6

DATA AND RELATED INFRASTRUCTURE NEEDS

While health services research has led to important policy changes and to improvements in health care delivery and health outcomes, and while current efforts continue to improve health care delivery, future advances will depend substantially on improving the national data infrastructure and leveraging routinely collected data. For that reason, one of the panels at the workshop considered the data infrastructure required to accelerate work on issues of quality, value, and equity in health care. The availability of exponentially more data than most envisioned a half century ago has been accompanied by the persistence of frustrating gaps, as well as formidable barriers due to technical, standards, regulatory, economic, and organizational asynchronies.

PROGRESS AND GAPS IN CLAIMS AND CLINICAL DATA

“The past decade has seen tremendous progress in assembling a data infrastructure with which to do interesting and impactful work,” said Niall Brennan, president and CEO of the Health Care Cost Institute (HCCI). More data are available, and more technology is available with which to work with data, than ever before. In the past five years, CMS launched a virtual research data center that significantly lowered the cost of access to claims data and allowed smaller universities and aspiring researchers access at a price that they could afford. It switched to quarterly data refreshes instead of annual refreshes, and public use files were released that many people have found extremely useful. A new policy also enables innovators and entrepreneurs to access CMS data through the Virtual Research Data Center. The Qualified Entity Program, which also provided Medicare data to a new set of actors, required that data be combined with private sector data to improve public reporting around cost and quality.⁶

___________________

⁶ https://www.cms.gov/Research-Statistics-Data-and-Systems/Monitoring-Programs/QEMedicareData/index.html

“There are more data on Medicare beneficiaries swimming around in the health data ecosystem than ever before,” he said.

At the same time, organizations like HCCI have emerged that aggregate and analyze private sector claims data and also make it available to academic researchers. HCCI has an open access data model that enables academic researchers at any university in the United States to use the data. In addition, organizations like FAIR Health and Truven Health Analytics make claims data available to researchers.

The greatest problems in leveraging data to address health care and health policy questions today, Brennan observed, are cultural rather than technological. These challenges include: (1) lack of follow through on commitments to release data; (2) legal challenges; and (3) the growth of proprietary data. Comprehensive national Medicaid data are still unavailable from CMS, despite significant time and monetary resources being devoted to a new data collection system. Medicare Advantage data exist but only one year of data is currently available to researchers.⁷ “Hundreds of thousands of physicians submit hundreds of unique quality measures to CMS, but the data are not widely available,” Brennan said.

Additionally, many data sets are still considered proprietary or are otherwise unavailable. For example, a Supreme Court decision has made it extremely difficult for state All Payer Claims Databases (APCDs) to incorporate all the data that they would like to incorporate because self-funded employers can no longer be legally required to contribute their data (Gobeille v. Liberty Mutual Insurance Company, 2016). Nor are data from Blue Cross Blue Shield plans systematically available to researchers. With regard to clinical data in EHRs, though progress has been made on the interoperability front, particularly as it relates to patient registry data, much of this data also remains siloed and proprietary, Brennan observed. Although claims data can be aggregated at scale and analyzed, he expressed doubts that EHR data will ever be able to achieve “universality” in the same way that claims data has.

In addition to clinical and claims data, nationally representative surveys that produce extremely valuable data continue to be cut in response to budget pressures. “We have to defend those surveys,” he said. “It is easy to forget about them. They are kind of old school, because a lot of them started in the 1980s, and everybody wants to talk about big data and AI [artificial intelligence] and things like that. But if we lose things like MCBS [the Medicare Current Beneficiary Survey] or MEPS [Medical Expenditure Panel Survey] or other surveys like that, we will all be very much the worse for it.”

