Jeffrey Clough, MD

Since 2019, the Medicare Shared Savings Program (MSSP) has allowed accountable care organizations (ACOs) to choose either retrospectively or prospectively attributed ACO populations. To understand how ACOs’ choice of attribution method affects incentives for care among seriously ill Medicare beneficiaries, this study compares beneficiary characteristics and Medicare per capita expenditures between prospective and retrospective ACO populations.

 This retrospective, cross-sectional analysis describes survival, patient characteristics, and Medicare spending for Medicare fee-for-service beneficiaries identified with serious illness (n = 1,600,629) using 100% Medicare Master Beneficiary Summary and MSSP beneficiary files (2014-2016).

 We used generalized linear models with ACO and year fixed effects to estimate the average within-ACO difference between potential retrospective and prospective ACO populations.

 Dying in the first 90 days of the performance year was associated with reduced odds of retrospective ACO attribution (odds ratio [OR], 0.24; 95% CI, 0.24-0.25) relative to beneficiaries surviving 270 days or longer. Similarly, hospice use was associated with reduced odds of retrospective assignment (OR, 0.80; 95% CI, 0.79-0.80). Among ACOs that did not achieve shared savings, average per capita Medicare expenditures (after truncation) were $2459 (95% CI, $2192-$2725) higher for prospective vs retrospective ACO populations. The difference was $834 (95% CI, $402-$1266) greater per capita among ACOs that achieved shared savings.

 The difference in survival and spending for ACO populations captured by prospective vs retrospective attribution methods means that ACOs may need to employ different care management strategies to improve performance depending on their attribution method.

Compared with retrospectively attributed Medicare accountable care organization (ACO) populations, prospectively attributed Medicare ACO populations included more decedents and had higher per-beneficiary per-year expenditures among seriously ill Medicare beneficiaries.

In evaluating ACO financial performance, the Medicare Shared Savings Program truncates the extremely high costs of 1 in 4 decedents, substantially reducing the difference in per capita expenditures between prospectively and retrospectively attributed ACO populations.

The difference in survival and spending for ACO populations captured by prospective vs retrospective attribution methods means that ACOs may need to employ different care management strategies to improve performance depending on their attribution method.

Since 2019, the Medicare Shared Savings Program (MSSP) has allowed accountable care organizations (ACOs) in the Basic or Enhanced tracks to choose either retrospectively or prospectively attributed ACO patient populations.

 Seriously ill patients with high needs, costs, and functional limitations have a high risk of death and may benefit from ACO care management strategies. Little is known about how ACOs’ choice of attribution method affects ACO attribution, and thus incentives for care, for seriously ill Medicare beneficiaries.

Attribution methods in MSSP have evolved over time. Since program inception, CMS required MSSP ACOs to use retrospective assignment, with the exception of the 8% of ACOs in Track 3 or the Next Generation Track, which both started in 2016.

 Beginning in 2019, the Pathways to Success MSSP rule allows ACOs an annual attribution choice: retrospective or prospective (potentially supplemented by voluntary patient attribution).

 Retrospective and prospective options both use claims-based algorithms to attribute patients to an ACO if a plurality of qualifying services over a 12-month period are received from a participating provider. In retrospective attribution, patients are assigned based on care provided during the contract year, and final attribution occurs at the end of the contract performance year (PY). In retrospective attribution, patients who received care from the ACO during the PY are captured in the ACO population. In prospective attribution, patients are assigned at the start of the contract PY based on claims in the prior calendar year, which allows ACOs to know their defined population and provide care management during the PY.

Improving serious illness care is critical to curb the trend of increasing health care expenditures, as the top 5% of spenders account for nearly 60% of health care costs.

 Most patients with serious illness have multiple comorbidities and functional limitations, which require medical and social services from providers and caregivers.

 As a result, serious illness care is often fragmented or incomplete.

 Polypharmacy increases with comorbidity, increasing risk of adverse reactions and inappropriate use.

