Patient Identification for Serious Illness Conversations: A Scoping Review

Serious illness conversations aim to align medical care and treatment with patients’ values, goals, priorities, and preferences. Timely and accurate identification of patients for serious illness conversations is essential; however, existent methods for patient identification in different settings and population groups have not been compared and contrasted. This study aimed to examine the current literature regarding patient identification for serious illness conversations within the context of the Serious Illness Care Program and/or the Serious Illness Conversation Guide. A scoping review was conducted using the Joanna Briggs Institute guidelines. A comprehensive search was undertaken in four databases for literature published between January 2014 and September 2021. In total, 39 articles met the criteria for inclusion. This review found that patients were primarily identified for serious illness conversations using clinical/diagnostic triggers, the ’surprise question’, or a combination of methods. A diverse assortment of clinicians and non-clinical resources were described in the identification process, including physicians, nurses, allied health staff, administrative staff, and automated algorithms. Facilitators and barriers to patient identification are elucidated. Future research should test the efficacy of adapted identification methods and explore how clinicians inform judgements surrounding patient identification.


Introduction
Conversations in serious illness are held to understand and support patients' values, goals, priorities, and preferences in relation to their health and medical care [1]. Kelley and Bollens-Lund [2] define the term 'serious illness' as 'a health condition that carries a high risk of mortality and either negatively impacts a person's daily function or quality of life, or excessively strains their caregivers' (p. S-8). Serious illness conversations have been associated with improved patient outcomes, such as reduced anxiety and suffering, in addition to improved quality of life and satisfaction [3][4][5]. Although scholars recommend having such conversations when patients are relatively stable, all too often eligible patients are not identified until late in the illness process [6,7]. To ensure patients and their families receive care that is concordant with their values, goals, priorities, and preferences, evidencebased approaches are required to identify eligible patients for serious illness conversations in a timely manner.
The Serious Illness Care Program (SICP), developed by Ariadne Labs, aims to equip clinicians with the knowledge and skills to undertake more, better, and earlier serious illness conversations [1,8]. This multicomponent program is comprised of patient identification, clinician training, workflow development, medical record documentation templates,

Selection of Evidence
The initial search results were imported for processing using the bibliographic reference management software EndNote X7.8 for Windows. The first author (RB) conducted the initial title and abstract screening based on the eligibility criteria. The full text was viewed in cases where the title and abstract did not provide sufficient material to inform a decision. Following the initial screening, all articles were read in full and evaluated for inclusion using the same criteria. Another author (AS) reviewed all articles marked for inclusion/exclusion and any uncertainty was discussed between the authors until consensus was reached. Several articles were noted to have originated from overarching study clusters and therefore used the same identification methods; however, these articles were deemed eligible for inclusion as they explored unique study aims, contained different descriptions of patient identification, and illustrated the evolution of how identification methods have changed over time.

Data Charting Process
Data were extracted using charting tables created by the authors, based on the guidelines proposed by The Joanna Briggs Institute [19]. The first charting table collected descriptive information, such as the author(s), publication year, study setting (country, clinical context), study aims, research methods, participants (if applicable), and study results/conclusions. A second charting table was used to collate data regarding patient identification for serious illness conversations, and any additional information relevant to the aim and research questions. Articles were grouped according to their original study cluster and then listed in chronological order of publication year to illuminate the evolution of patient identification methods in SICP/SICG-related research over time. The preliminary data charting tables were piloted on five articles to confirm extraction of relevant information, after which data from the remaining literature were extracted. preliminary data charting tables were piloted on five articles to confirm extraction of relevant information, after which data from the remaining literature were extracted.

Synthesis of Results
A deductive approach was used to organize and summarize information from the literature to address the study aim and questions. Extracted data were compared and contrasted to identify patterns, similarities, and differences in descriptions of patient identification for serious illness conversations. Emerging patterns were organized into categories related to the research questions. These groupings were discussed at length and all authors agreed upon the final results.

Included Articles
The initial database searches returned 444 results (CINAHL n = 105; MedLine n = 152; PsychInfo n = 29; PubMed n = 158). A list of 44 articles pertaining to the SICP published by Ariadne Labs was added to the raw list of articles from the initial database search as these articles were directly related to the SICP or SICG. Following the removal of duplicates, 181 articles progressed to title and abstract screening. Of these, 65 met the criteria for full-text review, and 39 met the inclusion criteria for the study. The reference lists of the articles marked for inclusion were examined, and an additional 16 articles were screened at title and abstract level. Of these, three underwent full-text review, but none met the inclusion criteria for the study. In total, 39 articles were eligible for inclusion in this scoping review ( Figure 1).

