BACKGROUND: Advance care planning is the process of discussing health care treatment preferences based on patients' personal values, and it often involves the completion of advance directives. In the first months of 2020, a novel coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), began circulating widely in the American state of Colorado, leading to widespread diagnosis of coronavirus disease (COVID-19), hospitalizations, and deaths. In this context, the importance of technology-based, non-face-to-face methods to conduct advance care planning via patient portals has increased.
OBJECTIVE: The aim of this study was to determine the rates of use of a web-based advance care planning tool through a health system-based electronic patient portal both before and in the early months of the COVID-19 pandemic.
METHODS: In 2017, we implemented web-based tools through the patient portal of UCHealth's electronic health record (EHR) for patients to learn about advance care planning and complete an electronically signed medical durable power of attorney (MDPOA) to legally appoint a medical decision maker. Patients accessing the portal can complete and submit a legally valid MDPOA, which becomes part of their medical record. We collected data on the patients' date of MDPOA completion, use of advance care planning messaging, age, sex, and geographic location during the early phase of the COVID-19 pandemic (December 29, 2019, to May 30, 2020).
RESULTS: Over a 5-month period that includes the early phase of the COVID-19 pandemic in Colorado, total monthly use of the advance care planning portal tool increased from 418 users in January to 1037 users in April and then decreased slightly to 815 users in May. The number of MDPOA forms submitted per week increased 2.4-fold after the stay-at-home order was issued in Colorado on March 26, 2020 (P<.001). The mean age of the advance care planning portal users was 47.7 years (SD 16.1), and 2206/3292 (67.0%) were female. Women were more likely than men to complete an MDPOA, particularly in younger age groups (P<.001). The primary use of the advance care planning portal tools was the completion of an MDPOA (3138/3292, 95.3%), compared to sending an electronic message (148/3292, 4.5%). Over 50% of patients who completed an MDPOA did not have a prior agent in the EHR.
CONCLUSIONS: Use of a web-based patient portal to complete an MDPOA increased substantially during the first months of the COVID-19 pandemic in Colorado. There was an increase in advance care planning that corresponded with state government shelter-in-place orders as well as public health reports of increased numbers of COVID-19 cases and deaths. Patient portals are an important tool for providing advance care planning resources and documenting medical decision makers during the pandemic to ensure that medical treatment aligns with patient goals and values.
BACKGROUND: Digital tools to document care preferences in serious illnesses are increasingly common, but their impact is unknown. We developed a web-based advance directive (AD) featuring (1) modular content eliciting detailed care preferences, (2) the ability to electronically transmit ADs to the electronic health record (EHR), and (3) use of nudges to promote document transmission and sharing.
OBJECTIVE: To compare a web-based, EHR-transmissible AD to a paper AD.
METHODS: Patients with gastrointestinal and lung malignancies were randomized to the web or paper AD. The primary outcome was the proportion of patients with newly documented advance care plans in the EHR at 8 weeks. Secondary outcomes assessed through an e-mail survey included the change in satisfaction with end-of-life plans, AD acceptability, and self-reported sharing with a surrogate.
RESULTS: Ninety-one participants were enrolled: 46 randomly allocated to the web AD and 45 to paper. Thirteen patients assigned to web AD (28%) had new documentation versus 7 (16%) assigned to paper (P = .14). Adjusted for demographic factors and primary diagnosis, the odds ratio of new documentation with web AD was 3.7 (95% CI: 0.8-17.0, P = .10). Satisfaction with advance care planning and AD acceptability were high in both groups and not significantly different. Among patients completing web ADs, 79% reported sharing plans with their caregivers, compared with 65% of those completing paper ADs (P = .40).
CONCLUSION: Web-based ADs hold promise for promoting documentation and sharing of preferences, but larger studies are needed to quantify effects on these intermediate end points and on patient-centered outcomes.
BACKGROUND: As part of a broader study to improve the capacity for advance care planning (ACP) in primary healthcare settings, the research team set out to develop and validate a computerized algorithm to help primary care physicians identify individuals at risk of death, and also carried out focus groups and interviews with relevant stakeholder groups. Interviews with patients and family caregivers were carried out in parallel to algorithm development and validation to examine (1) views on early identification of individuals at risk of deteriorating health or dying; (2) views on the use of a computerized algorithm for early identification; and (3) preferences and challenges for ACP.
