Association of Advance Care Planning Visits With Intensity of Health Care for Medicare Beneficiaries With Serious Illness at the End of Life

Key Points Question What is the association of a billed advance care planning (ACP) visit with intensive use of health care services at the end of life (EOL) for Medicare beneficiaries with serious illness? Findings In this cohort study of claims data of 955 777 Medicare beneficiaries with serious illness who died in 2017 and 2018, billed ACP visits that occurred during the decedents’ EOL course but before the last month of life were relatively uncommon. However, their occurrence was associated with less intensive use of EOL health care services. Meaning The findings of this cohort study suggest that ACP is associated with less intensive use of EOL health care services.


Introduction
Up to 76% of patients will be unable to participate in decisions about their care at the end of life (EOL). 1,2Advance care planning (ACP) is a process that enables the patient and health care system to better align preferences, values, and goals.Advance care planning has been shown to reduce the frequency with which patients die in the hospital 3 and are admitted to a hospital or an intensive care unit (ICU) in the last 30 days of life. 4,5Patients who have engaged in ACP are more likely to enroll in hospice care in a timely fashion, 3,4 avoiding crisis decisions in the last days before death.
A common ACP product may include documentation of EOL treatment preferences, including advance directives, physician orders for life-sustaining treatment, or assignment of a health care proxy.Advance care planning is intended to aid persons across their life course and may involve multiple points of engagement with clinicians. 6,7Advance care planning may be particularly helpful for persons who have a serious illness, who face a heightened risk of death, or whose condition adversely affects their daily quality of life. 8,9Although we know that persons living with serious illnesses are more likely than other older patients to have ACP visits, 10 less is known about how ACP affects their EOL care except among studies of small or local samples 11,12 and a limited sample of commercially insured patients. 13til recently, the lack of data on ACP visits in Medicare administrative claims limited the ability to examine these phenomena on a population level.In 2016, the Centers for Medicare & Medicaid Services initiated the use of time-based Current Procedural Terminology codes (99497 and 99498) to compensate clinicians for having ACP conversations with their patients. 14Uptake in the use of these billing codes among Medicare clinicians has been low but is gradually increasing. 15Within this context, the goal of this study was to test the hypothesis that having a billed ACP visit for seriously ill patients was associated with less intensive use of health care at EOL.

Data Source and Study Sample
We used claims from January 1 to December 31, 2016, for all continuously enrolled fee-for-service Medicare beneficiaries, excluding patients with any managed care coverage.Seriously ill patients who were expected to have a median survival of less than 2 years or significant limitations as a result of their disease were prospectively identified using an algorithm that has been used in previous studies and clinical trials. 10,17We included patients who had claims that contained the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnosis codes of chronic obstructive pulmonary disease and other lung diseases, heart failure, kidney failure, cancer, dementia, and neurodegenerative diseases who also had at least 2 inpatient admissions in the prior in the cohort is available from the authors.Next, we identified beneficiaries 65 years or older who died in 2017 or 2018.Claims for these individual patients were linked with the Medicare beneficiary summary file and inpatient, outpatient, skilled nursing facility, hospice, and carrier files using the beneficiary identification code.

Use of EOL Health Services
From the Medicare records, we obtained 6 evidence-based measures of use of EOL health care services or intensity 3,[19][20][21][22] : (1) in-hospital death, for which hospital discharge status was death; (2)   hospital admission, defined as at least 1 inpatient hospital admission within the last 30 days of life; (3) ICU admission, defined as at least 1 admission to the ICU within the last 30 days of life; (4) emergency department (ED) visit, defined as at least 1 ED visit within the last 30 days of life; (5) hospice timing, categorized as no hospice, hospice care initiated at least 4 days before death, or hospice care initiated fewer than 4 days before death; and (6) expenditures, calculated as total Medicare costs for all inpatient, outpatient, skilled nursing facility, and hospice care in the last 30 days of life.

ACP Billed Visit
Our primary exposure variable was receipt of a billed ACP visit.Beneficiaries having 1 or more billed visits from January 1, 2016, until death using either the primary code 99497 (first 30 minutes) or the secondary code 99498 (extended time beyond 30 minutes) or both were counted once for the analysis.Beneficiaries with their first ACP visit within 30 days of death were classified separately given that the period of exposure overlapped with the measurement period for EOL outcomes.Thus, we classified decedents as having no ACP (if neither the primary nor the secondary billing code was used in any visit from 2016-2018), timely ACP (if Ն1 ACP claim appeared in any visit from 2016-2018 until 30 days before the death), or late ACP (if the first ACP claim appeared within 30 days before death).

