SHARING Choices: Design and rationale for a pragmatic trial of an advance care planning intervention for older adults with and without dementia in primary care

Abstract

Background:

Advance care planning (ACP) and involving family are particularly important in dementia, and primary care is a key setting. The purpose of this trial is to examine the impact and implementation of SHARING Choices, an intervention to improve communication for older adults with and without dementia through proactively supporting ACP and family engagement in primary care.

Methods:

We cluster-randomized 55 diverse primary care practices across two health systems to the intervention or usual care. SHARING Choices is a multicomponent intervention that aims to improve communication through patient and family engagement in ACP, agenda setting, and shared access to the patient portal for all patients over 65 years of age. The primary outcomes include documentation of an advance directive or medical orders for life-sustaining treatment in the electronic health record (EHR) at 12 months for all patients and receipt of potentially burdensome care within 6 months of death for the subgroup of patients with serious illness. We plan a priori sub-analysis for patients with dementia. Data sources include the health system EHRs and the Maryland health information exchange. We use a mixed-methods approach to evaluate uptake, fidelity and adaptation of the intervention and implementation facilitators and barriers.

Conclusions:

This cluster-randomized pragmatic trial examines ACP with a focus on the key population of those with dementia, implementation in diverse settings and innovative approaches to trial design and outcome abstraction. Mixed-methods approaches enable understanding of intervention delivery and facilitators and barriers to implementation in rapidly changing health care systems.

ClinicalTrials.gov Identifier: NCT04819191

1. Introduction

Alzheimer’s Disease and Related Dementias (ADRD) are among the most profoundly disabling and costly health conditions [1] and a leading cause of death in the United States. [2] ACP can be defined as a communication process to understand and share personal values, life goals, and preferences for future medical care and designate health care surrogates, that many experts and US organizations recommend throughout the course of serious illness. [3,4] Although persons with ADRD often meaningfully participate in ACP, [6,7] they are less likely than those without to participate in ACP and complete advance directives. [8,9] As a result, persons with ADRD are at greater risk for lack of concordance of preferences with end-of-life care. [10]

Family and close friends (family, as broadly defined by each person, referred to as “family” hereafter) are foundational in managing ADRD. Because family is particularly important in clinical decision-making for persons with ADRD, [11] engaging them in primary care and ACP is critical. [12–15] [16–20] Almost all community-living persons with ADRD rely on help from family [21,22] to coordinate information and care, and participate increasingly in shared decision-making as disease progresses. [21–23] Yet ambiguity about families’ roles and primary care clinician time and skill limitations amplify ACP challenges for persons with ADRD. [15,24]

Across populations, support and education can increase ACP [10,25] and systematic reviews of ACP show effectiveness for increasing advance directive documentation. [25,26] However, most interventional research has been limited to evaluation of participation and advance directive documentation rather than patient-centered outcomes such as end-of-life care. Moreover, key reviews found that ACP interventions alone improve advance directive documentation but have not shown effectiveness for person-centered outcomes; one potential reason is that these interventions do not address the wide scope of communication and systems issues needed to improve end-of-life care. [3,27–29] Prior studies of ACP in primary care have mostly targeted general populations [10,25,30] and generally excluded those with ADRD; most interventions in persons with ADRD have been in nursing homes. [31,32]

With this backdrop, we undertook the SHARING Choices study with the goal of improving communication for older adults with and without ADRD through proactively engaging family in interactions with primary care and supporting ACP and attention to ADRD in primary care. Our aims were [1] to test the effects of SHARING Choices in 3 patient groups: all older adults, those with ADRD, and those who die, and [2] to evaluate implementation and contextual factors, using mixed methods. Our hypothesis is based on evidence that older adults and families appreciate and expect primary care practices to initiate ACP [33] but that patient, family, and system factors often limit these conversations. [10] As a pragmatic trial, we addressed ACP for all older adults in intervention practices with particular attention for those with ADRD, given that cognitive impairment is common and often challenging to discern from medical records and that improving end-of-life communication and outcomes requires system-level changes. [34]

