Effect of Physical Activity Coaching on Acute Care and Survival Among Patients With Chronic Obstructive Pulmonary Disease

Key Points Question What is the real-world effectiveness of a 12-month community-based physical activity (PA) coaching intervention on reducing all-cause acute care use and death in patients with a history of a chronic obstructive pulmonary disease (COPD) exacerbation? Findings In this multisite, randomized clinical trial that included a population-based sample of 2707 patients with COPD, 321 of 1358 patients participated in the PA coaching intervention and increased PA significantly, but there were no significant differences in the all-cause primary outcome (compostite measure of all-cause hospitalizations, observation stays, emergency department visits, and death) at 12 months. Meaning Most patients with a COPD exacerbation did not engage in PA, and the limited PA did not lead to significant benefit in 12-month health care use.


EXECUTIVE SUMMARY
Chronic obstructive pulmonary disease (COPD) is the third leading cause of the death in the US. The 42 personal, social and economic costs of the disease are tremendous, with annual expenditures of nearly 43 $50 billion, mostly from hospitalizations for exacerbations of COPD. For the vast majority of patients, 44 despite optimal pharmacological therapy, living with COPD is characterized by unrelieved dyspnea, 45 physical inactivity, deconditioning, and an insidious downward spiral of social isolation and depression. 46 There is mounting epidemiological evidence that physical inactivity is associated with more frequent 47 hospitalizations and increased mortality in COPD even after adjusting for disease severity. The evidence 48 is unequivocal that intensive supervised exercise training as part of pulmonary rehabilitation improves 49 outcomes of importance to patients. However, patient participation in supervised exercise at center-50 based rehabilitation programs is very low (1-3% of eligible patients) which undermines the wide scale 51 adoption of this approach in real world clinical settings for large numbers of patients. A paradigm shift is 52 needed in the non-pharmacological care of COPD from traditional rehabilitation to a more patient-53 centered, scalable, and sustainable model of promoting active lifestyles to improve outcomes for COPD 54 and its common co-morbidities. Identifying alternative, more flexible models that honors patients' 55 preferences and needs is of intense interest to patients and their caregivers. 56 57

OBJECTIVES 58
We propose a pragmatic randomized controlled trial in a large integrated health care system to 59 determine the effectiveness of a 12-month patient-centered, physical activity coaching intervention 60 (Walk On!) for patients with COPD on the primary composite outcome of all-cause hospitalizations, 61 emergency department (ED) visits, observation stays and death compared to standard care. We will 62 also examine the secondary outcomes of physical activity, symptom burden, quality of life, COPD-related 63 health care utilization, and cardio-metabolic markers. The long-term objective is to scale-up and spread 64 the implementation of this model into existing care management efforts across Kaiser Permanente and 65 other health care systems should the findings be positive. 66 67

METHODS 68
A randomized controlled design was used to test the effectiveness of Walk On! compared to standard 69 care (SC) in patients with a history of COPD-related hospitalization, ED or observation visits in the 70 previous 12 months. Eligible patients (n=1650; revised to 2707) were automatically identified from our 71 electronic medical records (EMR) system and randomized to either the Walk On! program or SC from 72 July 2015-July 2017. SC patients continue to have access to all the health services they would receive 73 such as a readmission reduction bundle, pulmonary rehabilitation, and health education programs. Walk 74 On! patients will receive SC plus the individually tailored Walk On! program over 12 months which 75 includes four components: baseline orientation/functional assessment, intensive coaching, and pro-76 active professional and peer support and monitoring via semi-automated outreach by telephone or 77 Internet as well as group visits. Outcomes will be analyzed using conventional statistical methods for 78 binary and continuous variables and according to intention to treat principles. We will also examine 79 heterogeneity of effects in patient subgroups. 80 81

PATIENT OUTCOMES 82
Our primary composite outcome will be all-cause hospitalizations, emergency department (ED) visits, 83 observation stays and death in the 12 months following enrollment in Walk On! Secondary outcomes 84 include physical activity, COPD-related health care utilization, and cardio-metabolic markers such as 85 BACKGROUND 133 Chronic obstructive pulmonary disease (COPD) is the third leading cause of the death in the US. 1

