Is There An Association Between Bundled Payments and “Cherry Picking” and “Lemon Dropping” in Orthopaedic Surgery? A Systematic Review

Abstract

Background

The goal of bundled payments—lump monetary sums designed to cover the full set of services needed to provide care for a condition or medical event—is to provide a reimbursement structure that incentivizes improved value for patients. There is concern that such a payment mechanism may lead to patient screening and denying or providing orthopaedic care to patients based on the number and severity of comorbid conditions present associated with complications after surgery. Currently, however, there is no clear consensus about whether such an association exists.

Questions/purposes

In this systematic review, we asked: (1) Is the implementation of a bundled payment model associated with a change in the sociodemographic characteristics of patients undergoing an orthopaedic procedure? (2) Is the implementation of a bundled payment model associated with a change in the comorbidities and/or case-complexity characteristics of patients undergoing an orthopaedic procedure? (3) Is the implementation of a bundled payment model associated with a change in the recent use of healthcare resources characteristics of patients undergoing an orthopaedic procedure?

Methods

This systematic review was registered in PROSPERO before data collection (CRD42020189416). Our systematic review included scientific manuscripts published in MEDLINE, Embase, Web of Science, Econlit, Policyfile, and Google Scholar through March 2020. Of the 30 studies undergoing full-text review, 20 were excluded because they did not evaluate the outcome of interest (patient selection) (n = 8); were editorial, commentary, or review articles (n = 5); did not evaluate the appropriate intervention (introduction of a bundled payment program) (n = 4); or assessed the wrong patient population (not orthopaedic surgery patients) (n = 3). This led to 10 studies included in this systematic review. For each study, patient factors analyzed in the included studies were grouped into the following three categories: sociodemographics, comorbidities and/or case complexity, or recent use of healthcare resources characteristics. Next, each patient factor falling into one of these three categories was examined to evaluate for changes from before to after implementation of a bundled payment initiative. In most cases, studies utilized a difference-in-difference (DID) statistical technique to assess for changes. Determination of whether the bundled payment initiative required mandatory participation or not was also noted. Scientific quality using the Adapted Newcastle-Ottawa Scale had a median (range) score of 8 (7 to 8; highest possible score: 9), and the quality of the total body of evidence for each patient characteristic group was found to be low using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) tool. We could not assess the likelihood of publication using funnel plots because of the variation of patient factors analyzed in each study and the heterogeneity of data precluded a meta-analysis.

Results

Of the nine included studies that reported on the sociodemographic characteristics of patients selected for care, seven showed no change with the implementation of bundled payments, and two demonstrated a difference. Most notably, the studies identified a decrease in the percentage of patients undergoing an orthopaedic operative intervention who were dual-eligible (range DID estimate -0.4% [95% CI -0.75% to -0.1%]; p < 0.05 to DID estimate -1.0% [95% CI -1.7% to -0.2%]; p = 0.01), which means they qualified for both Medicare and Medicaid insurance coverage. Of the 10 included studies that reported on comorbidities and case-complexity characteristics, six reported no change in such characteristics with the implementation of bundled payments, and four studies noted differences. Most notably, one study showed a decrease in the number of treated patients with disabilities (DID estimate -0.6% [95% CI -0.97% to -0.18%]; p < 0.05) compared with before bundled payment implementation, while another demonstrated a lower number of Elixhauser comorbidities for those treated as part of a bundled payment program (before: score of 0-1 in 63.6%, 2-3 in 27.9%, > 3 in 8.5% versus after: score of 0-1 in 50.1%, 2-3 in 38.7%, > 3 in 11.2%; p = 0.033). Of the three included studies that reported on the recent use of healthcare resources of patients, one study found no difference in the use of healthcare resources with the implementation of bundled payments, and two studies did find differences. Both studies found a decrease in patients undergoing operative management who recently received care at a skilled nursing facility (range DID estimate -0.50% [95% CI -1.0% to 0.0%]; p = 0.04 to DID estimate: -0.53% [95% CI -0.96% to -0.10%]; p = 0.01), while one of the studies also found a decrease in patients undergoing operative management who recently received care at an acute care hospital (DID estimate -0.8% [95% CI -1.6% to -0.1%]; p = 0.03) or as part of home healthcare (DID estimate -1.3% [95% CI -2.0% to -0.6%]; p < 0.001).

