Final Research Report

PATIENT-CENTERED OUTCOMES RESEARCH INSTITUTE FINAL RESEARCH REPORT

Comparing Two Treatments for Aortic Valve Disease

J. Matthew Brennan, MD, MPH1; Rachel S. Dokholyan, MPH1; Felicia Graham, MBA1; Laine Thomas, PhD1; David J. Cohen, MD, MSc2; David Shahian, MD3; Alice Wang, MD4; Michael J. Mack, MD5; David R. Holmes, MD6; Fred H. Edwards, MD7; Naftali Z. Frankel, MS*; Suzanne J. Baron, MD8; John Carroll, MD9; Vinod H. Thourani, MD10; E. Murat Tuzcu, MD11; Suzanne V. Arnold, MD2,12; Roberta Cohn*; Todd Maser*; Brenda Schawe*; Susan Strong*; Allen Stickfort*; Elizabeth Patrick-Lake*; Dadi Dai, PhD1; Fan (Frank) Li, PhD13; Roland A. Matsouaka, PhD1; Sean O’Brien, PhD1; Michael J. Pencina, PhD1; Eric D. Peterson, MD, MPH1

AFFILIATIONS: 1Duke University, Duke Clinical Research Institute, Durham, North Carolina 2University of Missouri–Kansas City, Kansas City 3Massachusetts General Hospital, Codman Center, Boston, Massachusetts 4Duke University School of Medicine, Duke Surgery, Durham, North Carolina 5Baylor Scott & White The Heart Hospital, Plano, Texas 6Mayo Clinic, Rochester, Minnesota 7University of Florida College of Medicine, Jacksonville 8Beth Israel Lahey Health, Burlington, Massachusetts 9University of Colorado School of Medicine, Aurora 10Piedmont Heart Institute, Atlanta, Georgia 11Cleveland Clinic, Heart, Vascular & Thoracic Institute, Cleveland, Ohio 12Saint Luke’s Cardiovascular Consultants, Kansas City, Missouri 13Yale University, Yale School of Medicine, New Haven, Connecticut *Patient and caregiver collaborators

Institution Receiving Award: Duke University Original Project Title: Optimizing Health Outcomes in Patients with Symptomatic Aortic Valve Disease PCORI ID: CER-1306-04350 HSRProj ID: HSRP20143549 ClinicalTrials.gov ID: NCT02266251

______To cite this document, please use: Brennan JM, Dokholyan RS, Graham F, et al. (2020). Comparing Two Treatments for Aortic Valve Disease. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/08.2020.CER-1306-04350

TABLE OF CONTENTS

ABSTRACT ...... 4 BACKGROUND ...... 6 Report Organization and Deviation From Initial Aims ...... 6 PARTICIPATION OF PATIENTS AND OTHER STAKEHOLDERS ...... 8 Table 1. Patient and Caregiver Collaborators ...... 8 Table 2. Stakeholders and Collaborators Involved in the Project ...... 9 METHODS ...... 11 Aim 1—Comparative Analysis ...... 11 Figure 1. Propensity Score Distribution Before Matchinga ...... 14 Figure 2. Unmeasured Confounders—Falsification Outcome (UTI)a ...... 15 Figure 3. Standardized Differences on the Matched Populationa ...... 18 Aim 2—Decision Aid Development ...... 20 Aim 3—Educational Resource ...... 31 RESULTS ...... 35 Aim 1—Comparative Analysis ...... 35 Aim 2—Decision Aid Development ...... 35 Table 3. Model for 1-Year Mortalitya ...... 36 Figure 4. One-Year Mortality Model Calibrationa ...... 37 Table 4. One-Year Stroke Model ...... 38 Figure 5a. One-Year Stroke Model: Matched Populationa ...... 38 Figure 5b. One-Year Stroke Model: Unmatched Populationa ...... 39 Figure 6. Discharge-to-Home Subgroupsa ...... 40 Figure 7. Discharge-to-Home Patient Characteristicsa ...... 41 Table 5. Discharge Location Model Parameter Estimatesa ...... 42 Figure 8a. DAOH: SAVR vs TAVRa ...... 43 Figure 8b. DAOH: Associated Risk Prediction Modela ...... 43 Table 6. QOL Model Coefficients ...... 45 Figure 9. Validation of the Modela ...... 46 Aim 3—Educational Resource ...... 49 Analysis of Website Use ...... 53 Figure 10a. Website Use Overview ...... 53

Figure 10b. Website Engaged Usea ...... 54 Figure 10c. Website Acquisition Usea ...... 54 DISCUSSION ...... 55 Aim 1—Comparative Analysis ...... 55 Aim 2—Decision Aid Development ...... 55 Aim 3—Educational Resource ...... 59 CONCLUSIONS ...... 63 REFERENCES ...... 64 RELATED PUBLICATION ...... 67 ACKNOWLEDGMENTS ...... 68 APPENDICES ...... 69 Appendix 1. PCORI AV Replacement Study Recruitment Brochure ...... 69 Appendix 2. Covariate Definitions ...... 69 Appendix 3. PCORI ADVICE Study Meeting Slides ...... 69 Appendix 4. ADVICE Quick Reference For Patients ...... 69 Appendix 5. ADVICE Website Informational Card for Providers ...... 69 Appendix 6. Supplementary Methods, Tables, and Figures ...... 69

ABSTRACT

Background: Transcatheter aortic valve replacement (TAVR) is a relatively new treatment for aortic valve (AV) stenosis in inoperable patients and those at high and intermediate risk of mortality with surgical AV replacement (SAVR). While randomized trials support the use of TAVR, the generalizability of those results in clinical practice has been challenged. Beyond average treatment effects, no mechanism exists to estimate an individual’s expected 1-year outcomes following TAVR or SAVR in the United States, and no independent educational resource exists to help patients and their caregivers evaluate the options of SAVR or TAVR for the treatment of AV stenosis.

Objectives: In this project, we sought to address these evidence gaps through 3 aims: (1) to determine the safety and effectiveness of TAVR vs SAVR in a nationally representative real- world cohort; (2) to develop a decision aid to help individualize the assessment of expected 1- year outcomes among intermediate-risk, high-risk, and inoperable patients; and (3) to create a peer-to-peer educational resource, including patient and caregiver testimonials and clear translations of state-of-the-art science for patients diagnosed with AV stenosis.

Methods: To achieve the first and second aims, we linked data from the Transcatheter Valve Therapy Registry and Society of Thoracic Surgeons (STS) National Database to Medicare administrative claims for follow-up. In a propensity-matched cohort of 9464 intermediate- and high-risk (STS predicted risk of operative mortality [PROM] ≥3%) US patients who underwent commercial TAVR or SAVR, we compared death, stroke, and days alive and out of hospital (DAOH) at 1 year, as well as discharge to home, with subgroup analyses by surgical risk, demographics, and comorbidities.

In this propensity-matched cohort, we developed the ADVICE study decision aid using standard techniques to model the probabilities of discharge to home and death or stroke at 1 year as well as the expected DAOH at 1 year following treatment. These models were then validated within the overall cohort of intermediate- or high- risk patients (STS PROM ≥3%) eligible for propensity matching, allowing internal validation in a broader cohort of patients more representative of US clinical practice. A cohort of TAVR patients who had data from the Kansas City Cardiomyopathy Questionnaire short form (KCCQ-12) at both baseline and 1 year was used to model the likelihood of survival at 1 year with at least equivalent quality of life.

Finally, with the direction of patient and caregiver collaborators, risk models were translated into a publicly available web- and mobile application–enabled decision aid for clinical use (available online at http://www.valveadvice.org and through both iOS and Android app sources). To accompany the decision aid, this peer-to-peer educational website, which includes patient testimonials (video and text), was developed by a working group of patients with AV stenosis, caregivers, clinician educators, researchers, and communication specialists.

Results: In a propensity-matched cohort (median age 82 years; 48% female; median STS PROM 5.6%), TAVR and SAVR patients experienced no statistically significant difference in 1-year rates

of death (17.3% vs 17.9%, respectively; hazard ratio [HR], 0.93; 95% CI, 0.83-1.04) and stroke (4.2% vs 3.3%, respectively; HR, 1.18; 95% CI, 0.95-1.47), and no difference was observed in the proportion of DAOH at 1 year (rate ratio = 1.00; 95% CI, 0.98-1.02). However, TAVR patients were more likely to be discharged to home after treatment than SAVR patients (69.9% vs 41.2%, respectively; odds ratio, 3.19; 95% CI, 2.84-3.58). Results were consistent across most subgroups, including among intermediate- and high-risk patients. Subsequently, 5 risk prediction models were developed with excellent calibration and variable discrimination. Four of the 5 models were translated into a publicly available decision aid for use by patients, caregivers, and providers in the process of shared decision-making. Finally, the ADVICE study educational website was launched at the American College of Cardiology (ACC) National Meeting in March 2017. Information regarding use patterns for the first year following website release demonstrates a need for further dissemination efforts.

Conclusions: Among unselected intermediate- and high-risk patients with AV stenosis, TAVR and SAVR resulted in similar rates of death, stroke, and DAOH at 1 year, but TAVR patients were more likely to be discharged to home. Relative risks and benefits of TAVR vs SAVR varied widely across the cohort, and creation of the ADVICE study decision aid has achieved a new paradigm of individualized, comparative risk assessment for the care of patients with AV stenosis. Finally, a peer-to-peer educational resource has been created by and for patients and caregivers. Future research should focus on further refinement of the decision aid and an evaluation of its implementation and dissemination.

Study Limitations: This study has important limitations. The comparative effectiveness analysis used nonrandomized real-world data that are subject to selection bias and other biases. Other limitations of the comparative analysis are discussed in detail in the publicly available primary manuscript publication (see Related Publication). The decision aid is subject to inaccuracies of the underlying models, and implementation has not been tested. The educational resource is primarily limited by its comprehensiveness, accessibility, and static nature. Each of these 3 products were developed and publicly released using data and information available at the time of release in 2016. Because this is a rapidly advancing field, these data are now largely outdated for clinical use.

BACKGROUND

Although many patients with aortic valve (AV) stenosis will never progress to a symptomatic state, symptomatic AV disease is common, occurring in 2% to 5% of the older adult US population. When untreated, symptom progression is both rapid and lethal, with 50% mortality expected at 2 years with AV replacement (AVR) once heart failure symptoms have begun. Medical therapy is largely ineffective, and AVR remains the standard of care. Surgical AVR (SAVR) is less attractive among older and higher-risk patients, so transcatheter AVR (TAVR) offers an alternative for most intermediate- and high-risk patients. With the commercial approval of TAVR in 2011 for the treatment of inoperable patients, the clinical community gained an important alternative to SAVR for treating patients with symptomatic AV stenosis in the United States. Additionally, the patient community faced a new treatment dilemma with limited resources to facilitate the choice between 2 very different treatment approaches. Between 2011 and 2016, the FDA extended their TAVR approval to a much larger pool of patients at intermediate risk of operative mortality, and ongoing clinical trials are evaluating the utility of TAVR in low-risk patients with severe symptomatic AV stenosis. Consequently, there is a growing need for tools to help educate patients and their caregivers, as well as to individualize treatment decisions.

Report Organization and Deviation From Initial Aims Our initial award identified 3 specific aims: (1) to compare contemporary health outcomes with SAVR vs TAVR among operable patients in the United States; (2) to create and assess a personalized decision aid to evaluate expected health outcomes with SAVR vs TAVR for operable patients with AV disease; and (3) to develop and assess a personalized risk assessment tool to evaluate expected health outcomes with TAVR for inoperable patients with AV disease. Additionally, we planned to achieve a general objective to create and evaluate web- and print- based educational resources for targeted dissemination to patients with AV stenosis, their caregivers, and their health care providers.

To achieve these aims, we organized our study into 3 workstreams, including a comparative analysis group to address aim 1, a modeling and personalized decision aid group to

address aims 2 and 3, and an educational-content group to address our general objective. This organizational structure allowed added efficiencies through parallel processing of our research activities, and it allowed our group to more fully capitalize on the divergent strengths of our patient and caregiver collaborators, adding emphasis to the educational component of our work—a recognized need for the clinical community. This report is organized to describe the output from these 3 workstreams.

Due to delays in multiparty contracting and data collection/maturation, data access and decision aid development were delayed. As a result, our final work product did not include a formal assessment of the impact of our decision aid. However, AV clinic patients and caregivers were involved in semistructured interviews (described below) to evaluate the design and functionality of the decision aid and educational resources. A larger-scale evaluation of the impact of these resources as tools to support shared decision-making remains a priority. Likewise, we did not create a parallel multi-outcome risk assessment tool for inoperable patients (ie, patients for whom SAVR was not a viable option). However, we did create a risk assessment model for an “alive and better” outcome at 1 year among the full spectrum of patients treated with TAVR. Translation of this model into a decision aid for the clinical community remains a priority.

In this report, we present the final resulting aims for the project:

• Aim 1: Compare SAVR vs TAVR outcomes at 1 year among intermediate- or high-risk patients in the United States, including among important patient subgroups (see Related Publication for complete list of subgroups).

• Aim 2: Create a personalized decision aid to help patients understand their expected outcomes at 1 year with SAVR or TAVR.

• Aim 3: Create a web-based peer-to-peer educational resource for patients with AV disease and their caregivers to explain the disease process and available treatment options.

PARTICIPATION OF PATIENTS AND OTHER STAKEHOLDERS

The PCORI ADVICE study engaged a broad group of stakeholders, including representatives from the patient and caregiver community, the American College of Cardiology (ACC) and Society of Thoracic Surgeons (STS) professional societies, the Centers for Medicare & Medicaid Services (CMS), the FDA, and clinician-researchers from 9 different academic medical centers.

Seven patients and 3 caregivers were actively engaged in every aspect of our study. Our patient and caregiver collaborators were identified and recruited via the project investigators’ collective clinical and patient networks. A study recruitment brochure was developed and has been attached as Appendix 1. Recruitment and management of interaction was overseen by Bray Patrick-Lake, Director of Patient Engagement for the Duke Clinical Research Institute (DCRI).

Collaborators were nominated and recruited from AV clinics throughout the United States; they included patients who received SAVR as well as those who received TAVR and were a combination of male, female, young, and old. While recruitment was initially challenging, the result was an enthusiastic group of patients and caregivers who added meaningfully to the decision aid design and offered critical clinical “grounding” to the overall effort. All study members have agreed to be publicly recognized for their contributions to this project and are listed in Tables 1 and 2.

Table 1. Patient and Caregiver Collaborators

Roberta Cohn Todd Maser Naftali Zvi Frankel Bray Patrick-Lake

Ellie Gully Brenda Schawe

Stewart Gully Allen Stickfort

James Lee Susan Strong

Table 2. Stakeholders and Collaborators Involved in the Project

Felicia Graham, Project Manager Vinod Thourani, Cardiac Surgeon—Emory University School of Medicine, Atlanta, GA Alice Wang, Research Fellow/Surgery Murat Tuzcu, Cardiologist—Cleveland Clinic Resident Abu Dhabi, Abu Dhabi, United Arab Emirates Michael Pencina, Statistician Suzanne Baron, Cardiologist—Saint Luke’s Mid America Heart Institute, Kansas City, MO Laine Thomas, Statistician Suzanne Arnold, Cardiologist—Saint Luke’s Mid America Heart Institute, Kansas City, MO Roland Matsouaka, Statistician Eric Peterson, Cardiologist—Duke University School of Medicine, Durham, NC; DCRI, Durham, NC Sean O’Brien, Statistician Daniel Mark, Cardiologist—Duke University School of Medicine, Durham, NC; DCRI, Durham, NC David (Dadi) Dai, Statistician Dan Canos, FDA/CMS Fan (Frank) Li, Statistician Danica Marinac-Dabic, FDA David Cohen, Cardiologist—Saint Luke’s Mid Rosemarie Hakim, CMS America Heart Institute, Kansas City, MO David Shahian, Cardiac Surgeon— Kevin Monroe, SignalPath IT Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts— representing the STS John Carroll, Cardiologist—University of Matt Tibbitt, SignalPath IT Colorado Hospital, Aurora, Colorado— representing the ACC Michael Mack, Cardiac Surgeon—The Heart Margaret (Patty) McAdams, Hospital Baylor Plano Research Center, Communications, DCRI Plano, TX David Holmes, Cardiologist—Mayo Clinic, Susan Poulos, Public Relations Rochester, MN Fred Edwards, Cardiac Surgeon—University Erin Campbell, Communications, DCRI of Florida Health Science Center, Jacksonville, FL Abbreviations: ACC, American College of Cardiology; DCRI, Duke Clinical Research Institute; STS, Society of Thoracic Surgeons.

