Table of Contents. Page 2 of 29

Transcription

1 Page 1 of 29

2 Table of Contents Table of Contents 2 Overview 3 Why is This Important? 3 Defining Value 3 Representing and Displaying Value 4 Project Goals 4 Current State of Reporting on Quality and Cost Data 5 Clinical Data Collection & Analysis 5 Background 5 Current Process Overview 5 Future Considerations 6 Cost of Care Data 7 Pilot Discovery and Solution Selection Process 7 Clinical Data Format Selection 7 Clinical Data Collection Method 10 Electronic Quality Measure (ecqm) Software Selection 11 Cost of Care Data Format and Collection Methods 11 Clinical Quality and Cost Measures Selection 11 Patient Identification and Matching System Selection 14 Pilot Solution 14 Overview 14 Listing of Tools & Technologies Utilized 15 Clinical Data Collection Metrics 15 Quality Measure Implementation 16 Patient Matching 17 Patient and Provider Attribution and Minimum Sample Sizes 17 Value Calculation 17 Value Reporting 18 Results & Findings 18 Quality & Cost Data Acquisition & Aggregation 19 Patient Matching 20 Value Calculation & Display 21 Broad Applicability of Our Approach 21 Limitations 21 Discussion 22 Whitepaper Authors 24 Calculated Measures 25 Page 2 of 29

3 Overview The Center for Healthcare Transparency (CHT) led by the Network for Regional Healthcare Improvement (NRHI) and Pacific Business Group on Health (PBGH) is focused on increasing transparency on the relative cost and quality of healthcare services available to 50 percent of the U.S. population by The Laura and John Arnold Foundation (LJAF), which focuses on addressing society s most pressing and persistent challenges using evidence-based, multidisciplinary approaches, provided the source funds for this initiative. The Health Collaborative conducted one of CHT s regional innovation pilots, working to answer critical questions that will enable CHT and its partners to reach this goal. Specifically, the project described in this white paper investigated new methods for collecting and linking quality and cost data and the best strategies to present this value-based information in a meaningful way for consumers. Why is This Important? Studies show that aligning quality with costs in payment programs and quality initiatives can lead to improved patient outcomes, while maintaining or even lowering costs. 2,3,4 In particular, recent attempts at improving clinical quality and/or controlling cost in isolation have failed to effectively address provider efficiency, a key component to defining value. 5 As the healthcare industry moves to a value-based payment, it becomes essential that the measurement of value be done in an efficient and reliable fashion. This effort is not sustainable without timely, integrated, comprehensive and valid information that can be easily provided, maintained, and interpreted. This data needs to be understandable from a provider, patient/consumer, and payer perspective. It is particularly important that providers have access to the data and analyses needed to help them understand the relationship between improved clinical outcomes and cost of care for their specific subsets of patients. It is with this understanding that the data becomes most actionable. Therefore, it could be argued that, without such an infrastructure, payment reform is unsustainable. Defining Value Value in healthcare is a consumer-centered and results-focused (clinical outcomes and cost) measure. This definition differs from approaches to healthcare delivery defined by processes of care or volume of services. 6 Rather, it is a measure of a specified stakeholder s (such as an individual patient s, consumer organizations, payer s, provider s, government s, or society s) preference-weighted assessment of a particular combination of quality and cost of care performance (Figure 1). 7,8 This concept of value continues to be refined with clinical outcomes (e.g., health status) now being distinguished from patient experience. 9 Figure 1. Value 8 Page 3 of 29

4 Representing and Displaying Value Measuring and representing value in a manner that is both meaningful to consumers and informative to clinicians is a relatively new concept, but has become widely supported by stakeholders across the healthcare spectrum. For this reason, research continues to understand how best to present quality and cost in a meaningful way. In 2012, the National Committee for Quality Assurance (NCQA) specifically examined consumers understanding of healthcare value, the best way to present this value-based data, and how it would inform consumer healthcare decisions. Their focus group results showed that while education was an important component of these initiatives, individuals found value-based data to be informative and were able to make distinctions between quality of care and cost. This work did not provide specific recommendations on how to present this data; although, high-level representations of quality and cost appeared to be preferable. 10 A similar study completed by Hibbard and colleagues found that providing guidance on what is considered high-value, along with these representations of quality and cost can assist individuals in decision-making. 11 Consumer testing completed by the American Institutes for Research through a grant funded by the Robert Wood Johnson Foundation confirmed that cost and quality should be displayed together and when costs were displayed alone, consumers assumed that high costs correlated with high quality, which is not necessarily a true depiction of value. 12 Information regarding value (clinical outcomes, patient experience, and cost of care) will soon be readily available and publicly reported. However, significant challenges must be addressed with solutions that can be adopted widely. Three of these challenges are: 1) improving the ease of access to real-time, comprehensive data; 2) determining how to best display this information for consumers; and 3) linking clinical outcomes with cost of care for cohorts of patients that have the same chronic disease. Project Goals CHT and The Health Collaborative initiated this project as part of their efforts to improve the accessibility and timeliness of quality and cost data. Specifically, the project s goals were to: 1. Study and test new methods for scalable electronic clinical data reporting and linking it to administrative claims data; specifically: Determine whether the Continuity of Care Document (CCD), which is a document template under the Consolidated-Clinical Document Architecture (C-CDA) standard or the Quality Reporting Document Architecture (QRDA) are feasible and cost effective methods for automating electronic clinical quality measure calculation. Both documents are based on the Clinical Document Architecture (CDA) standard. Determine the clinical quality and cost measures that can be generated based on available data Identify methods for data collection and extraction that are more easily scalable by others, relying on less manual processes Identify how to best combine quality and cost data through patient matching software and study the correlation between the two Capture the benefits and limitations to the approach 2. Study new methods for providing value information to consumers in a meaningful way and aggregated from multiple payers to obtain a more comprehensive provider profile Identify how to best display value quality and cost together Vet this proof of concept with key advisors Page 4 of 29

