Abstract
INTRODUCTION
Running a hospital is a tough business. Patients must be attracted, admitted, treated, and discharged; physicians and staff hired, trained, and retained; new technology mastered; and most importantly, reimbursement dollars from payors captured and retained. The hospital Board of Directors and top administrators must also keep an eagle eye on staff practices and patient outcomes or hospital revenue will suffer. The United States has over 6,000 hospitals, 1 many in health care systems that provide almost ninety percent of U.S. hospital beds. 2 These hospitals are the major providers of emergency care and highly complicated surgical and other procedures in the United States. 3 They are also producers of patient harms. 4
Hospitals in the United States now face a perfect storm—the confluence of (1) a rising tide of hospital adverse events, 5 (2) a tsunami of Artificial Intelligence (“AI”) technological developments changing hospital practice, 6 and (3) the rapid onslaught of Medicare reimbursement tools 7 . These three pressure fronts together are battering the management of hospitals. Each corresponding technology—patient adverse event spotting, the use of powerful AI data analytic tools to improve health care delivery, and pay-for-performance reimbursement using incentives to improve institutional behavior—has promise, and one day they will work well together. At present, however, I will argue that all three—adverse event discovery, AI tools, and pay-for-performance (also known as “P4P”)—are halfway technologies 8 , aligning in such a way that they create a fog that obscures safety, distorts hospital spending, and pushes institutions toward imperfect patient safety systems.
The original meaning of the phrase “halfway technology,” as used by Lewis Thomas, was as follows:
Halfway technology … represents the kinds of things that must be done after the fact, in efforts to compensate for the incapacitating effects of certain diseases whose course one is unable to do very much about…. By its nature, it is at the same time highly sophisticated and profoundly primitive…. It is a characteristic of this kind of technology that it costs an enormous amount of money and requires a continuing expansion of hospital facilities.
9
Thomas uses as his examples the Iron Lung, a negative pressure ventilator that helps a person to breathe on his or her own in a normal manner when muscle control is lost which was used to treat patients with Polio in the 1950s, as well as dialysis, heart transplants, coronary artery bypass grafts, coronary stents, total hip and knee replacements, and drugs for hypertension. 10 These “primitive” tools are used “when physicians are bogged down by their incomplete technologies, by the innumerable things they are obliged to do in medicine when they lack a clear understanding of disease mechanisms, [when] the deficiencies of the health care system are most conspicuous.” 11
I use the phrase “halfway technology” here to describe a hospital environment in which tools are deployed too rapidly before they are mature, driven by both market and government pressures. 12 Hospitals adopt full use of these tools, in Lewis Thomas' words, before they have “a clear understanding” of their flaws, limits, and side-effects. 13 Because they are only “halfway tools” not completely tested, they risk poor results, or no results at all. These tools are however “halfway” only in the sense that they need further testing; they are innovative tools that improve over time—it is the early implementation that often creates problems for patient safety uses.
I will argue in this article that the entanglement of these metrics—adverse event tracking, AI tools, and pay-for-performance metrics—requires two counterforces to untangle them. First, hospital Boards of Directors must take a forceful ethical stance on the role of quality and patient safety as a central goal of hospital actions. The onslaught of expensive data analytics tools, coupled with premature P4P metrics, requires Boards to avoid distractions and focus on quality of care first. Boards can be distracted by the financial focus of hospitals, as they worry about reimbursement and reputation. 14 Second, Medicare needs to develop a better compensation model for the often harmed and often undercompensated Medicare patient. A new model needs to be developed and tested that uses Medicare tools to offer compensation to insured patientsa Medicare Patient Compensation Fund that produces a new stream of data as to patient harms from adverse events and creates the sting of a new set of financial penalties that start with current P4P metrics and improves them.
I. PATIENT SAFETY: SPOTTING ADVERSE EVENTS IN THE FOG OF HOSPITAL PRACTICE
A. Halfway Technology I: Adverse Event Detection
Adverse events are as old as medicine itself, and their frequency has increased as medicine has become technologically complex. An adverse event is defined by the federal Agency for Healthcare Research and Quality (“AHRQ”) as “[a]ny injury caused by medical care.” 15 AHRQ notes that this definition means “an undesirable clinical outcome,” ranging from hospital-induced infections to a pneumothorax from central venous catheter placement, without regard to what caused the event. 16
Production of medical adverse events has traditionally focused on the individual physician as the primary causal agent of patient harm. Primary care practice certainly is marked by frequent misdiagnoses that cause harmful delays in treatment and by adverse drug reactions—and outpatient care can also do harm. 17 I will however focus my attention on hospitals, where most high-risk care is now delivered and most adverse events are produced. 18 Patient harms in hospitals continue to be seriously underreported. A report by the Office of Inspector General (“OIG”) in 2011 found that hospital staff did not report eighty-six percent of events to incident reporting systems. 19 This failure to report is caused in large part by the lack of a federal comprehensive mandated national reporting system for adverse events, and half of states also lack any kind of reporting system. 20
The hospital environment lacks proper policing and is surprisingly full of risks to patients. 21 Even the best hospitals generate a high volume of patient harms. 22 A 2019 study based on data from Leapfrog, which grades hospitals semi-annually on patient outcomes, concludes that “[t]he number of avoidable deaths per 1,000 admissions ranged from 3.24 lives per 1,000 admissions in ‘A’ hospitals to 6.21 lives per 1,000 admissions in ‘D’ and ‘F’ hospitals.” 23 If all hospitals had scored an A performance, more than 50,000 patient lives could be saved, and there would still be considerable room for improvement. 24 A recent study concluded that medical error was the third most common cause of death in the United States. 25
The Institute of Medicine report, To Err Is Human (“IOM Report”), published in 1998, estimated a maximum of 100,000 patient deaths annually occurred due to medical errors. 26 The IOM Report—with its extrapolation of high levels of patient harms—created and energized the Patient Safety Movement. Three years later, the IOM published Crossing the Quality Chasm: A New Health System for the 21st Century, which stressed the importance of health care systems design. 27 It argued that most errors and adverse events in health care are caused by problems with system processes and not solely by provider error. 28
The high level of adverse events among Medicare patients in hospitals has been well documented by the OIG. 29 The OIG Report in 2012 found that “[a]n estimated 13.5 percent of hospitalized Medicare beneficiaries experienced adverse events during their hospital stays. Of the nearly 1 million Medicare beneficiaries discharged from hospitals in October 2008, about 1 in 7 experienced an adverse event that met at least 1 of our criteria (13.5 percent).” The OIG's conclusion: 27% of beneficiaries experienced care-related harm, and half of these harms were preventable. 30
B. Halfway Technology II. Data Analytics Tools
1. Electronic Health Record Utility
Adverse events are hard to detect.
