Clinician-Identified Health Characteristics and Palliative Care Eligibility: Is Dementia Overlooked?

ABSTRACT

Objectives: Community-based palliative care provides support for community-dwelling individuals with elevated mortality risk, including those with dementia, who are underserved by palliative care. However, clinicians’ eligibility assessment processes are not well understood. This study evaluates the relationship between the factors that clinicians indicate are important and the eligibility determinations for a community-based palliative care program.

Study Design: Retrospective cohort analysis of July 2022 to December 2023 Medicare administrative claims data for a Medicare Advantage plan offering community-based palliative care. Participants included all members of the Medicare Advantage insurance plan who were identified as being at elevated risk for mortality and evaluated for palliative care need.

Methods: Multivariate logistic regression examined the relationship between eligibility determinations for a community-based palliative care program (outcome) and 4 factors that palliative care team members identify as important for determining palliative care need: diagnoses, symptom management, functional ability, and health care utilization.

Results: Of 343 palliative care evaluations (for 322 unique patients), 38% were of patients who identified as African American/Black, 38% as White, 9% as Asian/Pacific Islander, and 15% as other races; 41% were of patients who identified as Hispanic/Latino (measured separately from race); 80% were of female patients; and the mean patient age was 88 years. Of these, 169 (49%) evaluations were for patients who were eligible for palliative care. In multivariate logistic regression analysis of the factors team members deemed important when determining eligibility, only dementia was significantly associated—and negatively—with the final eligibility decision (adjusted OR, 0.45; 95% CI, 0.26-0.75; P = .003).

Conclusions: There is a potential mismatch between what clinicians identify as important in determining palliative care need and final eligibility determinations. Patients with dementia were less likely to be referred for palliative care despite elevated risk of mortality, indicating a potential missed opportunity.

Am J Manag Care. 2025;31(11):In Press

Takeaway Points

Although community-based palliative care can support seriously ill individuals, including those with dementia, to remain safely in the community, clinicians’ processes for determining eligibility for this service are poorly understood. We examine the relationship between factors that clinicians say are important in determining eligibility for community-based palliative care and final eligibility determinations.

• Among diagnoses, symptom burden, functional ability, and recent health care utilization, only dementia was significantly associated with final eligibility.
• The negative association between dementia diagnoses and palliative care eligibility determinations signals a potential missed opportunity to provide these patients support afforded through palliative care.

Palliative care, including outpatient or community-based palliative care (CBPC), is an evidence-based, interdisciplinary care delivery model that can support seriously ill, community-dwelling older adults in maintaining quality of life and remaining safely at home.^1-7 Nearly 2 million people in the US are homebound, and another 5.5 million require assistance to leave home.⁸ Between 60% and 80% of US adults 65 years and older have 2 or more chronic conditions, contributing to the need for CBPC.^9-12 CBPC improves outcomes for seriously ill individuals and their caregivers by reducing symptom burden, preventing hospitalization, and improving quality of life.^5,7,13-17 Palliative care is consistent with goals of person-centered care for patients with dementia,¹⁸ but underenrollment and lack of interventions to facilitate connections to CBPC for these individuals are indications of a critical quality issue.^19-22

However, it can be challenging to identify seriously ill individuals with high symptom burden and low levels of functional ability who may benefit from CBPC from among the larger pool who might just need specialized primary care, such as that provided by geriatricians.²³ Machine learning 12-month mortality prediction algorithms based on administrative claims data can be a useful clinical decision support tool for identifying individuals for targeted care in community-based settings, such as CBPC.^24-26 These algorithms use thousands of variables in claims data related to diagnoses, emergency department (ED) use and hospitalization, prescriptions, and durable medical equipment (DME) use.

Ultimately, clinicians make final determinations about whether individuals meet criteria for specific CBPC programs. Clinicians may utilize information about an individual’s health status that is not easily captured in claims data–based algorithms, including information from phone screenings, patient self-reports, and electronic health records, to determine CBPC eligibility.^27,28 Claims data provide information about individuals’ chronic conditions but may not contain details about disease stage or severity. Patients’ symptom burden and functional status are key to helping clinicians distinguish between older adults with well-managed chronic conditions and seriously ill, low-functioning individuals with high symptom burden who may benefit from CBPC.²⁹ However, symptom burden and functional status are not obviously evident in all administrative claims data. As such, proxy measures for each must be used. International Statistical Classification of Diseases, Tenth Revision (ICD-10) codes for pain, dyspnea, and other symptom management may signal greater symptom burden. Functional status is routinely captured for home health care recipients and nursing home residents^30,31 but not for older adults who do not receive these services. Evidence of diminished functional status in claims data has been captured by assessing in-home support such as home health aides and DME for respiratory and enteral support.³² Finally, patients who have recently visited the ED or been hospitalized and discharged to the community may require additional support as they recover from acute health episodes.