Conclusion

In six of 10 studies in which differences in patient characteristics were detected among those undergoing operative orthopaedic intervention once a bundled payment program was initiated, the effect was found to be minimal (approximately 1% or less). However, our findings still suggest some level of adverse patient selection, potentially worsening health inequities when considered on a large scale. It is also possible that our findings reflect better care, whereby the financial incentives lead to fewer patients with a high risk of complications undergoing surgical intervention and vice versa for patients with a low risk of complications postoperatively. However, this is a fine line, and it may also be that patients with a high risk of complications postoperatively are not being offered surgery enough, while patients at low risk of complications postoperatively are being offered surgery too frequently. Evaluation of the longer-term effect of these preliminary bundled payment programs on patient selection is warranted to determine whether adverse patient selection changes over time as health systems and orthopaedic surgeons become accustomed to such reimbursement models.

Introduction

Healthcare spending is growing at an alarming rate across high-, middle-, and low-income countries [37, 49], resulting in an increased focus on value, defined as health outcomes achieved per dollar spent across the entire cycle of care for a given condition [41, 43]. One proposed element of a value-based healthcare system is transitioning away from traditional fee-for-service payments toward bundled payments [6], which are defined as lump sums designed to cover the full set of services needed for the overall care of a condition or medical event [7, 42]. Bundled payments have been proposed as a means of triggering competition based on value by motivating and rewarding providers to deliver the best outcomes at the lowest cost [42]. Initial results appear promising [1, 31, 46], with mounting evidence suggesting that bundled payments for lower-extremity joint replacement maintains or improves quality while lowering costs [1, 3, 11-13, 16, 21, 25, 26, 31, 35, 44-46]. However, in fairness, there have been studies on bundled payments that did not report any difference in episode cost [15, 40]. Additionally, it is also important to note that much of the current savings is often achieved through avoidance of postoperative admission to skilled nursing facilities and/or reduction in length of stay, for example, and the quality metrics measured are often fairly crude (such as, readmission over a given time period) and vulnerable to “gaming” by surgeons and/or health systems.

Despite early positive findings in orthopaedic surgery, many surgeons likely remain skeptical that bundled payments will truly lead to higher value care for all patients without limiting care for certain higher risk patients who may still benefit from surgery [28] or are not simply a mechanism to cut costs and services. A major concern is that physicians and hospitals will preferentially select only healthier patients with a lower risk of complications (“cherry picking”) and/or refuse to provide care to patients with a higher risk of complications (“lemon dropping”) who may still benefit from surgical intervention because current bundled payments may make it financially advantageous to do so [4, 23, 38]. Cherry picking and lemon dropping can be considered two different explanations of the same phenomenon—adverse patient selection. However, although these two terms are commonly used, we do not feel they are appropriate to use in the remainder of our manuscript, as they are dehumanizing. Orthopaedic surgeons deliver care to patients, which is more than just a commercial exchange; therefore, we will use adverse patient selection moving forward. Nonetheless, if bundled payments do lead to a decrease in care access for patients at higher risk for complications or of certain racial or gender make-up, for example, these alternative payment models could worsen health outcomes disparities and inequity for those who may need care equally as much or even more than others. Currently, however, there is no such consensus about whether such an association exists between bundled payments and patient selection.

We therefore examined whether the implementation of bundled payments for orthopaedic surgery has led to adverse patient selection by orthopaedic surgeons. In this systematic review, we asked: (1) Is the implementation of a bundled payment model associated with a change in the sociodemographic characteristics of patients undergoing an orthopaedic procedure? (2) Is the implementation of a bundled payment model associated with a change in the comorbidities and/or case-complexity characteristics of patients undergoing an orthopaedic procedure? (3) Is the implementation of a bundled payment model associated with a change in the recent use of healthcare resources of patients undergoing an orthopaedic procedure?

Materials and Methods

Search Strategy and Criteria

We registered our protocol in PROSPERO (CRD42020189416). Our systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement [33].

Database searches were performed by a medical librarian (LLP) in MEDLINE, Embase, Web of Science, Econlit, Policyfile, and Google Scholar. Keywords and controlled vocabulary related to the concepts of bundled payments and adverse patient selection were combined, and all databases were searched from their inception dates to March 2020. To ensure no relevant study was excluded, orthopaedic surgery–related terms were excluded from the search strategy and only considered during the study selection process. No date or study type limits were used (Supplemental Digital Content 1, http://links.lww.com/CORR/A558).