31
Adverse event tracking, while it is improving, is still evolving … and expensive. Hospitals need to commit to well-tested AI tools to systematically find adverse events, because reliance on provider self-reporting has proved to be limited. Before Medicare can use reimbursement incentives to push hospitals to reduce adverse events, adverse events have to be discovered, tallied and stored in an accessible way for hospitals to track and then take action to reduce or eliminate. This has been a goal since the early 1900s, and progress has been slow.
32
In 2004 the Institute of Medicine report Patient Safety: Achieving a New Standard for Care (“IOM Patient Safety Report”) made a strong recommendation for computer systems to improve patient safety:
Traditional reporting systems can grossly under detect injuries, significantly impeding the ability to improve. A balanced detection system necessarily relies on case finding through surveillance, working together with voluntary incident reporting systems. Injury surveillance uses data-based clinical trigger systems that lead to prospective expert review, as well as retrospective review of patient records.
33
The IOM Patient Safety Report acknowledged the need for robust Electronic Health Record (“EHR”) systems to generate patient safety reports automatically during the patient care process. 34 The usefulness of these patient safety reporting systems, however, has been limited by the lack of EHR systems. The IOM Patient Safety Report notes: “[w]ithout EHRs, reporting systems typically rely on special data collection mechanisms (both human- and computer-based), making reporting a cumbersome, costly, and sporadic exercise. The data collected in these systems are neither complete nor standardized, making it difficult to aggregate the data or identify trends or patterns.” 35
The HITECH Act of 2009 was intended to speed up the slow adoption of electronic health records by hospitals. 36 The Act created $30 billion in incentives for adoption of EHRs: it subsidized adoption costs, provided technical support, and altered reimbursement rules, so long as an adopting physician was able to show “meaningful use” of the EHR system to improve patient care. 37 By 2012 the OIG had issued a clarion call to use EHRs and data analytics to improve patient safety, recommending that AHRQ and the Centers for Medicare & Medicaid Services (“CMS”) should work to identify all types of adverse events. 38 The OIG added that “CMS should provide further incentives for hospitals to reduce the incidence of adverse events.” 39 CMS had already begun its use of value-based-purchasing tools, and they increased their use after 2012. 40 It is as if the cart was loaded before the horse was even ready, given the continued problems with identifying adverse events. 41
EHRs are central tools for patient safety improvement. 42 Once an EHR system is in place, data analytic programs can be used to store, organize, and retrieve patient information; EHR applications can support healthcare providers and improve their ability to make both diagnostic and treatment decisions. 43 EHRs have tremendous potential, but have been slow to deliver. EHR software errors, coupled with user errors, have caused medical errors and several patient deaths. 44 EHR use also has produced surprisingly high levels of drug errors. 45
This rapid adoption of EHRs has created a range of problems—lack of standards, poor usability, and poor interoperability, among others. 46 The rapid push to adopt electronic health record technology by complex hospital systems has created a weak foundation for the application of pay-for-performance tools.
2. Training Modern Data Analytics
Today's hospitals are working to digitize and connect their facilities and hospital systems, and their take-up of data analytic tools is accelerating. 47 The AI health market may reach almost $7 billion by 2021—a compound annual growth rate of forty percent. 48 Vendors claim that this growth in use of AI by hospitals could cut U.S. healthcare costs by $150 billion a year by 2026, in addition to improving care. 49 The lure of AI and data analytics systems in hospitals is powerful, providing new and swift ways to discover and track abuses of patient harms. 50 This AI capability, if and when it works as promised, will ultimately alter the standard of care in hospitals, allowing for a real reduction of adverse patient events. 51 Hospital systems will have available to them the technology to discover adverse events and their causes, and these new technologies will create an intensified duty to implement policies to reduce such patient harms. 52 Hospital litigation already has begun to reflect new patient safety programs and their legal consequences. 53 AI diagnostic tools will also force hospitals to confront new standards of care for diagnosing patient diseases. 54
Data analytics connects statistics—the study of data relationships using numbers; artificial intelligence—the use of software and/or machines that display human-like behaviors; and machine learning or deep learning—algorithms that learn from existing data to make reliable predictions. 55 Health care is a perfect candidate for the power of Big Data, considering data source, the need to manage complex diseases, and Medicare's linking reimbursement to patient outcomes. 56
The critics argue that the capabilities of AI are overestimated at present as to patient care benefits. 57 EHRs are full of “messy” data 58 that is hard to read without natural language processing tools. 59 Data analytics provides the computing power and software to sort through large volumes of messy data; deep learning neural networks can learn from the data itself. 60 These tools can potentially work well in healthcare. 61 Fans of deep learning argue that it can incorporate the entire EHR to produce predictions for clinical problems and outperform state-of-the-art traditional predictive models. 62 Much of this, however, is stillthe promise of vendors of deep leaning tools and researchers who are exploring methods to improve deep learning. 63
A recent National Academy of Medicine report notes some of the limitations of implementing AI tools too early:
AI tools will also produce challenges that are entirely related to the novelty of the technology. Even at this early stage of AI implementation in health care, the use of AI tools has raised questions about the expectations of clinicians and health systems regarding transparency of the data models, the clinical plausibility of the underlying data assumptions, whether AI tools are suitable for discovery of new causal links, and the ethics of how, where, when, and under what circumstances AI should be deployed []. At this time in the development cycle, methods to estimate the requirements, care, and maintenance of these tools and their underlying data needs remain a rudimentary management science.