2. Methods

2.1. Conceptual framework and development of the intervention and trial

The design of SHARING Choices was informed by the fields of patient-provider communication, [17,35] family caregiving, [36,37] health services, [38,39] and health informatics, [40,41] acknowledging multiple ways that interpersonal relationships influence decision-making and end-of-life care. Our study design and evaluation are grounded in the Consolidated Framework for Implementation Research, a widely applied conceptual framework of contextual factors for adoption and spread of complex interventions in practice, as adapted for primary care, complex interventions and implementation outcomes. [42,43] The intervention is based on extensive research on feasibility and implementation of ACP and incorporation of ACP facilitators (staff trained in ACP that work with patients and families) and the other components of the intervention (Table 1). [44,45] The intervention also builds on health behavior change models of ACP that incorporate preparation to enhance uptake, [46,47] and is aligned with broader quality improvement approaches in our health systems and advanced primary care models. [48] Finally, the intervention and approach involved extensive stakeholder engagement, including patient and family interviews and patient advocate involvement, piloting with patients and families to refine the intervention, and model development within partner systems. [33,49–53] Groups of both local and national advisors also guide the trial.

Table 1.

SHARING choices components, content, rationale, evidence.

Content	Rationale	Evidence
1. Primary care engagement in voluntary advance care planning (ACP)	Most older adults and families appreciate when primary care practices engage them in ACP. Proactively introducing ACP through postal mailings normalizes these conversations.	Primary care initiatives to increase ACP are well-received. [52,66]
2. ACP information and availability of facilitator- led ACP conversations	ACP education increases patient & family awareness, knowledge and skill. Respecting Choices™ is a structured educational program to train facilitators to facilitate ACP conversations. In SHARING Choices, ACP facilitators initiate outreach and/or respond to clinician referrals.	ACP and Respecting Choices™ are associated with increased advance directive documentation and patient satisfaction in primary care and for those with serious illness. [26,44,45]
3. Person-Family Agenda Setting	Individuals & families often have different concerns. Agenda-setting stimulates conversations about ACP & the role of family. In SHARING Choices, agenda-setting tools are included in initial mailed outreach and can be supported by ACP facilitators.	Agenda-setting helps clarify concerns, goals, and expectations, and increase engagement in care. [67–69]
4. Patient/Family Access to Electronic Health Record	The patient portal facilitates timely and accurate information about health care issues, including ACP. In many health systems, with patient permission, families can have their own identity credentials to access information and communicate with clinicians. In SHARING Choices, information about the patient portal is included in initial mailed outreach and can be supported by ACP facilitators.	The patient portal operates through mechanisms of convenience, continuity, activation, and understanding. [40] Prior studies find benefit of supporting family through technology such as portals. [70]
5. Alzheimer’s Disease and Related Dementias (ADRD) Resources and Training of Staff and Clinicians	Patients and family appreciate when primary care practices provide ADRD information but practices often do not incorporate this into care. [71,72] [73] In SHARING Choices, ACP facilitators are provided additional training and support to engage in conversations in the context of ADRD.	Training curricula [74] and referral resources [75] support ADRD-friendly primary care practice.

2.2. Intervention: components of SHARING Choices

The intervention comprises 5 components (rationale and evidence in Table 1):

1. outreach from the primary care practice introducing SHARING Choices to prepare patients and families to engage in ACP conversations (Fig. 1),

2. access to a facilitator trained in all elements of SHARING Choices, including Respecting CHOICES and leading ACP for persons with ADRD and their families,

3. person-family agenda-setting to align perspectives about the role of family and stimulate conversation about health care issues and ACP (Fig. 2),

4. information about registration for the patient portal to enable and extend electronic interactions and information access to patients and family, and.

5. ADRD educational materials and resources for staff and clinicians.

Fig. 1.

Content of outreach letter to patients [Johns Hopkins version).

Fig. 2.

Agenda-setting checklist (Johns Hopkins version).