This diverse group of stakeholders and collaborators convened via teleconference every other month between April 2015 and May 2017. Smaller groups focusing more deeply on a specific topic held separate teleconferences in addition to the bimonthly meetings of the entire team. For instance, the Educational Working Group, led by Dr. John Carroll and Patty McAdams (DCRI Communications), focused on design and implementation of the online patient educational website and convened 4 times between November 2015 and June 2016.

Patients and caregivers worked with our core research team to determine the clinical outcomes of interest for the comparative analysis and design, as well as to tailor the educational website (https://valveadvice.org/) and the personalized ADVICE study decision aid (described here: https://valveadvice.org/about-us). Patient and caregiver input were critical in focusing our analytics on evaluation of less traditional outcomes, including discharge location, days alive and out of hospital (DAOH), and quality of life (QOL) assessment.

Many of our patients and caregivers engaged heavily in the creation of this educational resource. There was a strong sense among our patient and caregiver collaborators of the tremendous amount of knowledge that they wished they had known along the way and the desire to communicate this knowledge in a peer-to-peer manner to future patients and caregivers. Although education about the disease process and available treatments was important to our patient and caregiver collaborators, the questions that seemed to interest them the most were often the ones that our physician collaborators were least comfortable asking, for example, “What steps can a patient take to prepare their home and family for the procedural recovery process?” and “What is the best way to ask for a second opinion?” As a result, our patient and caregiver collaborators were largely responsible for driving much of the educational content, resulting in a true peer-to-peer resource for the AV disease community.

METHODS Aim 1—Comparative Analysis Aim 1 was to compare SAVR vs TAVR outcomes at 1 year among intermediate- or high- risk patients with AV disease in the United States, including among important patient subgroups.

Study Overview At the time of the ADVICE study analysis, 3 pivotal randomized trials had compared SAVR vs TAVR outcomes in a controlled setting in the United States. In these 3 trials, SAVR and TAVR outcomes were generally similar, with a few notable differences; however, the relatively small sample sizes included in these trials precluded detailed evaluation of many important patient subgroups. Additionally, a large, nonrandomized analysis of German Aortic Valve Registry data demonstrated an increased risk of mortality among patients treated with TAVR (vs SAVR), particularly among low- and intermediate-risk patients. Our primary comparative analysis was designed to address the safety and effectiveness of these 2 treatments among intermediate- and high-risk patients from the overall US clinical experience and among important patient subgroups.

The results of this analysis were published in 2017 in the Journal of the American College of Cardiology.1 The analysis is described briefly in the Results section.

Study Design This was a retrospective, propensity-matched study of US experience among patients considered at intermediate or high risk with surgical treatment of AV stenosis.

Study Participants

Data sources. In the United States, 2 large clinical registries collect baseline patient demographics, comorbidities, procedural details, and in-hospital outcomes during the initial SAVR or TAVR treatment hospitalization. Details of these clinical registries have been reported

previously and are publicly available online. For SAVR, the STS Adult Cardiac Surgery Database (ACSD) is a voluntary participation database that helps with quality improvement and assurance at participating sites (https://www.sts.org/registries-research-center/sts-national- database/adult-cardiac-surgery-database).2 The STS ACSD does not follow patients beyond 30 days after the SAVR procedure. More than 90% of US cardiac surgery programs participate in the STS ACSD. For TAVR, the Transcatheter Valve Therapy (TVT) Registry is jointly maintained by the STS and ACC (https://www.sts.org/registries-research-center/stsacc-tvt-registry).3 As a provision of the CMS National Coverage Determination, the TVT Registry collects information on nearly all Medicare patients treated with TAVR in the United States. In addition to baseline data, the TVT Registry also collects a QOL metric (the 12-item Kansas City Cardiomyopathy Questionnaire short form [KCCQ-12]) at baseline, 1 month, and 1 year following the TAVR procedure.4,5 Both the TVT Registry and STS ACSD employ routine data audits to ensure high- quality data input. Data elements are harmonized between the 2 data registries.

For long-term patient follow-up, both the TVT Registry and STS ACSD have been linked to CMS administrative billing records for fee-for-service Medicare participants. These records can be used to track rehospitalization events, medication use, and (on a limited basis) outpatient services following SAVR or TAVR. Medicare administrative billing records are linked to the Social Security Administration’s Death Master File.

Linkage of the TVT Registry with CMS administrative claims is performed by the CMS National Coverage Determination group and uses both direct and indirect patient identifiers. Direct patient identifiers are then removed from the analysis files. Linkage of the STS ACSD with CMS administrative claims is performed by the DCRI using deterministic linkage with indirect patient identifiers.6,7

All analyses for the ADVICE study were performed by the DCRI and approved by the Duke University School of Medicine IRB.

Balancing risk. Development of the analysis cohort involved 3 iterative steps: (1) application of clinical exclusion criteria; (2) creation and application of a propensity score model

with 1:1 greedy propensity score matching; and (3) balance assessment of measured clinical risk (using standardized difference assessment) and unmeasured clinical risk (using negative control outcomes).

Exclusion criteria focused on clinical characteristics that would strongly favor one treatment or the other, or preclude treatment with one or the other therapy.8 Additionally, patients were excluded if they had a low risk of operative mortality with SAVR (STS predicted risk of operative mortality [PROM] <3%, N = 2131 patients), because during the time interval of data collection, the TAVR devices were not FDA approved for, or reimbursed by, CMS for treatment of these individuals.9,10 As a result, our team deduced that there must be some compelling clinical reason (likely not included in the standard data collection) why these low- risk patients would have been treated with TAVR. This hypothesis was validated as we analyzed falsification outcomes data (discussed below). Finally, 20 patients from hospitals with <10 total SAVR or TAVR records were excluded.

Early cohort. The initial cohort included SAVR patients treated from July 1, 2011, through December 31, 2013, and TAVR patients treated from November 1, 2011, through June 30, 2014. Before propensity score matching, there was a strong preference for one or the other treatment in each of the 2 treatment groups (Figure 1). Nevertheless, the greedy matching protocol11,12 produced excellent balance across observed covariates, with <10% standardized difference for all demographic and comorbidity covariates of interest.13 Propensity scores included a large list of demographics and clinical comorbidities that were common to the 2 clinical registries (table of covariate definitions in Appendix 2). Details of propensity-score development are discussed in the Statistical Appendix in the related publication.1

Figure 1. Propensity Score Distribution Before Matchinga

Abbreviations: SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aThis figure displays a distribution of likelihood of treatment with TAVR based on a propensity score model that was designed to predict treatment, ranging from 0.00% to 100% likelihood of receiving TAVR. The red bars represent patients who were actually treated with TAVR, while the blue bars represent those treated with SAVR. From this figure, it is evident that (1) patients generally received the treatment that would have been predicted; and (2) there is a sizable proportion of the treated population that could have been treated with either SAVR or TAVR (the area of overlap between the red and blue bars). The patients in this area of overlap were used to form our study cohort (shaded area). This study cohort is generalizable to a large proportion of the overall population.

Despite excellent balance across measured characteristics in the 2 cohorts (SAVR, TAVR), there remained clinical concern that patients in the TAVR cohort might still have a higher baseline risk due to characteristics that were not collected by the clinical registries or included in the propensity score (eg, frailty and physical or mental impairment). To help evaluate balance across unmeasured characteristics, 2 falsification outcomes (urinary tract infections [UTIs] and lower-extremity fractures [LEFs]) were examined.14 Falsification outcomes are clinical outcomes that should not be affected by the treatment of interest (SAVR, TAVR) but may be affected by the unmeasured characteristic of interest. In this case, these outcomes were thought to be affected by patient frailty and debilitation, because higher rates of UTI and LEF have been documented in patients with these characteristics.15,16 If a higher rate of UTI or LEF was detected in the TAVR cohort, for example, this finding would suggest that (despite

propensity matching) the TAVR cohort retained a higher prevalence of frailty or debilitation, which could bias the long-term analysis against the TAVR treatment arm. Ultimately, analysis of the falsification outcomes in this early cohort demonstrated a higher incidence of rehospitalization for both UTI and LEF among TAVR patients (Figure 2). This effect was most prominent among low- and intermediate-risk patients, suggesting that the resulting bias would be greatest in these subgroups if we moved forward with analysis of this early cohort.

Figure 2. Unmeasured Confounders—Falsification Outcome (UTI)a

Abbreviations: HR, hazard ratio; PROM, predicted risk of operative mortality; STS, Society of Thoracic Surgeons (SAVR patients); TVT, transcatheter valve therapy; UTI, urinary tract infection; w, with; w/o, without. aThe top panel demonstrates that patients treated with TAVR had a higher likelihood of rehospitalization for UTI compared with patients treated with SAVR. There is no good biologic mechanism for this, raising concerns that patients treated with TAVR were different from those treated with SAVR—namely, that TAVR patients were more frail than SAVR patients, as frail patients have a higher proclivity for future UTIs. The bottom panel demonstrates that this increased hazard was seen across the spectrum of risk, as defined by the STS predicted risk of operative mortality (PROM). It was not simply a problem among high- or low-risk patients.

Hazard ratio (TAVR vs SAVR) over the STS PROM. Based on the findings of the falsification analysis from the early cohort and an ongoing clinical evolution in the treatment of patients with AV disease (such that clinical equipoise was more common in intermediate-risk patients treated after this initial data collection interval, 2011 to 2013), we decided to wait for additional data collection among TAVR patients to repeat the process of cohort development. We therefore started using data that were collected from 2014 through 2015 (as noted in the next paragraph).

Analysis cohort. The final analysis cohort included data from SAVR patients collected from July 1, 2011, through December 31, 2013, and data from TAVR patients collected from January 2, 2014, through September 30, 2015; this cohort has been described in detail in the related publication.1 In this final analysis cohort, identical exclusion criteria were applied,

yielding a large cohort of patients eligible for propensity matching that included 22 618 SAVR patients from 627 hospital sites and 17 910 TAVR patients from 383 sites; 1:1 propensity matching yielded a final study cohort of 4732 SAVR patients from 627 sites and 4732 TAVR patients from 365 sites (Table 1 in the published manuscript1). Propensity score overlap was excellent for the 2 treatment groups, and standardized differences were <10% for all demographics and comorbidities of interest, demonstrating balance across observed characteristics (Figure 3). In this cohort, the incidence of both UTI and LEF was statistically similar at 1 year following treatment, suggesting balance across unobserved characteristics. This balance was observed for all patients, except for those at lowest risk (STS PROM <3%). Given the strength of these findings, this propensity-matched cohort of 9464 patients became our analysis cohort.

Figure 3. Standardized Differences on the Matched Populationa

Abbreviations: CABG, coronary artery bypass graft; CAD, coronary artery disease; CV, cardiovascular; CVA, cerebrovascular attack; MI, myocardial infarction; PA, pulmonary artery; PCI, percutaneous coronary intervention. aBlue dots represent standardized differences between surgical aortic valve replacement (SAVR) cohort and transcatheter aortic valve replacement (TAVR) cohort before propensity matching. Orange dots represent standardized differences after propensity matching. This figure demonstrates that in the updated cohort, there was good balance across the patient characteristics that were available through the registries. This suggests that the 2 cohorts (TAVR and SAVR) were similar after applying propensity methods.

Discussion of cohort development. The process used to develop our analysis cohort follows best practices of both PCORI and the FDA, assessing balance for observed characteristics through conventional means (standardized differences) and balance of unobserved characteristics through additional appropriate means (falsification outcomes) before analysis of clinical outcomes of interest. This process alerted our analysis team to the likelihood of residual bias before the analysis of clinical outcomes of interest, thereby avoiding the likelihood of accepting a biased estimate. Our team was fortunate that data abstracted from records of procedures done at later dates yielded less biased data (coincident with increased equipoise in the treatment of lower-risk patients in clinical practice), allowing us to complete our predetermined analysis plan.

Other approaches to address this potential bias would have included the use of instrumental variables,17 adjustment of the biased results to account for the expected impact of the unobserved characteristic (only possible if unobserved variables had been well described), or the resource-intensive retrospective collection of additional data elements. Unmeasured but influential variables and uncontrolled bias are a reality of using nonrandomized data sets to perform causal inference analyses; the use of best practices to avoid the generation of biased estimates is imperative.

Interventions and Comparators or Controls We compared SAVR vs TAVR.

Study Outcomes Outcomes were reported at 1 year following AVR and included discharge location (to home or intermediate care facility), DAOH, perioperative stroke, and mortality. Details of outcome measures are available in the related publication.1 Discharge location was collected directly by the STS and TVT registries, while DAOH was collected from Medicare claims data. Both registry and claims data were used to detect postoperative mortality and stroke at 1 year.

Study Setting Treatment of AV disease took place as part of routine clinical care at hospital centers in the United States offering either or both SAVR and TAVR.

Time Frame for the Study AV replacement occurred between July 1, 2011, and December 31, 2013, for SAVR patients and between January 2, 2014, and September 30, 2015, for TAVR patients.

Data Collection and Sources See the “Study Participants” section.

Analytical and Statistical Approaches Details of the statistical analysis are reported in the related publication.1 Before the analysis, due to input from our patient and caregiver collaborators, outcomes of interest were modified to include discharge location and DAOH to help describe the posttreatment clinical course. The TAVR cohort included only patients with procedures performed after January 2014 due to evidence suggesting substantial residual bias following propensity matching using the early TAVR cohort (2011-2014), as described previously (see the “Study Participants” section).

Aim 2—Decision Aid Development Aim 2 was to create a personalized decision aid to help patients understand their expected outcomes at 1 year with SAVR or TAVR.

Study Overview “Personalized medicine…is the tailoring of medical treatment to the individual characteristics, needs and preferences of each patient,” according to Margaret Hamburg, MD, Past-Commissioner of the FDA.32

Shared decision-making is the clearest path to achieving personalized medicine. Tailoring treatment plans to an individual patient’s values can only be achieved by first understanding an individual’s risks and benefits with each of the available treatment options.

In the PCORI ADVICE study, our team developed a decision aid to help intermediate- and high-risk patients with symptomatic AV stenosis better understand their individual expected outcomes at 1 year following SAVR or TAVR. To achieve this aim, we used the International Patient Decision Aids Standards (IPDAS) as a guide, and we sought, and responded to, patient and caregiver feedback. We first surveyed patients and caregivers treated at Duke University Medical Center and our study team members to determine which of the available clinical outcomes were most important to patients facing this treatment decision. A survey of 20 patients (frequently with feedback from their caregivers) undergoing AVR (SAVR, n = 10; TAVR, n = 10) at Duke University Medical Center was conducted using a series of 4 open-ended questions and 6 structured questions on the day before or within 1 week following their AVR procedure. This process yielded some expected outcomes (death, stroke, QOL) and some unexpected outcomes (discharge location, DAOH). Second, we identified patient characteristics that were present in both the TVT Registry and STS ACSD; these characteristics were candidates for inclusion as predictors in the risk models. Third, a questionnaire was completed by the patient and caregiver collaborators to determine which of these patient characteristics would be most readily available to a patient using the decision aid without needing to consult a physician or their medical record. Three caregivers and 5 patients filled out the questionnaire. As a result of this questionnaire, we learned that certain variables (such as pulmonary artery systolic pressure [PASP] and last albumin level) that might be predictive of outcome are not pieces of information that all patients will know. In response to this patient feedback, we created the tool to give risk estimates despite missing information. Where information is missing, the tool uses cohort-specific medians and means in place of the missing items. In this way, estimates can be tailored to an individual to the extent that they know their health information. The result is that patients and caregivers can receive risk estimates without having to provide information for all of the risk predictors. Fourth, using these patient characteristics, we developed and validated statistical models to predict the likelihood of each outcome

(discharge site, DAOH, stroke, and death) for a given patient if treated with SAVR or TAVR (the description of the risk model development begins in the Study Design section below).