___________________

⁷ CMS plans to release a final version of 2015 Medicare Advantage encounter data by the end of 2018. (Ravindranath, 2018)

Finally, building on points made by Blumenthal and others, and underscoring the need for changes in incentives that reward accelerating the pace of research conduct and dissemination, Brennan expanded on the point about incentives with a not entirely overstated observation that: “If you asked every health services researcher in the country what they would prefer—to be published in 18 months in the Journal of the American Medical Association (JAMA) and win an award at the [AcademyHealth Annual Research Meeting] using 10-year-old data, or to have some findings on a blog that nobody noticed but that a health care system could use to save either lives or dollars—I think the vast majority would choose JAMA and the AcademyHealth award because that is how their incentive structure operates.”

CHALLENGE OF LINKING DATA FROM DIFFERENT SOURCES

Extending Brennan’s remarks, Karl Bilimoria, a surgical oncologist and a health services and quality improvement researcher at Northwestern University’s Feinberg School of Medicine and Vice President of Quality for the Northwestern Medicine health system, commented on the poor quality of data, which people are using to make important decisions. Patients are using it to decide where to go for health care. Payers are using it to decide which hospitals to direct patients and establish contracts. Hospitals are using it to set their quality targets. The problem is exemplified by the finding that the multiple public rating systems often disagree about the same hospital: one rating system may rate a certain hospital a 5 star while another rates the same hospital a 1 star. This is confusing for the end users.

He noted that currently available data used in quality measurement have some serious limitations and long delays plague the availability of data, the opportunity to use the data for change, and to monitor the results of subsequent process improvement efforts. Moreover, getting to measures that really matter is limited by the quality of the available data. While a lot of administrative data are available and are useful for measuring readmissions and mortality, they are far less useful in measuring other outcomes. Compared with chart review or clinical data, administrative data produce large numbers of false negatives and false positives. “There is miscoding in both directions that limits the validity of those data.” They also are limited in doing risk adjustment, since they do not have the level of granularity needed to describe, for example, the clinical severity of the spectrum of diabetes. Yet they serve as the basis for much health care quality and public measurement systems as well as value-based purchasing systems, Bilimoria noted.

The available data are also expensive. “As somebody who writes a check for CMS data once a year, I am painfully aware,” Bilimoria said. Ways to get good payer data exist and are also expensive, but CMS data are separate. Thus, there is not one place where data on all patients (all payers and ages) is easily available for quality and research uses.

On the other hand, clinical registry data can answer many clinical questions, have much more validity than administrative data, and can be extracted in a standardized fashion for quality measurement and research. But they, too, are expensive, and they typically are limited, Bilimoria noted. Due to their expense and the work required for abstraction, registries generally do not capture all the patients at a hospital, and each specialty is establishing a registry, so large hospitals are often being asked by the clinicians to participate in 50 or more different registries costing millions of dollars per year. This is likely not sustainable. The other data that are missing, Bilimoria said, are patient-reported outcomes. Everybody wants to listen to the patient, he said, “but we are far behind in being able to capture this in a standardized fashion.” Greater effort is needed to move forward on these measures, he concluded, “because they do reflect the most important aspects of care to doctors and to patients.”

Importantly, the future of better quality measurement requires us to much more effectively pull data directly from the EHR. This will alleviate the limitations and expense of manual abstraction. Bilimoria remarked, “While much is said of HL7 standards, natural language processing, artificial intelligence, and standardization of EHR data elements, almost no impact of this has been appreciated by the quality measurement community.” Getting better data will require innovation around how the data are put in and how they are pulled out of the EHR. Quality measurement is stalled until this change can occur.

Building on earlier points made by Jack Westfall, another related issue raised by Andrew Bazemore, a practicing family physician and the director of the Robert Graham Center, was the lack of alignment between current health data infrastructure and the ecology of where patients seek and receive medical care. To demonstrate this point and its immutability over the past 50 years, Bazemore returned to the earlier cited paper by White et al. (1961) (see “Moving Research into Communities” in chapter 5). That paper “helped us to establish, in a fairly elegant way, a sense of the patient care seeking universe in the United States,” said Bazemore, and that universe has not changed as much as some might think over the 50 years since.