Covariates
Covariates were drawn from potential confounders used in previous EOL outcome studies [23][24][25] and included sex (male or female), age (65-69, 70-74, 75-79, 80-84, 85-89, 90-94, and Ն95 years), Charlson Comorbidity Index (CCI) using a 6-month look-back period from date of death (0, 1, 2, 3, 4, or Ն5), and dual eligibility status for Medicare and Medicaid as a proxy for socioeconomic status.We included race and ethnicity as a covariate because of its association with use of EOL health care services 26,27 and used the categories as listed in the Master Beneficiary Summary File.Hospital referral regions (HRRs), a standard geographic unit developed by the Dartmouth Institute for Health Policy and Clinical Practice, Hanover, New Hampshire, were used to control for underlying geographic variation in health service use.Total Medicare spending by HRR was calculated as median age-, sex-, and race and ethnicity-adjusted expenses for all Medicare beneficiaries during a 5-year period from 2012 to 2016. 28The HRRs were then categorized into high (>75th percentile), medium (25th-75th percentile), and low (<25th percentile) levels of spending.The HRR spending level variable was assigned to each individual beneficiary.

Statistical Analysis
In unadjusted analyses, we used χ 2 tests of proportions or 1-way analysis of variance.For analyses adjusting for all covariates, we calculated separate multivariable logistic regression models for binary outcomes to produce the odds ratios (ORs) and 95% CIs of an outcome for timely vs no ACP and late vs no ACP.For EOL expenditures, a multivariable linear regression model was computed to produce adjusted mean differences in expenditure.All P values were 2-sided and adjusted using the Bonferroni correction 29 for multiple hypothesis testing, and P < .008indicated statistical significance.
Because participants who chose to have an ACP visit could be predisposed toward less intensive EOL care regardless of their ACP experience, we performed a matched propensity score adjustment as a sensitivity analysis. 30We used a multivariable logistic model to calculate the probability of receiving an ACP (the propensity score) including the aforementioned covariates.The no-ACP group was then matched 1:1 without replacement to the timely ACP group using a greedy matching algorithm. 31The estimated propensity score was used for the matching with a caliper size of 0.01.To evaluate the propensity score model, we compared the overlap of propensity scores in each group using graphs (eFigure in the Supplement) and compared propensity score-matched baseline characteristics using standardized differences, with a standardized difference of less than 20% indicating adequate balance between the 2 groups.

Sample Characteristics
A total of 955 777 fee-for-service Medicare beneficiaries identified as seriously ill in 2016 died between January 1, 2017, and December 31, 2018 (Table 1).They were more likely to be women

Association of ACP With EOL Outcomes Unadjusted Results
Compared with patients with no billed ACP visits, patients who had a timely ACP consistently experienced less intensive EOL care, including lower rates of in-hospital death (17  Patients with timely ACP visits accounted for slightly higher EOL expenditures ($18 205) than those without an ACP ($17 141), whereas patients with a late ACP had much higher EOL expenditures ($27 187; P < .001).This pattern persisted across HRR spending categories (eTable 8 in the Supplement).

Adjusted Results
In adjusted multivariable analyses (Table 2), seriously ill beneficiaries who had a timely ACP were significantly less likely to die in the hospital (adjusted OR [aOR], 0.85; 95% CI, 0.84-0.87),be admitted to a hospital in their last 30 days of life (aOR, 0.84; 95% CI, 0.83-0.85),have an ICU admission in their last 30 days of life (aOR, 0.87; 95% CI, 0.85-0.88),or have an ED visit in their last 30 days of life (aOR, 0.83; 95% CI, 0.82-0.84).However, decedents with timely ACP visits were somewhat more likely to be enrolled late to hospice (aOR, 1.06; 95% CI, 1.03-1.08).For every measure, decedents with a late ACP experienced more intensive use of EOL services.eTables 1 to 7

JAMA Health Forum | Original Investigation
in the Supplement show the complete results of multivariable models with all covariates represented.Differences in adjusted total mean expenditures followed a similar pattern as that of the unadjusted results (Table 3).Compared with patients with no ACP, there were small additional expenditures for patients with a timely ACP in the 2 higher spending regions (mean difference for medium HRR, $245 [95% CI, $46-$445]; mean difference for high HRR, $1411 [95% CI, $1117-$1704]) and large additional expenditures for patients with late ACP (mean difference for medium HRR, $7931 [95% CI, $7573-$8289]; mean difference for high HRR, $10 166 [95% CI, $9615-$10 717]).

Sensitivity Analyses Using Propensity Score Methods
After performing the propensity score-matched analysis, the results were nearly identical to those from the traditionally adjusted analyses with 2 exceptions (eTable 9 in the Supplement).There were no longer significant associations between timely ACP and late hospice referral or in mean total expenditures in the medium HRRs.

