Karen E. Joynt Maddox

Concentration of Potentially Preventable Spending Among High-Cost Medicare Subpopulations: An Observational Study

To increase the efficiency of the U.S. health care system, policymakers from both sides of the political aisle have supported the development and implementation of alternative payment models, such as accountable care organizations and patient-centered medical homes. In addition, under the Medicare Access and CHIP Reauthorization Act of 2015 (MACRA), physicians will have even stronger incentives to join alternative payment models. A key feature of these models is the notion that if health care providers assume greater responsibility for patients health care costs and clinical outcomes, they will make better clinical decisions that may lead to more efficient and effective care. For clinical leaders, finding areas of care where money can be saved and quality improved is an imperative but has often proved difficult. One approach that has received substantial attention recently has been focusing on high-need, high-cost patients (1, 2). Although definitions of this population vary, it is commonly considered to be the 10% of beneficiaries who account for the most Medicare spending (3). These patients often have several chronic conditions with physical and cognitive limitations and may struggle to care for themselves independently (2). A key advancement of recent work has been to recognize that these high-need, high-cost patients are not a monolithic population and thus, for more effective targeting, should be characterized into subpopulations (15). By definition, this population is very expensive to care for (6); however, we know less about whether their spending is preventable. Although prior work suggests that the proportion of preventable spending among high-need, high-cost patients may be small (3), it did not account for much of the downstream spending generated by the initial preventable event. Furthermore, given the heterogeneity of the high-need, high-cost population, preventable spending likely varies substantially among subgroups. Better understanding where that spending is concentrated and how clinical leaders might target their efforts would be immensely helpful, yet we lack the empirical data to guide our efforts. Therefore, in this study, we sought to answer 3 questions. We used validated algorithms to identify preventable spending to determine what proportion of total spending may be potentially preventable across clinically distinct subpopulations of high-cost Medicare beneficiaries and how it compares with that of nonhigh-cost beneficiaries. We sought to define how the amount and type of potentially preventable spending differ by care setting (preventable emergency department [ED] visits, inpatient hospitalizations, and postacute care use) among these high-cost subpopulations. Finally, because prior work suggests that most preventable spending comprises hospitalizations for ambulatory caresensitive conditions (ACSCs), we attempted to clarify the specific types of ACSC hospitalizations that drive potentially preventable spending in each Medicare subpopulation. Methods Data We obtained and merged Medicare claims files from 2011 and 2012, including the Medicare beneficiary denominator file (demographic and enrollment data); 20% of the carrier file (claims submitted by providers for inpatient and outpatient services); the inpatient file (facility claims for inpatient hospitalizations); 20% of the outpatient file (for example, facility claims for outpatient visits, testing, and surgeries); 20% of the skilled-nursing facility, home health agency, hospice, and durable medical equipment files (facility claims for each of these settings); the Part D file (pharmaceutical claims); and the impact file (wage index and other hospital payment information). All Medicare patients in the denominator database were considered for inclusion in the study. We excluded those with Medicare Advantage coverage for any portion of the study period, who were not continuously enrolled in Part A or B for 12 months, and who lacked a valid beneficiary identification number or sex designation. We excluded beneficiaries who died during the study period because they could not contribute 12 months of costs. They were also excluded because assessing preventability of end-of-life costs was beyond the scope of this analysis, so those data might bias us toward overestimating preventable spending. Our final sample consisted of 6112450 beneficiaries. We calculated costs for each patient in 2012 using methods described by the Centers for Medicare & Medicaid Services (7). We focused on standardized costs, in which each type of service is assigned a cost based on national Medicare rates; this allowed us to examine patterns of use across geographies independent of differences in price. We then classified high-cost patients as those in the highest 10% of spending in 2012. Segmenting the Medicare Population Using a previously described claims-based algorithm based on expert opinion (3) and informed by the Bridges to Health taxonomy proposed by Lynn and colleagues (5), we categorized Medicare beneficiaries into the following 6 mutually exclusive subpopulations: nonelderly disabled, frail elderly, major complex chronic illness, minor complex chronic illness, simple chronic illness, and relatively healthy (those with no chronic conditions). This segmentation strategy was developed with input from a multidisciplinary team of clinicians in general medicine, cardiology, surgery, and emergency medicine and from experts at a 3-part workshop series focused on high-need patients convened by the National Academy of Medicine (810). The Appendix and Appendix Figure provide further details on this approach. Appendix Figure. Population segments. Segments were assigned in a waterfall fashion in the order shown, such that the groups are mutually exclusive. First, beneficiaries aged <65 y were assigned to a group (nonelderly disabled). Then, of remaining beneficiaries, those with >2 frailty indicators were assigned to a group (frail elderly). Then, on the basis of the number of chronic conditions present, the remaining beneficiaries were divided into the 4 remaining groups shown. ESRD = end-stage renal disease. Identifying Preventable Hospitalizations and ED Visits We examined potentially preventable hospitalizations, ED visits, and associated costs by subpopulation. To identify potentially preventable hospitalizations, we used the Agency for Healthcare Research and Quality Prevention Quality Indicators software (11). This algorithm defines potentially preventable hospitalizations related to specific conditions, such as heart failure, diabetes, hypertension, and asthma, for which good outpatient care can likely prevent the need for hospitalization. The algorithm has been validated and used in prior work on the Medicare population. A full list of the potentially preventable hospitalization diagnoses and their associated International Classification of Diseases, Ninth Revision, codes is in Appendix Table 1. Appendix Table 1. Agency for Healthcare Research and Quality PQIs To identify potentially preventable ED visits, we used an algorithm created by Billings and colleagues (12, 13). This algorithm, which has been validated and used in prior published work (6), employs diagnosis codes to separate ED visits into the following 4 categories: nonemergent; emergent but primary caretreatable; emergent, ED care needed, but preventable; and emergent, ED care needed, and not preventable. Similar to prior work, we classified as potentially preventable the nonemergent; emergent but primary caretreatable; and emergent, ED care needed, but preventable ED visits. Because Medicare data combine ED costs with inpatient costs if a patient is hospitalized, we limited our sample of independent ED visits to those not leading to an admission. Statistical Analysis We categorized spending for 30 days after an admission or ED visit for ACSCs into the following major categories: inpatient care, ambulatory care, rehabilitative or long-term care, hospice care, skilled-nursing facilities, home health, physician services and tests, and durable medical equipment. We first compared demographics and comorbid conditions across the 6 subpopulations. We then calculated the amount and proportion of potentially preventable spending incurred overall and for each high-cost and nonhigh-cost subpopulation. Variations in spending patterns were then examined by setting (for example, inpatient, outpatient, and postacute) across each group. Finally, we examined spending for individual ACSCs across the subpopulations. All analyses were done using SAS software (SAS Institute). This study was approved by the Harvard School of Public Health Office of Human Research Administration. Role of the Funding Source This work was funded and supported by The Commonwealth Fund, which had no role in the studys design, conduct, or reporting. Results Patient Characteristics Our sample included 6112450 Medicare beneficiaries, all of whom were assigned to 1 of the 6 subpopulations. Using our claims-based algorithm, we assigned: 1093542 patients (17.9%) to the nonelderly disabled group, 522960 (8.6%) to the frail elderly group, 1102200 (18.0%) to the major complex chronic group, 1697784 (27.8%) to the minor complex chronic group, 1101447 (18.0%) to the simple chronic group, and 594517 (9.7%) to the relatively healthy group (Appendix Table 2. Appendix Table 2. Patient Characteristics of the 6 Medicare Subpopulations The top 10% of spenders, or 611240 beneficiaries, were designated as high-cost. The proportion of patients designated as high-cost was high in the frail elderly (46.2%), nonelderly disabled (14.3%), and major complex chronic groups (11.1%). It was low for the minor complex chronic (3.7%) and simple chronic (2.0%) groups, although it was still higher than for the relatively healthy group (1.1%). High-cost patients in each subpopulation were more likely to be dually eligible and have higher rates of chronic medical conditio

