John M. Hollingsworth, MD, MS

: Surgical care among older adults is costly. Although Medicare accountable care organizations (ACOs) are designed around primary care, there are reasons to believe that participation may also affect spending on surgery. This study examines the impact that Medicare Shared Savings Program ACO alignment has on spending for inpatient and outpatient surgical care.

 We conducted a retrospective cohort study using national Medicare claims (2008-2015).

 Among a 20% random national sample of beneficiaries, we identified adults 65 years or older enrolled in fee-for-service Medicare, distinguishing between those aligned and unaligned with a Medicare Shared Savings Program ACO. We then measured payments for surgical services made on their behalf. Finally, we fit multivariable regression models to evaluate the association between ACO alignment and spending for inpatient and outpatient surgical care.

 We identified 37,249,845 beneficiary-year observations, of which 2,950,188 (7.9%) were aligned with an ACO. After adjustment for patient factors, ACO alignment was associated with $181 lower spending per beneficiary-year (95% CI, –$243 to –$118; 

 < .001). ACO alignment was associated with 2.9% fewer inpatient surgical episodes per year (incidence rate ratio, 0.97; 95% CI, 0.96-0.98; 

 < .001), but 2.3% more outpatient episodes per year (incidence rate ratio, 1.02; 95% CI, 1.02-1.03; 

 < .001). Among inpatient surgical episodes, average payments were $956 lower for ACO-aligned beneficiaries (95% CI, –$1218 to –$694; 

 Medicare Shared Savings Program ACO alignment was associated with savings on surgical care. These savings resulted from increased outpatient surgery and reduced use of inpatient surgery, as well as reduced spending per inpatient surgical episode. Greater focus on surgical care may improve the ability of ACOs to control health care spending.

Surgery is highly prevalent in older adults and, as the population ages, accounts for an increasing share of health care spending in the United States.

 In fact, recent estimates suggest that more than half of total annual Medicare payments are attributable to surgical care.

 Moreover, there is considerable variation in spending on surgery across geographic regions that cannot be explained by differences in disease incidence or severity,

 suggesting an opportunity for CMS to reduce spending growth and maintain (or even enhance) quality. One major initiative that may help achieve this is Medicare’s Shared Savings Program accountable care organization (ACO) model.

Although Medicare ACOs are designed around the provision of primary care services, participation may influence spending on surgery in several ways. Medicare ACOs use shared accountability and collective incentives to encourage deeper clinical integration among physicians, hospitals, and other health care providers.

 Because they are held accountable for all costs of aligned beneficiaries, participating primary care providers (PCPs) may consider how costly surgeons are in their referral decisions. Further, when an aligned beneficiary undergoes surgery, participating PCPs’ focus on care coordination may smooth care transitions during the surgical episode, reducing costly readmissions. The potential of shared savings payments may motivate participating surgeons to be more cost conscious. In response, they may raise their thresholds for intervention or shift care to lower-cost settings (eg, ambulatory surgery centers) when surgery is deemed necessary.

To understand what impact, if any, the ACO model has had on surgical spending, we analyzed national Medicare data over 2008 to 2015. After identifying all inpatient and outpatient surgical episodes that occurred in our sample, we measured total spending made on the behalf of beneficiaries during these episodes. We then fit multivariable regression models to evaluate for differences in surgical spending, as well as the probability of undergoing surgery, between beneficiaries who are either aligned or unaligned with an ACO.


We analyzed claims data from the Medicare Provider and Analysis Review, Carrier, and Outpatient Research Identifiable Files (RIFs) over 2008 to 2015 for a 20% national sample of beneficiaries. We restricted this cohort to beneficiaries who were 65 years or older and enrolled in fee-for-service Medicare. To be included, a beneficiary needed continuous Part A and Part B coverage throughout a given study year or until death during that year. We excluded beneficiaries participating in Medicare managed care plans, as well as those who resided outside the United States or Puerto Rico due to the incomplete availability of their claims. Finally, we excluded beneficiaries who did not receive any primary care services in a given study year (12.5% of otherwise eligible beneficiaries) because they would have been ineligible for ACO alignment according to Medicare policy.

Our exposure of interest was beneficiary alignment to a Medicare Shared Savings Program ACO. To distinguish beneficiaries who were aligned to an ACO from those who were not, we used the Shared Savings Program Beneficiary-level RIF. Specifically, we defined a time-varying indicator, set to 1 during a given study year if a patient was aligned to an ACO that year. It took the value of 0 otherwise. For 2012, corresponding to the first year of the Shared Savings Program, there were 2 ACO start dates. For beneficiaries aligned to an ACO that year with an April 1 start date, our indicator took the value of 0.75; for those aligned to an ACO that year with a July 1 start date, our indicator took the value of 0.5.

The primary outcome was per beneficiary spending on surgery, which we calculated for each study year. Our secondary outcome was the occurrence of a surgical episode within a given study year. To measure per beneficiary spending on surgery, we first determined whether a beneficiary underwent a surgical procedure within a given year. We did this using Health Care Procedure Coding System (HCPCS) codes for surgery on the integumentary (10000-19999), musculoskeletal (20000-29999), respiratory (30000-32999), cardiovascular (33000-37799, 93451-93662), hemic and lymphatic (38000-38999), mediastinum and diaphragm (39000-39999), digestive (40000-49999), urinary (50000-53999), male (54000-55899) and female genital (55900-59999), endocrine (60000-60999), nervous (61000-64999), ocular (65000-68999), and auditory (69000-69999) systems to identify relevant claims in the Carrier RIF.

