BCBSM Pay-for-Performance Measure Technical Document (Version 2.0)

Transcription

1 BCBSM Pay-for-Performance Measure Technical Document (Version 2.0) Developed by Michigan Value Collaborative July 2017

2 ACKNOWLEDGEMENTS P4P Measure Methodology Report 2 July 2017

3 TABLE OF CONTENTS LIST OF TABLES 5 LIST OF FIGURES 5 EXECUTIVE SUMMARY 6 INTRODUCTION 8 BACKGROUND... 8 CHARGE FOR THE MICHIGAN VALUE COLLABORATIVE COORDINATING CENTER... 8 MEASURE DEVELOPMENT TIMELINE... 8 GUIDING PRINCIPLES FOR MEASURE DEVELOPMENT... 9 DESCRIPTION OF MVC DATA 10 DEFINING THE EPISODE OF CARE Claims Attribution Process Standard Inclusions.. 10 Site of Service Transfer Cases Policy for inclusion of end-stage renal disease (ESRD) and cancer patients Standard Exclusions PRICE STANDARDIZATION Inpatient Facility Claims Post- Acute Care Claims...13 Skilled Nursing Facility (SNF) Claims Other Facility Claims Professional Claims Limitations RISK-ADJUSTMENT Background What is risk-adjustment? How does MVC calculate risk-adjusted episode payments? How does MVC calculate expected payments? Required Variables Non-required variables Condition Specific Risk Adjustment Variables (CSRAV) DESCRIPTION OF P4P MEASURE 17 OVERVIEW DATA SOURCES EPISODE DURATION SELECTION OF SERVICE LINES QUALITY REQUIREMENTS STRUCTURE AND SCORING SYSTEM IMPROVEMENT TARGET METHODOLOGY P4P Measure Methodology Report 3 July 2017

4 MVC COHORTS...27 SUPPORT FOR HOSPITALS ANTICIPATING UNINTENDED CONSEQUENCES APPENDICES 30 P4P Measure Methodology Report 4 July 2017

5 LIST OF TABLES Table 1: BCBSM Pay-for-Performance Program... 8 Table 2: Eligible service lines and minimum annual case requirements Table 3: P4P measure structure and scoring system Table 4: Potential unintended consequences and proposed interventions LIST OF FIGURES Figure 1: Cohorts Developed for the Service Lines of: Joint, Pneumonia, CHF, & Colectomy. 49 Figure 2: AMI Cohort Designation. 49 Figure 3: CABG Cohort Designation. 50 Figure 4: Spine Cohort Designation. 50 P4P Measure Methodology Report 5 July 2017

6 EXECUTIVE SUMMARY The Michigan Value Collaborative (MVC) is a Collaborative Quality Initiative (CQI) funded by Blue Cross Blue Shield of Michigan s (BCBSM) Value Partnerships program. MVC aims to help Michigan hospitals achieve the best possible patient outcomes at the lowest reasonable cost by adhering to the Value Partnerships philosophy of using high quality data to drive collaborative quality improvement. BCBSM included a new Pay-for-Performance (P4P) measure derived from MVC data in their 2016 Hospital P4P Program. In developing and recommending a measure to BCBSM, the MVC coordinating center has been guided by the following core principles: 1. The measure will reflect the BCBSM Value Partnerships philosophy of using high quality data to drive collaborative quality improvement. 2. The measure will be fair, simple, and transparent. 3. The measure will align with existing BCBSM and CMS hospital quality measures when possible and be consistent with Value Partnerships CQI principles. 4. The measure will encourage examination and use of MVC data to drive value improvement and reward those efforts. Outlined below is a summary of the proposed P4P measure. Measure: Data sources: Episode duration: Number of service lines: Eligible service line pool: Minimum case requirement: Baseline year: Risk adjusted, price standardized total episode cost BCBSM PPO plus Medicare fee-for-service claims Index hospitalization plus 30 days post-discharge Voluntary selection of two service lines from a pool of seven Acute myocardial infarction, congestive heart failure, pneumonia, joint replacement (hip and knee replacement episodes combined), colectomy (non-cancer), coronary artery bypass graft, spine surgery 1 The minimum case volume must be twenty cases (including both BCBSM PPO and Medicare) for each service line over the most recently available twelve-month period Hospital service line total episode cost for a twelve-month period prior to the start of the program year 1 On the MVC registry website, this service line is labeled Other Spine. P4P Measure Methodology Report 6 July 2017

7 Performance year: Program year: Quality requirement: Scoring system: Program year 2016: Program year 2017: Program year 2018 and 2019: Collaboration goals: Hospital service line total episode costs for the most recently available twelve-month period at the end of the program year Twelve-month period for which BCBSM P4P payments will commence No points for this performance measure if a hospital, adjusted for case volume, is ranked below the 10 th percentile in the performance year for condition-specific in-hospital mortality or related readmissions. Confidence intervals will be used to ensure that these hospitals are true statistical outliers A three-year active scoring system is outlined below. The final 2018 and 2019 scoring systems will be determined based on ongoing empirical analyses and feedback from hospital partners. The two service lines selected by hospitals for quality improvement action plans and risk adjustment feedback in 2016 will be the same two services lines that will be measured for performance in 2018 and Refer to Appendix I for more details. Structured preparatory measures Scoring based on site visit and MVC meeting participation; providing formal feedback on the P4P measure and risk adjustment; and developing service line improvement action plans Transition year No scoring will be earned in this year; however, performance in 2017 will be captured in subsequent P4P program years. Performance measure Scoring based on internal year-over-year improvement or achievement relative to MVC cohort group Bonus points for all hospitals working on the same service line if those hospitals achieve a 5% service line cost improvement 2 2 The maximum points attainable is ten even for those hospitals earning a bonus point. P4P Measure Methodology Report 7 July 2017

8 INTRODUCTION Background Since 2000, Blue Cross Blue Shield of Michigan has included a pay-for-performance program in its hospital contracts. This P4P program is currently structured as a potential 5% bonus payment added to baseline annual contracted rates, and it is measured during a January 1 through December 31 performance period. Across all 80 Michigan hospitals currently eligible to participate in the P4P program, the annual bonus pool is valued at approximately $200 million. From 2006 through 2013, the P4P program comprised two domains: quality and cost efficiency. In 2014, BCBSM modified the structure to include five distinct measures, as outlined in Table 1. These new measures provide further emphasis on the value of care delivered in the state, and they are intended to incentivize high quality, low cost, coordinated care. Table 1: BCBSM Pay-for-Performance Program measures Collaborative Quality Initiatives (CQI) performance 40% Population-based per-member/per-month (PMPM) costs 20% Hospital inpatient cost/case efficiency 20% 30-day readmissions 10% Admission/discharge/transfer (ADT) notification system participation 10% Charge for the Michigan Value Collaborative Coordinating Center The Michigan Value Collaborative is a quality improvement (QI) collaborative funded by BCBSM. Established in 2013, MVC aims to help Michigan hospitals achieve the best possible patient outcomes at the lowest reasonable cost by providing hospital leaders with claims-based utilization and episode cost data to drive local quality improvement activities. In recognition of MVC s emerging role as a convener of utilization data and infrastructure for sharing best practices for improving episode cost-efficiency, BCBSM implement a new P4P measure based on MVC data in the 2016 Hospital Pay-for-Performance program. The requested measure will be weighted at 10% of the total bonus pool and will be a freestanding measure, not part of the CQI performance domain. Measure development timeline The request for the MVC coordinating center to develop a new measure was delivered on July 16, A proposal was due to BCBSM for discussion at its P4P Measurement Workgroup meeting on September 8, There was a specific request for feedback from participating MVC hospitals prior to submission of the proposed measure. P4P Measure Methodology Report 8 July 2017

9 Guiding principles for measure development MVC is one of BCBSM s Value Partnerships Program Collaborative Quality Initiatives. The CQIs have long followed the Value Partnerships philosophy of using high quality data to drive collaborative quality improvement. The measure development team, comprised of members of the MVC Coordinating Center and several external advisors, was guided by this philosophy throughout the process (Appendix A). Specifically, the measure development group adhered to the following core principles: 1. The measure will reflect the BCBSM Value Partnerships philosophy of using high quality data to drive collaborative quality improvement. 2. The measure will be fair, simple, and transparent. 3. The measure will align with existing BCBSM and CMS hospital quality measures when possible and be consistent with Value Partnerships CQI principles. 4. The measure will encourage examination and use of MVC data to drive value improvement and reward those efforts. P4P Measure Methodology Report 9 July 2017