64
Optimistic assumptions about the use of EHR data sources assume that such EHRs are “interoperable,” i.e. easily exchanged across health systems. This is often not the case. 65 Many of the theoretical benefits of EHRs depend on their being interoperable. 66 A truly interoperable EHR system could reduce medical errors by, for example, equipping emergency room doctors admitting an unconscious patient to quickly look up the patient's medical history and current medications. 67 Such care coordination allows providers working in different health systems who care for a single patient to share notes and records. 68 It could make possible detection of patient problems before they even materialize and prevent them. Data analytic AI tools may plausibly evolve from halfway technologies to powerful, effective, and bug-free tools over time, but in the interim, regulators need to be attentive to the limits of their tools. 69
C. Halfway Technology III. Pay-for -Performance in Health Care
1. The Evolution of Medicare's Regulatory Model
Payment matters in health care and so does the design of payment. Medicare pays for almost half of hospital care in the United States. 70 Unlike other federal benefit programs that provide health care services and items, Medicare has gone from a payor without question in the years after its passage in 1965, to a purchaser of covered items and services from independent vendors, and finally, as health care costs came to consume more and more of the federal budget, to a payor who uses reimbursement penalties to squeeze hospital revenues. 71
The use of economic incentives in the Medicare program is not new. First came the Diagnosis-Related Groups (“DRG”) program in 1983, a program designed to slow what had become rapid hospital price inflation. 72 It set maximum charges for hundreds of medical treatments by establishing 467 diagnostic payment rates for the 4,800 hospitals of that time, using historical Medicare data. 73 The Social Security Amendments of 1983 established the DRG prospective payment system for inpatient hospital services, in which Medicare pays hospitals a fixed fee for each type of case, determined in advance and based on the relative average cost of treating that type of case in hospitals nationwide instead of the hospital's own costs. 74 It forced hospitals to look at these procedures and see which ones were losing revenue and which were producing an excess of revenue over cost. 75
Medicare's DRG payment system shifted the balance of economic power between the providers of medical care (hospitals and physicians) and those who paid for it, power that providers had acquired for more than half a century. 76 The DRG program was in fact successful in rationalizing hospital budgeting and taming inflation. 77
With the publication of the Institute of Medicine report, Crossing the Quality Chasm, 78 Medicare turned its attention to more powerful reimbursement tools to both control Medicare expenditures and improve hospital quality of care. 79 CMS developed specific reimbursement measures as a regulatory tool to incentivize good hospital care and to punish hospitals for substandard performance: pay-for-performance 80 and value-based-purchasing 81 models were developed to create incentive systems where quality shortcomings trigger hospital reimbursement penalties. 82 The use of pay-for-performance metrics are designed to score hospitals on measurable specific conditions like Hospital-Acquired Conditions (“HACs”), through the HAC Reduction Program 83 , and through the use of patient outcome measures, such as patient safety indicators (“PSIs”). 84 As regulator, CMS has come to demand of hospitals that they fix excessive readmissions, reduce PSIs, and meet quality metrics or they will experience a Medicare reimbursement hit. 85
Data mining is used, for example, by hospitals to find HACs and to prevent them, in order to avoid CMS financial penalties. 86 Such tracking efforts will increase as (1) hospitals adopt data analytics and (2) ever-expanding federal pay-for-performance programs condition larger fractions of a hospital's reimbursement on performance standards for hospital-acquired events. Data analytics can aid clinicians in eliminating adverse events and resulting excessive costs. 87 It can also assist clinicians in determining care plans, for example, based on estimating disease trajectories of cancer patients from unstructured, free text in EHRs. 88 These tools have power to improve the care of Medicare patients in hospitals, without a doubt. The Medicare program has pushed these P4P tools to promote quality in hospitals but also to reduce costs to the program. 89 Medicare has enthusiastically noted that its non-payment policy for HACs has saved Medicare almost $350 million each year. 90 Medicare indeed has saved money, but these dollars represent revenue lost to hospitals that could have been spent by hospitals on other areas of quality improvement. 91 If the reimbursement measures used are well justified and fairly applied, then the Medicare system gains in both quality and cost; if however these metrics are poorly tested and are unproven, hospitals are wasting scarce staff time and resources. 92
Hospitals today face what has been called “confusing complexity—safety initiatives focused on a broad array of specific safety targets with interventions for each one: process standardization, checklists for surgery, bundles for central lines, electronic prescribing and order entry, and medication barcoding.” 93 For the hospital statisticians, epidemiologists, and data analytic officers, there are too many metrics and systems to easily keep track of, as they are of differing validity, and often hard to apply to the clinical setting.