Next, we used appropriate Place of Service and Type of Service codes appearing on a beneficiary’s claims to distinguish between inpatient and outpatient procedures, as well as outpatient procedures carried out in hospital outpatient departments vs freestanding ambulatory surgery centers vs the physician’s office. We excluded surgical procedures that took place in a skilled nursing facility or extended-stay rehabilitation facility.

We then summed all payments for claims filed on a beneficiary’s behalf during the surgical episode. For inpatient procedures, we summed all payments for claims beginning 3 days prior to admission and extending 90 days after discharge. Any beneficiary who had an inpatient surgical episode that would extend into the next calendar year was required to have Medicare eligibility for the months included in the episode window, and spending within that window was included in the study year of the index procedure. We further categorized the major components of payments, including those for the index hospitalization, readmissions, professional payments, and postacute care.

Given that postoperative complications and planned readmissions are uncommon following outpatient procedures,

 we defined the outpatient surgical episode as including all claims for services delivered within a 3-day window around the index outpatient procedure. For measurement purposes, if a beneficiary underwent an outpatient procedure within 90 days of discharge from a surgical admission, we considered it part of the inpatient episode. We summed all surgical episode payments made on behalf of a given beneficiary each study year and inflation-adjusted these payments to 2015 US dollars using the Consumer Price Index.

For our initial analytic step, we compared ACO-aligned and -unaligned beneficiaries across a variety of sociodemographic characteristics, including age, sex, race, urban/rural residence, and socioeconomic class (measured at the beneficiary zip code level).

 We also compared beneficiaries with respect to their level of comorbid illness (based on CMS’ Hierarchical Condition Categories).

 To determine whether observed differences were significant, we used parametric and nonparametric tests where appropriate.

With the beneficiary-year serving as our unit of analysis, we fit multivariable linear models regressing per beneficiary spending on our ACO indicator to evaluate for differences between aligned and unaligned beneficiaries. Next, to understand the extent to which differences in procedure mix may drive differences in surgical spending, we repeated the analysis of inpatient spending, including indicators for each diagnosis-related group (DRG) as fixed effects. We then used negative binomial regression to measure the association between the number of surgical procedures a beneficiary undergoes in a given study year and their ACO alignment status. To explore mechanisms for savings, we fit separate models, including only beneficiaries who underwent an inpatient surgical episode, and examined the association between ACO alignment and the components of inpatient spending (as defined previously). We estimated all models with robust standard errors to account for potential correlation of observations.

 In our models, we included ACO fixed effects, with unaligned beneficiaries considered part of a single group. We also included the sociodemographic and comorbid illness variables described previously, hospital referral region,

Finally, we performed 3 sensitivity analyses to examine the influence of ACO characteristics on the association between per beneficiary spending on surgery and alignment. Given emerging data suggesting that physician-led ACOs are associated with greater savings than hospital-led ACOs and hospital-physician collaborations,

 we first repeated our models including an interaction between our exposure and ACO leadership structure. We distinguished the latter using the Leavitt Partners ACO Database.

 Second, because ACO savings may increase with more experience participating as an ACO,

 we repeated our models including an interaction term between our exposure and an ACO’s duration of participation in the Shared Savings Program. For our third sensitivity analysis, we examined the effects of early (contract start date in 2012 or 2013) versus late (2014 or 2015) ACO model adoption on performance.

As an additional robustness check, we repeated outpatient surgical spending models to account for the type of outpatient procedure performed. We did this by adjusting models with groupings of HCPCS codes according to organ site and surgical specialty (see 

We performed all statistical tests using SAS version 9.4 (SAS Institute) and Stata 14 (StataCorp). We derived all predicted spending and episode counts from our models using the margins postestimation command in Stata. All our statistical tests were 2-tailed, and we considered a 

 value of less than .05 to be statistically significant. This study was deemed by the University of Michigan Medical School’s institutional review board to be exempt from its oversight.

This study included 37,249,845 beneficiary-year observations, of which 2,950,188 (7.9%) were aligned with a Medicare Shared Savings Program ACO. 

 presents the demographic characteristics and unadjusted outcomes for these beneficiaries. Compared with unaligned beneficiaries, those aligned to an ACO tended to be older and come from higher socioeconomic strata (

 < .0001 for each comparison). Unadjusted spending for ACO and non-ACO beneficiaries by year is represented in 

 displays results from our multivariable models, examining the association between per beneficiary spending on surgery and ACO alignment. After accounting for the measured differences described previously, overall spending on surgery was $181 lower per year for ACO-aligned beneficiaries (95% CI, –$243 to –$118; 

< .001). This difference was a result of lower inpatient surgical spending for ACO beneficiaries (–$198; 95% CI, –$261 to –$134; 

 < .001), which was offset by slightly higher average outpatient surgical spending ($17; 95% CI, $8-$26; 

 < .001). Table 2 demonstrates that these savings were due, in part, to increasing delivery of surgical care in less costly outpatient settings. Specifically, ACO alignment was associated with 2.9% fewer inpatient surgical episodes per year (incidence rate ratio [IRR], 0.97; 95% CI, 0.96-0.98; 

< .001) but 2.3% more outpatient surgical episodes per year (IRR, 1.02; 95% CI, 1.02-1.03; 

 < .001). Among inpatient surgical episodes (

), average payments were $956 lower for ACO-aligned beneficiaries (95% CI, –$1218 to –$694; 

This increased inpatient episode efficiency was driven by lower payments in each part of the inpatient episode: the index hospitalization, readmissions, postacute care, and professional payments (Table 3). In the model where we adjusted for differences in procedure mix (using indicator variables for each DRG), we found that a significant difference in favor of ACOs remained but was attenuated (Table 3). The index hospitalization component of inpatient episode spending (driven primarily by DRG-based payment to hospitals) was not significantly different in the model adjusting for DRG. The full results of all regression models are available in 

 displays the results of the 3 remaining sensitivity analyses. There was no significant association between ACO leadership structure or years of ACO experience on surgical spending. However, ACOs with contract start dates in 2012 realized increased surgical savings relative to later adopters of the ACO model.