134
The personal, social and economic costs of the disease are tremendous, with annual expenditures of 135 nearly $50 billion, mostly from hospitalizations for exacerbations of COPD and associated sequelae. 2 For 136 the vast majority of patients, despite optimal pharmacological therapy, living with COPD is characterized 137 by unrelieved dyspnea, physical inactivity, deconditioning, and an insidious downward spiral of social 138 isolation and depression that has a profound impact on the lives of patients and their caregivers. 3,4 139 140 Physical inactivity is significantly associated with more frequent hospitalizations and increased 141 mortality in COPD even after adjusting for disease severity. 5-10 Our recent findings further extend these 142 observations by showing that hospitalized COPD patients who reported engaging in any level of 143 moderate to vigorous exercise prior to the index admission had a 34% lower risk of 30-day readmission 144 compared to inactive patients. 11 Reducing 30-day hospital readmissions has become a major focus of 145 many health care systems, with most efforts targeted at addressing deficiencies in care transitions and 146 short-term outpatient management following discharge. [12][13][14] Missing from many of these efforts is the 147 recognition that a majority of hospitalizations for chronic illnesses like COPD reflect failures in aggressive 148 and proactive outpatient management of COPD exacerbations, comorbidities, and behavioral risk 149 factors. 150 151 The evidence is unequivocal that pulmonary rehabilitation (PR) improves symptoms, health-152 related quality of life, and exercise capacity in COPD. 15

201
A conceptual model guiding this study is illustrated in Figure 1. We hypothesize that increased 202 physical activity leads to improvements in both physiological (decreased ventilatory requirements and  203 breathlessness, improved cardio-metabolic management, and lower levels of inflammation) and 204 psychological (improved mood, lower anxiety and increased self-efficacy for self-care) factors. Changes 205 in these mediators are associated with increased quality of life, fewer COPD exacerbations, decreased 206 acute care utilization, and improved survival. This study is a pragmatic randomized controlled trial to determine the effectiveness of a physical 211 activity coaching intervention (Walk On!) compared to standard care for patients at high risk for COPD 212 exacerbations ( Figure 2). Our research question, study design, and proposed methods are aligned with 213 methodological standards for pragmatic or real-world clinical trials (Table 1). 35

Participants
All eligible participants enrolled, regardless of risk, responsiveness, comorbidities or past compliance.
All adult health plan members who had a COPD hospitalization/ED visit/Ob Stay in the past 12 months (excluding patients who clearly would not benefit from Walk On) were enrolled. Intervention Condition Interventions are highly flexible, offering providers leeway in formulation and application.
Walk On! allowed for tailoring to patients' needs and preferences. Varying levels of participation were expected.

Intervention Practitioners
Interventions are applied by the full range of practitioners in the full range of settings with only ordinary attention to dose and side effects.
Intervention clinicians were recruited from the existing local workforce (respiratory therapists). The sites were responsible for selection and supervision of clinicians (using standard quality control tools). Comparison Condition "Usual Practice" (or the best alternative), offering practitioners considerable leeway in application.
Walk On! was compared to standard of care that members with COPD receive based on their existing health plan benefits.

Comparison Practitioners
The control intervention is applied by the full range of clinicians in the full range of settings, with only ordinary attention to training, experience, and performance. Standard care was provided by real-world providers under usual practice conditions -with no additional training or supervision.

Follow-Up Assessments
There are no research assessments; existing databases are searched for outcomes.
All utilization and clinical data were collected from existing electronic medical records and insurance claims data. Limited patient-reported outcomes were collected from intervention patients and a randomly selected subgroup of standard care patients Outcome Definition The primary outcome is objectively measured, meaningful to study participants, and does not depend on central adjudication.
Primary and secondary outcomes were defined by utilization, pharmacy, and clinical data (exercise; cardiometabolic markers). No additional clinical assessment was required.

Intervention Compliance
There are no special strategies to improve compliance, and compliance is unobtrusively measured.
Quality of implementation were assessed using a study dashboard.