 Consequently, improving care for individuals with serious illness has been a principal motivation for recent ACO reforms. Understanding the factors and outcomes associated with ACO attribution, specifically for seriously ill Medicare beneficiaries, will inform the evolving MSSP program and care reforms for the seriously ill.

Evidence from the first year of the MSSP suggested that decedents are differentially excluded from retrospectively attributed ACO populations,

 but the relationship between choice of attribution method and patient case mix in retrospectively compared with prospectively attributed ACO populations is unclear. We compared patient characteristics and Medicare per capita expenditures between ACO populations defined using prospective and retrospective claims-based attribution methods. Additionally, we compared observed and truncated per-beneficiary per-year (PBPY) spending using the rules applied by CMS in calculating MSSP ACO financial performance, which truncates spending for patients exceeding the 99th percentile nationally by eligibility category. We hypothesized that PBPY spending will be higher in prospective rather than retrospective ACO populations, with and without truncation, because of greater inclusion of decedents.

To evaluate the probability of attribution of decedents to ACO populations, we tested the association of beneficiary survival during the PY with (1) retrospective attribution to an MSSP ACO and (2) prospective attribution to an MSSP ACO. Outcomes for ACO attributed populations are evaluated cross-sectionally by ACO-year rather than following patients over time to reflect the way ACOs’ performance is evaluated by CMS and, as a result, financial incentives for ACO decision-making.

Using 100% Medicare beneficiary summary and MSSP beneficiary files for years 2014-2016, we included fee-for-service (FFS) Medicare beneficiaries identified as seriously ill in the year prior to the PY. We applied the inclusion algorithm to the year prior to the ACO PY to prevent introducing bias by conditioning on colliders, or effects of the outcome and exposure. We identified the serious illness population by adapting a preexisting definition using Medicare claims data with 3 eligibility criteria: (1) any inpatient claim, (2) any claim for home health or skilled nursing or durable medical equipment, and (3) either a serious chronic illness or multicomorbidity.

 Comorbidities were defined using the Chronic Conditions Warehouse algorithms for 27 common conditions and other chronic and disabling conditions.

 Serious chronic illnesses included advanced liver disease or cirrhosis; lung, colorectal, or endometrial cancer; chronic obstructive pulmonary disease; congestive heart failure; Alzheimer disease or related dementias; diabetes with complications including ischemic heart disease or peripheral vascular disease; hip fracture; or renal failure as indicated by any use of dialysis.

The primary outcome is ACO Medicare spending PBPY. For each ACO in each PY, we described average Medicare spending for 2 possible definitions of ACO populations, prospective and retrospective, regardless of the attribution method that CMS actually used to evaluate the ACO. Medicare spending was evaluated cross-sectionally by ACO-year rather than following patients over time to reflect the way ACOs’ performance is evaluated by CMS and perceived by ACO administrators. To measure spending, we summed Medicare parts A and B PBPY spending reported in annual Medicare Beneficiary Summary files, then annualized based on eligibility months to mirror MSSP calculations, and then standardized spending at the county level using the CMS geographic variation public use file to remove variation due to inflation and geography. We applied the MSSP stop-loss rule by truncating spending at 0 and the average 99th percentile within each Medicare eligibility category over the study period.

