Abstract
Accountable Care Organizations (ACOs) are exemplars of so-called value-based care in the US. In this model, healthcare providers bear the financial risk of their patients’ health outcomes: ACOs are rewarded for meeting specific quality and cost-efficiency benchmarks, or penalized if improvements are not demonstrated. While the aim is to make providers more accountable to payers and patients, this is a sea-change in payment and delivery systems, requiring new infrastructures and practices. To manage risk, ACOs employ data-intensive sourcing and big data analytics to identify individuals within their populations and sort them using novel categories, which are then utilized to tailor interventions. The article uses an STS lens to analyze the assemblage involved in the enactment of population health management through practices of data collection, the creation of new metrics and tools for analysis, and novel ways of sorting individuals within populations. The processes and practices of implementing accountability technologies thus produce particular kinds of knowledge and reshape concepts of accountability and care. In the process, account-giving becomes as much a procedural ritual of verification as an accounting for health outcomes.
This article concerns the way populations are constructed through the processes of dataveillance and within the set of institutional relations designed to produce value and accountability. Value-based care (VBC), defined as health outcomes achieved per dollar spent, is becoming a widely embraced policy strategy to contain healthcare costs while improving patients’ care experience (Porter, 2010; Porter and Teisberg, 2006). As one physician remarked: ‘For healthcare providers, value-based care isn’t just an operational incentive anymore, it’s an imperative for basic survival. [It is] vitally important to redesign health system services for population health’ (Michael Blackman, MD as cited in Garvin, 2016). Implicit in this new conceptual landscape is an actuarial way of thinking, involving both economic and health risk calculations when considering how to achieve outcomes. Risks affect the health of individuals, populations and healthcare organizations.
To ensure that ‘value’ outcomes of improved cost and care are achieved, policy analyst Elliott Fisher contended, an arrangement was needed in which providers share risks, rewards and penalties with payers. This would make providers more accountable for the outcomes of their patients, (Fisher et al., 2007). In the US, this led to a new institutional form, the Accountable Care Organization (ACO). The inherent political rationalities on which ACOs are built manage risk by compiling comprehensive dossiers on individuals within defined populations and redistributing accountability for their health outcomes. The complex notion of accountability in contemporary American medicine is situated within a particular historical, political and sociotechnical moment with assemblages of technologies, concepts and practices that constitute value-based care. It takes a unique form in the US, with its market-based healthcare system and no guaranteed access to care. 1
My central argument is that accountability has become a foil through which systems of managing population health become entangled with particular concepts of value and risk, in ways that are consequential for population health and clinical care. ACOs consist of particular kinds of virtual populations to be managed with targeted interventions. This relies on the ability to make visible individuals who may be harbingers of risk and reassembling them into unique categories. Yet this rests on certain assumptions about what characteristics constitute current and future risk and how best to ameliorate it.
The administrative and algorithmic techniques to classify individuals and sort them into groups will determine who may receive what kind of care – something that is very much at stake in current political efforts to dismantle certain patient protections provided by recent health law. In the process, relations among providers, payers and patients are also reordered: Basing care on monetary incentives (and penalties) for the outcomes of patients puts providers in the position of arbitrating financial and health risks, blurring their role with that of insurers. Boundaries between clinical care and public health are blurred as population health management becomes a matter of data-intensive sourcing about individuals in their everyday lives, and individuals are viewed as risk objects in relation to others within unconventionally defined populations (Jacobson and Dahlen, 2016). At the same time, infrastructures established to document accountability facilitate data-intensive sourcing of personal health information for broader purposes of data collection beyond healthcare.
My analysis contributes to STS by bringing together perspectives from valuation studies and the social study of risk practices to consider the complexities of crossed economic domains and changing legal and financial practices (Birch, 2017; Dussauge et al., 2015; Power, 2007, 2016). Hilgartner (1992) distinguishes objects of risk (people or things potentially experiencing harm) and risk objects (potential causes of harm). From a value-based perspective, patients are both, since their health risks may also be financial risks to the ACO. The ability of ACOs to identify potential sources of risk is thus critical.
I use the analytical frame of assemblage – that is, the arrangements of practices, technologies, and theories that configure action in a sociotechnical space – to analyze interactions shaping the way accountability is manifested in particular institutional forms in the US (Law and Urry, 2004; Ruppert, 2011). This enables a broader view of value-based healthcare concepts and activities not only in relation to each other, but also in relation to activities and structures that pre-figured the current situation, plus phenomena in other social domains, such as increased dataveillance in consumer and finance domains. The assemblage includes the big data analytics, changing health information technology (HIT) infrastructures, novel cost accounting techniques, historical and political policy contexts, and intensified public health focus on social and behavioral influences on health as much as genomics. VBC in the US would likely not have existed in its current form without the interaction among these parts, and within the context of a market-based healthcare system.
Relations in an assemblage are dynamic: Actants may enter or depart, laws may be enacted or repealed, benchmarks may move, and measurement tools may be adapted by local users. I show how interactions in such an unsettled (and potentially unsettling) assemblage enact concepts such as ‘accountable care’ and ‘population health management’ through practices such as data-intensive sourcing (Hoeyer, 2016; Kitchin and Lauriault, 2014). At the same time, population databases created for accountable care materialize particular kinds of subjects within framings of future risk and value. Showing how a particular American version of accountability came to be manifested as algorithmic risk sorting of defined populations, I address a fundamental STS question of how new forms of knowledge and models of social control develop together.
Managing population health entails acquiring much more data about patients, of many more types, collected and analyzed in new ways. Such intensified data sourcing includes getting data from sources beyond that which is usually considered to be ‘health-related’ (Hoeyer, 2016; Hogle, 2016a; Van Dijk, 2014). As I show, information about individuals ‘in the wild’, rather than in experimental or treatment spaces in the clinic, is used to identify, characterize, and intervene in individuals’ health with care coordination, prevention efforts aimed at behaviors, and more.
At the same time, as value-based partnerships, ACOs have to prove they are providing care that has relative worth according to federally-set benchmarks that measure both quality and cost-efficiency. To do this requires metrics that order data in particular ways to demonstrate improved outcomes with which to support their claims to shared savings. Yet ‘data are not simply “collected”, but are the result of multiple sociotechnical arrangements of technological and human actors that configure agency and action’ (Ruppert, 2011: 7). It is important, then, to attend to the practices of collecting, measuring, analyzing, sorting and representing data in interaction with value-based payment and reporting structures. Metrics for measuring outcomes are performative, not only in terms of generating potential interventions with selected members of populations, but also in terms of the institutional forms, work practices, and meanings that emerge; in particular, they perform understandings of what comes to count as ‘population health’, ‘health risk’, and ‘accountability’. The vision may have been to deliver the so-called ‘Triple Aim’ goals – to improve population health and individuals’ care experience while reducing per capita costs (Berwick et al., 2008) – but as I show, the way accountability practices are being enacted has other effects.