Finally, after adjusting for surgical procedure type, ACO alignment was associated with approximately $6 lower outpatient surgical spending per beneficiary per year (

As a major driver of Medicare spending, surgical care is an important target for ACOs aiming to bend the cost curve. We found that surgical spending was significantly lower for ACO-aligned beneficiaries than for unaligned beneficiaries, representing $181 in savings per beneficiary-year. We identified 2 potential explanations for this spending reduction. First, ACO-aligned beneficiaries underwent fewer costly inpatient procedures than their unaligned counterparts. There was evidence to suggest that ACOs provided more outpatient surgery than non-ACOs, potentially as a mechanism to reduce inpatient surgery use. Second, when ACO-aligned beneficiaries required inpatient surgery, their episode spending was, on average, $956 lower than that of unaligned beneficiaries. This observed increased inpatient episode efficiency for ACO-aligned beneficiaries held across all components of spending, but the bulk of the savings was due to lower payments for postacute care.

To date, most evaluations of Medicare ACOs have focused on changes in spending for general medical care and chronic disease management.

 Studies of ACO effects on surgical care delivery are more limited, and those that have been published demonstrate underwhelming results. Data suggest that hospital participation in an ACO is not associated with reductions in the use of low-value procedural care,

 Moreover, in contrast to our findings, prior empirical work demonstrated comparable inpatient surgical episode spending for ACO and non-ACO hospitals.

 One reason for this discordance relates to the fact that we evaluated surgical spending among ACO beneficiaries more globally, identifying possible drivers of reduced spending that are independent of hospital ACO participation. These include reduced referrals for elective surgery, shifting surgical procedures outside the hospital setting, and improved preventive and perioperative care.

Our results suggest potential strategies that ACOs use to lower surgical spending. One possibility is that our findings result from ACOs shifting care from the inpatient to the outpatient setting for their aligned beneficiaries. Outpatient surgical care delivered in hospital outpatient departments, ambulatory surgery centers, and the physician’s office has been shown to be of comparable quality to inpatient procedural care but substantially less costly.

 We also found that, after adjusting for outpatient surgical procedure (by organ system or surgical specialty), ACO alignment was associated with slightly lower spending on outpatient surgery. This could provide further evidence that the higher observed outpatient surgical spending in ACOs is due, in part, to case mix. That is, because of a shift of some inpatient procedures to the outpatient setting, ACOs actually spend more on outpatient surgery. It is also notable that the observed inpatient episode efficiency was due, in large part, to lower postacute care payments. This likely reflects the more judicious use of expensive postoperative ancillary services. Similar results were previously demonstrated in the general Medicare population.

 Another potential explanation for these findings is residual differences in case selection by ACO providers. If ACOs select healthier or less-costly patients for inpatient surgery, our results may be somewhat overestimating the effect of ACO participation on surgical spending. However, to mitigate this possibility, we adjusted our models with variables strongly associated with surgical spending, such as age, DRG, and comorbidities.

Our study findings must be considered in the context of several limitations. First, we examined differences in Medicare spending for surgical care and found that ACO participation was associated with reduced spending. However, we did not examine whether this decreased spending had any effect on surgical quality. Overall spending offers an important and quantifiable measure of performance, but it is 1 of several measures that are important to Medicare beneficiaries, policy makers, and society at large. Existing evidence for the impact that Medicare ACOs have on surgical outcomes is mixed, and additional work is needed to more comprehensively understand the relationship between ACO policy and surgical quality. Second, we defined surgical care episodes to include 90 days after discharge from an inpatient surgery and a 3-day window around the time of an outpatient procedure. We chose these definitions because they are widely used in both research and Medicare policy and can be expected to capture the majority of spending associated with surgical care.

 Nonetheless, they may exclude some relevant spending, particularly in the time frame preceding an inpatient surgical admission. Finally, the Medicare Shared Savings Program is a voluntary program, and ACOs participating in this program may differ from physicians and hospitals that do not form ACOs. In order to address this limitation, we controlled for patient factors and used ACO fixed effects in our models to account for time invariant differences in provider groups. Nonetheless, as our sensitivity analysis on early vs late ACO model adopters shows, selection effects remain a potential confounder of the association between ACO participation and surgical spending.

These limitations notwithstanding, our analysis has important policy implications. Medicare Shared Savings Program ACOs have generated modest savings to date, which have fallen short of expectations for some. However, we demonstrate that surgical care is a potential source of savings for ACOs. By performing more outpatient and fewer inpatient procedures and by more efficiently delivering inpatient surgical care, ACOs can reduce surgical spending at a population level. Early data suggest considerable variation in the number of surgeons participating in Medicare Shared Savings Program ACOs.

 Heterogeneity in surgeon involvement in ACOs is reflected in the literature on ACO effects on surgical quality and costs. Insofar as attention to surgical care has not been the focus of ACO leaders,

 the increased engagement of surgeons and a focus on improving the value of surgical care may represent an opportunity to magnify the impact of ACO policy.