2. Limits of Halfway Technologies Such as P4P
Value-based-purchasing and pay-for-performance are growing in importance for hospitals, as described above. The rationality of using financial incentives to improve the Medicare program appeals to authors of the IOM reports and to policymakers—it is a potent mix of the ideology of market-like incentives to affect institutional behavior and the attraction of CMS to continued accumulation of political/financial muscle over often-resistant hospitals. 94 In the process, CMS is forcing hospitals to face increasing amounts of money at risk with every patient admission. 95 The Medicare program hopes that Big Data and data analytics will achieve remarkable efficiencies in the messy U.S. health care system, improving diagnoses and reducing costs and adverse events. 96
Here we see the fusion of economic tools and computer power shifting the fundamental role of hospitals from fiduciaries of their patients' care—putting patient safety and welfare first—to economic entities driven by the lure of increased value-based payments and the skittish avoidance of the sting of penalties. Neither pay-for-performance or data analytics tools perform miracles.
CMS, private insurance, and many states have taken an enthusiastic approach to the use of data analytic tools and measures to compare or benchmark hospitals' performances for patient safety. 97 Such tools are certainly necessary to track reimbursement driven incentive and penalty systems under Medicare. The problem is that the shortcomings of AI and its applications suggest a halfway technology that still has limitations along with its promise. 98 AI has not yet been proven to be clinically effective in many clinical applications. AI requires high quality data sources, 99 but most clinical data, whether from EHRs or medical billing claims, remain ill-defined and not easily exploited by AI technologies. EHRs for example lack specificity measures such as duration, severity, and pathophysiologic mechanism. 100 Such records also create their own risks to patient safety: adverse events including deaths are traceable to software glitches, user errors or other system flaws, and are not yet well understood. 101 Like adoption of many new technologies, safety depends on slow adoption with careful attention to the risks created and constant recalibration of the metrics and assumptions of such systems. 102
Hospitals are generally scored according to patient outcome measures. 103 These measures include specific PSIs. 104 Total Medicare payments to the worst-scoring facilities with HAC Reduction Program scores greater than the 75th percentile are reduced by 1%, and this percentage is likely to increase. 105 The hospital industry now faces financial pressures to reduce these harms or suffer both reimbursement penalties and loss of patients due to published hospital rankings and posting of quality scores. 106
The authors of one study found that only one measure out of twenty-one scientific criteria considered as true indicators of hospital safety—accidental punctures or lacerations obtained during surgery—met the researchers' validity criteria. 107 This clearly shows the rush by CMS to impose metrics lacking in evidence in order to validate the concept of P4P penalties and generate savings in the Medicare budget. If CMS measures are invalid, then reimbursement penalties are not empirically justified. And public rating systems magnify the effect of these CMS measures, as U.S. News and World Report's Best Hospitals, Leapfrog's Hospital Safety Score, and CMS's Star Ratings end up broadcasting and amplifying measures that lack validity. 108
The allocation of significant resources to PSI and HAC reimbursement penalties has a signification budgetary effect on hospitals, both positive and negative. 109 The benefit is more attention paid to patient safety—hospital leadership hires more staff, upgrades software, and demands better teamwork. 110 These are generally positive, increasing staff adherence to evidence-based guidelines (“EBGs”) and new protocols for prevention and prophylactic care treatment, new safety procedures for HAC prevention, and so on. 111
The negatives however are substantial. First, the HAC penalties in particular distort investments in patient safety and damage hospital reputations unfairly, while having little or no effect on the reduction of HACs. 112 Hospital staff spend time and resources; more specialists are hired; new oversight structures are created; productivity of clinical staff and coders decrease. 113 Some of these changes may be valuable, but others are a waste of time and money if few specific changes in HAC incidence patterns occur. As one commentator has noted, “[t]hese measures have the ability to misinform patients, misclassify hospitals, misapply financial data and cause unwarranted reputational harm to hospitals.” 114 These unvalidated metrics distract from patient safety efforts. 115 Second, the HAC penalties, while failing to drive effective clinical improvements, appear to penalize hospitals who care for more disadvantaged patients; one study concluded that this “could exacerbate inequities in care.” 116
The problem is not just that the metrics used to justify P4P penalties lack validity, but also that AI tools have yet to be proven clinically effective in many clinical applications. 117 AI will succeed if used with high quality data sources that allows it to “learn” and classify data in relation to outcomes. 118 Most clinical data, on the other hand, whether from EHRs or medical billing claims, is poorly defined and therefore insufficient for effective exploitation by AI technologies. 119 EHR data on demographics, clinical conditions, and treatment plans, for example, are generally of low dimensionality and are recorded in limited, broad categorizations (i.e., diabetes) that omit specificity (e.g., duration, severity, and pathophysiologic mechanism). 120 Without specificity and high dimensionality, the EHR data is of little value. 121
One conclusion is that reimbursement metrics should be reformed to use clinical rather than billing data. 122 HAC and PSI penalties are flawed and need to be fixed before reimbursement penalties are imposed. 123 Hospitals are slow learners, and the pressure of poorly validated reimbursement penalties in the U.S. hospital system further impedes their ability to learn. 124 Pay-for-performance has so far largely failed to produce improvements in patient health, especially as it cuts or otherwise reallocates hospital income away from productive patient safety initiatives. 125
II. COUNTERACTING THE PRESSURES
A. Creating A Compensation Program for Medicare
Why develop a compensation program for Medicare patients when the existing tort system is available to such patients? It is clear that elderly patients suffer from more preventable adverse events. Adverse drug events, procedures, and falls are common in the Medicare population, and in the words of one study, these events “should be targets for efforts to prevent errors.” 126 The incidence of adverse drug events was found to be 14.8% among hospitalized elderly patients with a mean age of 78. 127 “This figure is much higher than the reported rate of adverse drug events in patients of all ages from other studies.” 128
Medicare patients, in other words, fare poorly as a patient population compared with other age groups. 129 They also fare poorly with the medical malpractice claiming process. 130 The elderly suffer from lower claiming rates in part because of the impact of tort reform, which has reduced potential recoveries and the incentives of plaintiff lawyers to take the cases. 131 One study concluded that “the share of med mal payouts to the elderly remains well below their share of health care use. And total payouts to elderly claimants, after rising steadily during the pre-reform period, have dropped back to the low levels that prevailed at the start of our sample period.” 132 The elderly lack lost wages, have low life expectancy, and present an unsympathetic case for pain and suffering in their old age. The reality of a low claim for damages means that trial lawyers, litigating based on a contingency fee, find the Medicare elderly unrewarding plaintiffs.
