Mapping the Landscape of Advance Care Planning in Adolescents and Young Adults Receiving Allogeneic Hematopoietic Stem Cell Transplantation: a 5-Year Retrospective Review

Abstract

Background:

Allogeneic hematopoietic stem cell transplantation (HSCT) carries significant risk of morbidity and mortality. Participation in advance care planning (ACP) is crucial to promote patient-centered care and has been shown to have positive impacts on patients, caregivers, and providers. Historically, both HSCT recipients and adolescents and young adults (AYAs) are significantly less likely to engage in ACP.

Objectives:

We sought to characterize ACP utilization in AYAs undergoing HSCT by evaluating the frequency of different types of ACP documentation over time and identifying demographic and clinical factors associated with documentation of each type of ACP.

Study Design:

We conducted a single center, retrospective review of the electronic health records (EHR) of AYAs (ages 15 to 39 years) who received allogeneic HSCT between 2015 and 2020. EHR documents were screened for three pre-defined categories of ACP: 1) advance directives or medical orders (ADs/MOs) which included proof of signed paper directives, expressions of preferred code status, and identification of a healthcare proxy; 2) goals of care (GOC) conversations which included discussions of medical care in a specific situation informed by patients’ priorities; and 3) other ACP conversations which included more general discussions of patients’ values regarding their care or legacy wishes. Documents were coded by two researchers and discrepant categorizations were reviewed by a third researcher. Patients younger than 18 years on the day of transplantation were excluded in the analyses of AD/MO documentation. Univariate and multivariate logistic regressions tested for associations between patient factors and documentation of each type of ACP. For deceased patients, Kaplan-Meier curves were created to illustrate the time-to-event relationship between days before death and documentation of each type of ACP.

Results:

Sixty-eight thousand documents associated with 219 patients were reviewed and 666 ACP documents associated with 190 patients were identified. Few of the 219 patients had documented GOC (n=29, 13%) or other ACP conversations (n=81, 37%). A subset of patients (n=28, 13%) had no documented ACP. Most of the 201 patients aged 18 years or older had a documented AD/MO (n=172, 86%). No tested factors were significantly associated with documentation of ADs/MOs. GOC and other ACP conversations were more likely to occur in patients with a palliative care consult, and patients with a malignant diagnosis were also more likely to engage in GOC conversations. Over 50% of the documentation occurred in the subset of 39 deceased patients, with half of AD/MO documentation in the last 67 days of life, half of other ACP documentation in the last 20 days of life, and half of GOC documentation in the final 2 days of life.

Conclusions:

While the majority of AYA patients receiving HSCT did have documentation of ADs/MOs, few patients had documented GOC or other ACP conversations. The bulk of all ACP conversations occurred in patients that ultimately died and who were very close to the end of life. Our results support ongoing efforts to improve implementation of ACP in this vulnerable population, particularly for those receiving HSCT for non-malignant conditions.

Graphical Abstract

INTRODUCTION

Between 2014 and 2018, more than 7,000 adolescent and young adult (AYA) patients, defined as those between the ages of 15 and 39 years old, received an allogeneic hematopoietic stem cell transplantation (HSCT) in the United States.¹ AYAs with serious illnesses represent a historically understudied and highly vulnerable population. AYAs acquire more autonomy and responsibility in their personal lives, and those with chronic illnesses must simultaneously navigate the transition and the physical and philosophical divide between pediatric and adult silos of medical care. As summarized in a review by Wiener et al., multiple studies have revealed that AYAs with cancer may receive less quality care due in part to variety of systemic factors including poor access to care or health insurance, lack of participation in clinical trials, and challenges to adherence or follow-up.² For those receiving HSCT, AYAs’ precarious position can amplify the acute and chronic medical and psychosocial intensities of this endeavor. Compared to younger pediatric patients, multiple retrospective studies have shown that AYAs receiving HSCT for leukemia have worse outcomes, including lower 5-year overall survival (OS) and higher transplantation-related mortality (TRM).^3–5 AYA HSCT survivors are also at significant risk of transplant-related morbidity as they confront high rates of infertility and inherently high cumulative risk of a myriad of late effects.⁶ Given these significant risks of short- and long-term morbidity and mortality, AYAs receiving HSCT require open and longitudinal communication with healthcare providers to promote patient-centered and goal-concordant care.