231
Figure 2. Walk On! Study Design 232 233 Patients with a COPD-related hospitalization, emergency department visit, or observation stay in the previous 12 months and meet all other inclusion/exclusion criteria will be identified from the EMR (n=2707) There are no special strategies to maintain practitioner adherence, and adherence is unobtrusively measured.
Clinical supervision of physical activity coaches are at the same intensity as any operational clinical programs.

Primary Comparison
The analysis includes all patients regardless of compliance, eligibility, or others. All outcomes will be analyzed according to initial assignment -regardless of intervention participation or compliance.
(Original up to 9/31/15) COPD-related hospital-based utilization are defined according to the 248 Centers for Medicare and Medicaid Services (CMS) and National Quality Forum (NQF) criteria for the 249 Hospital Readmission Reduction Program. The following principal discharge diagnoses of COPD will 250 be included (ICD- Randomization 284 We considered and rejected the option of group-or cluster-level randomization. The core Walk 285 On! components were applied at the level of the individual patient rather than the provider or clinic, so 286 cross-over or spill-over of intervention effects within clinics or providers should not occur. Consequently, 287 there was no scientific advantage to cluster-level randomization. 288 289 All eligible patients were randomly assigned 1:1 to Walk On! or to continue with standard care, 290 stratified by medical center, time from hospitalization/ED/Ob Stay (<6months vs. ≥ 6months), level of 291 physical activity obtained from the exercise vital sign closest to the cohort identification date (inactive: 0 292 min/week of moderate/vigorous physical activity, MVPA vs. active: at least 1 min/week of MVPA; 293 patient will be assumed to be inactive if no EVS available), and median age (<72 vs. ≥72) by random 294 permuted blocks to ensure balance and reduce bias. The randomization was pre-generated and included 295 two steps for each stratum. First, the block size was randomly selected among a set of pre-selected 296 block. Second, within each block randomly selected, the overall number of treatment assignments was 297 balanced between groups, but the order of treatment assignment was randomly assigned. We repeated 298 this process until the number of maximum expected patients was reached. We reviewed every 6 months 299 to determine balance across the two groups on these characteristics and did not make any modifications 300 to the scheme. 301 302 For standard care patients (n=250; revised to n=537 due to a lower than expected response 303 rate) who were approached to complete surveys, with each recruitment wave, we randomly selected 304 patients from the six sites proportional to the number of patients randomized to Walk On! We sent 305 surveys to all SC patients on the final wave due to the low response rate from the earlier waves. 306 307 The anchor date for all patients is the date they are identified and randomized to treatment 308 arms.  334 We chose to compare the Walk On! intervention to standard of care for the following reasons: 335 1) there were insufficient data from large scale studies to support any specific PA intervention model for 336

Rationale for Standard Care as a Comparator
COPD and most small efficacy studies have had restrictive inclusion criteria; and 2) since there are no 337 scalable programs available, standard care (which includes access to pulmonary rehabilitation) was the 338 most appropriate comparator from the perspective of the healthcare system that is considering 339 implementing the intervention and the individual patient who is considering participating in the 340 intervention. We considered an active control group with mail outreach to remind members of the 341 programs available to them. However, this would not increase the scientific value of the study, given 342 that such low intensity touches are historically known to be ineffective. 343 344

