A Randomized Controlled Trial of the Implementation of BREASTChoice, a Multilevel Breast Reconstruction Decision Support Tool With Personalized Risk Prediction

Structured Abstract

Objective:

To implement the BREASTChoice decision tool into the electronic health record and evaluate its effectiveness.

Background:

BREASTChoice, is a multilevel decision tool that: 1) educates patients about breast reconstruction; 2) estimates personalized risk of complications; 3) clarifies patient preferences; and 4) informs clinicians about patients’ risk and preferences.

Methods:

A multisite randomized controlled trial enrolled adult women with stage 0-III breast malignancy undergoing mastectomy. Participants were randomized to BREASTChoice or a control website. A survey assessed knowledge, preferences, decisional conflict, shared decision-making, preferred treatment, and usability. We conducted intent-to-treat (ITT), per-protocol (PP) analyses (those randomized to BREASTChoice who accessed the tool), and stratified analyses.

Results:

23/25 eligible clinicians enrolled. 369/761 (48%) contacted patients enrolled and were randomized. Patients’ average age was 51 years; 15% were older than 65. BREASTChoice participants had higher knowledge than control participants (ITT: mean 70.6 vs. 67.4, p=0.08; PP: mean 71.4 vs. 67.4, p=0.03), especially when stratified by site (ITT: p=0.04, PP: p=0.01), age (ITT: p=0.04, PP: p=0.02), and race (ITT: p=0.04, PP: p=0.01). BREASTChoice did not improve decisional conflict, match between preferences and treatment, or shared decision-making. In PP analyses, fewer high-risk patients using BREASTChoice chose reconstruction. BREASTChoice had high usability.

Conclusions:

BREASTChoice is a novel decision tool incorporating risk prediction, patient education, and clinician engagement. Patients using BREASTChoice had higher knowledge; older adults and those from racially minoritized backgrounds especially benefitted. There was no impact on other decision outcomes. Future studies should overcome implementation barriers and specifically examine decision outcomes among high-risk patients.

Mini Abstract:

BREASTChoice is the first breast reconstruction decision tool, to our knowledge, that incorporates personalized risk prediction, evidence-based patient education, and clinician decision support. This paper reports on a multisite randomized controlled trial evaluating the implementation of the BREASTChoice tool into the electronic health record, examining its impact on decision quality, surgical choice, and shared decision-making.

Background:

Breast reconstruction includes a series of surgical procedures to create a breast shape after mastectomy. A clear understanding of the procedures is critical to comprehensive breast cancer treatment^{1 2}. Reconstruction decisions are considered preference-sensitive; patients’ goals and priorities are central to their choice. Patients can opt to have reconstruction or not, implant-based or autologous (flap) reconstruction, and immediate or delayed reconstruction. Although reconstruction can improve quality of life and body image among some women, it can increase the risk of complications and adds multiple procedures at a time when women are undergoing primary treatment for breast cancer.³ Each type of reconstruction carries specific tradeoffs such as risk to donor sites and risk of a longer surgery for autologous reconstruction, or risk of leaking or replacement surgery over time for implant-based reconstruction.⁴ Unfortunately, patients often have knowledge gaps about reconstruction and its risks⁵, and their decisions can be discordant with their preferences about risk, appearance, recovery, and the number of procedures involved. Many women experience regret about their reconstruction choice⁶.

There are also age- and race-based disparities in reconstruction decisions^7–9. Women over age 65 are less likely to undergo reconstruction and often report they were not offered reconstruction.¹⁰ Although some race-based disparities may have improved in recent years, Black and Latina patients overall are less likely to undergo breast reconstruction or to be referred to a reconstructive surgeon, which appears to be related to multiple factors, including social determinants of health, distance to a reconstructive surgeon, and insurance coverage.^11–13 Black women may be less satisfied with reconstruction decision-making as well^{7 14}. Overall, patients considering breast reconstruction would benefit from decision support, which can improve decision quality¹⁵. Yet decision support is seldom used in practice^{16 17}, especially in minoritized populations¹⁸.

BREASTChoice is a multilevel decision tool designed to: 1) educate patients about breast reconstruction options, including information about reconstruction compared to no reconstruction, implant-based or autologous (flap) reconstruction, and immediate or delayed reconstruction; 2) estimate risks of major wound complications, especially from immediate reconstruction; 3) help patients explore preferences for the various available options; and 4) send clinicians information about patients’ risk and preferences^19–21. Based on patient and community feedback, BREASTChoice was designed for women of all ages, races, and body types, with a photo library of diverse reconstruction results including no reconstruction. A single-site randomized trial found that patients using BREASTChoice had higher reconstruction knowledge compared to those in usual care¹⁹.