Data analysts are going further than collecting genomic or clinical data to create risk classifications, using big data analytics to associate social variables with health, deriving new categories such as ‘superutilizer’, ‘nonadherent’, ‘socially isolated’, or ‘aging-focused household’. Here the STS literature on classifications is helpful (Bowker and Star, 1999). Social classifications take on meanings that arise through interactions of scientific, administrative and popular definitions, and change the way individuals experience themselves (Hacking, 2006). In the ACO case, patients may be unaware of how they have been classified (or that they are being classified), but the sorting may nonetheless be consequential for their care (Pasquale, 2014; Solove, 2004). In a time of national policy precarity, practices of measuring and sorting risk thus become a key focus for study. In VBC, the locus of responsibility becomes more fluid and interactive among the state, various kinds of public and private corporate entities, and the patients themselves. As such, it problematizes simplistic understandings of neoliberalism. While explicitly a market-based model with incentivized competition at its core, responsibility and accountability are more complex in the new models and need to be examined.
This article proceeds in two parts. First, I provide background on the US system, reviewing current and emerging financial and legal infrastructures and including key laws marrying healthcare payment and delivery with information technologies. These prefigurations shape the form that accountability takes in the US in distinct ways. Second, I analyze emerging practices to produce accountability using data-intensive sourcing. As I show, the question of to whom ACOs are accountable and for what purposes is debatable.
Methods
This article overviews policy accountability practices in process. I do not include activities or responses of patients, although this is an area ripe for STS analysis (cf. Lupton and Michael, 2017). Rather, I focus on providers and payers, the focus of value-based practice changes. My data come from document analysis of policies and recommendations from expert governmental and nongovernmental policy advisory bodies (including the Institute of Medicine [IOM], Centers for Medicare & Medicaid Innovation Center, among others), laws (the ACA, the Health Information Technology for Economic and Clinical Health Act, and other relevant laws) to situate the emergence of ACOs politically and historically. Third-party white papers plus promotional literature from analytics companies and VBC consultancies, and reports from ACOs shed light on how concepts of accountability are framed by various actors. I also interviewed representatives of data analytics companies, and practicing professionals in health informatics, public health and hospital administration, primarily at clinical medical informatics conferences and workshops between 2015–2017. Population health management and patient stratification tools (both commercially available and those designed in-house by providers) were demonstrated at these meetings, using actual patient and provider data. Patient identifiers were masked, but patient-specific and provider-specific scores and comparisons were demonstrated. My observations illuminate the way accountability is being interpreted and assumptions are being built in to concepts about financial and health risk, and ultimately, into products that will be used to characterize individuals and stratify populations.
Background: Conditions of possibility for value-based care
Value-based care concepts are being introduced into many existing healthcare systems, but each is historically and politically distinct. There are different stakes for providers, payers, and patients. I therefore begin by reviewing salient features of the US system.
Roots of the cost-quality-outcomes conundrum
Most significantly, the US consistently has among the highest costs of care in the world, yet worse health indicators than many other countries: the US ranks 50th out of 55 countries based on cost per person, longevity and health indices (Du and Lu, 2016; Office of Economic Co-Operation and Development (OECD), 2017). Reform efforts since the second half of the 20th century attempted to remedy this conundrum, but have been thwarted by the fact that without a single-payer system, millions of Americans do not have insurance or are not covered by federal plans, and care is paid on a fee-for-service basis through many providers operating under various kinds of contracts.
Under fee-for-service, payers (typically government or private insurers) pay, based on multiple negotiated rates, for healthcare services when ordered by clinicians. The paradox of this volume-based model is that patients who are more sick and who are using more services bring in more revenue for providers. As a result, there is overuse of some services, especially those with higher revenue margins, while there is under-use by those who cannot afford certain procedures or medications. Providers are left to pay for expended services for uninsured or under-insured patients.
Providers have little incentive to contain costs, since their income comes from services. Payers, on the other hand, are often more focused on containing cost than improving quality. Payers are usually unwilling to increase their financial burden and risk, and quality initiatives may not pay off until far in the future. Private insurers experience significant churn (up to a quarter of their customers change insurers each year), so the return on investment may not be considered worth it. Policymakers saw this essential conflict as a lack of accountability on both sides, and it became the basis of policies to change payment and clinical practices. Yet, previous organizational experiments to cap costs (such as Health Maintenance Organizations [HMOs]) were highly unpopular, and efforts to standardize quality resulted in a proliferation of thousands of quality measures without substantially improving health indices (IOM, 2006, 2016). Pay-for-performance (P4P) models tried to incentivize change by paying physicians a bonus to meet or exceed performance benchmarks, but benchmarks mostly dealt with practice efficiency rather than quality, and had mixed results (Grossbart, 2006; Porter and Teisberg, 2006). In sum, efforts to improve quality or reduce cost have been neither comprehensive nor effective.
Between 2008 and 2016, attempts to devise a federally subsidized system for universal access to care, which might have helped standardized payment and quality requirements, met with intense resistance from insurance and medical product industries for which existing, predictable payment models were deeply entrenched, and from those who wanted to maintain private markets rather than move to a public, single payer. Ultimately the Affordable Care Act (PL 111-148, more commonly known as the ACA) was a middle (albeit potholed) road. Individuals not otherwise covered could gain access to insurance with subsidies (for some). Until the ACA, private insurers could refuse to cover individuals they deemed to be too risky, such as those likely to require more care and incur more costs because they had pre-existing or chronic conditions, had unhealthy lifestyles, or had other characteristics associated with higher risks. Likewise, physicians could refuse to accept certain patients, such as those who had insurance with low provider reimbursement rates (in particular, Medicare). While the ACA guaranteed the possibility of access to health insurance, there is no guarantee of access to care: individuals can still refuse to buy insurance and physicians can still limit their patient panels. Payers can no longer refuse patients due to pre-existing conditions. Plus, of those millions of people added to insurers’ rolls, many previously had no insurance (so likely had not sought care). The result was a new mixture of patients, many of whom were sicker and about whom there was little medical history, hence a different and potentially riskier pool. As I show, payers and providers are adapting their strategies accordingly, using value-based programs to adjust formulas for this new landscape.