10 DESCRIPTION OF MVC DATA Defining the episode of care Claims Attribution Process The Michigan Value Collaborative (MVC) defines each service line episode using ICD 9/10 procedure and diagnosis codes or CPT codes. The transition to ICD-10 coding was successful using CMS General Equivalence Mappings (GEMs). With each data update, we look at different data trends and have observed no unexpected deviations. Certain exclusion criteria are applied to ensure patient groups are as homogeneous as possible and thus comparable across hospitals in contrast to using diagnosis related groups (DRG), CPT or revenue codes to define each episode. The MVC validation project of 2015 identified an important opportunity to update the process by which post-discharge claims are attributed to index admissions and thus included in a 30 or 90 day episode window. For each claim in the post discharge period, the first and second Dx code of that claim is examined against an episode-specific document listing the Dx codes that are considered related to that episode. Post discharge claims are considered related to the index admission if the first or second Dx code listed on the claim is considered related to the episode type. The MVC validation project identified that there were Dx codes in the episode-specific document that were currently considered unrelated to the episode type, but should be considered related. This finding triggered a comprehensive review of each episode-specific document to examine the Dx codes listed as related or unrelated to the episode. Each episode-specific document was updated based on a multi-round review by clinical content experts and MVC Coordinating Center staff. Disagreements were discussed as a group, consensus was reached, and the final episode-specific documents were approved and transmitted to ArborMetrix for processing. Standard Inclusions These standard inclusions (Appendix B) are based on complications as defined by CMS for their complication measures, CMS Serious Reportable Events, American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP), and clinical experts Claims within 90 days with the following diagnosis codes should be considered related to the index hospitalization for all episodes. 3 Blue Cross Blue Shield of Michigan Hospital Pay-for-Performance Program Peer Groups 1-4. April Michigan Value Collaborative. Hospital P4P Survey. May CMS Medicare Severity Diagnosis-Related Groups (MS-DRGS), relative weighting factors and geometric, and arithmetic mean length of stay (Table 5) American Hospital Association survey database. P4P Measure Methodology Report 10 July 2017

11 Site of Service Consistent with CMS methodology, any facility claim originating from one of the below sites of service, or any professional claim associated with outpatient rehab, will be considered related to the index admission. It will be included in the 90 day episode cost and utilization calculations (see MVC Claims Categorization Rules document). This site of service relationship supersedes the Dx codes described above. Facility claim SNF HH IP Rehab OP Rehab Professional claims OP Rehab Transfer Cases MVC attributes transfer patients to the hospital where the index admission began. If a patient is transferred from the originating hospital before an MVC episode has been triggered, then the patient is attributed to the receiving hospital. Transfer cases represent a small percentage of overall cases, but since they represent real patients, inclusion in the MVC analytics ensures this population s outcomes have the ability to be measured and improved as part of overall quality improvement. Policy for inclusion of end-stage renal disease (ESRD) and cancer patients MVC has chosen to include end-stage renal disease patients and most cancer patients in our episode cost calculations. These conditions are included in the risk adjustment model as candidate variables. When those variables are found to significantly impact costs, they become part of the final models. Since potential high costs from these conditions are accounted for by the risk-adjustment, we do not exclude them. A detailed description of the MVC risk-adjustment model is discussed in a later section of this document. Standard Exclusions Chemotherapy is excluded using the following ICD-9 and ICD-10 codes. ICD 9 DX CC ICD9 CODE DESCRIPTION V RADIOTHERAPY ENCOUNTER V CHEMOTHERAPY ENCOUNTER (End 2005) V ANTINEOPLASTIC CHEMO ENC (Begin 2005) V IMMUNOTHERAPY ENCOUNTER (Begin 2005) V RADIOTHERAPY CONVALESCEN V CHEMOTHERAPY CONVALESCEN V RADIOTHERAPY FOLLOW-UP V CHEMOTHERAPY FOLLOW-UP P4P Measure Methodology Report 11 July 2017

12 ICD 10 CCS Description Z Encounter for antineoplastic radiation therapy Z Encounter for antineoplastic chemotherapy Z Encounter for antineoplastic immunotherapy MVC applies an algorithm to identify clinically related readmissions that occur within 30 and 90 days post-discharge. For example, claims data with primary and secondary diagnosis codes for chemotherapy are classified as an unrelated readmission and excluded from readmission rate calculations. Price Standardization ArborMetrix has developed a process to standardize medical claim payments for the purpose of analyzing hospital level variation in utilization. The goal of our approach is to eliminate the extent to which price variations are a result of contractual differences by assigning the average amount for a given service for each instance of the service in the data. This does not account for geographic variation that may be caused by cost structure differences and would tend to overstate the actual prices in rural or low-wage areas and understate actual prices in urban and high-wage areas. This particular process was developed for the joint analysis of BlueCross BlueShield of Michigan and Michigan Medicare data and likely contains some idiosyncrasies specific to BCBSM and Medicare data. The overarching goal is to standardize the prices of both the BlueCross and Medicare claims to be priced the average amount for that service in Medicare. Reconciling the occasionally different reimbursement practices of BCBCM and Medicare is an ongoing challenge and this process is subject to further refinement. However, the basic practices should remain consistent. Our price standardization process divides up the data into three parts: facility claims are comprised of 1) inpatient facility claims and 2) other facility claims, while 3) professional claims are treated as one group. This document will describe price standardization for each group separately. Inpatient Facility Claims We calculate two payment amounts for inpatient claims: DRG base payment and outlier payment. DRG Adjustment P4P Measure Methodology Report 12 July 2017

13 One complication in pricing data over multiple years is that the DRG definitions (and relative weights) change over time. To account for this, we used 3 rd party DRG grouping software. This takes information from five data elements (patient sex, patient age, patient discharge disposition, ICD9 diagnoses, ICD9 procedures) and defines DRGs for each inpatient claim using definitions from whatever year the user chooses. For the development, we took all BCBSM PPO and all Medicare claims and assign them DRGs as if those claims had been processed in Over time, the DRG mapping should be updated. DRG Base Payment Inpatient claims are assigned payments by DRG. Each DRG is associated with the average price for that DRG in the Medicare data. Outlier Payment Outlier payments are made separately from the base payment to providers to compensate for particularly complicated patients (i.e., when the level of treatment greatly exceeds the expected average for a given DRG s relative weight). As a general rule, these outlier payments are triggered when the claim s length of stay is significantly longer than the average length of stay for its DRG. As with the base payment, our outlier payment calculation uses information from TRICARE to standardize patients. The TRICARE DRG schedule includes a national long stay threshold. Inpatient claims associated with a length of stay that exceed the national long stay threshold will be flagged as outliers. The outlier payment is calculated as $2,500 per day over the length of stay threshold. If LOS>LOS Threshold DRG, then Outlier Payment = (LOS-LOS Threshold DRG)*$2,500 Otherwise, Outlier Payment = 0 Post- Acute Care Claims Skilled Nursing Facility (SNF) Claims For CMS patients, SNF payments are calculated based on length of stay and the Resource Utilization Groups (RUG) reported in the individual SNF claims. Daily RUG rates range from $180 to $743. For BCBSM patients, SNF payments are calculated based on length of stay and a standard daily payment rate of $479 per day that was calculated based on three independent analyses. Other Facility Claims These constitute the wide variety of facility claims that are not inpatient care. When possible, we use the CPT code associated with the claim. In cases where the CPT code is not available, we use the revenue code on the claim line. Each CPT or revenue code is associated with a quantity. The total payments for each code are summed and then divided by the sum of quantities to create a code rate for each CPT and revenue code. Prior to the rate calculations, payments for Other Facility Claims were trended at 3% per year up to $2012. Standardized payment = Code Rate * Quantity where Code Rate = Total payment for code / Total code quantity and quantity is capped at 0.95*max code quantity P4P Measure Methodology Report 13 July 2017

14 Professional Claims This process is similar to the one used for Other Facility Claims. The only difference is that all the claims have CPT codes. Ideally, this would be the simplest of the pricing standardization procedures. Each professional claim is associated with a CPT code, quantity, and unit. Each quantity*unit payment would be the same for a given CPT code. However, our version of the BCBSM data does not include unit amounts. Thus we have claims associated with a quantity, but no unit for that quantity. For most claims this is not a problem because the same unit is used for all claims (surgery, for example). However, for some pharmaceuticals the unit is not always constant and simply taking the average payment and dividing by average quantity can skew the rate calculation. To get around this we cap the top allowed quantity for both rate calculation and payment calculation purposes at 95% of the top quantity, yielding a simple formula: Standardized payment = CPT Rate * Quantity where CPT Rate = Total payment for CPT/ Total CPT quantity and quantity is capped at 0.95*max CPT quantity This process also incorporates CPT modifiers codes when calculating CPT rates. The CPT rates are calculated separately for CPTs with certain modifier codes 7 to account for the fact that reimbursement differs based on the use of these codes. Limitations One challenge is when Medicare does not reimburse certain codes we see in the BCBSM data. When that occurs, we rely on commercial claims data to which we have access. The pending addition of Medicaid data may have its own pricing peculiarities that need to be explored and accounted for. Lack of units in the professional data significantly limits our confidence in our pricing of certain high quantity services (e.g., prescription drugs). In the future, we should make sure that any data used on the website include all the fields necessary for our price-standardization process. The outlier payment calculation could be improved. The current $2,500 per-day rate functions well for most cases, but for extremely long lengths of stay (>1 year), the methodology likely overstates the actual outlier payment. MVC received a one-time feed of outlier indicators from BCBSM. ArborMetrix analyzed that file as was to show that reasonable overlap with what MVC were calling outlier and what BCBSM had as an outlier. However, it was difficult to rigorously compare the pricing for those because BCBSM could only supply the outlier payments in broad categories, and not the specific amounts. We only use the most common CPT modifiers as part of our rate schedule creation. Some evidence from users is that surgical assistant modifiers are not being accounted for in all instances. We may need to review which modifiers are being used and which CPT modifier fields (currently only the first modifier) are being included when calculating professional claim rate schedules. 7 The following are the modifier codes currently included during rate schedule creation: AS,80,81,82,26,25,TC,GP,NU,59,RR,AM,RH,76,SH,51,RT,LT,AA,GC,HN,AT,U1,QK,SQ,50,HH,QX,QW,33,SA,AI,30, NH,57,GO,24,GN,58,KX,SL,Q8,79,U2,91,KS,NJ,JN,AJ,QY,AH,PH,A1,PT,77,SW,U5,JR,RJ,52,HR,QS,GR,RG,Q9,EC P4P Measure Methodology Report 14 July 2017