Our key interest is ACO-level decision-making, and ACOs make strategic decisions by viewing the attributed population as a whole and comparing the potential retrospective and prospective ACO populations. For each eligible beneficiary, CMS provides an attribution status at the beginning of the PY (prospective) and at the end of the PY (retrospective) regardless of the ACO’s decision to use retrospective or prospective attribution for performance evaluation. Thus, we were able to define potential prospective and potential retrospective assignment for each beneficiary and compare characteristics and outcomes for the 2 potential ACO populations within each ACO. Retrospective attribution was defined using a binary indicator for the CMS final assignment indicator in the MSSP beneficiary-level administrative data files for each ACO PY. Prospective attribution was defined using a binary indicator for CMS preliminary assignment of a beneficiary to an MSSP ACO in the first quarter of the PY. We considered using the final assignment flag from the prior year to define prospective assignment; however, this would have restricted the sample to ACOs with multiple years in the MSSP, and we would not be able to define prospective attribution in the first year of an ACO contract. Because of the growth of the MSSP during this period, more than a third of the sample would have been excluded. The first-quarter preliminary assignment flag we used and the prior-year final assignment flags apply the same algorithm to the same period. The only difference we can discern is that fewer claims are submitted and available to inform the algorithm within 7 days (current-year preliminary assignment) than within 90 days (prior-year final assignment) of the end of the PY. As a result, we decided the improved generalizability was more valuable to the analysis than the small difference in precision that may be gained by using the prior-year flag.

Covariates included patient demographics (gender, age, race), Medicare eligibility category (age, disabled, end-stage renal disease [ESRD], or both disabled and ESRD), comorbidities (27 chronic conditions), county factors from the CMS geographic variation file (mean Medicaid and Medicare Advantage payments, average Hierarchical Condition Category scores), and year fixed effects. In the model describing the association between beneficiary survival and retrospective attribution, 4 survival categories were defined where the date of death was in 0 to 89 days, 90 to 179 days, 180 to 269 days, or 270 or more days from the start of the PY (January 1 of each year). Beneficiaries surviving the entire year were assigned to the referent category of 270 or more days survived from the start of the PY.

The association between patient survival and retrospective attribution to an ACO was assessed with logistic regression. The beneficiary-level model included an interaction of survival category and prospective assignment to improve model fit and was adjusted for patient demographics, comorbidities, county factors, and year fixed effects.

ACO-level observations were used to compare Medicare spending between the 2 potential ACO populations, retrospective and prospective. We aggregated truncated PBPY expenditures among prospective and retrospective populations for each ACO by year. The objective was to describe the difference in spending resulting from the difference in case mix due to selection; thus, the models are intentionally not adjusted for beneficiary characteristics. We used generalized linear models with ACO and year fixed effects to estimate the average within-ACO difference between retrospective and prospective ACO populations. The model presented included an interaction of attribution type (prospective or retrospective) with ACO shared savings success in that year. ACO shared savings success is defined by the indicator provided by CMS for achieved shared savings in the MSSP PY files. Correlations of outcomes within ACOs were modeled using exchangeable correlation structures. Analyses were conducted in SAS Studio Enterprise Edition 3.7 (SAS Institute), and this study was approved by the Duke University Institutional Review Board.

Despite similarity in demographics and comorbidities, spending and mortality differed substantially between prospective and retrospective ACO populations. Of 5,564,982 Medicare FFS beneficiaries identified with serious illness, beneficiaries attributed to prospective MSSP ACO populations (n = 1,336,562) are similar to beneficiaries attributed to retrospective ACO populations (n = 1,306,168) in age, gender, and disability status. Of 1,600,629 observations attributed either prospectively or retrospectively, two-thirds (65%) (n = 1,042,655) have common membership in retrospective and prospective ACO populations. The prevalence of chronic conditions was not meaningfully different between the groups, and most differences were less than 2 percentage points.

Mean unadjusted annualized Medicare expenditures PBPY are $10,000 higher in the prospective than in retrospective ACO populations. After applying the MSSP stop-loss rule to truncate outlier spending, the difference in unadjusted PBPY spending between prospective and retrospective ACO populations is less than $3000. A smaller proportion of beneficiaries died during the PY in retrospective vs prospective ACO populations (12.5% vs 15.6%, respectively) (

). The differences in expenditures and survival are driven by the subgroup members who are assigned prospectively but not retrospectively, referred to as prospective only (n = 293,907). Compared with the retrospective-only group (n = 264,067), the PY death rate in the prospective-only subgroup is nearly double, at 27.6% and 14.0%, respectively.

