Shared Decision-Making in the Emergency Department: The Chest Pain Choice Trial

Erik P Hess1,2, Judd E Hollander2, Jason T Schaffer3, Jeffrey A Kline5, Carlos A Torres1,5, Deborah B Diercks4, Russell Jones5, Kelly P Owen5, Zachary F Meisel6, Michel Demers,10, Annie Leblanc1, Nilay D Shah1, Jonathan Inselman1, Jeph Herrin9, Ana Castaneda-Guarderas1,10, Victor M Montori1

1. Department of Emergency Medicine, Mayo Clinic College of Medicine, Knowledge and Evaluation Research Unit, Mayo Clinic, and Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Healthcare Delivery, Rochester, MN 55905, USA

2. Department of Emergency Medicine; Thomas Jefferson University, Philadelphia, PA 19107, USA, 3. Department of Emergency Medicine; Indiana University, Indianapolis, IN 46202, USA 4. Department of Emergency Medicine; University of Texas Southwestern Medical Center, Dallas, TX, USA 5. Department of Emergency Medicine; University of California Davis Medical Center, Sacramento, CA, USA 6. Department of Emergency Medicine, Perelman School of Medicine, Philadelphia, PA, USA 7. Patient representative, Rochester, MN, USA 8. Caregiver representative, Rochester, MN, USA 9. Yale University School of Medicine, New Haven, CT 06510, USA and Health Research & Educational Trust,

Chicago IL 60606, USA 10. Department of Emergency Medicine, Aventura Hospital and Medical Center, Aventura, FL 33180, USA

Correspondence to: Erik P Hess, Department of Emergency Medicine, Mayo Clinic, 200 First Street SW, Rochester, MN 55906, USA, hess.erik@mayo.edu PCORI Project ID: 952 HRSProj ID: 20142216 CT.Gov ID: NCT01969240

2

Table of Contents ABSTRACT .......................................................................................................................................... 3 BACKGROUND ................................................................................................................................... 4

Stakeholder involvement .......................................................................................................................... 5 METHODS .......................................................................................................................................... 5

Study design and setting ........................................................................................................................... 5 Participants ............................................................................................................................................... 6 Randomization and masking ..................................................................................................................... 6 Study treatments ...................................................................................................................................... 7

Intervention ..................................................................................................................................... 7 Delivery of the intervention ............................................................................................................. 9 Usual care ...................................................................................................................................... 10 Data collection ............................................................................................................................... 10

Outcomes ................................................................................................................................................ 11 Primary outcomes .......................................................................................................................... 11 Secondary outcomes ...................................................................................................................... 12 Clarification of primary outcome ................................................................................................... 14 Statistical analysis .......................................................................................................................... 14

RESULTS .......................................................................................................................................... 17 Patient knowledge, decisional conflict, and trust, and satisfaction ....................................................... 22 Patient participation and acceptability ................................................................................................... 22 Clinician acceptability ............................................................................................................................. 22 Management and 30-day outcomes ....................................................................................................... 23 Healthcare utilization .............................................................................................................................. 25

DISCUSSION ..................................................................................................................................... 34 Principal findings ..................................................................................................................................... 34 Meaning of the study .............................................................................................................................. 34 Limitations and strengths of the study ................................................................................................... 35 Unanswered questions and future research .......................................................................................... 36

CONCLUSION ................................................................................................................................... 37 REFERENCES .................................................................................................................................... 39

3

ABSTRACT

Background: Patients at low risk for acute coronary syndrome (ACS) are frequently admitted for

observation and cardiac testing at a substantial burden and cost to the patient and the health care

system. We compared the effectiveness of shared decision making facilitated by the Chest Pain Choice

(CPC) decision aid with usual care (UC) in the choice of admission for observation and further cardiac

testing or referral for outpatient evaluation in patients with possible ACS.

Methods: This was a multicenter pragmatic parallel randomized controlled trial conducted in 6

geographically diverse US emergency departments. Participants included adults (> 17 years of age) with

a primary complaint of chest pain who were being considered for observation unit admission for cardiac

testing. Patients were randomly assigned (1:1) to CPC or to UC. The primary outcome, selected by

patient and caregiver representatives, was patient knowledge; secondary outcomes were involvement

in the decision to be admitted, proportion of patients admitted for cardiac testing, the 30-day rate of

major adverse cardiac events, and 45-day health care utilization. We also conducted a prespecified

analysis to assess the heterogeneity of effect of CPC in potentially vulnerable patient groups.

Results: We assessed 3236 patients for eligibility and enrolled 898 patients (451 CPC, 447 UC) from

October 2013 to August 2015. Compared with UC, CPC patients had greater knowledge (questions

correct: 4.2 CPS versus 3.6 UC; mean difference [MD] 0.66; 95% confidence interval [CI], 0.46, 0.86),

were more involved in the decision to be admitted (observing patient involvement [OPTION] scores:

18.3 CPC versus 7.9 UC; MD 10.3; 95% CI, 9.1, 11.5), and less frequently decided with their clinician to be

admitted for cardiac testing (37% CPS versus 52% UC; absolute difference [AD] 15%; p < 0.001). There

was no difference in the emergency department (ED) length of stay (LOS) between the CPC and UC

arms, although CPC patients had a significantly shorter median LOS in the ED observation unit (824.5 [SD

= 517.5] minutes versus 920.2 [SD = 482.4] minutes; p = 0.0305) and underwent fewer tests within 45

days (mean [SD] 5.6 [5.4] CPC versus 6.4 [5.8] UC; p = 0.0465). No major adverse cardiac events occurred

due to the intervention. When assessing the effect of the decision aid in potentially vulnerable patient

groups, we observed similar effectiveness of the decision aid to the trial population in the elderly and in

those with lower levels of education and less income on the outcomes of patient knowledge, decisional

conflict, and involvement in the decision (p for interaction = ns). CPC increased knowledge to a greater

degree in whites compared with nonwhites (11.0% versus 4.8%, AD 6.2%; p for interaction = 0.018) and

in patients with typical numeracy compared with those with low numeracy (10.6% versus 4.7%, AD

5.9%; p for interaction = 0.025). However, CPC increased physician trust to a greater degree in patients

4

with “low” health literacy compared with those with “typical” literacy (3.7% versus –1.4%, AD 5.1%; p

for interaction = 0.011).

Conclusions: Use of CPC in patients at low risk for ACS increased patient knowledge and engagement

and safely decreased health care utilization. All subgroups benefited to a similar extent from use of CPC;

white patients and those self-reporting better numeracy had greater knowledge gains, while physician

trust increased more in patients with low health literacy.

BACKGROUND

Chest pain is the second most common reason patients visit emergency departments (EDs) for

evaluation, accounting for more than 8 million visits annually.1 Over the past decade, the proportion of

patients diagnosed with acute coronary syndrome (ACS) in the emergency setting decreased from 26%

to 13%. Despite the decreasing incidence of ACS, advanced cardiac imaging for chest pain has increased

nearly 4-fold.2

Current clinical, electrocardiogram (ECG), and laboratory data do not identify all patients who present to

the ED with ACS, resulting in a 1.5% miss rate.3 Given the medical, legal, and psychological sequelae

associated with missing a diagnosis of ACS, clinicians have a very low threshold to admit patients for

prolonged observation and advanced cardiac testing.4 As a consequence, low-risk patients are

frequently admitted for observation and cardiac stress testing or coronary computed tomography

angiography (CCTA). This results in unnecessary hospital admissions, false positive test results, and

unnecessary invasive downstream investigations, at an estimated cost to the health care system of more

than 7 billion US dollars annually.5

Decision aids are patient-centered tools designed to facilitate shared decision making between a patient

and his or her clinician such that patients’ values and preferences are incorporated into health care

decisions.6 To assist clinicians and patients with possible ACS in making risk-informed decisions about

testing and follow-up and to engage patients in the decision-making process, we included validated7 8

45-day risk estimates for ACS into a decision aid, Chest Pain Choice (CPC).9 In a single-center pilot

randomized trial of CPC, we observed increased patient knowledge, increased patient engagement,

decreased decisional conflict, and a 19% lower rate of observation unit admission for cardiac stress

testing in the CPC arm compared with usual care (UC), with no adverse events in either study arm.10 We

conducted this pilot randomized trial in a single tertiary care academic ED in the Midwest United States.

5

In order to test the effectiveness of CPC in a broader population of patients with greater socioeconomic

diversity and in a variety of clinical contexts, we conducted a multicenter pragmatic11 randomized trial in

6 geographically diverse EDs across the United States.

Stakeholder involvement

Patients seeking emergency care for chest pain, a patient representative (MD), and a caregiver

representative (AL) were involved in the design of the study, design of the intervention, submission of

the application for funding, monitoring of study conduct, interpretation of the data, review of the

manuscript for important intellectual content, and approval of the final manuscript for publication.

When designing the trial, the patient representative, the caregiver representative, and the ED patient

advisory council at the Saint Mary’s Hospital at Mayo Clinic provided input regarding the prioritization

and selection of outcomes. As the primary purpose of the study was to educate and empower patients

to participate in decisions regarding their emergency care, we prioritized the patient’s viewpoint over

outcomes of potential interest to other stakeholders. We included outcomes of interest to other key

stakeholders as secondary outcomes. When designing the intervention, we sought input from the

patient and caregiver representatives, the ED patient advisory council, and patients receiving emergency

care for potential ACS regarding the clarity, helpfulness, and usefulness of the information included in

the CPC, and the CPC was iteratively refined based on this input. As the patients and patient and

caregiver representatives involved in the trial had no prior diagnosis of coronary artery disease and thus

no engagement in a heart disease–specific support group or organization, patient representatives were

not directly involved in dissemination of the study findings. However, the patient and caregiver

representatives were engaged at the highest level possible—partner—and included as coinvestigators

on the application for funding, members of the investigative steering committee, and coauthors in the

manuscripts generated from this research.