15 Risk-Adjustment Background As discussed in the November 2015 collaborative meeting, the Coordinating Center performed extensive research on different risk adjustment approaches. Given its findings, MVC recommended transitioning to CMS Condition Categories for identifying and accounting for comorbidities in the calculation of episode payments and other MVC metrics. The MVC risk adjustment models now employ the 79 Hierarchical Condition Categories that CMS has empirically shown to be predictive of expenditures for Medicare beneficiaries. Our methodology utilizes all 25 Dx codes reported on a Medicare and BCBS claim. What is risk-adjustment? Hospitals treat a variety of patients, and some patients are costlier than others. Hospitals that treat a disproportionate number of costly patients may be unfairly classified as high cost hospitals simply because of the type of patients that they treat. Risk-adjustment is a statistical method that levels the playing field by accounting for differences in case-mix. How does MVC calculate risk-adjusted episode payments? MVC performs risk-adjustment using observed/expected (O/E) ratios. The numerator in this ratio is the aggregate of the all observed payments for a particular hospital. The denominator is the aggregate of the all expected payments. This ratio is multiplied by the statewide expected mean payment to arrive at the risk-adjusted payment for that hospital. How does MVC calculate expected payments? MVC calculates expected payments for each condition (e.g., AMI, pneumonia, CABG) and each component (e.g., total episode payments, readmission payments) separately. Condition and component-specific expected payments are based on a statistical model that uses a combination of required variables and non-required variables. Required variables The following required variables are always included in the final model: age, gender, history of prior high spending, end-stage renal disease. Non-required variables Non-required variables include 79 comorbidities based on hierarchical condition categories (Appendix C), and condition specific risk adjusters (Appendix D). Non-required variables are selected using a model specification technique that occurs in two steps: 1. All candidate variables are individually tested using a univariate regression model to see if they predict payment. Non-required variables with a p-value < 0.10 are retained. 2. All of the retained variables are included in a multivariable regression model and variables with a p < 0.05 are used for the final model. The final variables for all conditions are listed Appendix E. P4P Measure Methodology Report 15 July 2017

16 Condition Specific Risk Adjustment Variables (CSRAV) MVC incorporates several condition-specific variables into our risk-adjustment model. CSRAVs have been suggested by participating hospitals and clinicians. For each CSRAV that is suggested, MVC evaluates the appropriateness of including the variable by following the four principles below: Principle 1: All CSRAVs will be considered as a candidate for the risk-adjustment model. In other words, any CSRAVs may be excluded in the final model if they are not found to be statistically significant. Principle 2: For surgical conditions, treatment decisions (e.g., laparoscopic vs open) are typically not considered CSRAVs. Principle 3: For all conditions, we will consider certain diagnosis codes as a CSRAV: Cancer diagnosis Reoperation diagnosis Principle 4: Variables that represent small variations of a disease process should not be considered a CSRAV. However, these variables can be grouped into broad categories: Simple case Complex case Examples Variable Category Appropriate riskadjustment variable? Lap vs Open for Colectomy Treatment decision (rarely both) No Cancer for Colectomy Severity of illness Yes Dialysis for AMI Treatment decision or complication No Emergency intubation for AMI Treatment decision or complication No GI bleed for colectomy Severity of illness Yes Re-operative CABG Severity of illness Yes Cardiac surgery for AMI Severity of illness Yes* Trach for pneumonia Treatment decision or complication No *In select instances, a treatment decision strongly reflects severity of illness P4P Measure Methodology Report 16 July 2017

17 DESCRIPTION OF P4P MEASURE Overview Description: The proposed efficiency measure is price-standardized, risk-adjusted 30 day episode costs. Rationale: The selection of total episode payments for two selected service lines reflects the centrality of this measure to the work of MVC. This measure is also immediately available for hospitals in reports from the MVC registry. Other considerations: The selection of total episode costs, as opposed to one or more component costs (e.g., readmissions or post-acute care), is consistent with MVC s efforts to develop and maintain a quality improvement focus across the continuum of care settings. Alignment: In Appendix F, we describe key characteristics of several value-based incentive programs from CMS that were reviewed during the measure development process. A condition-specific 30-day total episode spending aligns directly with the format of condition-specific episode-based cost measures being considered by CMS as supplements to the MSPB metric in the Hospital Value-Based Purchasing (HVBP) program. Hospital Feedback: There was no specific feedback on total episode versus component costs. Final Decision: MVC will measure total episode payments. Data sources Description: Based on ongoing analyses and feedback from hospital partners, it is recommended the measure use BCBSM PPO and Medicare fee-for-service claims. Discussion: As of July 2017, data available in MVC reports will include BCBSM PPO claims for January 2011 through June 2016, and Michigan Medicare fee-for-service claims for January 2011 through March We receive quarterly feeds of updated Medicare data that will allow MVC reports to include more synchronous and timely data from both BCBSM PPO and Medicare. We anticipate incorporating data fourth quarter 2016 BCBSM PPO and Medicare claims into the MVC registry by end of The consideration of using BCBSM PPO and Medicare claims versus only BCBSM PPO claims hinges on the tradeoff between an increased number of episodes for each hospital and the reality that hospitals cannot currently use the MVC registry to drill down into patient-level details of the Medicare cases (due to Medicare data privacy requirements). If the measure includes Medicare data, there will likely be more internal record review required for hospitals to understand the root causes of high-cost episodes among Medicare beneficiaries. P4P Measure Methodology Report 17 July 2017

18 Including both BCBSM PPO and Medicare claims in the measure increases the number of episodes available and will enable the measure to better focus on two services lines. The inclusion of more admissions in the calculation of episode costs will also ensure greater reliability of cost estimates. This assumes, however, that our anticipated timeline for receipt of updated Medicare claims comes to fruition. Including BCBSM PPO claims alone in the measure would require that the measure become an aggregate of the seven eligible service lines. However, hospitals could drill down into patient level details for all patients covered by the measure. Feedback from participating hospitals and CQI partners has consistently indicated that condition or service-line specific payment measures are more actionable from a quality improvement perspective, so an aggregate measure could limit quality improvement. Alignment: Using both BCBSM and Medicare claims in the measure would provide greater alignment between MVC activities and related metrics from CMS, including the condition-specific episode measures developed by CMS to supplement MSPB in the efficiency domain of the HVBP program and the Comprehensive Care for Joint Replacement (CCJR) project (Appendix F). A composite spending measure would be more analogous to CMS Medicare Spending per Beneficiary (MSPB) that reflects spending across multiple Diagnosis Related Groups (DRGs). Hospital Feedback: From a survey sent to all participating MVC hospitals, 54/66, or 81.8%, preferred BCBSM PPO plus Medicare data. Comments in support for BCBSM and Medicare data: - Alignment with current CMS activities and goals - Drill down will reveal opportunities for both populations - Statistically meaningful data Comments in support for BCBSM alone - Lack of detail and timeliness with CMS data - Better alignment with other BCBSM CQIs - Better ability to partner with post-care entities such as physician offices when data is relevant - Root causes may be determined more easily (with drill-down), less internal record review Final Decision: MVC will use BCBSM PPO plus Medicare Fee-for-Service claims data. Episode duration Description: The episode duration will include the index hospital stay plus 30 days post-discharge. Rationale: The main decision point was the selection of a 30 versus 90 day episode. Ultimately, selection of a proposed measure based on 30 day episodes was influenced by several considerations P4P Measure Methodology Report 18 July 2017