After additional stakeholder engagement, the tool was refined and found to be usable, feasible, and acceptable to users²². To better support implementation, BREASTChoice was integrated into the electronic health record (EHR) and patient portal for ease of use and access during clinical care²³. To our knowledge, BREASTChoice is the first breast reconstruction decision tool that incorporates personalized risk prediction, evidence-based patient education, and clinician decision support (see figure 1).

Figure 1:

BREASTChoice Patient Homepage

This paper reports on a multisite randomized controlled trial evaluating the implementation of BREASTChoice into care. We hypothesized that compared to those in a control group receiving a website about breast reconstruction, patients randomized to BREASTChoice would report higher decision quality (higher knowledge, lower decisional conflict, and a better match between preferences and treatment). We explored whether women at higher risk for complications from breast reconstruction would be more likely to choose no or delayed reconstruction, and whether those in the BREASTChoice group would report more shared decision-making with their plastic surgeon.

Methods:

Recruitment and Eligibility:

We recruited patients at two sites in the Midwest of the United States, hereafter referred to as Sites A and B. Both sites are academic medical centers with urban, suburban, and rural-residing patients in a large catchment area and both offer all types of reconstruction. We planned to include a community health center. However, the study began at the start of the COVID-19 pandemic in the U.S. The informatics team at the community site needed to prioritize pandemic-related activities and eventually declined participation.

The Institutional Review Boards at Sites A (ID 202005217) and B (ID 2020C0154) approved the study. The trial was registered with clinicaltrials.gov (NCT04491591). A data safety and monitoring board with representation from both sites met quarterly. Study procedures, workflow, and randomization were piloted before the trial from November 2020-January 2021 at Site A, and from April-June 2021 at Site B. Study procedures and BREASTChoice programming were revised based on feedback during the pilot.

Enrollment occurred between January 2021-July 2022. Eligible patients were adults (aged 18+), female sex at birth, with incident or recurrent stage 0-III breast malignancy, who were considering mastectomy.

Patients were ineligible if they had metastatic disease, histology type other than ductal or lobular carcinoma, or had already undergone mastectomy. Patients who did not have a malignancy, were planning to have partial mastectomy, could not read or speak English, or cognitive impairment limiting ability to complete study procedures were also ineligible.

Eligible clinicians were plastic surgeons who performed breast reconstruction. Physician assistants were included at site B because they sometimes first counseled patients when a plastic surgeon was unavailable.

Enrollment and Randomization:

Consenting clinicians received a brief training (virtual or in-person) on BREASTChoice and its features at the start of the study. The training included instructions to locate the EHR summary for patients randomized to BREASTChoice. After the study began, we allowed patients and clinicians to use the tool as it worked for them, as evidence suggests this process allows for integration into routine care.²⁴ Clinicians were sent pre- and post-trial surveys to assess attitudes toward shared decision-making and intention to use shared decision-making and BREASTChoice in routine care. Clinician participants received a $50 gift card incentive.

Patients were screened for eligibility through clinicians’ schedules in the EHR. The research team approached eligible patients in person in clinic, by phone, or via a patient portal message. Surgical oncologists assisted by giving study flyers to patients who might be eligible. Eligibility was confirmed with a screening questionnaire.

Participants were randomized using computer-generated random assignment in blocks of 2 and 4 to the BREASTChoice or control (a National Cancer Institute website about reconstruction²⁵) group. Participants received links to their assigned condition by secure email or patient portal. We sought to enroll patients before their plastic surgery appointment. However, some patients saw a plastic surgeon within hours of seeing the surgical oncologist and received study materials after the visit. We included such patients if they had not yet made a decision about reconstruction.

After viewing BREASTChoice or the control website, participants received an electronic survey that assessed decision quality, knowledge, decisional conflict, shared decision-making, preferred treatment (if known), and usability. Participants received a $20 gift card. We abstracted participants’ reconstruction choices (whether they had reconstruction, reconstruction type, and reconstruction timing) from the EHR, updating this data through February 2023, seven months after the last participant enrolled.

Audio Recording Consultations:

Because randomization occurred at the patient level, to check for possible contamination into the control group, a random sample of clinician-patient encounters were audio recorded with consent. We sought to record up to 20% of the encounters. Due to COVID surges and restrictions on non-essential personnel in clinics, 10% of encounters were recorded across sites. The study team reviewed transcripts and did not observe any contamination or reference to BREASTChoice during control encounters.

EHR Integration:

Each site worked closely with their institution’s informatics teams to integrate BREASTChoice into the EHR and ensure security compliance. Details about the EHR integration process were published previously²³.