The authors would like to acknowledge Andrew M. Ryan, PhD, for his comments on earlier versions of this work and his useful suggestions.


 Dow Division of Health Services Research, Department of Urology, University of Michigan Medical School (PKM, NM, LAH, BKH, JMH), Ann Arbor, MI.

This work was supported by grants from the National Cancer Institute to Dr Modi (F32CA232332), the National Institute on Aging (NIA) to Dr Hollenbeck (R01AG048071), and the Agency for Healthcare Research and Quality (AHRQ) to Dr Hollingsworth (R01HS024525 and R01HS024728). The views expressed in this article do not represent the views of the US federal government.

 Dr Hollenbeck has received an NIA grant on accountable care organizations (ACOs). Dr Hollingsworth has received 2 grants from AHRQ (R01HS024525 and R01HS024728) on ACOs. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (PKM, NM, LAH, BKH, JMH); acquisition of data (BKH, JMH); analysis and interpretation of data (PKM, NM, LAH, BKH, JMH); drafting of the manuscript (PKM, NM, BKH); critical revision of the manuscript for important intellectual content (PKM, NM, LAH, BKH, JMH); statistical analysis (PKM, NM); obtaining funding (BKH, JMH); administrative, technical, or logistic support (BKH); and supervision (JMH).

 John M. Hollingsworth, MD, MS, Dow Division of Health Services Research, Department of Urology, University of Michigan Medical School, 2800 Plymouth Rd, Bldg 16, 1st Floor, Room 112W, Ann Arbor, MI 48109. Email: kinks@med.umich.edu.

1. Berian JR, Rosenthal RA, Baker TL, et al. Hospital standards to promote optimal surgical care of the older adult: a report from the Coalition for Quality in Geriatric Surgery. 

. 2018;267(2):280-290. doi:10.1097/SLA.0000000000002185

2. Kaye DR, Luckenbaugh AN, Oerline M, et al. Understanding the costs associated with surgical care delivery in the Medicare population. 

. 2020;271(1):23-28. doi:10.1097/SLA.0000000000003165

3. Birkmeyer JD, Gust C, Baser O, Dimick JB, Sutherland JM, Skinner JS. Medicare payments for common inpatient procedures: implications for episode-based payment bundling. 

. 2010;45(6, pt 1):1783-1795. doi:10.1111/j.1475-6773.2010.01150.x

4. Chen LM, Norton EC, Banerjee M, Regenbogen SE, Cain-Nielsen AH, Birkmeyer JD. Spending on care after surgery driven by choice of care settings instead of intensity of services. 

. 2017;36(1):83-90. doi:10.1377/hlthaff.2016.0668

5. Fisher ES, Shortell SM. Accountable care organizations: accountable for what, to whom, and how. 

. 2010;304(15):1715-1716. doi:10.1001/jama.2010.1513

6. Medicare Shared Savings Program: shared savings and losses and assignment methodology. CMS. April 2017. Accessed May 30, 2019. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/sharedsavingsprogram/Downloads/Shared-Savings-Losses-Assignment-Spec-V5.pdf

7. Hollingsworth JM, Saigal CS, Lai JC, Dunn RL, Strope SA, Hollenbeck BK; Urologic Diseases in America Project. Medicare payments for outpatient urological surgery by location of care. 

. 2012;188(6):2323-2327. doi:10.1016/j.juro.2012.08.031

8. Hollingsworth JM, Saigal CS, Lai JC, Dunn RL, Strope SA, Hollenbeck BK; Urologic Diseases in America Project. Surgical quality among Medicare beneficiaries undergoing outpatient urological surgery. 

. 2012;188(4):1274-1278. doi:10.1016/j.juro.2012.06.031

9. CPI inflation calculator. US Bureau of Labor Statistics. Accessed April 29, 2019. https://www.bls.gov/data/inflation_calculator.htm

10. Diez Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. 

. 2001;345(2):99-106. doi:10.1056/NEJM200107123450205

11. Mehta HB, Dimou F, Adhikari D, et al. Comparison of comorbidity scores in predicting surgical outcomes. 

. 2016;54(2):180-187. doi:10.1097/MLR.0000000000000465

12. Froot KA. Consistent covariance-matrix estimation with cross-sectional dependence and heteroskedasticity in financial data. 

13. The Center for the Evaluative Clinical Sciences, Dartmouth Medical School. 

14. McWilliams JM, Hatfield LA, Landon BE, Hamed P, Chernew ME. Medicare spending after 3 years of the Medicare Shared Savings Program. 

. 2018;379(12):1139-1149. doi:10.1056/NEJMsa1803388

15. Colla CH, Lewis VA, Tierney E, Muhlestein DB. Hospitals participating in ACOs tend to be large and urban, allowing access to capital and data. 

. 2016;35(3):431-439. doi:10.1377/hlthaff.2015.0919

16. Sinha SS, Moloci NM, Ryan AM, et al. The effect of Medicare accountable care organizations on early and late payments for cardiovascular disease episodes. 

. 2018;11(8):e004495. doi:10.1161/CIRCOUTCOMES.117.004495

17. Schwartz AL, Chernew ME, Landon BE, McWilliams JM. Changes in low-value services in year 1 of the Medicare Pioneer accountable care organization program. 

. 2015;175(11):1815-1825. doi:10.1001/jamainternmed.
2015.4525

18. Resnick MJ, Graves AJ, Thapa S, et al. Medicare accountable care organization enrollment and appropriateness of cancer screening. 