19 including the following: 1) feedback from MVC hospitals has consistently suggested that a 30 day window is more clinically relevant and actionable for most of the MVC service lines; and 2) the 30 day episode better aligns with several concurrent initiatives from CMS and BCBSM, allowing hospitals to focus their improvement efforts on this time window. Other considerations: In addition to the rationale provided above, internal empirical analyses and review of claims submitted from days after hospital admissions suggest declining face validity with respect to attribution of services to the index hospital stay. We felt that inclusion of distantly related claims as part of a longer 90 day episode could undermine buy-in with respect to the proposed measure. Alignment: As described above, the selection and specification of 30-day total episode spending aligns well with measures in Medicare s HVBP program and the Hospital Readmissions Reduction Program (HRRP). In addition, this episode duration is consistent with the 30-day readmission measure already included in the BCBSM P4P program, thereby allowing hospitals, clinicians, and post-acute care providers participating in MVC to maintain a focus on care processes and transitions during this discrete time period. Although the recently announced Comprehensive Care for Joint Replacement Program (CCJR) from CMS involves a 90 day episode, there was widespread agreement among the development team that a 30 day episode is more consistent with the overarching principles for measure development. Moreover, only two Metropolitan Statistical Areas in Michigan (Flint and Saginaw) are participating in the CCJR, we did not feel that the different episode durations would be a major concern for most hospitals in the state. Hospital Feedback: There was no specific feedback on episode duration. Final Decision: MVC will use a 30 day measure. Selection of service lines Description: If the final measure includes BCBSM PPO and Medicare claims, hospitals will be required to select two service lines from a pool of seven eligible diagnoses and procedures (Table 2). Hospitals will select two service lines in 2016 for quality improvement action plans and will be evaluated on the same two service lines during the 2018 and 2019 performance periods. If the final measure includes BCBSM PPO claims alone, then the seven eligible services lines below will be combined into one aggregate measure. The seven service lines below were selected because they met the following criteria: 1. The service line was included in the 2015 MVC data validation project 2. The service line represents an area of focus for existing BCBSM clinical CQI programs P4P Measure Methodology Report 19 July 2017

20 Service Line Table 2: Eligible service lines and minimum annual case requirements Numbers of episodes reviewed in 2015 MVC data validation project Minimum annual case requirement* Percent of 2012 cases from BCBSM PPO patients Acute myocardial infarction % Congestive heart failure % Pneumonia % Hip and knee replacement % Colectomy (non-cancer) % Coronary artery bypass graft % Spine surgery^ % *Minimum annual case requirement includes both Medicare FFS and BCBSM PPO patients. A hospital s conditionspecific case volume will be measured during the most recently available 12-month period. ^Labeled as Other Spine on MVC registry Rationale: Number of service lines Our measure development group and external advisors reached a consensus that two service lines is a reasonable initial number for hospitals to work toward both achievement and improvement targets. A single service line might not sufficiently encourage hospitals to pursue crosscutting interventions and infrastructural changes aimed at improving efficiency. On the other hand, if the proposed measure involves too many service lines, hospitals may be overwhelmed and could have difficulty deciding where and how to focus limited resources. Selected service lines The selection of eligible service lines reflects the dual goals of 1) maximizing a hospital s choice in terms of where to focus its efforts, and 2) alignment of MVC measures with existing cost and quality improvement initiatives from CMS and BCBSM. To this end, we first considered all service lines that are already part of a similar CMS initiative [i.e., HVBP, HRRP, and CCJR, which together cover acute myocardial infarction (AMI), congestive heart failure (CHF), pneumonia, and joint replacement]. We next considered clinical service lines that represent areas of focus for existing BCBSM clinical CQI programs. From this initial larger roster, we then selected for final inclusion only those service lines that were also included in the 2015 MVC Data Validation project. Data Validation Project As part of the 2015 data validation project, we asked all MVC hospitals to submit information on utilization of relevant services (e.g., post-acute care, readmissions) during and after an index admission based on review of local medical records. We then compared the submitted information to data in the MVC registry based on BCBSM claims. All 63 MVC hospitals participated in the data validation, and this P4P Measure Methodology Report 20 July 2017

21 project examined the claims classification algorithms for eleven service lines. 8 This process was completed to ensure that the MVC Coordinating Center and participating hospitals had an opportunity for transparent examination, and subsequent refinement, of the claims classification algorithms for the included service lines. Detailed results from the validation project, and consequent modifications to MVC methods, was shared individually with hospitals during fall 2015 and with the entire collaborative at the November 2015 meeting. We anticipate that other service lines will become eligible for the MVC P4P measure as we extend the data validation process to more diagnoses and procedures in the future. Minimum case requirements We selected the minimum episode volume requirements based on several empirical analyses. First, we calculated the total episode volume for each MVC hospital with each service line. Second, we calculated the year-to-year reliability of our payment measure for all service lines. We then tested the reliability of the condition-specific measure across three years ( ). Ultimately, we selected minimum case thresholds that simultaneously maximize the reliability of the episode cost metric and the number of eligible hospitals for each service line. Appendix G provides additional details around these calculations. Other considerations: We considered many service line options before deciding on the approach described above. For example, we considered basing the proposed P4P measure on two standard highvolume service lines (i.e., pneumonia and AMI) for all hospitals. We also considered asking hospitals to select from more restricted pools (e.g., only service lines overlapping with CMS initiatives). However, the MVC development team and external advisors felt that such approaches were too restrictive, especially for hospitals that already have existing efficiency improvement initiatives. We also considered using statistical methods to improve the reliability of the proposed measure for hospitals with small case volumes. Using this approach, episode costs for hospitals with a smaller number of eligible cases would be shrunk to the mean to reduce the effect of random variation due to small sample sizes. While our working group acknowledged that this approach is used by CMS for public reporting of outcome measures on Hospital Compare, we felt that this strategy might also blunt the year-over-year improvement for some hospitals (i.e., small hospitals) and thus make it difficult for low volume hospitals to demonstrate high performance. Alignment: The eligible service lines align with many of the conditions that CMS includes in its valuebased incentive programs (Appendix F). For example, CMS recently developed 30-day episode payment measures for AMI, CHF, and pneumonia, as well as the CCJR program for episode-based bundled payments for hip and knee replacement. A requirement for minimum case thresholds is also used in the HVBP and MSPB programs. 8 At the time of the data validation project there were only 63 participating hospitals; however, there are currently 75 member hospitals. P4P Measure Methodology Report 21 July 2017

22 Hospital Feedback: From a survey sent to all participating MVC hospitals, 41/66, or 62.1% of hospitals, preferred a selection of two service lines from a pool of seven. Nine hospitals selected a performance measure based on selection of two service lines and use of BCBSM data alone for the P4P measure. These choices are mutually exclusive, but 6/9, or 66.7% of hospitals say the selection of two service lines is more important to them than using BCBSM data alone. Comments in support for selection of two service lines: - Difficult to find opportunities for improvement that span all service lines - More specific, actionable data - Desire to be evaluated based on specific initiatives in problem areas; aggregate would dilute efforts - Hard to make meaningful changes in 1-2 years for all service lines - Can allocate resources to service lines in greatest need - Suggestion that hospitals should be able to select ANY service line that meets case volume criteria Comments in support for aggregate measure: - Alignment with CMS and BCBSM metrics - Greater flexibility to manage cost - Smaller hospitals forced to choose service lines with adequate volume; low volume makes episode costs less reliable Final Decision: MVC will measure performance based the hospital s voluntary selection of two service lines from the pool of seven. Quality requirements Description: In our proposed measure, hospitals will not be eligible for a bonus if they are ranked in the bottom 10 th percentile in the performance year for condition-specific, risk-adjusted in-hospital mortality or related readmissions. Statistical adjustments will be made for hospitals ranked in the bottom 10th percentile if their case volumes are too low to allow for a reliable estimate. Accordingly, we will utilize confidence intervals to ensure that these hospitals are true statistical outliers. This approach is aligned with the method used by CMS for public reporting of outcomes on Hospital Compare. We recognize that future analyses may lead the 10 th percentile benchmark to be shifted to other levels and/or help us decide whether multiple years of data should be considered for calculation of this quality requirement. We plan to work towards the eventual replacement of percentile-based quality thresholds with criterion-based quality thresholds. Rationale: It was the consensus of our measure development team that the MVC episode cost measure should be linked with a required minimum quality standard. The group felt it would be inconsistent with MVC and BCBSM principles to reward hospitals that are low cost, but also potentially lower quality. We selected condition-specific mortality and related readmissions because these P4P Measure Methodology Report 22 July 2017

23 outcomes can be measured from claims data, are consistent with quality measures already used by CMS and BCBSM, and are endorsed by the National Quality Forum. Other considerations: We reviewed several approaches to defining a quality minimum for the proposed MVC measure. In particular, we considered linking the MVC measure with quality measures submitted by the BCBSM clinical CQIs. However, this approach would be difficult to implement since some Michigan hospitals are not currently participating in the clinical CQIs. In the future, we would like to incorporate CQI-specific outcomes measures as part of the required quality standard whenever possible. Alignment: The quality standards that we selected are used by CMS for several programs including HVBP, the Physician Value-Based Payment Modifier, and the Medicare Shared Savings Program for ACOs. In addition, the CCJR program requires hospitals to meet three quality metrics to be eligible for bonus payments. These quality metrics are based on hospital level risk-standardized complication rates for joint replacement, 30-day all-cause readmission rates and the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey. Hospital Feedback: Only 3/66, or 4.5%, of the MVC hospitals surveyed indicated no quality threshold should be applied. Four hospitals suggested a penalty in place of forgoing the bonus for those hospitals not meeting the quality threshold. One hospital, however, recommended raising the threshold. While most surveyed agreed with the proposed standards, a few hospitals suggested adding the following: Complication rate Provision of preventative health services Correlation to performance improvement plans Hospital acquired infections and conditions Other CQI clinical measures Length of Stay Physician follow up post-discharge Avoidable emergency department utilization Final Decision: MVC will impose a quality requirement whereby a hospital s adjusted case volume must be above the 10 th percentile in the performance year for the condition-specific in-hospital mortality or related readmissions to receive performance points. Structure and scoring system The new P4P measure was developed according to the principles outlined earlier in this document, and aims to be as simple, fair, and transparent as possible. The proposed structure and scoring system for this measure is depicted in Table 3 and Appendix H. P4P Measure Methodology Report 23 July 2017