Advance Care Planning (ACP), formally defined by a recent multidisciplinary Delphi panel of experts, is “a process that support adults at any age or stage of health in understanding and sharing their personal values, life goals, and preferences regarding future medical care, [with the goal of helping] ensure that people receive medical care that is consistent with their values, goals and preferences during serious and chronic illness.”⁷ When ACP discussions occur between AYAs and healthcare providers, they often have a significant positive impact on the care of patients and families. In one study of pediatric and AYA patients with complex chronic conditions who died between 2006 and 2015, documented ACP was associated with higher parental report of the child’s quality of life at end of life, decreased reported suffering, decreased parental regret,⁸ and higher congruence between patient/guardian treatment preferences.⁹ Inversely, poor prognostic communication has been shown to exacerbate grief, impair trust between family and provider, and foster lingering regret for bereaved families.^{10, 11}

Despite the established benefit of these conversations, review of the literature highlights many barriers to ACP discussions in both AYAs and HSCT recipients. Many clinicians feel unprepared to discuss resuscitation status with adolescent or AYA patients,¹² with one study among 144 providers caring for AYAs with organ failure or history of solid organ transplantation identifying the following barriers: a belief that parents do not want to discuss prognosis with their child (59%), unrealistic treatment expectations held by the patient and family (50%), difficulty balancing hope with realism (49%), feeling that prognosis is too uncertain to discuss given unknown disease trajectory (48%), and limited training in how to have ACP discussions (46%).¹³ One retrospective review of 163 patients that died following HSCT found that 40% had no documented ACP, while only 53% had palliative care (PC) involvement with the first PC consult occurring a median 15 days before death.¹⁴ In a larger sample of over 600 HSCT recipients, only 19% had PC involvement and 44% had documented ACP.¹⁵

In addition to the specific barriers to initiating and completing ACP conversations in AYAs and HSCT recipients, the effective electronic documentation of these discussions presents another challenge, with insufficient time, disruption of workflow and concerns about recording patient beliefs or views that might cause conflict with other members of the medical team or the patient’s family cited as specific barriers to documentation of ACP conversations in a previous study.¹⁶ A recent systematic review of available tools to help providers complete ACP discussions found very few studies focused on improving documentation, and no examples addressing this challenge in adolescent patients.¹⁷

Given the distinct challenges AYAs undergoing HSCT face, ACP is of interest to all providers caring for this population. To date, there is no quantitative study which addresses ACP utilization in AYA HSCT recipients. In this study, we report on the results of a single center, retrospective review of the electronic health records (EHR) of a diverse panel of AYA patients who received HSCT in the past 5 years to characterize the frequency and context of documented ACP conversations. We elucidate patterns and deficiencies in this process, and explore factors associated with successful documentation of different sub-types of ACP discussions.

METHODS

Study Participants

Participants included AYAs between the ages of 15 and 39 years old at the time of HSCT who had received HSCT from January 1, 2015, through January 1, 2020 at our center. The following information was obtained for all participants: age at the time of transplant, gender, self-identified race and ethnicity, primary language spoken, religious affiliation, indication for HSCT, survival status as of October 1, 2020, and date of death when applicable. Access was granted to available EHR documentation for all participants. This study was viewed as a quality assurance project and considered exempt from review by the Institutional Review Board of the National Cancer Institute.