Risk Scoring

The validated risk prediction model included the following risk factors for major wound complications after mastectomy: immediate reconstruction, bilateral mastectomy, obesity, depression in the past two years, psychosis, congestive heart failure, use of any nicotine products in the past 6 months, diabetes, and hypertension^{21 26}.

In the BREASTChoice condition, the tool automatically calculated the patients’ risk of major wound complications, using data from the EHR. The same approach was taken for patients in the control condition, but control patients and clinicians did not view the risk estimate. The tool gave participants in the BREASTChoice condition the option to review and edit their risk factors.

Retrieving history of chest radiation from the EHR has poor sensitivity at our sites. Some women receive radiation outside our institutions, and there is high variability between institutions, within institutions, and even within surgeons regarding where and when chest radiation is documented in the EHR. Thus we left this out of the model. We erred on the side of under-counting rather than over-counting risk factors.

Site B had a programming issue estimating risk for the 78 control participants and 13 intervention participants. For analyses purposes, we reviewed charts to obtain risk factor data for those participants, and calculated risk scores using the same regression equation used in BREASTChoice.

BREASTChoice Summary

After participants used BREASTChoice, they received a printable summary that included their personalized risk estimate, elicited preferences, and patient-inputted questions for the surgeon. The risk estimate was shown numerically and with an icon array (Figure 2). The BREASTChoice summary was sent to the clinician via the EHR (Figure 3). At Site B, this clinician summary was delivered on paper during the first 2 months of the trial because electronic delivery was still being programmed. Once integrated, the clinician summary was delivered as a best practice alert. Site B had a programming issue with the preference section for patients who used the web browser Safari. The tool did not always save those preferences, so the clinician received a blank preference summary.

Figure 2:

Sample Risk Display

Figure 3:

Sample Clinician Summary

Measures

Primary Outcomes

a. Knowledge

Nine questions were used and adapted from the Decision Quality Index (DQI) to assess participants’ knowledge about mastectomy with and without reconstruction²⁷.

b. Preference concordance

Surgery preferences:

After each BREASTChoice module (whether to have reconstruction, reconstruction type, and reconstruction timing), or in a survey for control participants, participants selected which surgical option they were leaning towards (reconstruction vs. no reconstruction, implant vs. flap-based reconstruction, immediate vs. delayed). For the BREASTChoice group, if the preference changed as someone used the tool, we used the last stated preference. We also explored the percentage of BREASTChoice users whose preferences changed as they used the tool.

Reconstruction versus no reconstruction:

We reviewed the EHR for breast cancer surgery and reconstruction through February 2023, seven months after the last enrolled participant. If no breast cancer surgery (mastectomy, partial mastectomy) was performed during this period, the patient was excluded from reconstruction analyses. If a patient had breast cancer surgery but no reconstruction started, we considered this “no reconstruction by the time of chart review.”

Reconstruction type:

Placement of a permanent breast implant or a tissue expander followed by a permanent implant was classified as implant-based reconstruction. Any procedure that involved a flap (including placing a tissue expander followed by a flap) was considered flap-based reconstruction. If a tissue expander or implant was placed under a flap, this was also classified as flap-based reconstruction. Other more complex cases (e.g., tissue expander placed but removed) were reviewed in detail for documented final procedures. Incomplete final procedures at the time of chart review were not coded for reconstruction type.

Reconstruction timing:

Patients who underwent mastectomy and reconstruction on the same day were classified as having immediate reconstruction. Patients who had mastectomy with no reconstruction that day, followed by reconstruction later in the study period, were classified as having delayed reconstruction.

Preference Concordance:

We assessed preference concordance for those who had documented preferences and known surgical procedures. Preference concordance was defined as agreement between preferred and actual treatment for reconstruction vs. no reconstruction, type of reconstruction, and timing of reconstruction.

c. Decisional conflict

The 4-item SURE measure was used to assess participants’ certainty about choice. Higher scores indicated less decisional conflict, or more certainty²⁸. Outcomes were dichotomized as a score of 4 (no decisional conflict) vs. ≤3 (decisional conflict) per scoring guidelines.

Secondary and Exploratory Outcomes

d. BREASTChoice-specific knowledge

After each BREASTChoice module, or by survey for control participants, participants answered eleven true/false/unsure questions, which covered more knowledge domains than the DQI, including reconstruction type and timing. They were designed as part of the tool’s education and developed in past work¹⁹. BREASTChoice-specific knowledge was scored as a percentage correct out of total items answered, if more than 50% of items were answered. Selecting ”unsure” was considered incorrect.

e. Number of high-risk people who choose reconstruction

We considered “high risk” to be two times the average risk using the validated model. The a priori average risk for mastectomy with immediate reconstruction at Site A was 16%, so 32% was selected as the threshold for “high-risk.”

f. Intentions to engage in shared decision-making

We adapted a measure examining how interventions impact clinicians’ intentions to change clinical behavior²⁹. We used mixed effects models to examine change in mean score pre- to post-trial.

g. Usability

Usability of BREASTChoice was measured in the BREASTChoice group using the 10-item System Usability Scale³⁰. Scores above 68 are considered strong usability³⁰.

h. Shared decision-making

Shared decision-making was evaluated using the 3-item CollaboRATE measure^{31 32}. The top score method was used indicating whether “every effort was made” or “less than every effort was made” to engage patients in decision-making^{31 32}.