24 Measure theme Table 3: P4P measure structure and scoring system and 2019 Transition Year Year over year or improvement Deep engagement with MVC data Absolute achievement Measure scoring Collaboration goal 2 pts - Participate in site visit with MVC 1 pt - Provide feedback on risk adjustments 1 pt - Provide feedback on P4P measure 1 pt - Attend MVC meetings 5 pts - Create specific QI action plans for 2017 No Score ^ = scoring for each service line assumes use of both BCBSM PPO and Medicare claims SD = standard deviation Baseline: Hospital service line total episode costs for the most recently available 12-month period 2018/2019 scoring:^ 1 pt = baseline mean 2 pts = baseline mean * SD based target 3 pts = baseline mean * SD based target 4 pts = baseline mean * SD based target 5 pts = baseline mean * SD based target Baseline: MVC cohort group service line total episode costs for the most recently available 12-month period 2018/2019 scoring:^ 50 th %tile = 1 pt 60 th %tile = 2 pts 70 th %tile = 3 pts 80 th %tile = 4 pts 90 th %tile = 5 pts +1 pt per service line if 5% improvement among hospitals choosing the specific service line We proposed that program year 2016 be a formal preparatory year that encourages MVC hospitals to become more familiar with the data, organize resources to enact change, and create infrastructure and care processes that enable success in 2017 and beyond. In this year, two service lines will be selected by hospitals prior to submission of the QI action plan. Those two service lines will remain the target service lines for the hospital in 2018 and Hospital performance was initially set to be measured in program year 2017; however, based on feedback from the P4P Measurement Workgroup, scoring was delayed until to allow hospitals more time to prepare. The proposed metrics for program years 2018 and 2019 aim to provide Michigan hospitals with an opportunity to be measured fairly and to be rewarded for high performance and collaboration. In program years 2018 and 2019, each hospital s service line-specific total episode cost for a twelvemonth period two years prior to the performance year would serve as its own improvement baseline. P4P Measure Methodology Report 24 July 2017

25 Appendix I provides additional details regarding the proposed method for calculating baseline performance data. During these years, improvement would be measured against the baseline. This calculation will be performed separately for each service line. As illustrated in Table 3, the improvement target for each hospital will be scaled based on case volume to account for differences in the reliability of MVC cost estimates across hospitals. In addition to measuring year-over-year improvement, hospitals may earn bonus payments based on their achievement. The MVC cohort group service line mean total episode cost from the twelve-month period two years prior to the performance year would serve as the achievement baseline for calculating percentile-based achievement targets. For 2018 and 2019 payments, hospitals would receive the higher of their improvement or achievement points. Performance points for improvement or achievement will only be awarded to hospitals that also meet the quality requirements. Collaboration goal: The MVC measure development team felt that it was important to follow closely the Value Partnership s philosophy of using high quality data to drive collaborative quality improvement. In order to foster collaboration at the local, regional, and state levels, a collaboration goal is also built into the scoring system for 2018 and 2019 (Table 3). Among hospitals selecting the same service line (if applicable), each hospital will receive one additional bonus point if that group of hospitals achieves a 5% or greater improvement in total episode costs compared to their collective average episode cost during the baseline period. However, a hospital would not receive a collaboration bonus if its own performance declined from one year to the next. Appendix H provides a more detailed example of how the proposed scoring system would be applied in all three performance periods. Hospital Feedback: There were many comments on the survey distributed to MVC hospitals about the scoring system: - Year-over-year improvement should be used. Will encourage collaboration, data transparency, and sharing of best practices - Outlier high costs should be removed - MVC patient population should be homogenous (ESRD should be excluded, and potentially cancer patients as well) - Clear and concise date range to base improvement targets against - Does not recognize or award organizations that are already high performers - Percent improvement expected over 2 years seems unreasonable Comments regarding how continued collaboration can be incentivized within MVC: - Bonus points for hospitals sharing cost reduction strategies - Incentivize attending MVC events - Reward year-over-year improvements - Incentive points for low performing hospitals to partner with successful mentoring hospital - Grant opportunities for specific themes P4P Measure Methodology Report 25 July 2017

26 Final Decision: In 2016, hospitals will earn points based on engagement and contributions to MVC. For 2017, hospitals will be measured based on year-over-year improvement, and in 2018, hospitals will be scored based on improvement or achievement, whichever yields the highest score. Improvement target methodology Description: The improvement targets for each hospital is measured by the service line total episode payments for the baseline year. To earn one P4P point, a hospital must have equal or lower total episode costs during the performance year compared to the baseline year for the measured service line. Subsequent bonus points will be calculated based on the ratio of each individual hospital s total episode payment to MVC total episode payment multiplied by a proportion of the MVC standard deviation. Baseline Mean x% * MVC Winsorized Standard Deviation where Baseline Mean = Hospital Mean Cost/MVC Mean Cost and x% represents.05 through.20 The intent of the formula is to account for each hospital s baseline mean costs and the service linespecific variability. The MVC mean and standard deviation will include all cases, and the MVC standard deviation will be winsorized at the 99 th percentile. The utility of winsorization is to mitigate the impact of extreme outlier cases. The percentage reduction of the MVC standard deviation required to earn P4P points was determined based on extensive internal modeling and comparisons to provider performance in the BPCI Model 2 program. Other considerations: We considered alternatively applying winsorization to both the hospital cost ratio and standard deviation as proposed by members of the P4P Measurement Workgroup. In this scenario, the MVC average cost (the denominator) will be winsorized for all hospitals; however, not all hospitals will have their individual costs (the numerator) winsorized. The individual hospital cost will only be winsorized if its case(s) contributed to the top 1% of MVC episode costs and the hospital has at least 100 cases. As a result, the majority of hospitals will have a slightly higher benchmark (ranging from approximately $1-$200 greater). After applying the P4P reduction targets (5-20%), the cost difference is minimized ($0-$20 on average, per patient) between options one and two but still greater compared to only winsorizing the MVC standard deviation. Hospitals with cases in the top 1%, however, will benefit more noticeably from winsorizing the cost ratio. Rationale: The decision to winsorize at the 99th percentile was based on empirical analysis showing that many of these top 1 percent of cases appear to be outliers. Cases winsorized at the 95th percentile, however, were not extreme cases. In reviewing other incentive-based programs, winsorizing at the 99th percentile is consistent with CMS MSBP methodology. As part of its original proposal, the MVC measure development team did not recommend winsorizing the hospital cost ratio primarily because it presents a bias to low volume hospitals. If each hospital s costs were winsorized, the hospital must have at least 100 cases during the baseline and performance P4P Measure Methodology Report 26 July 2017

27 periods to be impacted. For five of the seven service lines more than half of the hospitals have less than 100 cases in a twelve-month period. Hospital Feedback: Both options were shared with BCBSM s P4P Measurement Workgroup, and only one hospital expressed a preference, which was to only winsorize the standard deviation. Final Decision: Each hospital s P4P baseline mean will be computed as a ratio of the hospital s average cost to the MVC average cost and multiplied by the winsorized MVC standard deviation. Cases will be winsorized at the 99th percentile. MVC cohorts Description: Beginning in 2018, hospitals may earn year-over-year achievement points as compared to their respective cohorts. The MVC cohort methodology comprises two years of data that has empirically demonstrated stability in the groupings. In general, each MVC cohort is comprised of structurally similar hospitals identified by case mix index (CMI), bed size and teaching status. Methodology: CMI is defined based on the hospital s index admissions for the P4P service lines for Medicare FFS and BCBSM PPO patients from July 1, 2013 through June 30, For the purposes of classification, episodes without a Medicare Severity-Diagnosis Related Group (MS-DRG) associated with the index admission were excluded. The Centers for Medicare and Medicaid Services (CMS) MS- DRG relative weights are applied to all inpatient admissions to calculate the mean relative CMI weight for each hospital. Using the calculated mean relative CMI weights, hospitals are sorted from highest to lowest to establish a median threshold. The median CMI was 1.67; therefore, all hospitals with a CMI of 1.67 or greater will be classified as either Cohort 1 or 2. Hospitals with a low CMI (below 1.67) will be grouped as either Cohort 3 or 4. CMI is used as a primary grouper to account for the differences in the complexity of DRGs observed at each hospital. The cohorts are further refined based on bed size and teaching status. The presence of 300 or more beds and teaching status are used to distinguish between Cohorts 1 and 2. A hospital must meet both criteria to be categorized as Cohort 1. The presence of 100 beds or more and teaching status are used to differentiate between Cohorts 3 and 4. Unlike distinguishing between Cohorts 1 and 2, a hospital only needs to meet one of these criteria to be classified as Cohort 3 to ensure balance in the grouping dispersion. See Appendix J for the cohort flow chart and distribution. Rationale: The purpose of the MVC cohort groups is to provide for a more equitable and effective comparison of each hospital s performance. Hospital Feedback: Comments provided by BCBSM s P4P Measurement Workgroup on the original cohort structure was to develop service line specific cohort groups for Spine, AMI and CABG. The Workgroup believed a greater amount of variability with respect to cost is found within these conditions due to hospital structural characteristics. Therefore, a single cohort for across all service lines would not adequately capture these variances. P4P Measure Methodology Report 27 July 2017