5 As a result of this work, we developed a series of findings and recommendations for Regional Health Improvement Collaboratives (RHICs), policymakers, national thought leaders and other stakeholders, and identified where further work remains. Current State of Reporting on Quality and Cost Data Many groups, including RHICs, collect and display quality (primarily process measures) and/or patient experience data on ambulatory practices and hospitals. Some entities expanded into providing information on the costs associated with a specific condition or the total cost of care. The vast majority of this cost and quality information is not displayed in relationship to each other in order to represent value. For the few state or community-based initiatives that do report quality and cost together, the information is primarily limited to one setting (e.g., hospital) or one condition (e.g., maternity care). Also, the relationships rely on a weaker correlation in time between an averaged clinical outcome and an averaged overall cost of care for a given condition. Innovative methods are needed to address one of the key barriers faced when attempting to move forward on these initiatives expanding the access to and enhancing the richness and timeliness of the quality and cost data. Clinical Data Collection & Analysis Background Past research shows that data obtained from the gold standard the medical record is vastly superior to administrative claims data in its reliability and validity, but the time and resources required to acquire medical record data has limited its widespread use. 13,14,15 Frequently, administrative data was used for this purpose because it was more cost-effective and provided a larger representation of the patient population for which a physician, hospital, or other healthcare entity provided care. Unfortunately, this use of claims data constrained quality measurement to processes of care. Replacing or supplementing claims data with medical record data enables access to richer information including patient outcomes. Electronic health record (EHR) data is superior to paper medical records since there is the potential for access to real-time information across the entire patient population rather than a point-in-time picture gathered from a representative sample through chart abstraction. There is great opportunity to leverage EHR technology as the source of truth now that EHR adoption is more widespread. Current Process Overview The Health Collaborative leads a voluntary public reporting and performance measurement initiative called Your Health Matters (YHM). Representatives of independent medical practices, federally-qualified health centers (FQHCs), and health systems submit clinical data from EHR systems and/or data repositories to The Health Collaborative s secure data portal. Over 180 adult primary care and internal medicine practices across 20 counties in southwest Ohio, northern Kentucky, and southeastern Indiana voluntarily submit medical record data to compute and report on select clinical quality measures. These measures are based on NQF-endorsed measures with modifications on which the clinical community agrees. Data are currently collected for the following clinical conditions and measures: diabetes, ischemic vascular disease, colorectal cancer screening, high blood pressure, and the CG-CAHPS patient experience survey. Results are publicly reported on the YHM consumer web site at the practice level along with an aggregate weighted average of all participating practices in the region. As with other public reporting initiatives, The Health Collaborative works with the data available to them. The Health Collaborative began reporting on the quality of care provided by primary care practices and hospitals for specific health conditions in 2009 and patient experiences in To date, clinical data for the quality measures are derived from EHRs by data query, manual chart abstraction or a combination of the two to ensure accuracy and completeness. Since the process requires significant manual effort, the frequency of reporting is limited. The Health Collaborative is only able to generate a report once per year for any given metric as the process quickly overwhelms the practice personnel given the number of measures presently reported. Page 5 of 29

6 In general, the methods used to collect the clinical data needed for public reporting consist of multiphased processes that are very labor intensive and highly detailed. Each reporting organization is required to submit a flat file (spreadsheet) of clinical data obtained through an iterative process of manual or semi-automated data retrieval from EHRs and/or clinical data repositories. Before submission, medical groups must review the measure specifications, develop data extraction queries, and perform manual abstraction on missing data not discoverable by query. After extraction, they must also review the data for accuracy, clean the file of random formatting issues and errors, and revise any inaccuracies in the data before submission for scoring and analysis. The governance of the measures selected for reporting is managed by a group of clinicians in the community. The measures selected are all based on standard measure logic, but have been modified. While this modification provides what best suits this community, it does not allow for an apples to apples comparison of measures across the country with other organizations. The manual effort, chart abstraction process, and refining of the data allow for measures that are very accurate, but do not resolve the core issue of poor data collection in the various EHR systems. As quality reporting becomes more standardized, accurate collection of data in the EHR will only become a more critical component. The data are typically extracted by SQL queries developed by data analysts and/or manually abstracted by quality improvement department staff who must review each patient s chart to fill in missing or inaccurately documented data, or both. Due to the number of data elements required for some measures, such as the composite measure for assessing diabetes care, several queries must be developed and tested to retrieve all clinical data required for scoring. Future Considerations Leveraging EHR data is clearly the solution to ensuring data for the quality measures are broadly representative of the provider s performance; there is great potential for real-time access to this data that is not yet realized. While automating the calculation of electronic clinical quality measures (ecqms) will not completely eliminate the need for data review, verification and validation, it will streamline the data collection process by reducing the number of steps and iterations as well as the people involved. 16 Additionally, it will reduce the number of opportunities for human error at each step and mitigate the impact of staff turnover in the participating medical groups. Often, these medical groups require significant ongoing training for new lead personnel and IT staff to ensure consistent understanding of the measure specifications and reporting requirements. ecqms enable the capture of this information by defining the patient population and clinical aspects of care being measured in a standardized, machine-readable format with the associated coding and logic. ecqms use nationally accepted standards, which enable consistent implementation of measures across multiple healthcare entities. Among these national standards, the Health Quality Measure Format (HQMF) standard defines a measure including the clinical data elements necessary to calculate the measure as well as the various value sets. The clinical data is described in a standardized format called the Quality Data Model, which was initially developed by NQF and now managed by the Centers for Medicare and Medicaid Services (CMS). Finally, the Quality Reporting Document Architecture (QRDA) provides document format standards to support the exchange of ecqm aggregated results and/or patient-level data for purposes of calculation.. 17,18,19 The Office of the National Coordinator for Health Information Technology (ONC) adopted these standards for use in the EHR certification program. 20 ONC adopted QRDA as the standard to support both QRDA Category I (individual patient) and QRDA Category III (aggregate) data submission approaches. 21 Both of these standards are designed to enable the capture and exchange of patient-level data. This data not only allows an entity to calculate the overall performance (numerator and denominator) for the quality measures but it also provides the opportunity for more in-depth analysis. Meaningful Use Stage 3 further emphasizes this ability to drill down to the individual patient level, while still enabling aggregation of data for populations. Page 6 of 29