Data Analyses:

For the DQI knowledge outcome, an overall score was calculated by dividing the number of correct responses (range 0–9) by 9 to rescale to a score from 0–100. If more than 50% of items were unanswered, knowledge was not scored and treated as missing. The Wilcoxon rank-sum test compared DQI knowledge between intervention and control groups due to skewness. The Hodges-Lehmann estimator identified the location shift of the median score and a 95% confidence interval (CI). Two-sample proportion tests compared differences in binary outcomes (preference concordance, SURE, proportion of high-risk patients choosing reconstruction) between study groups and 95% CIs for difference in proportions. Adjusted models assessed the robustness of results accounting for a priori selection of age (<65 vs. ≥65 years), race (White, Black/African American, or another race(s)), and site.^{33 34} Separate stratified Wilcoxon tests provided adjusted analyses of DQI knowledge, while logistic regression models including age, race, and site provided adjusted analyses of binary outcomes. The Student’s T-test estimated differences between groups’ BREASTChoice-specific knowledge. Mixed effects models including a random effect for clinician (with adjusted analysis including age, race, and site) estimated differences in shared decision-making between groups. We present intent-to-treat (ITT) analyses and per-protocol (PP) analyses that excluded patients randomized to BREASTChoice who never accessed the tool. Results were considered statistically significant if p < 0.05 and were done using SAS Statistical Software 9.4 (SAS Institute Inc., Cary, NC).

Results

23/25 approached clinicians enrolled (15 plastic surgeons, 8 physician assistants). Between January 2021-July 2022, 1,426 patients were screened (745 at Site A, 681 at Site B). Of these, 689 did not meet inclusion criteria. Some eligible patients declined participation (232, 16%) or were not reached (132, 9%). Of the 761 eligible and approached, 369 were enrolled and randomized (48%). Of those 369, 48 did not complete a survey and were excluded from most analyses. Figure 4 displays a CONSORT flow diagram. The average age of the 369 randomized patient participants was about 51 years; 15% were over age 65. Most were White, non-Hispanic, had a college degree, and had adequate health literacy (Table 1).

Figure 4:

CONSORT Flow Diagram

Table 1:

Patient Participant Characteristics by Randomized Group, Including All Randomized and Enrolled (N=369)

	BREASTChoice (n=184)	Control (n=185)
Age (years)
Mean (SD, Range)	51.0 (10.8, 31–75)	51.2 (11.2, 25–75)
<65 years	156 (84.8%)	158 (85.4%)
65+ years	28 (15.2%)	27 (14.6%)
Race
White	167 (90.8%)	156 (84.3%)
Black/African American	11 (6.0%)	21 (11.4%)
Asian-American	2 (1.1%)	3 (1.6%)
More than one race	4 (2.2%)	3 (1.6%)
Another race or Unknown	0 (0.0%)	2 (1.1%)
Ethnicity
Non-Hispanic	182 (98.9%)	180 (97.3%)
Hispanic	2 (1.1%)	5 (2.7%)
Health Literacy (SILS)	n=142	n=158
Limited	5 (3.5%)	11 (7.0%)
Adequate	137 (96.5%)	147 (93.0%)
Annual household income 1	n=152	n=158
<$30,000	12 (7.9%)	17 (10.8%)
$30,000 to $74,999	33 (21.7%)	28 (17.7%)
$75,000 or more	91 (59.9%)	96 (60.8%)
Prefer not to answer	16 (10.5%)	17 (10.8%)
Educational attainment 1	n=142	n=158
High school or less	15 (10.6%)	16 (10.1%)
Technical training/Some college	31 (21.8%)	27 (17.1%)
College degree	47 (33.1%)	58 (36.7%)
Graduate/professional degree	49 (34.5%)	57 (36.1%)
When was BREASTCHOICE accessed?
Prior to appointment	102 (55.4%)
After appointment	50 (27.2%)	----
Never accessed	32 (17.4%)
Estimated Risk of Complication (%) 2
Mean (SD, Range)	17.8 (19.5, 7–90)	16.4 (16.3, 7–90)
Median (IQR)	12 (9–17)	12 (9–17)
N (%) at high risk (score 32+)	16 (8.7%)	13 (7.0%)
Breast cancer Stage
Stage 0	41 (22.3%)	32 (17.3%)
Stage I	75 (40.8%)	77 (41.6%)
Stage II	45 (24.5%)	54 (29.2%)
Stage III	22 (12.0%)	19 (10.3%)
Stage IV3	1 (0.5%)	3 (1.6%)
Breast Cancer Surgery	n=184	n=185
Had mastectomy	156 (84.8%)	149 (80.5%)
Had breast conserving surgery	23 (12.5%)	29 (15.7%)
Not yet had breast cancer surgery at time of chart review	5 (2.8%)	7 (3.8%)
Reconstructive Surgery (of those who had mastectomy)	n=156	n=149
Had reconstruction	117 (75.0%)	118 (79.2%)
Had no Reconstruction at time of chart review	39 (25.0%)	31 (20.8%)
Timing of Reconstruction (of those who had reconstruction)	n=117	n=118
Immediate Reconstruction	112 (95.7%)	114 (96.6%)
Delayed Reconstruction	5 (4.3%)	4 (3.4%)
Type of Reconstruction (of those who had reconstruction)	n=117	n=118
Implant-based	55 (47.0%)	59 (50.0%)
Flap-based	32 (27.4%)	36 (30.5%)
Unsure (final reconstruction procedure not completed at time of chart review)	30 (25.7%)	23 (19.5%)