Finally, we collaborated with an information technology vendor (SignalPath: https://www.signalpath.com/, SignalPath, LLC) to build web-based and mobile (iOS and Android) applications for implementation of these decision aids. We worked closely with our patient and caregiver stakeholder collaborators to understand the most effective way to display and optimize the clinical transfer of these data to those seeking treatment. When the website containing the decision aid was functional and ready for testing, we held individual critique sessions with each of our patient and caregiver stakeholders. To supplement the feedback from our patient and caregiver collaborators, our team conducted 4 rounds of semistructured interviews with a convenience sample of TAVR and SAVR patients and caregivers (n = 6) at the Duke University Valve Clinic. Of those who were approached to participate, none refused, and all provided informed consent to participate. Feedback from the clinic patients and patient and caregiver collaborators was used to tailor the decision aid before its public release.

In the next section, we describe the process and results of risk-model development and internal validation. The goal of the risk model development was to serve as the basis for the decision aid.

Study Design Aim 2 used the propensity-matched retrospective data set created for aim 1 as the development cohort for a series of 4 clinical outcomes, including discharge location, death at 1 year, stroke at 1 year, and DAOH at 1 year. Each of these models was then validated in the overall cohort of individuals who were eligible for propensity matching (after clinical exclusions). Modeling of a composite QOL (“alive and better”) included a broader cohort of patients, as described below.

Participants The study cohort for aim 2, which included patients with AV stenosis who were considered at intermediate or high risk for SAVR, was identical to that developed for aim 1, except for the composite QOL outcome (alive and better), which included a broader cohort encompassing intermediate- and high-risk patients and those deemed inoperable, as described below. This analytic file was a subset of the Medicare-linked observational data set that was used in aim 1.

Interventions and Comparators or Controls In this analysis, we modeled in-hospital and 1-year outcomes for SAVR vs TAVR.

Study Outcomes In this analysis, we developed multivariate risk models comparing SAVR vs TAVR for 4 outcomes—discharge location, stroke at 1 year, death at 1 year, and DAOH at 1 year.

Study Setting As in aim 1, the AVR operations (SAVR and TAVR) occurred as part of routine clinical care in both academic and community hospital centers in the United States.

Time Frame for the Study AV replacement occurred between July 1, 2011, and December 31, 2013, for SAVR patients and between January 2, 2014, and September 30, 2015, for TAVR patients.

Data Collection and Sources STS ACSD and TVT Registry procedural records were linked to Medicare claims as described in aim 1. For each model, validation cohorts included all treated patients (propensity matched and unmatched) meeting the aim 1 inclusion and exclusion criteria. For the QOL outcome, this cohort was expanded to include a broader group of patients eligible for only TAVR, as described below.

Analytical and Statistical Approaches Modeling methods varied across the outcomes of interest. Details are listed below by outcome. Generally, we developed a multivariate risk-prediction model for each of the 5 outcomes of interest. Four of the 5 models included interactions by treatment, allowing the estimation of predicted risk for each of the 2 treatments.

Modeling Methods by Outcome Mortality. The goal of the study was to produce a prognostic model with predictors of mortality. As the proportional hazards assumption was not violated for this outcome, a Cox model analysis was used.18 The candidate variables that were clinically selected for the mortality model included the following:

• Age • AV mean gradient • Chronic lung disease • Creatinine <2 mg/dL and no dialysis • Extent of coronary artery disease (CAD) • Home oxygen use • Immunosuppression • Left ventricular ejection fraction (LVEF) • Mean PASP • Mitral insufficiency • Peripheral vascular disease (PVD) • Previous cardiac surgery • Previous stroke • Sex

We first ran the Cox model with only SAVR/TAVR and each of the candidate variables, testing the interaction between SAVR/TAVR and the single-candidate variable. We found that the P values for the interactions with home oxygen and previous cardiac surgery were .15 and .0, respectively. Consequently, we forced the treatment by home oxygen and the treatment by previous cardiac surgery interaction terms into the final model. To obtain the final model, we started from the candidate variables; multivariate Cox models with backward selection19 were

performed to identify the independent predictors of 1-year and 30-day mortality. The robust empirical variance20 was used to account for within-hospital clustering.

The mortality-model calibration was assessed in 2 ways: (1) Plot the predicted 1-year mortality among SAVR vs TAVR patients; and (2) plot the observed mortality vs predicted mortality, which is similar to the calibration of the stroke models.21 The discrimination of the model was assessed by using the C statistic.22,23

Stroke. The stroke analyses focused on estimating the cumulative incidence function (CIF), which is the cumulative probability of an end point occurring over a patient’s lifetime. The CIF is the appropriate parameter for describing nonfatal events from the patient’s perspective in the setting of a high competing risk of mortality.24 Unlike standard time-to-event methods, which describe the probability of a nonfatal end point occurring in a hypothetical death-free environment, the CIF models the probability that an end point will actually occur, given that death may preclude an event from happening. For each subgroup of interest, the CIF for stroke was estimated nonparametrically using the Fine and Gray method.25 Differences in stroke incidence across subgroups were assessed using the Fine and Gray proportional subdistributions hazards model.25 Because the proportional hazards assumption was violated, a separate 1-month model was developed. Covariates for this model were identical to the mortality Cox model. Hazard ratios describe the relative risk of experiencing a stroke in a setting in which stroke events may be precluded by early death in some patients. The cluster effect of a hospital was taken into account with the robust sandwich estimate of Lin and Wei.20 The model with a backward selection method (P < .10) was then performed to identify independent predictors of the 1-month and 1-year stroke event. The candidate variables that were clinically selected included the following:

• Age • Aortic insufficiency • AV mean gradient • Body surface area • Cerebrovascular disease other than stroke • Chronic lung disease

• Creatinine • Dialysis • Hispanic ethnicity • Immunosuppression • Last hematocrit (hct) test • Left main CAD • LVEF • Mitral insufficiency • Mitral stenosis • Preoperative atrial fibrillation/flutter • Previous AVR • Previous CAD • Previous cardiovascular surgery • Previous implantable cardioverter-defibrillator (ICD) • Previous myocardial infarction • Previous PVD • Previous stroke • Previous shock • Systolic pressure • Tricuspid insufficiency • Urgent or preoperative intra-aortic balloon pump (IABP)/inotropes

The calibration and discrimination of 30-day and 1-year stroke models were evaluated in both a propensity score-matched and -unmatched population. The predicted stroke rate of each patient was calculated by applying regression estimates to the model. To assess model calibration, we rank-ordered patients from the lowest- to highest-predicted risk. We then compared the observed event rates within each risk stratum. The model discrimination was assessed by calculating the C statistic, which represents the probability that, between 2 randomly selected patients, the patient who survived longer had a lower predicted risk of stroke.

DAOH. Differences in DAOH between SAVR patients and TAVR patients were estimated as rate ratios with 95% CIs using Poisson regression26 with the generalized estimating equations (GEEs).27 An offset was used to adjust for differential follow-up time between patients.26 The

robust sandwich variance was used to calculate the standard error of the rate ratios and the associated CIs.27 Like the discharge-to-home outcome, the Poisson GEEs produce estimates similar to those from an ordinary log-linear model, but the estimated variance accounts for within-hospital clustering.27 The candidate variables included the following:

• Age • AV mean gradient • Body surface area • Chronic lung disease • Congestive heart failure (CHF) with New York Heart Association (NYHA) class IV • Commercial insurance • Creatinine • Diabetes control • Dialysis • Home oxygen test • Last hct test • Left main CAD • Mitral stenosis • Preoperative atrial fibrillation/flutter • Preoperative total albumin • Previous ICD • Previous PVD • Previous shock • PASP • Resuscitation • Tricuspid insufficiency • Urgent or preoperative IABP/inotropes

We first examined the interaction between each candidate adjustment variable with treatment by fitting individual Poisson GEEs with the robust variance estimator. Because of too few variables remaining in the model after backward selection with a P < .05 criterion, we loosened the criterion to P ≤ .10 and found that left main CAD, mitral stenosis, and resuscitation had significant interactions with treatment in predicting the DAOH outcome. We included these interaction terms in the model-building process. The full model included

treatment variables and all candidate adjustment variables together with the identified interaction terms. The final model was chosen by applying backwards selection with P value threshold of <.1 on the full model. The calibration was assessed by plotting the predicted DAOH vs the observed DAOH.

Discharge location. In examining the association between the variables and discharge home (vs discharge to an intermediate-care facility) as outcome, logistic regression with GEEs was used with an exchangeable working correlation structure27 to account for within-hospital clustering because patients at the same hospital were more likely to have similar responses relative to patients in other hospitals. This method produces estimates similar to those from ordinary logistic regression, but the approximate variances of the estimates are adjusted for the correlation of discharge home within each hospital.27 The candidate variables for the model included the following:

• Age • AV mean gradient • Chronic lung disease • Creatinine <2 mg/dL and no dialysis • Home oxygen use • Immunosuppression • LVEF • Mitral insufficiency • Previous cardiac surgery • Previous stroke • PASP

We first ran 1 model for each candidate variable, including the variable itself (SAVR/TAVR), and the interaction between the 2 treatments. Consequently, we chose interactions between the following variables with a P value ≤.15 in the above-described models:

• AV mean gradient • Extent of CAD • Mitral insufficiency (moderate/severe) • Previous cardiac surgery

• PASP • Sex • Treatment and age

Then the GEE model with backward selection was used with the criteria P ≤ .05 for main effects and P ≤ .15 for interaction terms.

The final model calibration was assessed by plotting the observed event rates vs the predicted event rates in each risk group, rank-ordered by the predicted event rates. The model discrimination was assessed using the C statistic.

Alive and better (a QOL outcome). The TAVR Poor Outcome risk model was developed using data from Cohorts A and B of the Placement of AoRTic TraNscathetER Valve Trial (PARTNER) pivotal clinical trial (the randomized clinical trial that tested the Edwards Sapien TAVR device), calibrated in the CoreValve clinical trial data set, as previously described.28 This model was used as the basis for the ADVICE study QOL risk model after recalibration in the TVT Registry data. We examined the rate of poor outcome (defined as death, poor QOL, or decline in QOL) at 1 year after TAVR and the performance of the TAVR Poor Outcome risk model among all patients included in the TVT Registry between November 9, 2011, and June 30, 2015. We then reestimated the intercept and coefficients in the model and retested model performance, both overall and in several key clinical subgroups1.

In the TVT Registry, a QOL assessment was available at baseline and 1 year after TAVR in most patients and is described using the KCCQ-12 overall summary score (KCCQ-OS), which is a previously validated measure.28 The primary outcome of interest was described as a “poor outcome,” defined as a composite of death, poor QOL (KCCQ-OS score < 60), or decline in QOL (decrease of ≥10 points in KCCQ-OS score from baseline) at 1 year following TAVR; the algorithm for defining the composite outcome was similar to other such algorithms developed in this field.

Consistent with other analyses from the TVT Registry, sites <50% completion rates for the KCCQ-OS (125 sites) were excluded. Patients from the remaining 245 sites made up the

study cohort.28 Methods for imputing missing data are described in detail in the related publication.1 A model that included the original intercept and variable coefficients from the previously developed TAVR Poor Outcome risk model was tested in the TVT Registry data. Discrimination was evaluated using a C index, and calibration was evaluated by plotting observed vs predicted risks by decile of predicted risk. The model parameters were then reestimated and model performance was reexamined in the overall cohort and among clinically meaningful patient subgroups.

Personalized Decision Aid Development Following development and internal validation of outcome-specific risk prediction models, the statistical algorithms were translated to both web and mobile applications for clinical dissemination in collaboration with SignalPath. The design for the decision aid was planned in close collaboration with patients and caregivers on the study team, as described previously. In general, patients and caregivers preferred a display that began with a high-level summary of risk and then stepped through increasing detail, including a measure of uncertainty. They requested the ability to evaluate risks in multiple ways, including with numeric presentation and using pictographs. Finally, they requested a personalized interpretation of the data. Each of these elements was incorporated into the final design of the website and accompanying iOS and Android applications.

The ADVICE study risk calculator is accompanied by comprehensive instructions for use, a description of the process of development of the decision aid, an explanation of how the risk calculator achieves approximations of estimated treatment-specific outcomes, and answers to many frequently asked questions. Importantly, the risk calculator explicitly states that it is “designed to facilitate a discussion between you and your physician on which treatment option is right for you,” (see https://valveadvice.org for calculator description), which emphasizes that the tool is intended to be used in consultation with the patient’s health care provider.

The IPDAS is the result of an academic collaboration that has developed minimal standards for decision aids.29 The IPDAS minimal criteria include 44 items that are being

included in plans by various entities to certify that patient decision aids meet these minimal criteria. Although certification of our personalized decision aid was considered, certification is currently being provided on a limited basis by the Washington State Health Care Authority and only in certain priority clinical areas.

Changes to the original study protocol. DAOH and discharge location were identified as important outcomes of interest by our patient and caregiver collaborators. These were added to our list of outcomes for multivariate modeling at the statistical planning stage before data analysis. Additionally, a composite QOL risk prediction model was developed for a broad TAVR cohort (including those at intermediate or high risk of mortality with surgical treatment and those considered inoperable by the AV teams, based on clinical parameters). Due to delays with data acquisition and contracting, we did not develop a full suite of models for patients considered clinically inoperable, as initially planned. The majority of our models only applied to intermediate- and high-risk patients (excluding inoperable patients, a category encompassing those with an unacceptably high risk for surgery). However, the development cohort for the composite QOL outcome included inoperable patients, thus allowing generalization of the results for this model to the inoperable cohort.

Aim 3—Educational Resource Aim 3 was to create a web-based peer-to-peer educational resource for patients with AV disease and their caregivers to explain the disease process and available treatment options.

Study Overview Faced with a new diagnosis, patients and their caregivers look to trusted sources of information for education. Among the aims of the PCORI ADVICE study, we sought to create an unbiased educational resource (online brochure) for patients with symptomatic AV stenosis. This web-based resource is intended to be a companion to the personalized risk assessment tool (described previously) and is directed towards patients and caregivers, with the goal of promoting a process of shared decision-making. The decision aid created under aim 2 of this grant is also accessible via this website. Below, we will describe the process of website

development; representative screen shots of the web-based content are provided in the Results section to give a sense of the results of our efforts. However, the full website is publicly available and can be accessed at http://www.valveadvice.org.

The educational patient website was planned, designed, and refined during 4 webinars of the Educational Working Group, led by Dr. John Carroll and Patty McAdams, which convened 4 times between November 2015 and June 2016. Discussions and decisions about the website also took place during project investigator webinars that occurred between April 2015 and May 2017. The patient and caregiver collaborators were invited to all calls of the Educational Working Group and project investigator calls. See Participation of Patients and Other Stakeholders for more information about patient and caregiver collaborator recruitment efforts.

An initial step during these conference calls was to identify topics of interest to patients facing a new diagnosis of AV stenosis. These topics included the following:

Background

• What is the AV? • What is AV disease? • Living with AV disease • Do I need treatment?

Treatment options

• Treatment approaches – What is TAVR? What is SAVR? – Am I a candidate for 1 or both of the AV approaches? – Common complications or issues • Valve types – Comparison of pros and cons of mechanical and biologic valves • Making a decision – Making a decision about treatment based on what is important to you – Getting a second opinion • Where to go for treatment – Common issues that may help guide the choice of treatment center

• Preparing for the procedure – Physical preparation – Mental and spiritual preparation – Clarification of your wishes – Organizing your personal life – Assembling your support system – Planning for postprocedure recovery (with a planning postprocedure checklist) Posttreatment

• Recovery from the procedure – Early recovery When will I feel the benefit of having a new AV?

– In-hospital services – Recovery process—what to expect (with a “usual course” activity chart for SAVR and TAVR procedures) • Living with a new valve – Including the importance of certain key medications like blood thinners and antibiotic prophylaxis Additional resources

• Quick-reference cards for patients and clinicians • Resource links

After identifying topics of interest, the team worked with the Communications Department at DCRI to develop and tailor written content and video testimonials to populate this educational resource. The website was built on a platform that enables tracking of use patterns, and a summary of use to date is provided in aim 3 in the Results section. A mechanism to contact our study team was included as part of the website and has been maintained by our project manager.