Most Americans buy health insurance through their employers (who partially subsidize the payments and either contract with insurance companies to provide coverage or take on the financial risk themselves with their own insurance plan). For older and poorer citizens, the Department of Health and Human Services (DHHS) Center for Medicare and Medicaid Services (CMS) contracts with private insurance companies to provide care. Individuals older than 65 who have worked and paid payroll taxes are covered under the Medicare program, while the poor and some disabled are covered under the Medicaid program (these are adults and children, but all must be US citizens, and being poor alone is insufficient for eligibility). 2 While Medicaid was expanded by the ACA to extend insurance to low-income individuals nationally, as of 2012, states are allowed to opt out (Kaiser Family Foundation, 2018) – as of this writing, eighteen states have opted out. As a result, policies and gaps in care vary among states. Finally, about 9% of Americans are still uninsured (28.8 million) compared to 16% (48.6 million) in 2010 (Zamitti et al., 2017). Medicare patients are sicker, older, and the most costly. Payment for their care constitutes about one-third of hospital revenues under the current fee-for-service system, one-quarter of this being for in-patient hospital care. Medicare patients alone accounted for about 15% of the entire federal budget in 2015, making this program a prime target for cuts during debates over federal budget deficits. Unlike private insurers, who can create risk pools with which to determine differential policy rates and benefits for members (based on variables such as health status, age, or other factors), Medicare must pay for ‘reasonable and necessary’ care for all who qualify. Of course, the interpretation of these terms can vary in practice. The Center for Medicare & Medicaid Services (CMS) sets reimbursement rates for services, drugs and devices, which also influence the rates of private payers. Providers complain that their costs for Medicare patients are often not fully reimbursed (about 12–48% of charges). This has led to a type of gaming the system called ‘upcharging’ or ‘upcoding’ (classifying patients using more lucrative diagnostic billing codes to enhance cost recovery), or outright fraud (claims for services not rendered, altering medical records) (Dafny and Dranove, 2009). It is unsurprising, then, that Medicare is a ripe target for alternative payment plans. Roughly half of ACOs operate in a Medicare Shared Savings Program (CMS- Aug 2017).
As for providers, just under a quarter of hospitals are for-profit, about 58% are private but not-for-profit and the remainder are state or locally-owned. In contrast to other countries, this affects how ‘value’ is tied to revenues and practices. The institutions that perform worse on both quality and cost metrics typically care for greater numbers of vulnerable patients, particularly elderly minority and Medicaid patients. These so-called ‘safety-net’ hospitals consistently look worse on quality and cost measures because of the complexity of cases, high costs and low revenues (Jha et al., 2011).
The upshot of diverse providers and payers for Americans is that healthcare varies widely across regions as they receive different levels of quality, cost, and comprehensiveness of care. Employers, payers (especially CMS) and providers (especially for-profit) are intent on lowering their financial risk, particularly with the new mix of patients, and anything that lowers costs provides competitive advantage. Risk-sharing initiatives become attractive in this scenario.
It is within these historical and political-economic environments that American healthcare is transitioning to value-based care, materialized in innovations such as the ACO. The VBC concept evolved long before passage of the ACA, but serves the so-called ‘Triple Aim’ goals that were a key feature of the law. Framing best care as best value is a very different way of thinking about managing a population’s health than universal care or healthcare as a human right, but fits with the market-economy basis of American healthcare. Focusing on value also decenters debates about rights to care that continue to plague American politics. At the time of writing, efforts to dismantle the law are ongoing; however, infrastructures already installed to execute value-based care will affect clinical care and public health for years to come.
The road to value-based care is a digital trail
Beyond structures for providing and paying for care, there are relevant histories of laws, policies, technologies and assumptions about quality paired with legislative requirements to embed information technologies bringing cost and quality together, rebranded as ‘value’ (for a more thorough discussion, see Hogle, 2016b).
In the 1990s, quality care became a priority in the US not only due to poor national indices, but also because of heightened concerns about medical error and institutional liability. Yet measuring ‘quality’ is tricky because the term itself is ambiguous. The most frequently cited definition of quality is: ‘the degree to which health services for individuals and populations increase the likelihood of desired health outcomes and are consistent with current professional knowledge’ (IOM, 1990). This vague definition allowed considerable leeway to institute a plethora of policies, incorporating different values regarding risk under different political and social environments. 3
By 2007, recommendations from the IOM (2007) to standardize quality measures were accompanied by calls for the expanded use of electronic health records (eHR) and for more data, with which to provide evidence of outcomes, to be collected in each clinical encounter. 4 The US had been slower than other countries to take up eHR. However, promoters of the rapidly-evolving healthcare information technology (HIT) field argued that electronic data capture was crucial to facilitate the transfer of patient data across clinics, compare treatments or physician practices for the purpose of evidence-based medicine, and to conduct operations or outcomes research (IOM, 2011).
Subsequently, policymakers in President Bush’s administration flagged electronic medical records as a critical national infrastructure need. By 2009, the Health Information Technology for Economic and Clinical Health (HITECH) Act was passed, requiring providers to adopt HIT to achieve ‘meaningful use’ of health information. Providers must prove that they are using certified eHR to communicate electronically with patients and other providers in ways that can be quantitatively measured. The stated aim was to enhance data exchange for purposes of care coordination, population health management and consumer engagement. In 2018, ‘meaningful use’ was rebranded as ‘promoting interoperability’, to underscore data sharing. New scoring and measurement policies were proposed as a condition for participation in Medicare, with penalties for not sharing data. This directly links to the 21st Century Cures Act of 2016 (dubbed the ‘Cures Act’), which penalizes data blocking and mandates open application program interfaces (APIs, the means through which software applications can interact). 5 Together, provisions in the HITECH and Cures Acts laid the infrastructural foundation for facilitating exchange of data about patients among providers, payers, but also third parties, including data aggregators and analytics companies.
It was also during President Bush’s tenure (2001–09) that healthcare policymakers began arguing that care should be ‘value-based’: services should have worth relative to outcomes, not based on cost-cutting alone or simply adding more quality measures. This occurred in an era of political efforts to ‘downsize government’ and contain federal costs for welfare and social programs. The trifecta of information technology and data analytics, a renewed emphasis on quantifiable quality measures, and cost containment pressures in this political climate came together at the starting block in the race to healthcare reform in 2010.
To ensure that value-based principles would be put into practice, the ACA built in measures to hold providers accountable for improving patient health outcomes as a condition of receiving payment. Additionally, the Medicare Access and Childrens’ Health Insurance Program Reauthorization Act of 2015 (MACRA) specified alternative payment models for Medicare. 6 The models offered providers financial incentives if they could demonstrate improvements in their patients’ outcomes and in cost-efficiency. 7 MACRA further stipulated that to get incentives providers must also abide by the meaningful use provisions of the HITECH Act, which, as described above, expanded electronic data use and sharing. The CMS goal was to convert 50% of providers to some form of alternate payment model by the end of 2018 – an aggressive timeline.
