Medicare Accountable Care Organizations and the Adoption of New Surgical Technology

Abstract

Background:

The dissemination of new surgical technology is a major contributor to health care spending growth. Accountable care organization (ACO) policy aims to control spending while maintaining quality. As a result, ACOs provide an incentive for hospitals to selectively adopt newer procedures with high value.

Study Design:

We conducted a retrospective cohort study using a 20% sample of national Medicare claims from 2010–2015. We identified hospitals that performed one of 6 sets of procedures: abdominal aortic aneurysm repair, aortic valve replacement, carotid endarterectomy or stent, lung lobectomy, colectomy, and prostatectomy. We identified hospitals participating in a Medicare Shared Savings Program ACO and a set of matched non-ACO control hospitals. We used a difference-in-differences approach to compare the rate of surgical treatment and the use of newer surgical technology for each set of procedures in ACO and non-ACO hospitals.

Results:

We included 707 ACO-hospitals and 1770 control hospitals. ACO-hospitals performed surgery for carotid stenosis at a lower rate than non-ACO hospitals. There was no difference in the rate of surgical treatment for all other procedure sets. ACO-hospitals were less likely to use an endovascular approach for abdominal aortic aneurysm repair (85.2% v 88.2%, p<0.001) and more likely to use a minimally invasive approach for lung lobectomy (42.2% v 34.7%, p=0.004) than non-ACO hospitals. In a difference-in-differences analysis, ACO participation was not associated with any significant difference in the use of surgical care for any of the 6 procedure sets, nor with any significant difference in the use of newer surgical technology.

Conclusion:

Despite ACO policy incentives to selectively adopt newer surgical technology, ACO participation was not associated with differences in the rate of surgery or the use of newer surgical technology for 6 major surgical procedures.

Introduction

The development and dissemination of new surgical technology is an underappreciated contributor to health care expenditures in the United States. Surgical care accounts for approximately half of all Medicare spending,¹ and new services are the largest driver of growth in the volume of major procedures.² While new surgical procedures enabled by technological advancements can represent substantial improvements, it is well established that new procedures vary considerably with respect to their costs and quality.³ Nevertheless, surgical technologies are often disseminated widely without sufficient evidence of their efficacy and safety.⁴ As a result, new surgical technology is often adopted by hospitals to improve market share and generate a competitive advantage without regard for its value.^5,6

Medicare’s Shared Savings Program Accountable Care Organizations (ACOs) aim to improve the efficiency of health care by encouraging health systems to consider population health and to enhance their financial stewardship. Specifically, hospitals in the fee-for-service system are incentivized to increase their capacity to offer profitable services, without regard for value or the needs of their populations. On the other hand, decision-making regarding investments in new technology under the ACO model may be better aligned with the needs of the population, due to a hospital system’s accountability for the overall costs and quality of health care for a population.⁷ If the perception of new surgical technology as a “profit-center” is shifted by ACO participation to a “cost-center” of uncertain value, the result may be a decrease in the use of new surgical technology by ACO participating hospitals. As the value of any given surgical procedure and of a new technology are heterogeneous, this calculation may vary based on the procedure in question. In this sense, ACO participating hospitals have the opportunity to be selective about the adoption of new technology based on its value.

We evaluated the effect of Medicare Shared Savings Program ACO participation on the use of 6 major surgical procedures and the use of newer- versus older- surgical technology for these procedures with varying levels of evidence for their value. We hypothesized that hospitals participating in an ACO would perform surgery at the same rate as non-ACO hospitals, but would differ in their use of newer surgical procedures enabled by advanced technology than those not participating in an ACO. As ACO hospitals are incentivized to be more selective with regard to the use of new technology, the differences in the rate of new technology use may be mixed, depending on the incremental value a given procedure offers over the older procedure it replaces.

Methods

Data source and study population

We performed a retrospective cohort study using claims from a 20% sample of Medicare fee-for-service beneficiaries between January 1, 2010 and December 31, 2015. We included claims from the Medicare Provider Analysis and Review, Carrier (Part B), and Outpatient files. We included 6 sets of major surgical procedures that included both an “older” technology as well as a “newer” technology. These procedure sets included abdominal aortic aneurysm repair (open and endovascular), aortic valve replacement (surgical and transcatheter), carotid (endarterectomy and stent), colectomy (open and laparoscopic/robotic), lung lobectomy (open and video-assisted thoracoscopic), and prostatectomy (open and laparoscopic/robotic). We chose these procedures because they are common in the Medicare population, typically performed electively, and usually performed in a hospital setting. The newer-technology enabled procedures have varying degrees of evidence supporting their use. For example, endovascular aortic aneurysm repair^8,9 and transcatheter aortic valve replacement^10–12 are supported by high quality randomized controlled trials that have found them to be equivalent or superior to traditional approaches. Alternatively, carotid artery stenting for carotid stenosis has mixed or negative randomized trial evidence;^13,14 and robotic prostatectomy for prostate cancer has no high quality trial evidence supporting its use (Supplemental Table 1).

