cover story one

Health Care Utilization

Finding the right balance

By Zeke McKinney, MD, MHI, MPH

Under-testing in high-risk patients and over-testing in low-risk patients were divergent problems.

Underutilization by Physicians 

There is no shortage of studies evaluating the issue of variation in physician utilization of health care services. To a large degree, physicians suffer from cognitive overload as it pertains to medical decision-making. In particular, the exponential increase in new research, testing and treatment paired with systemic pressures on increasing efficiency and volume, implicitly portend an eventual mismatch in a physician’s ability to make the “right” decision in a relatively small amount of time. As such, physicians’ decisions suffer from cognitive bias in a number of ways, including the inability to continually be up-to-date on the latest recommendations, they may focus on clinical “horses” while dismissing clinical “zebras”. There are challenges communicating with and correctly understanding patients from a myriad of intersecting biopsychosocial identities. These can lead to the potential and presence of differential treatment for patients based on various identifying factors like race, ethnicity, gender, age, socioeconomic status, educational background, or preferred language. Though even when physicians are making (or want to make) the “right” decisions for their patients, often the financial barriers of limited reimbursement, or even coverage in any capacity, preclude them from doing so. The financial mechanisms driving health care administrators’ decisions around specifying patient volumes, basing physician salaries on relative value units (RVUs; not always congruous with volumes), providing staff for administrative tasks, and opening/closing clinical sites, are major forces in affecting access. These factors have an impact on when and how physicians provide care, and in extreme cases may cause physicians to leave an institution or the field of medicine altogether. The clinical and financial outcomes of physician variation in utilization based on these and other factors have been evaluated in a variety of studies, with some striking results.

Early studies of utilization often focused on cost savings, while more recent studies tend to evaluate clinical or patient-oriented outcomes. In part, this focus on financial outcomes was a function of considering the economic theory of “moral hazard”, which implies one party in a financial transaction has an incentive to increase its risk because it has a disproportionate share of protection from the risk. Moral hazard is more clear in banking and insurance than in health care. This is due largely to a lack of health care price transparency. Physicians are often unaware of the cost of the care they provide and patients often do not find out their costs for those services until well after they were provided. Lacking the fundamental element of cost, or risk, in assessing how best to utilize health care, moral hazard can occur for both patient and physician in both consuming and delivering care. It is often assumed, by both parties, that a patient’s degree of insurance coverage and amount of out-of-pocket costs is proportional to the amount of health care they consume. Unfortunately this is not always the case and economists criticize interpretations of moral hazard in terms of health care delivery, because they suggest health care mirrors utilization of other goods or services. When people are very ill, they use health care only because it is absolutely necessary, whereas fully-covered preventive services often are avoided simply because the value is difficult to perceive when someone is feeling completely well. As moral hazard applies, if patients don’t have to pay for care, they don’t care how much it costs. If they have to pay for it, they may choose not to seek care. One instance of this effect was observed in a study evaluating emergency department (ED) visits in Oregon, where it was posited that Medicaid patients (who essentially have no out-of-pocket costs), would utilize more preventive (high-value) care versus ED visits (considered low-value), but the exact opposite effect was found: ED visits increased 40 percent in the presence of Medicaid. This finding is likely not surprising to physicians (as much as to economists), in terms of understanding individual behavior as a driver of health care utilization.

The classic example of moral hazard for physicians stems from treatment recommendations and their cost. Related to this is potential profit motivation in the recommendations. With the decline of physician –owned practices, the aspirational and eventual transition to outcomes driven, value-based care, the moral hazard around cost-shifting may become less of a factor. Though profit driven medicine continues to be a significant problem, particularly in privately owned procedural based specialties which are reimbursed at a much higher rate that non-procedural care. Motivation from profit is likely behind volume based (doing more) as opposed to value based (doing better) care. Well documented increases in spinal fusions from 1990 through 2010 is an example of overutilization. However, physicians do not always follow the expectations of a moral hazard model for reasons such as patient preference, clinician experience, or treatment availability. This begs the question: how do we resolve these patterns of over or underutilization if patients or physicians are not only driven by economic forces?

Assessing Moral Hazard

Assessment of moral hazard in medicine acknowledges the conceptual limitations of this economic theory relative to the behavioral factors driving patients and physicians, and resolves these issues by using “behavioral hazard” models. At the simplest level, even if humans were motivated only by financial considerations, people can make mistakes. However, divergences from expectations of the maximum financial benefit can be observed to have both positive and negative effects on outcomes, and can be explained by the fact that both patients and physicians proceed with health care utilization for a variety of logical and nonlogical reasons. Patients may be motivated by disinterest in following treatment recommendations, for example not wanting to take medications, having challenges with modifying addictive behaviors, distrust/mistrust of health care entities, and/or by overly or underly misperceiving the severity of their afflictions. Despite preventive care having no out-of-pocket cost under the Affordable Care Act, there are still many instances where such care is avoided, such as colonoscopies. Physicians similarly are impacted by various considerations, whether lacking awareness of contemporary “best” practices, overly relying on their own treatment patterns, seeking to limit costs for their patients, lacking transparency in health system or patient costs, and/or inadvertently weighting some clinical signs/symptoms as more/less specific than others. This effect was well-demonstrated in a 2022 study by Mullainathan and Obermeyer, where both overutilization and underutilization of acute coronary syndrome (ACS) in an ED setting were evaluated using machine learning, a form of artificial intelligence (AI) algorithm that considered a larger number of variables than a physician could.

Variables being evaluated and the goals being sought may be misaligned.