METHODS Study design and setting

The background and methods of the trial were described in a previous publication.12 This was a

pragmatic parallel randomized controlled trial in low-risk patients presenting to the ED with a potential

ACS. The trial compared an intervention group receiving a structured risk assessment using a

quantitative pretest probability Web tool13 incorporated into a decision aid (CPC) with a control group

receiving UC.14 We did not believe cross-contamination would be a problem because access to the

6

quantitative pretest probability instrument was password protected and not easily reproduced by

participating clinicians, and we observed in our pilot trial that clinicians reverted to their usual pattern of

interacting with patients without the decision aid. For these reasons we opted to randomize at the

patient level. The Institutional Review Boards (IRBs) at each of the participating hospitals approved all

study procedures. The IRB-approved study protocol is included as a supplementary file with this report.

We enrolled patients and clinicians from the EDs at 6 US sites (University of California Davis, Mayo Clinic

Rochester, Indiana University, University of Pennsylvania, Thomas Jefferson University, and Mayo Clinic

Florida). Each of the sites, with the exception of Mayo Clinic Florida, had access to an emergency

department observation unit in which protocols to provide care for patients with potential acute

coronary syndrome existed as part of routine practice. At Mayo Clinic Florida, when a decision is made

to admit a patient for observation, the patient’s status is changed from an emergency department visit

to an observation stay without a physical change in location. Otherwise we collected all data

consistently across sites.

Participants

We included all physicians, nurse practitioners, and physician assistants caring for patients with chest

pain. Eligible patients included adults (> 17 years of age) who presented to the ED with a chief complaint

of chest pain and were being considered by the treating clinician for observation unit admission for

cardiac stress testing or CCTA. Registration staff identified adults with a chief complaint of chest pain on

arrival to each participating ED, and study coordinators worked with the treating clinician to assess

eligibility. We excluded patients for the following reasons: ischemic changes on the initial ECG (e.g., ST-

segment depression, T-wave inversion, or new left bundle branch block), initial cardiac troponin of less

than the 99th percentile, known coronary artery disease, cocaine use in the past 72 hours (confirmed by

history or testing), prior plan for cardiac intervention or admission, barriers to outpatient follow-up,

primary institution of care other than the enrolling institution, prisoners, pregnancy, hearing or visual

impairment or other inability to use the CPC. We classified as postrandomization exclusions those

patients deemed to meet the exclusion criteria after randomization but before the patient–clinician

disposition discussion.15

Randomization and masking

7

An online password-protected randomization algorithm (Medidata BalanceTM [computer program] New

York City, NY: Medidata Solutions concealed allocation). We randomized patients 1:1 and dynamically

stratified16 them by age, gender, and site because of the known associations of age and gender with

cardiovascular risk, potential unmeasured differences between sites, and availability of these data at the

time of enrollment. We did not randomize clinicians. We did not mask to allocation patients, study

coordinators, and treating clinicians. We blinded all other investigators to allocation. We blinded study

coordinators to allocation at the time of the 45-day phone call.

Study treatments

Intervention

We sought to assist patients and clinicians in making a risk-informed shared decision in the emergency

setting, in which patients typically do not have the opportunity to learn about their condition prior to

the visit and clinicians frequently make decisions unilaterally to facilitate patient safety and rapid

treatment of life-threatening conditions. For these contextually specific reasons, we designed CPC for

use during the clinical encounter.17 CPC was developed9 in Rochester, Minnesota, through a

participatory action research methodology18 in which feedback was intentionally and iteratively sought

from patients, clinicians, an expert in health care design, and the investigative team and field-tested

until thematic saturation was achieved. Prior to conducting the trial, we refined the CPC to ensure

contextual fit with each practice setting. Figure 1 depicts the refined decision aid. At 2 of the sites, CCTA

was available and frequently obtained in the evaluation of patients with possible ACS. For these 2 sites,

we added the option of CCTA to the CPC. Figure 2 displays the decision aid that includes this option.

8

Figure 1: The Chest Pain Choice Decision Aid Decision aid used to facilitate a discussion between clinicians and patients regarding whether to be admitted to the emergency department observation unit for cardiac stress testing or to follow up with a clinician in 24-72 hours.

9

Delivery of the intervention

For patients randomized to CPC, a study coordinator collected each of the variables needed to populate

the quantitative pretest probability Web tool,13 asked the treating clinician to sign off on their accuracy,

and calculated the patient’s pretest probability of acute coronary syndrome, incorporating the result of

the first troponin test but prior to subsequent biomarker testing. Next, the study coordinator selected

the decision aid that depicted the level of risk corresponding to the pretest probability generated by the

Web tool. The study coordinator then offered to provide the clinician a concise refresher of the content.

The treating clinician, after evaluating the patient and receiving the results of the initial ECG and cardiac

troponin, then brought the CPC decision aid to the patient’s bedside and used it to educate the patient

about the results of the 2 tests; the potential need for observation and further cardiac testing;

subsequent cardiac troponin testing to definitively rule out acute myocardial infarction, if required; and

Figure 2: Chest Pain Choice Decision Aid that includes the option of coronary computed tomography angiography (CCTA). Decision aid used to facilitate a discussion between clinicians and patients regarding whether to be admitted to the emergency department observation unit for cardiac stress testing or CCTA or to follow up with a clinician in 24-72 hours.

10

his or her personalized 45-day risk for ACS. The clinician then engaged the patient to select the

management option most closely aligned with his or her values and preferences.

Usual care

For patients randomized to UC, a study coordinator instructed the clinician to discuss the results of

diagnostic investigations and management options in his or her usual way. When clinicians treated

patients who were randomized to UC, they did not have access to the quantitative pretest probability

Web tool or to CPC. As the trial was intentionally pragmatic in design, UC was not standardized.11

Data collection

We collected data documenting the screening process, randomization, and outcome assessment in

compliance with CONSORT guidelines.19 An immediate post visit survey collected data on patients’

knowledge regarding their risk for ACS and the available management options, decisional conflict, and

trust in their physician.12 The clinician–patient discussion was video and audio recorded to assess, using

the validated OPTION scale, the degree to which the clinician made efforts to engage the patient in the

decision-making process.20 Video and audio recordings were time stamped, and we determined the

duration of the clinician–patient discussion from these recordings. The recordings were uploaded to a

secure server and deleted from the portable devices after upload. The server was protected by 2-step

access: (1) password-protected access to all the Mayo clinic’s computers, and (2) password-protected

access to the secure server. Audio and video files from facilities outside of Mayo Clinic were downloaded

onto an encrypted password-protected flash drive, sent securely to the prime site by FedEx, and

uploaded to a secure server on receipt. We also collected data on cardiac risk factors, post-ED

management, and further cardiac investigations by reviewing the electronic medical record (EMR) at

each site.

We collected patient health care utilization data using 2 separate methods to ensure we had the most

complete data for each patient. Each patient filled out a health care diary documenting his or her

utilization of health care services in the 45 days after the ED visit. Study coordinators, blinded to the arm

to which the patient was randomized, contacted patients starting at 45 days after enrollment to query

patients about this diary and any additional testing, outpatient or inpatient visits, or procedures at

health care facilities other than the original institution. Study coordinators made at least 5 attempts to

contact each patient by phone for follow-up during different times of the day and on different days of

11

the week. If the patient was unable to be reached by phone or email and the EMR documented no

subsequent visits, mortality status was verified using Accurint, a national database frequently used by

banks and other businesses to track individuals and ensure payment collection.21 We included in the

analysis all other data collected on patients unable to be reached by phone. In addition, each site

provided all the billing data for each of the enrolled patients for the initial ED visit and the 45 days after

the initial visit. .

To facilitate assessment of the heterogeneity of CPC effect in vulnerable patients, we collected data on

patient age, gender, race, annual income, insurance status, education level, health literacy, and

numeracy. To collect data on race, we used the categories recommended by the Institute of Medicine

(American Indian/Alaska Native, Asian, Black or African American, Native Hawaiian or other Pacific

Islander, White/Caucasian, Other).22 To assess health literacy, patients answered 3 questions, each with

a 5-point Likert response (Subjective Literacy Scale),23 prior to their encounter with the clinician. For the

purposes of this study, we summed the 3 items into a total score after reverse coding 1 item: The higher

the summed score, the lower the patient’s subjective assessment of his or her general health literacy

skills. To assess numeracy, which quantifies the ability to understand and use numbers in daily life,24

patients completed an 8-item questionnaire (Subjective Numeracy Scale).25 We reversed and averaged

numeracy responses to all 8 questions, creating an overall score ranging from 1 to 6, where higher

scores are indicative of lower levels of numeracy.

Outcomes

Primary outcome

Because the goal of patient-centered outcomes research is to provide patients and the public with the

information they need to help them make decisions that affect their desired health outcomes,26 we

prioritized the perspective of the patient over the perspective of other stakeholders in determining the

primary outcome. Because knowledge emerged as the outcome of greatest importance during meetings

with patient and caregiver representatives, we selected patient knowledge as the primary outcome. As

in our pilot trial10 and in prior work,27 we assessed patient knowledge with an immediate postvisit survey

(Figure 3). Each question was classified as true or false, with the exception of Question 9, which was

open ended and asked the patient to provide a numerical answer.

12

Figure 3. Nine Knowledge Questions Included in the Post-Visit Patient Questionnaire*

Questions used to assess patient knowledge immediately after the emergency department encounter.