My proposal is to take advantage of two features of modern Medicare patient care. The first step is to take advantage of the tools of Big Data as they improve as a way to discover adverse events that may not otherwise be visible. 133 Automated search programs are already spotting such adverse events. 134
The second step is to develop a Medicare-based administrative claim system for harms suffered by patients due to preventable adverse events, such as the Nordic countries have developed. 135 One proposal is a compensation system using Medicare's Quality Improvement Organizations (“QIOs”), drawing on a proposal by Eleanor Kinney and William Sage. 136 Kinney and Sage propose that these QIOs would be tasked with “hearing and resolving cases of patient injury, with mediated discussions, reasonable awards of compensation, and feedback to participating institutional providers regarding the safety and quality of their care.” 137 QIOs at present have nothing to do with patient compensation, focusing on the metrics of quality improvement and cost reductions, but they do offer administrative frameworks within which a fair compensation system could be imbedded. 138
Such a compensation system might consist of several features. First, hospitals would be required to disclose Preventable Adverse Events (“PAEs”). 139 Second, the system would be funded through Medicare by an assessment on hospitals. Third, hospitals' assessments would be linked over time to their level of adverse event compensation, with reductions of their assessments as adverse events decreased. 140 Fourth, a fixed schedule of payments would be designed as part of a benefit structure, as opposed to a damage schedule modeled on tort reform. 141 Fifth, Medicare patients would be foreclosed from suing in civil court for medical malpractice. Sixth, the QIO program, by existing within the federal Medicare program, 142 would avoid the problem of trying to get fifty states to buy into a national compensation system of some kind. 143 This is a pure adverse event-driven benefit system that broadens the use of data analytic tools to develop effective adverse event search tools. 144
Various models of enterprise liability have been considered before as ways to improve the patient compensation system. 145 It is time again to develop a model of Medicare patient compensation, as a new approach to preventable adverse events. Ironically, it is the promise of the tools of AI and P4P that can underpin such a new benefit system, informed by more robust data and by an incentive structure that finally motivates hospitals to make patient safety central to their decision-making.
B. Articulating A Robust Hospital Fiduciary Duty
Hospitals are increasingly under pressure from multiple sources to track and prevent adverse event creation. 146 At the same time they face ethical mandates to manage AI problems that create ethical violations. 147 Hospital Boards of Directors, under the Medicare Conditions of Participation, are required to have hospital bylaws that reflect the accountability of the medical staff to hospital governing board or “governing body for the quality of care provided to patients.” 148 It isn't clear, however, that Medicare enforces these Conditions in any serious way.
Cultural changes in hospital patient safety require changes in the culture of the Boards of Directors as well. 149 Changes in cultures do produce changes in patient care and patient safety. 150 Studies have found that a range of board practices can improve hospital care and mortality. Such practices include (1) having a board quality committee; (2) establishing strategic goals for quality improvement; (3) being involved in setting the quality agenda for the hospital; (4) including a specific item on quality in board meetings; (5) using a dashboard with national benchmarks that includes indicators for clinical quality, patient safety, and patient satisfaction; and (6) linking senior executives' performance evaluation to quality and patient safety indicators. 151
A hospital Board of Trustees is a co-fiduciary with its physicians and staff. This creates an ethical obligation to protect patient safety and health. 152 Hospitals have a developing fiduciary duty to patients, with clear obligations to coordinate care to minimize patient safety risks, including informing patients that they have experienced adverse events and offering compensation for such events. 153 The expansion of corporate negligence doctrine in many states contains the kernel of an expanding fiduciary duty on hospitals. 154 The hospital is viewed as a corporate management structure with responsibilities to its patients, measured by its own internal practices as well as those of other similar hospitals. 155
These duties give new content to the existing duties of loyalty traditionally applied in corporate law to hospitals boards of directors. 156 This means that when hospitals take on expanded norms of fiduciary obligations to patients, this includes the obligation of “creating and sustaining an organizational culture that unites administration and physicians in the task of protecting and promoting the health-related interests of patients.” 157 This ethical role of the hospital Board of Directors should be expanded, as discussed above, as part of Board duties. It provides a strong counterpressure to the temptations of Boards to focus only on reimbursement penalties, arming the Board to demand more of hospital managers in developing a patient safety culture.
CONCLUSION
The technologies that hospitals are now deploying for regular use are often at odds with the goal of high quality and patient safety. Patient adverse event detection is improving with the development of new software programs to hunt for variation in hospital infections and other patient safety indicators. The adoption of EHRs and data analytics by U.S. hospitals continues, as hospitals struggle with the often problematic nature of such tools. And Medicare continues to ramp up its use of pay-for-performance tools to incentivize hospitals to reduce their readmissions, infections, and adverse events. The problem, as we have seen above, is not that these goals are not worthy, but that the tools are only halfway ready; hospitals are pressured to adopt these tools of data analytics and P4P metrics by sheer competition in the health law marketplace and by CMS as it pushes for visible progress in hospital quality and cost control of rising health care costs.