2.3. Study overview

We implement and evaluate the effectiveness of SHARING Choices on advance directive documentation and end-of-life outcomes by conducting a cluster-randomized pragmatic trial in 55 diverse primary care practices from the two partner health systems. We use a mixed-methods evaluation to examine not only “if” the intervention is effective but also “how” and “why” it had successes and limitations and to identify facilitators and barriers of implementation. We test the effects of SHARING Choices on ACP outcomes for all older patients, including persons with ADRD diagnoses in their electronic health record (EHR). Using information collected during routine care, we assess primary outcomes of advance directive documentation for all included patients and potentially burdensome care (defined as specific life-sustaining procedures that may not always be appropriate or consistent with patient preferences) within 6 months of death⁵³ for the subgroup of patients with serious illness (defined as those with at-risk diagnoses or conditions that may limit prognosis, capacity or function or increase disease burden) who die [54,55] (Appendix). We evaluate participation by patients and families in each element of the intervention overall and for those with ADRD. After the trial, we also conduct key informant interviews to assess various stakeholder perspectives (patient, family, primary care teams, administrators) on SHARING Choices.

2.4. Setting

We are conducting the study in partnership with primary care health systems operated by MedStar Health System and Johns Hopkins Community Physicians (JHCP) which together operate >75 primary care practices in the Baltimore-Washington DC area. Baseline population characteristics are in Table 3. Both systems have established EHRs with consumer-facing patient/family portals and many practices participating in the advanced primary care models that encourage ACP and advance directive documentation as part of value-based care. MedStar Health operates a centralized palliative care program in which the new grant-funded ACP facilitators were embedded. [48] The new JHCP grant-funded ACP facilitators functioned in a mixed centralized and local primary care environment with staff from JHCP’s Department of Education, building on many years of incorporating ACP into care management programs and primary care. [49,50,56]

Table 3.

Baseline Characteristics by Health System and Practice.

	Johns Hopkins Community Physicians	MedStar Health
Number of Clinics Randomized	20	35
Location
Urban	4 (20%)	10 (28.6%)
Rural	6 (30%)	5 (14.3%)
Suburban	10 (50%)	20 (57.1%)
Primary care clinicians, total (range by practice)	143 (4, 18)	342(5, 50)
Number of Patients	204,607	175,558
Number of Patients Age 65+	35,898	49,183
Percent Female (range by practice)	61% (55, 73)	61% (47, 73)
Percent Age 65+ (range by practice)	20% (5, 34)	30% (02, 95)
Race and Ethnicitv
Percent Hispanic/Latino (range by practice)	6% (2, 11)	2% (0, 8)
Percent White (range by practice)	60% (25,90)	53% (4, 90)
Percent Black (range by practice)	24% (4, 64)	34% (4, 90)
Percent Other Race (range by practice) Paver Mix	10% (3, 21)	13% (3, 36)
Percent Commercial (range by practice)	83% (57,91)	51% (6, 85)
Percent Medicare (range by practice)	14% (6, 29)	32% (01, 92)
Percent Medicaid (range by practice)	2% (0, 9)	12% (1, 43)
Percent with Advance Directive in Electronic Health Record (EHR)	4% (5, 10)	19% (7, 44)
Percent with Dementia Diagnosis in EHR (range by practice)	6% (2, 23)	8% (2, 49)
Percent Registered for EHR portal (range by practice)	65% (50, 82)	31% (11, 60)

2.5. Inclusion criteria

Eligible primary care practices have at least 2 clinicians whose panels collectively include >400 older adults. In this pragmatic trial, we target all new and established primary care patients ages 65 and over with a scheduled visit to the practice for an annual or follow-up in-person or telehealth appointment, including those with and without ADRD. There is no formal enrollment or consent as this is a practice-level initiative offered to all eligible patients at intervention practices.

3. ACP facilitation

Facilitators include lay facilitators, medical assistants, nurses or social workers embedded within partner health systems. Grant-funded intervention facilitators were trained in all elements of SHARING Choices, including The Respecting Choices^™ First Steps^™ curriculum for ACP, documenting advance directives, using the person-family agenda-setting checklist, and registering and accessing the patient portal (for both patients and family) (Table 1). The core of the required ACP training is Respecting Choices^™ (https://respectingchoices.org), a structured educational curriculum to train non-physicians in ACP. [57] Prior studies have shown that Respecting Choices^™ improves advance directive documentation, including in diverse communities. [44,45,57,58]

We reinforced facilitators’ ACP competency with traditional didactics, case scenarios, and modeling and mentored role play in one-hour biweekly mandatory meetings. Facilitators needed to successfully conduct at least one conversation before facilitating intervention conversations and received regular mentoring, monitoring and feedback, tailored to each health system. We also supplemented the Respecting Choices curriculum with targeted content on ADRD and ACP for persons with ADRD (Table 2). This supplemental training included communication techniques for a person-centered approach and supported decision making, recognition of limited patient understanding and need for communication with the primary care clinician on the need for further ACP conversation. We also trained facilitators on language and engagement of Black patients and their family.