*Questions 1-8 included the answer options “True,” “False,” and “Unsure.” Question 9 was an open-ended question with no suggested answers.

Secondary outcomes

We measured how informed patients felt about their management options using the decisional conflict

scale28 and patients’ trust in their clinician using the Trust in Physician Scale.29 The decisional conflict

scale includes 16 items scored from 0 to 4; the items are summed, divided by 16, and then multiplied by

25. The scale is from 0 to 100, where higher scores reflect increased patient uncertainty about choice.

One study found that for every unit increase in decisional conflict scale scores, patients were 19% more

likely to blame their doctor for bad outcomes.30 The Trust in Physician scale consists of 9 items scored

from 1 to 5; the items are subtracted by 1, summed, divided by 9, and then multiplied by 25. The scale

ranges from 0 to 100, where higher values reflect higher values of patient trust in his or her physician.

To the best of our knowledge, a clinically meaningful change in Trust in Physician scale has not been

published. We surveyed participating patients and clinicians regarding the clarity and helpfulness of the

information shared and the acceptability of the decision aid using a 7-point Likert scale. Finally, 5 trained

raters independently viewed videos of the patient–clinician discussion and assessed the degree to which

clinicians engaged patients in the decision-making process using the OPTION scale.20 This scale is

composed of 12 items with a value of 0 to 4; they are summed, divided by 48, and then multiplied by

100. Scores range from 0 to 100, where higher scores reflect higher levels of patient engagement.

Although a clinically meaningful change in OPTION scale score has not been defined, the mean score for

1. My current chest pain may indicate possible warning signs of a FUTURE heart attack. 2. I can follow-up with my own doctor at the next available visit and decide with him or her whether I need additional tests. 3. I can go home and have a follow-up visit with a heart specialist (cardiologist) within 24 to 72 hours for additional tests. 4. A stress test evaluation will tell me my risk of having a heart attack now. 5. The result of a stress test evaluation can be a false positive and lead to additional testing that is unnecessary. 6. Having a stress test now will not lengthen my emergency stay. 7. None of the stress tests involve exposure to radiation. 8. Radiation may increase my future risk of cancer. 9. If you take 100 people just like you, how many do you think will develop a heart attack or pre-heart attack within the next 45 days?

13

outpatient clinicians in the original development investigation was 16.9 (standard deviation [SD] 7.68).31

Given that we conducted the current trial in the emergency setting, in which time pressures and patient

acuity often impact the clinician–patient interaction, we anticipated OPTION scale scores in the current

investigation to be lower than the originally published mean.

We assessed management by recording whether patients were admitted to the ED observation unit or

hospital, or discharged home; whether cardiac stress testing or CCTA was conducted; the results of

testing; and whether the patient underwent percutaneous coronary intervention or coronary artery

bypass grafting by review of the EMR at each participating site.

To assess safety, we determined whether a patient experienced a major adverse cardiac event (MACE).

Consistent with a consensus document on ACS research in EDs,32 we defined MACE as acute myocardial

infarction (AMI),33 death due to a cardiac or unknown cause, emergency revascularization, ventricular

arrhythmia, or cardiogenic shock. Potential MACEs were shared with the data safety monitoring board

(DSMB), which reviewed the cases in detail and provided its judgment to the investigative team. The

entire investigative team also discussed potential MACE cases on monthly conference calls and

adjudicated the cases based on consensus among site investigators. We excluded MACEs occurring

during the index visit to the ED or hospital, as these events were considered appropriately diagnosed

during the index visit. Events that occurred after discharge home, which could have potentially been

avoided, were classified as MACE. We collected data on all MACE occurring up to 45 days to be

consistent with the follow-up period used in the development of the quantitative pretest probability

instrument;34 but we compared 30-day event rates to be consistent with standardized reporting

guidelines for ED risk stratification studies of patients with potential ACS.35

For health care utilization, we assessed using both patient-reported data and hospital-level billing data

on the following health care services delivered during the ED visit and the subsequent 45 days:

hospitalizations, emergency department visits, office visits, testing, imaging, and other services. Patient-

reported data provided information on subsequent hospitalizations, emergency department visits, and

physician office visits and the number of visits for each. The billing data provided information on testing

and imaging during the initial visit and subsequent utilization in the next 45 days. We classified all

utilization from the billing data using the Berenson-Eggers Types of Service (BETOS) codes.36 We used

BETOS codes to separate out evaluation and management visits, imaging, testing, and procedures. In

14

addition, we separated out inpatient and emergency department visits. Further, we looked at the

specific types of testing and imaging, including stress testing and CCTA. Finally, we assessed the amount

of time spent in the ED and observation unit in each of the arms. This allowed us to assess whether the

intervention increased the amount of time spent in the ED or the observation unit. Although in the

original trial protocol12 we anticipated reporting 30-day health care utilization, we opted to report 45-

day health care utilization instead given the slightly more robust inferences that can be drawn using a

longer follow-up interval to assess utilization.

Clarification of primary outcome

The primary outcome registered at clinicaltrials.gov is the phrase “Test if Chest Pain Choice [the

decision aid] safely improves validated patient-centered outcome measures,” with the description “Test

if the intervention significantly increases patient knowledge.” There is only 1 primary outcome for the

study: patient knowledge. The phrase “Test if Chest Pain Choice safely improves validated patient-

centered outcome measures” refers to the 5 additional outcome measures (a through e) listed as

secondary outcomes at clinicaltrials.gov. This is documented in the study protocol,12 which was

published prior to completion of enrollment for the trial in August 2015.

Statistical analysis

We estimated that 884 patients would provide 99% power to detect a 16% difference in patient

knowledge between intervention and control arms and 90% power to detect a 10% difference in the

proportion of patients admitted to an observation unit for cardiac testing.12 To account for an estimated

5% potential loss to follow-up, we planned to enroll 930 patients. We summarized patient

characteristics by study group and tested for differences between groups using t-tests and chi-square

tests. To test for differences in outcomes, we estimated a series of regression models, each of which

included indicators for study group. For continuous outcomes we used linear models, and for categorical

outcomes we used multinomial (polytomous) logistic models. To account for nonindependence of

outcomes by site, we included indicators for study site in each model. We assessed for additional

correlation with clinicians by estimating a hierarchical generalized model for each outcome and

calculating the intraclinician correlation (ICC). Because all ICCs were less than 1%, we chose not to

account for this correlation in the final models.

15

Health care utilization

We tested for differences in both self-reported and billing-based utilization between groups using t-

tests, chi-square tests, or the Kruskal-Wallis test, as appropriate. To test for differences in outcomes, we

estimated a series of regression models, each of which included indicators for study group. As with the

primary analysis of the trial,37 we conducted outcome assessments using regression models (linear for

continuous outcomes, multinomial for categorical outcomes) that included indicators for study arm

assignment and study site. We used appropriate count data models for each of the utilization outcomes.

For outcomes that had less than 10% of the patients with zero utilization, we used Poisson or negative

binomial models depending on the extent of overdispersion. Among outcomes with more than 10% of

patients with no utilization, we used a 2-part model, sometimes referred to as a hurdle model,38 which

separates the utilization outcomes into 2 distinct parts: whether there was any utilization of that

service, and amount of utilization of that service conditional on any utilization. More specifically, the

first hurdle used a logistic model to estimate the probability of having any utilization for a type of

service, while the second hurdle used Poisson models to estimate the extent of utilization. We

conducted all analyses for the baseline utilization, utilization in the 45 days after the index visit, and

total utilization. To account for nonindependence of outcomes by site, we included indicators for study

site in each model.

Heterogeneity of CPC effect

In a previously published patient-level metaanalysis of 7 randomized trials assessing the efficacy of

patient DAs used during the clinical encounter,39 we observed efficacy estimates similar to the trial

population when DAs were used with vulnerable patients such as the elderly and those with less income

and less formal education. We also observed, however, a greater increase in knowledge of risk among

patients with higher education compared with those with lower levels of education. Subgroup effects

based on race yielded imprecise results with wide confidence intervals. Based on the findings in the

meta-analysis, we hypothesized that use of CPC would increase knowledge, decrease decisional conflict,

and increase patient engagement across all sociodemographic patient groups but would also increase

knowledge to a relatively greater degree in patients with higher levels of education, health literacy, and

numeracy. Moreover, given empiric data demonstrating high levels of physician trust in a cohort

consisting largely of Caucasian patients40 and less positive perceptions of physicians in patients from

racial and ethnic minority groups,41 we hypothesized that use of CPC would increase physician trust to a

16

greater degree in patients from racial and ethnic minority groups. We designed the heterogeneity of

CPC effect analyses to test these prespecified hypotheses.

For the analysis, we dichotomized each of the following variables to evaluate the differential effect of

CPC across patient groups: age, gender, race, annual income, insurance status, level of education, health

literacy, and numeracy. We chose to dichotomize for 2 reasons: (1) Given the overall enrollment of 898

patients, more than 2 classifications would have resulted in many subgroups that were too small to

analyze; and (2) binary classifications made the heterogeneity of CPC effect simpler to analyze and

interpret by way of interactions.42 We dichotomized gender naturally. Other classifications were as

follows: race as white versus nonwhite (to facilitate comparison of CPC effects between minorities and

the largest nonminority group); annual income as less than $40 000 versus greater than or equal to $40

000; insurance status as uninsured versus insured; education as less than or equal to high

school/Graduate Education Diploma (GED) versus greater than high school/Graduate Education

Diploma; literacy as typical (≤ 3) versus low (> 3); and numeracy as typical (≤ 4) versus low (> 4). We

based these classifications on both the distribution of the values and conceptual considerations

regarding the most likely contrasts to show heterogeneity of effect. We excluded patients missing a

subgroup variable from the analysis for that subgroup. For race, we included the “other” group with

“nonwhite,” as we interpreted “other” as implying not white. For education, we excluded the “other”

category from the dichotomous groups, as we could not assume it to indicate either of the 2 categories.