To demonstrate this, Bazemore cited work from the Graham Center in 2001, and follow up efforts by Johansen in 2016 revealing how most care-seeking continues to occur in community and primary care settings, with very little occurring in the large academic medical centers where most training, research

and data-gathering occur. (Green et al. 2001; Johansen et al. 2016). Since 1961, many new sources of data have become available, such as the Medical Expenditure Panel Survey (MEPS), the National Health Interview Survey, and data from EHRs, health information exchanges, and registries. Remarkably, said Bazemore, “the ecologies tend to map similarly over time.” These efforts also point out the continued value, even in an age of myriad new data sources, of nationally representative surveys, including the MEPS from which the Ecology studies were derived. These surveys are under siege, said Bazemore, while many of the presumably richer new sources in the era of Big Data suffer proprietary lockdown, lack of sustainability models, or the inability to comment on the entire US population in a representative fashion.

That said, Bazemore noted that many new data sources are becoming available that can help to fill some of these gaps. He was particularly enthusiastic about PCORI’s investment in PCORnet, a large, highly representative, national “network of networks” that collects data routinely gathered in a variety of health care settings, including hospitals, doctors’ offices, and community clinics. One of its sites, the Accelerating Data Value Across a National Community Health Center Network (ADVANCE), a clinical data research network in Oregon, has even added information on the social determinants of health to the records of safety net patients on a large scale. Merging the uniform data system of community health centers with claims data would enhance understanding of the ecology of health care. For example, providers could be funded to support an upfront infrastructure that makes it easier for them to send their data to a primary care registry. Such steps would help make up for the losses of data occurring in nationally representative surveys, which, in the past, have been the main way to understand the primary care environment.

Bazemore concluded that with current advances in the national data infrastructure, it should be possible to say “that this is the county where asthma outcomes are worst, and smoking is at its highest prevalence, and here is the provider in the practice that most needs our help in caring for patients according to NHLBI guidelines for asthma action plans or smoking cessation”—a capability that the Graham Center and Community Care of North Carolina was testing 5 years ago by merging data from the state Behavioral Risk Factor Surveillance System and Medicaid to more effectively target clinicians in need of support. Yet, like many other pilots, the effort was terminated due to budget cut-backs. Additional obstacles stand in the way of such uses of data, including the limited availability of proprietary data, the sustainability of data sources, and dissemination of information and the results derived from data not only to policy makers but also to health care providers in useable ways.

COLLECTING DATA IMPORTANT TO CONSUMERS

Introducing a topic of central importance to the future of health services and systems research, Katie Martin, vice president for health policy and programs at the National Partnership for Women and Families, addressed some of the broader issues associated with data from the perspective of health care consumers. Consumers of health care have the same objectives that most stakeholders do, she said. They want a health system that keeps them healthy, that takes care of them when they are sick, and that does not threaten their financial security. However, the health care system of providers, administrators, and payers does not collect data that directly address these objectives, she pointed out. “If we were to look at the challenges of the health system from a consumer lens, it would lead us to collect a different, or maybe an additional, body of evidence.”

She used alternative payment models as an example. The extent to which such models save money is an important issue, she noted, particularly if cost savings translate to lower premiums and lower cost sharing. However, the question less often asked is whether alternative payment models better meet the needs of consumers. “What do consumers want from their health care? We could use more assessment even on that basic question.” Metrics might include convenient access to care, coordinated care, a trusted relationship with a provider, and care that treats patients with dignity and is consistent with their family’s beliefs and goals. Once such criteria were established, alternative payment models could be measured against them and compared with fee-for-service care on these measures.

Another example involves high-deductible health plans with health savings accounts or health reimbursement arrangements. The assumption is that consumers, by having more financial risk, will express their priorities, preferences, and assessments of quality through their market power. Martin argued that a young professional who is healthy one day and diagnosed with cancer the next, or a parent whose baby was born with a congenital heart defect, does not think in these terms. These consumers of health care have considerations other than the entirely rational ones dictated by economics. Martin also suggested collecting evidence on what such products do for the costs and quality of care and the care experience.