Table 2.

Supplemental Alzheimer’s Disease and Related Dementias Training.

Topic	Content	Methods
Living with Dementia: The Person	Patient experience in early-stage dementia	Video; discussion; reading
How to Talk so Alzheimer’s can Hear You	Patient experience in moderate- to late-stage dementia Recognizing and facilitating communication challenges between patient, family, and advance care planning facilitator	Video; discussion
Family	Adult family experience with dementia from early to moderate stages of dementia	Video; discussion
Communication	Reflective listening Empathy Facilitating conversation with patient and family	Videos; discussion; case examples; role play
Communication with Diverse Patients and Engagement with Black Patients	Use of language and avoidance of medical jargon Potential influence of religiousness/spirituality Cultural sensitivity	Discussion; case examples
Repetitive Behaviors	Understanding family experience with repetitive questions, phone calls	Video; discussion
The Role of Medical Orders for Life-Sustaining Treatment (POLST) in the Care of People with Dementia	Symptoms and staging of dementia Overview of MOLST MOLST role in decision making	Video; discussion

4. Intervention embedding

We presented information on SHARING Choices to practice leaders before the study addressing intervention components, planning support and timeline, and resources and support for implementation and program monitoring. SHARING Choices champions were primary contacts for the investigators during implementation for intervention practices. At JHCP, the investigators worked with practice leadership to identify champions (practitioners, nurses and case managers); at Medstar Health, champions were practice management and medical director quality improvement leaders. Intervention components and workflows were tailored with practice champions over 2–3 meetings in the three months preceding the trial. Champions connected SHARING Choices staff with key practice staff to guide processes for interacting with patients and communicating about ACP with clinicians. We invited champions and leadership to bi-monthly meetings in the trial’s first six months to review progress and potential adaptations.

5. Intervention delivery characteristics and processes

SHARING Choices is designed for compatibility with primary care and balances core elements with adaptability to health system context and practice preferences. We modified the protocol for COVID-19 pandemic logistical issues (mainly focusing on remote facilitation). All patients 65 years of age and older under care of intervention practice clinicians with an upcoming new, physical, or follow-up visit within the next month are mailed SHARING Choices information. Working with practice stakeholders, grant-funded facilitators prioritize patients for outreach based on Annual Wellness Visit eligibility, lack of advance directive in the EHR, or an ADRD diagnosis. Approximately two weeks before the visit, facilitators contact pattients by phone or portal messaging about an ACP conversation. Intervention practice clinicians and case managers can also refer patients to facilitators.

ACP conversations occur by phone or telehealth or at the practice and are not necessarily linked to a visit. The facilitator follows the Respecting Choices^™ structured conversation guide and employs motivational interviewing techniques to evaluate readiness and facilitate discussion of SHARING Choices components with patients (and family when available), focusing on communication techniques for those with ADRD. The facilitator records the duration (which varies per patient preferences) and content of ACP meetings and resulting decisions in the EHR. Advance directive documentation, if brought to a visit or sent in, is uploaded to the EHR.

6. Fidelity monitoring and adaptations

Guided by the NIH Behavior Change Consortium [59] we address fidelity through a.) design, by selecting distinct intervention elements based on theory, b.) training, by relying on a protocolized curriculum to train facilitators in ACP and other elements of SHARING Choices, c.) monitoring measures of implementation, and d.) ongoing training and resources to support facilitators and maximize fidelity to procedures within the pragmatic nature of the trial.