As with the primary analysis of the trial,37 we conducted outcome assessments for this study using

regression models (linear for continuous outcomes, multinomial for categorical outcomes) that included

indicators for study arm assignment and study site. In addition, to assess for heterogeneity of effect

across each of the subgroups, we included an interaction term for group assignment and subgroup. To

improve interpretation, we also replicated the main trial analysis (i.e., with no interaction term) within

each subgroup, and reported the group effect and whether the effect differed significantly from zero.

We reported this group effect as a coefficient for continuous outcomes and odds ratios for dichotomous

or multinomial outcomes.

Our statistical approach to subgroup analysis was informed by publication guidelines for reporting

subgroup analyses.42 We prespecified interaction testing between patient characteristics and the

outcomes of patient knowledge, decisional conflict, involvement in decision making, and physician trust,

and we used a significance level of 5% to identify significant interactions for these subgroup effects.

17

Given that a total of 80 comparisons were performed, we anticipated up to 0.05 multiplied by 80, or 4,

significant interactions based on chance alone. As such, we considered subgroup analyses that were not

prespecified to be hypothesis generating. We performed all analyses using Stata Version 14.1 (College

Station, TX: Stata Corporation; 2016). We followed the principle of intention to treat in the conduct of

the trial and in all analyses.

RESULTS

Main trial

We assessed 3236 patients for eligibility from October 2013 to August 2015 (Figure 4), and 361 clinicians

participated in the study. We randomized 913 patients, with 13 postrandomization exclusions and 2

patients who withdrew consent, leaving 898 patients (451 CPC, 447 UC) in the final analysis. In all 13

postrandomization exclusions, additional information became available after randomization, but before

the patient–clinician disposition discussion, indicating that the patient was not eligible. We audio or

video recorded the patient–clinician disposition discussion in 536 (59.7%) encounters. The main reasons

recordings were not obtained were clinician and patient refusal and technical difficulties with recording

equipment. We contacted 828 (92.2%, 413 CPC) patients by telephone and/or email for follow-up. Of

the 70 (7.8%) remaining patients, 68 had mortality data available in the EMR and/or Accurint,21 which

confirmed that none of these 68 patients died within 45 days. The 2 patients with missing mortality data

were in the UC arm.

Table 1 summarizes patient characteristics. Mean age (SD) was 50.3 (14.5) years, and 534 (59.5%) were

women. Most patients were white (531, 59.1%) or black (309, 34.4%). There were 285 (31.7%) patients

whose highest level of education was high school, GED, or less. The mean (SD) pretest probability of ACS

in the usual care arm was 3.8 (4.3) and 3.6 (3.7) in the decision aid arm (p = 0.46). There were no

significant differences in baseline characteristics between the study arms.

18

Table 1. Baseline Characteristics

Characteristic Usual care (n=447)

Decision aid (n=451)

Age, mean (SD) median (IQR)

50.6 (14.1) 51.0 (44.0, 59.0)

50.0 (15.0) 51.0 (43.0, 58.0)

Female 260 (58.2) 274 (60.8)

Race

American Indian/Alaska Native 4 (0.9) 4 (0.9)

Asian 9 (2.0) 6 (1.3)

Black or African American 154 (34.5) 155 (34.4)

Native Hawaiian or other Pacific Islander 0 (0.0) 2 (0.4)

White/Caucasian 269 (60.2) 262 (58.1)

Other 11 (2.5) 22 (4.9)

Annual income (n=850) Less than $20,000 20-30 30-40 40-60 60-80 80-100 100+

84 (18.8) 39 (8.7)

46 (10.3) 65 (14.5) 59 (13.2) 37 (8.3)

93 (20.8)

92 (20.4) 36 (8.0) 44 (9.8)

56 (12.4) 58 (12.9) 50 (11.1)

100 (22.2)

Highest level of education completed (n= 881)

Some high school or less High school or GED Some College or Vocational School College Graduate (4 year) Graduate Degree

38 (8.5) 109 (24.4) 157 (35.1) 82 (18.3) 54 (12.1)

47 (10.4) 91 (20.2)

150 (33.3) 98 (21.7) 55 (12.2)

Literacy screening questions (often/always)

Need help reading medical instructions? 36 (8.1) 44 (9.8)

Confident filling out medical forms? 358 (80.1) 375 (83.1)

Have difficulty understanding written information about your medical condition?

28 (6.3) 40 (8.9)

Numeracy, mean (SD) 4.3 (1.1) 4.2 (1.1)

median (IQR) 4.4 [3.5, 5.1] 4.4 [3.4, 5.1]

Hypertension 202 (45.1) 198 (43.9)

Dyslipidemia 137 (30.6) 114 (25.3)

Diabetes Mellitus 71 (15.8) 61 (13.5)

Family history of cardiac disease 182 (40.6) 176 (39.0) Smoking (current, recent cessation, or former) 165 (36.8) 181 (40.1) Renal insufficiency 9 (2.0) 7 (1.6)

History of stroke or transient ischemic attack 16 (3.6) 18 (4.0)

Chest pain duration in hours, mean (SD) 3.3 (5.5) 3.1 (5.0)

median (IQR) 1.0 [0.3, 4.0] 1.0 [0.2, 3.5]

19

Pre-test probability of ACS, mean (SD) 3.8 (4.3) 3.6 (3.7)

median (IQR) 2.8 [0.6, 5.2] 2.8 [0.6, 5.2]

Friend or family member present 244 (54.6) 257 (57.0)

Healthcare literacy Mean (SD)

Median (IQR) Missing

5.1 (2.4)

4.0 [3.0-7.0] 5 (1.1)

5.2 (2.6)

4.0 [3.0-7.0] 2 (0.4)

Numeracy

Mean (SD) Median (IQR)

4.3 (1.1) 4.4 [3.5-5.1]

4.2 (1.1) 4.4 [3.4-5.1]

Data are presented as n (%) unless otherwise indicated. ACS, acute coronary syndrome; GED, graduate education diploma; IQR, interquartile range; SD, standard deviation. *Unless otherwise specified.

20

Table 2. Effect of Decision Aid on Patient knowledge, Decisional conflict, Trust in the physician, Patient

Involvement in the Decision, and Decision Aid Acceptability

Outcome Usual Care (n=447)

Decision Aid (n=451)

Mean difference (95% CI) or P-value

Patient knowledge

Eight knowledge questions 3.6 (1.5) 4.2 (1.5) 0.66 (0.46, 0.86)

Correctly assessed 45-day risk for ACS 2 (0.4) 10 (2.2) 0.039

Correctly assessed 45-day risk for ACS within 10%

81 (18.1) 293 (65.0)

<0.001

Decisional Conflict and Trust

Decisional conflict scale 46.4 (14.8) 43.5 (15.3) -2.9 (-4.8,-0.90)

Trust in physician scale 87.7 (16.0) 89.5 (13.4) 1.7 (-0.2, 3.6)

Patient Involvement in the Decision

OPTION scale (n=536) 7.9 (5.4) 18.3 (9.4) 10.3 (9.1,11.5)

Acceptability

Patient Amount of information

Too little (1-2) Just right (3-5) Too much (6-7)

24 (5.5)

401 (91.6) 13 (3.0)

12 (2.7)

416 (94.3) 13 (2.9)

0.133

Clarity of information Not clear at all (1-2) Somewhat clear (3-5) Extremely Clear (6-7)

5 (1.1)

137 (31.3) 296 (67.6)

7 (1.6)

98 (22.3) 335 (76.1)

0.011

Helpfulness of the information Not helpful at all (1-2) Somewhat helpful (3-5) Extremely Helpful (6-7)

10 (2.3)

125 (28.5) 303 (69.2)

7 (1.6)

114 (25.9) 320 (72.6)

0.506

Would recommend to others Yes (1-2) Not Sure (3-5) No (6-7)

349 (79.9) 77 (17.6) 11 (2.5)

387 (88.0) 44 (10.0)

9 (2.0)

0.004

Would want to use for other decisions

Yes (1-2) Not Sure (3-5) No (6-7)

335 (76.7) 83 (19.0) 19 (4.3)

346 (78.6) 77 (17.5) 17 (3.9)

0.813

Clinician

21

Helpfulness of the information Not helpful at all (1-2) Somewhat helpful (3-5) Extremely Helpful (6-7)

13 (3.1)

265 (63.2) 141 (33.7)

24 (5.5)

175 (40.3) 235 (54.1)

<0.001

Would recommend to others: Yes (1-2) Not Sure (3-5) No (6-7)

175 (41.9) 234 (56.0)

9 (2.2)

271 (62.7) 148 (34.3)

13 (3.0)

<0.001

Would want to use for other decisions

Yes (1-2) Not Sure (3-5) No (6-7)

183 (43.8) 229 (54.8)

6 (1.4)

273 (62.9) 148 (34.1)

13 (3.0)

<0.001

Data are presented as n (%) or mean (standard deviation) unless otherwise indicated. ACS, acute coronary syndrome; CI, Confidence Interval.