7 Cost of Care Data In conjunction with efforts to collect and analyze clinical data, groups continue to examine the best methodologies to capture data on the costs accrued for a specific clinical condition, procedure or time period. Organizations interested in understanding and evaluating how well practices, hospitals and other providers perform developed measures on total cost of care or resource use for specific clinical conditions or procedures. These measures include HealthPartners Total Cost of Care, NCQA s Relative Resource Use, and the CMS Medicare Spending per Beneficiary Other potential approaches to capturing and understanding the costs incurred exist and continue to evolve. For example, many recommend that this information focus on shoppable" healthcare services those healthcare services that are often less complex (e.g., prescriptions, colonoscopies), elective (e.g., joint replacement, maternity care), and/or do not require quick decisions (i.e., nonemergent). 22 Administrative claims serve as the primary source of this data and provide information on the resources used by a provider, hospital, or other healthcare entity. This data is often difficult for those outside of a health plan to access and delays in processing the claims limit the data s timeliness. The amount of time between service delivery and availability of a claim has been largely predicated on the needs of the health plan to accomplish the most accurate payment. More timely information could be gathered if payers would agree to provide claims after a 30- or 60-day processing delay rather than the more typical 90-day delay. This reduction in time could move us closer toward real-time data, particularly in light of findings from a 2013 study by America s Health Insurance Plans: 98% of all claims were processed within 30 days and 99% within 60 days of receipt from the provider. Additional analyses should be conducted to understand whether this improvement in processing could yield earlier access to cost data without compromising data accuracy. It could still provide adequate cost and utilization information, but would also allow more timely correlation with quality data for value reporting purposes. Some costs, such as outof-pocket expenses for patients, should also be included but cost data derived from claims is limited in its ability to provide this information beyond patient co-pays. Work continues to ensure that cost data is aggregated and attributed appropriately to providers and addresses these ongoing challenges due to data limitations. Beyond the timeliness concern, health plans are very sensitive to sharing their utilization and cost data, presenting a significant hindrance to comprehensive reporting. Pilot Discovery and Solution Selection Process Clinical Data Format Selection Implementation of a cost effective ecqm system requires that patient data be readily available in standardized formats. Standardization not only implies consistency amongst various implementers but lowers the overall interface cost required to normalize or map data to a common format. Both the ONC Certification program and Meaningful Use program suggest the use of the QRDA Category I format as the mechanism to supply patient-level data to an ecqm calculation system. However, both programs also implement other data formats for alternative use cases such as data portability, transitions of care, and patient download. In this project, we set out to test if an alternative (and potentially more readily available) clinical data format could be leveraged as the patient-level data source. While the QRDA is a format many consider the document best suited to communicate quality measure data, there are limitations and constraints to this format. Below is a list of pros/cons between QRDA Category I and CCD in table 1 and an example diagram in figure 2. Page 7 of 29 Function QRDA Cat I CCD MU Requirement for EHR to generate Yes Yes Simple for EHR to configure No Yes Less Privacy Issues Yes No Ability to generate multiple measures per document type No Yes Ability to use document for many other uses No Yes Table 1. Pros and Cons of QRDA Category I and CCDs

8 CCD/QRDA CAT I Workflow Comparison One Source, Three Measures CCD Workflow EHR QRDA CAT I Workflow EHR CCD QRDA Measure 1 File QRDA Measure 2 File QRDA Measure 3 File Measures Engine Measures Engine QRDA Measures 1-3 MU Measures ACO Measures Custom Measures QRDA Measure 1-3 Figure 2. QRDA/CCD Workflow Comparison The ONC Certification Program ensures that a certified EHR system has the capability to generate and send patient-level data for various use cases. In this project we chose to test the patient-level data that is sent for purposes of care coordination during a care transition. Specifically, we leveraged the following 23 : Certification Category Criterion Description Summary Type Care Coordination Transition of Care (b) (b)(1)&(2) When transitioning a patient to another care setting, the EP or EH/CAH should provide a summary of care record Transition of Care Aka: Referral Summary Continuity of Care Document (CCD) Page 8 of 29

9 The certification criteria further specify the types of patient data that should be included in the CCD document and it is generally referred to as the MU Common Data Set. The data included the following 23 : Patient Name Care Team Members Laboratory Values Gender Medications Procedures Date of Birth Medication Allergies Vital Signs Race Care Plan Provider Ethnicity Problems Discharge Instructions (Inpatient) Preferred Language Laboratory Tests Immunizations By choosing the CCD, we simplified the implementation requirements for the providers and their EHR vendors because they already have mechanisms and processes in place to automate the sending of CCDs. Furthermore, the nature of QRDA Category I documents requires an extra level of customization that we sought to avoid as part of our cost effective measurement system. Figure 3. Relationships between the templates and standards. The data importer in our ecqm system receives patient data as CDA-based xml documents and parses them into a database. The base format of CCDs for transitions of care is derived from a template specified in the C-CDA. The C-CDA defines a range of document templates that contain various levels of data for specific use cases. The C-CDA is a constraint of the Clinical Document Architecture (CDA) that defines the common format for clinical data. Both the C-CDA and the QRDA Category I standards use the same underlying formats defined by the CDA. For this reason, we were able to feed the CCDs used from Transitions of Care into the same calculation system that accepts QRDA Category I. The clinical and claims data used in this project was provided by TriHealth, a Cincinnati-area integrated healthcare system formed as a partnership between Good Samaritan Hospital and Bethesda Hospital. TriHealth is a full-service, not-for-profit health system that provides a wide range of clinical, educational, preventive and social programs. TriHealth is one of the top integrated health systems in Greater Page 9 of 29