I have argued that the above three pressure fronts are battering the management of hospitals. Each technology has great potential, once they are coordinated and mature, but I have argued that all three are still halfway technologies, 158 applied while immature to distract hospitals from safety considerations in a more coherent and global way, distorting hospital spending, and misdirecting some of that spending; this pushes hospitals toward imperfect patient safety systems.
I have proposed two major counterforces that need to be developed. First, a new model of ethical obligations of hospital Boards of Directors is needed, in which the Board takes an aggressive role in mandating patient safety reforms while challenging the blind application of CMS P4P metrics that may not achieve their quality goals. Second, I advocate a new model of Medicare patient compensation that uses QIOs for more than appeals, turning QIOs into a benefit payment role for preventable adverse events.
The hope is that the two counterforces will slow down the adoption of the halfway technologies and drive them to a more developed set of tools that will make Big Data powerful and effective, and P4P meaningful in turning the incentives into real quality improvements. This is a large task, but the status quo at the moment suggests that federal dollars are being spent to a large extent in vain—and that is a waste of our resources.
Footnotes
1
See Fast Facts on U.S. Hospitals, A
].
2
See A].
3
Robert E. Suter, Emergency Medicine in the United States: A Systematic Review, 3 W
4
See Barry R. Furrow, Adverse Events and Patient Injury: Coupling Detection, Disclosure, and Compensation, 46 N
5
Id.
6
Barry R. Furrow, The Limits of Current AI in Health Care: Patient Safety Policy in Hospitals, 12 N
7
Charles N. Kahn III et al., Assessing Medicare's Hospital Pay-For-Performance Programs and Whether They Are Achieving Their Goals, 34 H
8
L
9
Id.
10
Id. (recalling “the early 1950s, just before the emergence of the basic research that made the [polio] vaccine possible … the cost of those institutes for rehabilitation, with all those ceremonially applied hot fomentations, and the debates about whether the affected limbs should be totally immobilized or kept in passive motion as frequently as possible, and the masses of statistically tormented data mobilized to support one view or the other?”).
11
Id. at 20.
12
Id. at 18–19.
13
Id. at 20.
14
Cheryl L. Wagonhurst and M. Leeann Habte, Health Care Boards of Directors' Legal Responsibilities for Quality, H].
15
Agency for Healthcare Research and Quality, Glossary, PSN] [hereinafter AHRQ Glossary].
16
AHRQ Glossary, supra note 15 (click “Adverse Event”).
17
David E. Newman-Toker et al., Serious Misdiagnosis-Related Harms in Malpractice Claims: The “Big Three” – Vascular Events, Infections, and Cancers, 6 D
18
René Schwendimann, et al., The Occurrence, Types, Consequences and Preventability of Inhospital Adverse Events – a Systematic Scoping Review, 18 BMC H
19
O
20
Furrow, supra note 4 at 451–53.
21
See Eric Nalder & Cathleen F. Crowley, Patients Beware: Hospital Safety's a Wilderness of Data, H] (illustrating that hospitals often underreport adverse events and showing that, in some instances, hospitals have missed cases where patients were killed).
22
David C. Classen et al., ‘Global Trigger Tool’ Shows that Adverse Events in Hospitals May Be Ten Times Greater than Previously Measured, 30 H
23
M
24
Id. at 6.
25
Martin A. Makary & Michael Daniel, Medical Error—The Third Leading Cause of Death in the US, 353 BMJ 2139, 2140 (2016).
26
I
27
I
28
Id. at 25.
29
O
30
Id; Spotlight On … Adverse Events, O].
31
See Furrow, supra note 6, at 6.
32
Id. at 7–9.
33
P
34
Id. at 7.
35
Id. at 5.
36
Julia Adler-Milstein & Ashish K. Jha, HITECH Act Drove Large Gains in Hospital Electronic Health Record Adoption, 36 H
37
Brian Schilling, The Federal Government Has Put Billions into Promoting Electronic Health Record Use: How Is It Going? C].
38
OIG Report, supra note 29, at iii.
39
Id. at 32 (emphasis added).
40
Id. at 40.
41
See Furrow, supra note 4, at 448.
42
Seth Freedman et al., Information Technology and Patient Health: Analyzing Outcomes, Populations, and Mechanisms, 4 A
43
Id. at 54.
44
See Fred Schulte & Erika Fry, Death By 1,000 Clicks: Where Electronic Health Records Went Wrong, K].
45
Raj. M. Ratwani et al., Identifying Electronic Health Record Usability and Safety Challenges in Pediatric Settings, 37 H
46
See generally Ross Koppel, Uses of the Legal System That Attenuate Patient Safety, 68 D
47
Internet of Things, A].
48
Artificial Intelligence: Healthcare's New Nervous System, A].
49
Id. at 1.
50
I focus on patient safety here, but AI has many other potential benefits in health care. See, e.g., id.
51
For a good discussion of AI history and definitions, see A
52
See, e.g., Barry R. Furrow, Searching for Adverse Events: Big Data and Beyond, 27 Annals Health L. 149, 160, 178-79 (2018).
53
See id. at 167-68.
54
A. Michael Froomkin et al., When AIs Outperform Doctors: Confronting the Challenges of a Tort-Induced Over-Reliance on Machine Learning, 61 A
55
Data Mining: What it is and Why it Matters, SAS, https://www.sas.com/en_us/insights/analytics/data-mining.html [].
56
W. Nicholson Price, Artificial Intelligence in the Medical System: Four Roles for Potential Transformation, 18 Y
57
Karen Weintraub, The Power of AI, W
58
See Laura Lovett, Organizing Messy Data, a Google Developer's View, M].