In order to compare treatment patterns associated with ACO participation, we first identified the ACO-participation status of hospitals based on the Medicare Shared Savings Program ACO Provider-level Research Identifiable File. Then, to determine the appropriate “denominator” of Medicare beneficiaries served by that hospital, we created empirically defined “physician-hospital networks” using established methodology.¹⁵ Briefly, by first assigning Medicare beneficiaries to their predominant ambulatory care physician and then associating physicians with the acute care hospital where they provide the plurality of inpatient care, we were able to associate Medicare beneficiaries with an acute care hospital even if they did not have an inpatient admission.

We included beneficiaries with continuous enrollment in Medicare Parts A and B for at least 1 year prior to the index date to facilitate risk adjustment. We excluded patients younger than 66 years and those in Medicare Advantage plans to ensure availability of complete claims data. We used Current Procedural Terminology codes to identify all patients who underwent an included procedure from 2010 through 2015 (Supplemental Table 2). For each beneficiary, we identified an “index date.” For each beneficiary who underwent surgery, the index date was the date of surgery. For those who did not undergo surgery, index dates were assigned randomly to proportionately match the dates of surgery among those undergoing surgery.

Analysis

We assessed differences in patient characteristics among those managed at an ACO and non-ACO hospital before and after ACO policy implementation using chi-squared tests. For each beneficiary, we identified age, sex (except for prostatectomy), self-reported race, socioeconomic status at the zip code level,^16,17 and comorbidity level using the Klabunde modification of the Charlson comorbidity index.¹⁸

In order to evaluate the effect of ACO policy on the rate of surgery and the rate of newer-technology use, we performed a difference-in-differences analysis.¹⁹ This method allows us to estimate the change in rate of surgery before and after ACO-policy initiation, as well as quantify this difference in hospitals affiliated with an ACO and those not participating in an ACO. Furthermore, this approach allows us to control for contemporaneous secular trends in the rate of surgical procedures. Because ACO-participating hospitals are systematically different from non-ACO hospitals, we performed propensity score matching to identify a set of non-ACO control hospitals for each procedure studied. We performed variable ratio propensity score matching to identify up to 4 control hospitals for each ACO-hospital, based on teaching hospital status and volume of procedures performed in 2010–2011 (the years prior to ACO policy implementation). After matching, these hospitals had similar trends for the use of surgery and newer-surgical procedures in the years prior to ACO implementation.

For each beneficiary associated with an ACO or control hospital, we then specified a binary time variable indicating whether the “index date” took place before or after the start of ACO participation. Because ACO implementation took place at multiple time points during our study period, we assigned an ACO start date for each patient attributed to an ACO-participating hospital based on the specific ACO contract. For non-ACO control hospitals, we randomly assigned start dates in same proportions as the matched ACO hospitals.

We then used generalized estimating equation regression models to calculate the rate of each study procedure for beneficiaries at ACO and non-ACO hospitals in the time period before and after ACO implementation. We adjusted each model for age, sex, race, socioeconomic class, and comorbidity score. All models used an exchangeable correlation structure based on the hospital as the repeating unit to account for hospital-level clustering. The models also included an interaction term for hospital ACO participation and time period, which allowed us to estimate the rate of surgery for ACO and non-ACO hospitals in the time periods before and after ACO policy implementation. We then used similar models to estimate the use of newer versus older surgical technology for these procedures.

This study used deidentified claims data and was therefore deemed exempt from IRB review. All analysis was performed using STATA version 15 (College Station, TX).

Results

We identified 708 ACO hospitals (707 for abdominal aortic aneurysm repair) which were matched with 1770 control non-ACO hospitals (1768 for abdominal aortic aneurysm repair). Table 1 represents the total number of beneficiaries in the population of patients attributed to ACO and matched non-ACO control groups for each set of procedures. While there were minimal differences in these populations, ACO hospitals served fewer beneficiaries with low socioeconomic status.

Table 1.

Population characteristics of all beneficiaries attributed to ACO and non-ACO hospitals in cohort.