In this study, CBPC team members identified factors they say weigh heavily in assessments of CBPC need within a population of older adults who are part of a Medicare Advantage (MA) plan, including (1) diagnoses for 7 chronic conditions that, in advanced stages, are indicative of palliative care need; (2) symptom management as a proxy for symptom burden; (3) functional ability using measures developed by data scientists at the MA plan; and (4) health care utilization. This pilot study assessed whether these factors are associated with actual CBPC eligibility determinations above and beyond what is controlled for in an existing mortality predictive algorithm used by the MA plan to identify individuals who may benefit from CBPC but require further assessment to determine final eligibility. We also compared the area under the receiver operating characteristic curve (AUROC) for the clinician-informed predictive model by diagnosis category. Understanding the relationship between factors that clinicians say are important and how those relate to eligibility determinations is an important antecedent step in understanding how to eventually connect individuals with CBPC.

METHODS

Data and Study Sample

This study analyzed Medicare administrative claims data from an MA plan with more than 4000 members run by a nonprofit health plan and care management organization serving a diverse population in New York, including New York City and upstate. We used a case-control design for the study. All MA plan members are evaluated monthly using a validated predictive algorithm developed by the data science team at the MA plan. The algorithm, which is explained in greater detail later, flags individuals at risk of 12-month mortality who also experienced a triggering event associated with a palliative care need. We limited our analysis to individuals who were flagged by the 12-month mortality algorithm and evaluated by palliative care team members for eligibility for a CBPC program that is part of the MA plan. We excluded individuals who were determined to be ineligible for the CBPC program prior to evaluation due to exclusion criteria such as receiving hospice or residing in a long-term care facility or skilled nursing facility. We provide descriptive statistics of 343 CBPC evaluations for 322 unique individuals—all of whom were enrolled in the MA plan from July 2022 to December 2023 and received a CBPC evaluation. Nineteen individuals had more than 1 palliative care evaluation during this period. All study activities were determined to be exempt from institutional review board (IRB) review under federal regulations category 2(ii) by the VNS Health IRB.

Outcomes

CBPC eligibility. Eligibility for CBPC for each MA plan member who was evaluated for the service was measured as a binary variable (eligible = 1; ineligible = 0).

Twelve-month mortality. To establish how well the current 12-month mortality algorithm performs among different patient populations, we calculated the AUROC for multiple disease categories.

Key Predictors

In a study by Sheldon et al, 2 focus groups with 7 palliative care team members involved in determining CBPC eligibility for MA plan members (including outreach/care managers and nurse practitioners) noted that it is helpful to know an individual’s diagnoses as part of the evaluation process.³³ Additionally, symptom burden and functional ability were identified as key indicators in making palliative care eligibility determinations.³³

Functional ability measures. The MA plan’s data science team identified Current Procedural Terminology codes related to functional ability, which were then conceptually grouped into 5 categories, measured as binary variables (yes = 1; no = 0). Activities of daily living (ADL) support refers to nonskilled home health, personal care, or homemaker services. DME: feeding support refers to artificial hydration, feeding tubes, enteral supplies, or nutritional deficiency. DME: respiratory support refers to oxygen or ventilator use. PT/OT services refers to physical or occupational services. Skilled nursing refers to skilled home health services.

Symptom management. In multivariate models, we controlled separately for 3 binary measures of symptom management (yes = 1; no = 0). Symptom burden was identified using ICD-10 codes for symptom management for nausea (R11.0, R11.2) and vomiting (R11.10) (which we combined into a single variable), pain (G89.0, R52), and dyspnea (R06.00).

Diagnoses. Diagnoses of cancer (C00.0-C96.9, D00.0-D48.9), end-stage kidney disease (ESKD; N18.4-N18.6), chronic obstructive pulmonary disease (COPD; J44.0-J44.9), cardiovascular disease (congestive heart failure, end-stage heart failure, coronary artery disease) (I20.0-I25.9, I50.1-I50.9), stroke (I60.00-I63.9), dementia (F01-F03.91), and anemia (D50.0-D64.9) were identified using ICD-10 diagnostic codes (provided in parentheses) and measured as binary variables (yes = 1; no = 0).