Defining Outcome Variables

Prior to reviewing available data, definitions were established for EHR documents addressing the following key concepts within the realm of advance care planning: 1) advance directives and medical orders (ADs/MOs), 2) goals of care (GOC) conversations, and 3) other advance care planning (ACP) conversations. Conversations could either be between the patient and provider directly, or between a proxy for the patient and the provider. Using the definition proposed by the Institute of Medicine, ADs were viewed as “patient-initiated documents [including] living wills and documents that name a health care agent.”¹⁸ MOs were viewed as provider-generated documents indicating a patient’s “preferences with respect to treatment, preferred site for receiving care, and death.”¹⁸ These include physician orders for life-sustaining treatment and indications of code status, such as do-not-resuscitate or do-not-intubate orders. We elected to combine EHR documents pertaining to ADs or MOs into a single category as these both contain written instructions about how to make medical decisions for a given patient, but do not offer additional information about the conversation surrounding these decisions. Given the legal nature of ADs and MOs and their limited use in minors, we elected to only include documents from patients 18 years or older at the time of transplantation in the final analyses of AD/MO documentation.

Our definition of a GOC conversation was informed by a recent systematic literature review that generated an operational definition for this concept, stating that discussing GOC involves “[establishing] medical care for a patient that [is] informed by patients’ underlying values and priorities, established within the existing clinical context, and used to guide decisions about the use of or limitation(s) on specific medical interventions.”¹⁹ Examples of GOC conversations included descriptions of a patient or family’s decision process to change code status or revisions to a planned medical treatment based on the patient’s current clinical status.

Lastly, other ACP conversations were defined using the aforementioned consensus report that sought to build upon the description of ACP offered by the Institute of Medicine.⁷ Importantly, AD/MO and GOC discussions were excluded. Examples of other ACP conversations included discussions of patients’ values and goals related to treatment decisions that were not constrained to a current medical decision, which included discussions about transitioning to Hospice care or preferences for end of life care. Other ACP conversations also included discussions of patient legacy, and any direct reference to established ACP tools (e.g., Five Wishes, Voicing My Choices).

Constructing the Data Set

The Biomedical Translational Research Information System (BTRIS) at the NIH searched the records of eligible participants (patients aged 15–39 years at the time of HSCT who were enrolled on an allogeneic HSCT protocol between 1/1/2015 and 1/1/2020). All EHR note types available for the patients were reviewed by BTRIS, and any EHR note types which could possibly contain relevant data were selected. A total of 68,860 notes from 219 AYA HSCT recipients were captured.

All documents were reviewed for text pertaining to ADs/MOs, GOC conversations, and other ACP conversations. First, screening terms were used to assist in rapid identification of potentially relevant documents, and included the following: “directive, “palliative,” “comfort,” “family meeting,” “ethic, “interdisciplinary or IDT,” “goals,” “code status,” and “prognosis.” Documents that did not contain screening terms or for which screening terms did not reflect pertinent content were then individually read. Documents were defined as either relevant and grouped into the most applicable category (AD/MO, GOC, or other ACP) or irrelevant. Subsequently, relevant documents were arranged chronologically for each individual patient. Those that contained pertinent text that was directly copied from an earlier note were removed to produce a subset of relevant, unique documents.

Finally, all remaining relevant documents were reviewed independently by the primary researcher (BWP) and a second researcher (AF) and categorized using the established definitions. Discrepant categorizations were reviewed between these two researchers and a third researcher (LW) to establish consensus.

Statistical Analyses

Demographic and clinical information were summarized using frequencies and percentages or means and standard deviations. Age was summarized as a binary variable (18 years or older, or younger than 18 years on day of transplantation); age was also summarized and analyzed as a continuous variable. The following characteristics were collapsed into binary variables prior to analysis: religious affiliation (any religion or no religion), primary language spoken (English or non-English), combined race and ethnicity (identify as white and non-Hispanic or identify as non-white and/or Hispanic), indication for allogeneic HSCT (malignant condition or non-malignant condition), history of palliative care consult (yes or no), and survival status as of 10/1/2020 (living or deceased).

Using available EHR data, three binary outcome variables were created for each individual patient, indicating whether or not the patient had documentation of an AD/MO, GOC, or other ACP conversation. Prior to all sub-analyses of AD/MO documentation, patients of age less than 18 years were excluded. These patients were included in all sub-analyses of GOC and other ACP documentation. First, univariate logistic regression analyses were performed to assess for associations between the three outcome variables and independent variables of interest. Second, variables significantly associated with the outcome variables in univariate logistic regression analyses (p<0.05) and other select variables of interest (0.05 < p <0.10) were used to build multivariate logistic regression models.