When the website containing the patient educational materials and the decision aid was functional and ready for testing, Dr. Alice Wang, one of the co-investigators on the project, followed up individually with each of the patient and caregiver stakeholders to review the website to determine whether the information was useful. Dr. Wang also visited the Duke University Valve Clinic on separate occasions and met with patients scheduled to have a TAVR procedure to obtain feedback via semistructured patient interviews. Six patients were

interviewed. Of those who were approached to participate, none refused, and all provided informed consent to participate. Feedback from the clinic patients and patient and caregiver stakeholder collaborators was given to SignalPath for final incorporation into the decision aid and to the DCRI Communications Department for overall changes to the website before making it public. Formal pilot testing of the website did not take place.

The website was publicly released in March 2017 at the ACC Annual Meeting. An informational card was developed by the DCRI Communications Department for dissemination at the meeting (see Appendix 5). No additional wide-scale dissemination effort accompanied this release.

RESULTS Aim 1—Comparative Analysis Generally, patients treated with TAVR were much more likely to be discharged to home than those treated with SAVR. The rates of stroke and mortality were similar at 1 year between treatments, with important subgroup differences. Likewise, DAOH was similar between the 2 treatments. A more complete presentation of the analysis results can be found in the related publication.1 In a propensity-matched cohort (median age, 82 years; 48% female; median STS PROM, 5.6%), TAVR and SAVR patients experienced no difference in 1-year rates of death (17.3% vs 17.9%, respectively; HR, 0.93; 95% CI, 0.83-1.04) and stroke (4.2% vs 3.3%, respectively; HR, 1.18; 95% CI, 0.95-1.47), and no difference was observed in the proportion of DAOH at 1 year (rate ratio = 1.00; 95% CI, 0.98-1.02). However, TAVR patients were more likely to be discharged home after treatment (69.9% vs 41.2%; odds ratio [OR], 3.19; 95% CI, 2.84- 3.58). The results were consistent across most subgroups, including among intermediate- and high-risk patients.

Aim 2—Decision Aid Development

Mortality at 1 Year The model for 1-year mortality was developed among 9464 patients, of whom 1363 died. The model included 22 patient characteristics (Table 3). For Tables 3 to 6, specifics of formatting of the model variables are discussed in the statistical supplement included in this report (see Related Publication).1 Model calibration was excellent among the development and broader cohorts (Figure 4).

Table 3. Model for 1-Year Mortalitya

β- Patient characteristic coefficient SE SE ratio χ2 P value TAVR vs SAVR .03 0.07 1.02 0.2 .7 Age .02 0.005 1.03 16.5 .00005 Preop atrial .3 0.06 1.05 21.0 .000005 fibrillation/flutter Previous CAD .2 0.07 1.09 5.0 .02 Body surface area –.4 0.1 0.99 10.4 .001 Chronic lung disease, mild .1 0.07 0.97 2.4 .1 Chronic lung disease, .4 0.09 1.06 21.9 .000003 moderate Chronic lung disease, severe .5 0.09 1.07 27.9 .0000001 CHF with NYHA class IV .4 0.06 0.96 38.1 .1 Creatinine .1 0.02 0.97 51.9 0 Last hct test –.02 0.006 1.01 16.6 .00005 Home oxygen use .4 0.1 0.96 11.9 .0006 Immunosuppression .3 0.09 0.99 13.1 .0003 PA systolic pressure .009 0.002 1.06 17.7 .00003 Previous ICD .3 0.1 0.98 8.5 .004 Previous PVD .3 0.06 0.96 19.2 .00001 Preop albumin –.5 0.06 1.12 64.6 0 Tricuspid insufficiency, .2 0.07 0.98 7.9 .005 moderate/severe AV mean gradient –.5 0.06 1.12 64.6 0 TAVR vs SAVR home oxygen –.3 0.2 0.99 2.8 .09 use Previous CV surgeries .1 0.08 0.99 1.5 .2 TAVR vs SAVR previous CV –.2 0.1 1.04 3.6 .06 surgeries Abbreviations: AV, aortic valve; CAD, coronary artery disease; CHF, congestive heart failure; CV, cardiovascular; hct, hematocrit; ICD, implantable cardioverter-defibrillator; NYHA, New York Heart Association; PA, pulmonary artery; Preop, preoperative; PVD, peripheral vascular disease; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aThis table is included to allow independent implementation of the mortality risk model. The numbers in this table are critical for that task, but they are unlikely to be interpretable (or interesting) outside of that context.

Figure 4. One-Year Mortality Model Calibrationa

Abbreviations: SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aIn a model with perfect calibration, the 10 patient groups for each treatment (represented by dots on the graph) would fall along the diagonal line. In fact, this is the case for this mortality model. Therefore, the expected result is generally close to the observed result—suggesting excellent model calibration (C statistic = 0.67). This is an important measure of whether the model gives accurate estimates of risk.

Stroke In the development cohort, 307 patients had experienced a stroke at 1 year. Many patient characteristics were associated with an increased risk of stroke at 1 year, but the final model included 5 characteristics (Table 4). Model calibration was good (Figures 5a and 5b), indicating that the ability of the model to assign accurate probabilities of a stroke occurring is high; however, model discrimination was relatively poor (C index = 0.57 [95% CI, 0.54-0.60] for the matched population, and C index = 0.56 [95% CI, 0.54-0.59] in the unmatched population; Figure 5b), indicating that the model’s ability to rank multiple patients from high to low risk of stroke is more limited. In this context, the model calibration may be considered more applicable than model discrimination.30

Table 4. One-Year Stroke Model

Patient characteristic HR 95% CI P value TAVR vs SAVR 1.225 0.986-1.523 .070 Previous stroke 1.483 1.104-1.994 <.01 Previous AV replacement 0.438 0.201-0.957 .040 Urgent or preop 0.762 0.574-1.013 .060 IABP/inotropes Severe mitral (3 to ³4) or 0.468 0.261-0.840 .010 tricuspid regurgitation Abbreviations: AV, aortic valve; preop, preoperative; HR, hazard ratio; IABP, intra-aortic balloon pump; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement.

Figure 5a. One-Year Stroke Model: Matched Populationa

Figure 5b. One-Year Stroke Model: Unmatched Populationa

Abbreviations: SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement aIn both panels of this figure, the groups of patients (dots on the graph) deviate somewhat from the diagonal line—demonstrating that while the calibration is still good, it is not as good as in the mortality model.

Discharge Location Of the 9464 propensity-matched patients with recorded discharge location from the index hospitalization, 70% of TAVR patients and 41% of SAVR patients were discharged to home (including with home care, in some cases; Figure 6). No patient subgroup was identified as more likely to be discharged to home if treated with SAVR vs TAVR (Figure 7); however, patient demographics and comorbidities affected the likelihood of discharge to home. Model parameters are reported in Table 5.

Figure 6. Discharge-to-Home Subgroupsa

Abbreviations: CAD, coronary artery disease; Dz, disease; PA, pulmonary artery; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aIn this figure, results to the right of the line of unity (odds ratio, 1.0) favor TAVR. Each of the subgroups evaluated had higher odds of discharge to home if treated with TAVR (vs SAVR) in multivariable modeling.

Figure 7. Discharge-to-Home Patient Characteristicsa

Abbreviations: CAD, coronary artery disease; CHF, congestive heart failure; CV, cardiovascular; DM, diabetes mellitus; IABP, intra-aortic balloon pump; LVEF, left ventricular ejection fraction; NYHA IV, New York Heart Association Class IV; O.R., odds ratio; Preop, PreOp, preoperative; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aThis figure shows whether each patient characteristic is associated with higher or lower odds of discharge to home (compared with not having the characteristic) in univariate modeling. For example, men (vs women) were more likely to be discharged to home, while those with a previous stroke (vs no previous stroke) were less likely to be discharged to home.

Table 5. Discharge Location Model Parameter Estimatesa

Patient characteristic Parameter estimate TAVR vs SAVR −0.87 Age −0.08 Sex (male) 0.58 Body surface area −0.45 Preop atrial fibrillation/flutter −0.23 Chronic lung disease (severe) −0.32 Chronic lung disease (moderate) −0.21 Chronic lung disease (mild) −0.09 CHF with NYHA class IV −0.41 Diabetes (no therapy) 0.04 Diabetes (insulin) −0.25 Diabetes (noninsulin) −0.08 Last hct test 0.02 Home oxygen use −0.28 LVEF 0.00 Prior stroke −0.30 Prior CAD −0.22 Previous PVD −0.33 Previous CV surgeries 0.28 Previous shock −0.63 Urgent or preop IABP/inotropes −0.36 Preop albumin 0.48 AV mean gradient 0.00 Abbreviations: AV, aortic valve; CAD, coronary artery disease; CHF, congestive heart failure; CV, cardiovascular; hct, hematocrit; LVEF, left ventricular ejection fraction; NYHA, New York Heart Association; Preop, preoperative; PVD, peripheral vascular disease; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aThis table gives model parameters for discharge location. These parameter estimates are only useful in their current form if an individual wanted to independently implement the discharge-to-home model. These parameter estimates are not meant to be interpreted outside that situation.

The number of DAOH in the first year following treatment was similar for patients treated with TAVR or SAVR (Figure 8). The associated risk prediction model follows (Figure 8b).

Figure 8a. DAOH: SAVR vs TAVRa

Abbreviations: DAOH, days alive and out of hospital; SAVR, surgical aortic valve replacement; TAVR, transcatheter aortic valve replacement. aThis figure demonstrates that a large proportion of patients are not rehospitalized in the first year after treatment (the large spike at 365 days), but the distribution of rehospitalizations among those rehospitalized in the first year is similar with SAVR vs TAVR.

Figure 8b. DAOH: Associated Risk Prediction Modela

aThis model is primarily of use for independent implementation of the DAOH model.

Patient characteristics used in the model associated with DAOH are the following:

1. source_tvt: 1 if TAVR and 0 if SAVR

2. age: age of patient

3. iafibflutter: preoperative atrial fibrillation/flutter

4. ibsa: body surface area

5. ichfnyha4: CHF with NYHA class IV

6. ichrlungd: chronic lung disease

7. icreatlst: creatinine

8. idialysis: dialysis

9. idmctrl: diabetes control

10. ihct: hct test

11. ihmo2: home oxygen use

12. iimmsupp: immunosuppression

13. iresusc: resuscitation

14. ipasys: PASP

15. ipayorcom: commercial insurance

16. iprioricd: previous ICD

17. ipriorpvd: previous PVD

18. itotalbumin: preoperative total albumin

19. itvmodsev: tricuspid insufficiency, moderate/severe

20. ivdgrada: AV mean gradient

21. priorCAD: previous CAD

22. urgentiab: urgent or preoperative IABP/inotropes

QOL Among 13 351 patients who underwent TAVR at 252 US sites, the rate of poor outcome at 1 year after TAVR was 38.9%, which was due to death in 20.7% and poor QOL or decline in QOL in 18.2%. The rate of poor outcome decreased slightly over time, from 42.0% in 2012 to 37.8% in 2015 (P for trend = .076). The original TAVR Poor Outcome risk model did not calibrate well on this population and was therefore reestimated—after which it performed well, with a C index of 0.65 and excellent calibration (Figure 9). Table 6 presents QOL model patient characteristics and coefficients. This table is primarily of interest for independent implementation of the model. The coefficient estimates were obtained by running a logistic model for the overall study cohort (Figure 9).

Table 6. QOL Model Coefficients

Patient characteristic Coeff. (IPW) Coeff. (IPW) P value Intercept: poor outcome = 1 −0.1392 .1274 Diabetes mellitus 0.0799 .0729 Previous atrial fibrillation/flutter 0.3329 <.0001 Preprocedure: creatinine level, mg/dL 0.1540 <.0001 Home oxygen 0.5955 <.0001 Preprocedure: AV stenosis—AV mean −0.0040 .0068 gradient (mm Hg) Baseline KCCQ-12 score −0.0145 <.0001 Abbreviations: AV, aortic valve; Coeff., coefficient; IPW, inverse probability weighting; KCCQ-12, Kansas City Cardiomyopathy Questionnaire short form; QOL, quality of life.

Figure 9. Validation of the Modela

aThe blue solid line represents regression with restriction of intercept = 0; the red dashed line represents regression without restriction. All validations were conducted on the population by taking normalized inverse probability weighting into account. As with the other calibration plots in this report, this figure evaluates how well the model-predicted likelihood of QOL improvement matched the observed results. In this case, the calibration of the model is excellent.

Personalized Decision Aid Development The online decision aid displays multiple tabs, including About, How It Works, FAQs, and Calculator. The welcome page contains a button for patients and a button for providers. The content provided to patients and providers is identical—the buttons were provided for the tracking. Below, we have provided a small sample of the online content via screenshots.

The About tab contains details about how the calculator was created, by whom, and how it was funded.

• The How It Works tab contains a basic description of what the calculator does. • The FAQs tab contains answers to several frequently asked questions about the calculator. • The Calculator tab contains the calculator itself, including detailed instructions for how to fill in the information that the calculator needs to function. Calculator users are asked to enter demographic and clinical information on items like age, race, insurance, history of diabetes, and previous stroke. After all information is entered, the user clicks on a “calculate” button, and the results are provided in both a “Fundamentals” version and an “Advanced” version by way of internal subtabs. The Advanced page contains the same results but provides additional statistical information.

The readability of the website was assessed using the services available at http://jucystudio.com.1 The Flesch Reading Ease score is 60.93. The Flesch Reading Ease score is a 100-point scale on which the higher the score, the easier the text is to understand. Authors are encouraged to aim for a score of approximately 60 to 70. The Flesch-Kincaid Grade level of the text is 7.02. As a general practice, a Flesch-Kincaid Grade level between sixth and eighth grade is recommended for a public audience.

Aim 3—Educational Resource

Educational Website Aim 3 was to create a web-based, peer-to-peer educational resource for AV disease patients and their caregivers to explain the disease process and available treatment options. The resulting website, titled ADVICE: Navigating Aortic Valve Treatment Choices, is located at http://www.valveadvice.org. The Advice study risk calculator (aim 2 of this project) is incorporated into this educational website and can be reached directly at calc.valveadvice.org2.

1 Ed. note: this website does not appear to be correct. Author did not provide updated website address, but we believe it is juicystudio.com. 2 Ed. note: this website is no longer active.

These resources were built on a web-based platform that tracks use patterns, including the frequency of use by patients or providers and the regional location of user internet provider addresses. Importantly, no protected health information is stored by these applications.

In an appendix, we have provided a full set of screenshots of the web-based educational resource for patients with AV disease (see Results Aim 2 section for screenshots of the web- based calculator) and graphics depicting website use analytics from the DCRI Communications department.

The tabs to navigate to pages on the website are Home, About, Treatment Options, Post Treatment, Resources, and Contact Us.

• The Home tab contains basic information about AV disease and the designers of the website. Note that the decision aid calculator mobile apps are linked in the top right corner of the Home tab. • The About tab provides more details about the making of this website and includes video of the lead investigator as well as a list of the various collaborators and stakeholders. • The Treatment Options tab contains several subpages that provide details related to treatment approaches, AV types, making a decision about treatment, where to go for treatment, and preparing for the procedure. Several videos of interviews with patients supplement the written information that was designed with patients in mind. • The Post Treatment tab contains further subpages with information about recovery after the procedure and living with a new AV. • The Resources tab contains 2 downloadable quick-reference documents for patients (Appendix 4) and providers (Appendix 5), as well as multiple links to external resources (eg, the American Heart Association website).

Analysis of Website Use Figures 10a to 10c characterize use of the website between March 15, 2017 (release), and February 14, 2018. There were 2589 users of the site who averaged 1.2 sessions each. The average session was 1 minute and 29 seconds. Of all users, 712 were considered engaged users, spending more than 10 seconds on the site. We expect that with full advertising efforts these results will improve substantially.

Figure 10a. Website Use Overview

Figure 10b. Website Engaged Usea

aIndividuals who visited multiple pages or spent >10 seconds on the site. Engaged users comprised 29% of all users and 26% of all sessions.

Figure 10c. Website Acquisition Usea

aHow individuals arrived at the site. Organic search and direct traffic (ie, they already knew the URL) drove >90% of traffic to the site.

DISCUSSION Aim 1—Comparative Analysis A full discussion of the context, generalizability, implementation, subpopulation implications, and limitations of the results of the first (comparative) aim of this project is available in the related publication.1

Aim 2—Decision Aid Development

Context for Study Results To help both patients and providers better understand the probable outcomes with each of the available treatments for AV disease (TAVR or SAVR), we developed and internally validated risk prediction models to serve as the basis of a personalized risk assessment tool (decision aid).