METHODS: Fourteen participants were recruited from two Canadian provinces. Participants included individuals aged 65 and older with declining health and self-identified caregivers of individuals aged 65 and older with declining health. Semi-structured interviews were conducted via telephone. A qualitative descriptive analytic approach was employed, which focused on summarizing and describing the informational contents of the data.
RESULTS: Participants supported the early identification of patients at risk of deteriorating health or dying. Early identification was viewed as conducive to planning not only for death, but for the remainder of life. Participants were also supportive of the use of a computerized algorithm to assist with early identification, although limitations were recognized. While participants felt that having family physicians assume responsibility for early identification and ACP was appropriate, questions arose around feasibility, including whether family physicians have sufficient time for ACP. Preferences related to the content of and approach to ACP discussions were highly individualized. Required supports during ACP include informational and emotional supports.
CONCLUSIONS: This work supports the role of primary care providers in the early identification of individuals at risk of deteriorating health or death and the process of ACP. To improve ACP capacity in primary healthcare settings, compensation systems for primary care providers should be adjusted to ensure appropriate compensation and to accommodate longer ACP appointments. Additional resources and more established links to community organizations and services will also be required to facilitate referrals to relevant community services as part of the ACP process.
BACKGROUND: Historically, it has been assumed that the Emergency Department (ED) is a place for maximally aggressive care and that Emergency Medicine Providers (EMPs) are biased towards life-prolonging care. However, emphasis on early recognition of code status preferences is increasingly making the ED a venue for code status discussions (CSDs). In 2018, our hospital implemented a policy requiring EMPs to place a code status order (CSO) for all patients admitted through the ED. We hypothesized that if EMPs enter CSDs with a bias toward life-prolonging care, or if the venue of the ED biases CSDs towards life-prolonging care, then we would observe a decrease in the percentage of patients selecting DNR status following our institution's aforementioned CSO mandate.
METHODS: We present a retrospective analysis of rates of DNR orders placed for patients admitted through our ED comparing six-month periods before and after the implementation of the above policy.
RESULTS: Using quality improvement data, we identified patients admitted through the ED during pre (n=7,858) and post (n=8,069) study periods. We observed the following: after implementation DNR preference identified prior to hospital admission from the ED increased from 0.4% to 5.3% (relative risk (RR) 12.5; 95% CI: 5.2-29.9), defining CS in the ED setting at the time of admission increased from 2.4% to 98.6% (p <0.001), and DNR orders placed during inpatient admission was unchanged (RR=0.97 (95% CI = 0.88-1.07)).
DISCUSSION: Our results suggest that the ED can be an appropriate venue for CSDs.
Background: Many deaths in Norway occur in medical wards organized to provide curative treatment. Still, medical departments are obliged to meet the needs of patients at the end of life. Here, we analyse the electronic patient record regarding documentation of the transition from curative to palliative care (i.e. the ‘turning point’). Considering the consequences of these decisions for patients, they have received surprisingly little attention from researchers. This study aims to investigate how the patient record denotes reasons for the shift from curative treatment to palliation and how texts involve voices of the patient and their families.
Methods: The study comprised excerpts from electronic patient records retrieved from medical wards in three urban hospitals in Norway. We executed a retrospective analysis of anonymized extracts from 16 electronic patient records, searching for documentation on the transition from curative to palliative care.
Results: In the development of the turning point, the texts usually shift from statements about the patient’s clinical status and technical findings to displaying uncertainty and openness to negotiation with different textual voices. This shift may represent a need to align or harmonize the attitudes of colleagues, family, and patient towards the turning-point decision. The patient’s voice is mostly absent or reported only briefly when, in their notes, nurses gave an account of the patient’s opinion. None of the physicians’ notes provided a detailed account of patient attitudes, wishes, and experiences.
Conclusion: In this article, we have analysed textual representations of patient transitions from curative to end-of-life care. The ‘reality’ behind the text has not been our concern. As the only documentation left, the patient record is an adequate basis for considering how patients are estimated and cared for in their last days of life.
Tout usager du système de santé français, majeur, pas obligatoirement malade, peut demander l’aide de son médecin traitant pour rédiger ses directives anticipées (DA).