As a pragmatic trial, tracking of processes is monitored through data collected in routine interactions. The study team uses reporting systems to regularly monitor measures of implementation such as number of mailings and scheduled and completed ACP facilitator meetings. ACP facilitators document meeting content using structured EHR fields, including structure (presence of family, meeting duration) and ACP topics discussed. EHR Data monitored includes documented facilitator conversations, completed advance directives, and patient portal use. Additionally, the study team collects limited information from ACP facilitators related to safety monitoring and practice-level field observations during routine check-ins with managers and the study team for the implementation evaluation. Key stakeholders including administrators, champions, facilitators, patients (including those with ADRD) and families will participate in interviews after the trial to evaluate implementation processes and facilitators and barriers. We will conduct interviews using an interview guide based on our pilot work, literature review and stakeholder input. We will use a concurrent mixed-methods approach, with quantitative evaluation of measures of implementation (Table 4) and a reflexive thematic analysis approach for evaluation of interviews.

Table 4.

Schedule of data elements, purpose and sources.

	Purpose and Timing
Assessment (Data Source)	Practice Eligibility and Randomization (Day – 180)	Patient eligibility and Intervention: Day 0–456	Outcomes at 12 Months (±30 Days)
Primary Care Practice Characteristics, including size and age and race distributions (Health Systems)	X
Patient Characteristics, including age, race and ADRD diagnosis (Health System EHR)		X
Measures of Implementation (ACP Facilitators contacts, ACP conversations, family presence, content, and AD completion or uploading) (Health System EHR and facilitator field notes)		X
Primary Outcome: Advance Directive or MO(L)ST Documentation in EHR (Health System EHR)			X
Primary Outcome: Potentially burdensome Care, last 6 months of life (Health System EHR/Health Information Exchange)			X
Adverse Events		X
Implementation Interviews			X

ACP = Advance care planning, ADRD = Alzheimer’s Disease and Related Dementias, EHR = Electronic Health Record, MO(L)ST = Medical Orders for Life-Sustaining Treatment/Medical Orders for Scope of Treatment.

The Chair, ACP Facilitator Training and Oversight Workgroup (VC) regularly convenes ACP facilitators. Frequency and content of meetings varied based on study phase and facilitators’ specific needs. Before trial launch, the focus was educating facilitators about study objectives, workflows, and protocols and building ACP skills. After launch, we provided facilitators with study updates and asked them to present cases, troubleshoot challenges, engage in collaborative problem-solving, and discuss strategies for resolution. Ad-hoc support is available from investigators (VC, KS, KW) and the health systems. Through ongoing supervision, investigators probe for feedback regarding practice-level systems and workflows that work well or pose barriers, identifying needs for adaptations.

We evaluated facilitators’ skills mid-trial using observation of standardized patient encounters evaluating communication skills for ACP in those with ADRD. Each facilitator completed two case-based simulations, one with cognitive impairment and one with ADRD and a family member, followed by a debrief and structured assessment to evaluate adherence to the intervention, quality of delivery, and areas for improvement, which were incorporated into supplemental training.

7. Description of evaluations

7.1. Outcomes

We selected measures based on: 1.) known reliability and validity, 2.) sensitivity to change, 3.) clinical relevance, 4.) reflection of objective indicators of domains of interest, and 5.) feasibility of data extraction within the pragmatic trial mechanism. Our primary outcomes of advance directive documentation or MO(L)ST at 12 months and receipt of potentially burdensome care for those with serious illness within 6 months of death are extracted from JHCP and MedStar Health EHRs and the regional health information exchange (Table 4). We additionally assess measures of implementation, including facilitators’ patient contacts, ACP conversations, inclusion of family, and conversation content.

Advance directive or MO(L)ST documentation

Advance directive or MO(L)ST documentation is defined as a composite measure of EHR documentation of either [1] advance directive: a health care agent form (durable medical power of attorney/surrogate) or living will, or [2] MO(L)ST: Maryland Medical Order for Life-Sustaining Treatment (MOLST), or District of Columbia Medical Order for Sustaining Treatment (MOST). Documentation is based on audited data fields available through each health system’s EHR 12 months after study entry.