Studies were eligible if they met the following criteria [30]: Population: patients undergoing surgery for an orthopaedic condition; Intervention: the implementation of a bundled payment program; Control: standard or conventional healthcare payment model (for example, fee-for-service); Outcomes: comparison of patient characteristics for those undergoing orthopaedic surgical intervention before and after the implementation of the bundled payment program (PICOS). The following patient characteristics and information were considered: age, gender, race, education level, American Society of Anesthesiologists score, socioeconomic status (for example, income or area deprivation index), patient-reported outcomes, comorbidities (such as the Charlson Comorbidity Index [a weighted model of 19 medical conditions], Elixhauser Comorbidity Index [a weighted model of 31 medical conditions that has been shown to outperform the Charlson Comorbidity Index in predicting inpatient death after orthopaedic surgery] [32], obesity, diabetes [Type I or Type II], or other comorbidities or measures of comorbidities), smoking status, and other patient factors analyzed in the included studies. Studies were considered for inclusion if they were randomized trials, observational studies, case-control studies, cross-sectional studies, or other scientific manuscripts. We excluded animal studies, editorials, commentaries, review articles, and those not subjected to peer review.

Study Review Methods

Abstract and full-text study selection was performed independently by two authors (DNB, CR), with disagreements resolved by a third author (RWP). Reference lists from selected articles were hand-searched for additional relevant studies. Data were abstracted by one author (DNB) (Table 1). A second author (CR) confirmed the accuracy of the abstracted data through an independent review of the included articles.

Table 1.

Characteristics of the included studies

	Dummit et al. [12]	Navathe et al. [36]	Finkelstein et al. [15]	Bronson et al. [5]	Barnett et al. [3]	Haas et al. [21]	Plate et al. [40]	Murphy et al. [34]	Humbyrd et al. [24]	Navathe et al. [35]
Publication year	2016	2018	2018	2018	2019	2019	2019	2019	2020	2020
Journal	JAMA	JAMA	JAMA	Spine	New England Journal of Medicine	JAMA Internal Medicine	Journal of Arthroplasty	Clinical Orthopaedics and Related Research	Journal of Bone and Joint Surgery, American Volume	Health Affairs
Country	USA	USA	USA	USA	USA	USA	USA	USA	USA	USA
Type of study	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)	Observational (retrospective)
Data source/details	Medicare claims (2011 to 2015)	Medicare claims (2011 to 2015)	Medicare claims (2012 to 2014, 2016)	Single-center (institutional data) (2013 to 2014)	Medicare claims (2015 to 2017)	Medicare claims (2014 to 2018)	Single-center (institutional data) (2015 to 2017)	Medicare claims (2013 to 2016)	Medicare claims (2015 to 2016)	Medicare claims (2011 to 2016)
CMS program?	Yes, BPCI (voluntary)	Yes, BPCI (voluntary)	Yes, CJR (mandatory)	Yes, BPCI (voluntary)	Yes, CJR (mandatory)	Yes, CJR (mandatory)	Yes, CJR (mandatory)	Yes, BPCI (voluntary)	Yes, CJR (mandatory)	Yes, BPCI (voluntary)
Orthopaedic procedure	Lower extremity joint replacement	Lower extremity joint replacement	Lower extremity joint replacement	Spine fusion (except cervical)	Lower extremity joint replacement	Lower extremity joint replacement	Lower extremity joint replacement	Lower extremity joint replacement (elective THA without major comorbidity [DRG 470] only)	Lower extremity joint replacement	Lower extremity joint replacement
Sample sizea	122,277	1,717,243	131,285	868	657,439	338,422	751	371,586	76,537	288

^aAll sample sizes are the number of care episodes or patients except for Navathe et al. (2020) [35], which provides a sample size of the number of total hospitals assessed; CMS = Centers for Medicare and Medicaid Services; BPCI = Bundled Payments for Care Improvement; CJR = Comprehensive Care for Joint Replacement; DRG = Diagnosis-Related Group.

Study Selection

A total of 440 studies were identified via the database search and known records relevant to the systematic review. After duplicates were removed, 429 records remained. These records were screened, and 399 records were removed because they were irrelevant. The full text of the remaining 30 records was reviewed; of those, 20 were excluded because they did not evaluate the outcome of interest (patient selection) (n = 8); were editorial, commentary, or review articles (n = 5); did not evaluate the appropriate intervention (introduction of a bundled payment program) (n = 4); or assessed the wrong patient population (not orthopaedic surgery patients) (n = 3). Ultimately, 10 studies were included in the final analysis (Fig. 1) [3, 5, 12, 15, 21, 24, 34-36, 40].

Fig. 1.

This flowchart shows the studies that were included in the analysis. We reviewed the full-text records of 30 studies, 10 of which were ultimately included in the final analysis.

Study Characteristics

All studies were published from 2016 to 2020 and were performed in the United States (Table 1). Eight studies used Centers for Medicare and Medicaid Services (CMS) Medicare claims data, while two studies analyzed single-institution data. All studies analyzed the impact of one of two bundled payment programs implemented by the CMS: Bundled Payments for Care Improvement (BPCI) [8] or the Comprehensive Care for Joint Replacement Model (CJR) [9]. Nine studies analyzed bundled payments for lower extremity joint arthroplasty (hip or knee arthroplasty), and one study evaluated bundled payments for spinal fusions.