Health care utilization. ED visits and hospitalizations were identified based on New York’s Medicaid Managed Care Operational Reports requirements³⁴ and measured as binary variables (yes = 1; no = 0).

Covariates

Sociodemographic factors. We controlled for beneficiary demographic factors commonly associated with disparities in health outcomes, including age (in years), sex (male, female [reference]), race (Asian, Black or African American [hereafter, Black], other, White [reference]), and Hispanic ethnicity. Younger age (<65 years), being male, and identifying as Black or Hispanic are all associated with less access to palliative care and poorer health outcomes among seriously ill individuals.³⁵

Current mortality prediction model. We also controlled for probability of 12-month mortality as a means of controlling for the performance of the existing machine learning algorithm. The model was trained using the XGBoost³⁶ algorithm on 3500 candidate, time-windowed features capturing recency and intensity of health care utilization across broad categories of service (eg, inpatient, pharmacy, etc), ICD-10 classifications, and specific clinical items preidentified by CBPC team members, also derived from claims data. Hyperparameters were optimized to explore the complex feature space using Optuna.³⁷ The AUROC for the current model was 0.84 (95% CI, 0.83-0.85).

Analysis

To assess CBPC eligibility determinations, for ease of interpretation, we performed a bivariate analysis using χ² or Fisher exact tests for categorical variables and t tests for age to compare evaluations of individuals who were determined to be eligible and those who were not from July 2022 to December 2023. We used a 12-month look-back period to identify diagnoses, symptom burden, functional ability, and health care utilization. Because the sample size is relatively small, we first performed a series of multivariate logistic regression analyses using CBPC evaluation from July 2022 to December 2023 as the unit of analysis to determine the extent to which diagnoses (model 1), symptom management (model 2), functional ability measures identified by clinicians and codified by the data science team (model 3), and health care utilization (model 4) are associated with final CBPC eligibility determinations. All models controlled for sociodemographic factors and probability of 12-month mortality as predicted by the current mortality algorithm. We then conducted a final multivariable logistic regression model controlling for sociodemographic characteristics, existing mortality algorithm performance, and factors that were statistically significant from models 1 to 4. Finally, to assess the 12-month mortality model’s performance by patient population, we calculated and compared AUROC for the current predictive model by diagnosis category.

RESULTS

Descriptive Statistics and Bivariate Analysis

Of the 343 CBPC evaluations for 322 unique MA plan members during the study period, 38% were for patients who identified as Black, 38% as White, 9% as Asian/Pacific Islander, and 15% as other races (American Indian/Alaska Native, more than 1 race, unknown race); 41% were for patients who identified as Hispanic/Latino (measured separately from race); 80% were for female patients; and the mean (SD) patient age was 88.0 (10.5) years. Of these, 169 (49%) evaluations were for patients determined to be eligible for CBPC and 174 (51%) evaluations were for patients deemed ineligible.

In bivariate analyses, individuals who were determined to be eligible for CBPC were younger than expected (P < .001) but did not differ on other sociodemographic characteristics. Individuals with cancer (P = .012), cardiovascular disease (P = .001), COPD (P = .011), and ESKD (P = .016) were determined eligible for CBPC more often than expected, whereas those with dementia were determined ineligible more often than expected (P < .001). Individuals receiving assistance for pain (P < .001) and dyspnea (P < .001) management were eligible more often than expected. Those with skilled nursing were eligible more often than expected (P = .002), and individuals who experienced an ED visit (P = .001) or hospitalization (P < .001) were eligible more often than expected (Table 1).

