition, because of the learning effects of a clinician-focused intervention such as a conversation-based decision aid, the stepped-wedge design limits contamination in the control group. Five 3-month periods were planned; we extended the length of some periods after trial initiation due to few eligible patients during the onset of the COVID-19 pandemic, resulting in five periods of 5, 3, 5, 6, and 3 months, respectively. We enroled independent eligible patients in each period.

| Settings
We conducted the study in outpatient urology clinics affiliated with a large academic medical center in the Midwest region of the United States.

| Participants
We included five urologic surgeons who practiced at a participating clinic and routinely discussed management options for low-risk prostate cancer with patients.
We included English-speaking adult patients scheduled to visit a participating urologic surgeon to discuss a new diagnosis of low-risk prostate cancer. Eligible patients had at least one of the following: (1) a Gleason score of 6 or 7 (3 + 4); (2) a prostate-specific antigen level less than 10 ng/mL; (3) a surgeon's referral for study eligibility. We excluded patients unable to give informed consent due to cognitive or emotional barriers and those discussing recurrent or ongoing prostate cancer management. A research coordinator screened potentially eligible patients based on inclusion criteria and confirmed eligibility with the surgeon before arranging to contact the patient to gain consent to enrol in the study; surgeons could also refer patients directly to the study if they used the intervention with a patient during clinical care.

| Patient and stakeholder engagement
At the start of the study, a patient and stakeholder advisory board was formed consisting of a survivor of prostate cancer, a patient advocate and leader of a local prostate cancer advocacy and education group, a urologist, a community engagement leader at the cancer center and expert on prostate cancer disparities, and an oncologist with expertise in financial toxicity. This team met monthly throughout the duration of the project to advise on the study design, intervention adaptation to include cost-related information, outcome measures selected, social service organisations and personnel for cost-related referrals, data interpretation, and dissemination of study findings.

| Intervention and comparator
The intervention comprised an OG conversation-based decision aid comparing management options for low-risk prostate cancer, with relative cost information included for each option in addition to referral information for general and local resources for navigating care costs (Appendix A). OGs are brief, tabular comparisons of options written at an accessible reading level and organised by common patient questions. 29 OGs are used collaboratively by clinicians and patients to facilitate conversations and decision dialogue, while providing evidence-based information. 38 They are particularly useful in situations when patients might not have had time to prepare for a decision discussion, such as low-risk prostate cancer when patients often receive biopsy results from a clinician immediately before discussing management options. At the study start, we merged previously developed and tested OG information into a research version of an OG to compare active surveillance, surgical treatment and radiation treatment for low-risk prostate cancer. The cost information was added to the common questions about trade-offs between options and was generated from the literature available at the time of the study start. 24,25,39 To display comparative cost-related information, we included a visual icon to represent relative costs to patients ($-$$, $$-$$$, $$$) across options. We reviewed intervention adaptations with our stakeholder advisors, clinical partners, and study team members for clarity, feasibility, acceptability and appropriateness, and tracked adaptations systematically based on standards in implementation science. 40 Future work is ongoing to quantify more precise cost-level estimates across treatment options, but in this work, we encouraged clinicians and patients to weigh relative costs, and then refer patients to discuss more precise, personalised costs or resources with social workers, financial navigators, billing specialists or social service organisations. The study coordinator (K. P.), clinician, or administrative clinic personnel delivered the intervention to individual patients before or during a first discussion of management options following a new diagnosis of low-risk prostate cancer as defined above. After trial initiation, we adapted to allow clinicians who forgot to introduce the intervention before or during the initial patient visit to send the intervention to a patient postvisit, tracking adaptations using a standard framework. 40 The comparator was usual care.
Before study initiation, participating clinicians were trained in the study protocol. At the step initiating their entry into the intervention arm, each clinician attended a 30 min virtual training session in shared decision-making, use of the intervention, and cost-related resource and referral information.

| Primary outcomes
We measured patient reports of cost conversations and whether or not a referral was made to discuss costs in a questionnaire collected immediately postvisit (T1).
Soon after trial initiation and before any observational data were collected, we discontinued planned observational data collection using an observer-reported cost conversation checklist. 41 This change occurred because of the COVID-19 pandemic which prompted mostly telehealth visits at participating clinics during the early months of the study and compromised the feasibility of having study team members audio record clinic visits in person for any inperson consultations. We attempted to have clinicians record the consultations using the telehealth software, but the new and changing software, combined with the challenges and pressures of the COVID-19 pandemic, and patients' hesitation to be recorded in their own homes with family members often present (even if clinicians were only to save the audio recordings, the recordings had to be created with video to start if patients were having a consultation with video turned on) made this process cumbersome.
Instead, we decided to rely on the patient-reported questionnaire to simplify the process for all stakeholders.
At 3-month follow-up (T2), we measured decisional conflict (SURE 42 ), decision regret (decision regret scale 46 ), treatment choice received, and financial toxicity (COST 44,45 ) in a questionnaire distributed to participants by email or telephone.