. 2018;178(5):648-654. doi:10.1001/jamainternmed.2017.8087

19. Busch AB, Huskamp HA, McWilliams JM. Early efforts by Medicare accountable care organizations have limited effect on mental illness care and management. 

. 2016;35(7):1247-1256. doi:10.1377/hlthaff.2015.1669

20. Modi PK, Kaufman SR, Borza T, et al. Medicare accountable care organizations and use of potentially low-value procedures. 

. 2019;26(2):227-233. doi:10.1177/1553350618816594

21. Herrel LA, Norton EC, Hawken SR, Ye Z, Hollenbeck BK, Miller DC. Early impact of Medicare accountable care organizations on cancer surgery outcomes. 

. 2016;122(17):2739-2746. doi:10.1002/cncr.30111

22. Schoenfeld AJ, Sturgeon DJ, Blucher JA, Haider AH, Kang JD. Alterations in 90-day morbidity, mortality, and readmission rates following spine surgery in Medicare Accountable Care Organizations (2009-2014). 

. 2019;19(1):8-14. doi:10.1016/j.spinee.2018.06.367

23. Nathan H, Thumma JR, Ryan AM, Dimick JB. Early impact of Medicare accountable care organizations on inpatient surgical spending. 

. 2019;269(2):191-196. doi:10.1097/SLA.0000000000002819

24. Chukmaitov AS, Menachemi N, Brown LS, Saunders C, Brooks RG. A comparative study of quality outcomes in freestanding ambulatory surgery centers and hospital-based outpatient departments: 1997-2004. 

. 2008;43(5, pt 1):1485-1504. doi:10.1111/j.1475-6773.2007.00809.x

25. Agarwal D, Werner RM. Effect of hospital and post-acute care provider participation in accountable care organizations on patient outcomes and Medicare spending. 

. 2018;53(6):5035-5056. doi:10.1111/1475-6773.13023

26. Bundled Payments for Care Improvement (BPCI) initiative: general information. CMS. Accessed January 29, 2019. https://innovation.cms.gov/initiatives/bundled-payments/

27. Hawken SR, Ryan AM, Miller DC. Surgery and Medicare Shared Savings Program accountable care organizations. 

. 2016;151(1):5-6. doi:10.1001/jamasurg.2015.2772

28. Resnick MJ, Graves AJ, Buntin MB, Richards MR, Penson DF. Surgeon participation in early accountable care organizations. 

. 2018;267(3):401-407. doi:10.1097/SLA.0000000000002233

29. Dupree JM, Patel K, Singer SJ, et al. Attention to surgeons and surgical care is largely missing from early Medicare accountable care organizations. 

. 2014;33(6):972-979. doi:10.1377/hlthaff.2013.1300

Articles

Medicare Shared Savings Program accountable care organizations spent less on surgical care by reducing inpatient surgery, increasing outpatient surgery, and reducing spending on postacute care after inpatient surgery.

Medicare Accountable Care Organizations Reduce Spending on Surgery

 To assess whether hospital participation in alternative payment models (APMs) is associated with greater engagement in health information exchange (HIE) along 4 dimensions: volume of patients for whom information is exchanged, diversity of information types, breadth of partner types, and depth of technical approach.

 Pooled, cross-sectional analysis of data on US hospitals from 2014 to 2015.

 APM participation came from Leavitt Partners data, Medicare public use files, and the American Hospital Association (AHA) Annual Survey. We used Medicare data to measure HIE volume for 798 hospitals attesting to stage 2 Meaningful Use and the AHA Information Technology Supplement to measure HIE diversity, breadth, and depth for 1730 hospitals. We used mixed-effects regression to estimate the association between participation in APMs and each dimension of HIE.

 Compared with nonparticipating hospitals, full-year APM participation was associated with lower HIE volume (data were sent for 11 percentage points fewer discharges; 

 = .003), greater HIE diversity (of 4 data types, 0.3 more were transmitted; 

 <.001), greater HIE breadth (of 3 partner types, data were sent to 0.3 more; 

 <.001), and greater HIE depth (the odds of using a push and pull approach were 1.68 times greater; 

 Our finding that APM participation was associated with greater HIE diversity, breadth, and depth suggests that value-based payment may be spurring improvements in HIE infrastructure. However, our finding that APM participation is associated with lower HIE volume suggests that there may be an incentive to focus HIE investments on a limited number of partners.

Under alternative payment models (APMs), hospitals are incentivized to improve health information exchange (HIE) engagement in order to facilitate better healthcare quality and reduce cost. However, even under value-based reimbursement, substantial challenges to improving HIE may still remain.

Hospital participation in APMs was associated with greater engagement in 3 of 4 HIE dimensions: diversity of data types, breadth of partner types, and depth of exchange approach.

APM participation was associated with lower HIE volume.

Our work suggests that under value-based reimbursement, indirect incentives may improve HIE infrastructure, but hospitals may be limiting HIE efforts to a few partner-specific connections, resulting in lower overall HIE volume.

Failures in information sharing between hospitals and postacute care providers following hospital discharge can result in higher-cost, lower-quality care.

 Electronic health information exchange (HIE) can improve the accessibility of information during hospital discharges, leading to cost savings and better outcomes.

 However, volume-based reimbursement does not create incentives for provider organizations to engage in HIE.

 Medicare’s alternative payment models (APMs) are expected to change this dynamic: By rewarding hospitals for improving the quality and cost-efficiency of care received across the care continuum, APMs create financial incentives for hospitals to engage in greater HIE.

 However, even under APMs, significant barriers, such as lack of technical standards across electronic health record (EHR) products, poor usability of HIE solutions, concerns about data security, and potential loss of profitable fee-for-service patients to competitors, may impede hospital pursuit of HIE.