	ACO beneficiaries	Matched Non-ACO Controls
Abdominal aortic aneurysm repair	1,240,963	2,405,221
Age (median, IQR)	73 (12)	73 (12)
Sex (% Male)	540946 (43.6)	1058868 (44.0)
Race
White	1119438 (90.2)	2155942 (89.6)
Black	76011 (6.1)	146469 (6.1)
Other	45514 (3.7)	102810 (4.3)
Socioeconomic status (% lowest tertile)	350974 (28.3)	831939 (34.6)
Charlson Comorbidity Index
0	628867 (50.7)	1211471 (50.4)
1	264080 (21.3)	515360 (21.4)
2	167878 (13.5)	325506 (13.5)
3+	180138 (14.5)	352884 (14.7)
Aortic valve replacement	1,230,076	2,374,741
Age (median, IQR)	73 (13)	73 (13)
Sex (% Male)	536756 (43.6)	1047200 (44.1)
Race
White	1109381 (90.2)	2127496 (89.6)
Black	75392 (6.1)	145896 (6.1)
Other	45303 (3.7)	101349 (4.3)
Socioeconomic status (% lowest tertile)	357702 (29.1)	812135 (34.2)
Charlson Comorbidity Index
0	614519 (50.0)	1178603 (49.6)
1	260757 (21.2)	505163 (21.3)
2	167581 (13.6)	325374 (13.7)
3+	187219 (15.2)	365601 (15.4)
Carotid endarterectomy / stent	1,243,106	2,457,790
Age (median, IQR)	73 (12)	73 (12)
Sex (% Male)	542064 (43.6)	1082921 (44.1)
Race
White	1121255 (90.2)	2193112 (89.2)
Black	76271 (6.1)	156459 (6.4)
Other	45580 (3.7)	108219 (4.4)
Socioeconomic status (% lowest tertile)	350554 (28.2)	849255 (34.6)
Charlson Comorbidity Index
0	630158 (50.7)	1234616 (50.2)
1	264164 (21.3)	525429 (21.4)
2	168399 (13.5)	333762 (13.6)
3+	180385 (14.5)	363983 (14.8)
Colectomy	1,241,832	2,611,046
Age (median, IQR)	73 (12)	73 (12)
Sex (% Male)	541555 (43.6)	1150759 (44.1)
Race
White	1120344 (90.2)	2340406 (89.6)
Black	75859 (6.1)	157842 (6.0)
Other	45629 (3.7)	112798 (4.3)
Socioeconomic status (% lowest tertile)	354906 (28.6)	894555 (34.3)
Charlson Comorbidity Index
0	627303 (50.5)	1311157 (50.2)
1	263928 (21.3)	557637 (21.4)
2	168521 (13.6)	355146 (13.6)
3+	182080 (14.7)	387106 (14.8)
Lung lobectomy	1,240,478	2,273,976
Age (median, IQR)	73 (12)	73 (12)
Sex (% Male)	541291 (43.6)	1002078 (44.1)
Race
White	1119102 (90.2)	2036456 (89.6)
Black	75785 (6.1)	141636 (6.2)
Other	45591 (3.7)	95884 (4.2)
Socioeconomic status (% lowest tertile)	357618 (28.8)	783020 (34.4)
Charlson Comorbidity Index
0	628370 (50.7)	1147183 (50.4)
1	264417 (21.3)	484497 (21.3)
2	167430 (13.5)	308014 (13.5)
3+	180261 (14.5)	334282 (14.7)
Prostatectomy	1,239,126	2,375,642
Age (median, IQR)	73 (12)	73 (12)
Sex (% Male)	540612 (43.6)	1046454 (44.0)
Race
White	1117961 (90.2)	2126532 (89.5)
Black	75681 (6.1)	149663 (6.3)
Other	45484 (3.7)	99447 (4.2)
Socioeconomic status (% lowest tertile)	357095 (28.8)	816288 (34.4)
Charlson Comorbidity Index
0	628100 (50.7)	1202674 (50.6)
1	263073 (21.2)	506919 (21.3)
2	168352 (13.6)	321509 (13.5)
3+	179601 (14.5)	344540 (14.5)

ACO and matched non-ACO hospitals offered surgical treatment at similar rates for each set of procedures, except carotid endarterectomy/stent (Table 2). For carotid disease, ACO hospitals performed surgical treatment at a slightly lower rate, both before (9.3 per 1,000 beneficiaries in ACO hospitals versus 10 per 1,000 beneficiaries in non-ACO hospitals, p=0.009) and after (9 per 1,000 versus 9.6 per 1,000, p=0.027) ACO policy implementation. For each of the 6 procedure groups examined, ACO participation had no significant association with surgical treatment rate, when accounting for background trends using a difference-in-differences approach (Figure 1).