Although our primary outcome is operationalized as a composite measure, we exoamine advance directives and MO(L)ST in parallel for older adults ages 65 and older with and without a diagnosis of ADRD as defined by ICD-10-CM diagnosis codes (Appendix). By separately examining between-group differences in advance directive and MO(L)ST completion, we assess whether the study differentially affected these two outcomes for relevant subgroups.

Potentially burdensome care

Potentially burdensome care is measured as any (yes/no) life-sustaining procedures within the 6 months preceding death using dates and validated ICD-10-CM codes for hospital services (Appendix). Specific validated codes for potentially burdensome care include intubation and mechanical ventilation, tracheostomy, gastrostomy feeding tube placement, hemodialysis, enteral and parenteral nutrition, and cardiopulmonary resuscitation (Appendix). [60] The measure denominator comprises patients who died during the 12 months post-analytic sample with serious illness (as defined above, under study overview) using a modified list of validated ICD-10-CM codes (Appendix). [56]

7.2. Sample size

Sample size estimates use the following considerations: (a) the cluster-randomized trial design, (b) ability to detect effect sizes comparable to treatment effect estimates from published literature, [25] and (c) two-tailed type 1 error rate of 0.025 given two trial primary outcomes. We expect approximately 87,000 older adults from 55 primary care practices to be eligible, of whom approximately 9200 will have a diagnosis of ADRD (10.6%) and 7000 will die within 12 months of study entry. Power calculations incorporate two features of SHARING Choices: 1) that practices are randomized to intervention versus control at a 1:2 ratio, and 2) that effects of clustering are accounted for by assuming variability in the control group success rate (analogous to an intraclass correlation assumption for a continuous outcome). We performed power calculations using the Optimal Design software package for cluster randomized trials. [61,62]

For our primary outcome, advance directive or MO(L)ST documentation, preliminary data indicates that advance directive documentation in the EHR is relatively low and varies by practice across our two health system partners. If we assume advance directive documentation in the EHR of 15% at baseline (reported base rate at JHCP at the time of trial design) [56] and a plausible range of ±10% (i.e. from 5% to 25%) based on nationally-representative estimates, we have power of 0.80 to detect an increase from 15% to 21% or greater. Because 29,000 patients are expected to be eligible for the intervention, an increase from 15% (4350 persons) to 21% (6090 persons) represents an increase in advance directive documentation among 1740 older adults at intervention practices. For the 9200 older adults expected to have an ADRD diagnosis, we have power of 0.80 to detect an increase from 15% to 22% or greater. Of these patients, 3064 are expected to be randomized to the intervention, and an increase from 15% (460 persons) to 22% (674 persons) represents an increase in advance directive documentation for 214 ADRD patients in intervention practices.

Based on US data, we project that 7000 of the 87,000 patients in the trial (8%) will die within 12 months of study entry, [63] and assume a base rate of potentially burdensome care of 18%, with a plausible range of ±5% (i.e., from 13% to 23%). [9] From these assumptions, we have power of 0.80 to detect a reduction in potentially burdensome care from 18% to 15% for the patients who die.

7.3. Randomization

The unit of randomization is the primary care practice, which reduces potential for contamination of controls and enables practice-wide deployment. We randomized practices to intervention or control in a 1:2 ratio because control practices incur no study cost and having more control practices affords greater precision in study outcome estimates. We first stratified practices by participating health system (JHCP or Medstar Health) since variation in system-level processes, culture, and intervention implementation could affect outcomes. Second, we balanced intervention and control groups on practice characteristics: geography (urban, suburban, rural location), percent of patients of Black race, percent of patients age ≥ 65, and number of primary care clinicians.

Because we include 20 JHCP and 35 Medstar Health practices, notable and problematic imbalances on practice-level characteristics may occur by chance in a single randomization. Methodological innovations are needed to achieve balance on multiple practice-level characteristics while maintaining the experimental rigor and inferential value of true random assignment.