The reformulation of care-as-value was fully embraced by the time the report ‘Best Care at Lower Cost’ appeared (IOM, 2013). Significantly, it advocated the creation of a national, searchable database of information about individuals, real-time collection of data during each clinical encounter, and greater use of genomics and social indicators. To deal with such complex, large datasets, advocates heralded the nascent field of big data analytics in many policy reports (Manyika et al., 2011). Notably, the report was authored by the Roundtable for Evidence-Based Medicine (charged with best practices in knowledge production), which then renamed itself the Roundtable on Value and Science-driven Healthcare. Members include health policy experts plus representatives from CMS, the Office of National Coordinator for Health IT, the pharmaceutical industry, and the head of the largest US electronic medical record firm.
To make such sweeping transformations in a system with deeply entrenched payment structures and revenue streams would take drastic measures to change the way care was paid for. As Berwick et al. (2003) put it, ‘systematic changes will not come forth quickly enough unless strong financial incentives are offered to get the attention of managers and governing boards’ (p. 8). A ‘carrot and stick’ approach was built into alternative payment systems to incentivize ‘high-value’ care, but, importantly, new models shift responsibility for costs and risks. Whereas payers (public and private) previously shouldered most of the financial risk of patients becoming or staying ill, shifting the responsibility to providers for both patients’ health risks and their own financial risk would make providers bear consequences for their actions, in the value-based way of thinking. In contrast to earlier efforts to change payment and service delivery models, risk-sharing constitutes a substantial rethinking of accounting systems, organizational relationships, and expertise, entailing novel institutional forms and practices for analyzing and managing care and its costs (Shortell, 2013). 8
Accountable Care Organizations
The Accountable Care Organization (ACO) is one such form. An ACO is most simply defined as a partnership of healthcare providers (physicians, hospitals, clinics and other care providers) held jointly accountable to payers (federal government or private insurers) for the quality, cost and health outcomes of a defined population (McClellan et al., 2010). ACOs essentially expand the pay-for-performance concept to defined populations (Grossbart, 2006; Kindig, 2006; Stoto, 2014). They are predominantly organized around Medicare patients (due to their overall higher costs and risks), but there are also Medicaid ACOs, and private (commercial) ACOs (typically consisting of large regional healthcare organizations and large private insurers). ACOs can be physician-based or hospital-based, but this poses a conundrum for the latter in the US. Hospitals measure overall success by getting physician business and having patients in hospitals, not keeping them out.
Managing ‘defined populations’ is key. ‘Population’ today usually refers to a subgroup based on specific characteristics for particular statistical purposes (Armstrong, 2017; Holmberg et al., 2013). For American ACOs, those characteristics are not bounded by citizenship, geography, similarity of attributes or diseases; rather, population means individuals who are in an ACO’s defined service area who are attributed to that population by the commercial or federal payer. Attribution is done using several methods, which can affect not only the way populations are constituted (e.g. for whom are providers accountable), but also how providers may manipulate data. 9 About 30 million Americans are now in ACOs – although many of them would be unaware, since payers virtually attribute them to an ACO based on what services they have used and where they have used them (from claims data). How populations are defined has everything to do with understandings of risk distribution within a group. Furthermore, compensation for services in accountability care is based on the health outcomes of individuals in relation to the defined populations for which the organization is responsible, not the national population (Porter and Teisberg, 2006).
ACO population management includes coordinating care among the various clinical entities patients encounter (using social workers or nurses as ‘practice extenders’), and monitoring patients and promoting health behaviors through continual electronic communications, including data from devices or apps per HITECH mandates. This is consistent with Fisher’s vision of organizations engaging patients in activities that should keep them healthier so they would not need to use clinical care in the first place. ACOs also negotiate with pharmaceutical, device and service suppliers to set prices contingent on clinical outcomes of treatments. 10
However, ACO population management also means assessing current and future health outcomes of patients within an ACO’s attributed population, identifying those who are high utilizers of costly care, and making interventions based on individual profiles (Casalino et al., 2015; Kindig, 2015). This requires more than routine epidemiological information, or records of individuals’ past clinical encounters from medical records or insurance claims data. Rather, more data collected from more sources would be needed, along with sophisticated analytics. The term population health management has thus also come to imply the incorporation of data tools used to characterize individuals as risk objects in relation to others in the defined population. Analytics consultants, for example, use the following definitions: any activity that improves the health or care of a single patient by viewing that patient as one small piece of a broad group of his or her peers. This may mean using a risk score calculator based on big data analytics to pinpoint one patient’s likelihood of developing heart disease, or helping another patient manage her diabetes by drawing on engagement and adherence lessons learned from previous cases. (HealthIT Analytics, 2015) [population health management is] the information technology (IT) component of the clinical and administrative aspects of care. This … requires IT resources and tools to collect data on individual health status; stratify and target populations based on their risk and need for care; and engage people in their health using patient health records or online portals. (DeVore and Champion, 2011)
Population health management is directly tied to alternative payment models that require a provider to statistically prove an improvement in a given population’s health. Several models exist. 11 Some continue to use fee-for-service and have modest incentives for simply reporting quality and efficiency data (such as hospital readmission rates, total cost of care, patient satisfaction ratings). More drastic models allow ACOs to take on greater financial risk in exchange for potentially greater shared savings and bonuses, but may have additional penalties if they fail to meet targets. Capitation models, similar to HMOs, give a fixed rate to providers, who bear the consequences of costs being below or above the negotiated rate. This is a high-stakes game for providers: essentially, hospitals and physician groups who have had a reasonably predictable revenue stream are betting that any population management measures they take may save them money in the long run, but they are taking on greater financial uncertainty.
CMS payments are budget-neutral; there is a single pool of funds for the incentives, so the penalties on poor performers are used to reward the best performers. An additional five-star rating system from CMS signifies which providers are the best performers; this also affects reimbursement rates. Additionally, payers may steer patients to providers who perform better, by limiting which physicians and hospitals the insured can use or tiering the amount a patient must self-pay. In fact, several presentations I attended demonstrated products enabling comparisons of patient outcomes metrics by physician. These features, along with the way ACOs decide to allocate any savings among participating providers, set up a competitive environment among providers. Porter and Teisberg (2006) argue that this would produce the best results, but it is certain to change relationships among physicians within and across ACOs and may affect the aggressiveness with which they may pursue value-based objectives. In the uniquely American assemblage, as described in the background section, value-based care doubles as a strategic marketing tool.