Description of Articles
The articles were set in a range of inpatient and outpatient clinical settings and comprised of staff, patient, and relative participants. A variety of research methods were used, and the majority of studies originated from the United States (n = 34). The full characteristics of the included articles are detailed in Appendix B.

Description of Patient Identification
Patient identification for serious illness conversations was described in various ways. Some articles explicitly outlined the entire identification process, clearly stating who was responsible for patient identification, the guidelines for patient identification, the procedures by which patients were identified, the training provided for patient identification (if any), and justification for these procedures. However, in some cases, it was not possible to delineate the separate parts of this process, for example, if it was not specified whether the clinician who held the serious illness conversation was the same person who identified the patient. Detailed descriptions and excerpts regarding how patients were identified and who performed the identification are presented in Table 1.   (2) Response to a triggering medical event or assessment of the patient's health status, which led clinicians to initiate a discussion; (3) Responding to patient-or family-initiated statements that clinicians interpreted as a sign of readiness for the conversation (p. 461).

University of Pennsylvania Machine Learning Cluster Randomized Trial (n = 2)
Manz et al.
(2020) [30] Clinical/diagnostic triggers An EHR-based machine learning algorithm uses real-time patient data, including demographic information, comorbidities, lab values, and encounters with the health system over the prior six months, to estimate individuals' risk of dying in the subsequent six months (p. 2). Clinicians could view a list of up to six patients scheduled for a visit in the coming week with the highest-risk of machine-predicted six-month mortality (p. 4).  [31] Clinical/diagnostic triggers Clinicians could review a list of patients scheduled for the following week in their clinic who had a high risk of mortality. Mortality risk was determined by a machine learning algorithm, which used structured EHR data to predict risk of 180-day mortality. Clinicians could view a list of up to 6 patients with the highest predicted 180-day mortality risk (p. 3).

EHR-based machine learning
Clinicians [physicians, nurse practitioners, physician assistants] could review patients scheduled for the following week in their clinic who had a 'high risk' of mortality (p. 2).

Massachusetts General Hospital Cluster (n = 2)
Gace et al.   team coded lists of patients as receiving palliative treatment. Clinicians reviewed their individual lists of pre-screened patients and used the SQ to identify those felt to be at risk of death in the next 1-2 years. At primary care sites, general practitioners used a practice register of patients thought to be in the last 12 months of life to identify patients who they felt should be offered a serious illness conversation (p. 5). Brigham Health: Initially, screened patients deemed eligible for the iCMP were asked the SQ on a paper survey as part of the enrolment process, but this missed many patients. The second patient selection algorithm expanded the timeframe of the SQ to 2 years and asked it as part of a SICP-specific electronic screening survey (p. 6). The patient selection algorithm expanded the SQ and asked it as part of an electronic screening survey sent to doctors, care coordination nurses and social workers (p. 6).

University of Pennsylvania Health
Lally et al.
(2020) [41] Clinical/diagnostic triggers A daily dashboard identifies when ACO patients are admitted to the hospital, and patients who meet the criteria for CCM were enrolled. Any patient identified on this daily report is added to a spreadsheet and the data analyst looks for a documented serious illness conversation within 14 days of discharge from the hospital (p. 113).
A dashboard identified when ACO patients are admitted. Then nurse case managers enrolled patients who met the criteria (p. 113).

Ma et al. (2020) [42]
Clinical/diagnostic triggers Patients were eligible to be enrolled in the SICP if they were admitted to a medical ward, had a stay of at least 48 hours, and received a score of 5 or 6 on the interRAI Emergency Department Screener on admission (p. E449).
The unit champion (former bedside nurse from the medical ward) screened medical inpatients for eligibility. The unit champion triggered clinicians to have the conversations (p. E449).   Almost half of the articles (n = 17) described specific clinical-and/or diagnostic-related triggers as their primary method for identifying patients for serious illness conversations. Several articles (n = 9) reported using the SQ (one or two years) as their principal identification method, and the remaining articles (n = 13) described using some combination of the SQ, clinical/diagnostic-related triggers, patient/family request, and clinician judgement. Physicians were the most frequently named clinicians in the identification process, followed by physician's assistants, nurse practitioners, medical assistants, nurses, social workers, care coordinators, and allied health staff. Research and administration staff were also said to be actively involved in identifying eligible patients, and several articles indicated that Electronic Health Record (EHR)/Electronic Medical Record (EMR) systems/algorithms were instrumental in the patient identification process.