If hospitals participating in APMs are not engaging in greater HIE, it is an ominous sign about the potential for HIE growth, as it suggests that even aligned financial incentives are not strong enough to overcome these barriers. It is therefore important to not only assess whether APM participation is associated with greater hospital HIE, but to do so in a way that reflects the multiple ways that HIE can generate value under APMs. Given that avoiding hospital readmissions is a core performance metric under APMs, improving HIE between hospitals and postacute care providers is likely to be a prioritized use case for HIE.

The extent of hospital HIE engagement with postacute care can be measured along 4 dimensions: volume, diversity, breadth, and depth.

 Volume refers to the proportion of discharged patients for whom data are transmitted electronically. Diversity refers to the types of data that are transmitted electronically. Breadth refers to the types of trading partners to whom data are electronically transmitted. Finally, depth refers to the nature of the technical approach through which data are transmitted (ie, push and/or pull).

Prior research on the association between APMs and HIE suggests that HIE is perceived as valuable to the achievement of financial rewards under APMs.

 However, these studies have not systematically examined whether hospitals in APMs engage in greater HIE or whether HIE efforts focus specifically on supporting care coordination following discharges.

To assess whether participation in APMs is associated with greater HIE engagement with postacute care providers along 4 dimensions (volume, diversity, breadth, and depth), we used mixed-effects regression analysis on data on US hospitals from 2014 to 2015. We defined APM participation as the proportion of the calendar year that a hospital participated in at least 1 of the following programs: Medicare or commercial accountable care organizations (ACOs), the Medicare Bundled Payments for Care Improvement (BPCI) initiative, and any type of medical home. Results from this study are critical to informing policy efforts aimed at improving interorganizational care coordination—a national policy priority. In particular, this work sheds light on challenges that may exist in reaching that goal through greater HIE under value-based incentives, such as those initiated under the Medicare Access and CHIP Reauthorization Act of 2015 (MACRA).

The study population includes all nonfederal acute care hospitals in the continental United States with HIE data for at least 1 year of the study period. Data for this study came from the Leavitt Partners ACO database (current as of 2016), 2014-2015 Medicare BPCI public use file, 2014-2015 American Hospital Association (AHA) Annual Survey, 2014-2015 Meaningful Use public use file, and 2017 Area Health Resource File.

Under MACRA, hospitals can participate in APMs that provide value-based incentives for hospitals to engage in HIE. With APM participation, hospitals are held accountable for the cost and quality of care provided by their outpatient partners. APMs include ACOs, the BPCI program, and medical homes. ACOs offer shared savings to participating hospitals if per-patient spending is lower than a targeted amount. The BPCI program bundles inpatient and outpatient services into single episodes of care. Lastly, hospitals can participate in medical homes as part of their hospital-based ambulatory care practices. Medical homes provide performance-based rewards, along with per-member-per-month fees for care management.

Independent variable: APM participation. 

We measured hospital participation in APMs by calculating the portion of the calendar year in which a hospital participated in any APM (a Medicare or commercial ACO, phase 2 of any BPCI model, and/or a medical home). ACO start dates for all ACOs formed as of 2016 were obtained from Leavitt Partners on Medicare or commercial ACO participation. Quarterly participation in BPCI was obtained from the Medicare public use file. Medical home participation was obtained from the AHA survey; because start and end dates were unavailable, we assumed full calendar year participation.

Dependent variables: 4 dimensions of HIE. 

Drawing on a framework developed by Massetti and Zmud, we measured HIE engagement along 4 dimensions: volume, diversity, breadth, and depth.

We measured volume as the proportion of discharged patients for whom summary of care records (SCRs) were sent electronically during a hospital’s stage 2 Meaningful Use attestation period. (For more on this measure and methodology, refer to Lin et al.

) Data on this measure came from the Medicare Meaningful Use public use file.

 we measured diversity as the number of data types (ie, structured SCRs, radiology results, laboratory results, and medical history) that are routinely sent electronically by a hospital to ambulatory providers outside its system.

Breadth indicates how successful a hospital is at creating and maintaining connections with a variety of partners. Following prior work,

 we measured breadth as the number of partner types (ie, out-of-system ambulatory care providers, long-term care providers, and behavioral health providers) to which a hospital routinely sends structured SCRs.

We measured depth by assessing whether or not a hospital routinely transmits SCRs using either push or pull, or both approaches. In the push approach, patient data are directly sent to another electronic system, whereas the pull approach aggregates data from multiple sources into a database that a provider can query. Hospitals that used neither push nor pull (eg, eFax, mail, or fax) were dropped from the analysis. Both push and pull approaches are useful for care coordination; however, the availability of both approaches is associated with greater provider satisfaction and higher use.

 A hospital was considered as using a push approach if it used secure messaging to routinely transmit SCRs and a pull approach if it used patient portals or Epic’s Care Everywhere HIE platform (ie, the hospital reported using Epic as their primary inpatient EHR vendor and using a third party to routinely transmit SCRs) to routinely transmit SCRs. Data on HIE diversity, breadth, and depth came from the AHA Information Technology Supplement.

To control for potential confounding, we included controls that could be associated with both HIE engagement and APM participation: hospital size, urbanicity, teaching status, system membership, network membership, ownership, disproportionate share hospital (DSH) percentage, case mix index, critical access hospital status, and market share from the AHA Annual Survey and the Medicare Impact File.