59
Lauren E. Sweet & Heather L. Moulaison, Electronic Health Records Data and Metadata, 1 B
60
Artificial Intelligence: What it is and Why it Matters, SAS, https://www.sas.com/en_us/insights/analytics/what-is-artificial-intelligence.html [].
61
One such tool is the use of automated electronic search strategies. Automated extraction of data from electronic health records (EHRs) conducts high-quality retrospective analysis of large patient cohorts; these automated techniques can predict with high accuracy preoperative predictors and identify postoperative complications, such as postoperative myocardial infarction in large cohorts of surgical patients. Oludare O. Olatoye et al., Derivation and Validation of an Automated Electronic Search Algorithm to Identify Patients at Risk for Obstructive Sleep Apnea, 13 S
62
See Andrew J. Steele et al., Machine Learning Models in Electronic Health Records Can Outperform Conventional Survival Models for Predicting Patient Mortality in Coronary Artery Disease, 13 PLOS O
63
See, e.g., M
64
Stephan Fihn et al., Chapter 6 Deploying AI in Clinical Settings, in A
65
See Sharona Hoffman & Andy Podgurski, E-Health Hazards: Provider Liability and Electronic Health Record Systems, 24 B
66
See S
67
S
68
See Craig Konnoth & Gabriel Scheffler, Can Electronic Health Records Be Saved?, 46 A
69
See Sabyasachi Dash et al., Big Data in Healthcare: Management, Analysis and Future Prospects, 6 J. B
70
Ricardo Alonso-Zaldivar, Government headed for close to half of nation's health tab, AP (Feb. 20, 2019), https://apnews.com/5dc460ae8d8b4c8a93c6c3108fd71e9c [].
71
See Eleanor D. Kinney, Medicare Payment to Hospitals for a Return on Capital: The Influence of Federal Budget Policy on Judicial Decision-Making, 11 J. C
72
Judith Mistichelli, Diagnosis Related Groups (DRGs) and the Prospective Payment System: Forecasting Social Implications, 4 B
73
Id.
74
Kinney, supra note 71, at 455-57.
75
Id. at 456-57.
76
Rick Mayes, The Origins, Development, and Passage of Medicare's Revolutionary Prospective Payment System, 62 J. H
77
Natasa Mihailovic et al., Review of Diagnosis-Related Group-Based Financing of Hospital Care, 3 H
78
I
79
See Eleanor D. Kinney, The Accidental Administrative Law of the Medicare Program, 15 Y
80
I
81
“The Hospital VBP Program rewards acute care hospitals with incentive payments for the quality of care provided in the inpatient hospital setting. This program adjusts payments to hospitals under the Inpatient Prospective Payment System (IPPS) based on the quality of care they deliver, paying based on the quality of care provided to Medicare patients.” The Hospital Value-Based Purchasing (VBP) Program, C]. The Hospital VBP Program “withholds participating hospitals' Medicare payments by a percentage specified by law (2%); uses these reductions to fund incentive payments to hospitals based on their performance; applies the net result of the reduction and the incentive as a claim-by-claim adjustment factor to the base operating Medicare severity diagnosis-related group (MS-DRG) payment amount for Medicare fee-for-service claims in the fiscal year associated with the performance period.” Id. The measures in the VBP Program include “mortality and complications; healthcare-associated infections; patient safety; patient experience; process; and efficiency and cost reduction.” Id.
82
I
83
Hospital-Acquired Condition Reduction Program (HACRP), C].
84
Fact Sheet on Patient Safety Indicators, A] [hereinafter Fact Sheet on Patient Safety].
85
Hospital-Acquired Condition Reduction Program Fiscal Year 2020 Fact Sheet, C] [hereinafter HAC Reduction Program Fact Sheet].
86
See Jennifer Bresnick, Using Big Data Analytics for Patient Safety, Hospital Acquired Conditions, H].
87
See, e.g., Kasper Jensen et al., Analysis of Free Text in Electronic Health Records for Identification of Cancer Patient Trajectories, 7 S] (noting that data-driven decision-making may decrease adverse events and readmissions, as well as provide higher quality care).
88
Id. (“By using these disease trajectories, we predict 80% of patient events ahead in time. … We believe that the presented methodology and findings could be used to improve clinical decision support and personalize trajectories, thereby decreasing adverse events and optimizing cancer treatment.”).
89
See Medicare “Pay for Performance (P4P)” Initiatives, C].
90
Hospital-Acquired Condition (HAC) Reduction Program, C].
91
See Jessica Martin, Substantial Economic Burden Attributed to Recurrent Clostridium Difficile, I] (explaining the money lost by hospitals for treatment of C. diff infections, i.e. the money lost by hospitals as a result of CMS' non-payment policy for HACs).
92
See Elizabeth A. Fehlberg et al., Impact of the CMS No-Pay Policy on Hospital-Acquired Fall Prevention Related Practice Patterns, 00 I] (“The CMS no-pay policy increased utilization of fall prevention strategies despite little evidence that these measures prevent falls.”).
93
Tejal K. Gandhi et al., Patient Safety at the Crossroads, 315 JAMA 1829, 1829 (2016).
94
See Gerard Anderson et al., Medicare Payment Reform: Aligning Incentives for Better Care, C].
95
See id. (discussing penalties for hospitals with poor performance or higher-than-expected rates of hospital-acquired conditions).
96
See Jessica Kent, Medicare ACOs Use Analytics for Care Coordination, Population Health, H].
97
See generally C].
98
See Wullianallur Raghupathi & Viju Raghupathi, Big Data Analytics in Healthcare: Promise and Potential, 2 H
99
Thomas Maddox et al., Questions for Artificial Intelligence in Health Care, 321 JAMA 31, 31 (2019).
100
Id. at 31.