Martin pointed out that, while the federal government is spending a great deal on health care—28 percent of federal health expenditures—US households spend just as much “and we don’t think of them in the same way as we do other stakeholders” (Hartman et al. 2018). With that perspective as a lens, she had three suggestions for health services research. The first is to evaluate payment models and innovation through the additional lens of patient-centered metrics.

Such evaluations would consider such factors as health equity, meaningful patient engagement and partnership, and patient-generated information. Her second suggestion is to work on understanding the correlation between patient experiences, patient partnerships, and cost and quality outcomes. Her third suggestion is to conduct research looking at some of the foundational assumptions in health care and health financing. “What if you were to conduct [the RAND Health Insurance Experiment (Brook et al. 2006)] again in the current health care environment and with current health insurance products? Would the conclusions be the same, or would we learn some new ones about the way the world has evolved over the past 40 years? And what if we were to add racial, ethnic, and income granularity to every survey so that we can understand beyond the averages what is happening to different people across the country?”

IMPROVING DATA ACCESS, PRIVACY, AND INFRASTRUCTURE

A major topic of discussion, moderated by Adaeze Enekwechi, vice president at McDermott+Consulting and former head of health programs at the White House Office of Management and Budget, focused on the policy levers needed to make more data available and useful for accomplishing the broader goal of improving quality, access, and equity in healthcare. Brennan, who spent time as the CMS Chief Data Officer from 2010-2017, noted that “money has a tendency to solve a lot of problems.” At CMS, for example, the infrastructure that provides data to researchers is a largely self-funding mechanism. A line item in the CMS budget could theoretically reduce the cost of the data to zero, though Brennan thought that charging something would inhibit frivolous uses of the data.

Continuing efforts are being made to access the data residing in EHRs, Brennan also pointed out. Upfront infrastructure funding could give providers an incentive to make their data accessible. He also mentioned the need for a “zero-tolerance policy around data blocking by either EHR vendors or individual providers and systems.”

Enekwechi mentioned the possibility of combining such data with data from large surveys such as the Health and Retirement Study, which “are probably some of the best longitudinal datasets we have in the country.” And Bazemore, in addition to his point about the need to create uniform data reporting systems to gather data from rural primary health clinics, spoke to the potential of bringing informaticists and clinicians together to design systems based on patient-centered outcome questions. Such systems could incorporate information about the social determinants of health and provide an opportunity “to use clinical data merged

across multiple levels—inpatient, outpatient, laboratory, and community—to understand how we need to risk adjust and adapt to the social drivers of health care.”

The discussion also focused on privacy issues. Brennan commented that maintaining privacy is critically important. However, he also made the point that the Health Insurance Portability and Accountability Act (HIPAA) is, in fact, a permissive regulation that is often misinterpreted by nonresearchers to block data sharing. Further, he noted that there are methods that allow for statistical deidentification while maintaining the utility of the data. With use of the appropriate security protocols, “you can do a lot and not worry that you are compromising or threatening people’s privacy.” However, other workshop participants countered that fears of how the data might be misused and other privacy concerns are legitimate barriers to sharing and worthy of attention. One proposed approach was a model policy that would provide health care entities with a safe harbor in case of a data breach if they followed all of the HIPAA requirements and other relevant guidance. Resolving these issues is core to the transformative progress of health system research and improvement.

The discussion then broadened to a consideration of public health data and data on the social determinants of health, including the possibility of a “meaningful use revolution in the public health sector to tie those data together.” For example, data on both the social service and health care needs of an individual could help determine whether more money spent on social services could reduce health care spending. Martin pointed out that this kind of work is going on at the federal level, but the social determinants data are receiving less emphasis than the clinical data, and “there is opportunity to encourage the administration to accelerate those data elements and data collections.”

Ultimately, as several participants noted, the most critical step to promoting policy changes to improve the data infrastructure and data access is demonstrating the value in leveraging data to end users, including health care consumers, clinicians, health systems leaders, payers, and policy makers. Without first demonstrating value and creating demand, it will continue to be challenging to address the cultural barriers mentioned by the panelists.