Therefore, we used a covariate-constrained randomization (CCR) method to assure approximate balance on selected site characteristics across intervention and control groups. In this CCR, for each organization (JHCP and Medstar Health), 1000 randomizations were undertaken using a 1:2 assignment ratio. We then identified a subset of randomizations meeting 5 balancing characteristics (urban vs. suburban, rural vs. suburban, # of clinicians, % age ≥ 65, and % Black) and the following 3 criteria (Table 3):

1. The overall balance index, I, based on the half-normal distribution of the absolute value of SHARING Choices vs. Control differences on the balancing variables, was in the bottom 10%. That is, only the best 100 of the 1000 randomizations were further considered, following Huang and Roth. [64]

2. The p-value from a non-parametric test, the Kruskal-Wallis test, had to be >0.30 for each balancing variable, as recommended by Ciolino et al. for CCR. [65]

3. For JHCP, 1–2 urban and 2–3 rural practices were assigned to SHARING Choices; for MedStar Health, 3–4 urban and 2–3 rural practices were assigned to SHARING Choices.

For JHCP, 79 of the 1000 randomizations met all three criteria, and for MedStar Health, 41 of the 1000 randomizations met all three criteria. From the 79 JHCP and 41 MedStar Health eligible randomizations, one was randomly selected for each organization for preliminary randomization. Of the 55 practices, 20 were randomized to SHARING Choices and 35 to control.

7.4. Data analyses for primary outcomes

We compare SHARING Choices and control practices by analyzing the distribution of practice-level characteristics at baseline using appropriate graphical procedures, summary statistics, and multivariate methods. We examine consistency of effects stratified by patient age group, gender, race, ADRD diagnosis, and primary care practice.

We test the effects of SHARING Choices on all patients and on a subsample of patients with a diagnosis of ADRD, defined by ICD-10-CM diagnosis codes, as described above. For both analyses, we hypothesize that patients at intervention practices are more likely to have a documented advance directive and/or MO(L)ST in the EHR than patients at control practices at 12 months. We additionally hypothesize that among patients who die, those at intervention (versus control) practices are less likely to experience potentially burdensome end-of-life care within 6 months of death. All analyses are performed using SAS statistical software and account for the cluster design, with the patient as the unit of analysis. All primary analyses comparing the effect of SHARING Choices to usual care follow the principle of intention-to-treat, including all eligible study subjects with measurements at all available time points as appropriate to the outcome. The primary independent variable is group assignment at the initial time point that eligible patients are identified, consistent with intention-to-treat.

We assess comparability of patients in intervention and control groups by comparing the distribution of practice-level and patient-level characteristics at baseline using appropriate graphical procedures, summary statistics, and multivariate methods. The study is designed to produce balance on important covariates at the practice level (unit of randomization) but not necessarily at the patient level. Baseline patient and practice characteristics that differ between groups are included in our models of between-group differences. We use SAS Proc GLIMMIX to examine binary primary outcomes. The model includes practice variables as fixed effect practice-level covariates. Initial analyses examine baseline variables distributions and differences by intervention and control group. We build multilevel models accounting for practice-level randomization and group assignment and clustering of outcome distributions within practices. Multilevel models also include and control for baseline patient and practice-level covariates. In addition to overall effects of the intervention, we examine consistency of effects stratified by patient age group (< 75, ≥ 75) and primary care practice. We conduct parallel analyses to determine whether effects vary by primary outcome definition (AD alone versus AD+MOLST combined).

8. Conclusion

The design of this pragmatic trial of SHARING Choices has several advantages for evaluating potential improvements in processes and care for older adults with and without ADRD in primary care. First, the pragmatic nature of the trial allows for a flexible approach to fidelity at the practice level and adaptations as the trial continues. This was particularly important as many COVID-19 restrictions were still in place at trial launch, and as health system, local and federal initiatives to improve ACP and changes in EHR and portal and proxy systems and use evolve during the trial. Second, emphasizing training facilitators and incorporating strategies for those with ADRD is particularly important since these patients are often excluded and may require additional time to engage in effective ACP and document advance directives and/or MO (L)ST for the EHR. Finally, although EHR data has many limitations in outcome measurement, monitoring advance directive documentation and potentially burdensome care in EHRs across health systems and health information exchanges is an important advance in trials of care for large populations with serious illness.

The pragmatic nature of the trial also has limitations. Some parts of the intervention may be less successfully implemented than others given emerging health system priorities, measurement of implementation outcomes may be less complete and differ across health systems, and many aspects of ongoing ACP and related initiatives already integrated into health systems are not easily measurable, [27] but the randomized design should account for these issues.