Value-based reimbursements are tied to specific metrics set primarily by the CMS and are based on recommendations from various sources, including Institute of Medicine reports (noted above) and the federal Agency for Healthcare Research and Quality, among others. 12 They consist of a complex set of quality and cost efficiency measures that are weighted to get composite scores. The scores determine how much savings and incentive rewards an ACO will get, or how much they will be penalized, depending on the model.
There are currently 23 required quality measures in 4 domains: patient and caregiver experience, care coordination and patient safety, preventive health, and at-risk populations (CMS.gov, 2019). 13 In 2018, there were 31, about half for clinical outcomes; that is, assessment of intervention effectiveness (such as whether a diabetic patient’s A1c level is maintained after being given medications), obtained from payment claims. For 2019, 10 of the 23 required measures relate to patient and caregiver experiences, which are derived from satisfaction surveys. For example, ACO-45 assesses staff courtesy and helpfulness as rated by patients. The four domains are now more equally weighted, suggesting that qualitative evaluations are now co-equal to clinical measures in calculating scores and hence, payments in shared services programs. This shift will likely encourage changes in administrative procedures and patient experiences, but whether actual health outcomes are improved is yet to be seen. Measures with particularly high cost impact, such as adherence to treatment or medication protocols and readmission to the hospital within 30 days after discharge, are targeted in additional policy documents. 14 Other measures assess structural issues (Were eHR and data infrastructures adopted?) and process issues (Were patients immunized? Screened for alcohol abuse?). If scores on these measures qualify, then the ACO is rewarded accordingly. However, many of the measures do not speak to actual health outcomes. Rather, they relate to whether procedures were followed and the percent of times an action was taken. Notably, while some measures are mandatory, providers may choose from a menu of others to submit for review. Furthermore, commercial ACOs can set their own metrics and thresholds for achieving them. ACOs can also select the time period of the minimum 90-day reporting period, which may affect the metrics. These features can lead to ‘gaming the system’ by controlling what gets measured and when.
Such metrics are inscriptions of risk (Hilgartner, 1992; Power, 2016). That is, they reflect what kinds of things are viewed as having potential for harm (to patients or ACOs) and how they might be mitigated. The choice of which things rise to the level of being important to measure (or not) thus participate in shaping knowledge about what quality care is. Metrics are also performative organizational artifacts (Power, 2016). That is, they channel the way organizations enact routines and work flows in a way that suits CMS reporting requirements: collecting particular kinds of data and entering it into eHR in a particular way, configuring databases to compare individuals against each other within a population or to compare providers against each other to get performance scores, or triggering reports in certain formats. Paradoxically, setting measures as a target to be met through changes in work flows and protocols makes the measure less effective, because altered interactions among actors and organizations changes the context. In effect, the system as a whole is being assessed, rather than individual performance (Strathern, 1997). 15
The checklist of measures provides records to be compared over time and across institutional entities: An audit trail that documents accountability according to value-based rationalities. Performance indicators, guidelines and outcomes measures can thus be thought of as ‘accountability devices’ (Jerak-Zuiderent and Bal, 2011; Wallenberg et al., 2016). They provide accountability to the governance system over ACOs, but the extent to which they provide accountability to patients or payers is unclear (Fisher and Shortell, 2010).
However, demonstrating value requires different strategies using different kinds of data than simply documenting whether a procedure is followed or follows quality guidelines. A value calculus only works if the worth of outcomes can be measured. Definitions – and metrics for – outcomes are thus crucial. Yet decisions about what should be measured and how it should be measured embed values and assumptions about what matters for the governance of health. Measurements as accountability devices perform several roles: They provide auditable quantifications for reporting requirements, but they also visible manifestations of both which health conditions and behaviors are problems in need of intervention and what data forms come to count as relevant evidence. If accountable care is as much about shared risk as improving health, then sources of modifiable risk have to be made visible before they can be made amenable to intervention. In the emerging biopolitics of population health management, this involves algorithmic tools and expertise designed to create new types of risk scores and categorizations. In the next section, I show how intertwined concepts of risk and accountability are materialized with specific informatic practices. The actants in this part of the assemblage bridge healthcare with other commercial consumer domains.
Intensified data sourcing for population health management
Large volumes of data are already being collected for analysis of populations, most easily from insurance claims and medical records (Halamka, 2014; IOM, 2013; Thompson et al., 2016), but these have limitations. Diagnostic codes from claims are retrospective, socio-legally constructed categories for billing purposes, not accountability-work. Medical records documents what gets ordered and recorded in the clinic (diagnostic tests, pain scales, prescriptions, etc), but much of this is unstructured data (text, image, rating scales). Digital health monitoring with phone apps or dedicated devices is being used to add real-time data on health behaviors and status (Ruckenstein and Schüll, 2017), but these are snapshots of specific activities.
If ACOs are accountable for risk and outcomes (that also affect organizational risk and outcomes), then they need information about what happens beyond clinic walls that affects their defined population. In particular, social determinants (food insecurity, social isolation, financial stress) are increasingly linked to health outcomes (Braverman and Gottlieb, 2014; IOM, 2015; Kindig, 2006). Social determinants, however, are variably interpreted at the local level. A clinician said that her clinic uses a company to screen patients for them as a ‘top priority for their agenda’, but explained that this includes whether they pay their bills regularly, plus motivational factors: ‘How interested are they in changing? We can get this from social media’(ACO conference attendee, 2017). One ACO manager in a rural area with a large proportion of ethnic minorities explained that patients in his ACO ‘are sick because of stupid lifestyles. It’s their food and culture’ (attendee at the same conference, 2017). Structural issues (systematic discrimination, poverty, exposures to pollutants, working conditions) also clearly affect health, but are largely out of reach for ACO interventions as they are currently designed for targeted outcomes. Information is being gathered to act as a proxy for some of these social and structural effects, and aggregated at both individual and defined population levels to create profiles of health status and risk (Gottlieb et al., 2016).
At one extreme, Barrett et al. (2013) propose matching metabolic profiles to socio-demographic profiles using data produced at all scales, from personal devices to environmental sensing: ‘The social and economic environment can be quantified using spatially explicit socioeconomic data, such as from the US Census, American Community Survey, or publicly available crime data. And social connectedness can be assessed through online social networks’ (p 171). Although similar grand schemes have been proposed by precision medicine initiatives, the impracticality of collecting and analyzing such vast troves of data belies the optimism and naiveté of such big data proponents.