The relationship between patient survival status as of 10/1/2020 and whether or not the patient had documentation of each different subtype of ACP was evaluated by Fisher’s Exact Test. Survival status was excluded from univariate and multivariate analyses as subsequent death could not be viewed as a predictive factor for previously occurring ACP documentation. The proportions of documentation of AD/MO, GOC, or other ACP conversations were determined for living and deceased patients. Lastly, in the subset of the deceased patients, Kaplan-Meier curves were created to illustrate the time-to-event relationship between days before death (time) and documentation of AD/MO, GOC, or other ACP conversations (event). STATA IC 14 (StatCorp LP) was used for all analyses, with p < 0.05 considered statistically significant.

RESULTS

Documentation Characteristics:

Over 3,000 types of EHR documents were screened and 531 document types were selected for content analysis (Figure 1). A total of 219 patients receiving allogeneic HSCT between 1/1/2015 and 1/1/2020 were identified, for whom there were a total of 68,860 potentially relevant documents. Following the removal of documents that were non-relevant (n=67,810), contained duplicated pertinent text (n=369) or were related to ADs/MOs in patients <18 years old at the time of transplantation (n=11), a total of 666 relevant documents from 190 patients were identified. A subset of 28 patients had no documented ACP. Following the review of documents with discrepant categorization (n=34), the final distribution was obtained: 330 AD/MO documents from 172 patients, 90 GOC documents from 29 patients, and 246 other ACP documents from 81 patients. Samples of each type of ACP document are provided (Table 1).

Figure 1. Selection of documents.

Consort diagram detailing selection process of advance care planning (ACP) documents. AYA: adolescent and young adult; EHR: electronic health record; AD/MO: advance directive or medical order; GOC: goals of care.

Table 1.

Examples of Subtypes of Advance Care Planning (ACP) Documentation

Subtype of ACP	Definition	Examples
Advance Directive or Medical Order (AD/MO)	AD: “patient-initiated documents [including] living wills and documents that name a health care agent.”18 MO: provider generated “patient preferences with respect to treatment, preferred site for receiving care, and death.”18	Advance directive: Completed and in the chart. Contacted wife to inform her of patient’s ICU transfer, also inquired about decision maker as she is next of kin. She indicated she wishes that patient’s sister remains as patient’s medical decision maker as had been the case while she was present on campus. Code Status: Do not resuscitate, do not intubate (DNR/DNI)
Goals of Care (GOC)	“[a discussion establishing] medical care for a patient that [is] informed by patients’ underlying values and priorities, established within the existing clinical context, and used to guide decisions about the use of or limitation(s) on specific medical interventions.”19	Patient will be now under a DNR/DNI status…Patient’s mother shared that they have come to this decision and are at peace with this decision. This patient and his mother believe that if he was intubated that he will not be able to come off the ventilator so he refused that option. A discussion was had with the family regarding the poor prognosis particularly in the current setting of no known pathology for his respiratory compromise. A lung biopsy has been proposed…The biopsy however is not without risks…and the patient could in fact not survive the biopsy procedure itself. The family vocalized understanding of the risks, and agree to proceed.
Other ACP	“[documentation of] a process that support adults at any age or stage of health in understanding and sharing their personal values, life goals, and preferences regarding future medical care, [with the goal of helping] ensure that people receive medical care that is consistent with their values, goals and preferences during serious and chronic illness.”7 *Excluding AD/MO or GOC documentation	It appears that patient is understanding and accepting that this treatment was not successful. Brother repeatedly stated God can still save him and he holds on to the slim hope. This worker and family discussed about the grieving and the difficulties of not accepting this outcome. He shared that he has told his family that he is not doing well and that “this is it” but his family is in denial and holds on to the “possible miracle”. Although he understands his cancer “has no cure” it appears at times that he has a hard time understanding the gravity of these next few days. I reintroduced use of advanced care planning guide, Voicing My Choices, which was developed by young adults to enhance communication with loved ones and articulate goals for end-of-life planning. She is interested in participating Pt requested a “time” that he will be alive but it was explained that no one would be able to give him a “time”. Pt was very emotional and shared that his wish would be to go back to his home country El Salvador and pass at home.