Most used a difference-in-difference (DID) technique, which is a quasi-experimental statistical approach that compares changes in two groups before and after an intervention (such as the introduction of a bundled payment program) that only impacts one of those groups; thus, one group is an experimental group (like hospitals that entered a bundled payment arrangement) and the other group is a control group (for example, hospitals that did not enter a bundled payment arrangement) [10]. When such analyses are conducted, the value reported is the percentage difference in the experimental group pre- to postintervention compared with the control group. For example, if the calculated DID estimate is -1.0%, this means that the intervention was associated with a 1% decrease in the characteristic being measured in the experimental group compared with the control group.

Nine of the included studies [3, 12, 15, 21, 24, 34-36, 40] evaluated changes in sociodemographic characteristics, while all 10 studies in the systematic review [3, 5, 12, 15, 21, 24, 34-36, 40] assessed comorbidities and case complexity characteristics (Table 2). In addition, three studies [12, 35, 36] examined changes in the recent use of healthcare resources characteristics following the implementation of a bundled payment initiative. Overall, four studies [15, 21, 24, 35] found no change in any patient characteristics from before to after the implementation of bundled payments (Table 3). Six studies [3, 5, 12, 34, 36, 40] found evidence that the implementation of a bundled payment program was associated with changes in patient characteristics broadly (Supplemental Digital Content 2, http://links.lww.com/CORR/A559). The heterogeneity of data precluded a meta-analysis.

Table 2.

Patient selection characteristics evaluated by study

	Dummit et al. [12]	Navathe et al. [36]	Finkelstein et al. [15]	Bronson et al. [5]	Barnett et al. [3]	Haas et al. [21]	Plate et al. [40]	Murphy et al. [34]	Humbyrd et al. [24]	Navathe et al. [35]
Sociodemographic characteristics	Age, gender, Medicaid eligibility	Age, race, gender, Medicaid eligibility, ZIP code (median income and education proxies)	Age, race, gender, disabled, Medicaid eligibility		Age, gender, race or ethnicity, urban residence, original reason for Medicare entitlement (age > 65, disability, end-stage renal disease), Medicaid eligibility	Age, gender	Age, gender	Age	Race, Medicaid eligibility	Age, race, gender, Medicaid eligibility
Comorbidities and procedure complexity characteristics	Disabled (with no end-stage renal disease), MS-DRG 470	High complexity, obesity, diabetes, diabetes with complications, coronary artery disease, congestive heart failure, alcohol use, depression, psychoses, frailty, presence of at least two comorbidities	Number of Charlson comorbidities, Elixhauser Comorbidity Index	Case complexity	Original reason for Medicare entitlement (disability or end-stage renal disease)	Presence of hip fracture episode, CMS-HCC risk score	BMI, American Society of Anesthesiologists class, Elixhauser comorbidities	Elixhauser Comorbidity Index	Charlson Comorbidity Index, current tobacco use, history of tobacco use, obesity, diabetes with complications, diabetes without complications	Elixhauser Comorbidity Index
Characteristics of recent use of healthcare resources	Acute-care hospital, emergency department admission, home health care, inpatient rehabilitation facility, skilled nursing facility, psychiatric hospital, long-term care hospital, no institutional care, no post-acute care, institutional nursing facility	Acute-care hospital, inpatient rehabilitation facility, skilled nursing facility								Acute care hospital, inpatient rehabilitation facility, skilled nursing facility

MS-DRG = Medicare Severity Diagnosis-Related Group; HCC = hierarchical condition category.

Table 3.

Assessment of whether sociodemographic, comorbidities and/or case-complexity, and/or recent use of healthcare resources characteristics change for patients receiving surgery after bundled payment implementation by study

Author	Sociodemographic		Comorbidities and/or case-complexity		Recent use of healthcare resources		Any characteristic
	Assessed?	Change appreciated?	Assessed?	Change appreciated?	Assessed?	Change appreciated?	Assessed?	Change appreciated?
Dummit et al. [12]	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes
Navathe et al. [36]	Yes	No	Yes	No	Yes	Yes	Yes	Yes
Finkelstein et al. [15]	Yes	No	Yes	No			Yes	No
Bronson et al. [5]			Yes	Yes			Yes	Yes
Barnett et al. [3]	Yes	Yes	Yes	Yes			Yes	Yes
Haas et al. [21]	Yes	No	Yes	No			Yes	No
Plate et al. [40]	Yes	No	Yes	Yes			Yes	Yes
Murphy et al. [34]	Yes	No	Yes	Yes			Yes	Yes
Humbyrd et al. [24]	Yes	No	Yes	No			Yes	No
Navathe et al. [35]	Yes	No	Yes	No	Yes	No	Yes	No

Risk of Publication Bias

Although we planned to assess the likelihood of publication bias using funnel plots, given the limited number of studies and variation in analyzed patient factors, the data were insufficient to appropriately evaluate publication bias.