| Exploratory implementation outcomes
Among clinician participants, we measured feasibility, appropriateness, and acceptability of sustained intervention use with the Feasibility of Implementation Measure, Appropriateness of Implementation Measure, and Acceptability of Implementation Measure. 47 These four-item validated measures use a five-point ordinal scale, ranging from 'completely disagree' (score = 1) to 'completely agree' (score = 5). Higher scores indicate greater feasibility, appropriateness, and acceptability. We also measured clinicians' attitudes toward shared decision-making and cost conversations using the ADOPT scale 48 which lists words to describe using an intervention (e.g., CostTalk) and asks people how they feel about it.
Among patient participants, we assessed preferences for having cost conversations, with whom they prefer having cost conversations using an adapted validated measure on a five-point ordinal scale. 14 We assessed their confidence having cost conversations using a fouritem measure on a four-point ordinal scale. 49 Among patient participants who received the CostTalk intervention, we measured acceptability of the intervention using the Acceptability of Implementation Measure. 47 These exploratory outcomes were added after study initiation.
Sample size. We assumed an intraclass correlation coefficient (ICC) value of 0.05 and a feasible total sample size of 200 patients across the five participating urologic surgeons. We estimated power directly for a stepped-wedge design by solving for its value given the specified ICC of 0.05 and the within-cluster variance of a Bernoulli (binary-valued) random variable (conservatively assumed to be its maximum value of 0.25. We based effect size assumptions on a prior study demonstrating a significant effect when comparing the impact of a decision aid with comparative cost information to a decision aid without cost information or usual care on cost conversation frequency (66.7% vs. 33.3%). 34 Using a two-sided test, the power to detect a difference in our primary outcome of cost conversation frequency was estimated to be 0.804.

| Randomisation
The principal investigator (M. P.) and study coordinator (K. P.) enroled clusters. The study statistician (A. J. O.), masked to cluster identity, generated the randomisation schedule and randomly allocated clusters (urologic surgeons) to the intervention sequences with a simple randomisation approach using R statistical software. Across the control and intervention arms, the study coordinator (K. P.) enroled consecutive eligible patients who provided informed consent to participate.

| Statistical analysis
We first performed unadjusted bivariate comparisons of outcomes and predictors across intervention groups using t-tests for continuous variables and χ 2 or Fisher's exact tests (where indicated due to small sample sizes) for categorical variables.
For primary and secondary binary-valued outcomes (e.g., cost conversations), we conducted logistic mixed-effectss regression analysis adjusting for patient educational attainment (less than college degree vs. college degree or more), patient employment status (full-time work vs. other), telehealth versus in-person visit, visit date, and the binary indicator variables of the study time period when the patient began follow-up as fixed effects, and clinician random effects to account for clustering of patient participants by urologic surgeon. We use analogous linear mixedeffect regression models to analyse outcomes with multilevel scales (e.g., decision regret scale 46 ). In both types of models, we accounted for the stepped-wedge study design as well as the above-mentioned patient covariates. Formerly, let Y ijt denote an outcome measured on the ith patient of the jth surgeon in time period t, OG jt indicate whether surgeon j has transitioned from usual care to the OG by time period t, and X ijt denotes a vector of covariates on the ith patient of the jth surgeon in time period t.
The logistic and linear mixed-effect regression using models have the general form: where in both models λ { } t t=2:5 adjusts for time periods 2 through 5 (time period 1 is the baseline period) and θ j is a random effect specific to surgeon j assumed to be drawn from a normal distribution with mean 0 and an unknown variance. For the linear regression model only,  ijt is an idiosyncratic error term assumed to be drawn from a normal distribution with mean 0 and an unknown variance. In checking for outliers, we identified a few erroneously coded variables that we corrected. While we were prepared to use multiple imputation methods if the missing data was extensive, because drop-out only occurred at time period 2 (i.e., there was no dropout up to time period 1) and most outcomes were analysed at time period 1, we favoured the use of complete-case analyses for all analyses.  Figure 2 shows participant flow details.