Association of Practice-Level Social and Medical Risk With Performance in the Medicare Physician Value-Based Payment Modifier Program

Importance Medicare recently launched the Physician Value-Based Payment Modifier (PVBM) Program, a mandatory pay-for-performance program for physician practices. Little is known about performance by practices that serve socially or medically high-risk patients. Objective To compare performance in the PVBM Program by practice characteristics. Design, Setting, and Participants Cross-sectional observational study using PVBM Program data for payments made in 2015 based on performance of large US physician practices caring for fee-for-service Medicare beneficiaries in 2013. Exposures High social risk (defined as practices in the top quartile of proportion of patients dually eligible for Medicare and Medicaid) and high medical risk (defined as practices in the top quartile of mean Hierarchical Condition Category risk score among fee-for-service beneficiaries). Main Outcomes and Measures Quality and cost z scores based on a composite of individual measures. Higher z scores reflect better performance on quality; lower scores, better performance on costs. Results Among 899 physician practices with 5u2009189u2009880 beneficiaries, 547 practices were categorized as low risk (neither high social nor high medical risk) (mean, 7909 beneficiaries; mean, 320 clinicians), 128 were high medical risk only (mean, 3675 beneficiaries; mean, 370 clinicians), 102 were high social risk only (mean, 1635 beneficiaries; mean, 284 clinicians), and 122 were high medical and social risk (mean, 1858 beneficiaries; mean, 269 clinicians). Practices categorized as low risk performed the best on the composite quality score (z score, 0.18 [95% CI, 0.09 to 0.28]) compared with each of the practices categorized as high risk (high medical risk only: z score, −0.55 [95% CI, −0.77 to −0.32]; high social risk only: z score, −0.86 [95% CI, −1.17 to −0.54]; and high medical and social risk: −0.78 [95% CI, −1.04 to −0.51]) (Pu2009<u2009.001 across groups). Practices categorized as high social risk only performed the best on the composite cost score (z score, −0.52 [95% CI, −0.71 to −0.33]), low risk had the next best cost score (z score, −0.18 [95% CI, −0.25 to −0.10]), then high medical and social risk (z score, 0.40 [95% CI, 0.23 to 0.57]), and then high medical risk only (z score, 0.82 [95% CI, 0.65 to 0.99]) (Pu2009<u2009.001 across groups). Total per capita costs were