Evaluation of Scientific Quality

The Adapted Newcastle-Ottawa Scale was used to assess the risk of bias [48]. The scale ranges from 0 to 9, with higher scores preferred. The initial scoring was performed by one author (DNB) and independently verified by another author (CR). The median (range) Newcastle-Ottawa Scale score was 8 (7 to 8) (Supplemental Digital Content 2, http://links.lww.com/CORR/A559).

Grading the Evidence

An overall assessment of the quality of the total body of evidence was performed for each patient selection characteristic using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) tool (http://www.gradeworkinggroup.org/) [18-20]. For each patient selection characteristic, we began our assessment assuming a high quality of evidence and then downgraded this level of evidence, if needed, based on the risk of bias, indirectness, imprecision, and inconsistencies.

All evidence on the effects of bundled payments on patient selection characteristics was downgraded directly from a high to low quality of evidence because all studies were observational. There were no major concerns about indirectness or imprecision; however, across studies on patient selection characteristics, there were some inconsistencies in results. Overall, we graded the quality of evidence for each of the three patient selection characteristics (sociodemographic, comorbidities and case complexity, and recent use of healthcare resources) as low (Supplemental Digital Content 3, http://links.lww.com/CORR/A560).

Results

Sociodemographic Characteristics

Of the nine included studies that reported on the sociodemographic characteristics of patients selected for care, seven showed no change with the implementation of bundled payments [15, 21, 24, 34-36, 40], while two demonstrated a difference [3, 12]. One notable sociodemographic characteristic evaluated was dual-eligibility (that is, eligible for both Medicare and Medicaid [insurance for the poor or underserved] insurance). Overall, six studies [3, 12, 15, 24, 35, 36] evaluated whether the implementation of a bundled payment program led to a change in the percentage of patients treated who were dual-eligible. Four studies found no such association with DID estimates; in such cases, the authors reported the 95% CIs, which were shown to cross zero, and/or they reported p values, which showed a p value > 0.05 [15, 24, 35, 36]. In contrast, one study found that hospitals participating in BPCI treated a decreased number of dual-eligible patients (DID estimate -1.0% [95% CI -1.7% to -0.2%]; p = 0.01) compared with those not participating in BPCI after the implementation of the bundled payment program [12]. In another study, fewer patients who were younger than 65 years (DID estimate -0.50% [95% CI -0.82% to -0.24%]; p < 0.05) and who were dual-eligible (DID estimate -0.4% [95% CI -0.75% to -0.10%]; p < 0.05) were treated at facilities participating in the CJR program 2 years after implementation compared with institutions not participating in the bundled payment initiative [3]. In addition, the same authors found an increased number of patients of another race or ethnicity (that is, patients who do not fall into the following categories: non-Hispanic white, non-Hispanic Black, Asian, or Hispanic) (DID estimate 0.2% [95% CI 0.0% to 0.34%]; p < 0.05) and patients who became eligible for Medicare entitlement because of their age (age older than 65 years) (DID estimate 0.6% [95% CI 0.18% to 0.97%]; p < 0.05) [3].