345
For those patients randomized to Walk On!, a recruitment packet that included a letter signed 346 by the principal investigator and the pulmonary physician in charge for the medical service area, a study 347 brochure describing the Walk On! program, and a 3-min DVD video "testimonial" 348 (http://abc7.com/archive/9501102/) about the importance of PA by one of the Patient Advisory Board 349 (PAB) members was mailed within 1-6 business days of cohort selection and randomization. (Note: We 350 stopped mailing the DVD after wave 4 due to patient and coaches' feedback that they were not being 351 watched. Uptake of Walk On! was not negatively impacted by eliminating the DVD from our recruitment 352 packet.) Patients had the option of calling the physical activity coach to either agree to participate, or to 353 opt-out in response to the mailing. If patients did not actively call the PA coach for more information or 354 to opt-in to the study, the coaches conducted a total of two outreach contacts via phone and/or secure 355 message seven business days after the mailing. When contact was made with the patient, the coach 356 described the purpose of the study, how the patient was selected for participation, the Walk On! 357 intervention, and time commitment. After addressing the patient's questions, the coach obtained oral 358 consent and scheduled the baseline orientation intake visit. Patients did not have any further contact to 359 recruit them once the recruitment packet was sent and two contact attempts were made. 360 Patients who agreed to participate in Walk On! activities were sent a baseline packet 361 approximately ten days before their scheduled baseline visit that included: 1) a consent form, 2) an 362 activity sensor, 3) an additional copy of the study brochure, and 4) surveys to assess their physical 363 activity, symptom burden (COPD Assessment Test, CAT), depression (PHQ-8), and anxiety (GAD-7), and 364 quality of life (PROMIS-10). Patients were asked to wear one of two available sensors for up to seven 365 days prior to the baseline visit. After the visit, patients received four weekly coaching phone calls. 366 Outreach by the PA coaches for the remaining 11 months were individualized and targeted based on 367 patients' progress with their walking program. Patients were also encouraged to attend monthly peer 368 support meetings. 369 At six and 12 months after their randomization date, patients were sent a survey packet and a 370 $5 gift card. A reminder letter was sent if the surveys are not received within two weeks. Finally, a 371 phone follow-up was made seven business days after the reminder letter. Patients had the option of 372 providing their survey responses over the phone. 373 For standard care patients, a random sample (n=250, revised to 537) was invited to complete 374 the same set of surveys at baseline, six and 12 months, with the exception of the satisfaction questions, 375 in order to compare changes in PROs between intervention and standard care patients. These patients 376 were only informed that their medical center is participating in a study to improve outcomes for 377 members with COPD. We used the same mailing and phone-based follow up procedures described 378 above. 379