Patient Identification among Population Groups and Clinical Settings/Contexts
The ways in which patients were identified for serious illness conversations varied across population groups and clinical settings/contexts ( Table 2). The SQ (1 or 2 years) was described in the oncology setting (n = 7), as were clinical/diagnostic triggers (n = 2), and a combination of methods (n = 4). Medical (i.e., acute, inpatient, outpatient) and other specialties (i.e., intensive care, pediatrics) clinical settings/contexts primarily identified patients using clinical triggers (n = 11). The primary care setting revealed the greatest diversity in identification methods.

Facilitators and Barriers to Patient Identification
Twenty-one articles specified facilitators and/or barriers relevant to patient identification. Potential facilitators were described as including simple and structured screening systems [37], EHR/EMR support and reminders [45], and clinician education [5,29]. Tools such as the SQ were said to improve clinician buy-in and contemplation surrounding recruitment for, and conduction of, serious illness conversations [26]. With regards to barriers, several studies outlined potential discrepancies in the interpretation of identification criteria. Billie and Letizia [39] wrote that there were 'several situations in which a case manager evaluated the patient as appropriate for an SI [serious illness] conversation, although he or she did not meet the established SI criteria' (p. 226). Other studies also indicated ambiguity surrounding eligibility criteria, for example, variation in the interpretation of clinical characteristics [34] and differences in understanding what constituted a 'serious illness' [50]. Uncertainty surrounding the ideal timing of the conversation, and lack of time to have the conversation, were also stated to be barriers to identification, as recruitment could be limited by patient number or urgency [12,52]. Lakin and colleagues described disparities in the ways in which clinicians identified patients, with staff stating 'no, we don't have a process for patient selection', 'when I do patient selection, I sit down and look at a list of patients and just choose', and 'when I do patient selection, I sit down along with a nurse and we look together at a list of patients choose who needs the conversation' (p. 760) [27]. It could also be challenging to answer the SQ for patients with multi-morbidities, cognitive impairment, or frailty as life expectancy can vary [29]. Furthermore, among larger, sicker patient groups, the SQ could be inadequate or difficult to operationalize [12]. It was suggested that relying solely on the SQ could overlook some patients who would benefit from a palliative approach [25,26,50]; similarly, replying 'no' to the SQ was not always thought to require a serious illness conversation [50]. Triggering criteria for a conversation did not guarantee that a conversation would be held, and without a structured tracking system it could be difficult for clinicians to know who had, or had not, completed serious illness conversations [35,43].
Lack of a systematic approach to identification (i.e., EHR/EMR queries, use of simple trigger thresholds) was said to be a barrier to identifying appropriate patients for serious illness conversations [47]. Studies stated that it could be difficult for clinicians to manually identify patients, particularly when there was no structured EHR/EMR support [12,25,47,48,52]. However, EHR/EMR systems may neglect to flag seriously unwell patients with poor prognoses [31] as not all trigger criteria are available for algorithmic computation [35]. Additionally, it takes time and (human) resources to support such systems [35]. Another potential issue was the efficacy and reliability of EHR/EMR algorithmic triggers, as some have not undergone formal validation and may therefore under-(or over-) identify patients for serious illness conversations [32,33].