Comorbidities and Case-Complexity Characteristics

Of the 10 included studies that reported on comorbidities and case-complexity characteristics, six studies [12, 15, 21, 24, 35, 36] reported no change in such characteristics with the implementation of bundled payments, and four studies [3, 5, 34, 40] noted differences. In a study evaluating spine care at a single institution, the authors actually found an increase in spine fusion complexity after the introduction of BPCI (45% of 350 patients versus 23% of 518 patients; p < 0.001) [5]. Additionally, in a study using Medicare claims data, the authors reported that CJR-participating hospitals treated a decreased number of patients with disabilities (DID estimate -0.6% [95% CI -0.97% to -0.18%]; p < 0.05) [3]. In a single-center study evaluating patients undergoing THA, the authors found that after the implementation of CJR, patients at their institution who had commercial insurance had more Elixhauser comorbidities than they did before the CJR was implemented (before: score of 0-1 in 63.6% [82 of 129] of patients, 2-3 in 27.9% [36 of 129] of patients, > 3 in 8.5% [11 of 129] of patients versus after: score of 0-1 in 50.1% [175 of 349] of patients, 2-3 in 38.7% [135 of 349] of patients, > 3 in 11.2% [39 of 349] of patients; p = 0.033); there was no difference in the patient sample covered by Medicare and its bundled payment program (before: score of 0-1 in 38.4% [33 of 86] of patients, 2-3 in 40.7% [35 of 86] of patients, > 3 in 20.9% [18 of 86] of patients versus after: score of 0-1 in 39.0% [73 of 187] of patients, 2-3 in 36.9% [69 of 187] of patients, > 3 in 24.1% [45 of 187] of patients; p = 0.786) [40]. In another study assessing the BPCI program, the authors found that patients treated through the BPCI program had a lower Elixhauser Comorbidity Index score (mean Elixhauser Comorbidity Index difference 0.02 [95% CI -0.04 to -0.01]; p = 0.01) than did patients who were not treated under a BPCI bundled payment [34].

Recent Use of Healthcare Resources Characteristics

Of the three included studies that reported on the recent use of healthcare resources of patients, one study [35] found no difference in the use of healthcare resources with the implementation of bundled payments, while two others did find differences [12, 36]. In one study assessing the BPCI, hospitals participating in the BPCI treated a decreased number of patients with recent use of either acute-care hospitals (DID estimate -0.8% [95% CI -1.6% to -0.1%]; p = 0.03), home healthcare (DID estimate -1.3% [95% CI -2.0% to -0.6%]; p < 0.001), or skilled nursing facilities (DID estimate -0.5% [95% CI -1.0% to 0.0%]; p = 0.04) [12]. Additionally, they found an increased number of patients who did not recently receive institutional care (DID estimate 0.8% [95% CI 0.0% to 1.6%]; p = 0.04) after the implementation of bundled payments [12]. In another study examining the association of the BPCI with patient selection, the authors reported that BPCI-participating hospitals treated a decreased number of patients recently receiving care at a skilled nursing facility (DID estimate -0.53% [95% CI -0.96% to -0.10%]; p = 0.01) [36].

Discussion

As healthcare spending continues to grow exponentially without evidence of improved quality, value-based healthcare initiatives, such as bundled payments, have been proposed as a means of better aligning incentives to achieve improved clinical outcomes per dollar spent. Because bundled payments provide a lump sum for an episode of care for a condition or procedure, such as a total joint arthroplasty, one concern is that such a reimbursement structure encourages surgeons and hospitals/health systems to actively select patients for operative management who are less likely to have postoperative complications, while not offering surgical treatment to patients who are more likely to have postoperative complications. Although individual studies have sought to assess this proposed phenomenon, there is no consensus about whether such an association exists between bundled payments and patient selection. In the present systematic review that included 10 total studies, two [3, 12], four [3, 5, 34, 40], and two [12, 36] studies found evidence that the implementation of bundled payments led to a change in patients selected for surgery based on sociodemographic factors, comorbidities and case-complexity characteristics, and recent use of healthcare resources, respectively. Importantly, however, although statistical changes in certain patient characteristics after the implementation of bundled payment programs were appreciated in some—but not all—included studies, we must really consider whether such findings are clinically relevant. The degree to which patient selection changed after the implementation of bundled payments was found to be small (approximately 1% or less). To put the estimated impact into perspective, if 1000 patients seek a total joint arthroplasty, only 10 may see their care altered. Therefore, the clinical difference may be quite minimal, although the fact that it does exist in a measurable way suggests such an association between bundled payments and patient selection should continue to be monitored as this payment structure grows to be more routine in the United States and globally.

Limitations

The included studies in our review suffer from several limitations. First, moral hazard—when an entity has an incentive to change its behavior, but others bear the associate risk—may play a role in the structure of the CMS bundled payment programs (for example, the BPCI and CJR). Although the income of some orthopaedic surgeons depends on individual patient reimbursement, many others, including those at large medical centers, are salaried. Their incomes remain the same regardless of a fee-for-service or bundled payment arrangement, limiting the financial incentive for an individual surgeon to select only patients with low risk of postoperative complications for surgery. This may be a core explanation of limited adverse patient selection to date with the implementation of bundled payments. With the growth of bundled payments, employers may adjust compensation plans. Future analyses are recommended to determine whether adjustment in compensation plans leads to increased adverse patient selection. However, if the remuneration structure remains unaltered, we believe our findings are consistent with the level of adverse patient selection that can be expected. Next, the duration of follow-up in the included studies was limited. Future studies need to examine long-term changes in patient selection. As physicians, health systems, and other stakeholders adapt to bundled payments, the effect of such a payment structure on patient selection may become more apparent. Indeed, a rational actor—whether that be a health system or individual surgeon—may select patients based on incentives, either consciously or subconsciously. Third, for each included study, the association of bundled payments on patient selection only included patients who underwent operative intervention. Thus, patients who saw an orthopaedic surgeon but did not have surgery and patients who never even saw an orthopaedic surgeon for their musculoskeletal concern were not considered in analyses in the included studies. This could help explain the slight differences found in our systematic review, as more patients at higher risk of complications presenting to institutions involved in bundled payment arrangements may have been treated nonoperatively. However, given that many (but not all) orthopaedic surgeons are salaried, we believe such an impact is likely limited.