¹Clinical variables were obtained from EHRs; age, race, ethnicity, health literacy, income, education were participant-reported

²Risk was manually calculated for Site B controls (n=78) & Site B BREASTChoice participants who did not access the tool (n=13)

³Stage IV was a criterion for exclusion, but some participants were eligible at the time of enrollment and upstaged later

Table 2 summarizes the primary and secondary outcomes using intention-to-treat (ITT) analyses, including adjustment for site, race, age, and clinician as appropriate. We also display our a priori planned adjusted analyses by site, age, and race.

Table 2:

Primary and Secondary Patient-Level Outcomes by Group, ITT analysis

	BREASTChoice (n=156)	Control (n=165)	Unadjusted Analysis	Stratified Analysis
Primary Outcomes (continuous)
DQI Knowledge			Location shift = 5.5		By age: p=0.04
Mean (SD)	70.6 (13.2)	67.4 (14.7)	(0.0, 11.1)	By site: p=0.04	By race: p=0.04
Median (IQR)	66.7 (66.7–77.8)	66.7 (55.6–77.8)	p=0.08
Primary Outcomes (categorical)	BREASTChoice (n=156)	Control (n=165)	Difference in proportion or means (95% CI)	Adjusted (for site)	Adjusted (for site, age, race)
Reconstructive surgery 1
Yes	103 (76.3%)	109 (79.6%)
No	32 (23.7%)	28 (20.4%)	-	-	-
Preference Concordance
Reconstruction vs. No reconstruction 2	n=156	n=165	−6.2%
Treatment matches preference	83 (86.5%)	113 (92.6%)	(−14.3, 1.9%)
Treatment doesn’t match preference	13 (13.5%)	9 (7.4%)	p=0.14	p=0.04	p=0.04
Immediate vs. Delayed Reconstruction 3	n=103	n=109	4.5%
Treatment matches preference	48 (81.3%)	73 (76.8%)	(−9.0%, 18.0%)
Treatment doesn’t match preference	11 (18.6%)	22 (23.2%)	p=0.51	p=0.38	p=0.35
Flap vs. Implant Reconstruction 4	n=103	n=109	5.7%
Treatment matches preference	33 (84.6%)	60 (78.9%)	(−9.8%, 21.1%)
Treatment doesn’t match preference	6 (15.4%)	16 (21.1%)	p=0.45	p=0.39	p=0.40
SURE: Decisional Conflict 5	n=142	n=157	−6.2%
Decisional conflict	31 (21.8%)	44 (28.0%)	(−16.1%, 3.7%)
No decisional conflict	111 (78.2%)	113 (72.0%)	p=0.22	p=0.24	p=0.22
Secondary Outcomes
Proportion of high-risk patients choosing reconstruction 6	n=16	n=13
Chose reconstruction	10 (71.4%)	11 (100.0%)	−28.6%	-	-
Chose no reconstruction	4 (28.6%)	0 (0.0%)	(−57.9%, 0.8%)
			p=0.056
Knowledge as assessed in BREASTChoice
(Range 27.3–100%)	n=147	n=154	18.2%
Mean (SD)	84.7 (13.8)	66.5 (15.8)	(14.8, 21.6)	p<0.001	p<0.001
			p<0.001
Exploratory Outcome
CollaboRATE Top Score Method 7	n=141	n=156
Less than every effort was made	78 (55.3%)	92 (59.0%)	3.7%
Every effort was made	63 (44.7%)	64 (41.0%)	(−7.6%, 14.9%)
			p=0.53	p=0.26	p=0.37