22

Patient knowledge, decisional conflict, trust, and satisfaction As Table 2 depicts, patients randomized to CPC had greater knowledge (questions correct: 4.2 CPC

versus 3.6 UC; mean difference [MD] 0.66; 95% confidence interval [CI], 0.46, 0.86). A greater

proportion of patients in the intervention arm correctly reported their exact pretest probability of ACS

and their risk within 10% of the correct value (65.0% CPC versus 18.1% UC; absolute difference [AD]

46.8%; 95% CI, 41.2, 52.5). Patients in the intervention arm reported significantly less decisional conflict

(Decisional Conflict Scale: 43.5 [15.3] CPC versus 46.4 [14.8] UC; MD –2.9; 95% CI, –4.8, –0.90). Use of

CPC did not significantly impact patients’ trust in their physician. The proportion of patients who were

“strongly satisfied” with the information shared with them by their clinician did not significantly differ

between the study arms (49% CPC versus 43% UC; absolute difference 6%; P = 0.06).

Patient participation and acceptability

Interobserver agreement (Cohen’s kappa) between raters for OPTION scale assessments was 0.89 (95%

CI 0.84 to 0.93). Patients randomized to CPC were more engaged in the decision-making process as

indicated by higher OPTION scores (18.3 [9.4] CPC versus 7.9 [5.4] UC; MD 10.3, 95% CI 9.1, 11.5; see

Table 2). Patients randomized to CPC found the information discussed to be of greater clarity, and a

greater proportion (88.0% CPC versus 79.9% UC; AD 8.1%; P = 0.004) would recommend to others the

way they discussed management options with their clinician.

Clinician acceptability

A greater proportion of clinicians in the intervention arm reported the information shared with their

patient about the decision to be admitted for further observation and testing to be extremely helpful

(235 [54.1] CPC versus 141 [33.7] UC; AD 20.4%; P = < 0.001; see Table 2). Most (62.7%) clinicians would

recommend using the CPC to others, and 62.9% would want to use a CPC for other decisions. The mean

(SD) length of the clinician–patient discussion was 1.3 minutes longer in the intervention arm (4.4 [0.40]

minutes CPC versus 3.1 [0.29] minutes UC; MD 1.3; P = 0.008).

23

Management and 30-day outcomes

A significantly lower proportion of patients randomized to CPC decided, with their clinician, to be

admitted to the ED observation unit for cardiac stress testing or CCTA (37.3% CPC versus 52.1% UC; AD

14.8%; 95% CI 1.1, 13.9; see Table 3). A significantly lower proportion of patients in the CPC arm

underwent cardiac stress testing within 30 days, including tests obtained during the index ED visit and

the subsequent 30 days (38.1% CPC versus 45.6% UC; AD 7.5%; 95% CI 1.1, 13.9). Of those who

underwent cardiac stress testing, a significantly greater proportion of patients randomized to CPC had

testing performed in the outpatient setting (30.2% CPC versus 17.2% UC; AD 13.1%; 95% CI 4.5, 21.7).

There was no difference in the rate of coronary angiography, coronary revascularization, hospitalization,

repeat hospitalization, repeat ED visits, or outpatient clinic visits between study arms.

Table 3. Management and 30-day outcomes Characteristic Usual care

(n=447) Decision aid

(n=451) P-value

Shared management decision <0.001 Observation unit admission for stress testing or coronary CTA

225 (52.1) 165 (37.3)

Follow up with a cardiologist 52 (12.0) 101 (22.9) Follow up with a primary care Physician 100 (23.1) 138 (31.2)

Have emergency physician decide 55 (12.7) 38 (8.6) Cardiac stress test performed within 30 days 204 (45.6) 172 (38.1) 0.013 Outpatient stress testing Stress test type

35 (17.2) 52 (30.2) 0.001 0.779

Exercise treadmill testing 65 (31.9) 44 (25.6) Stress echocardiography 86 (42.2) 81 (47.1) Nuclear perfusion testing 39 (19.1) 37 (21.5) Other 14 (6.9) 10 (5.8)

Coronary CT angiography performed within 30 days

80 (17.9) 63 (14.0) 0.111

Coronary revascularization 4 (0.9) 7 (1.6) 0.366 Percutaneous coronary intervention 3 (75.0) 6 (85.7) Coronary artery bypass grafting 1 (25.0) 1 (14.3)

Admitted to the hospital from the emergency department observation unit

22 (4.9) 22 (4.9) 0.990

Repeat emergency department visit 39 (9.3) 52 (12.5) 0.156 Rehospitalization 19 (4.5) 20 (4.8) 0.884 Outpatient clinic visit 259 (62.0) 266 (64.1) 0.568

24

Cardiac events Acute myocardial infarction 1 (0.2) 4 (0.9) 0.215 Death of cardiac or unknown cause 0 (0.0) 0 (0.0) 1.00 Major adverse cardiac event within 30 days*

0 (0.0) 1 (0.2)

0.998

ED, emergency department; CTA, computed tomographic angiography; ACS, acute coronary syndrome, NS, no significant difference. *Excluding the index event.

A total of 5 patients had AMIs, 4 in the CPC arm and 1 in the UC arm (p-value for difference = 1.0). The 1

patient in the UC arm with AMI had the AMI diagnosed during the index ED visit on serial cardiac

troponins and was admitted to the hospital. Four of the 5 patients with AMI and all cardiac interventions

occurred during the index visit. Two of the 4 patients with acute myocardial infarction in the CPC arm

had an initial troponin less than the 99th percentile, no acute ischemic changes on the initial ECG, and a

subsequent increased troponin level detected on serial cardiac biomarker testing. These patients were

admitted to the hospital for further evaluation and management and received a diagnosis of non-ST

segment elevation myocardial infarction. A third case of myocardial infarction in the CPC arm occurred

in a patient who had negative serial cardiac troponin results and no acute ischemic changes on the ECG

but symptoms suggestive of acute coronary syndrome. The patient was admitted to the hospital,

underwent percutaneous coronary intervention, and subsequently developed in-stent thrombosis. This

in-stent thrombosis, which occurred in the hospital, was accompanied by increased troponin levels and

ST-segment elevation on ECG. The patient underwent a second percutaneous coronary intervention and

recovered uneventfully. There were no deaths of cardiac or unknown cause in either arm. The fourth

patient in the CPC arm with acute myocardial infarction was classified as a MACE. This patient decided

with her clinician to be admitted to the hospital, she underwent nuclear perfusion stress testing as an

inpatient, and the test was interpreted as negative. She was discharged from the hospital. Subsequently,

she developed recurrent chest pain and returned to the ED within 30 days of hospital discharge with a

non-ST segment myocardial infarction and underwent percutaneous coronary intervention. This was the

only patient who had a cardiac intervention performed after the index ED admission; all the remaining

cardiac interventions occurred during the index ED admission. The DSMB classified this MACE as not

related to the intervention.

25

Health care utilization

Patient-reported utilization

Of the 898 patients, patient-reported health care utilization data were available for all 834 (92.9%)

patients who were reached by phone for 45-day follow-up. Based on the 8 patient-reported utilization

questions in the health care diary, there was no significant difference in 45-day health care utilization

between the study arms (Table 4).

Billing-based utilization

Hospital-level billing data were available for all 898 (100%) enrolled patients. Overall, including the date

of the ED visit and the subsequent 45 days, there was no difference in physician evaluation and

management codes (level of physician services), number of imaging tests obtained, or the number of

procedures performed between arms of the study. However, the mean (SD) number of tests was lower

in the CPC arm (13.3 [6.9] CPC versus 14.7 [7.7] UC; p = 0.0432; see Table 5). During the index ED visit,

the mean (SD) number of imaging tests was lower in the CPC arm (1.4 [1.4] CPC versus 1.6 [1.2] UC; p =

0.02), and there was no difference in the median length of stay (LOS). There was no difference in the

frequency of observation unit admission between arms; however, for patients who were admitted to

the ED observation unit, the median LOS was 95 minutes shorter in the CPC arm (824.8 minutes versus

920.2 minutes; p = 0.03; see Table 6). In the 45 days following the ED visit (not including the ED visit),

patients randomized to CPC underwent fewer tests overall (mean [SD] 5.6 [5.4] CPC versus 6.4 [5.8] UC;

p = 0.0465; see Table 5).

Table 4. Healthcare utilization within 45 days of the emergency department visit collected by patient report by use of a healthcare diary

Healthcare diary question Decision Aid

(N=418) Usual Care

(N=416) Total

(N=834) p value Since your initial Emergency Department visit, 0.8818 have you been re-admitted to the hospital? Yes 20 (4.8%) 19 (4.6%) 39 (4.7%) Since your initial Emergency Department visit, 0.3603 have you had an office visit with a physician? Yes 271 (64.8%) 257 (61.8%) 528 (63.3%) Since your initial Emergency Department visit, 0.1240 have you returned to the Emergency Department?

26

Healthcare diary question Decision Aid

(N=418) Usual Care

(N=416) Total

(N=834) p value Yes 52 (12.4%) 38 (9.1%) 90 (10.8%) Since your initial Emergency Department visit, 0.1110 have you had any blood testing, x-rays, CTs, or cardiac (heart) stress testing? Yes 167 (40.0%) 144 (34.6%) 311 (37.3%) Since your Emergency Department visit, did a 0.9959 clinician tell you that you had a myocardial infarction or heart attack? Missing 0 1 1 Yes 1 (0.2%) 1 (0.2%) 2 (0.2%) How many hospital visits does the patient list? 0.2622 N 20 19 39 Mean (SD) 1.1 (0.2) 1.2 (0.5) 1.1 (0.4) How many office visits does the patient list? 0.6811 N 271 257 528 Mean (SD) 1.6 (1.1) 1.6 (1.0) 1.6 (1.0) How many ED visits does the patient list?