10 Cincinnati and is the fourth largest employer in the city of Cincinnati, with approximately 11,500 employees. TriHealth indicated that the CCDs used in this project as the source of patient-level data were easier to configure and a standard output from their EHR system, saving time and reducing cost and staff effort. We received CCDs from a wide range of encounter locations and encounter types: Organization Type Percentage Surgical Centers 32% Hospitals 35% Primary Care Offices 33% Table 2. Origin of CCDs All of the CCDs received are considered episodic because they cover the care of a patient during a single encounter such as a hospitalization or office visit. We did not request any particular clinical data element from TriHealth but rather asked for the default CCD implementation to be used. The purpose was to determine the types of data and measures available to us with the least amount of technical work imposed on TriHealth and their EHR vendor. It was critical for us to find the most efficient solution to obtain data from a health system to ensure the pilot provided a replicable solution for other communities across the country. Table 3 shows the data types and the average amount of data available across the patient population. Data Type Average Number of Entries Encounters 1* Medications 4 Allergies 3 Procedures 1 Care Goals 0 Conditions (Problems) 2 Vital Signs 5 Laboratory Results 3 Social History 0 Immunizations 0 Table 3. Data included. We received CCDs per day. (*The EHR was configured such that each CCD covered one encounter.) Clinical Data Collection Method The CCDs from TriHealth s Epic EHR supported several mechanisms to transport the data, and ultimately we used the DIRECT protocol. TriHealth has been an existing user of The Health Collaborative Health Information Service Provider (HISP) service, which provided connectivity via DIRECT XDM/XDR methods. These methods were in place to support TriHealth s MU initiatives by allowing a provider to send and receive a CCD without having to leave the Epic EHR. This connectivity method and protocols are in compliance with ONC s MU criteria. We also evaluated the possibility of using FHIR for collecting clinical data; however, TriHealth s Epic EHR does not yet have any FHIR services available. We modified the existing process because the pilot required a machine-to-machine approach of automatically sending messages at the close of each encounter. The Epic solution was configured to send all CCDs to a single DIRECT address, which was then sent to The Health Collaborative s DIRECT solution, MirthMail. The MirthMail solution was modified to copy all CCDs to a secure file location, where they could be used for later processing. It is critical moving forward to ensure that CCDs can be sent automatically and that the DIRECT solution is able to extract the data to another location for additional processing. Page 10 of 29

11 Electronic Quality Measure (ecqm) Software Selection Various software tools are available to compute and report on ecqms and other clinical metrics. For example, CitiusTech s BI-Clinical, Novobi CQM Engine and Spectramedix all provide ecqm calculations using proprietary software platforms. 24,25,26,27 For this project, we selected pophealth, an open source software that is widely available, maintained by a community of users, and requires no licensing fees. The pophealth software was originally developed by the ONC as an offshoot of the technology used to certify electronic health records. pophealth is now part of the Population Health Analytics Suite stewarded by the Open Source Electronic Record Alliance ( We selected pophealth for two reasons: 1. It has the potential to be the lowest overall cost solution 2. Because it is open source, it has the best chance of adding new enhancements and innovations to meet the evolving needs of pay-for-performance organizations. We installed pophealth and used it to perform ecqm measurement calculations against a stream of incoming patient data, which was acquired specifically for this project by The Health Collaborative. pophealth consists of a set of tools to import patient data, import ecqm definitions and calculate quality measures. We leveraged the template nature of the C-CDA based QRDA Category I to push CCDs into the file importer of the pophealth system. Cost of Care Data Format and Collection Methods As discussed previously, obtaining cost data from health plans is an arduous process often resulting in either no data, or incomplete data without the cost. Our approach for this pilot was to work with TriHealth to obtain claims for their self-insured patients. This allowed us to work with TriHealth and their health plan directly for access to the data. We approached several other plans for data in addition to evaluating whether CMS fee-for-service data (obtained through the Qualified Entity program) could be used but the initial discussions made it clear that the process for obtaining the data for this use would be longer than the allotted time for this pilot. As there is no widely available claims format for use in analysis, the data arrived in flat file format defined by the health plan. There are significant efforts underway by NRHI and the All-Payer Claims Database (APCD) Council to work towards a standard file format, but neither of these formats were supported by the health plan from TriHealth. Data transmission was accomplished with the sftp format, which is widely supported across the healthcare industry for transmitting protected health information. Clinical Quality and Cost Measures Selection Selecting measures that best represent the quality and cost components of the value equation was key in this project. Part of this work was to determine whether leveraging data from the CCDs would enable measurement that was clearly linked to improving patient care using data that is typically time-intensive to collect and not necessarily available real-time. A set of criteria was developed to evaluate various measures currently used in Meaningful Use and in the Physician Quality Reporting System (PQRS). Two hundred and fifty-five measures were evaluated and rated against these criteria: 1. Was the measure currently specified as an ecqm (i.e., used in MU program)? 2. Were the data elements available in CCDs? 3. Is the aspect of care or outcome being measured effective (i.e., is there potential to improve patient care)? 4. Does the measure apply to a broad initial patient population? 5. What is the complexity score of the ecqm rated by the BONNIE tool? 6. Would the measure be considered Holy Grail (i.e., measures ranked as highly effective and leverage multiple data sources and/or require interoperability or data sharing that previously may not have been feasible without the use of tools such as the CCDs)? We also compared the Meaningful Use and PQRS measures against those practice-level measures currently reported on YHM. While there was not a one-to-one comparison of all of the measures, there Page 11 of 29