380
Theoretical Foundations 381 The Walk On! intervention was designed based on learnings from a series of collective 382 studies 45,47,48,85 by the investigative team that were informed by early and deep engagement with 383 patient stakeholders and is grounded in social cognitive 38 and self-regulation theories 39,40 and core 384 principles of motivational interviewing (Table 2). 41 In self-efficacy theory, the impetus for change resides 385 in the individual's efficacy expectations or one's "confidence in one's ability to take and persist in 386 action." These expectations reflect a person's beliefs about how capable he or she is in performing a 387 task. External environmental supports, like professional, peer and family modeling and engagement in 388 similar behaviors also increases efficacy. Walk On! had three core components (baseline 389 assessment/orientation, intensive coaching, and pro-active support) with built in flexibility to 390 accommodate the diverse preferences and needs of patients as well as anticipated implementation 391 constraints. We focused on promoting walking as the primary mode of PA since nearly 90% of activities 392 that patients with COPD engage in are ambulatory in nature and it is a safe and accessible form of PA. 42 393 The estimated time commitment ranged from 6-18 hours over the course of 12 months depending on 394 patients' participation in various Walk On! activities. 395 396 average steps/day during the baseline 7-day monitoring period; they used the higher of the two step 415 counts as an initial step goal. Our previous data showed that patients typically perform at approximately 416 60% of the walking intensity achieved during the 6MWT. 44 Thus, we derived a step count goal of total 417 steps accrued during the 6MWT multiplied by a factor of 5 to achieve approximately 30 minutes of 418 walking per day. For instance, a patient who accrued 500 steps during the 6MWT might be asked to aim 419 for 1500 steps/day during week one (500 x 5 x 0.60 = 1500 steps/day). For patients who were more frail 420 and for whom walking would initially be difficult due to severe deconditioning, we loaned a portable 421 cycle ergometer for patients to use during the first four weeks to strengthen their walking muscles and 422 gradually progressed them to a walking program. 423 424 3) Training on use of activity sensors & resistance bands. Patients chose one of two devices to monitor 425 their step counts, the Omron HJ329 pedometer or Tractivity accelerometer based on their preference 426 and access to the Internet, and were trained on their proper use. The Omron has an on-device display 427 whereas the Tractivity device displays step count data via any Internet or Blue-tooth enabled device. 428 Patients who did not have Internet access or were Spanish speakers were encouraged to use the Omron 429 pedometer since they would be able to see their step counts more easily and could report their average 430 weekly step counts to our automated telephone interactive voice response (IVR) system (Spanish script), 431 respectively. The Omron pedometer was worn on the waist and had been validated and used in several 432 of our COPD studies. 30,45 (Appendix Section 2 ) 433 434 The Tractivity, worn on the ankle, was validated against a research grade accelerometer in a general 435 population of hospitalized medical-surgical patients (n=20) 46 and used by patients in our pilot with 436 acceptable concordance with a research grade accelerometer (Stepwatch). Patients who chose to use 437 Tractivity were shown how to download a small applet on their Internet-enabled device and how to 438 view their step counts. (Appendix Section 2) 439 440 Note: The vendor that provided the Tractivity sensors went out of business in December 2016. We were 441 only able to offer patients the Omron device with recruitment waves 10 and 11 while we worked on 442 evaluating alternative devices, selecting and testing our top selection and configuring and testing our 443 systems to accommodate a new device for the final wave. We were able to offer patients the option of 444 using a wrist-worn, FitBit Alta or an Omron device for wave 12. We also converted patients from the 445 earlier waves who had challenges with using their Omron device to the FitBit if their coaches felt that 446 having the FitBit device help with engagement and motivation. (Appendix Section 2) 447 448 Since breathlessness with daily activities that involve the upper extremities is common in this 449 population, patients were also instructed on arm exercises using study-issued resistance bands to 450 strengthen their upper extremities. They were asked to complete these arm exercises 3 times/week, 451 but these exercises were not closely tracked. 452 453 B. Intensive coaching (Weeks 1-4). We have found that the initial weeks of starting a walking program 454 are most critical and are a time when patients require significant support to solve problems and barriers 455 that arise as they integrate a new activity in their daily lives. Thus, the coach conducted weekly phone 456 calls to help patients progress with their PA goals, reinforce COPD self-care skills, support patients' 457 efforts to monitor their activities and symptoms, assist with problem solving PA barriers, and 458 troubleshoot any device or technology issues. The coaches were guided by key principles of 459 motivational interviewing such as expressing empathy, rolling with resistance, and supporting self-460 efficacy 41 to personalize the content of these calls according to the patients' progress. The coaches 461 made appropriate referrals to either the patient's primary care provider or pulmonologist regarding any 462 clinical issues that needed to be followed up on. Patients were closely guided on how to safely resume 463 their PA after experiencing a COPD exacerbation. In addition, participants were instructed during the 464 baseline orientation to know when to stop their PA and seek emergent care (e.g. significant increases in 465 their dyspnea, chest pain or tightness, or other severe pain associated with activity). 466 5-52) breathing and health status as asked of Omron users, which generated the step goal recommendation 483

C. Proactive follow-up and support (Weeks
for the subsequent week. Patients were encouraged to review the graphical summary of their step 484 counts, which displayed past step data and suggested step goals. 485 486 The personalized step goal algorithm (Figure 4) was designed to ensure that the step progression was 487 safe and minimized common adverse events such as increased muscle soreness, more dyspnea and 488 fatigue associated with increasing PA. Email alerts were generated to the coaches when patients 489 reported worsening breathing and health problems interfering with their PA. 490 491 Patients who used the FitBit Alta are not asked the weekly health and breathing questions due to our 492 inability to deploy these questions within the FitBit web application nor were we able automatically 493 generate suggested step goals based on their previous week's performance and survey responses. 494 However, patients have access to all the various tracking functionalities available on the FitBit website to 495 use at their discretion. 496 497

IVR System
Step Goal Target: Continue with reported average step count for the week. "Based on the information you've reported, we suggest you aim for [provide step count they reported for current week]." *If 0 PA/steps reported, suggest 500 steps/day

IVR System
Step Goal Target: Do not suggest any step goal "Based on your report of worsening breathing and health problems interfering with your activity, we recommend that you cut back on your activities. We will send a message to your coach for follow-up."