Nevertheless, for value-based reporting, providers and payers are collecting data far beyond what has typically been thought of as ‘medical’ data to serve as proxies for indicators of health status and potential risk. Associating broader, nonclinical information with health outcomes entails intensified collection of social, behavioral and lifestyle characteristics (Hoeyer, 2016; Hogle, 2016a). Sources include publicly available data (education records, property ownership, voter registration, criminal records, by geolocation for their defined service area). Public sector entities also profit from data sales: 33 states in the US currently sell hospital discharge data (Sweeney, 2015). Individuals’ digital traces (social media, consumer fitness trackers or apps) are also available from data aggregators who sell access to individuals’ credit card purchase transactions, loyalty card records, movie rentals, memberships in smoking cessation or wellness programs, and more (Singer, 2014; Terhune, 2008). 16 Searches can also glean information about social networks (people and activities with which an individual connects, or absence thereof).
As I will show, such information – structured and collected for entirely different purposes – is being repurposed in efforts to draw conclusions about patients and their health. Intensified surveillance and big data analytics (natural language processing machine learning, and predictive analytics) are used to handle the large volumes of complex data, and make it possible for information about individuals to be disaggregated and compared across population-based datasets, to compare individuals against themselves, or to compare providers against each other and to identify misalignments with the ACA quality and outcomes measures (Bates et al., 2014; Ryan et al., 2016). Once the infrastructure and arrangements are in place to collect and store data, consulting or insurance companies or ACO data analysts can then do endless variable querying, well beyond reporting requirements. It is important to add that entities such as data aggregators and business intelligence consultants collect patient data to sell for secondary purposes. IQVIA (a marketing research firm) claims to have more than 120,000 global sources to acquire patient data. Large pharmaceutical firms, for example, might pay millions of dollars per year to acquire (de-identified) data to mine for drug reactions or effectiveness. Firms supplying eHR platforms can also write into their contracts that they can use provider data for secondary purposes (Tanner, 2017).
The combination of mandated data-reporting requirements and urgency to make radical system changes have created lucrative markets for health IT products and consultants. According to several presentations I attended, this burgeoning industry is worth about $6 billion in the US, with about $1.5 billion of this in predictive analytics. Significantly, for cost accounting, these services will be embedded in cost of care, and so are not visible as non-direct costs, offsetting some savings that might be earned. Few providers (especially smaller group practices or hospitals) have the expertise or infrastructures to do the kind of analytics necessary for ACO reporting requirements. This creates entrepreneurial opportunities, and both small start-ups and large consumer consulting companies are entering the field. As a line I frequently heard in conferences puts it: ‘Data fishing leads to good fish stores.’ These new ‘fish stores’ do not sell the data sets they collect; rather, their analyses are based on proprietary algorithms. The content and designs of algorithms are opaque to the purchaser-client and to the individuals whose information is extracted.
Some firms offer branded population health management services with copyrighted algorithms. Many target advertising to ACOs, offering to do both economic risk analyses of an ACO’s population and to stratify populations for interventions (HealthCatalyst, Optum, Premier, Philips Wellcentive, Forecast Health). Insurance companies own some of these firms, so they already have access to claims and medical records data. Other firms aggregate patient data from provider clients with publicly available or proprietary databases.
Notably, consumer industry companies (which already have millions of data points on consumers) have entered the space. For example, Experian and FICO are credit rating firms that have tens of thousands of data on credit card transactions, home or car loans, income, investments, zip code (a poor but oft-used proxy for wealth) and more. Lexis-Nexis and Google are search engine firms that track transactions (search terms, online purchases, memberships, dating services) and aggregate them with publicly available data (voting records, property ownership, welfare or food aid enrolment). Acxiom claims to have an estimated 1,600 pieces of data on 98% of individuals in the US. Using big data analytics, they claim to be able to peg individuals as ‘highly stressed’ (based on credit rating scores, crime in their neighborhoods or other associated variables), ‘motivated’, ‘diabetes-aware’, ‘senior-oriented’ and other categories (Citron and Pasquale, 2014; Hogle, 2016a; Pasquale, 2014; Singer, 2014).
Making accountability devices
Such categories constructed from aggregated digital traces are offered as ways to estimate current and future health and cost risks. Profiles produced from associative analytics are being used to characterize individuals’ health status; for example, to see if they are likely to develop metabolic conditions or heart attacks (Steinberg et al., 2014). Researchers claim to be able to predict risk of cardiovascular disease by associating health behaviors with credit scores and factors based on questionable categories of cognitive ability, self-control and education (Israel et al., 2014). Furthermore, predictive analytics are being employed to project trajectories: for instance, not only if, but when individuals are likely to become ill (Weiss et al., 2012). While these claims are yet to be borne out with clinical evidence, they nonetheless produce hyperbole that stimulates hope that data can be acquired with which to respond to outcome reporting requirements.
Behavioral, financial and lifestyle information derived from memberships, product purchases, web surfing and more are being aggregated to create new risk categories (Citron and Pasquale, 2014; Hogle, 2016a). Such data is also being used as a proxy for other kinds of conditions, such as indicators that people are denying their diagnosis, not seeking care or seeking too much care, or not complying with regimens (Halamka, 2014; IOM, 2013; Murdoch and Detsky, 2013).
Increasingly, predictive analytics are used to stratify patients into groups according to certain understandings of risk. A 2015 survey by analytics firm Jvion reported that 92% of providers using predictive analytics use the tools to predict probability of hospital readmissions for specific patients or to project patient deterioration. Readmission rates are one of the major quality measures for CMS, so are a focus of attention. 17 For example, Kaggle competitions paid a $3 million prize for the best algorithm to predict who would be in the hospital in the following year – using actual (de-identified) patient data.
Many new firms have business models based on predictive analytics for population health management. Jvion promotes its predictive models thus: ‘The objective is simple – stop the waste of resources and lives by predicting and stopping losses before they ever happen’ (Jvion, 2015). Forecast Health illustrated applications on their website (no longer available after acquisition by Lumeris), using photos of individuals with associated captions such as “What does her home tell us about her risk of post-op complications? What does his last vacation tell us about his hospital length of stay? What does his marital status tell us about his risk of exceeding the bundled price target? More than you think.” The company claimed to use 4,000 person-specific data points, including individuals’ car ownership or public transportation use, retail purchasing habits (clothes, toiletries), ability to pay (student loan or other debt), lifestyle data (alcohol consumption, exercise) to predict health and cost outcomes. Companies add such data to medical records to predict first hospital admission, readmissions, and post-operative complications (among other risk factors). What is notable is that while much of analytics utilizes de-identified data and might be used for small cohorts of similarly-grouped patients, for many kinds of interventions, specific identity rather than de-identified data is necessary, raising questions about privacy.