12 were several reported by YHM that were either similar or identical to those used in national reporting programs. As a result, five measures included in the Optimal Diabetes Care composite currently reported by YHM were selected to represent the quality component. Table 4 provides an analysis of how the YHM measures aligned with current ecqms in national programs and the measure selection criteria. Measure ecqm used in MU Data available in CCD Blood pressure less than 140/90 Level of bad cholesterol (LDL) less than 100 mg/dl Blood sugar (A1c) Less than 7% 7%-8% 8%-9% Greater than 9% Remain tobacco-free CMS 165: Controlling High Blood Pressure CMS 163: Diabetes: Low Density Lipoprotein (LDL-C) Control (<100 mg/dl) CMS 122: Diabetes: Hemoglobin A1c Poor Control (>9.0%) CMS 138: Preventive Care and Screening: Tobacco Use: Screening and Cessation Intervention Effective Broad patient population ecqm complexity score Holy Grail Yes Yes Yes Simple Yes Yes Yes Yes Simple Yes Yes Yes Yes Simple Yes Yes Yes Yes Simple Yes Take aspirin daily as recommended CMS 164: Ischemic Vascular Disease (IVD): Use of Aspirin or Another Antithrombotic Table 4. YHM measures alignment with current ecqms Yes Yes Yes Simple Yes Page 12 of 29

13 We were unable to identify a one-to-one match of all five measures against those used in national programs. Because we leveraged patient-level data using CCDs and the data import was not limited to a specific set of data elements, this lack of identical ecqm specifications did not limit our ability to collect and analyze this information. Specific steps taken to leverage existing ecqm specifications are outlined in Table 5. Measure Blood pressure less than 140/90 Level of bad cholesterol (LDL) less than 100 mg/dl Blood sugar (A1c) Less than 7% 7%-8% 8%-9% Greater than 9% Remain tobacco-free Take aspirin daily as recommended ecqm used in MU CMS 165: Controlling High Blood Pressure CMS 163: Diabetes: Low Density Lipoprotein (LDL-C) Control (<100 mg/dl) CMS 122: Diabetes: Hemoglobin A1c Poor Control (>9.0%) CMS 138: Preventive Care and Screening: Tobacco Use: Screening and Cessation Intervention CMS 164: Ischemic Vascular Disease (IVD): Use of Aspirin or Another Antithrombotic Identical specification No Modifications made Patient population modified to capture patients with a diagnosis of diabetes rather than hypertension. Yes None required. Yes No No A1C level changed to capture Less than 7% 7%-8% 8%-9% Greater than 9%. Used subset of information collected in current ecqm measure: patients assessed in the measurement year and identified as tobacco user and modified patient population to only those patients with a diagnosis of diabetes aged years. Patient population modified to capture patients with a diagnosis of diabetes AND ischemic vascular disease Table 5. Specific steps taken to leverage existing ecqm specifications No Able to capture needed data elements using CCD data Yes While several measures around cost and resource use are publicly available (e.g., HealthPartners Total Cost of Care, NCQA s Relative Resource Use), most require the use of episode groupers and/or risk adjustment methodologies. Open source tools do not yet exist to enable broad implementation of any one of those measures; thus, limiting our ability to use nationally recognized cost measures. For this reason, we aggregated the total costs for each patient with a diagnosis of diabetes identified in one or more of the quality measures once patient matching was completed. Yes Yes Yes Page 13 of 29

14 Patient Identification and Matching System Selection There is typically no single, reliable patient identifier that can be used to match multiple CCDs and multiple claims belonging to a single patient. Hence, the use of patient matching software, or master patient index (MPI) technology was crucial to our project. We selected IBM s Master Data Management (MDM, formerly Initiate) product. The Health Collaborative already uses this solution for master patient, provider, and organization indexing. The existing MPI contains data for 4.4 million medical record numbers, representing 3 million unique individuals in the greater Cincinnati, Dayton, and Northern Kentucky area. The Health Collaborative uses HL7 Admission, Discharge, and Transfer (ADT) messages from the health systems to populate the MPI. The MDM solution has 34 US Patents specific to the solution and is highly effective at matching patients from disparate sources of data so it was a logical selection for this project. There are alternative MPI solutions available. ARGO and NextGate, for example, have commercially available products, and openempi is an open-source solution. In addition, Healthcare Information and Management Systems Society and the Department of Health and Human Services are collaborating to find new innovative ways for patient matching that may result in new technologies being available in the future. 28 Pilot Solution Overview The technical solution piloted focused on using standards-based data, measures, open source components and commercially available software to present a comprehensive solution that is scalable and allows rapid implementation. Figure 5 provides an overview of the technical flow of data. Figure 5. Overview of technical flow of data Page 14 of 29

15 Listing of Tools & Technologies Utilized Open Source / Freely Available Tools Measure Authoring Tool (MAT) - A web based tool which allows you to author ecqms using the Quality Data Model (QDM). Measures can be expressed and exported in common HQMF format among others. Value Set Authority Center (VSAC) - Machine readable listing of official vocabulary sets used in the 2014 Clinical Quality Measures (CQMs). BONNIE - A tool for testing electronic clinical quality measures (ecqms) particularly those used in the Meaningful Use (MU) program. Cypress - Cypress is the official testing tool for Meaningful Use (MU) Stage 2 Clinical Quality Measures (CQMs) under the 2014 EHR Certification program. pophealth & pophealth is an open source software service that calculates ecqm s. mongodb - A document-oriented database commonly referred to as a NoSQL database. pophealth utilized MongoDB to store documents such as CCD and QRDA and allow them to be accessed via JSON. MONAHRQ - A software tool that generates a health care reporting Website. Commercially Available Products MirthMail - A HIPAA-compliant, robust Direct message system featuring full compliance with DIRECT standards to securely send/receive messages including C-CDA documents. IBM Master Data Management IBM master data management solutions provide a single, trusted view of critical entities including patients, providers and organizations. Microsoft SQL Server - Robust relational database used to store and process claims data and clinical to claims combining Clinical Data Collection Metrics Once the clinical data collection mechanism was implemented, we received almost 400,000 CCD documents from TriHealth at an average of more than 5,000 documents per day. The CCDs collected represented over 350,000 unique patients at over 613 different TriHealth facilities. One reason for selecting CCDs was the large quantity and overall versatility of the data. This also meant that the documents were relatively large, averaging 200KB per file. Overall, we collected more than 79GB of data over a three-month period from a single health system. Data storage will be an important consideration for future work that includes multiple health systems over a larger time window. Claims Data Collection Metrics Below is a summary of the data received from TriHealth s claims processor: File Type Scope Utilization Method Medical Claims Detail 2013 Present Source to obtain cost data Pharmacy Claims 2013 Present Source to obtain cost data Detail Patient Census File 7/2014 Present Source as input to our Master Patient Index for patient identification Table 6. Data received from TriHealth s claims processor Page 15 of 29