The ADVICE models are unique within the field of valvular heart disease, providing a direct comparison of expected outcomes between 2 viable treatments for AV stenosis. Before the development of these models, patients and providers relied on average treatment effects from randomized clinical trials or short-term models predicting in-hospital outcomes, developed from cohorts of patients treated with either SAVR or TAVR. However, average treatment effects do not often apply to individuals, and comparisons of the estimates from available sources were necessarily biased by differences in the underlying development cohorts so that comparison of expected outcomes from the STS PROM with those from the TVT models would necessarily favor surgical treatment due to a sicker development cohort for the TVT models. The ADVICE models advance the state of the science in this field, providing a mechanism for individualized risk assessment for outcomes beyond the index hospitalization. They serve as the foundation for publicly available decision aids, designed for use and comprehension by both patients and providers as a part of a shared decision-making process. In doing so, they fulfill a clinical need within AV stenosis and serve as a prototype for development in other disease states.

Generalizability of the Findings The ADVICE models for discharge location, 1-year mortality, stroke, and DAOH are generalizable to patients with AV stenosis deemed to be at intermediate or high risk for surgical mortality (STS PROM ≥3%) in the United States. The QOL model is further generalizable to patients at intermediate or high risk and those deemed inoperable using a traditional surgical approach.31 As the state of the science progresses to include lower-risk patients (STS PROM <3%), these models will need to be redeveloped to include those cohorts.

Implementation of Study Results The ADVICE models are intended for direct clinical implementation as a part of a shared decision-making process. The publicly available decision aids (available online or through iOS and Android mobile applications) are designed to provide estimates of risk tailored to the level of available information. Beyond a few key pieces of information (eg, age, sex, height, weight), estimates continue to gain precision as new information is entered. However, complete information is not required to obtain a risk estimate. If data fields are left blank, the calculator will auto-populate those fields with population means or medians as appropriate. This function increases the usability of the tool by patients and their caregivers who may not know all of the fields necessary to obtain a fully tailored risk estimate. The more information that is known, the more tailored the estimate that can be produced. Of course, the most accurate and individualized estimates are available only through accurate completion of all data fields.

While these decision aids have tremendous clinical potential, there will be important barriers to their implementation. The 2 primary barriers to implementation are time in the clinical workflow and availability of data elements. For a busy clinician, data entry can be an obstacle in the use of a clinical tool. The ADVICE decision aid can take 3 to 8 minutes to complete, depending on how well a clinician knows the patient and the ease of review of the various tests needed to pull data elements (eg, catheterization report, echocardiogram report, laboratory data). This time investment can represent a substantial proportion of the time allotted to a patient visit. Efforts to reduce this time commitment (such as auto-population from the electronic health record [EHR]) would be expected to substantially improve the

success of implementation efforts. Likewise, availability of data elements to patients and caregivers may represent a barrier to the use of the full precision of estimates available through the decision aid. Although we have provided direction for how patients and caregivers can access the data elements needed to complete the decision aid data entry, it is unlikely that they will have access to all of the information needed to complete the tool. We have made allowance for this likelihood by designing the tool to auto-populate incomplete data fields with population-level means or medians. The less a patient knows about their medical history, the more their calculated estimates will resemble those of the general population (ie, less individualized—and less accurate—for the individual patient). To the extent that patients and caregivers know their medical history, the tool will individualize their risk away from the (broadly available) population average risks.

Subpopulation Considerations The ADVICE decision aid goes beyond subpopulation estimates to provide individualized estimates for patients. This tool was created in large part because of the inadequacies of subpopulation estimates for both patient counseling and risk prediction. Each of the patient characteristics included in the final risk models was drawn from a list of candidate variables that were prespecified in the analysis plan. The variable selection process was prespecified, as well.

Study Limitations Beyond the implementation challenges listed previously, the primary limitation of the ADVICE decision aid is the inaccuracy of the underlying models. While model calibration was excellent for each of the models, discrimination was not. Good calibration indicates that the model’s ability to assign accurate probabilities of a stroke occurring is high; however, poor model discrimination indicates that the model’s ability to rank multiple patients from low to high risk of stroke is more limited. This is expected to affect the precision of the risk estimate rather than the relative comparison of the treatments (SAVR vs TAVR). It is unclear at this point whether the model discrimination issues (as assessed by the C statistic) were due to a lack of important patient characteristics at the time of the procedure or whether 1-year outcomes are

simply more dependent on circumstances and characteristics that materialize during or after the performance of the AV replacement. Certainly, we know that some elements (such as markers of patient frailty and conditioning) were not available during the development of these models. As this field progresses and we gain a better understanding of the importance of additional patient characteristics on a patient’s long-term outcomes, these models will be updated. Likewise, as technology and technique improve, estimates of risk will need to be updated.

Reviewers noted that we may have introduced some forms of bias by including older data for SAVR patients than for TAVR patients in the comparative effective research aim (aim 1) as well as the model-building aim (aim 2). It is unclear what the effect of this decision would be, but given relatively stable outcomes for SAVR procedures, the effect is not expected to be large. Although we would have preferred to use fully overlapping cohorts for SAVR and TAVR, there were not good alternatives to the approach that we selected. This is discussed further in the Limitations section of the related publication.1

Additionally, implementation of this decision aid has not been tested. It remains unclear whether this information will be generally interpretable or useful to patients as they make their treatment decisions. Further usability testing (among patients, caregivers, and specific patient subsets) should be conducted to better direct broad implementation of this tool.

Finally, model development and validation are typically performed using separate data sets. In this case, there was not sufficient sample size to allow for testing the calibration and discrimination of the risk models in independent samples. For this reason, a portion of the testing sample included patients from the development cohort. This may have led to overestimating the expected fit of the risk prediction models in a truly independent model- testing sample. Future updates to the models should include independent development and testing samples when possible.

Aim 3—Educational Resource

Context for Study Results One of the general objectives of the ADVICE study was the creation of educational resources for patients, caregivers, and providers. The primary comparative analysis of the ADVICE study (aim 1) was tailored toward providers as an extension of existing research comparing SAVR vs TAVR in real-world practice in a broadened cohort of patients. Although the clinical impact of the ADVICE study decision aid (aim 2) has not been tested, it was developed to serve as a resource to facilitate shared decision-making. The decision aid was built to be comprehensible and usable by the entire community affected by AV disease, including patients, caregivers, and providers. Further testing is needed to validate clinical utility. In contrast, the ADVICE study educational website (aim 3) was designed by and for patients and their caregivers. Its development was facilitated by a team that included patients, caregivers, physician educators, researchers, and communications specialists. It is intended as a mode of peer-to-peer communication for patients considering treatment for AV stenosis, and it addresses issues unique to all stages of the process, from diagnosis through recovery.

Generalizability of the Findings The educational website is applicable to most patients with AV stenosis. However, it is primarily geared toward educating patients who have a clear treatment choice between SAVR and TAVR. In today’s practice, these patients make up >90% of the cohort of patients with AV stenosis. Through this resource, patients are directed to a deeper understanding of the coming treatment path. Both the process of development and the content of this resource can and should serve as a guide for other disease states and treatment decisions.

Implementation of Study Results The ADVICE study educational website is a public resource, available to patients and caregivers in the comfort of their homes. The website is best accessed by patients and caregivers outside of the clinical encounter—when there is time to explore the resource without the constraints of time. Ideally, this resource would be accessed by patients and

caregivers after diagnosis and before meeting with an AV specialist, and the information would help direct the discussion. To facilitate this process, our team has developed informational cards and a “quick-reference sheet” (available through the “Resources” tab on the website) that can be printed by a patient’s primary care provider or primary cardiologist, directing the patient to the website for more information. Incorporation of this resource (particularly the brochure available on the website) into the EHR after visit summary would allow a systematic approach to the implementation of this process.

Subpopulation Considerations The educational website does not address any specific subpopulation issues.

Study Limitations Limitations of the educational resource are primarily limited to its comprehensiveness, accessibility, and static nature. This resource was developed by a diverse group of stakeholders; however, despite the diversity of backgrounds, there may be issues of interest to some patients and caregivers that are not addressed (eg, expanded information about postoperative recovery, medication precautions). In recognition of this, the website provides a series of links to other web-based resources available to patients and caregivers. The second primary limitation pertains to accessibility. Although a quick-reference sheet is available to those without internet access, use of the website does require access to the internet. Because AV disease often affects older individuals, accessing this resource may be best facilitated by the patient’s family or other caregivers. Creation of a printable version of the information on the website may help facilitate this process and would be a future need for increasing the usability of this resource. Another access-related limitation of the resource pertains to language barriers. At this time, this resource has not been translated into languages other than English. Finally, the static nature of the information on this website is a limitation. Although much of the information is timeless, some of the discussion of the risks and benefits of SAVR and TAVR, as well as the discussion of treatment and recovery expectations, may change over time. To address this limitation and remain useful, this website will need to be periodically updated.

Future Research The next steps in this project will include (1) extension of the current tools to a longer- term horizon (3-5 years), (2) incorporation of additional outcomes (such as subacute nursing facility use), and (3) evaluation of the clinical utility of the ADVICE study decision aids (eg, educational website and personalized risk assessment tool). Currently, the comparative analysis and decision aid are designed to evaluate outcomes to 1-year posttreatment. However, as TAVR continues to be implemented in lower-risk patients, a longer-term horizon for evaluation has become increasingly important. In terms of further evaluation efforts, a better understanding of how patients and physicians interpret “close call” situations where there is not a clinically notable difference between treatments across certain outcomes would be useful.

Assuming these tools prove to be clinically useful, we will then focus on dissemination and implementation efforts. Dissemination efforts should focus on reaching 3 groups: (1) patients and their caregivers, (2) health care providers, and (3) payers. Dissemination efforts to patients and caregivers should target those at risk for AV disease (primarily older individuals through coordinated outreach with organizations like AARP) and those with (or caring for others with) newly diagnosed AV disease via (1) advertisements with online valvular heart disease forums, (2) search engine optimization, and (3) health care provider outreach. Dissemination to health care providers can occur through many venues, including professional societies (eg, ACC, STS, the American Heart Association), device manufacturers, health systems, and payers; this type of dissemination may take the form of direct-to-provider communication or through health care system implementation programs. Another alternative dissemination mechanism is through payer organizations. Beyond the obvious public health benefit of accurately matching patients with treatments, payers stand to benefit from improved outcomes through shared decision-making implementation. Through local and national coverage determinations, payers can encourage the use of tools that help facilitate a process involving open communication of personalized risks and benefits and matching treatments to patients’ individual goals and values.

Through our PCORI clinical effectiveness research grant funding, we developed both an educational resource (http://www.valveadvice.org) and a decision aid for patients who are newly diagnosed with AV disease. These tools are designed to promote a process of shared decision-making when considering SAVR or TAVR. Although patients and caregivers can obtain estimates of their expected outcomes with incomplete information, the most accurate estimates require entry of >30 data elements, which requires in-depth knowledge of a patient’s clinical history with data elements drawn from various time points throughout the medical record. This process is time-consuming for clinicians and may be prohibitive for patients and caregivers. The next step in implementation efforts will focus on developing EHR applications to create automated links to both the website and the decision aids, thereby prepopulating data elements in the decision aids. This next step is expected to improve the efficiency and accuracy of data entry and subsequently foster a more robust process of shared decision-making in this field.

CONCLUSIONS

TAVR is a viable alternative to SAVR for many patients with AV stenosis. Through the PCORI ADVICE study, we explored the risks and benefits of this procedure, and developed an educational website to help patients and caregivers better understand the relative risks and benefits of TAVR vs SAVR. Further work is anticipated to refine the available tools, particularly as procedural techniques, device technologies, and indications evolve.

REFERENCES

1. Brennan JM, Thomas L, Cohen DJ, et al. Transcatheter versus surgical aortic valve replacement: propensity-matched comparison. J Am Coll Cardiol. 2017;70(4):439-450. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5826727/

2. Shahian DM, Jacobs JP, Edwards FH, et al. The Society of Thoracic Surgeons National Database. Heart. 2013;99(20):1494-1501.

3. Carroll JD, Edwards FH, Marinac-Dabic D, et al. The STS-ACC Transcatheter Valve Therapy National Registry: a new partnership and infrastructure for the introduction and surveillance of medical devices and therapies. J Am Coll Cardiol. 2013;62(11):1026- 1034.

4. Green CP, Porter CB, Bresnahan DR, Spertus JA. Development and evaluation of the Kansas City Cardiomyopathy Questionnaire: a new health status measure for heart failure. J Am Coll Cardiol. 2000;35(5):1245-1255.

5. Pettersen KI, Reikvam A, Rollag A, Stavem K. Reliability and validity of the Kansas City Cardiomyopathy Questionnaire in patients with previous myocardial infarction. Eur J Heart Fail. 2005;7(2):235-242.

6. Jacobs JP, Edwards FH, Shahian DM, et al. Successful linking of the Society of Thoracic Surgeons Adult Cardiac Surgery Database to Centers for Medicare & Medicaid Services Medicare data. Ann Thorac Surg. 2010;90(4):1150-1157.

7. Hammill BG, Hernandez AF, Peterson ED, Fonarow GC, Schulman KA, Curtis LH. Linking inpatient clinical registry data to Medicare claims data using indirect identifiers. Am Heart J. 2009;157(6):995-1000.

8. Rothman KJ. Epidemiology: An Introduction. Oxford University Press; 2012.

9. Marinac-Dabic D. Postmarket update: focus on TVT Registry. Published June 13, 2012. Accessed November 16, 2018. https://wayback.archive- it.org/7993/20170404142130/https://www.fda.gov/downloads/AdvisoryCommittees/C ommitteesMeetingMaterials/MedicalDevices/MedicalDevicesAdvisoryCommittee/Circul atorySystemDevicesPanel/UCM308413.pdf

10. FDA expands use of Sapien 3 artificial heart valve for high-risk patients. News release. FDA; June 5, 2017. Accessed November 16, 2018. https://www.fda.gov/news- events/press-announcements/fda-expands-use-sapien-3-artificial-heart-valve-high-risk- patients

11. Rosenbaum PR. Observational Studies. 2nd ed. Springer-Verlag; 2002.

12. Austin PC. A comparison of 12 algorithms for matching on the propensity score. Stat Med. 2014;33(6):1057-1069.

13. Austin PC. Using the standardized difference to compare the prevalence of a binary variable between two groups in observational research. Commun Stat Simul Comput. 2009;38(6):1228-1234.

14. Prasad V, Jena AB. Prespecified falsification end points: can they validate true observational associations? JAMA. 2013;309(3):241-242.

15. Dharmarajan K, Hsieh AF, Lin Z, et al. Diagnoses and timing of 30-day readmissions after hospitalization for heart failure, acute myocardial infarction, or pneumonia. JAMA. 2013;309(4):355-363.

16. Ottenbacher KJ, Karmarkar A, Graham JE, et al. Thirty-day hospital readmission following discharge from postacute rehabilitation in fee-for-service Medicare patients. JAMA. 2014;311(6):604-614.

17. Baiocchi M, Cheng J, Small DS. Instrumental variable methods for causal inference. Stat Med. 2014;33(13):2297-2340.

18. Cox DR. Regression models and life-tables. J R Stat Soc Series B Stat Methodol. 1972;34(2):187-220.

19. Efroymson M. Multiple regression analysis. In: Mathematical Methods for Digital Computers. Wiley; 1960.

20. Lin DY, Wei LJ. The robust inference for the Cox proportional hazards model. J Am Stat Assoc. 1989;84(408):1074-1078.

21. Peterson ED, Dai D, DeLong ER, et al. Contemporary mortality risk prediction for percutaneous coronary intervention: results from 588,398 procedures in the National Cardiovascular Data Registry. J Am Coll Cardiol. 2010;55(18):1923-1932.

22. Steyerberg EW. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. Springer; 2009.

23. Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128- 138.

24. Grunkemeier GL, Jin R, Eijkemans MJC, Takkenberg JJM. Actual and actuarial probabilities of competing risks: apples and lemons. Ann Thorac Surg. 2007;83(5):1586- 1592.

25. Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94(446):496-509.

26. Aitkin M. Statistical Modelling in GLIM. Clarendon Press; 1989.

27. Liang KL, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73(1):13-22.

28. Arnold SV, Spertus JA, Vemulapalli S, et al. Quality-of-life outcomes after transcatheter aortic valve replacement in an unselected population: a report from the STS/ACC Transcatheter Valve Therapy Registry. JAMA Cardiol. 2017;2(4):409-416.

29. Washington State Health Care Authority. Patient decision aids (PDAs). Accessed November 16, 2018. https://www.hca.wa.gov/about-hca/healthier-washington/patient- decision-aids-pdas

30. Pencina MJ, D’Agostino RB. Evaluating discrimination of risk prediction models: the C statistic. JAMA. 2015;314(10):1063-1064.

31. Arnold SV, Cohen DJ, Dai D, et al. Predicting quality of life at 1 year after transcatheter aortic valve replacement in a real-world population. Circ Cardiovasc Qual Outcomes. 2018;11(10):e004693. doi: 10.1161/CIRCOUTCOMES.118.004693

32. Food and Drug Administration. Paving the Way for Personalized Medicine: FDA’s Role in a New Era of Medical Product Development. US Food and Drug Administration; 2013. Accessed August 18, 2020. https://www.fdanews.com/ext/resources/files/10/10-28-13- Personalized-Medicine.pdf

RELATED PUBLICATION

Brennan JM, Thomas L, Cohen DJ, et al. Transcatheter versus surgical aortic valve replacement: propensity-matched analysis from two United States registeries. J Am Coll Cardiol. 2017 Jul 25;70(4):439-450. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5826727/ Supplemental Appendix: NIHMS943753-supplement-Appendix.pdf

ACKNOWLEDGMENTS

Special thanks to DCRI Communications for their help in editing and formatting this report—in particular, Jenny Walker, MLS, senior medical writer; and Kerry Stenke, graphic designer.

APPENDICES Appendix 1. PCORI AV Replacement Study Recruitment Brochure Appendix 2. Covariate Definitions Appendix 3. PCORI ADVICE Study Meeting Slides Appendix 4. ADVICE Quick Reference For Patients Appendix 5. ADVICE Website Informational Card for Providers Appendix 6. Supplementary Methods, Tables, and Figures

Appendix 1. PCORI AV Replacement Study Recruitment Brochure

Are you a PATIENT or

Pro00062757 CAREGIVER of a

IN COLLABORATION WITH patient who has

Aortic Valve Replace- undergone aortic ment Patient-Centered valve replacement? Outcomes Research We need you to assist

Study researchers in understanding AND A Patient-Centered your experience in order to design tools that may help Outcomes Research future aortic valve replace- Institute and Duke ment patients and their loved Clinical Research ones have better information Institute Initiative with which to make treatment decisions.

SPONSORED BY CONDUCTED BY

WHAT IS THIS STUDY ABOUT?

The Patient-Centered Outcomes Research Institute (PCORI) Aortic Dr. Brennan is an Valve Replacement (AVR) Research Assistant Professor of Study was designed to develop AVR Medicine and an Inter- education resources for patients, CONTACT US ventional Cardiologist caregivers, and healthcare providers. at Duke University PCORI AVR’s ultimate goal is to help For more information, please Medical Center. He is AVR patients, caregivers, and provid- contact the PCORI AVR study also a member of the ers navigate a wide range of complex team by emailing: Health Services and choices with regard to AVR treatment Outcomes Research Group at the options by providing the resources [email protected] Duke Clinical Research Institute. necessary to make better informed or calling: (919) 668-8670 care decisions for individual patients.

HOW MUCH TIME WILL I NEED TO DEVOTE EACH MONTH?

We estimate that focus group participants will spend approximately 1–2 hours reviewing documents and participating in the focus group discussion. Surveys are generally designed to take about 30 minutes to complete. WILL I BE COMPENSATED FOR MY TIME?

Yes! Focus group participants will WHAT IS PCORI?

be compensated for participating in our study. PCORI is an independent, non-profit organization that was established in ARE THERE ANY RISKS? 2010 by the United States Congress WHAT WOULD BE MY ROLE? to facilitate patient-centered re- We foresee no risks involved in As a study participant, we will ask search. PCORI funds are distributed study participation because: that you provide your perspective nationwide across a variety of (as a patient or as a patient caregiv- • We will not ask personal ques- projects. Each initiative focuses on er) including what questions our tions (for example, about your the treatment of a unique disorder study needs to address, the pro- health). or condition. posed methodology, and sugges- • The study is free (no cost to you). Our PCORI AVR study aims to tions about print material revisions. • There are no medical procedures improve the care and treatment for or drugs involved. The input that you provide may be AVR patients worldwide. as simple as educating our team about your experience navigating the AVR treatment process and, as We hope that by engaging you feel comfortable, your input interested aortic valve disease may help guide methodology plan- patients and their caregivers ning and results interpretation. Ad- ditionally, you may be asked to con- in our study team, the results tribute input into the design of edu- of our research will be more cational materials to help future pa- useful to others who will tients and caregivers. follow in your footsteps.

Appendix 2. Covariate definitions

Covariate Definition Age, median (IQR) Patient's age in years, at time of surgery Female Patient with female gender identified at birth Race White Patient’s race, as determined by the patient or family; includes white Black Patient's race, as determined by the patient or family; includes black/African American. This includes a person having origins in any of the black racial groups of Africa. Terms such as "Haitian" or "negro" can be used in addition to "black or African American."

Hispanic Patient's race, as determined by the patient or family; includes Hispanic, Latino, or Spanish ethnicity

Other Patient's race, as determined by the patient or family; does not include white, black, or Hispanic

Commercial insurance Commercial insurance refers to all indemnity (fee-for-service) carriers and preferred provider organizations (eg, Blue Cross and Blue Shield)

Body surface area, Calculated as SQRT: (height [cm] x weight [kg]/3600) median (IQR)

Dialysis Patient is currently undergoing dialysis, including hemodialysis, peritoneal dialysis, and continuous veno-venous hemofiltration

Prior MI Patient has had at least 1 documented previous MI at any time before this surgery

Recent STS: MI occurred less than or equal to 21 days previous TVT: MI occurred less than 30 days previous

Old STS: MI occurred greater than 21 days ago previous TVT: MI occurred greater than or equal to 30 days ago previous

Prior CV surgeries Patient had open-heart cardiac surgeries before this procedure. This includes open-heart coronary artery bypass, or valve replacement/repairs.

Resuscitation STS: Patient required CPR within 1 hour before the start of the operative procedure, which includes the institution of anesthetic management. TVT: "Sudden" cardiac arrest is the sudden cessation of cardiac activity so that the victim becomes unresponsive, with no normal breathing and no signs of circulation. If corrective measures are not taken rapidly, this condition progresses to sudden death. Cardiac arrest should be used to signify an event as described above that is reversed, usually by CPR, and/or defibrillation or cardioversion, or cardiac pacing. Sudden cardiac death should not be used to describe events that are not fatal.

Covariate Definition Creatinine clearance, Indicate the creatinine level closest to the date and time prior to the median (IQR) procedure, but prior to anesthetic management in mg/dL.

Preoperative atrial Indicate whether atrial fibrillation or flutter was present within 30 days of the fibrillation/flutter procedure.

LVEF, median (IQR) Most recent determination before the surgical intervention, documented on a diagnostic report. If a percentage range is reported, whole-number mean is reported. Heart failure symptoms Physician documentation that the patient has been in a state of heart failure less than 2 weeks within the past 2 weeks. Heart failure is defined as physician documentation or previous report of clinical symptoms of heart failure, defined as unusual dyspnea on light exertion, recurrent dyspnea occurring in the supine position, fluid retention; or the description of rales, jugular venous distension, pulmonary edema on physical exam, or pulmonary edema on chest x-ray, presumed to be cardiac dysfunction. A low ejection fraction alone, without clinical evidence of heart failure does not qualify as heart failure.

None or Class I Patient has cardiac disease but without resulting limitations of ordinary physical activity. Ordinary physical activity (eg, walking several blocks or climbing stairs) does not cause undue fatigue, palpitation, dyspnea, or anginal pain. Limiting symptoms may occur with marked exertion.

Class II Patient has cardiac disease resulting in slight limitation of ordinary physical activity. Patient is comfortable at rest. Ordinary physical activity such as walking more than 2 blocks or climbing more than 1 flight of stairs results in limiting symptoms (eg, fatigue, palpitation, dyspnea, or anginal pain).

Class III Patient has cardiac disease resulting in marked limitation of physical activity. Patient is comfortable at rest. Less than ordinary physical activity (eg, walking 1-2 level blocks or climbing 1 flight of stairs) causes fatigue, palpitation, dyspnea, or anginal pain.

Chronic lung disease Patient has a history of chronic lung disease with severity documented as one of the following:

None No documented chronic lung disease.

Mild Mild: FEV1 60% to 75% of predicted, and/or on chronic inhaled or oral bronchodilator therapy.

Moderate Moderate: FEV1 50% to 59% of predicted, and/or on chronic steroid therapy aimed at lung disease.

Severe Severe: FEV1 less than 50% predicted, and/or room air pO2 less than 60 or room air pCO2 greater than 50.

Covariate Definition

Home oxygen use Patient uses supplemental oxygen at home. Prior stroke STS: Patient has a history of stroke (ie, any confirmed neurological deficit of abrupt onset caused by a disturbance in blood flow to the brain) that did not resolve within 24 hours. TVT: Defined as an acute episode of focal or global neurological dysfunction caused by brain, spinal cord, or retinal vascular injury as a result of hemorrhage or infarction. Cerebrovascular disease STS: Patient has a history of loss of neurological function that was abrupt in onset without prior CVA but with complete return of function within 24 hours. TVT: Patient has a history of a transient ischemic attack, defined as a transient episode of focal neurological dysfunction caused by brain, spinal cord, or retinal ischemia, without acute infarction.

Diabetes History of diabetes mellitus according to the American Diabetes Association criteria regardless of duration of disease or need for antidiabetic agents. None No treatment for diabetes.

Insulin Insulin treatment (includes any combination with insulin).

Non-insulin Oral agent treatment (includes oral agent with/without diet treatment). CAD: # diseased Number of diseased major native coronary vessel systems: LAD system, vessels circumflex system, and/or right system with greater than or equal to 0% narrowing of any vessel preoperatively. Left main disease (greater than or equal to 50%) is counted as 2 vessels (LAD and circumflex, which may include a ramus intermedius). For example, left main and RCA would count as 3 total.

Left main CAD Left main CAD is present when there is greater than or equal to 50% compromise of vessel diameter preoperatively.

Preoperative Patient had a mechanical assist device in place at the start of the IABP/inotropes procedure/medications the patient received 24 hours before the procedure.

Covariate Definition

Hypertension Patient has a diagnosis of hypertension, documented by one of the following: a) Documented history of hypertension diagnosed and treated with medication, diet, and/or exercise. b) Documentation of prior blood pressure greater than 140 mm Hg systolic or 90 mm Hg diastolic for patients without diabetes or chronic kidney disease, or documentation of prior blood pressure greater than 130 mm Hg systolic or 80 mm Hg diastolic on at least 2 occasions for patients with diabetes or chronic kidney disease. c) Currently on pharmacologic therapy to control hypertension.

Immunosuppression Indicate whether immunocompromise is present due to immunosuppressive medication therapy within 30 days preceding the operative procedure or existing medical condition (see training manual). This includes, but is not limited to, systemic steroid therapy, antirejection medications, and chemotherapy. This does not include topical steroid applications, 1-time systemic therapy, inhaled steroid therapy, or preoperative protocol. Aortic insufficiency Highest severity of aortic insufficiency between 12 months before the procedure (moderate/severe) and start of the procedure. Moderate: Qualitative measurements—angiographic grade of 2+; color Doppler jet width greater than mild but no signs of severe aortic regurgitation (insufficiency); Doppler vena contracta width 0.3–0.6 cm. Quantitative measurements (catheterization or echocardiogram)—regurgitant volume 30-59 mL/beat; regurgitant fraction 30–49%; regurgitant orifice area 0.10–0.29 cm2. Severe: Qualitative measurements—angiographic grade of 3–4+; color Doppler jet width (central jet) greater than 65% of LVOT; Doppler vena contracta width greater than 0.6 cm. Quantitative measurements (catheterization or echocardiogram)— regurgitant volume greater than or equal to 60 mL/beat; regurgitant fraction greater than or equal to 50%; regurgitant orifice area greater than or equal to 0.30 cm2.

Mitral insufficiency Severity of mitral valve regurgitation according to the American Society of (moderate/severe) Echocardiography Guidelines integrated approach.

Tricuspid insufficiency Evidence of tricuspid valve regurgitation. (moderate/severe)

ICD Patient had a previous implant of an ICD. This does not include lead placement only.

Covariate Definition

Prior PCI Previous PCI was performed any time prior to this surgical procedure. PCI refers to those treatment procedures that unblock narrowed coronary arteries without performing surgery. PCI may include, but is not limited to: Balloon catheter angioplasty, percutaneous transluminalcoronary angioplasty 1) Rotational atherectomy 2) Directional atherectomy 3) Extraction atherectomy 4) Laser atherectomy 5) Intracoronary stent placement Peripheral vascular Patient has a history of peripheral arterial disease (includes upper and lower disease extremity, renal, mesenteric, and abdominal aortic systems). This can include: 1) Claudication either with exertion or at rest 2) Amputation for arterial vascular insufficiency 3) Vascular reconstruction, bypass surgery, or percutaneousintervention to the extremities (excluding dialysis fistulas and vein stripping) 4) Documented aortic aneurysm, with or without repair 5) Positive noninvasive test (eg, ankle brachial index less than or equal to 0.9, ultrasound, magnetic resonance, or computed tomography imaging of greater than 50% diameter stenosis in any peripheral artery (ie, renal, subclavian, femoral, iliac) or angiographic imaging. Peripheral arterial disease excludes disease in the carotid or cerebrovascular arteries. Aortic valve mean Highest mean gradient (in mm Hg) across the aortic valve obtained from an gradient, median (IQR) echocardiogram or angiogram preoperatively between 12 months before and start of procedure. Status (elective, Clinical status of the patient before entering the operating room: urgent, emergent, Elective—the patient's cardiac function has been stable in the days or weeks emergent salvage) before the operation. The procedure could be deferred without increased risk of compromised cardiac outcome. Urgent—procedure required during same hospitalization in order to minimize chance of further clinical deterioration. Examples include but are not limited to: worsening, sudden chest pain, CHF, AMI, anatomy, IABP, unstable angina with IV NTG, or rest angina. Emergent—patients requiring emergency operations will have ongoing, refractory (difficult, complicated, and/or unmanageable) unrelenting cardiac compromise, with or without hemodynamic instability, and not responsive to any form of therapy except cardiac surgery. An emergency operation is one in which there should be no delay in providing operative intervention. Emergent salvage—patient is undergoing CPR en route to the OR or before anesthesia induction or has ongoing ECMO to maintain life.

Covariate Definition

Hematocrit Preoperative hematocrit level at the date and time closest to surgery but before anesthetic management (induction area or OR).

Preop total albumin, The total albumin closest to the date and time before surgery but before anesthetic median (IQR) management (induction area or operating room).

Prior CABG Patient had a previous CABG before the current admission. Patient had a previous surgical aortic valve replacement before the current admission. Prior AVR Patient had a previous AVR.

PA systolic pressure, Highest PA systolic pressure recorded before SAVR or TAVR procedure. median (IQR)

Mitral stenosis Patient has mitral stenosis present.