28 Final Decision: Separate cohort groups were created for Spine, AMI and CABG. See Appendix J for greater detail on how each cohort is structured. The remaining service lines (Joint Replacement, Pneumonia, CHF and Colectomy) will share the same grouping based on CMI, bed size and teaching status. Support for hospitals Two documents were made available to hospitals to support their participation in The first provided hospitals with a summary of the baseline costs for their two selected service lines. The document displayed component cost breakdowns similar to those provided on the MVC registry. The second document was a template for a specific quality improvement action plan. The template will help hospital leaders undertake a structured assessment of opportunities for improvement and provide a framework for identifying and deploying resources necessary to achieve episode cost improvements for the selected service lines. The ideas and strategies outlined in these templates will also serve as a foundation and framework for collaborative learning and best practice sharing at MVC meetings. Each year, the MVC Coordinating Center will also provide partner hospitals with an updated hospital P4P report. The P4P report will identify hospital baseline costs for the relevant performance year. This document outlines hospital-specific episode payment reduction targets for eligible service lines in that performance year to help hospitals identify what is needed to achieve successive P4P points. All data is adjusted, and a hospital must have a minimum of 20 episodes in a service line for it to be displayed in the report. The MVC Coordinating Center will also host a series of virtual workgroups based on input from its hospital partners. The primary goal of these workgroups is to provide hospital leaders with a highly accessible platform to share best practices and challenges facing hospitals throughout the state of Michigan. The ideas and strategies outlined in these discussions will serve as a foundation and framework for collaborative learning and best practice sharing at MVC meetings. The MVC Coordinating Center will also continue its work to improve the utility of the MVC data registry website and host semiannual meetings to prove a venue for the sharing of best practices and additional insights. Anticipating unintended consequences Despite best intentions, there are possible unintended consequences that can result from efforts and metrics aimed at improving quality and reducing costs in complicated health care settings. The MVC development group and external advisors considered possible unintended outcomes that could ensue from the structure of this proposed P4P measure. These discussions also focused on potential interventions that could be applied to minimize the risk of such adverse outcomes. P4P Measure Methodology Report 28 July 2017

29 Several potential unintended consequences, and proposed interventions intended to protect against such outcomes, are outlined in Table 4. Table 4: Potential unintended consequences and proposed interventions Possible unintended consequence Incentivizing competition Rewarding low quality hospitals Rewarding cost improvements that result from inappropriate reduction in services provided Rise in number of low-cost episodes driven by changes in admission thresholds Hospitals not responding to P4P measure because bonus payment is out of balance with efforts required to achieve improvements or opportunity costs associated with improvement efforts Focus on costs incurred outside of the hospital to the exclusion of improvements in hospital-based services Hospitals not responding to P4P measure because of lack of additional shared savings Proposed intervention(s) Initial focus on internal improvement and participation; include collaboration goal to encourage statewide collaboration and reward statewide improvement Include quality minimums that must be met to be eligible for performance bonus Include quality minimums that must be met to be eligible for performance points; possible future inclusion of CQI-based quality measures Monitoring year-to-year changes in hospital-level case mix index for selected service lines Ongoing analyses evaluating balance between bonus amount vs. reduction in costs Include index hospitalization costs and readmissions in the P4P measure Implement a group collaboration bonus for achieving state-wide improvement goals P4P Measure Methodology Report 29 July 2017

30 Appendices Appendix A: Measure Development Team Name Position Michigan Value Collaborative Coordinating Center Jim Dupree, MD, MPH Scott Regenbogen, MD, MPH Chad Ellimoottil, MD, MS Kristyn Vermeesch, MPP John Syrjamaki, MPH Edward Norton, PhD Director Co-Director Program Associate, Analytics Project Manager Associate Project Manager/Senior Analyst Program Economist External advisors Andrew Ryan, PhD Ellen Ward, MHSA John Ayanian, MD, MPP Anup Das Associate Professor, University of Michigan School of Public Health Manager, BCBSM Value Partnerships Program Director, University of Michigan Institute for Healthcare Policy and Innovation PhD candidate, University of Michigan School of Public Health P4P Measure Methodology Report 30 July 2017

31 Appendix B: Standard Inclusions Stroke + Transient Ischemic Attack (TIA) Sepsis/Infection Urinary Tract Infection (UTI) Acute Myocardial Infarction (AMI) P4P Measure Methodology Report 31 July 2017

32 Pneumonia Pulmonary Embolism (PE) Deep Vein Thrombosis (DVT) Acute gastrointestinal ulcerative disease Pressure Ulcers P4P Measure Methodology Report 32 July 2017

33 Electrolyte Imbalance Debility, malaise, fatigue, weakness Complications of surgical and medical care, not elsewhere classified E Pneumothorax, plural effusions Medication effects Aftercare V5789 V571 V5849 P4P Measure Methodology Report 33 July 2017

34 Acute exacerbations of chronic diseases Diabetes Mellitus (DM) Asthma Chronic Obstructive Pulmonary Disease (COPD) Congestive Heart Failure (CHF) Renal failure Hypertension P4P Measure Methodology Report 34 July 2017

35 Appendix C: Hierarchical Condition Categories Condition Categories Acute Myocardial Infarction Acute Renal Failure Amputation Status Complications Amyotrophic Lateral Sclerosis Angina Pectoris Artificial Openings for Feeding or Elimination Aspiration and Specified Bacterial Pneumonias Atherosclerosis of the Extremities Bone/Joint/Muscle Infections/Necrosis Breast, Prostate, and Other Cancers Cardio-Respiratory Failure and Shock Cerebral Hemorrhage Cerebral Palsy Chronic Hepatitis Chronic Kidney Disease, Stage 4 Chronic Kidney Disease, Stage 5 Chronic Obstructive Pulmonary Disease Chronic Pancreatitis Chronic Ulcer of Skin, Except Pressure Cirrhosis of Liver Coagulation Defects Colorectal, Bladder, and Other Cancers Coma, Brain Compression Complications of Implanted Device Congestive Heart Failure Cystic Fibrosis Depressive, Bipolar, and Paranoid Disorders Diabetes with Acute Complications Diabetes with Chronic Complications Diabetes without Complication Diabetic Retinopathy and Vitreous Hemorrhage Dialysis Status Disorders of Immunity Drug/Alcohol Dependence Drug/Alcohol Psychosis Endocrine and Metabolic Disorders Hemiplegia/Hemiparesis Hip Fracture/Dislocation HIV/AIDS Inflammatory Bowel Disease Intestinal Obstruction/Perforation Ischemic or Unspecified Stroke Lung and Other Severe Cancers Lymphoma and Other Cancers Major Head Injury Major Organ Transplant or Replacement Status Metastatic Cancer and Acute Leukemia Monoplegia, Other Paralytic Syndromes Morbid Obesity Multiple Sclerosis Muscular Dystrophy Myasthenia Gravis/Myoneural Disorders Opportunistic Infections Paraplegia Parkinson's and Huntington's Diseases Pneumococcal Pneumonia, Empyema, Lung Abscess Pressure Ulcer of Skin with Full Skin Loss Pressure Ulcer of Skin with Necrosis Protein-Calorie Malnutrition Quadriplegia Respirator Dependence Respiratory Arrest Rheumatoid Arthritis Schizophrenia Seizure Disorders and Convulsions Septicemia or Sepsis Severe Head Injury Severe Hematological Disorders Severe Skin Burn or Condition Specified Heart Arrhythmias Spinal Cord Disorders/Injuries Traumatic Amputations and Complications P4P Measure Methodology Report 35 July 2017

36 End-Stage Liver Disease Exudative Macular Degeneration Fibrosis of Lung Unstable Angina, Acute Ischemic Heart Disease Vascular Disease Vascular Disease with Complications Vertebral Fractures P4P Measure Methodology Report 36 July 2017

37 Appendix D: Condition Specific Risk-Adjustment Variables AMI Colectomy CABG lvad colectomy_other cabg_reop cardiac_surgery colectomy_inflam_bowel arterial_graft iabp iabp iabp lvad lvad emergency_intubation pci pci pci ptca ptca electrophys_test ptca ptca electrophys_test electrophys_test ptca colectomy_vasc_insuf colectomy_volvulus colectomy_diverticulitis colectomy_gibleed CHF, Pneumonia, Hip Replacement, Knee Replacement, and Other Spine: No condition-specific risk adjustors. P4P Measure Methodology Report 37 July 2017