Effect of a Hospital-wide Measure on the Readmissions Reduction Program

9506 for practices categorized as low risk,

Participation and Dropout in the Bundled Payments for Care Improvement Initiative

13u2009683 for high medical risk only,

30-Day Episode Payments and Heart Failure Outcomes Among Medicare Beneficiaries

8214 for high social risk only, and

Comparison of Accessibility, Cost, and Quality of Elective Coronary Revascularization Between Veterans Affairs and Community Care Hospitals

11u2009692 for high medical and social risk. These patterns were associated with fewer bonuses and more penalties for high-risk practices. Conclusions and Relevance During the first year of the Medicare Physician Value-Based Payment Modifier Program, physician practices that served more socially high-risk patients had lower quality and lower costs, and practices that served more medically high-risk patients had lower quality and higher costs.

Dually Enrolled Beneficiaries Have Higher Episode Costs On The Medicare Spending Per Beneficiary Measure

Background The Hospital Readmissions Reduction Program penalizes hospitals that have high 30‐day readmission rates across specific conditions. There is support for changing to a hospital‐wide readmission measure to broaden hospital eligibility and provide incentives for improvement across more conditions. Methods We used Medicare claims from 2011 through 2013 to evaluate the number of hospitals that were eligible for penalties, in that they met a volume threshold of 25 admissions over a 3‐year period for a specific condition or 25 admissions over a 1‐year period for the cohorts included in the hospital‐wide measure. We estimated the expected effects that changing from the condition‐specific readmission measures to a hospital‐wide measure would have on average penalties for safety‐net hospitals (i.e., hospitals that treat a large proportion of low‐income patients) and other hospitals. Results Our sample included 6,807,899 admissions for the hospital‐wide measure and 4,392,658 admissions for the condition‐specific measures. Of 3443 hospitals, 688 were considered to be safety‐net hospitals. Changing to the hospital‐wide measure would result in 76 more hospitals being eligible to receive penalties. The hospital‐wide measure would increase penalties (mean [±SE] Medicare payment reductions across all hospitals) from 0.42±0.01% to 0.89±0.01% of Medicare base diagnosis‐related‐group payments. It would also increase the disparity in penalties between safety‐net hospitals and other hospitals from ‐0.03±0.02 to 0.41±0.06 percentage points. Conclusions A transition to a hospital‐wide readmission measure would only modestly increase the number of hospitals eligible for penalties and would substantially increase the penalties for safety‐net hospitals.

Age trends in 30 day hospital readmissions: US national retrospective analysis

Participation and Dropout in the Bundled Payments for Care Improvement Initiative The Centers for Medicare & Medicaid Innovation launched the Bundled Payments for Care Improvement (BPCI) Initiative in 2013,1 a voluntary alternative payment model that holds participating hospitals, practices, or facilities accountable for quality and costs in 30-, 60-, or 90-day episodes of care. Participants can join for as many or as few of 48 eligible conditions as they wish and drop out without penalty. If cost targets are achieved, participants keep a portion of the savings; if cost targets are exceeded, participants reimburse Medicare a portion of the difference. To our knowledge, no published data characterize participation or dropout from the risk-bearing phase of this program. Such information is important because voluntary alternative payment models are a key element in Medicare’s strategy to improve quality and costs of care.2

Elements of Program Design in Medicare’s Value-based and Alternative Payment Models: a Narrative Review

OBJECTIVESnThe purpose of this study was to examine the association of 30-day payments for an episode of heartxa0failure (HF) care at the hospital level with patient outcomes.nnnBACKGROUNDnThere is increased focus among policymakers on improving value for HF care, given its rising prevalencexa0and associated financial burden in the United States; however, little is known about the relationship between payments and mortality for a 30-day episode of HF care.nnnMETHODSnUsing Medicare claims data for all fee-for-service beneficiaries hospitalized for HF between July 1, 2011, and Junexa030, 2014, we examined the association between 30-day Medicare payments at the hospital level (beginning with a hospital admission forxa0HFxa0and acrossxa0multiple settings following discharge) and patient 30-day mortality using mixed-effect logistic regressionxa0models.nnnRESULTSnWe included 1,343,792 patients hospitalized for HF across 2,948 hospitals. Mean hospital-level 30-day Medicare paymentsxa0per beneficiary were

Association Between Race, Neighborhood, and Medicaid Enrollment and Outcomes in Medicare Home Health Care

15,423 ±