Objectives: An increasing number of children are living with complex chronic diseases (CCDs) due to medical advances. Despite a need for code status discussions (CSDs), there is great variability in the frequency and documentation of such conversations. The objective was to identify gaps in the documentation of CSDs within the electronic health record (EHR), focusing on patients with CCDs.
Methods: This was a retrospective review of all patients admitted from the emergency department of a tertiary care children's hospital in 2016. An EHR query using the Apache Hadoop cluster and manual review identified documentation of CSDs, including (1) code status orders, (2) advance directives, and (3) CSDs in provider notes. Patient complexity was stratified using the Pediatric Medical Complexity Algorithm 3.0. Comparative analysis was performed using chi-square, Kruskal–Wallis tests and multivariable logistic regression.
Results: There were 12,648 unique patients of whom 4157 (32.9%) had CCD. Only 209 (1.7%) patients had a code status documented, of whom 200 (95.7%) had CCD. Of 528 (4.2%) patients =18 years of age, 428 (81.1%) had CCD and only 65 (12.3%) had CSDs. Palliative care consultation increased odds of CSDs (OR: 21.4, 95% CI: 13.8–33.2, p < 0.0001), whereas African American race decreased odds of CSDs (OR: 0.42, 95% CI: 0.27–0.64, p < 0.0001).
Conclusions: Among admitted pediatric patients, most do not have documentation of CSDs, including those with CCD and patients =18 years of age. Improvements in both frequency and consistency of CSD documentation are needed to inform the family-centered care of patients living with CCDs.
Access to data on quality metrics can better equip palliative care social workers to identify and address gaps in patient care, establish standards and accountability for social work functions on the interdisciplinary team, and evaluate the impact of interventions. The objective of this demonstration project was to create and pilot a data collection format in the patient electronic medical record (Epic) for documentation of social work metrics at each inpatient consultation, and to build corresponding pilot reports relevant to quality improvement goals. The successful implementation and initial pilot reports were reviewed for the feasibility of longer-term applications.
Opportunities for expanding advance care planning (ACP) throughout the health-care system make it critical that primary care (PC) providers have a basic understanding of how the electronic health record (EHR) can aid promoting ACP discussions and documentation. This article will offer PC providers 5 useful tips for implementing ACP in outpatient settings utilizing the EHR.
BACKGROUND: Advanced care planning through Physician Order For Life-Sustaining Therapies (POLST) has been encouraged by professional societies. But these documents may be overlooked or ignored during hospitalization and "full-code" orders written as a default, putting patients at risk for unwanted resuscitation. After 2 instances of unwanted resuscitation in which limited support POLSTs were ignored, a series of improvements were implemented. This study measured the effectiveness of those steps in reducing POLST code status discrepancy.
METHODS: Pre-post implementation chart review of randomly chosen medical admissions to determine the rate of discordance between POLST orders (when present) and admission code status orders. Physician Order For Life-Sustaining Therapies were classified as either "full" or "limited" based on orders for life-sustaining therapies on the form. Chi-square tests or Fisher exact tests were performed on binary data to identify statistically significant differences at the 95% confidence level between pre- and postimplementation admissions.
RESULTS: In all, 444 preimplementation and 448 postimplementation admissions were evaluated. Discrepant code status orders for those with limited POLST fell from 10 (22.7%) of 44 preimplementation to 3 (4.6%) of 65 after implementation, P = .006. The number of documented code status discussions in admission notes increased from 19.6% to 63.6% (P < .001). The median age of all POLST in the chart was 1.2 years.
CONCLUSIONS: Among those patients with limited POLST orders, discrepant full-code orders increase the potential for unwanted resuscitation. Multistep improvements including documentation templates improved the process of verifying end-of-life wishes and increased meaningful code status discussions. The rate of discrepant orders fell in response to process improvements.
Background: Precision health initiatives for end-of-life planning require robust methods for identifying patient risk for decline and mortality. The Outcome and Assessment Information Set (OASIS) surprise question (SQ; M1034 Overall Status) is the primary tool for evaluating risks in homebound older adults. However, the OASIS-D, Released in 2019, eliminates this question. This study examines the prognostic ability of 12- and 24-month mortality risk reflected in the OASIS-SQ and develops an alternative approach for classifying mortality risk to support decision-making in the absence of the OASIS-SQ.
Design: Retrospective secondary data analysis.