Among beneficiaries prospectively assigned to an ACO, death in the first 90 days was associated with a 76% reduction in odds of retrospective ACO attribution (odds ratio [OR], 0.24; 95% CI, 0.24-0.25), and dying in the second or third quarter was associated with 41% and 31% reductions in odds (OR, 0.59; 95% CI, 0.58-0.60; and OR, 0.69; 95% CI, 0.67-0.70, respectively) compared with survival of 270 days or more. Estimates for survival effects were independent of hospice enrollment, which was also associated with reductions in the odds of retrospective attribution (OR, 0.80; 95% CI, 0.79-0.80). Full results are reported in 

). Among beneficiaries who were prospectively assigned to an MSSP ACO, beneficiaries dying in the first quarter of the PY had a predicted probability of retrospective assignment of 0.45 compared with 0.77 for those surviving until the fourth quarter (

The difference in the odds of prospective attribution was less than 3% for beneficiaries dying in the first 90 days relative to those surviving 270 days or more (

Spending in Retrospective and Prospective ACO Populations

Using an ACO-fixed effect model, we evaluated the within-ACO difference in PBPY Medicare expenditures for prospective and retrospective ACO populations. After truncation of outlier spending, average PBPY Medicare expenditures were higher for prospective vs retrospective ACO populations with a mean difference of $2459 (95% CI, $2192-$2725) among ACOs that did not achieve shared savings (

). The mean difference was greater among ACOs that achieved shared savings that year, with an additional $834 (95% CI, $402-$1266) PBPY increase in the prospective vs retrospective ACO populations. Restricting the sample to the 1,355,777 ACO beneficiaries who survived the PY, we found that prospective ACO populations cost $261.58 (95% CI, $79.12-$416.65) PBPY more than retrospective ACO populations, on average.

Observed (not truncated) spending for prospective and retrospective ACO populations is presented by inpatient, postacute, hospice, and other care settings (

). In each category, spending is greater in prospective than retrospective populations. The MSSP stop-loss rule reduces the difference in PBPY expenditures between prospective and retrospective ACO populations overall (

), and greater changes are observed among subgroups of hospice users and decedents. The stop-loss threshold truncated spending for 5.1% of the seriously ill beneficiaries and 24.5% of decedents. Decedents’ actual annualized spending surpassed $150,000 PBPY, on average, in prospective populations vs $119,000 PBPY in retrospective populations. After applying the stop-loss threshold to truncate spending, the difference between prospective and retrospective populations is reduced to $3000 PBPY among decedents, hospice users, and the full sample. Furthermore, 80% of beneficiaries exceeding the stop-loss threshold died during the PY and 47% died during the first 90 days.

Compared with retrospectively attributed ACO populations, prospectively attributed MSSP ACO populations included more decedents and had higher PBPY expenditures. Beneficiaries who survived the PY contributed 10% of the difference between prospective and retrospective cohorts, suggesting that the majority of the difference is due to greater inclusion of decedents in prospective ACO populations. Although the MSSP stop-loss rule narrows the difference in spending between retrospective and prospective ACO populations by truncating outlier spending, which is often decedents’ spending, a significant difference in spending remains between the 2 potential ACO populations after truncation. These findings have implications for ACO care delivery for seriously ill patients, as well as MSSP rules and regulations.

We found that decedents and hospice users had reduced odds of retrospective attribution to an MSSP ACO in recent PYs. This relationship is a consequence of the mechanics of claims-based attribution, which relies on the accumulation of claims over time to identify attributable beneficiaries. The longer that seriously ill beneficiaries survive in a given PY, the more likely they will use services necessary for retrospective attribution. A study using 2012 Medicare data found that the majority of unattributable beneficiaries had no encounter with the health care system at all; however, 6% of those unattributed were decedents with high spending.

 Other studies evaluating end-of-life care in MSSP ACOs have applied prospective attribution instead of retrospective attribution to retain substantial proportions of hospice users and decedents in the study sample and reduce risk of bias.