Tractivity
Step Goal Target: Take average 7 day step count (Monday-Sunday) from previous week and do NOT increase step goal If 0 step count for all 7 days / no data step goal is 500 steps/day.

Tractivity
Step Goal Target: Do not suggest any step goal. Present the following message: "Based on your report of worsening breathing and health problems interfering with your activity, we recommend that you reduce your activities. We will send a message to your coach for follow-up." *No step goal suggestions in the first 30 days after first data upload *Min steps goal: 500 steps/day *Max steps goal: 15,000 steps/day *If no data upload for week, continue with step goal from previous week

Tractivity
Step Goal Target: Take average 7 day step count (Monday-Sunday) from previous week and do NOT increase step goal If 0 step count for all 7 days / no data step goal is 500 steps/day.

503
Data from both the IVR system and Tractivity web site were automatically retrieved and displayed on a 504 dashboard for regular review by the coaches (Figure 4). The dashboard facilitated population 505 management and targeted phone/secure message outreach to patients who were struggling to progress 506 with their walking goals and/or had more severe symptoms than usual, in which case the coach 507 communicated with the patient's provider as needed. The dashboard facilitated contact and workflow 508 management, standardized documentation to track intervention exposure and thus increased the 509 efficiency of the quality control/process evaluation efforts. 510 511 2) Monthly group visits for psychosocial support from peers, skill-building, and problem solving. Patients 512 had the option of attending monthly hour-long support sessions with their family member or friend. 513 These group visits started with 15-minutes of light exercise followed by 15-minutes of informal peer 514 interactions and networking. Peer support is especially important for patients who feel they have 515 limited support from their families. The meetings concluded with a 25-minute didactic/skill-building 516 component that was broadcasted via the web and tele-conference. 517 518 The session topics focused on practical strategies to overcome common barriers to staying active, e.g. 519 COPD exacerbations, weather, motivation and other relevant topics related to COPD management. The 520 coaches collaborated in creating power point slides for these topics and collectively reviewed and 521 approved the content for 12 topics. Other slide sets were developed on new topics that were either 522 suggested by patients or nominated by the coaches. 523 524 Our PAB members participated in these sessions as their time allowed and, along with other peers, 525 shared their successes with using community-based resources to stay active. Patients were entered into 526 a raffle for a $20 gift card at each monthly meeting. Sites that have an active pulmonary rehabilitation 527 program were encouraged to combine the Walk On group education visits with the rehabilitation 528 education sessions to increase sustainability and efficiency in early to mid-2017. Two sites were 529 successful in doing this mostly because they did not have a physical space constraint. 530 531 532 Figure 4. Coaches Dashboard (Note: Patient names are fictitious) Walk On! physical activity coach training 534 Walk On! coaches were recruited from the existing KPSC workforce of respiratory therapists, 535 pulmonary rehabilitation coordinators, and pulmonary care managers. The coaches participated in a 536 general motivational training workshop offered to all health care providers in our system and a half day 537 in-person project-specific training during the 6-month pilot phase prior to the study launch. The 538 coaches were also provided a detailed guide of the Walk On! program. Each coach implemented the 539 Walk On! protocol with 1-4 pilot patients for 3 months in preparation for the trial; the principal 540 investigator (HQN) or one of the lead PA coaches observed and provided feedback to the coaches during 541 their first 1-2 baseline visits. Issues or concerns with the phone coaching calls were discussed during 542 weekly to bi-weekly web conferences. This quality control structure continued throughout the 36-month 543 intervention period. 544 545 Intervention uptake and fidelity 546 Given the pragmatic design where all eligible patients were automatically randomized to 547 treatment arms, we closely tracked refusal rates and reasons for patients assigned to the Walk On! 548 intervention. For participants who agreed to actively participate in Walk On!, we used the study 549 dashboard to track uptake of the intervention components. Mild COPD exacerbations that are managed 550 on an outpatient basis and hospitalizations/ED visits/observation stays for moderate to severe 551 exacerbations are common in this cohort and were expected to be a major barrier to sustained PA and 552 participation in intervention activities. Temporary suspension of intervention activities as requested by 553 the patient or initiated by the coach due to COPD exacerbations or other acute illness as well as active 554 withdrawals were documented. In order to balance the pragmatic nature of the study, we instituted a 555 low intensity intervention fidelity assurance plan to include observations of up to five baseline intake 556 visits and reviewing up to 5% of the planned and as needed telephone coaching contacts across each 557 site. 558