Discussion
This scoping review examined the current literature regarding patient identification for serious illness conversations within the context of the SICP and/or the SICG. The findings revealed that patients were primarily identified using the SQ or clinical/diagnostic triggers. Combinations of criteria and development of automated systems/algorithms indicate ongoing evolution and adaptation of identification methods for different clinical settings/contexts. A diverse range of clinicians was involved in identifying and conducting serious illness conversations, with physicians, nurses, and automated EHR/EMR systems the most commonly named actors in the identification process. Barriers and facilitators were described regarding clinician understanding of the concepts and identification criteria, structured support systems, and training/education.
In recent years, the SQ has emerged as a useful screening tool to identify patients nearing the end of life who may benefit from a palliative approach to care. A major advantage of the SQ is that it encourages a level of closeness between the clinician and the patient, prompting active contemplation of the patient's unique situation and care needs [26]. It is, however, important to note that the SQ has reported mixed sensitivity (low to reasonable/good), and responses to the question are said to be impacted by the clinician's familiarity with both the question and the patient [50,[55][56][57]. Furthermore, repeatedly asking oneself the SQ is not only time consuming but can be emotionally exhausting given the gravity of the overarching topic [12,58]. These findings revealed how the use of the SQ in the SICP/SICG has evolved over time with 1-and 2-year alternatives, and combinations with clinical/diagnostic triggers, clinician judgement, and patient/family factors. Nevertheless, the efficacy of these adaptations and combinations to accurately identify patients for serious illness conversations has not yet been established.
The results show that clinical/diagnostic triggers have emerged as a popular identification method, particularly in acute and specialty clinical contexts. These criteria ranged in specificity from targeted lab values, to entire patient populations. As various phases of illness are often distinguished by changes in function, pain, perception, or physical ability, monitoring of clinical/diagnostic triggers provides valuable information to inform patient identification at the so-called 'right' time [59,60]. This is important because if (mis)identification occurs too early or too late in the illness trajectory it can result in undue physical, mental, emotional, and spiritual labor for both patients and clinicians [61]. However, according to Kelley and Bollens-Lund [2], identifying seriously ill patients using administrative data alone (i.e., diagnosis codes, hospitalizations) is not sufficient. This may support the use of combined methods for identification, such as prognosis-related triggers and indicators of critical loss, or clinical/diagnostic triggers and calculations of resource use [13,55,62]. Further research would therefore be useful to compare and contrast clinical/diagnostic triggers between specialties and explore the effectiveness of different combinations and hybrid methods in identifying patients for serious illness conversations.
This study found that of the numerous clinicians named in the identification process, physicians were the most common identifiers of patients for serious illness conversations. Other clinicians or non-clinical resources that were described in this process included physician's assistants, nurse practitioners, medical assistants, nurses, social workers, care coordinators, allied health staff, researchers, and EHR/EMR applied algorithms. It appears that the roles and responsibilities in relation to patient identification have evolved over time to include a more diverse range of clinicians and resources. However, in some articles it was not stated who performed the identification, and in others it was unclear if the clinicians who received SICP training performed the patient identification and held the subsequent conversation. Transparency was lacking regarding when in the care trajectory patients were considered for serious illness conversations, nor were there extensive justifications as to why particular triggers were selected or excluded. For example, it is interesting to note that few studies included patient/family benefit or readiness as a criterion for these conversations. Perhaps this is because it is still unclear whether the 'tipping point' for recognition of seriously ill patients is more closely linked to demographics, diagnosis, symptoms, prognosis, clinical context, and critical loss, or to the clinicians' own perceptions and experiences [62]. This reinforces the need to develop identification protocols that provide specific guidance regarding health/illness trajectories and their associated conversations [2,63]. It also seems important to distinguish (and report) each step in the identification process, namely (1) how potential patients are identified; (2) how this information is communicated to clinicians; and (3) how clinicians evaluate patient eligibility (and readiness) for conversations. While this scoping review only explored part of this process, it would be useful for future studies to examine how clinicians justify decisions regarding patient eligibility for serious illness conversations, including motivations as to why they did or did not initiate a conversation in practice.

Limitations
This scoping review has several limitations. First, the review was limited to the literature that described patient identification for serious illness conversations in the context of Ariadne Labs' SICP/SICG. Studies that used the SICP and/or the SICG, but did not describe patient identification, or that described patient identification but did not explicitly state their affiliation with the SICP/SICG, were therefore excluded or omitted. Other studies pertaining to serious illness conversations that used different conversation programs, tools, or guides may outline different identification methods. As the SICP informs the SICG, and vice versa, we did not separate the included articles into groups that used the SICP only, the SICG only, or some form of adaptation. Further, while the majority of studies in this review originated from the United States, a recent survey of the Serious Illness Care Community of Practice indicated that the program and the guide have been implemented in a wide range of clinical settings across 45 countries [11]. Research and publications from outside North America regarding the SICP and SICG are therefore ongoing.
It should also be noted that the number of articles written about the SICP/SICG outnumber the number of unique studies. This is because the SICP and SICG originated from a research group based out of Ariadne Labs, a joint center for health systems innovation at Brigham & Women's Hospital and the Harvard T.H. Chan School of Public Health, and has been subsequently adopted for implementation at several other clinical/research sites (i.e., Massachusetts General Hospital, the University of Pennsylvania Health System, etc.). These articles were reported and analyzed individually due to differences in descriptions of identification methods and changes to identification methods that may have occurred over time. The relationship between the articles and study clusters is highlighted in the tables and footnotes for transparency.
Finally, a distinguishing feature of scoping reviews is their focus on providing a broad overview of the existing literature, irrespective of type or quality; hence, a formal evaluation of the risk of bias of the included articles was not undertaken [17][18][19]. As such, these findings are exploratory/descriptive in nature and do not seek to explain or analyze the literature in relation to policy or practice [17]. This study did, however, take care to provide a detailed description of the characteristics of the included literature so it is left to the reader to decide the generalizability and relevance of these findings.