The difference in per capita spending between prospective and retrospective attribution methods may influence ACOs’ choice of attribution method; however, low per capita spending does not necessarily result in better financial performance relative to the benchmark. Savings or losses are assessed relative to the relevant benchmark. ACO benchmark calculations can be complex, incorporating historical benchmarks, regional trends, truncation, and other factors. In theory, if an ACO chooses prospective (or retrospective) attribution, and they are compared against a prospective (or retrospective) benchmark, then an ACO’s choice of attribution method should not affect its savings calculation. It is unclear whether ACO benchmarks adequately adjust for differences in case mix.

Prior studies have evaluated ACO selection of healthier patients with mixed results; our results suggest that sicker patients may be differentially excluded as an artifact of attribution methodologies rather than ACO activities.

 In the past, CMS would not upwardly adjust beneficiaries’ risk scores between PYs, which may have discouraged inclusion of high-risk beneficiaries in the ACO. The new MSSP rule allows up to 3% change, which balances accuracy against potential for upcoding.

The difference in survival and spending for ACO populations captured by prospective vs retrospective attribution methods means that ACOs may need to employ different care management strategies to improve performance depending on their attribution method. Despite evidence that early ACOs targeted preventive health in low-risk populations to achieve savings, a national survey found that 93% of ACOs self-reported partially implemented efforts to identify their serious-illness population.

 Seriously ill beneficiaries may benefit from the care management strategies employed by ACOs. Building on the success of ACOs, CMS is testing the next evolution of risk-sharing arrangements through Direct Contracting, with a specific focus on beneficiaries with complex, chronic conditions. Similarly, the Primary Care First CMS demonstration model encourages a seamless continuum of care for seriously ill Medicare beneficiaries, potentially led by palliative and hospice providers.

Prospective and retrospective attribution methods both have benefits that should be carefully weighed by policy makers and health system leaders. Greater inclusion of decedents and hospice users in prospectively attributed populations means that more of ACOs’ financial performance will depend on moving the needle among seriously ill patients approaching the end of life. In addition, prospective attribution methods allow ACOs to know which patients are included in their defined population at the beginning of the period, allowing them to target care programs more directly. Retrospectively attributed ACO populations are more likely to include beneficiaries newly enrolled in Medicare during the period and patients with newly diagnosed serious illness, and they are more likely to have a greater number of attributed patients overall.

 Unlike prospective assignment, retrospective attribution does not hold the ACO responsible for beneficiaries who have moved or changed their care and receive little or no care from ACO providers during the year.

Although there is growing evidence to inform ACO attribution strategies, the implications of technical contract specifications, like the MSSP rule for truncating outlier spending, are not well understood. Truncating outlier spending is necessary to protect ACOs against catastrophic claims; however, the current MSSP truncation rule affects spending calculations for 1 in 4 decedents, reducing ACO accountability for escalating end-of-life spending. By truncating the extremely high spending on decedents, the MSSP stop-loss rule substantially reduces the differences in PBPY expenditures between prospective and retrospective ACO populations. The current method of annualizing spending results in highly inflated values by applying end-of-life spending to the full year, and alternative models may better capture spending for decedents. As CMS indicated in the 2018 rules, new methodologies specifically targeting “the most complex patients, such as high-risk patients or those receiving care for chronic conditions,” are needed for both attribution and stop-loss rules.

Our conclusions are limited to the subgroup of the MSSP-attributed population that meets criteria for serious illness in the prior year and has continuous FFS eligibility during the study period. The population with serious illness changes with time, as individuals who are seriously ill in a given year often die or do not meet the serious illness criteria in the following year, resulting in substantial churn in the serious illness sample from year to year. This study benefits from a comprehensive national data source and MSSP beneficiary data that reflect CMS algorithms for attribution to ACOs. A further limitation is that we were unable to assess the sensitivity of our findings to the definition of prospective assignment using prior-year retrospective assignment rather than current-year preliminary assignment.