16 For the purpose of matching quality and cost, we focused on using the members from the active census file and related claims from the first six months of The table below summaries the analysis and processing of the claims data. Category Value Number of active members in TriHealth s employee group 21,477 Number of members known to The Health Collaborative 17,542 Number of known members with claims data between Jan-June 10, Table 7. Summary of Claims Data Analysis Quality and Cost Comparison Data Metrics Category Value Diabetic Patients found in clinical data 4,609 Diabetic Patients with non-zero, meaningful cost data 138 Table 8. Summary of Claims Data Analysis Quality Measure Implementation Once data flowed into the pophealth database, the next step was to implement the selected measures within the pophealth Quality Measure Engine. The pophealth tool includes a library that imports the HQMF measure and converts it to an executable set of functions. Locating and using HQMF measures requires a set of resources and tools that can be found at the ecqi Resource Center and are freely available. 29 Table 9 shows the process and tooling required to implement an ecqm. Step Tool 1. Locate the ecqm definition Measure Authoring Tool 2. Locate all the value sets required by the measure Value Set Authority Center 3. Export the ecqm measure package, which includes the Measure Authoring Tool emeasure, an HQMF version of the measure 4. Build test patients that fall into the numerator and BONNIE denominator of the measures to validate that the measure is performing properly 5. Export the measures as a measure bundle, which is a BONNIE specific implementation of the measures that includes an executable source code to process the measures 6. Install a version of pophealth, which includes a pophealth database for patient data, data importation tools and the QME 7. Import the measure bundle produced by BONNIE into pophealth pophealth 8. Start the Quality Measure Engine pophealth 9. Export the calculated results pophealth Table 9. Process and tooling to implement an ecqm For purposes of this project we validated that the clinical data received was compliant with the C-CDA standard and coded with the applicable ontologies. We did not validate the data back to the EHR to ensure that our ecqm calculations matched the true clinical record of the patients. This is an important step that would need to be added for a production implementation of the ecqm system tested in this project. Page 16 of 29

17 The QME inside of pophealth creates a set of results that shows each patient and which populations of a measure the patient meets. Table 10 shows an example of the results calculated by pophealth for CMS measure122 hemoglobin A1c poor control - for Location Numerator Denominator % of Patients with HBA1C > 9 Outpatient Clinic % Family Practice Clinic % Primary Care Clinic % ALL CLINICS % Table 10. Example of results calculated by pophealth for CMS 122 for 2014 Patient Matching The first step to combining claims and clinical data was to assign an Enterprise ID (EID) to each patient using The Health Collaborative s existing MPI system. The existing MPI solution had > 3 million lives already in it, which were used for the MPI to match against. The demographics from the CCDs and the claims files were loaded into the MPI software system which utilizes both deterministic and probabilistic matching methods to uniquely identify each patient. We were able to assign an EID to each patient associated with the CCDs. The documents included the patient medical record number as well as enough patient demographics to ensure a match of high confidence. The claims data provided by Humana contained only 4 data elements: Name, date of birth, gender, and subscriber zip code. This is a subset of what we would typically want to use for matching. By necessity, we lowered our MPI match threshold, and identified approximately 79% patients from the claims files. We also received a member file for the claims set which had all desired demographic data elements (name, date of birth, gender, full address, phone number, Social Security Number) to ensure high-quality matches with high confidence. This member file increased our hit rate to 82%. Once both sets of data contained EID s, we were able to join those data sets during the measure calculation process which occurred later in the project. Patient and Provider Attribution and Minimum Sample Sizes For the purposes of our pilot, we did not follow any conventions for minimum sample size, since we were receiving CCDs for all patients in a practice. Our patient attribution mechanism was simplified for this pilot. We simply associated each patient to the attending physician indicated in the CCD. Future analyses may need to spend more time with realistic patient attribution mechanisms, similar to what is used for public health reporting on YHM. Value Calculation The final step of the project was to display the value provided by each practice, which is the combination of the composite score of the clinical quality measures and the cost (claims) data.. One goal of the project was to test our ability to provide practices with meaningful and timely data regarding their populations of patients whereby they could see the results of process changes within their practices as they attempted to improve their level of care. Using actual data from their CCDA and claims data with a 60 day run out that we collected over a 6 month period, we used the list of patients that were included in the denominator of the Diabetes A1C Poor Control (>9.0) measure and calculated the total cost of care for each patient. The cost and the quality results were aggregated to three groups so as to vary the levels of blood sugar control. Table 11 shows a view of the cost and quality measures by practice. This approach allows a practice to correlate their quality of care with the cost of care for their particular patient population. While this was a preliminary test, it can be extended to any number of practices and additional analysis of the data. Page 17 of 29