¹Excluded: no mastectomy yet (BC n=4, Control n=4); breast-conserving surgery (BC n=17, Control n=24)

²Excluded: no mastectomy yet (BC n=4, Control n=4); “unsure” preference about having reconstruction reported (BC n=3, Control n=7); no preference about having reconstruction reported (BC n=43, Control n=13); breast-conserving surgery (BC n=10, Control n=19)

³Of those who had Reconstruction. Additional excluded participants: “unsure” timing preference (BC n=9, Control n=6); no timing preference (BC n=35, Control n=8)

⁴Of those who had Reconstruction. Additional excluded participants: final reconstruction type not available yet (BC n=23, Control n=21); “unsure” type preference reported (BC n=8, Control n=6); no type preference reported (BC n=33, Control n=6)

⁵Excluded: missing SURE questions (BC n=14, Control n=8)

⁶Includes all randomized participants, not just those with survey data. P-value and estimated differences are provided for the proportion who chose reconstruction versus no reconstruction. Additional excluded participants: had breast-conserving surgery (BC n=2, Control n=1); did not yet have mastectomy (Control n=1)

⁷Adjusted models for CollaboRATE also included a random effect for surgeon. 20 patients at Site B saw a PA to discuss reconstruction and did not have a reconstructive surgeon listed in the EHR, so were excluded from the CollaboRATE adjusted analysis

Table 3 summarizes our per-protocol (PP) analyses. For this analysis, we excluded BREASTChoice participants who never accessed the tool, but responded to surveys sent outside of the tool.

Table 3:

Per-protocol analysis considering those in the BREASTChoice group who accessed the tool compared to control participants, for selected outcomes

	BREASTChoice (n=150)	Control (n=165)	Unadjusted Analysis	Stratified Analysis
Primary Outcome (continuous)
DQI Knowledge					By age: p=0.02
Mean (SD)	71.4 (12.8)	67.4 (14.7)	Location shift = 5.5 (0.0, 11.1)	By site: p=0.01	By race: p=0.01
Median (IQR)	66.7 (66.7–77.8)	66.7 (55.6–77.8)	p=0.03
Primary Outcomes (categorical)	BREASTChoice (n=150)	Control (n=165)	Difference in proportions or means (95% CI)	Adjusted (for site)	Adjusted (for site, age, race)
Reconstructive surgery 1
Yes	100 (76.9%)	109 (79.6%)
No	30 (23.1%)	28 (20.4%)	-	-	-
Preference Concordance
Reconstruction vs. No reconstruction 2	n=150	n=165	−6.2%
Treatment matches preference	83 (86.5%)	113 (92.6%)	(−14.4, 1.9%)
Treatment does not match preference	13 (13.5%)	9 (7.4%)	p=0.14	p=0.04	p=0.04
Immediate vs. Delayed 3	n=100	n=109	4.5%
Treatment matches preference	48 (81.4%)	73 (76.8%)	(−9.0%, 18.0%)
Treatment doesn’t match preference	11 (18.6%)	22 (23.2%)	p=0.51	p=0.38	p=0.35
Flap vs. Implant 4	n=100	n=109	5.7%
Treatment matches preference	33 (84.6%)	60 (78.9%)	(−9.8%, 21.1%)
Treatment doesn’t match preference	6 (15.4%)	16 (21.1%)	p=0.47	p=0.40	p=0.40
SURE: Decisional Conflict 5	n=136	n=157	−5.2%
Decisional conflict	31 (22.8%)	44 (28.0%)	(−15.3, 4.0%)
No decisional conflict	105 (77.2%)	113 (72.0%)	p=0.31	p=0.32	p=0.30
Secondary Outcomes
Proportion of high-risk patients choosing reconstruction 6	n=13	n=13
Chose reconstruction	8 (66.7%)	11 (100.0%)	−33.3%	-	-
Chose no reconstruction	4 (33.3%)	0 (0.0%)	(−64.3%, 2.4%)
			p=0.04
Knowledge as assessed in BREASTChoice			18.2%
(Range 27.3–100%)	n=147	n=154	(14.8, 21.6)
Mean (SD)	84.7 (13.8)	66.5 (15.8)	p<0.001	p<0.001	p<0.001
Exploratory Outcome
CollaboRATE Top Score Method 7	n=135	n=156
Less than every effort was made	73 (54.1%)	92 (59.0%)	4.9%
Every effort was made	62 (45.0%)	64 (41.0%)	(−6.5%, 16.3%)	p=0.19	p=0.27
			p=0.40