N 52 37 89 0.6999 Mean (SD) 1.4 (0.7) 1.4 (0.7) 1.4 (0.7)

Table 5. Healthcare utilization data obtained from hospital-level billing data: index visit, subsequent 45-days, and total

Index ED visit

45-day follow-up Total

Utilization Category Decision Aid Usual Care

P-value Decision

Aid Usual Care

p-value Decision Aid Usual Care

P-value

Evaluation and Management (E&M) visits 1.3 (0.4) 1.4 (0.4) 0.3160* 2.1 (2.5) 2.0 (2.5) 0.6893 3.5 (2.6) 3.4 (2.5) 0.9344

Testing 7.7 (4.0) 8.1 (4.2) 0.1371* 5.6 (5.4) 6.4 (5.8) 0.0465 13.3 (6.9) 14.7 (7.7) 0.0432

Imaging 1.4 (0.3) 1.6 (0.3) 0.0202* 1.7 (2.1) 1.6 (2.0) 0.3446 3.1 (0.8) 3.2 (0.8) 0.4822

Procedures 0.5 (0.9) 0.4 (0.9)

0.6626* 0.6 (1.5) 0.7 (1.7) 0.3131* 1.1 (1.6) 1.2 (1.8) 0.3108*

Other (Ambulance, Chiro, Vision, Hearing, etc.) 0.4 (0.7) 0.3 (0.6) 0.0220* 0.8 (2.1) 1.0 (2.7) 0.1659* 1.2 (1.7) 1.4 (2.1) 0.1249*

*hurdle model used in analysis

Table 6. Chest Pain Specific Utilization

Decision Aid Usual Care P-value Stress Test (Total) 0.51 (1.1) 0.61 (1.2) 0.2563 Coronary CTA (Total) 0.27 (0.6) 0.33 (0.6) 0.1441 ED LOS (in minutes)* 354.7 (283.5) 350.2 (292.7) 0.8278 Admitted to Observation Unit, n(%)* 207 (45.9%) 228 (51.0%) 0.1256 Observation Unit LOS (in minutes)* 824.8 (517.5) 920.2 (482.4) 0.0305 LOS=length of stay *Unadjusted tests (since no zero counts)

Heterogeneity of CPC effect

There were significant interactions between patient characteristics and the outcomes of knowledge

(percentage of questions correct), knowledge of ACS risk, and trust in the physician (Table 7). For patient

knowledge (%), there were significant interactions based on race (p = 0.018) and numeracy (p = 0.025)

(Figure 5). Use of CPC increased knowledge in both the white and nonwhite subgroups, but to a greater

degree in the white subgroup (11.0%) compared with the nonwhite subgroup (4.8%), with the increase

in both subgroups reaching significance compared with UC. For numeracy, CPC increased knowledge

10.6% in the “typical” subgroup compared with 4.7% in the “low” subgroup, with both increases

reaching significance compared with UC. For knowledge of ACS risk, there was a significant interaction

with patient race (0.018) (Table 7). Use of CPC increased patient knowledge of ACS risk more in the

white subgroup (11.4%) compared with the nonwhite subgroup (7.2%), with the increase in the white

subgroup reaching significance compared with UC. Finally, for trust in the physician, we observed a

significant interaction with health literacy (p = 0.011) (Figure 6). Use of CPC increased trust in the

physician in patients with “low” health literacy (3.7%) while decreasing trust in the physician for patients

with “typical” literacy (–1.4%), with the increased trust reaching significance in the “low” subgroup

compared with UC. There were no significant interactions on based on decisional conflict or patient

involvement in the decision-making process (OPTION score).

29

Forest plot demonstrating the effect of the Chest Pain Choice decision aid on patient knowledge in subgroups according to patient characteristics.

30

Forest plot demonstrating the effect of the Chest Pain Choice decision aid on Trust in Physician scale scores in subgroups according to patient characteristics.

Table 7. Differential effect of the decision aid on patient knowledge, decisional conflict, trust in the physician, and involvement in the decision (OPTION Score) based on patient sociodemographic characteristics Characteristic Knowledge

(%correct) Knowledge of

ACS risk Decisional

Conflict Trust in

Physician OPTION Score

Decision Aid Effect

P value Decision Aid Effect

P value Decision Aid Effect

P value Decision Aid Effect

P value Decision Aid Effect

P value

Age 0.236 0.988 0.473 0.253 0.936 ≤ 50 years > 50 years

9.74 (1.75) 6.77 (1.83)

9.03 (2.12) 9.36 (2.14)

-3.79 (1.54) -2.18 (1.32)

2.92 (1.46) 0.58 (1.32)

10.31 (0.95) 10.23 (0.81)

Gender 0.928 0.08 0.439 0.679 0.42 Female Male

8.18 (1.65) 8.48 (1.99)

11.74 (2.64) 6.73 (1.65)

-3.47 (1.29) -2.06 (1.61)

2.05 (1.24) 1.19 (1.61)

10.68 (0.79) 9.67 (0.98)

Race 0.018 0.018 0.863 0.062 0.541 White Non-White

10.95 (1.63)* 4.79 (1.99)*

11.44(2.40)* 7.16 (1.94)*

-2.79 (1.22) -2.96 (1.75)

0.28 (1.08) 3.97 (1.83)

10.75 (0.75) 9.42 (1.10)

Annual income 0.317 0.519 0.366 0.371 0.832 < $40,000 ≥ $40,000

6.76 (2.07) 9.46 (1.66)

11.60 (3.64) 8.99 (1.86)

-1.05 (1.87) -3.14 (1.22)

2.43 (1.73) 0.89 (1.18)

10.06 (1.07) 10.25 (0.79)

Insurance 0.236 0.834 0.167 0.991 0.727 Uninsured Insured

3.76 (4.61) 8.72 (1.32)

14.31 (9.83) 8.96 (1.52)

-8.47 (3.65) -2.48 (1.06)

1.80 (4.37) 1.72 (1.00)

10.67 (2.88) 10.36 (0.63)

Level of Education 0.059 0.2 0.962 0.348 0.552 ≤ High school†

> High school†

5.17 (2.33) 10.31 (1.51)

17.14 (6.60) 9.00 (1.75)

-2.84 (1.87) -2.62 (1.21)

2.71 (1.97) 0.90 (1.11)

10.87 (1.03) 10.04 (0.77)

Literacy 0.095 0.626 0.829 0.011 0.812 Typical (3) Low (>3)

11.02 (2.12) 6.67 (1.55)

10.38 (2.77) 9.02 (1.94)

-2.47 (1.49) -3.07 (1.33)

-1.40 (1.41) 3.67 (1.31) *

10.47 (1.04) 10.17 (0.76)

Numeracy 0.025 0.282 0.454 0.09 0.885 Typical (≤4) Low (>4)

10.64 (1.63)* 4.65 (1.95)*

9.60 (2.00) 13.97 (4.50)

-2.47 (1.27) -3.97 (1.64)

0.61 (1.10) 3.85 (1.78)

10.38 (0.79) 10.20 (1.00)

P values are for an overall interaction between the DA and each characteristic. Indicates significant Decision Aid effect for each subgroup compared with its control (usual care) for the outcome – only provided for those outcomes for which there is an overall significant interaction (p<0.05) †Includes GED

Regarding management decisions, there were significant interactions between patient race (p = 0.004)

and annual income (p = 0.028) and the decision to have a stress test (Table 8). White patients had 52%

lower odds of having a stress test, while nonwhite patients had 19% greater odds of having a stress test,

with the decreased odds of having a stress test reaching significance in the white subgroup compared

with UC (Figure 7). For annual income, the group with income < $40 000 had 8% greater odds of having

a stress test, and those with income ≥ $40 000 or more had 46% lower odds of having a stress test, with

the decrease in the ≥ $40 000 income group reaching significance compared with UC. There were no

significant interactions on the decision to have a CT angiogram or to undergo revascularization.

Forest plot demonstrating the effect of the Chest Pain Choice decision aid on the rate of cardiac stress testing in subgroups according to patient characteristics.

33

Table 8. Differential effect of the decision aid on use of stress testing, coronary CT angiography, and revascularization based on patient sociodemographic characteristics

Characteristic Stress Test CT Angiogram Revascularization

Decision Aid Effect

P value Decision Aid Effect

P value Decision Aid Effect

P value

Age 0.551 0.917 0.542 ≤ 50 years > 50 years

0.62 (0.14) 0.76 (0.15)

0.73 (0.22) 0.71 (0.22)

0.98 (1.39) 2.28 (1.64)

Gender 0.271 0.717 0.149 Female Male

0.79 (0.15) 0.54 (0.13)

0.66 (0.19) 0.80 (0.26)

0.65 (0.60) 5.05 (5.60)

Race 0.004 0.842 na White Non-White

0.48 (0.09)* 1.19 (0.28)

0.74 (0.25) 0.70 (0.20)

1.34 (0.91) 1.00 (0.00)

Annual income 0.028 0.328 na < $40,000 ≥ $40,000

1.08 (0.27) 0.54 (0.10)*

0.88 (0.29) 0.56 (0.17)

1.00 (0.00) 1.05 (0.75)

Insurance 0.613 0.921 0.915 Uninsured Insured

0.75 (0.20) 0.63 (0.11)

0.73 (0.27) 0.65 (0.18)

2.01 (1.91) 2.01 (1.76)

Level of Education (n= 872) 0.054 0.499 0.549 ≤ High school > High school

0.48 (0.12) 0.88 (0.16)

0.79 (0.29) 0.64 (0.17)

1.09 (1.11) 2.41 (2.05)

Literacy 0.927 0.18 0.36 Typical (3) Low (>3)

0.68 (0.13) 0.70 (0.17)

0.55 (0.16) 1.00 (0.33)

1.02 (0.85) 3.30 (3.77)

Numeracy 0.249 0.069 na Typical (≤4) Low (>4)

1.08 (0.61) 0.66 (0.10)

3.59 (3.07) 0.63 (0.14)

1.00 (0.00) 1.31 (0.89)

34

DISCUSSION

Principal findings

In patients with chest pain who were otherwise being considered for observation unit admission and

advanced cardiac testing, use of the decision aid increased patient knowledge, increased patient

engagement, decreased decisional conflict, and did not significantly affect physician trust. The decision

aid was found to be acceptable to both patients and physicians, and its use, which took an average of 1

additional minute of clinician time, decreased the proportion of patients who decided, with their

clinician, to be admitted to the observation unit for advanced cardiac testing. Use of the decision aid

also decreased cardiac stress testing within 30 days of the ED visit and the 45-day rate of testing overall.