Another important issue is that CMS bundled payment programs are announced before their implementation. Thus, whether a program is voluntary (the BPCI) or mandatory (the CJR), stakeholders are able to prepare for the change in reimbursement method. Consequently, changes in patient selection could have occurred before the implementation of these models, so comparing immediately before and after BPCI or CJR may not truly represent fee-for-service practice patterns. However, it is hard to imagine that system-wide referral patterns were notably altered in preparation, although in the future they could be, if appropriate risk-adjustment is not made. Another limitation is that all studies included in this review used data from the United States and CMS bundled payment programs. Accordingly, the effect of bundled payments on changes in patient selection in other parts of the world remains unknown. Further, the impact of private payer bundled payment initiatives in the United Sates (or elsewhere) on patient selection is also unknown. There are ongoing bundled payment initiatives in eight countries and among private payers [47], and further analyses will hopefully address this current knowledge gap. An additional limitation is that nearly all (9 of 10) of the studies in this systematic review focused on lower extremity joint arthroplasty; thus, the generalizability of our findings to other orthopaedic conditions is unknown. However, given that lower extremity joint arthroplasty is elective in nature, our findings may be generalizable to other such procedures. Further, although we continue to assess whether these payment models are associated with certain health disparities, such as access to care, it is crucial that the sociodemographic data included in analyses are as accurate and granular as possible. For example, patient race may or may not be accurate for all patient subgroups when utilizing large databases, such as those available through Medicare [14]. Thus, although any findings using such data can help broadly guide further scholarly inquiry and move important policy discussions forward, we should continue to strive for fully accurate, appropriately granular, and complete sociodemographic data to ensure any interventions truly positively impact those they aim to assist. In the meantime, however, it is crucial to report how race data are gathered [27].

Our systematic review also has limitations. First, although our search strategy was developed with support from an experienced medical librarian, it is possible that we did not include all relevant databases and missed work from other fields (such as economics). However, we included the most commonly used databases to assess medical studies; thus, we feel the core scholarly work in this area was likely captured. Second, although two coauthors independently assessed the level of evidence, the GRADE methodology is subjective by design. Others may differ in their evaluation, but we believe our approach of assessing the included studies independently and then reconvening was as thorough as possible.

Sociodemographic Characteristics

We found low-grade evidence that the implementation of a bundled payment program was associated with a change in the sociodemographic characteristics of patients undergoing orthopaedic procedures in two of nine studies assessing the issue. In the two studies demonstrating an association, both found a decrease in the number of dual-eligible patients (that is, patients who qualified for both Medicare and Medicaid) after the implementation of a bundled payment program [3, 12]. Decreasing the number of dual-eligible patients undergoing surgery may occur because these patients often have worse outcomes after THA and TKA; therefore, these procedures are more costly [29]. It is important to consider that the percentage decrease appreciated was very small, suggesting that the difference may not be clinically relevant in daily practice at present. Nonetheless, we would argue it remains concerning because it still suggests that patients who are economically disadvantaged—many of whom fall into minority patient populations—were selected against. On the surface, it may be reasonable that such patient selection be made by surgeons to try to decrease postoperative complication as much as possible. However, the potential deleterious effect of increasing health inequities should not be ignored. Importantly, part of an orthopaedic surgeon’s job, in our opinion, is to help patients improve their health and not just operate on patients with exceptionally low risk of postoperative complications. It is also important to remember that given the data are national, it is difficult to evaluate whether this effect happened more often in certain patient settings or regions (such as urban versus rural). Currently, we have shown this association is small, but we believe it is reasonable to assume it may become a much larger issue as health systems and orthopaedic surgeons become more comfortable with this type of payment model and realize the incentives may be misaligned if they continue to treat patients at risk for worse outcomes and higher cost. This would be a rational reaction and one that should be addressed proactively in our opinion, especially because many of the sociodemographic factors are truly proxy variables for structural injustices and will likely lead to further health inequities.