Multivariate Logistic Regression

Table 2 and the Figure present the results of 4 multivariate logistic regression models. These models controlled for sociodemographic factors and 12-month mortality risk (as determined by the current algorithm) and examined the association between palliative care eligibility determinations and factors that clinicians indicate are important in making these determinations: diagnoses (model 1), symptom management (model 2), functional ability (model 3), and health care utilization (model 4). In model 1, dementia was associated with 54% lower odds of being eligible for palliative care (adjusted OR [AOR], 0.46; 95% CI, 0.26-0.79; P = .005). In model 2, assistance with dyspnea and pain management were both associated with 97% higher odds of being determined eligible for palliative care (dyspnea: AOR, 1.97; 95% CI, 1.21-3.21; P = .007; pain: AOR, 1.97; 95% CI, 1.17-3.35; P = .011). In model 3, skilled nursing, the only of 5 functional ability measures linked with CBPC eligibility, was associated with 65% higher odds of CBPC eligibility (AOR, 1.65; 95% CI, 1.03-2.66; P = .036). In model 4, both ED visits and hospitalization were associated with higher odds of CBPC eligibility (ED visit: AOR, 1.78; 95% CI, 1.10-2.88; P = .018; hospitalization: AOR, 1.82; 95% CI, 1.10-3.01; P = .019). In the multivariable logistic regression model containing all statistically significant variables from models 1 to 4, only dementia was significantly (and negatively) associated with CBPC eligibility—with 54% lower odds of being eligible for CBPC (AOR, 0.46; 95% CI, 0.27-0.77; P = .004) (Table 3). No sociodemographic factors, symptom management, functional ability, or health care utilization variables were associated with eligibility determinations. The current 12-month mortality predictive algorithm was not significant in any regression model. Akaike information criterion and Bayesian information criterion were lowest for our final model, indicating best model fit. The Hosmer-Lemeshow test, which assesses model fit for logistic regression and where a large P value indicates good fit, was not significant.

Model Performance by Disease

Table 4 provides the results of sensitivity analyses comparing performance (AUROC) of the currently deployed machine learning algorithm by disease category. Although there was some variation in how well the model performed by disease category and performance measure, overall, the mortality algorithm did a good job of identifying individuals who were eligible for CBPC. AUROC, a commonly used measure of model fit, ranged from 0.74 (95% CI, 0.71-0.77) among patients with ESKD to 0.83 among patients with cancer (95% CI, 0.81-0.85).

DISCUSSION

In a diverse sample of MA plan members identified as being at high risk for 12-month mortality and subsequently evaluated for CBPC eligibility over a 1-year period, relatively equal proportions of the 343 evaluations were for patients deemed by clinicians to be eligible (49%) and ineligible (51%) for the services. We novelly explore the association between factors that clinicians indicate are important in making CBPC eligibility determinations and whether an individual was deemed eligible for CBPC above and beyond probability of mortality as determined by a 12-month mortality prediction algorithm currently used by an MA plan serving a diverse population. Of factors that clinicians indicated are important in making CBPC eligibility determinations—diagnoses, symptom management, functional ability, and health care utilization—only dementia was associated (negatively) with whether an individual was determined to be eligible for palliative care.

This finding that dementia was associated with being ineligible for CBPC as determined by practicing clinicians is counterintuitive, given the high AUROC for individuals with dementia identified by the mortality prediction model and previous work indicating machine learning models can accurately predict mortality in persons with dementia.³⁸ However, it is consistent with prior work indicating that individuals with dementia may underutilize CBPC.³⁹ This may signal a missed opportunity to provide support to these individuals—whose symptom burden may be underreported⁴⁰ and who are at increased risk for mortality and candidates for palliative care⁴¹—and to their caregivers. CBPC eligibility processes should attend to the needs of persons with dementia and their families by carefully considering them as potential beneficiaries of CBPC.

Symptom management, functional ability, and health care utilization were not independently associated with clinicians’ final CBPC eligibility determinations, despite clinicians identifying these factors as important. However, the P values for pain management and ED visits do suggest a positive association with CBPC eligibility. More research is needed to determine what sample size would be required for a fully powered analysis showing significant effect sizes, and these factors should be considered potentially important in future work. This finding signals a potential disconnect between what clinicians say is important and how those factors influence actual eligibility decisions. Additional research is needed to better understand the relationship between clinician perceptions of which factors are important in decision-making and actual decisions. Our findings also underscore the importance of clinicians having access to information that may not be evident in claims data, such as individuals’ symptom burden and functional limitations, when making CBPC eligibility determinations.

Among sociodemographic factors, age was negatively associated with clinicians’ CBPC eligibility determinations in all models except the final model—a counterintuitive finding given that older adults are at increased risk for mortality and tend to have higher disease and symptom burdens and lower functional ability. However, it is consistent with prior literature finding that older patients may be overlooked in accessing primary and specialty palliative care,^42,43 particularly if they experience frailty but not a clear terminal diagnosis.⁴³ It is possible that the algorithm, which uses 12-month mortality as a proxy for CBPC eligibility that requires further clinical evaluation, identified individuals at risk for mortality on the basis of older age. Upon further clinical evaluation, these individuals may have been determined to be ineligible for CBPC because they were not experiencing symptoms or other precipitating events.