As MSSP saturates the health care landscape, the impact of ACO contractual incentives on care for seriously ill patients will intensify. ACOs using prospective attribution may have greater motivation to target high-risk groups with care management programs because of greater inclusion of decedents and hospice users in the ACO population. CMS should consider performance algorithms and benchmarks that increase accountability and reward health systems that are able to bend the spending curve near the end of life. Future research should evaluate potential models, including specialized ACOs, disease-specific medical homes, and other serious illness payment models.


 Margolis Center for Health Policy, Duke University (BGK, WKB, RS, DA, CHV, JC, MBM), Durham, NC, and Washington, DC; Department of Population Health Sciences, Duke University School of Medicine (BGK, CHV), Durham, NC; Center of Innovation to Accelerate Discovery and Practice Transformation (ADAPT) (BGK, CHV), Durham VA Medical Center, Durham, NC; Leavitt Partners (DBM), Salt Lake City, UT; Duke Clinical Research Institute (JC), Durham, NC.

 This research is funded by the Gordon and Betty Moore Foundation through Grant GBMF6879 to the Margolis Center for Health Policy at Duke University.

 Dr Bleser has previously received consulting fees from Merck on vaccine litigation unrelated to this work and serves as Board Vice President (uncompensated) for Shepherd’s Clinic, a clinic providing free health care to the uninsured in Baltimore, MD. Dr Muhlestein is employed by Leavitt Partners, which consults about accountable care organizations. Dr McClellan is an independent board member on the boards of Alignment Healthcare, Cigna, Johnson & Johnson, and PrognomIQ; cochairs the Guiding Committee for the Health Care Payment Learning and Action Network; and receives fees for serving as an adviser for Arsenal Capital Partners, Blackstone Life Sciences, and MITRE.

 Concept and design (BGK, WKB, RS, DA, CHV, DBM, JC, MBM); acquisition of data (BGK); analysis and interpretation of data (BGK, WKB); drafting of the manuscript (BGK, WKB); critical revision of the manuscript for important intellectual content (BGK, WKB, RS, DA, CHV, DBM, JC, MBM); statistical analysis (BGK); obtaining funding (WKB); and administrative, technical, or logistic support (BGK, WKB, RS, DA, CHV, DBM, JC, MBM).

 Brystana G. Kaufman, PhD, MSPH, Margolis Center for Health Policy, Duke University, 230 Science Dr, Durham, NC 27705. Email: Brystana.kaufman@duke.edu.

1. CMS, HHS. Medicare Program; Medicare Shared Savings Program; accountable care organizations—Pathways to Success and extreme and uncontrollable circumstances policies for performance year 2017. final rules. 

2. Jaffery J, Ronk K, Smith M. Does beneficiary switching create adverse selection for hospital-based ACOs? 

. April 15, 2019. Accessed November 3, 2020. https://www.healthaffairs.org/do/10.1377/hblog20190410.832542/full/

Dying in America: Improving Quality and Honoring Individual Preferences Near the End of Life

Individuals Living in the Community With Chronic Conditions and Functional Limitations: A Closer Look

5. Aldridge MD, Kelley AS. The myth regarding the high cost of end-of-life care. 

. 2015;105(12):2411-2415. doi:10.2105/AJPH.2015.302889

6. Kelley AS. Epidemiology of care for patients with serious illness. 

. 2013;16(7):730-733. doi:10.1089/jpm.2013.9498

7. Joynt KE, Figueroa JF, Beaulieu N, Wild RC, Orav EJ, Jha AK. Segmenting high-cost Medicare patients into potentially actionable cohorts. 

. 2017;5(1-2):62-67. doi:10.1016/j.hjdsi.2016.11.002

8. Figueroa JF, Zhou X, Jha AK. Characteristics and spending patterns of persistently high-cost Medicare patients. 

. 2019;38(1):107-114. doi:10.1377/hlthaff.2018.05160

9. Figueroa JF, Joynt Maddox KE, Beaulieu N, Wild RC, Jha AK. Concentration of potentially preventable spending among high-cost Medicare subpopulations: an observational study. 