18 Location Count of Patients % of Patients with Average Cost Per with HBA1C > 9 HBA1C > 9 Patient with HBA1C > 9 Outpatient Clinic 1 2% $1, Family Practice Clinic 2 4% $1, Primary Care Clinic 4 10% $5, ALL CLINICS 7 5% $3, Table 11. Cost and quality measures by practice Value Reporting Recognizing the need to identify a relatively easy-to-use software tool for displaying value, we evaluated three methods of public display. First, we considered extending our current website functionality at YHM. This website was custom-built by a website design company and therefore the custom nature of the solution makes changes very costly. The second option was purchasing and installing a Commercial Off- The-Shelf Business Intelligence (BI) platform (see Gartner s 2015 Magic Quadrant for Business Intelligence Tools at: Most modern BI platforms have the ability to embed results either statically or with interactive data visualization tools directly into existing websites. This solution required purchase of a tool and building expertise internally to master the functions. Instead, we installed and experimented with MONAHRQ (My Own Network, powered by the Agency for Healthcare Research and Quality [AHRQ]). 30 This free website-generating software is designed to enable an organization to create and host a healthcare reporting website that displays evidence-based healthcare performance reports for providers and consumers. We decided to use MONAHRQ because it was open source, more efficient and cost-effective, and could be more easily replicated by others. Figure 6 shows a simulation of the type of public website display that could be used to provide this information directly to consumers using the technology proven through this project in combination with MONAHRQ and the existing YHM technology. We worked with MONAHRQ Version 6.0. AHRQ released a newer version (Version 6.0, Build 2.0) in August that provides even more capabilities to tailor the types of reports by users (e.g., consumer-focused, side-by-side comparisons). Figure 6. Display of value by practice Results & Findings The Health Collaborative successfully tested a new method for electronic clinical data reporting using CCDs and linked this data with administrative claims data. We were then able to take the results, and using an open-source reporting vehicle, display value at the practice level. Our results suggest that this is a viable data analysis technique that can be used for reporting by multiple providers, multiple EHRs, and Page 18 of 29

19 for larger numbers of patients. We identified several findings that were key to our success and are relevant to others: Quality & Cost Data Acquisition & Aggregation Resource requirements are significantly lower than current manual method: For this project, TriHealth performed a onetime configuration change to output all CCD s for patients, and the digital exhaust began to flow to The Health Collaborative. TriHealth did not need to perform any other tasks throughout the project with regards to submitting clinical data. In other production activities to produce quality measures, TriHealth and other providers manually create custom data extracts specific to a clinical measure and send that data out manually at various intervals, which is far more labor intensive than the method tested in this pilot. The Health Collaborative also saw a significant reduction in resource requirements during the pilot when compared to existing manual methods. If we were to use the existing manual methods during this pilot, it would have taken approximately three to five times more resources to achieve the same result. Additionally the pilot process is a set it and forget it approach that allows the system to run in an automated fashion for most phases of measure generation. CCDs provided a broad set of data with which we could drilldown and apply multiple ecqms at one time: The default CCD output provided a broad set of data types such as laboratory tests (including values), medications, and care goals across many types of encounters including hospitalizations, home visits, and office visits. It was clear and confirmed by TriHealth that the effort to automate the delivery of the data in this format was trivial and that the effort to liberate an equivalent data set using QRDA Category I would require significant resources to implement. That said, because we received more than 79GB of data over three months from a single health system, data storage would be an important consideration for future analyses that include multiple health systems that span over a larger time window. This real-time process demonstrated that the error resolution time is shorter than the traditional manual methods and enabled data to be refreshed quickly and frequently. While this update could occur daily or any other desired timeframe, it was thought that the ideal schedule for calculating and updating performance rates may be on a monthly basis. This approach could encourage providers and practices to track and monitor performance toward desired benchmarks proactively and enable us to move toward longitudinal analyses of patient populations over time. We were able to migrate from a system of measures to a measurement system: Because we received a broad data set through the use of the CCDs as opposed to the targeted set of measurespecific data provided by QRDA, we were not limited on which measures we should use to represent quality nor did it require us to determine the measures at the start of the project. We leveraged the ecqi tools to easily modify several ecqms to narrow our analysis down to the diabetes population. This ability to modify and refine ecqms with real-time data allowed us to better represent the quality component of value. If we were limited to the current set of ecqms, many gaps addressing many relevant patient populations and outcomes would have limited our representation of quality. We also were able to calculate 95 measures prior to selecting the Optimal Diabetes Composite. See Appendix A for a list of measures we were able to calucate using the CCDs from TriHealth. The selection of this composite aligns with the measures proposed for a dashboard under a separate CHT grant to The Health Collaborative, where 20 measures are proposed to be reported by 2019, demonstrating a potential solution to help us achieve this goal. We were able to use claims data to represent some but not all of the total costs incurred by patients and the health system: We were able to load claims data, identify unique patients using The Health Collaborative s MPI software, and match claims data to CCD-identified patients. Due to the timing and resources available, we were only able to compute total claims dollar amount per patient and did not attempt to subdivide claims into related disease states. These adjudicated claims represent the cost of care, in this case, to the employer. Only some of the out-of-pocket costs were identified through the claims record so this approach presents an approximation of the true cost of care for these patients. In Page 19 of 29