Conclusions
The findings from this scoping review shed light on current methods, processes, and practices used to identify patients for serious illness conversations in the context of the SICP and/or the SICG. Identification methods varied among different clinical settings/contexts and included the SQ, clinical/diagnostic triggers, and combinations of criteria. A constellation of clinicians and resources were described in the identification process. Although this study provides an initial understanding of the existent patient identification methods for serious illness conversations, reporting methods for identification were inconsistent and there appears to be a lack of validated and standardized protocols for comparison. As timely patient identification is arguably one of the most challenging components of the SICP/SICG, future research is necessary to explore how clinicians justify and motivate decisions regarding patient identification and to establish the efficacy of these adapted/combined identification methods.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
Appendix A Table A1. Database search strategy.

Database Search Strategy
CINAHL TI ("serious illness communication" OR "serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") OR AB ("serious illness communication" OR"serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") Limiters: Date of publication 20140101-20210901; English language MedLine TI ("serious illness communication" OR "serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") OR AB ("serious illness communication" OR"serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") Limiters: Date of publication 20140101-20210901; English language PsychInfo TI ("serious illness communication" OR "serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") OR AB ("serious illness communication" OR "serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model") OR KW ("serious illness communication" OR "serious illness program *" OR "serious illness care" OR "serious illness conversation *" OR "serious illness model" Note: The terms "serious illness program *" and "serious illness model" were not recognized by PubMed. Oncology.

Appendix B
-- We believe that developing scalable models for improving SICs will contribute to better alignment of healthcare with the preferences of oncology patients, and eventual extension to other patient populations and care settings. Patients in the clinics with the program implemented were more likely than those in comparison clinics to have SICs-including discussion of values and goals-documented in patients' medical records. Clinicians who participated also reported high satisfaction with training they received as part of the program, which they regarded as effective. (To) (1) identify the barriers to SICs in the dialysis population, (2) review best practices in and specific approaches to conducting SICs, and (3) offer solutions to overcome barriers as well as practical advice, including specific language and tools, to implement SICs in the dialysis population.
Special issue article. End-stage renal disease.
--Implementing SICs for patients on dialysis involves identifying patients at the highest risk of adverse outcomes, triggering conversations, and conducting them routinely. The Guide provides a tested, scalable structure for conducting these conversations that can be used by nephrologists and other dialysis clinicians, and it can be adapted further to meet the needs of this population. Documentation and sharing of conversation content and identification of metrics to drive performance improvement are also essential to the successful implementation of SICs for patients on dialysis.  We are conducting a cluster randomized trial comparing team-based to clinician-focused ACP using the SICP.
Protocol for a cluster randomized trial. Primary care.
--Our dissemination will report the results of comparing the two models and the implementation experience of the practices to create guidance for the spread of ACP in primary care. Describes the design of a stepped-wedge cluster randomized trial to evaluate the impact of an intervention that employs machine learning-based prognostic algorithms and behavioral nudges to prompt oncologists to have SICs with patients at high risk of short-term mortality.
Stepped-wedge cluster randomized controlled trial. Oncology clinics.
--This trial represents a novel application of machine-generated mortality predictions combined with behavioral nudges in the routine care of outpatients with cancer.    The finalized PediSICP intervention includes a structured HCP and family ACP communication occasion supported by a 3-part communication tool and bolstered by focused HCP training. We also identified strategies to ameliorate implementation barriers.