. 2017;167(10):706-713. doi:10.7326/M17-0767

10. Kelley AS, Bollens-Lund E. Identifying the population with serious illness: the “denominator” challenge. 

. 2018;21(S2):S7-S16. doi:10.1089/jpm.2017.0548

11. Aldridge MD, Bradley EH. Epidemiology and patterns of care at the end of life: rising complexity, shifts in care patterns and sites of death. 

. 2017;36(7):1175-1183. doi:10.1377/hlthaff.2017.0182

12. Moore KL, Patel K, Boscardin WJ, Steinman MA, Ritchie C, Schwartz JB. Medication burden attributable to chronic co-morbid conditions in the very old and vulnerable. 

2018;13(4):e0196109. doi:10.1371/journal.pone.0196109

13. Ouayogodé MH, Meara ER, Chang CH, et al. Forgotten patients: ACO attribution omits those with low service use and the dying. 

14. Kelley AS, Covinsky KE, Gorges RJ, et al. Identifying older adults with serious illness: a critical step toward improving the value of health care. 

. 2017;52(1):113-131. doi:10.1111/1475-6773.12479

15. Kelley AS. Defining “serious illness.” 

. 2014;17(9):985. doi:10.1089/jpm.2014.0164

16. Condition categories. Chronic Conditions Data Warehouse. Accessed June 14, 2019.
https://www.ccwdata.org/web/guest/condition-categories

17. Kaufman BG, O’Brien EC, Stearns SC, et al. Medicare Shared Savings ACOs and hospice care for ischemic stroke patients. 

. 2019;67(7):1402-1409. doi:10.1111/jgs.15852

18. Gilstrap LG, Huskamp HA, Stevenson DG, Chernew ME, Grabowski DC, McWilliams JM. Changes in end-of-life care in the Medicare Shared Savings Program. 

. 2018;37(10):1693-1700. doi:10.1377/hlthaff.2018.0491

19. McWilliams JM, Hatfield LA, Landon BE, Chernew ME. Spending reductions in the Medicare Shared Savings Program: selection or savings? National Bureau of Economic Research working paper 26403. October 2019. Accessed November 3, 2020. https://www.nber.org/papers/w26403

20. Markovitz AA, Hollingsworth JM, Ayanian JZ, et al. Risk adjustment in Medicare ACO program deters coding increases but may lead ACOs to drop high-risk beneficiaries. 

. 2019;38(2):253-261. doi:10.1377/hlthaff.2018.05407

21. McWilliams JM, Chernew ME, Landon BE. Medicare ACO program savings not tied to preventable hospitalizations or concentrated among high-risk patients. 

. 2017;36(12):2085-2093. doi:10.1377/hlthaff.2017.0814

22. Driessen J, West T. Variation in end-of-life care is an open invitation for accountable care organization innovation. 

. August 25, 2017. Accessed November 3, 2020. https://www.healthaffairs.org/do/10.1377/hblog20170825.061646/full/

23. Bleser WK, Saunders RS, Winfield L, et al. ACO serious illness care: survey and case studies depict current challenges and future opportunities. 

. 2019;38(6):1011-1020. doi:10.1377/hlthaff.2019.00013

24. Lewis VA, McClurg AB, Smith J, Fisher ES, Bynum JP. Attributing patients to accountable care organizations: performance year approach aligns stakeholders’ interests. 

2013;32(3):587-595. doi:10.1377/hlthaff.2012.0489

Articles

This study compared beneficiary characteristics and Medicare per capita expenditures among seriously ill Medicare accountable care organization (ACO) populations defined using prospective and retrospective claims-based attribution methods.

Jeffrey Clough, MD

Articles

Prospective or Retrospective ACO Attribution Matters for Seriously Ill Patients