¹Excluded: no mastectomy yet (BC n=4, Control n=4); breast-conserving surgery (BC n=16, Control n=24)

²Excluded: no mastectomy yet (BC n=4, Control n=4); “unsure” preference about having reconstruction reported (BC n=3, Control n=7); no preference about having reconstruction reported (BC n=37, Control n=13); breast-conserving surgery (BC n=10, Control n=19)

³Of those who had Reconstruction. Additional Excluded participants: “unsure” timing preference reported (BC n=9, Control n=6); no timing preference reported (BC n=32, Control n=8)

⁴Of those who had Reconstruction. Additional excluded participants: final reconstruction type not available yet (BC n=22, Control n=21); “unsure” type preference reported (BC n=8, Control n=6); no type preference reported (BC n=31, Control n=6)

⁵Excluded: missing SURE questions (BC n=14, Control n=8)

⁶Includes all randomized participants, not just those with survey data. P-value and estimated differences are provided for the proportion who chose reconstruction versus no reconstruction. Additional excluded participants: had breast-conserving surgery (BC n=1, Control n=1); did not yet have mastectomy (Control n=1)

⁷Adjusted models for CollaboRATE also included a random effect for surgeon. 20 patients at Site B saw a PA to discuss reconstruction and did not have a reconstructive surgeon listed in the EHR, so were excluded from the CollaboRATE adjusted analysis

Primary Outcomes

a. Knowledge (DQI)

BREASTChoice participants had higher average knowledge (mean 70.6) compared to control participants (mean 67.4); however, this difference was not statistically significant (p= 0.08) in ITT analyses (Table 2). ITT analyses stratified by site, age, and race were statistically significant (p=0.04). In PP analyses, BREASTChoice participants had significantly higher knowledge (mean score 71.4) compared to control participants (mean 67.4; p=0.03; Table 3). PP analyses stratified by site (p=0.01), age (p=0.02), and race (p=0.01) were significant (Table 3).

b. Decisional conflict

Participants using BREASTChoice reported about the same level of decisional conflict (22%) compared to control participants (28%) in ITT (Table 2) and PP (Table 3) analyses.

c. Preferences

96/156 (61.5%) people using BREASTChoice compared to 121/165 (73.3%) control participants preferred to have reconstruction. 60/103 (58.2%) people using BREASTChoice compared to 80/109 (73.4%) control participants who wanted reconstruction preferred to have immediate reconstruction, and 38/103 (36.9%) people using BREASTChoice group compared to 56/109 (51.4%) control participants who wanted reconstruction preferred to have implant-based reconstruction.

Tables 2 and 3 display the percentage of people in each group whose treatment choice matched their preferences. Most people in both groups chose a treatment that matched their preferences; a slightly higher percentage of the control group received treatment that matched their preference for having reconstruction (p<0.04), although many participants in the BREASTChoice group were unsure of their preferences at survey completion (38.5% skipped a preference question versus 13.3% of control participants). In addition, 96/156 (61.5%) of BREASTChoice users had stable preferences across all three questions about surgery choice in the tool, 43 (27.6%) were unsure of their preferences, and 17 (10.9%) changed their preferences as they used the tool. In the control group, 148/165 (89.7%) had stable preferences from the first to last preference question in the survey, 13 (7.9%) were unsure of their preferences, and 4 (2.4%) changed their preference during the survey.

Secondary and Exploratory Outcomes

c. BREASTChoice-specific knowledge

In ITT analyses of BREASTChoice-specific knowledge¹⁹, BREASTChoice participants had higher knowledge (mean 84.7) compared to control participants (mean 66.5; p<0.001). In PP analyses, BREASTChoice users also had higher knowledge (mean = 85.3) compared to control participants (mean = 66.5; p<0.001).

d. Number of high-risk people who choose breast reconstruction

There was no difference in the proportion of high-risk patients in the BREASTChoice group who opted for reconstruction compared to the control group, in ITT analyses (71.4% in the BREASTChoice vs. 100% in the control group; p=0.11; Table 2). In PP analyses, this difference approached statistical significance (66.7% in the BREASTChoice group vs. 100% in the control group; p=0.056; Table 3).

e. Intentions to engage in shared decision-making

Clinicians’ intentions to engage in shared decision-making were similar from pre- (mean 6.4) to post-trial (mean 6.6). They reported similar social perceptions (mean 5.8 vs. 5.2), beliefs about capabilities (mean 6.2 vs. 6.0), moral norms (mean 6.7 vs. 6.4), and beliefs about consequences (mean 6.5 vs. 6.3) pertaining to shared decision-making (all p’s>0.05).

f. Usability

BREASTChoice was rated highly usable with a mean score of 84.6 (SD=14.3).

g. Shared decision-making:

BREASTChoice participants reported about the same levels of shared decision-making (44%) compared to the control participants (41%); this was not significantly different in ITT (Table 2) or PP (Table 3) analyses.