Table 2.

Rate of surgery use in ACO and non-ACO hospitals, pre- and post- ACO policy implementation. AAA = Abdominal aortic aneurysm; AVR = aortic valve replacement

		Pre-ACO	Post-ACO	p-value (pre v post)
AAA, n=3,646,184	Non-ACO (rate per 1,000)	4.5	4.4	0.438
	ACO (rate per 1,000)	4.4	4.4	0.831
	p-value (ACO v non-ACO)	0.268	0.814
		Pre-ACO	Post-ACO	p-value (pre v post)
AVR, n=3,604,817	Non-ACO (rate per 1,000)	4	5	<0.001
	ACO (rate per 1,000)	4.1	5.2	<0.001
	p-value (ACO v non-ACO)	0.517	0.399
		Pre-ACO	Post-ACO	p-value (pre v post)
Carotid, n=3,700,896	Non-ACO (rate per 1,000)	9.97	9.6	0.03
	ACO (rate per 1,000)	9.34	9	0.114
	p-value (ACO v non-ACO)	0.009	0.027
		Pre-ACO	Post-ACO	p-value (pre v post)
Colectomy, n=3,852,878	Non-ACO (rate per 1,000)	14.2	13.7	0.008
	ACO (rate per 1,000)	14.3	13.8	0.031
	p-value (ACO v non-ACO)	0.563	0.841
		Pre-ACO	Post-ACO	p-value (pre v post)
Lung lobectomy, n=3,514,454	Non-ACO (rate per 1,000)	2.97	2.68	<0.001
	ACO (rate per 1,000)	2.97	2.63	0.01
	p-value (ACO v non-ACO)	0.985	0.671
		Pre-ACO	Post-ACO	p-value (pre v post)
Prostatectomy, n=3,614,768	Non-ACO (rate per 1,000)	4.12	2.91	<0.001
	ACO (rate per 1,000)	3.86	2.69	<0.001
	p-value (ACO v non-ACO)	0.245	0.173

Figure 1 (a–f).

Adjusted rates of surgical treatment for each procedure set. Despite changes over time in the rate of surgical treatment, ACO-participation had no significant independent effect on the use of surgical treatment in any of these 6 procedure cohorts. Gold = ACO-hospitals, Blue = non-ACO hospitals

The use of newer surgical technology was also similar among ACO and non-ACO hospitals (Table 3). Among beneficiaries undergoing surgical treatment for AAA, ACO hospitals were less likely than non-ACO hospitals to use an endovascular approach in both pre-ACO and post-ACO policy implementation time periods. Prior to ACO implementation, ACO hospitals were more likely to use minimally invasive techniques for patients undergoing lung lobectomy (42.2% versus 34.7%, p=0.0004). However, after accounting for background trends in a difference-in-differences analysis, ACO alignment was not significantly associated with the use of new surgical technology for any of the 6 procedures examined (Figure 2).

Table 3.

Proportion of newer-technology use by ACO participation status, pre- and post- ACO policy implementation. AAA = Abdominal aortic aneurysm; AVR = aortic valve replacement

		Pre-ACO	Post-ACO	p-value (pre v post)
AAA, n=16,183	Non-ACO	0.882	0.927	<0.001
	ACO	0.852	0.904	<0.001
	p-value (ACO v non-ACO)	<0.001	0.013
		Pre-ACO	Post-ACO	p-value (pre v post)
AVR, n=16,169	Non-ACO	0.242	0.472	<0.001
	ACO	0.262	0.475	<0.001
	p-value (ACO v non-ACO)	0.117	0.83
		Pre-ACO	Post-ACO	p-value (pre v post)
Carotid, n=35,677	Non-ACO	0.133	0.124	0.185
	ACO	0.124	0.118	0.394
	p-value (ACO v non-ACO)	0.327	0.572
		Pre-ACO	Post-ACO	p-value (pre v post)
Colectomy, n=54,211	Non-ACO	0.376	0.435	<0.001
	ACO	0.375	0.447	<0.001
	p-value (ACO v non-ACO)	0.947	0.318
		Pre-ACO	Post-ACO	p-value (pre v post)
Lung lobectomy, n=10,029	Non-ACO	0.347	0.464	<0.001
	ACO	0.422	0.486	0.009
	p-value (ACO v non-ACO)	0.004	0.442
		Pre-ACO	Post-ACO	p-value (pre v post)
Prostatectomy, n=12,615	Non-ACO	0.725	0.791	<0.001
	ACO	0.734	0.804	<0.001
	p-value (ACO v non-ACO)	0.642	0.487

Figure 2 (a–f).