 To control for confounding that may arise from the number of trading partners, we included the number of skilled nursing facilities (SNFs), hospitals, and primary care providers (PCPs) in the county from the Area Health Resource File. Missing controls for a given year were imputed using forward, then backward, imputation. If no data were available on 1 or more control variables, the hospital was dropped.

Analytic Approach: Design and Statistical Analysis 

Using hospital-year data from 2014 to 2015, we used mixed-effects models regressing each dimension of HIE engagement on APM participation and controls; we included hospital mixed effects to control for unobserved heterogeneity over time and across hospitals, year fixed effects to control for temporal effects, and hospital-clustered standard errors to account for serial correlation. For our models of HIE volume, diversity, and breadth, we used linear regression; for our model of HIE depth, we used logistic regression.

Four sensitivity analyses were conducted: (1) To test whether our results were driven by concurrent participation in multiple APM types, we reran our analyses including the number of APM types in which a hospital participated; (2) To test whether participation in different APM types had different associations with HIE, we reran our analyses using a 3-way interaction term interacting participation in each of the 3 APM types in any given year; (3) To test whether using linear regression resulted in overestimated effect sizes, we reran the diversity and breadth models using negative binomial models; and (4) To test the sensitivity of our volume analysis to sample selection bias, we ran a Heckman model.

Of 6101 hospitals in the AHA Annual Survey Database from 2014 to 2015, we dropped 1537 noncontinental federal hospitals. Of the remaining 4564 hospitals, 2233 had data on all control variables. Of those, 798 had at least 1 year of data on HIE volume (397 had 1 year and 401 had 2 years) and were included in our sample for HIE volume analysis. There were 1730 hospitals that had at least 1 year of data on HIE diversity and breadth (550 had 1 year and 1180 had 2 years) and were included in our sample for those analyses. There were 1427 hospitals that had at least 1 year of data on HIE depth and used push and/or pull to transmit SCRs electronically (662 had 1 year and 765 had 2 years); these hospitals were included in our sample for HIE depth analysis (

Hospitals in our samples differed in important ways from hospitals that were missing HIE data (

]). Hospitals in all 3 samples were more likely to participate in APMs. They were also more likely to be nonprofit; to be in hospital networks; to be larger, urban, teaching hospitals; to have a higher case mix index and higher market share; and to be in counties with more SNFs and PCPs.

Unadjusted bivariate analyses revealed that, compared with nonparticipating hospitals, hospitals participating in APMs for any portion of the calendar year engaged in greater HIE on all dimensions except volume (

). The average value of HIE volume was 53% among nonparticipating and 48% among participating hospitals. The average value of HIE diversity was 2.4 data types among nonparticipating and 3.1 data types among participating hospitals. The average value of HIE breadth was 1.2 partner types among nonparticipating and 1.8 partner types among participating hospitals. Of the subsample of hospitals that used push and/or pull approaches, the percentage of hospitals that used both push and pull approaches was 54% among nonparticipating and 69% among participating hospitals.

After adjusting for controls and hospital-specific trends, full-year APM participation was associated with a —11 percentage point difference in HIE volume (21% fewer discharges than nonparticipating hospitals; 

 = .001), a difference of 0.31 data types in HIE diversity (13% more data types than nonparticipating hospitals; 

 <.001), a difference of 0.29 partner types in HIE breadth (24% more partner types than nonparticipating hospitals; 

 <.001), and greater HIE depth (the odds of using both push and pull approaches were 1.68 times greater in participating than nonparticipating hospitals; 

To make these results more interpretable, the adjusted predicted 2015 values for each HIE dimension for a common hospital type (medium-sized, urban, nonteaching, nonsystem, nonnetwork, nonprofit, non—critical access hospital, with DSH percentage, case mix index, and market share held at their population means) are presented in 

. The average predicted HIE volume for nonparticipating hospitals was 62% compared with 51% for hospitals participating for the full calendar year. The average predicted value for HIE diversity for nonparticipating hospitals was 2.8 data types compared with 3.1 data types for hospitals participating for the full calendar year. The adjusted predicted value for HIE breadth for nonparticipating hospitals was 1.4 partner types compared with 1.7 partner types for hospitals participating for the full calendar year. The probability of a nonparticipating hospital using both push and pull approaches versus only 1 approach was 72% compared with 80% among hospitals participating for the full calendar year.

Results from our first sensitivity analyses suggest that our main results were not driven by concurrent participation in multiple APM types (

). Results from our second sensitivity analysis suggest that the associations between APM participation types and dimensions of HIE access are similar to our main analysis (

). For ease of comparison, we used the model with 3-way interactions to estimate 95% CIs for each dimension of HIE for each pattern of APM participation. We then compared these CIs with that of our main analysis in Figure 2. Each of the CIs overlapped, suggesting that the associations between APM participation and dimensions of HIE did not significantly vary by APM participation type. Results from our third sensitivity analysis suggest that effect sizes from the analyses of HIE diversity and breadth may be slightly overestimated as a result of using linear models. However, because we were unable to run the negative binomial models using hospital mixed effects, we present the results of the negative binomial models in 

. Results from our final sensitivity analysis revealed evidence of moderate sample selection bias (athrho = —0.42; 

 ≤.001). However, because results were similar to our main results and we were unable to control for hospital mixed effects in the Heckman model, we present these results in 

Using national data on US hospitals from 2014 to 2015, we found that hospital APM participation was associated with greater HIE accessibility along 3 dimensions: diversity (number of transmitted data types), breadth (number of exchange partner types), and depth (number of technical approaches). However, APM participation had a negative association with HIE volume (percent of discharges in which an SCR was transmitted electronically). This finding suggests that challenges exist to increasing utilization of HIE for all patients under value-based reimbursement. There are several possible explanations. First, unlike other dimensions of HIE accessibility, greater HIE volume requires process and workflow changes that depend on clinician involvement. It may be that hospitals that participate in APMs increase the burden on clinicians to engage in multiple quality improvement initiatives at once, resulting in change fatigue and poorer performance on HIE volume. Change fatigue may be exacerbated by modifications to health information technology structures, which may be occurring under APMs.