560
Descriptive Variables 561 Socio-demographic variables: Age, gender, marital status, education and income (census-based), 562 race/ethnicity, and insurance status will be obtained from membership files 563 Medications: Pulmonary medications will be obtained from pharmacy databases. Supplemental oxygen 564 use will be obtained from durable medical equipment files. Long-term oxygen use is defined as the 565 patient being on oxygen >90 days, allowing a gap of no more than 14 days in the 12 months prior to 566 cohort identification. 567 Co-morbidities: All available diagnoses from outpatient and inpatient encounters in the 12 months prior 568 to cohort identification will be used to calculate the Charlson co-morbidity index 569 Injurious Falls (DSMB Report): The following E codes were used prior to 10/1/15: E880-E888. After 570 10/1/15, the following codes were used: W00-W19 571 572 Primary Outcome 573 The primary composite outcome is all-cause hospitalizations, emergency department (ED) visits, 574 observational stays, and mortality in the 12 months following randomization. Given the multiple 575 morbidities that patients with advanced COPD have and the known benefits of PA for these other 576 chronic conditions, it is reasonable to expect that Walk On! will have positive effects on hospitalizations, 577 ED visits, and observation stays for multiple causes. Walk On! is not expected to have its peak effects 578 until at least 6 months into the program and thus, follow-up of at least 12 months is needed for all 579 patients; and for those enrolled earlier, follow-up of up to 3 years will be available for secondary 580 analyses of long term adherence and effectiveness. 581 582 Cardio-metabolic Markers include body mass index, systolic blood pressure, diastolic blood pressure, 618 HbA1C, low density lipoprotein, high density lipoprotein, triglycerides, and total cholesterol. All 619 measurements available in the 12 months prior to identification will be averaged and used as baseline 620 values. 621 622 For follow-up assessments of systolic and diastolic blood pressure, we will use the average of all routine 623 clinic blood pressure readings taken between 6 and 12-months post-randomization. Blood pressures 624 obtained with temperatures of >100F and those obtained in urgent care are excluded. 625 626 For the others, we will use the measure that is closest to the 12-month post-study enrollment date. 627 Based Satisfaction. Overall satisfaction with Walk On! and its components including the baseline orientation, 652 intensive follow-up in the first 4 weeks, pro-active monitoring, step goal setting using the IVR and 653 Tractivity tools, reinforcement from the coach, and peer support will be measured at 6 and 12 months. 654 We are also conducting semi-structured exit interviews with a randomly selected 25% of the Walk On! 655 participants (or until thematic saturation) to understand the personal and ecological barriers and 656 facilitators to successful uptake of Walk On! 657 658 Descriptive statistics will be calculated prior to conducting the primary analyses. For all analyses, 661 data consistency and assumptions required, e.g., normality of responses will be checked. Any data 662 transformation or alternative methods necessary to analyze the data will be determined by examining 663 the data structure. Baseline characteristics will be compared between the two groups to assess whether 664 randomization balanced the group characteristics. The analyses will follow an intent-to-treat (ITT) 665 strategy, i.e. the analyses will include all randomized participants in the groups to which they were 666 randomly assigned, regardless of their adherence with the treatment and subsequent withdrawal. 667 668 Analysis for Aim #2 669 To test the primary hypothesis that the proportion of patients with any occurrence of all-cause 670 hospitalizations, ED visits, observation stays, and death 12 months after randomization will be 671 significantly lower in the Walk On! intervention group compared to standard care, we will use logistic 672 regression adjusted for randomization stratification variables (medical centers, time from 673 hospitalization/ED/Ob Stay, level of activity and age). For the secondary outcomes, logistic regression 674 will be used for categorical outcomes and analysis of variance will be used for continuous outcomes. 675 Baseline characteristics that are unbalanced between the two groups will be included as covariates. 676 Baseline characteristics for patients who do not complete the study due to health plan disenrollment 677 will be compared to the patients who complete the study and differential "drop outs" between the two 678 groups will be assessed by an interaction test between the intervention group and drop-out indicators. 