38 Appendix E: Expected Total Episode Cost 30-day Regression Variables Colectomy Variables female agegrp large_prior_total bcbsm colectomy_diverticulitis colectomy_gibleed colectomy_inflam_bowel colectomy_lap colectomy_open colectomy_ostomy colectomy_other colectomy_vasc_insuf hcc_arrhythmia hcc_artificial_openings hcc_bact_pneumonia hcc_bone_muscle_infect hcc_cardio_resp_failure hcc_chf hcc_chron_kidney_stg5 hcc_chron_pancreatitis hcc_coag_defects hcc_copd hcc_diabetes_no_cmp hcc_drug_alcohol_depend hcc_drug_alcohol_psych hcc_endo_metab_othr hcc_hemiplegia hcc_immunity_disorder hcc_intest_obstruct hcc_lower_amputation_cmp hcc_lymphoma hcc_major_depressive hcc_met_cancer_leuk hcc_morbid_obesity hcc_opp_infect hcc_parkins_huntingtons hcc_pdr hcc_press_ulcer_necrosis hcc_pro_cal_malnutr hcc_renal_failure hcc_rheum_arthritis hcc_schizophrenia hcc_seizure hcc_septicemia hcc_spinal_cord CABG Variables female agegrp large_prior_total bcbsm bypass cabg_reop hcc_ami hcc_angina_pectoris hcc_arrhythmia hcc_atherosclerosis hcc_cardio_resp_failure hcc_chf hcc_chron_kidney_stg4 hcc_chron_kidney_stg5 hcc_coma hcc_copd hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_ibd hcc_implant_cmp hcc_lymphoma hcc_macular_degen hcc_major_depressive hcc_morbid_obesity hcc_parkins_huntingtons hcc_pro_cal_malnutr hcc_renal_failure hcc_rheum_arthritis hcc_septicemia hcc_severe_hemat hcc_spinal_cord hcc_stroke hcc_trauma_amputation hcc_vascular hcc_vascular_cmp P4P Measure Methodology Report 38 July 2017

39 P4P Measure Methodology Report 39 July 2017 AMI Variables female agegrp large_prior_total bcbsm cardiac_surgery electrophys_test emergency_intubation hcc_arrhythmia hcc_atherosclerosis hcc_bact_pneumonia hcc_bone_muscle_infect hcc_cardio_resp_failure hcc_chf hcc_chron_kidney_stg4 hcc_cirrhosis_liver hcc_copd hcc_diabetes_acute_cmp hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_endo_metab_othr hcc_hemiplegia hcc_hip_fracture hcc_implant_cmp hcc_morbid_obesity hcc_organ_trans hcc_paraplegia hcc_pro_cal_malnutr hcc_renal_failure hcc_resp_depend hcc_seizure hcc_stroke hcc_vascular hcc_vascular_cmp hemodialysis lvad pci ptca Pneumonia Variables female agegrp large_prior_total bcbsm hcc_ami hcc_arrhythmia hcc_artificial_openings hcc_atherosclerosis hcc_bact_pneumonia hcc_bone_muscle_infect hcc_cardio_resp_failure hcc_cerebral_hem hcc_chf hcc_chron_kidney_stg5 hcc_chron_ulcer hcc_coag_defects hcc_copd hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_dialysis hcc_drug_alcohol_psych hcc_end_stage_liver hcc_endo_metab_othr hcc_hemiplegia hcc_ibd hcc_immunity_disorder hcc_implant_cmp hcc_lower_amputation_cmp hcc_lung_cancer hcc_lymphoma hcc_major_depressive hcc_met_cancer_leuk hcc_morbid_obesity hcc_ms hcc_opp_infect hcc_organ_trans hcc_paraplegia hcc_parkins_huntingtons hcc_pdr hcc_pneumococcal hcc_pro_cal_malnutr hcc_quadriplegia hcc_renal_failure hcc_schizophrenia hcc_seizure hcc_septicemia hcc_severe_burn hcc_severe_hemat hcc_spinal_cord hcc_stroke hcc_unstable_angina hcc_vascular hcc_verteb_fracture pneumonia_intubation pneumonia_mechvent pneumonia_trach

40 CHF Variables female agegrp large_prior_total electrophys_test emergency_intubation hcc_ami hcc_arrhythmia hcc_atherosclerosis hcc_cardio_resp_failure hcc_cerebral_palsy hcc_chron_hep hcc_chron_kidney_stg4 hcc_chron_kidney_stg5 hcc_chron_ulcer hcc_copd hcc_crectal_bldr_cancer hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_dialysis hcc_end_stage_liver hcc_endo_metab_othr hcc_hemiplegia hcc_ibd hcc_implant_cmp hcc_lung_cancer hcc_lung_fibrosis hcc_major_depressive hcc_met_cancer_leuk hcc_morbid_obesity hcc_muscular_dystrophy hcc_opp_infect hcc_organ_trans hcc_paraplegia hcc_parkins_huntingtons hcc_pneumococcal hcc_press_ulcer hcc_press_ulcer_necrosis hcc_pro_cal_malnutr hcc_quadriplegia hcc_renal_failure hcc_rheum_arthritis hcc_seizure hcc_septicemia hcc_severe_hemat hcc_vascular hcc_vascular_cmp hcc_verteb_fracture ptca Hip Replacement Variables female agegrp large_prior_total bcbsm hcc_arrhythmia hcc_artificial_openings hcc_bact_pneumonia hcc_breast_prost_cancer hcc_chf hcc_chron_kidney_stg4 hcc_chron_ulcer hcc_coag_defects hcc_copd hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_dialysis hcc_endo_metab_othr hcc_hemiplegia hcc_hip_fracture hcc_immunity_disorder hcc_lymphoma hcc_macular_degen hcc_major_depressive hcc_major_head_injury hcc_monoplegia hcc_morbid_obesity hcc_ms hcc_parkins_huntingtons hcc_pro_cal_malnutr hcc_rheum_arthritis hcc_schizophrenia hcc_seizure hcc_severe_hemat hcc_vascular hcc_verteb_fracture P4P Measure Methodology Report 40 July 2017

41 Knee Replacement Variables female agegrp large_prior_total bcbsm hcc_angina_pectoris hcc_arrhythmia hcc_bone_muscle_infect hcc_chf hcc_chron_kidney_stg4 hcc_chron_kidney_stg5 hcc_chron_ulcer hcc_coag_defects hcc_copd hcc_diabetes_chron_cmp hcc_diabetes_no_cmp hcc_dialysis hcc_drug_alcohol_psych hcc_end_stage_liver hcc_endo_metab_othr hcc_hemiplegia hcc_hip_fracture hcc_ibd hcc_implant_cmp hcc_lung_fibrosis hcc_macular_degen hcc_major_depressive hcc_major_head_injury hcc_morbid_obesity hcc_myasthenia_gravis hcc_paraplegia hcc_parkins_huntingtons hcc_pro_cal_malnutr hcc_renal_failure hcc_resp_depend hcc_schizophrenia hcc_seizure hcc_septicemia hcc_severe_head_injury hcc_severe_hemat hcc_unstable_angina hcc_vascular hcc_vascular_cmp hcc_verteb_fracture Spine Variables female agegrp large_prior_total bcbsm cervical_disc_degen_dx cervical_spinal_sten_dx hcc_angina_pectoris hcc_cardio_resp_failure hcc_cerebral_palsy hcc_chf hcc_chron_hep hcc_chron_kidney_stg5 hcc_coag_defects hcc_copd hcc_diabetes_chron_cmp hcc_drug_alcohol_depend hcc_endo_metab_othr hcc_hemiplegia hcc_implant_cmp hcc_lymphoma hcc_major_depressive hcc_monoplegia hcc_morbid_obesity hcc_organ_trans hcc_paraplegia hcc_parkins_huntingtons hcc_pneumococcal hcc_pro_cal_malnutr hcc_quadriplegia hcc_renal_failure hcc_rheum_arthritis hcc_spinal_cord hcc_stroke hcc_vascular hcc_vascular_cmp hcc_verteb_fracture lumbar_disc_dx lumbar_spinal_stenosis_dx lumbar_spondyl_dx myelopathy P4P Measure Methodology Report 41 July 2017

42 Appendix F: Value-based incentive programs from the Centers for Medicare & Medicaid Services (CMS) Bundled Payment for Care Improvement (BPCI) BPCI is a voluntary program developed and implemented by CMS to test the effectiveness of bundling payments for episodes of care. There are four separate models distinguished by the episode length, number of providers involved in the program, and how payments are distributed. Nationally, more than 2,115 hospitals, post-acute care facilities, and other providers are currently participating in BPCI. Hospital Value-Based Purchasing (HVBP) HVBP is a program developed by CMS that links Medicare payments to various quality and cost-efficiency measures. At present, value-based payments are distributed to over 3,500 hospitals. The magnitude of the value payments depends on hospital performance with measures defined by four specific categories: efficiency (20%), clinical process of care (20%), outcome (30%), and patient experience of care (30%) The specific efficiency measure used in this program is Medicare Spending per Beneficiary (MSPB). MSPB is a claims-based measure that is calculated using risk-adjusted, price-standardized payments for episode of care comprising the index hospital stay plus three days prior to admission and 30 days post-discharge. Condition-specific episode payment measures for AMI, Heart Failure, and Pneumonia CMS recently designed and released episode payment measures for three common medical diagnoses. The measure calculates risk-adjusted payments for 30-day episodes of care. At present these measures are not tied to value-based payments, however, they are publically reported as part of the Hospital Compare program. Comprehensive Care for Joint Replacement (CCJR) The proposed CCJR model was developed by CMS to test episode-based bundled payments for joint replacement surgery. Unlike BPCI (which is a voluntary program), CMS selected 75 geographic regions (metropolitan statistical areas) to participate in in CCJR. The participants include a wide range of hospitals with varying baseline costs. In this model, CMS will establish target episode prices and provide year-end reconciliation payments for hospitals that provide joint replacement at a lower cost. Hospitals that exceed the target price will be required to pay the difference back. To be eligible for reconciliation payments, hospitals must simultaneously meet quality standards for complications, readmissions and patient satisfaction. The table below provides additional details for each program. P4P Measure Methodology Report 42 July 2017