Patient Characteristics:

Demographic and clinical characteristics of the studied patient population are included in Table 2. Most patients were male (n=126, 58%), non-Hispanic/Latino (n=155, 71%), primarily English-speaking (n=172, 78%), and identified a form of Christianity as their religion (n=142, 65%), with a mean age of 26.7 years. Approximately half of the patients identified their race as white (n=101, 46%). The most common indication for transplantation was leukemia/lymphoma (n=54, 25%), followed closely by sickle cell disease (n=48, 22%). The majority of patients had a documented palliative care consult (n=124, 57%) and most were living at the time of data analysis (n=180, 82%). Regarding documentation patterns, most of the patients over the age of 18 years had a documented AD/MO (n=172, 86%), while a minority of all patients had documentation of GOC (n=29, 13%) or other ACP (n=81, 37%) conversations.

Table 2.

Demographic and Clinical Characteristics of the Allogenic Hematopoietic Stem Cell Transplantation (HSCT) Recipient Population, N=219

Age in years, mean (Std. Dev.)	26.7 (6.4)
Age group, n (%)
Younger than 18 years	18 (8%)
18 Years or Older	201 (92%)
Gender, n (%)
Female	93 (42%)
Male	126 (58%)
Race, n (%)
White	101 (46%)
Black/African-American	56 (26%)
Asian	13 (6%)
Native American/Alaska Native	3 (1%)
Multiracial	7 (3%)
Unknown race	39 (18%)
Ethnicity, n (%)
Non-Hispanic or Latino	155 (71%)
Hispanic or Latino	64 (29%)
Combined Race and Ethnicity, n (%)
White and Non-Hispanic/Latino	76 (35%)
Non-White and/or Hispanic	143 (65%)
Primary Language, n (%)
English	172 (78%)
Spanish	37 (17%)
Other Language	10 (5%)
Religion, n (%)
Christian	142 (65%)
Other religion	14 (6%)
None identified	63 (29%)
Indication for HSCT, n (%)
Leukemia or Lymphoma (Malignancy)	54 (25%)
Sickle Cell Disease	48 (22%)
GATA2 Deficiency	42 (19%)
Chronic Granulomatous Disease	20 (9%)
Aplastic Anemia	19 (9%)
Other immunodeficiencies	36 (16%)
Survival status as of 10/1/2020, n (%)
Living	180 (82%)
Deceased	39 (18%)
Patient had a palliative care consult, n (%)
Yes	124 (57%)
No	95 (43%)
Patient with documented goals of care (GOC) conversation, n (%)
Yes	29 (13%)
No	190 (87%)
Patient with documented other advance care planning (ACP) conversation, n(%)
Yes	81 (37%)
No	138 (63%)
Patient 18 years or older (n=201) with documented Advance Directive or Medical Order (AD/MO), n (%)*
Yes	172 (86%)*
No	29 (14%)*

^*Percentages of total of 201 patients 18 years or older

Factors Associated with ACP Documentation:

Multivariate models of factors associated with documentation of each type of conversation were constructed in a stepwise fashion, summarized in Table 3. Variables were assessed for univariate significance with each of the three outcome variables. Documentation of AD/MO in patients 18 years of age or older was not significantly associated with any of the tested factors. Documentation of GOC was significantly associated with palliative care consultation (OR=2.71, p=0.029) and malignant condition as indication for transplantation (OR=4.148, p=0.001) in univariate analyses. Primary language was not significantly associated with GOC documentation in univariate analyses, but this relationship was tested in the multivariate model. The final multivariate model confirmed significant associations between GOC documentation and a malignant condition as indication for transplantation (OR=4.80, p=0.001) and palliative care consultation (OR=3.33, p=0.012). Documentation of other ACP was only associated with palliative care consultation (OR=2.31, p=0.005).