 Another explanation is that, under APMs, hospitals focus HIE efforts on a few partner-specific connections, resulting in overall decreases in HIE, especially if HIE partners are not also those with whom hospitals share the highest volume of patients. On the other hand, APM-participating hospitals may be focusing HIE efforts on high-cost or complex patients to prevent readmissions. Future research should examine how the number of HIE partners changes under APMs, as well as patient-level factors that predict HIE use and the relationship to quality measures such as readmission rates.

Our finding that APM participation is associated with greater HIE diversity, breadth, and depth suggests that value-based payment reform may be spurring investments in HIE infrastructure as hospitals strive to achieve performance-based incentives through information-driven care coordination improvements for discharged patients.

 However, given the limitations of our cross-sectional design, we were unable to say with strong certainty whether these associations reflect a causal relationship. It is possible that instead of APM participation driving changes in HIE, hospitals that choose to join APMs are also those that exchange data for fewer patients and have more mature HIE systems. It will be important to continue monitoring HIE activities in APM-participating organizations in the future.

Our findings have important implications for policy makers. Prior efforts have primarily used direct incentives to promote HIE engagement through the Meaningful Use program. With the passage of the 21st Century Cures Act, efforts have shifted to developing interoperable infrastructures and reducing information blocking. This study suggests that indirect incentives, through the form of value-based payments, may be an effective driver of HIE infrastructure that can overcome competitive forces that drive information blocking, at least among the hospital population included in this study.

Our findings suggest that even under value-based reimbursement, challenges may exist to improving HIE use. We find that APM participation is associated with worse HIE volume, which may be because hospitals restrict HIE efforts to a limited number of affiliated partners. This suggests that policies should focus on promoting HIE connections among all partners, not just those with whom hospitals have a strategic reason to share information. For example, medical home accreditation programs currently require engagement in HIE, but not a specific threshold of engagement. Specifying a threshold based on patient volume may incentivize hospitals to improve overall interoperability rather than focus on partner-specific connections. Future studies should also examine whether APMs are promoting HIE connections with the most appropriate partners. Prior evidence suggests that although hospital HIE infrastructure is improving, these connections may not be serving the highest-volume partnerships.

 On the other hand, APM-participating hospitals may be targeting HIE efforts toward high-cost or complex patients to prevent readmissions. Future research should also examine patient-level factors that predict HIE use.

This study has important limitations. First, because measures of HIE were only available for 2014 to 2015, we were unable to use robust approaches to causal inference. Second, our measures of HIE dimensions do not take into account variation in other dimensions. For example, our measure of HIE breadth captures the number of partner types and not the number of partners within a single type with which a hospital shares information. It may be that hospitals only share information with a few partners of a specific type and we would not be able to detect this. Finally, because our analyses were restricted to hospitals with HIE data, hospitals in our sample differed from hospitals out of sample, limiting generalizability. Specifically, our sample for all models had a smaller proportion of smaller, rural, nonteaching, nonsystem, nonnetwork, or for-profit hospitals than hospitals out of sample (eAppendix Table 1). Therefore, our findings that APM participation is associated with lower HIE volume and higher HIE diversity, breadth, and depth may not apply to these hospitals. It will be important to assess whether our findings still hold for hospitals underrepresented in this study population.

Our study assesses the association between APM participation and HIE engagement along 4 important dimensions: volume of data exchange, diversity of data types, breadth of partner types, and depth of exchange approach. We find that APM participation is associated with lower HIE volume, but greater HIE diversity, breadth, and depth. This finding suggests that under value-based reimbursement, indirect incentives may improve HIE infrastructure, but significant challenges to achieving high HIE volume remain and may require more targeted policy actions.

School of Public Health (SCL) and Department of Urology (JMH), University of Michigan, Ann Arbor, MI; Department of Medicine, University of California, San Francisco (JA-M), San Francisco, CA.

 This study was supported by the Agency for Healthcare Research and Quality grants 1R01HS024525 01A1 and 1R01HS024728 01 (JMH).

 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.

 Concept and design (SCL, JA-M); acquisition of data (JMH); analysis and interpretation of data (SCL, JA-M); drafting of the manuscript (SCL, JA-M); critical revision of the manuscript for important intellectual content (SCL, JA-M); statistical analysis (SCL); provision of patients or study materials (JMH); administrative, technical, or logistic support (JMH); and supervision (JMH, JA-M).

 Sunny C. Lin, MS, School of Public Health, University of Michigan, 1415 Washington Heights, Ann Arbor, MI 48109. Email: sunnylin@umich.edu.

1. Moore C, Wisnivesky J, Williams S, McGinn T. Medical errors related to discontinuity of care from an inpatient to an outpatient setting. 

. 2003;18(8):646-651. doi: 10.1046/j.1525-1497.2003.20722.x.

Alternative payment models (APMs) introduce value-based incentives for greater hospital health information exchange (HIE) engagement. We find that APM participation is associated with lower HIE volume and greater HIE diversity, breadth, and depth.

Author | John M. Hollingsworth, MD, MS

Articles

New content coming soon, please check back later.