Given these strengths and limitations, evaluating the impact of SHARING Choices will contribute to knowledge to better equip primary care in effective communication and team-based ACP, particularly for the critically important population with ADRD.

Appendix

Table.

Codes for potential burdensome care and serious illness.

Codes for Potentially Burdensome Care at the End of Life (Numerator) [60]
Procedure	CPT/HCPCS Code	ICD-10-CM
Intubation and Mechanical Ventilation	31,500	OBH17EZ, OBH18EZ, OB717DZ, OB718DZ, 0BH07DZ, OBH07YZ, OBH172Z, OBH17YZ, OBH182Z, OBH18YZ, OBHK7YZ, OBHK8YZ, OBHL7YZ, OBHL7YZ, OBHL8YZ, OWHQ7YZ, 5A19, 5A1935Z, 5A1945Z, 5A1955Z, 5A12012, 5A19054
Tracheostomy	31,600, 31,601, 31,603	OB11, OB110F4, OB11OZ4, OB113F4, OB113Z4, OB114F4, OB114Z4,
Gastrostomy Tube Insertion	49,440, 43,761
Hemodialysis	90,935, 90,937	5A1D, 5A1D00Z, 5A1D60Z
Enteral or Parenteral Nutrition	43,750, 43,246, 44,372, 44,373, 74,350, 43,832, 43,830, 43,653, 49,440, 49,441, 49,446	0DH67UZ, 0DH68UZ, 3E0G36Z
Cardiopulmonary Resuscitation	92,950	5A120012, 5A19054

Codes for Serious Illness and Conditions (Denominator) [54,55]
	ICD-10-CM Code
Cancer	ICD-10-CM liver excluding HCC (C22.1–4), esophagus (C15.3–5, C158–9), stomach (C16.0–6, C16.8–9), pancreas (C25.0–3, C25.7–9), peritoneum (C48.0–2, C48.8), brain (C71, C71.0–9), secondary malignant neoplasms (C78.00, C78.1–2, C78.39, C78.4–7, C78.89, C79.00, C79.11, C79.19, C79.2, C79.31–2, C79.49, C79.51–2, C79.60, C79.70, C79.81, C79.82, C79.89, C79.9), malignant pleural effusion (J91.0), malignant ascites (R18.0), disseminated malignant neoplasm (C80.0), leptomeningeal carcinomatosis (G96.12), heme malignancy (C81.09, C81.19, C81.29, C81.39, C81.49, C81.79, C81.99, C85.19, C85.29, C85.89, C91.02, C91.12, C91.52, C91.62, C91.92, C91.A2, C92.02, C92.12, C95.12).
Heart failure	ICD-10-CM (I09.81, I11.0, I13.0, I13.2, I50.1, I50.9, I50.20–23, I50.30–33, I50.40–43).
Chronic obstructive lung disease	ICD-10-CM (J43.9, J44.0, J44.1, J44.9). Home oxygen: ICD-10-CM (J95.850, Z99.1, Z99.11–12, Z99.81).
Cirrhosis	ICD-10-CM (K70.3, K70.30, K70.31, K70.4, K70.40, K70.41, K74.3–5, K74.60, K74.69).
Hepatic decompensation	ICD-10-CM (K65.9, K67, K65.0, K65.2, K65.8); esophageal varices: ICD-9 (456, 456.2, 456.21); ICD-10-CM (I85.01, I85.10, I85.11); ascites: ICD-9 (789.5, 789.51, 789.59); ICD-10-CM (R18.8).
End-stage renal disease	ICD-10-CM (N18.5, N18.6, Z94.0, Z99.2, Z91.15, Z49.31, Z49.01, Z49.02, Z49.32).
Amyotrophic Lateral Sclerosis	ICD-10-CM (G12.21).
Alzheimer’s Disease and Related Dementias (ADRD)	ICD-10-CM (F01.50, F01.51, F02.80, F02.81, F03.90, F03.91, G30.0, G30.1, G30.9, G31.01, G31.09, G31.83, G31.84, R41.81) (includes code G31.84, Mild Cognitive Impairment)