Table 3.

Factors Associated with Types of Advance Care Planning (ACP) Documentation: Logistic Regression Analyses

Advance directive/medical order documentation in patients 18 years of age or older
Variable	Univariate logistic regression			Multivariate logistic regression
	OR	95% CI	p value	OR	95% CI	p value
Age	1.02	0.96–1.10	0.494	-	-	-
Gender, female	0.66	0.30 – 1.44	0.295	-	-	-
Religion, any	1.36	0.59 – 3.13	0.471	-	-	-
Primary Language, English	0.53	0.17 – 1.61	0.261	-	-	-
Race and ethnicity, white and non-Hispanic	0.53	0.24 – 1.18	0.113	-	-	-
Transplantation indication, malignancy	1.86	0.67 – 5.15	0.234	-	-	-
Palliative care consult, yes	2.22	1.00 – 4.94	0.051	-	-	-
Goals of care documentation
Variable	Univariate logistic regression			Multivariate logistic regression
	OR	95% CI	p value	OR	95% CI	p value
Age	1.03	0.97–1.09	0.365	-	-	-
Gender, female	0.57	0.25–1.31	0.185	-	-	-
Religion, any	1.31	0.53–3.25	0.555	-	-	-
Primary Language, English	0.46	0.20–1.07	0.072	0.97	0.35 – 2.68	0.951
Race and ethnicity, white and non-Hispanic	0.99	0.43 – 2.25	0.979	-	-	-
Transplantation indication, malignancy	4.15	1.85–9.32	0.001	4.80	1.83 – 12.54	0.001
Palliative care consult, yes	2.71	1.11–6.65	0.029	3.33	1.30 – 8.53	0.012
Other ACP documentation
Variable	Univariate logistic regression			Multivariate logistic regression
	OR	95% CI	p value	OR	95% CI	p value
Age	1.03	0.98–1.07	0.227	-	-	-
Gender, female	1.14	0.65–1.98	0.650	-	-	-
Religion, any	1.38	0.74–2.57	0.308	-	-	-
Primary Language, English	0.93	0.48–1.81	0.834	-	-	-
Race and ethnicity, white and non-Hispanic	1.08	0.61 – 1.92	0.793	-	-	-
Transplantation indication, malignancy	1.52	0.81–2.83	0.192	-	-	-
Palliative care consult, yes	2.31	1.29–4.11	0.005	2.31	1.29–4.11	0.005

All variables other than age were condensed to binary variables with the following categories: gender (female/male), religion (any/none), primary language (English/other), ethnicity (Hispanic/Non-Hispanic), race (white/other), survival status (deceased/living), transplantation indication (malignancy/non-malignancy), and palliative care consult (yes/no). OR: odds ratio; CI: confidence interval

Additional Characteristics of ACP Documentation in Deceased Patients

Compared to those surviving as of 10/1/2020, deceased patients were significantly more likely to have had documentation of AD/MO (p=0.006), GOC (p<0.001), or other ACP conversations (p<0.001, Table 4). Considering all subtypes of ACP together, the majority of documentation occurred in the deceased subset of patients (54.4% of the total, Table 5). In the subset of deceased patients, Kaplan-Meier curves show the timing of individual occurrences of each type of ACP prior to patient death (Figure 2). Considering types of documentation separately in deceased patents, half of AD/MO documentation occurred in the last 67 days of life, half of other ACP documentation occurred in the last 20 days of life, and half of GOC documentation occurred in the final 2 days of life.

Table 4.

Relationship Between Survival Status as of 10/1/2020 and Documentation of Advance Care Planning (ACP)

Survival status as of 10/1/2020	Patient 18 years of age or older with advance directive or medical order documentation*		Patient with goals of care documentation		Patient with other ACP documentation
	Yes	No	Yes	No	Yes	No
Deceased (n=39 patients)	34	0	24	15	33	6
Living (n=180 patients)	138	29	5	175	48	132
Fisher’s Exact Test, two-sided	p =0.006		p<0.001		p<0.001

^*Includes only the 201 patients 18 years or older at the time of transplant

Table 5.