Discussion
In this cohort study of US Medicare beneficiaries, we examined the use of EOL health care services among Medicare decedents identified prospectively as seriously ill and found that having timely ACP visits (>30 days before death) with their clinicians was associated with less aggressive care in the last 30 days of life, including fewer hospital deaths, fewer hospital and ICU admissions, and fewer visits to the ED.Timely ACP was associated, albeit slightly, with higher overall EOL expenditures for those living in higher-spending HRRs.Associations between ACP and hospice use varied depending on the comparison.Timely ACP led to timely admission to hospice (Ն4 days before death).The association between timely ACP and late hospice use (a typical measure of EOL intensity) was no longer significant after propensity score-matched analysis.The lack of association with late hospice use might be explained by the parallels between hospice use and ACP in the sense that they both may  represent actions recognizing the likelihood of death and, at least for some patients, a desire to prioritize quality over quantity of life.
These findings have at least 2 sets of policy implications.First, they address an evolving controversy over the value of ACP.Critics of ACP assert that the limited studies on the benefits of ACP fail to present evidence of changes in the use of health care services or other outcomes. 32Our study counters this skepticism with the finding that patients with documented clinician visits involving ACP appear to use care differently at the end of life compared with those without such a visit.Second, patients living with serious illness have a high burden of both physical and psychological symptoms and treatments that adversely affect their quality of life and that of their family members.Advocates of ACP recommend that conversations about values and preferences take place throughout adulthood. 1,33However, given limited resources, our results suggest that health systems may prioritize conversations for patients who are closer to death, including those identified as seriously ill.Therefore, understanding the EOL outcomes of these patients is essential.Studies of small, carefully selected cohorts have found that better communication can reduce the use of invasive treatments and can improve bereavement outcomes for care partners. 4,34,35Whereas some of these studies have relied on interviews in unique settings, our analysis of Medicare claims, representing visits specifically involving ACP discussions, provides a greater level of standardization to the exposure 36 and a much larger scale represented by our cohort of all Medicare patients with serious illness.
The disproportionate level of spending at the end of life has often been raised as a reason to more carefully consider how the US provides care. 37,38We found a small but significant increase in overall EOL spending for patients with timely ACP visits.0][41][42][43][44] However, our effect size was small and our sample was large, so this outcome does not necessarily represent a different conclusion.Regardless of the association with health care costs, the effect on patients' well-being at the end of life due to the intensity of potentially unwanted services is an important measure. 45r findings that patients receiving their first billed ACP visit within 30 days of dying had substantially higher use of health care should serve as a caution for research in this area.There are at least 3 possible explanations for such disparate results when compared with decedents having earlier ACP visits.First, the measurement periods of exposure and outcome overlap.Patients receiving multiple services in the last 30 days of life have more contact with the health care system and more opportunity to have an ACP visit, and this increased contact may have motivated them or their physicians to have a billed ACP conversation.Also, the late ACP visit might have involved a conversation about goals of care or occurred during a hospital stay, which is often associated with high costs.Second, the circumstances of patients who have their first ACP conversation so near to death may be different from those of other decedents (eg, perhaps the reason for the delay is an underlying preference for more life-extending services).Third, patients with late ACP visits may be more likely to be in crisis and thus may decide to have more services performed.Clearly, more studies are needed to determine the dynamics of this subset of Medicare beneficiaries.

Limitations
Our study has several limitations.First, health care claims do not contain information on the mechanics, content, or quality of the discussion.These discussions could contain preferences for more aggressive rather than less aggressive care.Knowledge of the content of ACP may help explain the contrary findings of late ACP.Second, the uptake of Medicare's ACP codes has been slow. 15In fact, many more patients likely had ACP discussions, but their clinicians did not bill Medicare for the service.Barriers to the adoption and use of these codes may include such factors as clinician lack of awareness of the new codes or unease with documentation requirements, reticence to incur patient cost-sharing, or limits placed on the type of clinical staff eligible to bill. 36Third, we lacked data on beneficiaries enrolled in non-fee-for-service Medicare Advantage plans.Because those plans may focus on value, the EOL experiences of that covered population may differ from our findings.Fourth, our findings may be subject to so-called decedent bias. 46However, those study designs typically analyze data beginning with the time of death.We identified our cohort of seriously ill patients in 2016 and followed them up until death.Finally, as with all administrative data sets, the findings are subject to the accuracy of the claims.

Conclusions
Advance care planning has been encouraged for several decades yet has failed to be adopted broadly, and its effect on goal-concordant care at EOL continues to be debated.Medicare's ACP codes were intended to incentivize ACP by offering clinician compensation.To the extent that many or most patients near EOL desire less aggressive care, the results of this cohort study suggest that the provision of ACP is associated with lower use of health care services.As provision of ACP and the payment of claims continues to grow, future studies using available EOL research methods may be necessary to be confident about its benefits.
This cohort study was approved by the institutional review board at Partners Healthcare.Data were provided by the Centers for Medicare & Medicaid Services and used under an appropriate data use agreement.The analyses were performed from November 1, 2020, to March 31, 2021, using data from January 1, 2016, to December 31, 2018.The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reportingguideline for cohort studies.16

Table 1 .
Characteristics of Fee-for-Service Medicare Beneficiaries With Serious Illness Who Died in 2017 or 2018, by ACP Grouping a Percentages have been rounded and may not total 100.b Specific racial and ethnic categories were unavailable.

Table 2 .
Adjusted Odds Ratios for Use of Health Care Resources in Last 30 Days of Life a Bonferroni corrected P = .008for 6 pairwise comparisons.All models were adjusted for sex, age group, race and ethnicity, Charlson Comorbidity Index score, hospital referral region median Medicare spending level, and dual eligibility status.

Table 3 .
Mean Difference in Expenditure Among HRR Groups a Bonferroni corrected P = .008for 6 pairwise comparisons.All models adjusted for sex, age group, race and ethnicity, Charlson Comorbidity Index, HRR median Medicare spending level, and dual eligibility status.