Cardiac presentation Worst type of angina present before this procedure: 1) no symptoms; 2) symptoms unlikely to be ischemia; 3) stable angina; 4) unstable angina; 5) non-ST Elevation MI; 6) ST elevation MI

AMI, acute myocardial infarction; CABG, coronary artery bypass graft; CAD, coronary artery disease; CHF, congestive heart failure; CPR, cardiopulmonary resuscitation; CV, cardiovascular; CVA, cerebrovascular accident; ECMO, extracorporeal membrane oxygenation; FEV1, forced expiratory volume in 1 second; IABP, intra-aortic balloon pump; ICD, implantable cardioverter defibrillator; IQR, interquartile range; IV, intravenous; LAD, left anterior descending; LVEF, left ventricular ejection fraction; LVOT, left ventricular outflow tract; MI, myocardial infarction; NTG, nitroglycerin; OR, operating room; PA, pulmonary artery; PCI, percutaneous coronary intervention; PCO2, partial pressure of carbon dioxide; PO2, partial pressure of oxygen; Preop, preoperative; RCA, right circumflex artery; SQRT, square root; STS, Society of Thoracic Surgeons; TVT, Transcatheter Valve Therapy (Registry); USA, unstable angina

Appendix 3. PCORI ADVICE study meeting slides

ANNUAL OCT. 31-NOV. 2, 2017 MEETING ARLINGTON, VA

Implementing Shared Decision-Making for Patients with Aortic Valve Disease

J. Matthew Brennan, MD MPH Susan Strong Interventional Cardiology Patient Representative Denver, CO Duke Triangle Heart Associates

Associate Professor of Medicine

Duke University School of Medicine

November 1, 2017 #PCORI2017

ANNUAL MEETING | #PCORI2017

Overview

• Patient’s Perspective: Personalizing the Treatment of Aortic Valve Disease

• PCORI ADVICE Study: Developing Decision Aids for Aortic Valve Disease

ANNUAL MEETING | #PCORI2017 Aortic Valve Disease: What are the treatment options?

Surgical Valve Replacement Transcatheter Valve Replacement (SAVR) (TAVR)

ANNUAL MEETING | #PCORI2017

The Counterfactual

ANNUAL MEETING | #PCORI2017

PCORI ADVICE Study: Data Sources & Design • Procedural registry data - STS National Database - STS/American College of Cardiology (ACC) Transcatheter Valve Therapy (TVT) Registry • Follow-up: Medicare claims • Observational Study (non-randomized) • Propensity-matched cohort

ANNUAL MEETING | #PCORI2017

Exclusions: Clinical Equipoise No TAVR No SAVR

 Age <65 yrs Age >90 yrs  Endocarditis Hostile chest  Emergency/ Salvage Aortic Insufficiency Other without stenosis Centers with <10 Moderate/Severe TAVR cases Mitral Stenosis Low risk for SAVR (STS PROM <3%)

Spectrum of Patient Risk

ANNUAL MEETING | #PCORI2017

Assessing Balance Measured Confounders Unmeasured Confounders Propensity Matching Falsification Outcome (UTI)

HR 1.45 (1.04-2.02) p=0.028

SAVR TAVR

ANNUAL MEETING | #PCORI2017

7 (Nov 2011 – June 2014)

ANNUAL MEETING | #PCORI2017 Assessing Balance

Urinary Tract Infection

SAVR TAVR

8 (Jan 2014 – Q3 2015)

ANNUAL MEETING | #PCORI2017

PCORI ADVICE Study: Risk Models (SAVR vs TAVR)

5 Clinical Outcomes Model Parameters 1. Discharge to home • Demographics 2. Death, 1 year • Medical History 3. Stroke, 1 year • Labs 4. Days alive and out of hospital, 1st year • Testing 5. Alive and better, 1 year • Clinical Status

Mortality – 1 year Stroke – 1 year “Alive and Better”

ANNUAL MEETING | #PCORI2017

PCORI ADVICE Study: Decision Assistance Tool

ANNUAL MEETING | #PCORI2017

Questions?

J. Matthew Brennan, MD MPH Interventional Cardiology Duke Triangle Heart Associates Associate Professor of Medicine Duke University School of Medicine [email protected]

November 1, 2017 11 ANNUAL MEETING | #PCORI2017

Appendix 4. ADVICE quick-reference for patients

What is the aortic valve? NORMAL AORTIC VALVE Simply put, the heart is a pump – moving blood through the lungs and to the body. To keep blood OPEN CLOSED moving in one direction, the heart has a series of four one-way valves (like doorways). The aortic valve is the last in the series, and it is the last stop for blood on the way out of the heart.

A healthy aortic valve is flexible and strong with three leaflets. As the heart pumps, the aortic valve opens—allowing blood to pass effortlessly to the body. DISEASED AORTIC VALV As the heart relaxes, the aortic valve closes abruptly – OPEN CLOSED stopping the flow of blood back into the heart.

“I had no symptoms or problems and was surprised to learn that my aortic valve was more than 90% blocked . I would encourage all adults to have an annual physical exam.” Stewart Gully, Surgical Aortic Valve Replacement Patient Symptoms In some patients, a heart murmur might What is aortic valve disease? indicate aortic valve disease. However, not all Aortic valve disease simply indicates an abnormality of people will experience symptoms and for some the aortic valve or the structure on which the valve sits. people, symptoms are not present until the Aortic valve disease may be the result of a birth defect, disease has progressed to the point that blood such as a bicuspid valve. The disease may be acquired flow is significantly reduced. following rheumatic fever or as a consequence of aging. Aortic valve stenosis indicates a ‘hardening’ • Chest pain (also referred to as angina) or narrowing of the valve. Over time, calcium deposits • Dyspnea (difficulty or labored breathing) within the valve leaflets and the valve turns to a hard • Syncope (black-out or fainting spells) substance, like bone. As the valve hardens, it does • Fatigue especially during exertion that not open normally and becomes increasingly difficult previously was effortless for the heart to push blood past the valve and out to • Palpitations (erratic heart beats) the body.

TREATMENT OPTIONS Two approaches-surgical or transcatheter- currently exist to replace diseased valves.

1. SAVR 2. TAVR

Surgical aortic valve replacement (SAVR) is an Transcatheter aortic valve replacement (TAVR) open-chest surgery where the surgeon removes is a relatively new technique that allows the diseased aortic valve and sews in a new one in implantation of a new valve through a very small its place. Less invasive surgical techniques have hole in the leg or chest wall. Unlike a surgical been developed to allow the surgeon to perform aortic valve replacement, a TAVR procedure does this procedure through a smaller incision at the not remove the diseased aortic valve – rather the top or to the right side of the chest wall. Both old valve is pushed aside, opening space to place metal (mechanical) and tissue valve prostheses a new valve. Only tissue valve prostheses can be can be inserted using a surgical approach. inserted using a transcatheter approach.

Making a Decision About Treatment Based on What is Important to You Today, many patients have the option of having their aortic valve replaced using either a surgical or transcatheter approach. Understanding your personal preferences and risks when deciding between these two treatments will help you to choose the option that is right for you.

An application will be available soon that estimates likelihood of death, stroke, being discharged to home, and number of days out of the hospital in the first year after treatment with each of the two treatment options.

Patients can work through the Aortic Valve LET’S GET STARTED ADVICE Calculator on their own or with their doctor. Check out www.valveadvice.org to learn more about aortic valve disease

Watch for the Aortic Valve ADVICE Calculator coming soon.

This work is funded through a Patient-Centered Outcomes Research Institute (PCORI) Award (CER- 1306-04350).

Appendix 5. ADVICE website informational card for providers

OPTIMIZING OUTCOMES THROUGH PERSONALIZED MEDICINE

Today, many patients have the option of treating their aortic valve disease with either a surgical or transcatheter approach. With new web-based tools, you can help your patients choose the option that is right for them.

In collaboration with the Society of Thoracic Surgeons (STS) and the American College of Cardiology (ACC), the Aortic Valve ADVICE team Patients can work through the has developed a personalized risk calculator and Aortic Valve ADVICE calculator an educational website for your patients with on their own or with their doctor. aortic valve disease.

The Aortic Valve ADVICE Calculator creates a personalized estimate of several clinical outcomes in the first year after valve replacement with either surgical or transcatheter approach.

Helping find the www.valveadvice.org best choice for you. Watch for the Aortic Valve Advice Share this website with your patients Calculator App coming soon. to learn more about aortic valve disease.

Appendix 6. Supplementary Methods, Tables, and Figures

I. METHODS

Methods for Covariate Adjustment Propensity score matching was selected a priori as the method for confounding adjustment. Other methods such as inverse probability of treatment weighting and regression adjustment were explored in preliminary analyses of the entire population on the falsification endpoints and mortality (no subgroups), with very similar results. Given the aim to evaluate subgroups, matching provided a transparent approach to identify small subgroups and evaluate covariate balance within subgroups. In addition, the extreme tails of the propensity distribution would be expected to generate large weights for some patients, which could be highly impactful to subgroup analyses (results may be sensitive to the weight of an individual). To preserve transparency and avoid sensitivity to large weights in the subgroup analyses, propensity matching was used throughout. This did not result in the systematic exclusion of any type of patients, as the propensity score distributions were overlapping. This did result in a reduction in sample size; however, the final sample size was still 10 times larger than previous clinical trials. Starting with very large registry samples that happened to include very different patients, propensity matching allowed us to identify a more comparable population that was still relatively large.

The propensity score was calculated by logistic regression for the probability of receiving transcatheter aortic valve replacement (TAVR), given measured covariates. Overlap in the covariate distribution and propensity scores between study groups was assessed. Since patients at the tails (<5%, >95%) of the propensity distribution were thought to represent individuals with an overwhelming likelihood of treatment with one or another of the two treatments, these patients were excluded. Propensity score matching was conducted in a 1:1 ratio, by greedy matching, using a caliper of 0.20 standard deviations in the linear predictor. The adequacy of the propensity model was confirmed by checking covariate balance before and after matching. Details are provided below.

Models for treatment on outcomes were fit to the matched sample using a robust empirical variance to account for matched pairs. Associations were estimated in pre-specified subgroups, along with 95% confidence intervals and tests of interaction. All subgroups that were tested are shown. To address the potential for unmeasured confounding, the analysis was repeated on two falsification endpoints: urinary tract infection and lower extremity fracture.

II. TABLES

Table S1. Covariate Definitions

Covariate STS TVT Age, median (IQR) Patient's age in years, at time of surgery Female Patient with female gender identified at birth Race White Patient’s race, as determined by the patient or family, includes white Black Patient's race, as determined by the patient or family, includes black/African American. This includes a person having origins in any of the black racial groups of Africa. Terms such as "Haitian" or "negro" can be used in addition to "black or African American." Hispanic Patient's race, as determined by the patient or family, includes Hispanic, Latino, or Spanish ethnicity Other Patient's race, as determined by the patient or family, does not include white, black, or Hispanic Commercial insurance Commercial insurance refers to all indemnity (fee-for-service) carriers and Preferred Provider Organizations (e.g., Blue Cross and Blue Shield). Body surface area, median Calculated as SQRT: (height [cm] x weight [kg]/3600) (IQR) Dialysis Patient is currently undergoing dialysis, including hemodialysis, peritoneal dialysis, and continuous veno-venous hemofiltration Prior MI Patient has had at least one documented previous MI at any time prior to this surgery Recent MI occurred ≤21 days MI occurred <30 days Old MI occurred >21 days MI occurred ≥30 days Prior CV surgeries Patient had open heart cardiac surgeries prior to this procedure. This includes open heart coronary artery bypass, or valve replacement/repairs.

Resuscitation Patient required CPR within one hour "Sudden" cardiac arrest is the sudden cessation of cardiac before the start of the operative activity so that the victim becomes unresponsive, with no procedure which includes the normal breathing and no signs of circulation. If corrective institution of anesthetic management. measures are not taken rapidly, this condition progresses to sudden death. Cardiac arrest should be used to signify an event as described above that is reversed, usually by CPR, and/or defibrillation or cardioversion, or cardiac pacing. Sudden cardiac death should not be used to describe events that are not fatal. Creatinine clearance, median Indicate the creatinine level closest to the date and time prior to the procedure, but prior to anesthetic (IQR) management in mg/dL. Pre-operative atrial Indicate whether atrial fibrillation or flutter was present within thirty days of the procedure. fibrillation/flutter LVEF, median (IQR) Most recent determination prior to the surgical intervention documented on a diagnostic report. If a percentage range is reported, whole number mean is reported. Heart failure symptoms <2 Physician documentation that the patient has been in a state of heart failure within the past 2 weeks. weeks Heart failure is defined as physician documentation or report of any of the following clinical symptoms of heart failure described as unusual dyspnea on light exertion, recurrent dyspnea occurring in the supine position, fluid retention; or the description of rales, jugular venous distension, pulmonary edema on physical exam, or pulmonary edema on chest x-ray presumed to be cardiac dysfunction. A low ejection fraction alone, without clinical evidence of heart failure does not qualify as heart failure. None or Class I Patient has cardiac disease but without resulting limitations of ordinary physical activity. Ordinary physical activity (e.g., walking several blocks or climbing stairs) does not cause undue fatigue, palpitation, dyspnea, or anginal pain. Limiting symptoms may occur with marked exertion. Class II Patient has cardiac disease resulting in slight limitation of ordinary physical activity. Patient is comfortable at rest. Ordinary physical activity such as walking more than two blocks or climbing more than one flight of stairs results in limiting symptoms (e.g., fatigue, palpitation, dyspnea, or anginal pain). Class III Patient has cardiac disease resulting in marked limitation of physical activity. Patient is comfortable at rest. Less than ordinary physical activity (e.g., walking one to two level blocks or climbing one flight of stairs) causes fatigue, palpitation, dyspnea, or anginal pain.

Class IV Patient has dyspnea at rest that increases with any physical activity. Patient has cardiac disease resulting in inability to perform any physical activity without discomfort. Symptoms may be present even at rest. If any physical activity is undertaken, discomfort is increased. Chronic lung disease Patient has a history of chronic lung disease with severity documented as one of the following: None No documented chronic lung disease Mild Mild: FEV1 60% to 75% of predicted, and/or on chronic inhaled or oral bronchodilator therapy. Moderate Moderate: FEV1 50% to 59% of predicted, and/or on chronic steroid therapy aimed at lung disease Severe Severe: FEV1 <50% predicted, and/or Room Air pO2 < 60 or Room Air pCO2 > 50. Home oxygen use Patient uses supplemental oxygen at home. Prior stroke Patient has a history of stroke (i.e., Defined as an acute episode of focal or global neurological any confirmed neurological deficit of dysfunction caused by brain, spinal cord, or retinal vascular abrupt onset caused by a disturbance injury as a result of hemorrhage or infarction. in blood flow to the brain) that did not resolve within 24 hours. Cerebrovascular disease Patient has a history of loss of Patient has a history of a transient ischemic attack, defined as without prior CVA neurological function that was abrupt a transient episode of focal neurological dysfunction caused in onset but with complete return of by brain, spinal cord, or retinal ischemia, without acute function within 24 hours. infarction. Diabetes History of diabetes mellitus according to the American Diabetes Association criteria regardless of duration of disease or need for antidiabetic agents. None No treatment for diabetes Insulin Insulin treatment (includes any combination with insulin). Non-insulin Oral agent treatment (includes oral agent with/without diet treatment). CAD: # diseased vessels Number of diseased major native coronary vessel systems: LAD system, circumflex system, and/or right system with ≥50% narrowing of any vessel preoperatively. Left main disease (≥50%) is counted as TWO vessels (LAD and circumflex, which may include a ramus intermedius). For example, left main and RCA would count as three total. Left main CAD Left main CAD is present when there is ≥50% compromise of vessel diameter preoperatively. Pre-operative IABP/inotropes Patient had a mechanical assist device in place at the start of the procedure. Medications the patient received 24 hours prior to the procedure.