20 addition, The Health Collaborative may have been able to align the timeframes used for the quality data with the cost data if we had been able to receive more timely cost data. Additional analyses are needed to understand whether shorter processing delays (e.g., 30 days, 60 days) could yield greater access to cost data without compromising data accuracy. The work by NRHI and the APCD Council to formalize and standardize the format of post-adjudicated claims data will aid in future efforts to meaningfully exchange this data in the future. Our approach would enable us to include all patients within a practice rather than a sample of patients: While the ability to completely represent a practice s performance was limited based on the timeframes in which the data was provided, this approach would represent a practice s performance based on the entire patient population rather than a representative sample. By broadening our ability to capture the total population of patients seen by an individual practice, we increase our confidence that the performance reported is reliable and reflects true practice performance. Practices did not need to participate in data collection: Practices were not required to submit data and, thus, eliminating the time and resources traditionally requested of them. In future efforts, this time could then be reallocated to reviewing the results produced from the data to ensure that it is valid and reflects the practice s true performance. We were able to successfully calculate measures using pophealth: The pophealth software was successfully installed and after some shaping of the CCD data and custom programming required to successfully ingest data, it was able to classify patients into the numerators, denominators, and evaluate exclusion criteria to generate the quality measures. HIE-Centric Data Collection: Our focus of the pilot was on TriHealth clinical and claims data. However, our analysis of the claims data indicated that a number of the TriHealth patients receive care from non- TriHealth facilities, such as the local children s hospital..we were therefore unable to include this data as part of our project. This work illustrates the value of an HIE-centric approach where clinical and claims data can be collected and evaluated from multiple sources. A production implementation of this project would include clinical data submission from multiple healthcare and insurance companies as well as nontraditional sources like behavioral health organization, specialty care clinics and independent laboratories. Patient Matching We did not need to change existing practices on patient matching: By using some of our existing practices and methodologies to attribute patients appropriately to the practice we can minimize the concerns of providers and practices that the quality and cost represented in the value display does not reflect true performance. Future analyses are needed to analyze realistic patient attribution mechanisms and sufficient sample sizes to further ensure confidence in the validity of the value calculations, similar to what is used for public health reporting on YHM. Patient matching of quality and cost data: We were fortunate that The Health Collaborative has many years of experience using patient matching MPI software for the purpose of identifying all of the unique patients within its service area. We were able to use The Health Collaborative s experience and expertise to uniquely identify the patients associated with the CCDs and claims we received from TriHealth. While the identification process, particularly for claims data, is not 100%, it was solidly at 80% to 90% and would be expected to improve over time should this process be moved to production status. MPI quality can be improved with sufficient feedback loops: If the process were to be scaled to include additional healthcare systems and payers, it will be important to develop appropriate feedback loops between the MPI and the health systems and payers to ensure that the quality of the MPI data is maintained to ensure high quality matches. This means that the data in the MPI should be occasionally reconciled with the sources feeding it and as source systems alter their data (e.g., patient merges), this information should be made available to the MPI to ensure the patient data is synchronized. Page 20 of 29

21 Value Calculation & Display Display and public reporting of value can be accomplished using an open-source tool MONAHRQ: We were able to aggregate the quality and cost results by practice to represent value using the MONAHRQ tool. Because of the limited time that we had to experiment with MONAHRQ we were unable to embed it within The Health Collaborative s existing YHM website. However, conversations with MONAHRQ s developers and knowledge of the YHM technology stack suggests that this approach is feasible. Alternatively, organizations that do not have a public facing site as mature as YHM may choose to use one or more MONAHRQ-generated sites for direct use by the public. Value can be displayed in a consumer-friendly manner, while remaining meaningful to practices: The authors believe that underlying data used to calculate ecqms and total cost of care can be useful to physicians and their practices particularly if examined using one of several business intelligence software services coupled with the advice of a knowledgeable data analyst. However, this project did not have the time or resources to demonstrate that belief. Also, the authors believe that a higher level summarization of that same ecqm and cost data can be represented in a useful form for consumers using some combination of MONAHRQ and a public facing website such as YHM. The authors were able to experiment with this approach but were not able to demonstrate its use on a live public facing site. However, we believe that the application of more resources and a few months of additional time would result in this use being demonstrated and would allow an assessment of its impact on consumer decisionmaking. Broad Applicability of Our Approach We were able to accomplish the goals of this project using open source tools with the exception of our already well-established patient matching approach: By leveraging publicly available tools, the approach modeled in this project could be successfully implemented by others. Because data from only one integrated system was used, additional work is needed to increase our confidence that the tools are ready for larger scale implementation. While a key goal of this project was to determine whether this process could be replicated across multiple EHR vendors, payers, and systems, The Health Collaborative was only able to identify one integrated health system that was willing to share EHR and claims data. While we were unable to demonstrate this possibility beyond a single health system and payer data, we have confidence that the solution is replicable and scalable to any organization with similar goals. Limitations Several limitations to our work exist that may impact the applicability and uptake of this approach. While these issues are discussed as limitations, the majority could be addressed through additional analyses and replication with different data and/or groups. First, we recognize that the data used was from only one source and for a set timeframe. Challenges such as data storage, provider attribution, and minimum sample sizes, could be encountered if the work was replicated using multiple data sources over longer periods of time. We were also limited in our ability to use nationally recognized measures on cost due to our limited data set and the lack of open source tools around risk adjustment and episode groupers. The dearth of these tools must be addressed to allow broader implementation and use of cost and outcome measures. As discussed above, we were not able to incorporate the patient experience data into the value calculation and we therefore unable to fully achieve the widely accepted definition of value. Validation was accomplished only through manual checks of the data to ensure there was consistency across the calculations but much work remains to determine that the data used to represent the quality and costs are truly representative of the care provided (i.e., validated against the gold standard EHR). It is assumed that the results produced from this project do not reflect true performance due to the limited timeframes for which data was sent, the data was from only one source, and only included data that was captured in coded, discrete fields. Additional sources could be used to supplement the current data set since The Health Collaborative serves as the HIE. Leveraging the existing relationships and multiple data sources through the HIE could move us toward quality and costs measures that truly reflect the care provided within the community. Page 21 of 29

22 Discussion This innovation pilot demonstrated that there is great potential to leverage existing national standards and tools to enable reporting on quality and cost data to represent value. Much of what was tested by The Health Collaborative is scalable and could be adopted by other RHICs and HIEs. As highlighted in Figure 7, HIEs have great potential to leverage multiple data sources for multiple uses. This approach is promising due to its ability to better represent the care provided by an individual physician, practice, hospital, or accountable care organization. As Figure 7 illustrates real time data can be collected, matched and reported within a reasonable time frame so as to allow meaningful information for improving the care provided within a practice, informing consumers, and rewarding performance in shared savings programs. Figure 7. Benefits of using an HIE for measure reporting Page 22 of 29