Setting/Participants: A nationally representative sample of 69 097 OASIS-C assessments (2012) linked to the Master Beneficiary Summary file (2012 and 2013).
Measurements: Survival analysis, k-means clustering, and Cohen coefficient with Z test.
Results: The OASIS-SQ predicts mortality (35% at 12 and 45% at 24 months; P < .001). Cluster analysis identified 2 risk groups: OASIS activity of daily living “ADL total scores” >15 = (lower risk) and =15 = (higher risk) for 24-month mortality. Model agreement is weak for both cluster 1 and cluster 2, the OASIS-SQ = 0.20, 95% confidence interval (CI) = .19 to .21, and “alive/not alive” = .17, 95% CI = .16 to .18.
Conclusion: The OASIS-SQ and “ADL total score” are almost equally likely to predict 24-month mortality; therefore, it was reasonable to use the “ADL total score” as a substitute for the OASIS-SQ. Removal of the OASIS-SQ leaves home care providers with few clear options for risk screening resulting in missed opportunities to refer to palliative or hospice services.
BACKGROUND: Electronic care coordination systems, known as the Key Information Summary (KIS) in Scotland, enable the creation of shared electronic records available across healthcare settings. A KIS provides clinicians with essential information to guide decision making for people likely to need emergency or out-of-hours care.
AIM: To estimate the proportion of people with an advanced progressive illness with a KIS by the time of death, to examine when planning information is documented, and suggest improvements for electronic care coordination systems.
DESIGN AND SETTING: This was a mixed-methods study involving 18 diverse general practices in Scotland.
METHOD: Retrospective review of medical records of patients who died in 2017, and semi-structured interviews with healthcare professionals were conducted.
RESULTS: Data on 1304 decedents were collected. Of those with an advanced progressive illness (79%, n = 1034), 69% (n = 712) had a KIS. These were started a median of 45 weeks before death. People with cancer were most likely to have a KIS (80%, n = 288), and those with organ failure least likely (47%, n = 125). Overall, 68% (n = 482) of KIS included resuscitation status and 55% (n = 390) preferred place of care. People with a KIS were more likely to die in the community compared to those without one (61% versus 30%). Most KIS were considered useful/highly useful. Up-to-date free-text information within the KIS was valued highly.
CONCLUSION: In Scotland, most people with an advanced progressive illness have an electronic care coordination record by the time of death. This is an achievement. To improve further, better informal carer information, regular updating, and a focus on generating a KIS for people with organ failure is warranted.
Le déploiement du Système d'Information et du Dossier Patient informatisé suscite craintes et interrogations au sein de l'Hôpital. Le rôle d'accompagnement du cadre de santé est alors essentiel. L'élaboration d'objectifs nécessite une stratégie qui repose sur différents leviers. La phase exploratoire a permis d'établir trois hypothèses répondant à : en quoi une stratégie dans l'accompagnement influence le déploiement du Dossier Patient informatisé ?
Une enquête basée sur une approche qualitative tend à les confirmer ou réfuter. L'étude repose sur l'analyse d'entretiens avec des cadres de santé. A l'issue, il s'avère que le diagnostic est une phase importante pour l'adaptation de l'accompagnement. La responsabilisation, l'implication des professionnels sont de véritables leviers. Cependant la finalité qui favorise l'adhésion n'est pas toujours celle autour de la prise en charge du patient mais plutôt la traçabilité et l'utilisation pratique. Deux autres leviers ont également été éclairés.
[Résumé auteure]
Advance care planning is a process that supports conversations about the values that matter most to patients and their family members. The documentation of advance directives and code status in a patient's electronic health record (EHR) is a critical step to ensure treatment preferences are honored in the medical care received. The current approach to advanced care planning documentation in electronic medical records often remains disparate within and across EHR systems. Without a standardized format for documentation or centralized location for documentation, advance directives and even code status content are often difficult to access within electronic medical records. This case report launched our palliative care team into partnership with the Information Technology team for implementation of a centralized, standardized, longitudinal, functional documentation of advance care planning and code status in the electronic medical record system.
Importance Earlier clinician-patient conversations about patients’ values, goals, and preferences in serious illness (ie, serious illness conversations) are associated with better outcomes but occur inconsistently in cancer care.
Objective To evaluate the efficacy of a communication quality-improvement intervention in improving the occurrence, timing, quality, and accessibility of documented serious illness conversations between oncology clinicians and patients with advanced cancer.