A high priority for quality measures is whether patients are likely to become non-adherent to medications or treatments. Attacking patient adherence is also low-hanging fruit for cost-savings to the medical industry and providers because it becomes an easy way to decrease system costs without having to decrease or renegotiate the cost of drugs or other products and services. Adherence was by far the biggest topic of a number of conferences I attended on value-based medicine and health IT. It is estimated that non-adherent patients cost the system about $100–300 billion annually, due to re-admissions and continued symptoms with subsequent use of services (Sulzicki et al., 2012; Volpp et al., 2008). Companies such as IMS and Intelliscript have long tracked prescription orders, obtained from pharmacies, physicians, and pharmacy benefits managers for marketing research. These databases contain information about not only who has filled what kind of prescriptions (and who has failed to fill or continue their prescription), but physicians’ prescribing trends and comparative costs. ACOs (and drug companies that may participate in ACO through purchasing contracts) are very interested in this data, especially for more expensive drugs. The quality measures to ensure electronic communications and patient engagement also get put to use here, encouraging the creation of ways to monitor potentially non-adherent patients and then to intervene. Such efforts are also being monetized: For example, the University of Pennsylvania hospital uses tracking devices both to remind patients of their need to take their blood thinning medications and to track when it was not taken. Those who did were entered into a lottery which gave them a chance to win up to $100, and those who did not were told how much they would have won if they had complied (O’Kane et al., 2012).
According to the CMS, 5% of Medicare and Medicaid patients account for 50% of the costs of health care (CMS.gov, 2018). Much of this is due to complex cases with co-morbidities, but a considerable portion is due to so-called ‘super-utilizers’ (those who disproportionately use health services). Public health experts suggest that much of this is due to behavioral and lifestyle factors in which an intervention could be made – and if so, this is a prime target to demonstrate ‘improved outcome’ metrics (Braverman and Gottlieb, 2014).
However, while targeting high utilizers seems self-evident, analysts now advocate more fine-grained targeting for value accounting. Many super-utilizers, such as some older or chronically ill patients, are unlikely ever to get better, but others may change their risk status over time. For the purpose of demonstrating quality outcome improvement, it is strategically advantageous to make interventions with patients who are more likely to improve, to demonstrate better outcomes in their metrics. For this reason, analysts are using predictive analytics to examine risk dynamics; that is, conditions in which individuals have rising or declining risk. ACOs can then employ interventions that are ‘worth it’. For example, Forecast Health creates ‘impactable risk’ scores to indicate patients for whom an intervention is likely to demonstrate an improved outcome. Strategic interventions on those who will likely demonstrate improved outcomes will produce more reportable value than interventions on those who will not.
Accountability devices and social sorting
Such scores and measures are opaque, arbitrary and discriminatory (Citron and Pasquale, 2014; Pasquale, 2014). Significantly, the examples I have given may provide precise information (e.g. more data points with which to profile an individual), but may not be accurate. Information sources used are partial at best, and can be misleading or spurious. Contexts in which individuals’ transactions are digitally traced can shape behaviors in ways that algorithms may not discern. As individuals upload new information, buy things, do online searches, or fail to pay bills on time, they enact social categories that may or may not be reflected in the algorithm (Cheney-Lippold, 2011). In the example of adherence, decisions not to adhere might include ability to pay, access to treatment centers and transportation, other socio-cultural issues, or side effects. These are highly relevant contexts that are missing from such analyses.
Categories based on associative data are problematic (Hoffman and Podgurski, 2013). One firm claims that mail-order and online shoppers were more likely to use emergency services. Another claims that buying a smaller home flags financial insecurity – and hence, likelihood of becoming non-adherent (Hogle, 2016a). Were the algorithm designers presuming sedentary lifestyles of online shoppers? Might a downsizer simply be an empty-nester? Could there be other spurious relationships among such data? Categories such as creditworthiness or proneness to adherence cannot directly be measured; rather, proxy information imputes characteristics with which to categorize and stratify populations. Nevertheless, the categories are artifacts of the way a problem has been defined and the way healthcare constructs systems of accountability, risk and value. Regardless, it is unlikely that most people know which of their behaviors or transactions affect their scores. While some behaviors can be gamed by altering such things as gym membership, reported dietary consumption, or time of day of glucose monitoring, other factors used to design scoring algorithms are obscure. When it comes to credit and property or life insurance scoring, it is virtually impossible for individuals to make changes to their scores. There is also disparate impact on low-income people who already suffer discrimination (Barocas and Selbst, 2016; Madden et al., 2017).
STS literature demonstrates how social classifications have taken on meanings that arise through interactions of scientific, administrative and popular definitions, and change the way individuals experience themselves (Bowker and Star, 1999). In this case, classifications in the service of accountability create a new form of automated social sorting (Lyon, 2003). The stigma and potentially unjustified classifications can have profound ramifications (Pasquale, 2014; Solove, 2004). Problematic categories as produced by associative analytics may be reified as proxies to explain social and biological phenomena and become part of the permanent record: a ‘health deny-er’ or a future ‘non-adherent’ person will likely carry that label for payers and providers for some time.
Outcomes of outcome-based systems and implications for accountability
The report card on ACOs is mixed (Grossbart, 2006; McWilliams et al., 2016). Some reports tout increases in numbers of participants covered and cost savings for some Medicare models (Muhlestein and McClellan, 2016). Others lost money, and most were unable to increase both quality scores and cost containment (Kocot and White, 2016; Muhlestein and Hall, 2014). Unsurprisingly, providers complain that quality measures are not based on solid methodology and are too difficult to achieve. In fact, some ACOs with significant cost savings nonetheless failed to meet CMS targets, so dropped out of the program (Blackstone and Fuhr, 2016). In a survey of 411 clinic and hospital managers, only 21% had achieved objectives (Numerof & Associates, 2018). The main problem cited was difficulty with outcome measures: While 78 percent could track standardized clinical outcomes (A1C or blood pressure), only 43 percent were able to track required ‘quality’ outcomes.
More than three-quarters of surveyed physicians felt the new models would not improve care and half felt that using metrics to assess provider performance has a negative impact on care. Ninety percent felt they could not spend the extra time or investment to implement new IT systems (Humana News, 2015; Kaiser Family Foundation, 2015). Many providers hesitate to enter risk-based agreements when their revenues depend on what patients do outside of the purview of the clinic.
Organizationally, the upfront expense and disruption of such major innovations can be difficult to justify, especially for smaller organizations, and benefits may not appear for years. New personnel are required to coordinate care across organizations and intervene directly with high-risk patients, and new forms of expertise and information technology infrastructures are necessary for managing and sharing sensitive data on the scale described. Data-sharing across facilities is expensive to implement and maintain, even with cloud-based data warehouses, and in the competitive American healthcare market, providers and vendors balk at sharing, even with the new CMS emphasis on interoperability. Administrative burdens to meet recent federal requirements designed to establish value-based systems are being blamed for declines in smaller independent practices and consolidations with hospital systems.