Comorbidities and Case-Complexity Characteristics

We found low-grade evidence that the implementation of bundled payments was associated with a change in the comorbidities and/or case-complexity characteristics of patients undergoing orthopaedic procedures in four of 10 studies assessing the issue. Interestingly, the single-institution study evaluating spine fusion found an increase in case complexity, defined by the authors as revisions, four to eight levels of fusion, or transforaminal lumbar interbody fusion [5]. This may suggest less, not more, adverse patient selection in the setting of a bundled payment program in the short-term. The study authors state this was likely driven by changes in referral patterns, and their findings suggest that the provided cost of care was higher than the negotiated bundled payment [5]. Thus, we can assume that rational actors would then adjust to ensure financial stability by decreasing their treatment of complex patients over time; however, such patients would still require care. Therefore, we believe more appropriate risk-adjustment payment amounts will be needed because the findings in this study are not compatible with a financially sustainable clinic. More consistent with adverse patient selection, another study focused on total joint arthroplasty found an association between patients with disabilities and decreased orthopaedic surgical intervention after the implementation of bundled payments [3]. Further, one study demonstrated an association between patients with fewer comorbidities and treatment with total joint arthroplasty after the implementation of a bundled payment program [34], while another found that patients with greater comorbidities who underwent total joint arthroplasty were more likely to have commercial insurance after the implementation of a bundled payment initiative [40]. Such associations were small, indicating that the differences may not be largely clinically relevant at present. However, financial incentives without appropriate risk-adjustment may lead to even fewer patients with comorbidities receiving surgical treatment over time. Although some may consider limiting surgery in such populations appropriate clinical decision-making, prior research has demonstrated that patients with certain comorbidities, such as severe or morbid obesity, still benefit greatly from a functional and pain perspective after total joint arthroplasty [28]. We argue that our job as orthopaedic surgeons is not just to treat those without comorbidities but also to help people improve their health and care for those who may be more comorbid but would still benefit from our expertise. Therefore, we believe much greater work in determining more appropriate risk-adjustment for bundled payments is warranted to ensure appropriate compensation for treating patients across the spectrum of complication risk.

Recent Use of Healthcare Resources Characteristics

We found low-grade evidence to suggest that the implementation of a bundled payment reimbursement model was associated with a change in the recent use of high-cost healthcare resources by patients undergoing orthopaedic procedures in two of the three studies assessing the issue. Similar to the changes in patient selection for surgery based on the other characteristics assessed, the effect size here was also small and may not be clinically appreciable at present. However, given the size of the patient population undergoing total joint arthroplasty or other orthopaedic procedures, especially as the population globally ages, the effect is still likely to become ever more noticeable. This is especially true as more bundled payment programs are implemented. Further, based on the incentives to minimize cost while also maintaining excellent clinical outcomes, it makes sense that surgeons and health systems would seek to limit care, when possible, to patients who use healthcare resources at a high rate. This is especially true if the bundled payment amount is consistent across all patients. Thus, this again reaffirms the importance of appropriate risk-adjustment as more bundled payment programs are implemented to ensure equitable access to high-quality orthopaedic care.

Conclusion

In the present systematic review, we found low-grade evidence that the implementation of bundled payments leads to a small effect on patient selection when assessing changes in sociodemographic, comorbidities and case- complexity, and recent use of healthcare resources characteristics. Key findings include a decrease in dual-eligible patients undergoing operative intervention as well a decrease in patients with more comorbidities or who had recently used high-cost healthcare resources receiving surgery. Importantly, these findings were not consistent across all included studies, and even when present, calculated differences were quite small. Thus, the current impact clinically is likely minimal. However, because the use of bundled payments is expected to increase up to 17% of all medical payments in the United States by 2021 [17], it will be vital to continue to monitor the potential impact of bundled payments on patient selection to make sure any signs of growing access disparities are expeditiously addressed as the variety of medical conditions and events covered by such a payment mechanism expands. This is a ripe area for future research. Practically, one approach to systematically evaluate the impact of bundled payments on patient selection over time is through the use of continuous improvement cycles, which allow for frequent adjustments, as needed, to make improvements, and they have been shown to be valuable in improving orthopaedic surgical care in other areas [22, 39]. Further, through collaboration, including between the American Academy of Orthopaedic Surgeons and the CMS [2], bundled payment initiatives can continue to be improved to not only benefit patients but also the surgeons providing the care as well. In addition to assessing the effect of new bundled payment programs as they are introduced, further research is warranted to assess the influence of bundled payments on patient selection in other countries to see if differences are more or less pronounced than in the United States.