Design, Setting, Participants This cluster randomized clinical trial in outpatient oncology was conducted at the Dana-Farber Cancer Institute and included physicians, advanced-practice clinicians, and patients with cancer who were at high risk of death.
Main Outcomes and Measures The primary outcomes (goal-concordant care and peacefulness at the end of life) are published elsewhere. Secondary outcomes are reported herein, including (1) documentation of at least 1 serious illness conversation before death, (2) timing of the initial conversation before death, (3) quality of conversations, and (4) their accessibility in the electronic medical record (EMR).
Results We enrolled 91 clinicians (48 intervention, 43 control) and 278 patients (134 intervention, 144 control). Of enrolled patients, 58% died during the study (n=161); mean age was 62.3 years (95% CI, 58.9-65.6 years); 55% were women (n=88). These patients were cared for by 76 of the 91 enrolled clinicians (37 intervention, 39 control); years in practice, 11.5 (95% CI, 9.2-13.8); 57% female (n=43). Medical record review after patients’ death demonstrated that a significantly higher proportion of intervention patients had a documented discussion compared with controls (96% vs 79%, P = .005) and intervention conversations occurred a median of 2.4 months earlier (median, 143 days vs 71 days, P < .001). Conversation documentation for intervention patients was significantly more comprehensive and patient centered, with a greater focus on values or goals (89% vs 44%, P < .001), prognosis or illness understanding (91% vs 48%, P < .001), and life-sustaining treatment preferences (63% vs 32%, P = .004). Documentation about end-of-life care planning did not differ between arms (80% intervention vs 68% control, P = .08). Significantly more intervention patients had documentation that was accessible in the EMR (61% vs 11%, P < .001).
Conclusions and Relevance This communication quality-improvement intervention resulted in more, earlier, better, and more accessible serious illness conversations documented in the EMR. To our knowledge, this is the first such study to demonstrate improvement in all 4 of these outcomes.
Objectives: Globally, healthcare systems are using the Electronic Health Record (EHR) and elements of clinical decision support (CDS) to facilitate palliative care (PC). Examination of published results is needed to determine if the EHR is successfully supporting the multidisciplinary nature and complexity of PC by identifying applications, methodology, outcomes, and barriers of active incorporation of the EHR in PC clinical workflow.
Methods: A systematic review using Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The data sources PubMed, CINAL, EBSCOhost, and Academic Search Premier were used to identify literature published 1999 - 2017 of human subject peer-reviewed articles in English containing original research about the EHR and PC.
Results: The search returned 433 articles, 30 of which met inclusion criteria. Most studies were feasibility studies or retrospective cohort analyses; one study incorporated prospective longitudinal mixed methods. Twenty-three of 30 (77%) were published after 2014. The review identified five major areas in which the EHR is used to support PC. Studies focused on CDS to: identify individuals who could benefit from PC; electronic advanced care planning (ACP) documentation; patient-reported outcome measures (PROMs) such as rapid, real-time pain feedback; to augment EHR PC data capture capabilities; ,and to enhance interdisciplinary communication and care.
Discussion: Beginning in 2015, there was a proliferation of articles about PC and EHRs, suggesting increasing incorporation of and research about the EHR with PC. This review indicates the EHR is underutilized for PC CDS, facilitating PROMs, and capturing ACPs.
Timely outreach to individuals in an advanced stage of illness offers opportunities to exercise decision control over health care. Predictive models built using Electronic health record (EHR) data are being explored as a way to anticipate such need with enough lead time for patient engagement. Prior studies have focused on hospitalized patients, who typically have more data available for predicting care needs. It is unclear if prediction driven outreach is feasible in the primary care setting. In this study, we apply predictive modeling to the primary care population of a large, regional health system and systematically examine the impact of technical choices, such as requiring a minimum number of health care encounters (data density requirements) and aggregating diagnosis codes using Clinical Classifications Software (CCS) groupings to reduce dimensionality, on model performance in terms of discrimination and positive predictive value. We assembled a cohort of 349,667 primary care patients between 65 and 90 years of age who sought care from Sutter Health between July 1, 2011 and June 30, 2014, of whom 2.1% died during the study period. EHR data comprising demographics, encounters, orders, and diagnoses for each patient from a 12 month observation window prior to the point when a prediction is made were extracted. L1 regularized logistic regression and gradient boosted tree models were fit to training data and tuned by cross validation. Model performance in predicting one year mortality was assessed using held-out test patients. Our experiments systematically varied three factors: model type, diagnosis coding, and data density requirements. We found substantial, consistent benefit from using gradient boosting vs logistic regression (mean AUROC over all other technical choices of 84.8% vs 80.7% respectively). There was no benefit from aggregation of ICD codes into CCS code groups (mean AUROC over all other technical choices of 82.9% vs 82.6% respectively). Likewise increasing data density requirements did not affect discrimination (mean AUROC over other technical choices ranged from 82.5% to 83%). We also examine model performance as a function of lead time, which is the interval between death and when a prediction was made. In subgroup analysis by lead time, mean AUROC over all other choices ranged from 87.9% for patients who died within 0 to 3 months to 83.6% for those who died 9 to 12 months after prediction time.