Overall, the 12-month mortality prediction model was not associated with eligibility determinations, suggesting that although it may be a useful referral mechanism for identifying individuals to evaluate for CBPC, it is not used in actual eligibility determinations.

Limitations

This study has some limitations. It was based on eligibility criteria and determinations at a single MA plan and may not be generalizable to other MA plans or CBPC programs. Although the MA plan we used to conduct the analysis is small (4000 members) relative to large MA plans, we analyzed data for the population of individuals enrolled in the plan who were evaluated for CBPC eligibility during the study period, providing a starting point for considering how mortality prediction models are implemented into clinical decision-making. Additional analysis in larger MA plans would further contextualize our results. The sample we used is also small relative to the number of factors important in clinical decision-making that we would like to explore. Although our analytic approach attempted to compensate for the sample size, it may have been underpowered to detect statistically significant differences in factors such as symptoms including pain and dyspnea, which are considered key factors in identifying palliative care need.⁴⁴ Additionally, symptom burden and functional ability levels are not directly measured in administrative claims data, and our process for codifying both concepts may not accurately identify individuals with high symptom burden and low levels of functional ability.

CONCLUSIONS

A growing proportion of community-dwelling older adults live with serious and life-limiting illnesses and could benefit from ongoing services such as CBPC to support them and their family caregivers. Individuals with dementia—who tend to have extended illness trajectories and require extensive caregiving support—may be at elevated risk for exclusion from CBPC services that may otherwise enable them to remain safely in the community. This is an important population to address as the proportion of older adults with dementia continues to grow. Additional information is needed to better understand the mechanisms that clinicians use in making CBPC eligibility determinations, particularly when there is a mismatch between what they indicate are important factors in decision-making and how those factors relate to actual decisions.

Author Affiliations: Rutgers University Institute for Health, Health Care Policy, and Aging Research (EAL), New Brunswick, NJ; VNS Health (CB, KB), New York, NY; University of Pennsylvania School of Nursing (KB), Philadelphia, PA.

Source of Funding: The study was supported by a pilot grant from the National Center for Advancing Translational Sciences, a component of the National Institutes of Health, under award number UL1TR003017.

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (EAL, CB, KB); acquisition of data (CB); analysis and interpretation of data (EAL, CB, KB); drafting of the manuscript (EAL, CB, KB); critical revision of the manuscript for important intellectual content (EAL, CB, KB); statistical analysis (CB); and obtaining funding (EAL).

Address Correspondence to: Elizabeth A. Luth, PhD, Rutgers University, 303 George St, New Brunswick, NJ 08901. Email: eluth@rwjms.rutgers.edu.

REFERENCES

1. Rabow MW, Dibble SL, Pantilat SZ, McPhee SJ. The comprehensive care team: a controlled trial of outpatient palliative medicine consultation. Arch Intern Med. 2004;164(1):83-91. doi:10.1001/archinte.164.1.83

2. Follwell M, Burman D, Le LW, et al. Phase II study of an outpatient palliative care intervention in patients with metastatic cancer. J Clin Oncol. 2009;27(2):206-213. doi:10.1200/JCO.2008.17.7568

3. Higginson IJ, Evans CJ. What is the evidence that palliative care teams improve outcomes for cancer patients and their families? Cancer J. 2010;16(5):423-435. doi:10.1097/PPO.0b013e3181f684e5

4. O’Connor NR, Junker P, Appel SM, Stetson RL, Rohrbach J, Meghani SH. Palliative care consultation for goals of care and future acute care costs: a propensity-matched study. Am J Hosp Palliat Care. 2018;35(7):966-971. doi:10.1177/1049909117743475

5. Vernon E, Hughes MC, Kowalczyk M. Measuring effectiveness in community-based palliative care programs: a systematic review. Soc Sci Med. 2022;296:114731. doi:10.1016/j.socscimed.2022.114731

6. Marshall D, Howell D, Brazil K, Howard M, Taniguchi A. Enhancing family physician capacity to deliver quality palliative home care: an end-of-life, shared-care model. Can Fam Physician. 2008;54(12):1703-1703.e7.