679 We expect little to no missing data for the health care utilization outcomes. Secondary as-treated 680 analyses will be conducted based on actual treatment received to assess the efficacy of the intervention. 681 Results from this analysis will be compared to the ITT analysis and any differences will be reported and 682 interpreted with caution. 683 Since some participants will have follow-up data as far as 3 years after randomization, we will 684 perform additional analyses to evaluate the intervention effect on long-term outcomes. We will use 685 Poisson regression to assess the intervention effect on the average events during the entire study period 686 and use survival analyses to assess the intervention effect on time to the first event. Generalized 687 estimating equation (GEE) and mixed effects models will be used to compare the average proportions 688 and mean changes for continuous outcomes between intervention groups, while taking into account 689 correlated measures. 690 691 Heterogeneity of treatment effect (Aim #3) 692 We will assess heterogeneity of treatment effect by testing for a limited number of interactions, 693 to determine whether intervention effects differ by patient subgroups, e.g. presence of other common 694 morbidities (heart failure, diabetes, depression, and anxiety), level of social support, race/ethnicity 695 (White vs. non-White), gender, age, and access to the Internet. These are pre-planned hypotheses and 696 significant treatment heterogeneity will be declared through interaction tests with a standard alpha-697 level. The nature of the heterogeneity will be further assessed through subgroup analysis. Point 698 estimates and appropriate confidence intervals will be presented. We may conduct other exploratory 699 interaction and subgroup analyses, for which, appropriate alpha adjustment will be made to minimize 700 the chance finding (type I error). Although these analyses will be exploratory in nature, it is critical that 701 we understand what patient characteristics are associated with response to Walk On! in order to 702 appropriately target the intervention in future dissemination efforts. 703 704 Missing data 705 Because of our integrated health delivery system and ability to capture all utilization internally 706 and externally, we expect little to no missing data for the primary outcome or other secondary measures 707 of health care utilization. For other EMR-based secondary measures such as self-reported physical 708 activity and cardio-metabolic markers, we also expect to have nearly complete data since this patient 709 cohort has on average 16 outpatient encounters with our health system annually. We will only analyze 710 A1C and lipid data in the subset of patients with diabetes and cardiovascular disease. We expect a 711 higher level of missing data for PROs (symptoms, quality of life). Baseline characteristics will be 712 compared between patients with and without PRO data. For any missing data, we will assess whether 713 data is likely missing completely at random (MCAR) or missing at random (MAR) or missing not at 714 random (MNAR) by comparing patient characteristics. Sensitivity analysis and appropriate missing data 715 imputation techniques will be deployed depending on the types of missing data. Results will be 716 compared and differences will be interpreted with caution. 717 718 Measuring and accounting for confounders 719 All characteristics we can obtain from EMR and membership files will be extracted. This includes 720 but is not limited to age, gender, marital status, insurance status, race/ethnicity, smoking status, BMI, 721 Charlson index, O2 use, FEV1%predicted, prior number of hospitalization, etc. Due to randomization, we 722 do not expect baseline values of these covariates will confound the data analysis assessing the 723 effectiveness of Walk On! using the ITT samples. However, it is likely that these variables may confound 724 the data analysis assessing the efficacy of Walk On! using the patients who actually receive the 725 intervention (as-treated sample). This is because intervention acceptance, uptake and adherence may 726 vary by patient characteristics and outcomes may vary by these characteristics. We will assess potential 727 confounding and report results with and without appropriate adjustment. 728 729 Qualitative analysis for Aim #4 730

802
The data and safety monitoring plan for this study included monitoring recruitment progress 803 and potential adverse events resulting from data collection and the intervention activities. Since the 804 pragmatic design precluded active outreach to the standard care patients, we relied on EMR-based data 805 to conduct ongoing surveillance of events that resulted in a care encounter for safety monitoring. All 806 serious adverse events related to study procedures were reported to the IRB and data safety monitoring 807 board (DSMB