43 Summary characteristics of CMS value-based incentive programs BPCI (Model 2) MSPB (HVBP) AMI/HF*/Pneumonia Episode payment measure CCJR Episode Length 30, 60, or 90 days post discharge 3 days prior to index admission through 30 days post discharge 30 day episode of care beginning with a hospitalization 90 day episode Episode Triggers Inpatient admission of eligible beneficiary to acute care hospital for one of the MS-DRGs in a selected episode Index inpatient hospital admission Index admission for AMI/HF/Pneumonia MS-DRG 469 or 470 Episode Inclusion Captures payments for all care covered under Medicare Part A and Part B within time of episode Captures payments for all inpatient, outpatient and postacute care claims Captures payments for all inpatient, outpatient and post-acute care claims Captures payments for all inpatient, outpatient and postacute care claims Medical/Surgical Both Both Medical Surgical Readmissions Included Included Included Included *HF Heart Failure P4P Measure Methodology Report 43 July 2017

44 Appendix G: Reliability calculations for minimum case requirements The reliability of a metric is the extent to which the variation in the measure represents actual change versus statistical noise. We were interested in studying whether year-to-year variation in collaborative-wide mean cost for a particular service line could be reliably distinguished from random variation. For this purpose, we calculated the reliability ratio for all eligible service lines using data from BCBSM and Medicare patients from The reliability ratio is measured on a scale from 0 (poor reliability) to 1 (good reliability). The table below presents reliability ratios for each eligible service line stratified by annual hospital case volume thresholds. Collectively, these data show that the aggregated (i.e., collaborative-wide) episode payment measure is very reliable at our 10 case minimum for all service lines, with ratios generally in the range of 0.7 to 0.9. There was minimum improvement in reliability with a higher case threshold. Reliability ratios for eight MVC service lines Condition All hospitals Hospitals with >10 cases Hospitals with >15 cases Hospitals with >20 cases Hospitals with >25 cases Hospitals with >30 cases Reliability Reliability Reliability Reliability Reliability Reliability n* ratio n* ratio n* ratio n* ratio n* ratio n* ratio CABG AMI Pneumonia CHF Hip replacement Knee replacement Spine surgery Colectomy *n represents the number of hospitals that fulfill the minimum case threshold reported in each column P4P Measure Methodology Report 44 July 2017

45 Appendix H: Application of proposed scoring system assuming use of both BCBSM PPO and Medicare Data (example) Measure theme P4P measure structure and scoring system and 2019 Transition Year Year over year or improvement Deep engagement with MVC data Absolute achievement Measure scoring Collaboration goal 2 pts - Participate in site visit with MVC 1 pt - Provide feedback on risk adjustments 1 pt - Provide feedback on P4P measure 1 pt - Attend MVC meetings 5 pts - Create specific QI action plans for 2017 No Score ^ = scoring for each service line assumes use of both BCBSM PPO and Medicare claims SD = standard deviation Program Year 2016 Hospital A selects pneumonia and joint replacement as its two service lines. Baseline: Hospital service line total episode costs for the most recently available 12 month period 2018/2019 scoring:^ 1 pt = baseline mean 2 pts = baseline mean * SD based target 3 pts = baseline mean * SD based target 4 pts = baseline mean * SD based target 5 pts = baseline mean * SD based target Baseline: MVC-wide service line total episode costs for the most recently available 12 month period 2018/2019 scoring:^ 50 th %tile = 1 pt 60 th %tile = 2 pts 70 th %tile = 3 pts 80 th %tile = 4 pts 90 th %tile = 5 pts +1 pt per service line if 5% improvement among hospitals choosing the specific service line In 2016, Hospital A participates in a site visit (2 pts), attends the MVC meetings (1 pt), and creates a specific QI action plan for improving efficiency of its pneumonia and joint replacement service lines (5 pts). However, Hospital A did not provide feedback on risk adjustment methods. In addition, Hospital A performs above the 10 th percentile on the mortality and related readmission measures, thereby meeting the quality requirements. Therefore, Hospital A will earn 8 out of 10 possible points for P4P Measure Methodology Report 45 July 2017

46 Program Years 2018 and 2019 In 2018, Hospital A again meets the quality requirement by performing above the 10 th percentile on the mortality and related readmission measure. Hospital A s 30-day mean episode costs for joint replacement are outlined below: Service line Mean costs for baseline period MVC Winsorized Standard Deviation based target for baseline period Mean costs in 2018 performance period Joint replacement $16,393 $2,100 $16,871 Hospital A s 2018 cost improvement targets for joint replacement are calculated as follows: Because Hospital A does not reduce its joint replacement costs any further in 2018, it does not earn any points for year-over-year improvement. However, because Hospital A s 2018 episode costs for joint replacement are well below its cohort s average episode cost for joint replacement (i.e., Hospital A is a high performer), it is still eligible for achievement points year over year improvement targets Points $16,393 1 $16,393 (0.05*$2,100) = $16,288 2 $16,393 (0.10*$2,100) = $16,183 3 $16,393 (0.15*$2,100) = $16,078 4 $16,393 (0.20*$2,100) = $15,973 5 Mean costs for baseline period Hospital A s Cohort $19,202 Hospital A $16, achievement targets Points $19,202 (50th percentile) 1 $18,883 (60th percentile) 2 $18,378 (70th percentile) 3 $17,502 (80th percentile) 4 $16,792 (90th percentile) 5 P4P Measure Methodology Report 46 July 2017

47 Hospital A will earn 4 points for absolute achievement because its 2018 joint replacement costs are ranked between the 80 th and 90 th percentiles. Separately, all hospitals in the state working on joint replacement reduced average episode costs by more than 5% (data not shown). However, since Hospital A s 2018 joint replacement costs were higher than its baseline costs, it would not be eligible to earn an additional bonus point as part of its collaboration goal. For 2018, Hospital A will earn a total of 4 points for its joint replacement service line. Its 2018 pneumonia service line performance would be scored separately using the same methodology. P4P Measure Methodology Report 47 July 2017

48 Appendix I: Method for calculating baseline performance data The schedule for determining baseline costs must take into account the claims rectification and distribution schedules from BCBSM and Medicare. The baseline cost will be calculated from the most recently available twelve-months of claims data and will be two-years prior to the performance year. The purpose of the twoyear window is to provide hospitals with prospective targets and allow more time for the impact of quality initiatives to be observed in episode costs. The performance costs will be derived from the most recently available twelve-months of claims data, which will be approximately twelve-months prior to the beginning of the 2017 and 2018 program years. This twelve-month lag period accounts for the time required for hospital claims submissions, processing of claims by BCBSM and Medicare, and analytics and reporting by MVC. Reporting Timeline 2015 P4P Program Year Baseline Period (Services provided 1/1/15 12/31/15) 2016 Baseline Period (Services provided 1/1/16 12/31/16) Calendar Year Performance Period (Services provided 1/1/17 12/31/17) Data Analysis/Claims Adjudication Performance Period (Services provided 1/1/18 12/31/18) 2019 Payment Applied 7/1/19 Data Analysis/Claims Adjudication 2020 Payment Applied 7/1/20 P4P Measure Methodology Report 48 July 2017

49 Appendix J: P4P Cohort Structure Figure 1. Cohorts developed for the service lines of: Joint, Pneumonia, CHF, and Colectomy. The presence of 300 or more beds and teaching status are used to distinguish between Cohorts 1 and 2. A hospital must meet both criteria to be categorized as Cohort 1. The presence of 100 beds or more and teaching status are used to differentiate between Cohorts 3 and 4. Unlike distinguishing between Cohorts 1 and 2, a hospital only needs to meet one of these criteria to be classified as Cohort 3 to ensure balance in the grouping dispersion. Figure 2. AMI Cohort Designation The AMI cohort groups are based upon clinical intervention services. N=33 N=14 N=9 P4P Measure Methodology Report 49 July 2017

50 Figure 3: CABG Cohort Designation The CABG cohort is based upon the structural characteristic of bed size. This decision was made because nearly half of MVC hospitals to not offer CABG services, and the CMI distribution does not produce distinct clusters of hospitals. N=17 N=15 Figure 4. Spine Cohort Designation The spine cohort groups are based upon whether a hospital performs complex spine surgery (as determined by Spine DRG CMI) as well as the structural characteristics of bed size and teaching status. N=15 N=17 N=18 P4P Measure Methodology Report 50 July 2017