7. Kavalieratos D, Corbelli J, Zhang D, et al. Association between palliative care and patient and caregiver outcomes: a systematic review and meta-analysis. JAMA. 2016;316(20):2104-2114. doi:10.1001/jama.2016.16840

8. Ornstein KA, Leff B, Covinsky KE, et al. Epidemiology of the homebound population in the United States. JAMA Intern Med. 2015;175(7):1180-1186. doi:10.1001/jamainternmed.2015.1849

9. Administration of Community Living. 2021 Profile of Older Americans. HHS; November 2022. Accessed December 29, 2023. https://acl.gov/sites/default/files/Profile%20of%20OA/2021%20Profile%20of%20OA/2021ProfileOlderAmericans_508.pdf

10. Boersma P, Black LI, Ward BW. Prevalence of multiple chronic conditions among US adults, 2018. Prev Chronic Dis. 2020;17:E106. doi:10.5888/pcd17.200130

11. Buttorff C, Ruder T, Bauman M. Multiple Chronic Conditions in the United States. RAND Corporation; 2017. doi:10.7249/TL221

12. Salive ME. Multimorbidity in older adults. Epidemiol Rev. 2013;35:75-83. doi:10.1093/epirev/mxs009

13. Rabow M, Kvale E, Barbour L, et al. Moving upstream: a review of the evidence of the impact of outpatient palliative care. J Palliat Med. 2013;16(12):1540-1549. doi:10.1089/jpm.2013.0153

14. Kamal AH, Currow DC, Ritchie CS, Bull J, Abernethy AP. Community-based palliative care: the natural evolution for palliative care delivery in the U.S. J Pain Symptom Manage. 2013;46(2):254-264. doi:10.1016/j.jpainsymman.2012.07.018

15. Peeler A, Doran A, Winter-Dean L, et al. Public health palliative care interventions that enable communities to support people who are dying and their carers: a scoping review of studies that assess person-centered outcomes. Front Public Health. 2023;11:1180571. doi:10.3389/fpubh.2023.1180571

16. Brumley R, Enguidanos S, Jamison P, et al. Increased satisfaction with care and lower costs: results of a randomized trial of in-home palliative care. J Am Geriatr Soc. 2007;55(7):993-1000. doi:10.1111/j.1532-5415.2007.01234.x

17. Brumley RD, Enguidanos S, Cherin DA. Effectiveness of a home-based palliative care program for end-of-life. J Palliat Med. 2003;6(5):715-724. doi:10.1089/109662103322515220

18. Hughes JC, Jolley D, Jordan A, Sampson EL. Palliative care in dementia: issues and evidence. Adv Psychiatr Treat. 2007;13(4):251-260. doi:10.1192/apt.bp.106.003442

19. Sampson EL, Ritchie CW, Lai R, Raven PW, Blanchard MR. A systematic review of the scientific evidence for the efficacy of a palliative care approach in advanced dementia. Int Psychogeriatr. 2005;17(1):31-40. doi:10.1017/s1041610205001018

20. Murphy E, Froggatt K, Connolly S, et al. Palliative care interventions in advanced dementia. Cochrane Database Syst Rev. 2016;12(12):CD011513. doi:10.1002/14651858.CD011513.pub2

21. Hashimie J, Schultz SK, Stewart JT. Palliative care for dementia: 2020 update. Clin Geriatr Med. 2020;36(2):329-339. doi:10.1016/j.cger.2019.11.011

22. Erel M, Marcus EL, Dekeyser-Ganz F. Barriers to palliative care for advanced dementia: a scoping review. Ann Palliat Med. 2017;6(4):365-379. doi:10.21037/apm.2017.06.13

23. Kelley AS, Bollens-Lund E. Identifying the population with serious illness: the “denominator” challenge. J Palliat Med. 2018;21(S2):S7-S16. doi:10.1089/jpm.2017.0548

24. Yourman LC, Lee SJ, Schonberg MA, Widera EW, Smith AK. Prognostic indices for older adults: a systematic review. JAMA. 2012;307(2):182-192. doi:10.1001/jama.2011.1966

25. Avati A, Jung K, Harman S, Downing L, Ng A, Shah NH. Improving palliative care with deep learning. BMC Med Inform Decis Mak. 2018;18(suppl 4):122. doi:10.1186/s12911-018-0677-8

26. Dowding D, Russell D, McDonald MV, et al. “A catalyst for action”: factors for implementing clinical risk prediction models of infection in home care settings. J Am Med Inform Assoc. 2021;28(2):334-341. doi:10.1093/jamia/ocaa267