Proportions of Advance Care Planning (ACP) Documents (N=666) in Deceased and Living Patients

Survival status as of 10/1/2020	Advance directives or medical orders, n (%)*	Goals of care documents, n (%)	Other ACP documents, n (%)	Total documents, n (%)
Deceased	106 (32.1%)	85 (94.4%)	171 (69.5%)	362 (54.4%)
Living	224 (67.9%)	5 (5.6%)	75 (30.5%)	304 (45.6%)

^*Includes only the 201 patients 18 years or older at the time of transplant

Figure 2. Cumulative occurrence of subtypes of advance care planning documents in deceased patients.

Kaplan-Meier curves of cumulative occurrence of individual advance care planning documents (N=362) over time in the subset of 39 deceased patients prior to death. AD/MO: advance directive/medical order; GOC: goals of care; ACP: advance care planning. *: 50% of AD/MO documents in the last 67 days before death in patients 18 years or older; **50% of other ACP conversations documented in the last 20 days before death; ***50% of GOC conversations documented in the last 2 days before death.

CONCLUSIONS

Our study reveals patterns of documentation of different types of ACP in a large sample of AYA patients who received allogenic HSCT, a population which has historically encountered many barriers to these conversations. Most patients (18 years of age or older) had documented ADs/MOs (86%). However, substantially fewer patients overall had documented GOC discussions (13%) or other discussions pertinent to ACP (37%). Previous studies in older adults receiving HSCT have identified similar differences in the frequency of documentation of subtypes of ACP,^{14, 15} which we also demonstrate in this AYA population. The low rates of documented GOC discussions and other ACP discussions compared to ADs/MOs seen in our study with AYAs suggest that these types of conversations may be harder for HSCT recipients to access and/or for providers to capture in the EHR. Our results also highlight the importance of studying subtypes of ACP, as conversations beyond ADs/MOs may be required to achieve stronger alignment between AYAs, families, and providers.

In further exploration of the subtypes of ACP, multivariate models demonstrated specific factors associated with documentation GOC and other ACP conversations. Of note, no factors were significantly associated with documentation of ADs/MOs in patients aged 18 years or older. This is likely a reflection of the requirement in some of the included clinical trials for baseline completion of ADs in adult patients. Palliative care consultation was an important factor significantly associated with documentation of GOC and other ACP conversations in univariate analyses and in the final multivariate model for GOC documentation, a finding seen in other studies.^20,21 Our results suggest that palliative care consultation promotes a wide variety of ACP discussions, both formal and informal. Separately, informed by the process of reviewing many documents for pertinent language, we also hypothesize that the structure of palliative care documentation increases the visibility and accessibility of these conversations.

Patients with an underlying malignant condition as the indication for HSCT were significantly more likely to have a documented GOC discussion in the multivariate analysis, which has also been demonstrated in previous studies in older adults.^{20, 22} As patients with a malignant condition were significantly more likely to die following HSCT (p=0.004, data not shown) and as the majority of GOC discussions occurred in patients that died, this association between GOC documentation and malignancy is likely related to the higher mortality of this group of patients. It seems likely that the preceding experiences and different prognostic expectations of patients with malignant conditions may facilitate or necessitate GOC conversations.

The association between survival status and documentation of any type of ACP suggests that ACP-related conversations more often occurred out of necessity and towards the end-of-life and not as a general practice for patients pursuing a life-threatening treatment. Accordingly, we found that the bulk of ACP discussions in subsequently deceased patients occurred very close to the end of life. Although no previous studies specifically consider this time-to-event relationship between days prior to death and occurrence of ACP documentation, our results illustrate the barriers to ACP discussions revealed in previous studies of pediatric and AYA patients.^{13, 23} In one qualitative study of transplant and ICU physicians caring for AYA HSCT recipients, the major barrier to engaging patients in ACP discussions was described as “the conflict between the conceptual support of ACP in allowing patients to make decisions earlier in their illness, with the belief by [providers] that patients can never truly understand the future implications of the decisions they make.”²³ The result of this philosophy, supported by our findings, is a rapid peak in ACP activity only when death is certain.