The current study explored the perceptions of health care providers' use of electronic advance directive (AD) forms in the electronic health record (EHR). The Technology Acceptance Model (TAM) was used to guide the study. Of 165 surveys distributed, 151 participants (92%) responded. A moderately strong positive correlation was noted between perceived usefulness and actual system usage (r = 0.70, p < 0.0001). Perceived ease of use and actual system usage also had a moderately strong positive correlation (r = 0.70, p < 0.0001). In contrast, the strength of the relationship between behavioral intention to use and actual system usage was more modest (r = 0.22, p < 0.004). There was a statistically significant difference in actual system usage of electronic ADs across six departments ( 2[5] = 79.325, p < 0.001). The relationships among primary TAM constructs found in this research are largely consistent with previous TAM studies, with the exception of behavioral intention to use, which is slightly lower. These data suggest that health care providers' perceptions have great influence on the use of electronic ADs. [
CONTEXT: Opportunities for patients to receive unnecessary, costly and potentially harmful care near the end-of-life abound. Advance care planning (ACP) can help to make this vulnerable period better for patients, caregivers and providers.
OBJECTIVE: The objective of this study was to determine whether older age predicted the presence of certain forms of retrievable ACP documentation in the Electronic Health Record (EHR) in a large sample of hospice-referred patients.
METHODS: This was a retrospective analysis of medical-record data on 3,595 patients referred to hospice between January 1st, 2013 and December 31, 2015. EHR documentation of an ACP note in the problem list, presence of a scanned Advance Directive (AD), and the presence of a verified Do Not Resuscitate order (DNR) were the outcome measures. Logistic regression was used to assess the effect of age, education, race, gender, cancer diagnosis, dementia diagnosis, palliative encounter and death on the outcome variables.
RESULTS: Our results suggest that when we control for prognosis, patients over age 70 may experience gaps in ACP communication. We found that as patients age, the odds of having documentation of a conversation (OR=0.56; P <0.001) or scanned AD decreased (OR=0.63; P<0.001), while the odds of having a verified DNR increased (OR=1.42; P<0.001).
CONCLUSION: The results of this study may imply some degree of unilateral and physician-driven decision making for end-of-life care among older adults. Collaborative efforts between an interdisciplinary medical team should focus on developing policies to address this potential disparity between younger and older adults at the end-of-life.
BACKGROUND: Access to palliative care is a key quality metric which most healthcare organizations strive to improve. The primary challenges to increasing palliative care access are a combination of physicians over-estimating patient prognoses, and a shortage of palliative staff in general. This, in combination with treatment inertia can result in a mismatch between patient wishes, and their actual care towards the end of life.
METHODS: In this work, we address this problem, with Institutional Review Board approval, using machine learning and Electronic Health Record (EHR) data of patients. We train a Deep Neural Network model on the EHR data of patients from previous years, to predict mortality of patients within the next 3-12 month period. This prediction is used as a proxy decision for identifying patients who could benefit from palliative care.
RESULTS: The EHR data of all admitted patients are evaluated every night by this algorithm, and the palliative care team is automatically notified of the list of patients with a positive prediction. In addition, we present a novel technique for decision interpretation, using which we provide explanations for the model's predictions.
CONCLUSION: The automatic screening and notification saves the palliative care team the burden of time consuming chart reviews of all patients, and allows them to take a proactive approach in reaching out to such patients rather then relying on referrals from the treating physicians.