101
Schulte & Fry, supra note 44.
102
See Venkataraman Palabindala et al., Adoption of Electronic Health Records and Barriers, 6 J. C
103
Fact Sheet on Patient Safety, supra note 84.
104
Id.
105
HAC Reduction Program Fact Sheet, supra note 85.
106
See Armineh Zohrabian et al., The Economic Case for U.S. Hospitals to Revise Their Approach to Heart Failure Readmission Reduction, 6 A
107
Bradford D. Winters et al., Validity of the Agency for Health Care Research and Quality Patient Safety Indicators and the Centers for Medicare and Medicaid Services Hospital-Acquired Conditions, 54 M
108
See id. at 1106; see also Austin, supra note 106, at 429.
109
See Asta Sorensen et al., HAC-POA Policy Effects on Hospitals, Other Payers, and Patients, 4 M
110
See id. at E11.
111
Id. at E7.
112
Kyle H. Sheetz et al., Hospital-Acquired Condition Reduction Program Is Not Associated With Additional Patient Safety Improvement, 38 H
113
Sorensen et al., supra note 109, at 6–7.
114
Common Hospital Safety Measures Are Often Misleading to Public, J].
115
Winters, supra note 107, at 1105.
116
See Roshun Sankaran et al., Changes in Hospital Safety Following Penalties in the U.S. Hospital Acquired Condition Reduction Program, 366 BMJ l4109 (2019).
117
Maddox et al., supra note 99, at 31.
118
Id.
119
Id.
120
For a more thorough treatment of EHRs and data problems, see Furrow, supra note 4.
121
Id.
122
Pronovost recently outlined additional fixes that could be implemented by the rating community in a commentary published in the April 2016 issue of JAMA. See Ashish Jha & Peter Pronovost, Toward a Safer Health Care System, 315 JAMA 1831, 1831 (2016) (designating a separate reporting entity to establish standards for data collection and making funds available for systems engineering research were listed as possible starting points).
123
Rishi K. Wadhera et al., The Hospital Readmissions Reduction Program—Time for a Reboot, 380 N
124
See Anita L. Tucker & Amy C. Edmondson, Why Hospitals Don't Learn from Failures: Organizational and Psychological Dynamics that Inhibit System Change, 45 C
125
Jake Miller, Pay-for-Performance Fails to Perform, H].
126
Eric J. Thomas & Troyen A. Brennan, Incidence and Types Of Preventable Adverse Events In Elderly Patients: Population Based Review Of Medical Records, 320 BMJ 741, 743 (2000).
127
Id.
128
Id.; see also Michelle M. Mello & Allen Kachalia, Medical Malpractice: Evidence on Reform Alternatives and Claims Involving Elderly Patients, A Report to the Medicare Payment Advisory Commission, M
129
Id. at 1, 23-26.
130
Id. at 22, 25.
131
Id. at 32, 37.
132
Myungho Paik et al., How Do the Elderly Fare in Medical Malpractice Litigation, Before and After Tort Reform? Evidence from Texas, 14 A
133
See Furrow, supra note 6 at 16–18.
134
Id.
135
See, e.g., Kenneth Watson & Rob Kottenhagen, Patients' Rights, Medical Error and Harmonisation of Compensation Mechanisms in Europe, 25 E
136
See generally Eleanor D. Kinney & William M. Sage, Dances with Elephants: Administrative Resolution of Medical Injury Claims by Medicare Beneficiaries, 5 I
137
Id. at 7.
138
See C
139
See, e.g., Allen Kachalia et al., Effects Of A Communication-And-Resolution Program On Hospitals' Malpractice Claims And Costs, 37 H
140
Randall R. Bovbjerg & Laurence R. Tancredi, Liability Reform Should Make Patients Safer: “Avoidable Classes of Events” are a Key Improvement, 33 J. L. M
141
Randall R. Bovbjerg, Reform of Medical Liability and Patient Safety: Are Health Courts and Medicare the Keys to Effective Change?, 9 J. H
142
William M. Sage & Eleanor D. Kinney, Medicare-Led Malpractice Reform, in M
143
See Michelle Mello et al., “Health Courts” and Accountability for Patient Safety, 84 M
144
See Barry R. Furrow, Searching for Adverse Events: Big Data and Beyond, 27 A
145
William M. Sage, The Role of Medicare in Medical Malpractice Reform, 9 J. H
146
See Ziad Obermeyer & Ezekiel J. Emanuel, Predicting the Future – Big Data, Machine Learning, and Clinical Medicine, 375 N
147
Elliott Crigger & Christopher Khoury, Making Policy on Augmented Intelligence in Health Care, 21 AMA J. E
148
42 C.F.R. § 482.12(a)(5) (2019).
149
See Timothy J. Vogus et al., Doing No Harm: Enabling, Enacting, and Elaborating a Culture of Safety in Health Care, 24 A
150
H. Joanna Jiang et al., Board Oversight of Quality: Any Differences in Process of Care and Mortality?, 54 J. H
151
Id. at 15.
152
Barry R. Furrow, Patient Safety and The Fiduciary Hospital: Sharpening Judicial Remedies, 1 D
153
See, e.g., T.A. Faunce & S.N. Bolsin, Fiduciary Disclosure of Medical Mistakes: The Duty to Promptly Notify Patients of Adverse Health Events, 12 J. L. & M
154
Furrow, supra note 152.
155
156
Thomas C. Tsai et al, Hospital Board and Management Practices Are Strongly Related to Hospital Performance on Clinical Quality Metrics, 34 H
157
Frank A. Chervenak et al, Physicians and Hospital Managers as Cofiduciaries of Patients: Rhetoric or Reality, 48 J. H
158
T