| Patient participant characteristics
We enroled 117 patients total, of which 51 were randomised to usual care and 66 to the intervention condition. Most participants were White (84%), non-Hispanic (88%), and had a college degree or more education (56%). Reflecting the onset and progression of the COVID-19 pandemic, 49% of patient visits in the usual care arm were conducted via telehealth; 15% of patient visits in the intervention arm were telehealth.    Table 3 summarises decision-related outcomes collected postvisit (T1). Table 4   to estimate models that adjusted for study time-period, other patient covariates, and physician random-effects was compromised with the effects of most such variables being inestimable. However, the one variable whose effect could be adjusted for was visit date and we found minimal evidence that it was associated with the resolution of treatment indecision.

| Implementation outcomes
Most clinicians had positive views on using the intervention.
Regarding acceptability, all clinicians stated that they approved of the intervention (100% agreed or strongly agreed; mean = 4.  However, a little more than half (3/5 or 60%) of clinicians described using it as easy and effective, only 2/5 (40%) described it as timesaving and collaborative, and 1/5 (20%) described it as necessary. One clinician (20%) described it as inefficient and laborious.
Patients stated that they preferred having cost discussions with a doctor (68% agreed or strongly agreed; mean = 3. It is possible that those enroled in the study early in the recruitment period had higher financial toxicity and financial uncertainty due to the onset of the COVID-19 pandemic. We did note that more participants early in the study reported lower incomes and higher financial toxicity scores; these challenges were faced by many individuals in 2020 as jobs required individuals to stay home to avoid spreading illness, and the economy suffered from widespread shutdowns which were necessary but placed financial strain on individuals and businesses. It is also possible that the use of telehealth (which was more common early in the study, common in 2020 for those with nonurgent needs and those who did not need inperson care) impacted the way in which clinicians and patients discussed options and costs openly. In our study's regression analyses adjusting for these factors, neither visit type (telehealth or in-person) nor visit date were significantly associated with our outcomes. It was important to adjust for these factors in our analyses, but we might not have been able to detect intervention effects above and beyond these differences.
In addition, it is possible that clinicians need more than a conversation-based decision aid and brief training to encourage cost conversations. Clinicians were enthusiastic about the intervention, but it is difficult to change typical conversational flow and content.
Perhaps more documented impact on patient outcomes and more indepth role-playing interventions could serve as stronger motivators and increase self-efficacy for discussing costs. Alternatively, perhaps the patients in our study were less concerned about costs because many were employed with relatively high incomes. A larger study could engage a more socioeconomically diverse group of patients.
It is interesting to note that the intervention might not have encouraged cost-related referrals. The conversation-based decision aid listed resources for patients to contact for assistance or questions with financial aspects of care (e.g., social workers, financial navigators, hospital billing representatives and community organisations). It also listed questions patients could ask if they wanted to learn more about their costs. Perhaps clinicians using the intervention felt that the conversation-based decision aid covered this information without having to bring it up directly, or that many patients in this study did not need additional referrals to discuss costs.
Changing logistics of care 7.8% (4) 6.1% (4) Choosing a lower cost procedure or scan 5.9% (3) 0% (0) Choosing a generic instead of a brand name drug 3.9% (2) 4.6% (3) Copay assistance, coupons, rebates, or samples 3.9% (2) 4.6% (3) Referring to hospital billing office 3.9% (2) 3.0% (2) Setting up a payment plan for bills 2.0% (1) 4.6% (3) Suggesting government or VA assistance 2.0% (1) 3.0% (2) Changing the dose (amount) of treatments, or how often you got treatments, or how the treatments were given (such as a pill instead of an infusion) 2.0% (1) 0% (0) Suggesting that you talk to human resources about your time-off or insurance benefits 2.0% (1) 0% (0) Suggesting a free drug programme 0% (0) 1.5% (1) Other 0% (0) 1.5% (1) Referral made to learn about costs .81 Strengths of the study included the highly engaged clinical team, many of whom thanked the research staff during and after the study for the intervention and resources. Some reported anecdotally that patients were bringing the intervention back to the clinic at their follow-up visit(s) months or even a year later. In addition, adding exact costs to decision aids is resource-intensive and often varies widely by patient. We used relative cost information and referrals to incorporate costs into an existing, previously tested conversation aid.
The engagement of patient and stakeholder advisors also allowed us to adapt an intervention and develop a study flow that met the needs of end users even during difficult times such as the onset of the COVID-19 pandemic. The intervention was rated feasible and acceptable, and some plan to continue using even beyond the study period.
However, limitations included the small sample size (of about 60% of that planned) and heterogeneity of the number of patients by   | 2039