Discussion

BREASTChoice is a novel breast reconstruction decision support tool that incorporates personalized risk prediction, evidence-based patient education, and clinician decision support. Participants randomized to this multilevel intervention demonstrated higher knowledge about reconstruction, including about type, timing, and complication risks. In addition, in PP analyses of those who used BREASTChoice, fewer high-risk patients chose to have reconstruction. BREASTChoice did not decrease decisional conflict, improve preference concordance, or increase shared decision-making in this study.

Findings are consistent with past work about decision aids improving knowledge¹⁵, including an earlier version of BREASTChoice. Specific knowledge about type and timing of reconstruction demonstrated the greatest improvement between the intervention and control groups. Specific information about implants, flaps, and risks of complications from immediate breast reconstruction are essential to informed choices; it is a unique strength of BREASTChoice that patients learned information that clinicians and patients in past work deemed important.^{27 33 35} The impact of BREASTChoice on patients’ knowledge was even more pronounced when stratified by age, race, and site. Older adults and those from racially minoritized backgrounds could benefit from evidence-based, accessible information in this context. Future research could specifically design or adapt decision aids for these groups.

Although BREASTChoice did not improve perceptions of shared decision-making, future research should explore methods to encourage implementation of decision tools.^{36 37} In our study, some clinicians did not access the BREASTChoice summary page at the point-of-care. Others viewed it, but not each time. At one site, technology challenges interfered with delivery of the BREASTChoice summary. Clinicians had high baseline support for shared decision-making. Future studies should work to overcome implementation barriers to shared decision-making, especially with clinicians who might not already support this approach.

BREASTChoice’s limited impact on preference concordance was surprising. Preferences tended to match choices in both groups, which is positive. Measuring preferences is challenging and time-sensitive; some BREASTChoice participants left the preference sections blank or changed their preferences as they proceeded through the tool. It is possible that BREASTChoice facilitated deliberation^{38 39} among options; perhaps preference and choice matched by the time a surgery choice was made. Some reconstructive surgeons might not offer reconstruction to someone who uses nicotine or has an elevated body mass index, which could reduce preference concordance. How BREASTChoice might affect preference concordance in that circumstance is not clear. Future work should study preferences and how they shift over time, measured at different times in the care pathway.

This study should be interpreted within the context of some limitations. It began in 2020 at the peak of the COVID-19 pandemic during which clinicians and patients had additional stressors on top of the stressors of usual practice (e.g., virtual visits, staffing shortages, anxiety about contracting COVID-19, clinician burnout). The pandemic also led to staffing shortages and increased demands on informatics teams. One community site could not participate, thus the final sample might not be representative of patients at community locations. The two included sites took slightly longer to launch than planned. We were not allowed to conduct observation of clinical workflows prior to the trial, which could have affected implementation. Programming bugs took longer to address given informatics resource challenges. Control participants were also fairly activated; in fact, some searched for photos or decision support outside of the study which could have affected study results. The final sample lacked diversity by health literacy and was not as racially diverse as planned. Although formative work included adaptations to ensure that the tool was applicable to women from diverse racial and ethnic groups (including the photo library and images), future work could revisit key study questions over time in a more diverse sample of patients and institutions, with participation from community sites. This study was limited to English-speaking women; future projects are adapting the tool for Spanish-speaking populations.⁴⁰ Some patients viewed the decision tool after while others viewed it before their appointment with a reconstructive surgeon; future work could explore how the timing of using a decision tool affects outcomes. Finally, the risk prediction model focuses on wound complications and does not address cosmetic outcomes over time; although the text and photos in the tool describe cosmetic outcomes, future work could build on this validated risk prediction model and focus on longer-term outcomes such as cosmetic results. Such work could also measure the impact of longer-term outcomes on decision regret.^{41 42}

Overall, BREASTChoice shows promise at improving decision outcomes, and its integration into the EHR helped prepare it for dissemination. However, with EHR integration of decision tools, some design choices do not work across sites and EHR systems, and programming bugs can impact use. Alert fatigue and workflow barriers could also lead some clinicians to ignore EHR-delivered summary data. Future work could test ways to implement a decision tool such as BREASTChoice with both patient- and clinician-facing components. Future studies of integrated risk prediction tools should assess their impact on high-risk patients’ choice of lower versus higher-risk procedures.