Adjusted proportion of surgical treatment using newer surgical technology for each procedure set. ACO-participation was not associated with a significant change in the use of newer surgical technology for these 6 procedure cohorts. Gold = ACO-hospitals, Blue = non-ACO hospitals

Discussion

This study has two main findings. First, ACO participation was not associated with any significant change in the use of surgical treatment for the included conditions. We hypothesized that ACO participation would not influence the use of surgical care and these results provide some reassurance that ACOs are not limiting the use of surgery to reduce costs. Second, ACO participation was not associated with any significant difference in the rate of newer surgical technology for these procedures. While we hypothesized that ACO hospitals may be more sensitive to the value of newer surgical technology than non-ACO hospitals, our results suggest that ACO and non-ACO hospitals are approaching the adoption of these newer procedures at similar rates.

It is well established that the payment model for health care services can influence the adoption of an expensive new technology.^20–22 In general, a payment model that incentivizes volume of procedures and sets reimbursement based on costs (e.g. Medicare’s traditional fee-for-service model) can make the adoption of new technology profitable by increasing reimbursement and improving a hospital’s market share and thus volume of procedures.⁶ On the other hand, a more restrictive payment model, such as a capitated model (e.g. health maintenance organizations), may limit the adoption of costly new technologies as hospitals in these payment models are incentivized to decrease volume.²⁰

Prior studies have demonstrated that, despite its major contribution to the overall costs of health care, ACOs have not prioritized surgical care.²³ Instead, the focus of early ACOs has been on improving care coordination, management of chronic conditions, and limiting readmissions and emergency department use.²³ Further, surgeon participation in ACOs has been low, with less than a quarter of US surgeons participating in an ACO in 2015.²⁴ Accordingly, the ability of ACOs to influence the quality and spending of surgical care is not well established. Nathan et al demonstrated that, for patients undergoing one of 6 major inpatient surgical procedures, ACO participation was not associated with lower surgical episode spending or improved clinical outcomes.²⁵

Our study suggests that, despite incentives to reduce costs and be more selective with regard to the adoption of surgical technology, ACO participating hospitals have not changed their use of surgery or the adoption of newer surgical technology as compared to non-ACO hospitals. There are several potential explanations for these findings. First, the incentive to reduce spending may not be large enough to overcome the incentive to expand surgical care and the use of expensive new technology. Indeed, most ACOs in the Medicare Shared Savings Program are in a 1-sided risk model, in which they may see some benefits but do not incur any penalties for high spending.²⁶ Therefore, the incentives inherent in the fee-for-service payment model may be considerably stronger than the incentive to reduce spending, even for ACO participants. Further, hospitals may not know the proportion of their surgical patients who will ultimately be attributed to their ACO. Taken together, these factors may limit the ability of the ACO incentive for cost cutting to drive major changes in capital investments in new surgical technology. This fits with existing data suggesting that hospital-integrated ACOs have not achieved significant spending reductions as compared to physician led ACOs without a hospital.²⁷ Second, newer technology for surgical care may be of high value, despite its costs. For example, transcatheter aortic valve replacement is more expensive, but more cost-effective, than traditional surgical aortic valve replacement for the treatment of aortic stenosis.²⁸ Therefore, it may be a worthwhile investment for ACO hospitals, despite high up-front costs. Finally, ACO hospitals may aim to reduce overall spending while maintaining hospital revenues by limiting expenditures outside of the hospital, such as for post-acute care services.²⁹ As a result, newer technology enabled surgical care (i.e. minimally invasive surgery) may allow for faster recovery and less need for post-acute care.^30,31

These results must be considered in the context of this study’s limitations. First, the adoption of some of the newer surgical technologies may have occurred well before the initiation of an ACO contract. To account for this, we used several different technologies with varying eras of adoption. Further, our analysis was designed to also identify de-adoption and could identify a reduction in the use of a newer technology if it occurred. Second, our study examines early ACOs in the first 3 years of initiation of the Medicare Shared Savings Program. Subsequent analyses may be able to identify changes in hospital behavior after more experience with the ACO program. Finally, our study did not examine spending or clinical outcomes related to the quality of surgical care provided by hospitals. Improving the quality of care is an important goal of ACOs, but was outside of the scope of this study. This question has been addressed by a number of other studies.^25,32,33

Conclusions

New surgical procedures are a significant contributor to health care spending. While ACO policy has demonstrated modest success in controlling overall health care costs, it does not appear to have influenced the use of major surgery nor the adoption of 6 newer surgical procedures. Additional attention to surgical care or other policy mechanisms for limiting spending growth may be needed to reduce spending on new surgical technology.