27. Doebbeling BN, Chou AF, Tierney WM. Priorities and strategies for the implementation of integrated informatics and communications technology to improve evidence-based practice. J Gen Intern Med. 2006;21(suppl 2):S50-S57. doi:10.1007/s11606-006-0275-9

28. Doebbeling B, Saleem J, Haggstrom D, et al. Integrating Clinical Decision Support Into Workflow—Final Report. Agency for Healthcare Research and Quality; September 2011. Accessed June 17, 2022. https://digital.ahrq.gov/sites/default/files/docs/citation/IntegratingCDSIntoWorkflow.pdf

29. Kelley AS, Covinsky KE, Gorges RJ, et al. Identifying older adults with serious illness: a critical step toward improving the value of health care. Health Serv Res. 2017;52(1):113-131. doi:10.1111/1475-6773.12479

30. Outcome and Assessment Information Set: OASIS-E Manual. CMS; January 1, 2023. Accessed February 21, 2024. https://www.cms.gov/files/document/oasis-e-manual-final.pdf

31. Home Health Outcome and Assessment Information Set (OASIS). CMS. Accessed July 23, 2024. https://resdac.org/cms-data/files/oasis

32. Iezzoni LI, Wint AJ, Tishler L, Palmisano J, Tripodis Y. Using claims for long-term services and support to predict mortality and hospital use. Disabil Health J. 2019;12(3):523-527. doi:10.1016/j.dhjo.2019.03.002

33. Sheldon KG, Bowles KH, Sage C, Stanley J, Luth EA. Making community-based palliative care eligibility determinations: palliative care team member perspectives on access to information and algorithm use. Am J Hosp Palliat Care. 2025;42(11):1187-1195. doi:10.1177/10499091241302044

34. Managed care reports. New York State Department of Health. Accessed July 23, 2024. https://www.health.ny.gov/health_care/managed_care/reports/

35. Nelson KE, Wright R, Peeler A, Brockie T, Davidson PM. Sociodemographic disparities in access to hospice and palliative care: an integrative review. Am J Hosp Palliat Care. 2021;38(11):1378-1390. doi:10.1177/1049909120985419

36. Chen T, Guestrin C. XGBoost: a scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery; 2016:785-794. doi:10.1145/2939672.2939785

37. Akiba T, Sano S, Yanase T, Ohta T, Koyama M. Optuna: a next-generation hyperparameter optimization framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. Association for Computing Machinery; 2019:2623-2631. doi:10.1145/3292500.3330701

38. Wang L, Sha L, Lakin JR, et al. Development and validation of a deep learning algorithm for mortality prediction in selecting patients with dementia for earlier palliative care interventions. JAMA Netw Open. 2019;2(7):e196972. doi:10.1001/jamanetworkopen.2019.6972

39. Rosenwax L, Spilsbury K, Arendts G, McNamara B, Semmens J. Community-based palliative care is associated with reduced emergency department use by people with dementia in their last year of life: a retrospective cohort study. Palliat Med. 2015;29(8):727-736. doi:10.1177/0269216315576309

40. Harrison KL, Bull JH, Garrett SB, et al. Community-based palliative care consultations: comparing dementia to nondementia serious illnesses. J Palliat Med. 2020;23(8):1021-1029. doi:10.1089/jpm.2019.0250

41. van der Steen JT, Radbruch L, Hertogh CM, et al; European Association for Palliative Care (EAPC). White paper defining optimal palliative care in older people with dementia: a Delphi study and recommendations from the European Association for Palliative Care. Palliat Med. 2014;28(3):197-209. doi:10.1177/0269216313493685

42. Rostoft S, Thomas MJ, Slaaen M, Møller B, Syse A. The effect of age on specialized palliative care use in the last year of life for patients who die of cancer: a nationwide study from Norway. J Geriatr Oncol. 2022;13(8):1103-1110. doi:10.1016/j.jgo.2022.08.002

43. Lloyd A, Kendall M, Carduff E, Cavers D, Kimbell B, Murray SA. Why do older people get less palliative care than younger people? Eur J Palliat Care. 2016;23(3):132-137.

44. Weissman DE, Meier DE. Identifying patients in need of a palliative care assessment in the hospital setting: a consensus report from the Center to Advance Palliative Care. J Palliat Med. 2011;14(1):17-23. doi:10.1089/jpm.2010.0347