Although there was no difference in the ED length of stay between study arms, patients randomized to

the decision aid who were admitted to the ED observation unit had a 90-minute-shorter median length

of observation unit stay. When assessing heterogeneity of effect, use of the decision aid increased

patient knowledge, decreased decisional conflict, and increased involvement in the decision similar to

the trial population in the elderly and in patients with less formal education and lower income.

However, the decision aid increased knowledge to a greater degree in whites compared with nonwhites

and in patients with greater numeracy. However, the decision aid increased physician trust to a greater

degree in patients with low self-reported health literacy. We do not think the 1 patient with a MACE in

the decision aid arm was related to the intervention, as the patient was admitted to the hospital during

the index ED admission.

Meaning of the study

Findings from this trial suggest that patients can be effectively educated and engaged in the emergency

care setting in decisions regarding testing and follow-up and that it is feasible to do so in the flow of

clinical care. In addition, when risk estimates from validated prediction models were shared with

patients, and patients were invited to apply their informed values and preferences to decisions, rates of

admission and testing did not increase. Rather, patient-centered interventions such as those tested in

this trial indicate that patients, when educated and informed of their risk, may choose with their

clinician to undergo less extensive evaluation more closely tailored to their personalized risk. Use of an

encounter-level CPC to facilitate risk-informed diagnosis in the emergency care setting also has potential

to decrease downstream health care utilization related to diagnostic testing.

35

When assessing the effect of the decision aid in potentially vulnerable patient groups, we observed

similar effectiveness of the decision aid to the trial population in the elderly and in those with lower

levels of education and less income on the outcomes of patient knowledge, decisional conflict, and

involvement in the decision. However, knowledge increased to a greater degree in patients with “high”

numeracy compared with those with “low” numeracy. Given that communication of risk in numerical

terms is frequently involved in shared decision-making conversations, it is important for clinicians to

follow best practices when communicating risk with patients, such as use of natural frequencies,

estimates of absolute risk, and a consistent denominator,43 and anticipate that patients with low

numeracy may have difficulty comprehending and applying numerical concepts to the decision. It is also

important for shared decision-making researchers to involve patients with limited numeracy in the

decision aid development process and test the effect of the decision aid in this subgroup of patients. We

also observed greater increases in knowledge in white compared with nonwhite patients. Given that

knowledge was the primary outcome of the study, this interaction is of potential importance. Racial and

ethnic differences between patients and physicians may impact knowledge transfer. However, we did

not prespecify an interaction between patient race and knowledge, and this observation could be due to

chance alone and should be interpreted with caution. We also observed a significant interaction

between patients with “low” health literacy and physician trust. Although this interaction is interesting

and supports the hypothesis that use of the decision aid increased physician trust to a greater degree in

potentially vulnerable patients with low health literacy, this interaction was not prespecified and should

be considered hypothesis generating. Finally, white patients and patients with annual incomes of > $40

000 were less likely to undergo stress testing than nonwhite patients and those with annual incomes of

≤ $40 000. This is a potentially important finding, as shared decision making may decrease utilization

when the knowledge of risk is effectively transferred, the patient understands the available

management options, and there is ready access to outpatient follow-up. However, these findings were

not prespecified and should be confirmed in future shared decision-making trials.

Limitations and strengths of the study

Several limitations of this trial should be taken into consideration. The quantitative pretest probability

Web tool7 8 applies only to patients with chest pain. As such, the decision aid cannot be used in patients

with potential ACS who present with non–chest pain syndromes (eg, shortness of breath and/or

diaphoresis). We used 2 different versions of the decision aid in the trial—1 that included the option of

CCTA and 1 that included only cardiac stress testing. Although this introduced a degree of heterogeneity

36

in the intervention, the trial was intentionally pragmatic in design, and contextual fit of the decision aid

to facilitate clinician–patient discussions relevant to the clinical settings enrolling patients in the trial

was essential. In addition, there is now evidence to support applying the shared decision-making tool in

a greater variety of clinical care contexts. We randomized at the patient level, increasing the risk of

contamination between intervention and control groups. To limit the risk of contamination, the

quantitative pretest probability Web tool was password protected, and coordinators did not provide

clinicians access to the decision aid. However, even if contamination were to occur, this would bias the

results of the trial toward the null, and we observed a positive effect of the intervention despite the

potential for contamination. We were unable to contact 70 patients (8%) for assessment of a secondary

outcome. Of these, 68 were confirmed alive at 45 days. The 92% phone follow-up rate supplemented by

mortality review from a national database, however, is robust and comparable to other ED high-quality

studies of patients with potential ACS. We were unable obtain video recordings in 40% of the

encounters. However, the 536 video recordings that were obtained exceeded the required sample size

of 221 needed to meet power estimates. The study had 78% power to detect a 5% difference in MACE

rate between study arms, using a 1-sided noninferiority test with an alpha of 0.05. Although this was

substantially greater power than the initial cohort of patients recruited in our single-center pilot trial,

greater power and precision would be optimal.

For the health care utilization analysis, utilization measures were defined prospectively as secondary

outcomes that would be evaluated, but the study design was not necessarily powered to detect a

difference in these measures.12 Although we reviewed the EMR and attempted to contact all enrolled

patients, we were unable to collect self-reported health care utilization data in the 70 (8%) patients who

were unable to be contacted by phone for follow-up. To increase the rigor of the health care utilization

analysis, we collected hospital-level billing data to help evaluate utilization. It is possible that patients

could have undergone further testing outside of these institutions that may not be captured in these

data. However, all of the patients identified the institution where they were enrolled as their primary

institution.

The primary limitations of the heterogeneity of decision aid effect analysis relate to issues of multiple

testing and imprecision around the estimates of subgroup effects. Given that a total of 80 comparisons

were performed, 1 could estimate that up to 4 interactions (80 x 0.05) could be observed based on

chance alone. To limit the risk of bias associated with multiple testing, we prespecified hypotheses

37

based on prior observations in Shared Decision-Making (SDM) trials39 and prior literature demonstrating

lower levels of physician trust in racial and ethnic minority groups compared with whites.41 We also

followed guideline recommendations for reporting subgroup analyses in clinical trials42 by presenting

only those subgroup analyses that were prespecified or based on a primary study outcome in the

abstract, distinguishing subgroup analyses of special interest in the methods, basing analyses of the

heterogeneity of effect on tests for interaction, and exercising caution in interpreting subgroup

differences. Our analyses often yielded imprecise results of potentially important subgroup effects.

However, this limitation is inherent in subgroup analyses of clinical trials, and, to the best of the

investigators’ knowledge, the current investigation represents the largest cohort of patients enrolled in

a shared decision-making trial to date and has potential to reveal important insights related to the effect

of a decision aid in vulnerable patients.

Unanswered questions and future research

To date, no shared decision-making interventions have been routinized and incorporated into clinical

protocols and emergency care delivery. While the findings from this multicenter trial suggest that the

decision aid may be effective across a variety of clinical settings, further implementation studies are

needed to determine how best to incorporate the decision aid in care pathways; how emergency

clinicians, cardiologists, and primary care clinicians can best work together to ensure incorporation and

implementation of risk-informed patient preferences into admission, testing, and follow-up decisions;

and how to ensure patient preferences guide decision making both during and after the ED encounter.

In addition, given the limited time available for clinician–patient interaction in the emergency setting

and variable levels of health care literacy between patients, time-efficient, vulnerable population–

targeted approaches to patient activation that involve education and preparation for engagement in

shared decisions with clinicians—such as a brief, standardized video–should be explored. Interventions

designed to ensure communication of the rationale for care decisions to family members who were not

present during the ED encounter are also needed to ensure effective implementation of the care

decisions made. Finally, a large-scale implementation trial may be needed to more definitively test the

safety of the intervention.

Conclusions

Use of a decision aid in patients with low-risk chest pain who were otherwise being considered for

observation unit admission for cardiac stress testing or CCTA increased patient knowledge, increased

38

patient engagement, and decreased decisional conflict. Shared decision making facilitated by the

decision aid was perceived to be acceptable to both patients and physicians, and its use decreased the

median length of stay for patients admitted to the ED observation unit, the 30-day rate of stress testing,

and the 45-day rate of all cause testing with no adverse events related to the intervention. All subgroups

benefited to a similar extent from use of the decision aid; white patients and those self-reporting better

numeracy had greater knowledge gains, while physician trust increased more in patients with low health

literacy. Future SDM studies using encounter-level decision aids should take into consideration tailored

approaches to empowering and engaging vulnerable patients in decisions regarding their care.

39

REFERENCES

1. Niska R, Bhuiya F, Xu J. National Hospital Ambulatory Medical Care Survey: 2007 emergency department summary. Natl Health Stat Report 2010(26):1-31.