ACS was well suited to study in this regard given the potential for severe outcomes if unrecognized, and as a whole, found overtesting was an issue. However, in considering the differences between high-risk and low-risk patients, a different picture emerged where undertesting in high-risk patients and overtesting in low-risk patients were divergent problems, with significant variability found in considering the ED shift where the patient presented. Such findings suggest that broad stroke averaging analyses of populations without considerations of more relevant clinical details may be overly sensitive for overutilization without appropriate specificity for underutilization.

Multiple Variables

At the heart of considering overutilization versus underutilization lies the fact that there are a multitude of variables that may be impossible for a physician to consider in real-time. Physicians are often left with the overly simple calculus of pursuing too much testing at the expense of unnecessary cost, or too little testing with the potential for significant morbidity/mortality. One of the most famous cases of overutilization is easily observed – and has been well studied – in the management of acute low back pain. Despite a limited set of instances where lumbar magnetic resonance imaging (MRI) is indicated, such as profound lower extremity weakness, fecal incontinence, or saddle anesthesia, this test is one of the (if not the most) overutilized tests in all of medicine in cases where such significant symptoms are not present. However, system pressures to reduce such tests over time may cause some physicians to fall into the opposite trap of not ordering an MRI when one is indicated. Even when considering this issue from these perspectives, there are still the multifaceted layers of bias, including a patient’s degree of perceived pain, a physician’s interest in reassuring a patient, a patient’s biopsychosocial background, and/or the physician attempting to be protected from malpractice claims. This relatively simple example (in comparison to ACS) demonstrates the perfect storm of factors that create distinct variations in how physicians provide different care to different patients. Thus, there continue to be opportunities for physicians to better understand the characteristics of these problems in various contexts.

The primary problem that these many examples orbit is one of information, where both the variables being evaluated and the goals being sought may be misaligned. In the present health care environment, where so much of utilization is driven by costs, whether to a patient or an insurer, there continues to be a significant focus on limiting or minimizing overutilization. The major risk of an approach driven only by consideration of financial incentives or moral hazard is that it fails to consider the human motivations in decision-making. Whether attributable to preferences, experience, or errors, this may be contributing to increasing underutilization in health care even when such utilization is appropriate. The immensely growing amount of available information via electronic health records (EHR) and medical literature is undigestible by human beings, even expert sources like physicians, especially within the rapid pace of on-the-ground US health care. Fortunately, modern technological solutions exist to aid the approaches to diagnosing and treating utilization mismatches in health care, particularly informatics-based solutions leveraging the ways in which computers can perform better than people.

The Role of Informatics

The field of informatics (a contraction of “information science”) includes how data is collected, stored, analyzed and translated into actions, where experts like physicians can use better information management to make better decisions. Computers differ from humans in that computers can perform massive amounts of calculations extremely quickly compared to humans (i.e., computers have greater processing capabilities), whereas humans can not only absorb and retrieve information extremely quickly, but also can draw connections between connected and seemingly-unconnected information in ways computers cannot. The aforementioned study on ACS using AI modeling to include many variables not considered by classic ACS diagnostic algorithms is a good example of when computers have an advantage over people. The same ACS study appropriately acknowledged that physician actions evaluated in that study likely included variables not represented in the EHR, such as a bruise explaining a patient’s chest pain, but nonetheless identified a divergence from expected clinical thresholds for high-risk testing. That study demonstrates a pertinent example of the so-called “fundamental theorem of informatics,” where the combination of human and computer expertise exceeds the expertise of a human alone, as well as a computer alone. Using the ACS study as a use case of informatics interventions, several possible solutions emerge to better identify and monitor divergences in care, including recurrent evaluation of clinical ACS algorithms in terms of reviewing physician adherence to such algorithms (physician variability), instances where clinical algorithms incorrectly identify high versus low risk (algorithm sensitivity/specificity), and patient selection in test orders (patient variability). Following such identification, classic informatics interventions such as the use of standardized ordersets, based on diagnostic algorithm output, or alerts, such as triggered by “chest pain” denoted in the EHR, can be aided by more modern applications of AI. The advantage of AI applications of this knowledge is the ability to include much more information than may be considered by humans, constructing probabilistic models of risk by combining existing diagnostic algorithms with computer-generated multivariable models, and continuously monitoring EHR data to provide updated degrees of risk as additional information is generated. However, there remain system-level issues that must be addressed by all stakeholders in order for such efforts to succeed.

System Level Change

First and foremost, physicians and patients must become greater stakeholders in driving how health care is delivered. Dissatisfaction exists for both physicians and patients in terms of the amount of time allocated to the provision of patient care. It may be an unreasonable expectation for physicians to become better informed and have better outcomes in the absence of more time to leverage the overwhelming amount of information being generated in health care settings. The other major facet of this problem is the ongoing focus on cost as the primary outcome driving system level decision-making, which often misses the individual and social determinants leading human behavior. Lastly, the fragmented nature of health care in the U.S. makes capitalizing on all of these various data sources difficult if not impossible, and is exacerbated by the value of data as a commodity, disincentivizing the sharing of data by various sources. Only when systems maximize using the expertise of all involved entities, not simply those holding the purse strings, will the optimal benefits of growing knowledge be realized. In such a world, the best outcome is neither overutilization nor underutilization, but rather providing the right care to the right people, and deriving appropriate financial thresholds or models from those outcomes rather than the current state of doing the reverse.

Zeke McKinney, MD, MHI, MPH, is  the program director of the HealthPartners occupational and environmental medicine residency and an affiliate assistant professor with the University of Minnesota School of Public Health in the division of environmental health sciences.