Public health advocates and social scientists have long argued that impactful issues such as violence, transportation and food insecurity should be included in health status assessments, but clinical institutions have little expertise and capacity for interventions (Casalino et al., 2015; Tannenbaum, 2017). Some health issues are beyond the control of providers and may be beyond the control of patients as well. Not everyone has the option of moving away from neighborhoods in toxic or unsafe environments to avoid being triggered by asthma or being stressed. Some patients may not have enough money near the end of a pay period to refill a prescription or buy healthier food, even if prompted to change behaviors to make outcomes measures look better. Dealing with problems of food insecurity issues and affordable transportation requires a government willing to make major investments in infrastructures and a long-term commitment beyond clinical care. It is unlikely that the political will to sustain such services will exist in the near future.
If some risks are addressed, others are not. Data (whether collected by payers, third parties, or well-meaning care providers) may be used for or against vulnerable groups. Medicare and Medicaid patients (the primary targets of ACO population management and most likely to be high utilizers of care) are targets for major funding cuts under the current US political administration. Proposals to repeal the ACA not only decrease support for these patients, but make coverage conditional on behavioral attributes and predicted risk. Specifically, one proposal allows insurers to use predictive analytics to identify potentially high-risk, high-cost patients and cede them to a high-risk pool, for which the federal government will pay a capped amount (Hall and Bagley, 2017).
At the time of this writing, cuts to cancer screening and other preventive care, social security, disability and child health programs, and more than $800 billion to Medicaid are proposed. Federal funding to help with social determinants (poverty, violence, housing and food insecurity, environmental toxins) is also being cut in the current administration. Additional severe cuts to environmental and consumer protection agencies undercut any downstream efforts to alleviate damaging health effects from social and environmental harms. Together, these represent the defunding of support to some of the most vulnerable in the US. Paradoxically, market-based healthcare qua value-based care is being asked to shoulder large-scale social justice issues the federal government demurs to take on.
Conclusions
In this article, I have shown how the assemblage of actants organized around value-based care enact accountability and value through practices of population health management using data-intensive sourcing, and I illuminated policy and political trajectories underpinning these new configurations. The socio-technical arrangements involved emerge within historical and political particularities of US market-based healthcare and the embedding of data-driven solutions to social problems such as public health.
Accountability devices such as outcomes measures and predictive analytics create particular kinds of groupings of individuals into populations, with the goal of managing population health and producing value in ways formulated by VBC. The unconventional processes of data collection, interpretation and category-making may change dynamically in the complex interactions of ACO assemblages, yet they have significant implications for patients and their care. Outcomes measures developed for official reporting requirements embed notions of what kinds of things come to count as risk objects, inscribe rules about how risk should be managed, and embody interpretations about what accountability means at a particular historical and political-economic moment. Patients are not passive in this picture: They are still held to account for their behaviors and health status. Value-creating activities rely on both producing new forms of knowledge through data-intensive sourcing and getting patients to act on ACOs’ filtered interpretations of those knowledge products in order to meet institutional benchmarks.
The practices of producing profiles for outcome measures may generate value according to defined outcomes per dollars spent, but they also generate a trove of data about individuals, which can be monetized and variably used for alternative purposes (Ruckenstein and Schüll, 2017). The process of implementing accountability devices thus reshapes the values they were supposed to set forth (Dussauge et al., 2015). The aim may have been to generate patient-centered improved outcomes and responsible cost-efficiency by determining what interventions on which individuals are prudent, but practices to instantiate the alternative payment system and document processes arguably serves the system as much as patients’ wellbeing.
In the ACO example, participants are pushed to change their practices rapidly to adapt to value-based care – but to do so under conditions shaped by interactions with other parts of the assemblage: laws and regulations changing not only healthcare payment systems but also mandating particular kinds of information technologies and circuits, expanded use of data analytics from consumer and finance industries, changing notions of acceptable evidence (risk modeling based on associative and machine-learning analytics rather than causal data and conventional epidemiological logics). While some of these conditions may exist elsewhere, the US assemblage also includes market competition for patients, entrepreneurial and proprietary control over databases, historical failed efforts to secure comprehensive healthcare reform and more.
The assemblage is already modulating. Since the first version of this article, CMS announced an overhaul of both MACRA and HITECH ‘meaningful use’ provisions, ostensibly to reduce regulatory burden. On closer inspection, this announcement by CMS head Seema Varma at the 2018 Health Information Management Systems Society conference is more of a reorientation from Federal reporting requirements to requirements to share data more broadly. As Jared Kushner (son-in-law and advisor to President Trump) put it: ‘The time is now to align every facet of the federal government and the private sector to ensure information is communicated and shared seamlessly.’ Varma plans to leverage CMS payment regulations and contracts with insurers to crack down on data blocking, with a primary focus on providing interfaces to allow the private sector to derive value from insurance claims data.
STS scholars have expertly analyzed the epistemology of emerging HIT and data analytics in specific scientific domains, and specific algorithmic means of digitizing persons and populations and the resulting effects. However, I encourage future STS scholarship to examine how emerging data infrastructures increasingly integrate domains. As I have shown, population health management spans clinical care, public health/epidemiology, biomedical research and behavioral economics. Following the data – as well as the money – shows how information flows beyond institutional or knowledge domain boundaries into everyday life. At the same time, modes of rationality, the social relations involved, and systems for governing and financing phenomena can be exposed by following data flows.
To this end, I used the idea of assemblage to point out some of the complex and dynamic interactions to bring an STS perspective to the analysis of public health. Assemblage makes the co-evolution of data and organizational infrastructures with value-based medicine more visible, and opens the way to follow such phenomena as they travel beyond activities related to ACOs themselves. In particular, more than merely creating an audit trail for accountability to public and private payers, the very acts of measuring and documenting outcomes serve to establish a broader networked platform with which to collect data about individuals far beyond any specific aims of population health management. Finally, if ACOs are accountable for creating value, then one has to ask whose sense of value counts, and to whom are ACOs really accountable?
Footnotes
Acknowledgements
I gratefully acknowledge the healthcare professionals, company representatives and other attendees at several value-based medicine conferences who shared their experiences, as well as helpful comments from anonymous reviewers. An early version of this article was presented at the 2016 4S/EASST conference in Barcelona Spain. Comments from attendees, and in particular, panel organizers Klaus Hoeyer and Martyn Pickersgill, were valuable contributions for the final article.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.