2. Bhuiya FA, Pitts SR, McCaig LF. Emergency department visits for chest pain and abdominal pain: United States, 1999-2008. NCHS Data Brief 2010(43):1-8.

3. Graff LG, Chern CH, Radford M. Emergency physicians' acute coronary syndrome testing threshold and diagnostic performance: acute coronary syndrome critical pathway with return visit feedback. Critical pathways in cardiology 2014;13(3):99-103. doi: 10.1097/HPC.0000000000000021

4. Than M, Herbert M, Flaws D, et al. What is an acceptable risk of major adverse cardiac event in chest pain patients soon after discharge from the Emergency Department?: a clinical survey. Int J Cardiol 2013;166(3):752-4. doi: 10.1016/j.ijcard.2012.09.171

5. Ladapo JA, Blecker S, Douglas PS. Physician decision making and trends in the use of cardiac stress testing in the United States: an analysis of repeated cross-sectional data. Ann Intern Med 2014;161(7):482-90. doi: 10.7326/m14-0296

6. An introduction to patient decision aids. BMJ 2013;347:f4147.(doi):10.1136/bmj.f4147.

7. Kline JA, Johnson CL, Pollack CV, Jr., et al. Pretest probability assessment derived from attribute matching. BMC medical informatics and decision making 2005;5:26.

8. Mitchell AM, Garvey JL, Chandra A, et al. Prospective multicenter study of quantitative pretest probability assessment to exclude acute coronary syndrome for patients evaluated in emergency department chest pain units. Ann Emerg Med 2006;47(5):447.

9. Pierce MA, Hess EP, Kline JA, et al. The Chest Pain Choice trial: a pilot randomized trial of a decision aid for patients with chest pain in the emergency department. Trials 2010;11:57. doi: 10.1186/1745-6215-11-57

10. Hess EP, Knoedler MA, Shah ND, et al. The chest pain choice decision aid: a randomized trial. Circ Cardiovasc Qual Outcomes 2012;5(3):251-9. doi: 10.1161/circoutcomes.111.964791

11. Tunis SR, Stryer DB, Clancy CM. Practical Clinical Trials. JAMA 2003;290(12):1624-32. doi: 10.1001/jama.290.12.1624

12. Anderson RT, Montori VM, Shah ND, et al. Effectiveness of the Chest Pain Choice decision aid in emergency department patients with low-risk chest pain: study protocol for a multicenter randomized trial. Trials 2014;15:166. doi: 10.1186/1745-6215-15-166.

40

13. Kline JA, Jones AE, Shapiro NI, et al. Multicenter Randomized Trial of Quantitative Pretest Probability to Reduce Unnecessary Medical Radiation Exposure in Emergency Department Patients with Chest Pain and Dyspnea. Circulation Cardiovascular imaging 2014;7(1):66-73. doi: 10.1161/circimaging.113.001080

14. Karanicolas PJ, Montori VM, Devereaux PJ, et al. A new 'mechanistic-practical" framework for designing and interpreting randomized trials. J Clin Epidemiol 2009;62(5):479-84. doi: 10.1016/j.jclinepi.2008.02.009

15. Fergusson D, Aaron SD, Guyatt G, et al. Post-randomisation exclusions: the intention to treat principle and excluding patients from analysis. BMJ 2002;325(7365):652-4. [published Online First: 2002/09/21]

16. Pocock SJ, Simon R. Sequential treatment assignment with balancing for prognostic factors in the controlled clinical trial. Biometrics 1975;31(1):103-15. [published Online First: 1975/03/01]

17. Hargraves I, LeBlanc A, Shah ND, et al. Shared Decision Making: The Need For Patient-Clinician Conversation, Not Just Information. Health Aff (Millwood) 2016;35(4):627-9. doi: 10.1377/hlthaff.2015.1354.

18. Montori VM, Breslin M, Maleska M, et al. Creating a conversation: insights from the development of a decision aid. PLoS Med 2007;4(8):e233. doi: 10.1371/journal.pmed.0040233

19. Schulz KF, Altman DG, Moher D. CONSORT 2010 statement: updated guidelines for reporting parallel group randomized trials. Ann Intern Med 2010;152(11):726-32. doi: 10.1059/0003-4819-152-11-201006010-00232

20. Elwyn G, Hutchings H, Edwards A, et al. The OPTION scale: measuring the extent that clinicians involve patients in decision-making tasks. Health Expect 2005;8(1):34-42.

21. Accurint for Healthcare, LexisNexis Risk Solutions. http://www.accurint.com/health_care.html

Accessed April 24, 2017. 1000 Alderman Drive. Suite 1000 Alpharetta, GA 30005.

22. IOM (Institute of Medicine). 2009. Race, Ethnicity, and Language Data: Standardization for Health Care Quality Improvement. Washington, D.C., 2009.

23. McNaughton C, Wallston KA, Rothman RL, et al. Short, subjective measures of numeracy and general health literacy in an adult emergency department. Acad Emerg Med 2011;18(11):1148-55. doi: 10.11/j.553-2712.011.01210.x.

24. Rothman RL, Montori VM, Cherrington A, et al. Perspective: the role of numeracy in health care. J Health Commun 2008;13(6):583-95. doi: 10.1080/10810730802281791.

41

25. Fagerlin A, Zikmund-Fisher BJ, Ubel PA, et al. Measuring numeracy without a math test: development of the Subjective Numeracy Scale. Med Decis Making 2007;27(5):672-80. Epub 2007 Jul 19.

26. Patient Centered Outcomes Research Institute. http://www.pcori.org/about-us Accessed June 8, 2017

27. Weymiller AJ, Montori VM, Jones LA, et al. Helping patients with type 2 diabetes mellitus make treatment decisions: statin choice randomized trial. Arch Intern Med 2007;167(10):1076-82. doi: 10.1001/archinte.167.10.1076

28. O'Connor AM. Validation of a decisional conflict scale. Med Decis Making 1995;15(1):25-30.

29. Thom DH, Ribisl KM, Stewart AL, et al. Further validation and reliability testing of the Trust in Physician Scale. The Stanford Trust Study Physicians. Med Care 1999;37(5):510-7.

30. Gattellari M, Ward JE. Will men attribute fault to their GP for adverse effects arising from controversial screening tests? An Australian study using scenarios about PSA screening. J Med Screen 2004;11(4):165-9.

31. Elwyn G, Edwards A, Wensing M, et al. Shared decision making: developing the OPTION scale for measuring patient involvement. Qual Saf Health Care 2003;12(2):93-9. [published Online First: 2003/04/08]

32. Cullen L, Than M, Brown AF, et al. Comprehensive standardized data definitions for acute coronary syndrome research in emergency departments in Australasia. Emerg Med Australas 2010;22(1):35-55. doi: 10.1111/j.1742-6723.2010.01256.x

33. Thygesen K, Alpert JS, Jaffe AS, et al. Third universal definition of myocardial infarction. Circulation 2012;126(16):2020-35. doi: 10.1161/CIR.0b013e31826e1058

34. Kline JA, Zeitouni RA, Hernandez-Nino J, et al. Randomized trial of computerized quantitative pretest probability in low-risk chest pain patients: effect on safety and resource use. Ann Emerg Med 2009;53(6):727-35 e1. doi: S0196-0644(08)01861-1 [pii]10.1016/j.annemergmed.2008.09.034 [doi]

35. Hollander JE, Blomkalns AL, Brogan GX, et al. Standardized reporting guidelines for studies evaluating risk stratification of emergency department patients with potential acute coronary syndromes. Ann Emerg Med 2004;44(6):589-98. doi: 10.1016/s0196064404012806

36. USHIK: HCPCS BERENSON-EGGERS TYPE OF SERVICE CODE. https://ushik.ahrq.gov/ViewItemDetails?system=mdr&itemKey=65356000 [accessed August 30 2016.

37. Hess EP, Hollander JE, Schaffer JT, et al. Shared Decision-Making in patients with low-risk chest pain: a prospective randomized pragmatic trial Under Revision 2016

42

38. Hu MC, Pavlicova M, Nunes EV. Zero-inflated and hurdle models of count data with extra zeros: examples from an HIV-risk reduction intervention trial. Am J Drug Alcohol Abuse 2011;37(5):367-75. doi: 10.3109/00952990.2011.597280.

39. Coylewright M, Branda M, Inselman JW, et al. Impact of sociodemographic patient characteristics on the efficacy of decision AIDS: a patient-level meta-analysis of 7 randomized trials. Circ Cardiovasc Qual Outcomes 2014;7(3):360-7. doi: 10.1161/HCQ.0000000000000006. Epub 2014 May 13.

40. Hess EP, Knoedler MA, Shah ND, et al. The chest pain choice decision aid: a randomized trial. Circ Cardiovasc Qual Outcomes 2012;5(3):251-9. doi: 10.1161/CIRCOUTCOMES.111.964791. Epub 2012 Apr 10.

41. Doescher MP, Saver BG, Franks P, et al. Racial and ethnic disparities in perceptions of physician style and trust. Arch Fam Med 2000;9(10):1156-63.

42. Wang R, Lagakos SW, Ware JH, et al. Statistics in medicine--reporting of subgroup analyses in clinical trials. N Engl J Med 2007;357(21):2189-94.

43. Lin GA, Fagerlin A. Shared decision making: state of the science. Circ Cardiovasc Qual Outcomes 2014;7(2):328-34. doi: 10.1161/CIRCOUTCOMES.113.000322. Epub 2014 Feb 4.

Copyright© 2018 Mayo Clinic. All Rights Reserved

Disclaimer:

The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.

Acknowledgement: Research reported in this report was [partially] funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#952).