Factors that have been associated with higher rates of ACP or AD/MO completion in previous studies were not significant in our analysis. Multiple studies in adults have considered an association between older age and completion of ACP and found conflicting results.^{20, 24–27} Several other studies have found that patients identifying as Black/African-American or Hispanic were significantly less likely to have engaged in various types of ACP,^{24, 28–30} less likely to be aware of palliative and hospice care,³¹ and more likely to want aggressive end of life care including CPR.³² Multiple other studies have found significant associations between participation in ACP and higher educational attainment^{22, 26, 29, 33} or socioeconomic status.^{31, 34}

Several limitations of our retrospective study are important to note. We were not able to access documentation from patients’ prior health care institutions, which may have included ACP documentation. As a referral institution, we captured data based on available EHR documentation. If patients had previously engaged in ACP discussions that were not captured in our EHR but were accessible to clinical staff at our institution, providers may have not engaged in potentially redundant conversations. It is also possible that patients and providers at our institution engaged in ACP discussions that were not documented. Our data set did not contain all demographic data shown in previous studies to be associated with ACP completion, with 18% of patients with unknown race and no information on educational attainment or socioeconomic status. We also recognize that although our sample included patients 15 to 39 years old, only 18 patients (8% of the total sample) were less than 18 years old at the time of transplant. These results may be less generalizable to younger adolescent patients. Finally, while the definitions we used for AD/MO, GOC, and other ACP discussions are rooted in the literature, there is some inherent subjectivity in how we chose to categorize documents into these three categories.

Our study is strengthened by the breadth, both in number and variety, of screened documents for the 219 HSCT recipients. It was important to include many types of documents to ensure that the maximum amount of available data was captured, as there is no single type of document shared amongst providers that captures ACP information. Through the additional steps of removing documents with duplicate text (a known pervasive problem in all EHR documentation) and employing two additional researchers to review (AF) and address discrepant categorizations of documents (LW), we distilled this large quantity of raw information into a clinically applicable data set. This process was completed manually without the assistance of natural language processing algorithms. Although this process was more labor intensive, this methodology was both more inclusive in the beginning and more selective in the end, allowing us to feel confident that the available EHR data was comprehensively reviewed. The heterogeneity of our population, with no single indication for transplantation applying to more than 25% of patients, makes our results more generalizable to all AYAs requiring allogeneic HSCT.

Our results have direct implications for future studies that hope to improve access to ACP and documentation of these discussions for AYA HSCT recipients. Considering the low frequencies of ACP discussions other than ADs or MOs seen in our analysis and the concentration of these conversations in dying patients at the end of life, there is room for improvement, particularly in patients receiving HSCT for non-malignant conditions as these patients had significantly less documentation of GOC conversations. Multiple groups are actively working on this quality gap in other AYA populations with serious illness through interventions that aim to improve patient, family, and provider engagement with ACP.^35–38 Others have focused on improving documentation, showing that the use of EHR note templates can significantly improve concordance between recorded ACP conversations and written text.³⁹ Consistent and predictable documentation in a single note type, shared amongst different care providers, would also make this information more visible and, hopefully, more impactful. For AYAs receiving HSCT, novel interventions are needed to tackle the unique challenges encountered by this vulnerable and highly diverse population. Further prospective studies, with a focus on longitudinal communication interventions with patient, family, and provider participation, clear and accessible documentation, and ideally with a randomized control design to increase rigor are needed to garner greater systemic support for these endeavors and to have the greatest positive impact on patients, caregivers, and providers.

HIGHLIGHTS:

• Adolescents and young adults (AYAs) face barriers to advance care planning (ACP).

• Hematopoietic stem cell transplantation (HSCT) carries high risk, necessitating ACP.

• 68,000 electronic record documents were reviewed in AYA HSCT recipients for ACP.

• Palliative care consults and a malignant diagnosis increased ACP documentation.

• Most ACP occurs in patients that ultimately died in the last weeks to days of life.