Prior shock Patient was, at the time of procedure, in a clinical state of end organ hypoperfusion due to cardiac failure according to the following criteria: persistent hypotension (systolic blood pressure <80–90 or mean arterial pressure 30 mmhg lower than baseline) and severe reduction in cardiac index (<1.8 without support or <2.2 with support). Hypertension Patient has a diagnosis of hypertension, documented by one of the following: a. Documented history of hypertension diagnosed and treated with medication, diet and/or exercise b. Prior documentation of blood pressure >140 mmHg systolic or 90 mmHg diastolic for patients without diabetes or chronic kidney disease, or prior documentation of blood pressure >130 mmHg systolic or 80 mmHg diastolic on at least 2 occasions for patients with diabetes or chronic kidney disease c. Currently on pharmacologic therapy to control hypertension. Immunosuppression Indicate whether immunocompromise is present due to immunosuppressive medication therapy within 30 days preceding the operative procedure or existing medical condition (see training manual). This includes, but is not limited to systemic steroid therapy, anti-rejection medications, and chemotherapy. This does not include topical steroid applications, one time systemic therapy, inhaled steroid therapy or preoperative protocol. Aortic insufficiency Highest severity of aortic insufficiency between 12 months prior to the procedure and start of the (moderate/severe) procedure. Moderate: Qualitative Measurements: Angiographic grade of 2+; Color Doppler jet width greater than mild but no signs of severe aortic regurgitation (insufficiency); Dopplar vena contracta width 0.3–0.6 cm. Quantitative Measures (cath or echo): Regurgitant volume 30-59 ml/beat; Regurgitant fraction 30–49%;Regurgitant orifice area 0.10–0.29 cm(2) Severe: Qualitative Measurements: Angiographic grade of 3–4+; Color Doppler jet width (Central jet) >65% of LVOT; Dopplar vena contracta width >0.6 cm. Quantitative Measures (cath or echo): Regurgitant volume ≥60 ml/beat; Regurgitant fraction ≥50%; Regurgitant orifice area ≥0.30 cm(2) Mitral insufficiency Severity of mitral valve regurgitation according to the American Society of Echocardiography (moderate/severe) Guidelines integrated approach. Tricuspid insufficiency Evidence of tricuspid valve regurgitation. (moderate/severe) ICD Patient had a previous implant of an ICD. This does not include lead placement only.

Status (elective, urgent) Clinical status of the patient prior to entering the operating room: Elective: The patient's cardiac function has been stable in the days or weeks prior to the operation. The procedure could be deferred without increased risk of compromised cardiac outcome.

Urgent: Procedure required during same hospitalization in order to minimize chance of further clinical deterioration. Examples include but are not limited to: Worsening, sudden chest pain, CHF, acute myocardial infarction (AMI), anatomy, IABP, unstable angina (USA) with intravenous (IV) nitroglycerin (NTG) or rest angina.

Emergent: Patients requiring emergency operations will have ongoing, refractory (difficult, complicated, and/or unmanageable) unrelenting cardiac compromise, with or without hemodynamic instability, and not responsive to any form of therapy except cardiac surgery. An emergency operation is one in which there should be no delay in providing operative intervention.

Emergent Salvage: The patient is undergoing CPR en route to the OR or prior to anesthesia induction or has ongoing ECMO to maintain life. Hematocrit Pre-operative hematocrit level at the date and time closest to surgery but prior to anesthetic management (induction area or operating room). Pre-op total albumin, median The total albumin closest to the date and time prior to surgery but prior to anesthetic management (IQR) (induction area or operating room). Prior CABG Patient had a previous coronary bypass graft prior to the current admission. Prior aortic valve replacement Patient had a previous surgical aortic valve replacement prior to the current admission. PA systolic pressure, median Highest PA systolic pressure recorded prior to SAVR or TAVR procedure. (IQR) Mitral stenosis Patient has mitral stenosis present. Cardiac presentation Worst type of angina present prior to this procedure: 1. No symptoms; 2. Symptoms unlikely to be ischemia; 3. Stable angina; 4. Unstable angina; 5. Non-ST Elevation MI; 6. ST Elevation MI CAD = coronary artery disease; CPR = cardiopulmonary resuscitation; CV = cardiovascular; CVA = cerebrovascular accident; FEV1 = forced expiratory volume in 1 second; IABP = intra-aortic balloon pump; IQR = interquartile range; LAD = left anterior descending; LVEF = left ventricular ejection fraction; MI = myocardial infarction; RCA = right circumflex artery; SQRT = square root; STS = Society of Thoracic Surgeons; TVT = Transcatheter Valve Therapy (Registry)

Table S2. Baseline Characteristics Before and After Matching Continuous variables are reported as medians, along with the 25th and 75th percentiles, and categorical variables include the number and percent in a category. Standardized differences were computed to summarize covariate balance across the SAVR and TAVR interventions.

Before Propensity Matching After Propensity Matching Overall SAVR TAVR SD SAVR TAVR SD TAVR (N=40,528) (N=22,618) (N=17,910) TAVR vs. (N=4,732) (N=4,732) vs. SAVR SAVR Age, median (IQR) 81 (76,85) 80 (75,84) 83 (79,87) 52.87% 82 (77,85) 81 (77,85) -1.01% Female 19,008 (46.9) 10,270 (45.4) 8,738 (48.8) 6.78% 2,278 (48.1) 2,256 (47.7) -0.93% Race 11.57% 2.70% White 37,210 (91.8) 20,708 (91.6) 16,502 (92.1) 4,355 (92.0) 4,354 (92.0) Black 1,332 (3.3) 730 (3.2) 602 (3.4) 151 (3.2) 165 (3.5) Hispanic 1,215 (3.0) 633 (2.8) 582 (3.2) 159 (3.4) 145 (3.1) Other 771 (1.9) 547 (2.4) 224 (1.3) 67 (1.4) 68 (1.4) Commercial insurance 24,612 (60.7) 13,415 (59.3) 11,197 (62.5) 6.58% 2,890 (61.1) 2,861 (60.5) -1.26% Body surface area, median (IQR) 1.9 (1.7,2.0) 1.9 (1.7,2.1) 1.8 (1.7,2.0) -23.17% 1.9 (1.7,2.1) 1.9 (1.7,2.0) 0.04% Dialysis 1,511 (3.7) 752 (3.3) 759 (4.2) 4.79% 186 (3.9) 179 (3.8) -0.77% Prior MI 37.36% 2.21% Recent 2,637 (6.5) 2,188 (9.7) 449 (2.5) 161 (3.4) 173 (3.7) Old 7,853 (19.4) 3,695 (16.3) 4,158 (23.2) 954 (20.2) 924 (19.5) Prior CV surgeries 10,080 (24.9) 4,132 (18.3) 5,948 (33.2) 34.69% 1,484 (31.4) 1,406 (29.7) -3.58% Resuscitation 139 (0.3) 102 (0.5) 37 (0.2) -4.27% 16 (0.3) 15 (0.3) -0.37% Creatinine clearance, median (IQR) 1.1 (0.9,1.4) 1.1 (0.9,1.4) 1.1 (0.9,1.5) 9.89% 1.1 (0.9,1.4) 1.1 (0.9,1.5) -0.32% Pre-operative atrial fibrillation/flutter 13,542 (33.4) 5,135 (22.7) 8,407 (46.9) 52.60% 1,619 (34.2) 1,572 (33.2) -2.10% LVEF, median (IQR) 55.0 (45.0,55.0) 55.0 (50.0,55.0) 55.0 (45.0,55.0) -12.74% 55.0 (45.0,55.0) 55.0 (45.0,55.0) -1.10% Heart failure symptoms <2 weeks 83.18% 4.28% None or Class I 11,246 (27.7) 10,731 (47.4) 515 (2.9) 447 (9.4) 335 (7.1) Class II 5,847 (14.4) 3,336 (14.7) 2,511 (14.0) 947 (20.0) 995 (21.0) Class III 17,282 (42.6) 6,137 (27.1) 11,145 (62.2) 2,499 (52.8) 2,509 (53.0) Class IV 6,153 (15.2) 2,414 (10.7) 3,739 (20.9) 839 (17.7) 893 (18.9) Chronic lung disease 27.62% 1.62%

None 23,777 (58.7) 14,719 (65.1) 9,058 (50.6) 2,793 (59.0) 2,784 (58.8) Mild 7,349 (18.1) 3,938 (17.4) 3,411 (19.0) 872 (18.4) 866 (18.3) Moderate 4,957 (12.2) 2,256 (10.0) 2,701 (15.1) 564 (11.9) 558 (11.8) Severe 4,445 (11.0) 1,705 (7.5) 2,740 (15.3) 503 (10.6) 524 (11.1) Home oxygen use 3,373 (8.3) 985 (4.4) 2,388 (13.3) 32.02% 385 (8.1) 378 (8.0) -0.54% Prior stroke 4,476 (11.0) 2,263 (10.0) 2,213 (12.4) 7.47% 524 (11.1) 506 (10.7) -1.22% Cerebrovascular disease without prior CVA 6,940 (17.1) 2,849 (12.6) 4,091 (22.8) 27.08% 725 (15.3) 712 (15.0) -0.77% Diabetes 10.95% 1.45% None 23,485 (57.9) 12,558 (55.5) 10,927 (61.0) 2,726 (57.6) 2,743 (58.0) Insulin 1,141 (2.8) 751 (3.3) 390 (2.2) 137 (2.9) 126 (2.7) Non-insulin 6,060 (15.0) 3,433 (15.2) 2,627 (14.7) 745 (15.7) 746 (15.8) CAD: # diseased vessels 16.17% 0.95% None 13,277 (32.8) 6,686 (29.6) 6,591 (36.8) 2,292 (48.4) 2,326 (49.2) 1 7,528 (18.6) 4,031 (17.8) 3,497 (19.5) 770 (16.3) 757 (16.0) 2 7,249 (17.9) 4,446 (19.7) 2,803 (15.7) 520 (11.0) 512 (10.8) 3 12,474 (30.8) 7,455 (33.0) 5,019 (28.0) 1,150 (24.3) 1,137 (24.0) Left main coronary artery disease 5,351 (13.2) 3,408 (15.1) 1,943 (10.8) -12.59% 424 (9.0) 434 (9.2) 0.74% Pre-operative IABP/inotropes 1,172 (2.9) 644 (2.8) 528 (2.9) 0.60% 128 (2.7) 123 (2.6) -0.66% Prior shock 215 (0.5) 129 (0.6) 86 (0.5) -1.25% 28 (0.6) 31 (0.7) 0.81% Hypertension 36,907 (91.1) 20,624 (91.2) 16,283 (90.9) -0.94% 4,291 (90.7) 4,285 (90.6) -0.43% Immunosuppression 3,250 (8.0) 1,198 (5.3) 2,052 (11.5) 22.38% 363 (7.7) 344 (7.3) -1.53% Aortic insufficiency (moderate/severe) 7,949 (19.6) 4,700 (20.8) 3,249 (18.1) -6.67% 956 (20.2) 947 (20.0) -0.47% Mitral insufficiency (moderate/severe) 10,333 (25.5) 3,951 (17.5) 6,382 (35.6) 42.03% 1,166 (24.6) 1,125 (23.8) -2.02% Tricuspid insufficiency 7,354 (18.1) 2,488 (11.0) 4,866 (27.2) 42.04% 803 (17.0) 783 (16.5) -1.13% (moderate/severe) ICD 1,510 (3.7) 465 (2.1) 1,045 (5.8) 19.50% 185 (3.9) 179 (3.8) -0.66% Prior PCI 11,650 (28.7) 5,188 (22.9) 6,462 (36.1) 29.12% 1,278 (27.0) 1,233 (26.1) -2.15% Peripheral vascular disease 10,312 (25.4) 4,711 (20.8) 5,601 (31.3) 23.97% 1,138 (24.0) 1,113 (23.5) -1.24% Aortic valve mean gradient, median 42.0 (34.0,50.0) 42.0 (33.0,51.0) 42.0 (35.0,49.0) -4.68% 42.0 (35.0,52.0) 42.0 (36.0,52.0) 0.46% (IQR) Status (elective, urgent) 30,806 (76.0) 14,542 (64.3) 16,264 (90.8) 67.02% 3,871 (81.8) 3,813 (80.6) -3.14%

Hematocrit 36.0 (32.4,39.3) 36.6 (33.0,40.0) 35.4 (31.5,39.0) -21.51% 36.0 (32.3,39.5) 36.0 (32.1,39.6) 0.27% Pre-op total albumin, median (IQR) 3.7 (3.5,4.0) 3.7 (3.5,4.0) 3.7 (3.5,3.9) -8.00% 3.7 (3.5,4.0) 3.7 (3.5,4.0) -0.50% CAD requiring CABG 19,967 (49.3) 14,447 (63.9) 5,520 (30.8) -70.15% 1,565 (33.1) 1,523 (32.2) -1.89% Prior aortic valve replacement 1,316 (3.2) 654 (2.9) 662 (3.7) 4.51% 219 (4.6) 214 (4.5) -0.51% PA systolic pressure, median (IQR) 41.0 (37.0,45.0) 41.0 (36.0,41.0) 41.0 (39.0,50.0) 29.44% 41.0 (37.0,46.0) 41.0 (37.0,46.0) 1.09% Mitral stenosis 2,069 (5.1) 673 (3.0) 1,396 (7.8) 21.47% 204 (4.3) 220 (4.6) 1.63% Cardiac presentation 4,316 (10.6) 3,662 (16.2) 654 (3.7) -42.90% 285 (6.0) 295 (6.2) 0.88% SD = standardized difference; All other abbreviations can be found in Table S1.

III. FIGURES

Figures S1. Consort Diagram This figure displays the beginning study population, through exclusions, to the final study population. CABG = coronary artery bypass grafting; PS = propensity score; SAVR = surgical aortic valve replacement; STS = Society of Thoracic Surgeons; STS PROM = Society of Thoracic Surgeons Predicted Risk of Mortality; TAVR = transcatheter aortic valve replacement; TVT = Transcatheter Valve Therapy (Registry)

Figure S2. Propensity Distribution Propensity distribution: A) before matching; and B) after matching (zoomed in; exclusion of propensities<0.05 and >0.95 already applied). The distribution of propensity scores demonstrates overlap, in that some patients could be identified at any level of the propensity score in both treatment arms. SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement

Percent

0 0.0 0.2 0.4 0.6 0.8 1.0 Propensity Score

SAVR patients TAVR patients

Figure S3. Exploration of Balance after Propensity Score Matching Standardized differences on the matched population (SAVR n=4732, TAVR n=4732). Prior to matching, extreme imbalances are observed between TAVR and SAVR patients, and these are resolved by matching. CABG = coronary artery bypass grafting; CAD = coronary artery disease; CV = cardiovascular; CVA = cerebrovascular accident; MI = myocardial infarction; PCI = percutaneous coronary intervention; SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement

Figure S4. Boxplots of Absolute Standardized Differences Boxplots of absolute standardized differences (percent), for all covariates (54 including categorical levels), across all 35 pre-specified subgroups (1,890 total). The horizontal line indicates the ideal threshold of 10%. Balance within subgroups is not guaranteed by propensity matching, unless the propensity score is sufficiently flexible (with respect to interactions between covariates and subgroups). We investigated balance within the subgroups of interest. Complete tables (1,890 rows) are available upon request and were used to examine balance. Standardized differences above 10% were addressed by adding interactions to the propensity model (i.e., an imbalanced covariate within a subgroup has a unique relationship to the propensity within that subgroup and, therefore, requires an interaction). A few factors remained >10% but did not exhibit a clinically relevant difference. The final result is summarized in the boxplot below.

Figure S5. Falsification Endpoints Falsification endpoints for LEF and UTI were identified as endpoints that should not be effected, causally, by TAVR versus SAVR. However, UTI would be related to potential sources of confounding in the TAVR versus SAVR comparison, such as factors related to general health (TAVR patients being generally sicker), access to preventative health care and support, or different health care (some hospitals, regions, or health systems providing better care). These sources of unmeasured confounding could be revealed by a non-zero association between TAVR versus SAVR on the endpoint of UTI, or systematic patterns across subgroups. LEF and UTI are displayed as follows: A) LEF cumulative incidence curve; B) LEF forest plot; C) UTI cumulative incidence curve; and D) UTI forest plot. The confidence intervals generally overlap one, and the point estimates are not systematically different from one. CAD = coronary artery disease; Dz = disease; LEF = lower extremity fracture(s); PA = pulmonary artery; SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement; UTI = urinary tract infection

Figure S6. Discharge to Home Forest plot of patients discharged to home and associated patient characteristics. CAD = coronary artery disease; Dz = disease; PA = pulmonary artery; SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement

Figure S7. One-year Stroke Patients suffering a 1-year stroke, and associated patient characteristics. CAD = coronary artery disease; Dz = disease; PA = pulmonary artery; SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement

Figure S8. Proportion (%) of DAOH by Treatment Group Proportion (%) of DAOH according SAVR versus TAVR. DAOH = days alive and out of hospital; SAVR = surgical aortic valve replacement; TAVR = transcatheter aortic valve replacement Copyright © 2020. Duke University. 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.

Acknowledgment: Research reported in this report was funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#CER-1306-04350). Further information available at: https://www.pcori.org/research-results/2013/comparing-two-treatments-aortic- valve-disease

125