Sorites paradox and persistence in overuse and underuse in healthcare delivery services

Healthcare around the world suffers from underuse (undertreatment and undertesting) and overuse (overtreatment and overtesting) that have persisted for decades. Studies have identified many drivers of suboptimal medical care, generally classified into three categories: (a) economic incentives affecting clinicians' behaviour; (b) professional knowledge, bias, the uncertainty inherent in clinical decision-making and (c) failure to include patients to capitalise on the power of human relationships.1 These drivers, however, only partially explain the persistence of suboptimal care. In this paper, we argue that the overuse and underuse of medical services are an inherent consequence of the nature of the relationship between scientific evidence and decision-making. Evidence about diagnosis or health outcomes exists on the probability continuum (from impossibility to virtual certainty), while decisions are categorical (e.g., we recommend treatment, or we don't).2 For example, risk for heart disease can range from close to zero (young, healthy people) to 100% when is clinically manifested as myocardial infarction. To help people reduce risk of heart disease, the American College of Cardiology and American Heart Association (ACC/AHA) recommends that people should be treated with statins if their risk for heart disease ≥7.5% over 10 years. American Society of Oncology recommends treatment with colony-stimulating factors (CSF) if the patients undergoing chemotherapy have a risk of developing febrile neutropenia ≥20%. ACC/AHA also recommends treatment for hypertension, defined as blood pressure at or above 130/80 mmHg. In these cases, correct recommendation according to these guidelines is to prescribe statins when risk for heart disease is ≥7.5%, administer CSF when risk of febrile neutropenia ≥20%, and antihypertensives when blood pressure exceeds 130/80 mmHg. If the estimated risk is below these thresholds, we should not recommend treatment. Providing treatment under such circumstances would reflect overuse/overtreatment. Conversely, not recommending treatment would constitute underuse/undertreatment. Naturally, in cases like these, we can ask why not administer statins at 7.4% (or, 7.3%, 7.2% … 0.01%), CSF at 19% (18.9%, 18.8% … 1%) and antihypertensive at 129/79, 128/78 … 120/70 mmHg? At which values, correct recommendations would become overuse/overtreatment or underuse/undertreatment? The problem we just described is related to an ancient epistemological puzzle known as the Sorites paradox,3 also known as ‘little-by-little arguments’. Sorites in the Greek language means heap; paradox reflects challenges in defining clear boundaries between borderline cases of the quantities of interest. At which point does the collection of grains becomes large enough to be called a heap and small enough to be classified as a few scattered grains of the sand? The Sorites paradox abounds in medicine. Because clinicians use scientific evidence (that exists on a continuum) but make categorical (yes/no) decisions based on numerical thresholds,4 Sorites paradox is unavoidable in medical practice. If most decisions are based on thresholds, one approach to a solution to the Sorites paradox is to apply the threshold decision models2, 5,* (Supporting Information Appendix). At the decision thresholds, we are indifferent in committing to one or another action course. This occurs when the overall benefits of health interventions are equal to its harms at a particular probability of a disease or health outcome coherent with the patient's values and preferences (V&P).2, 5 Thresholds of equal benefits and harms demarcate correct from incorrect management actions. Then, if the probability of disease or health outcome6 is above the threshold, we should recommend treatment. Not recommending treatment, in this case, would constitute underuse/undertreatment. Conversely, if the probability of disease or health outcome is below the threshold, we should not recommend treatment. Providing treatment under such conditions reflects overuse/overtreatment. But perfect demarcation is impossible. There will always be some people we should have treated, but did not, and vice versa. According to the threshold model, the higher the benefit/harm ratio, the lower probability of diagnostic or predictive certainty at which we should act. Acting at lower levels of clinical certainty means the unabated use of diagnostic and treatment interventions. As the regulators typically approve safe and effective therapeutics with favourable benefit/harm ratios to be used in practice, this implies that overuse is built into the modern practice of medicine. There are other factors that affect the determination of the threshold including (1) the lack of consensus on a theoretical framework on how to integrate the benefits and harms of competing treatment alternatives to derive decision thresholds, (2) the lack of agreed-upon methods for elicitation of patients' V&P and (3) frequent incompatibility between individual and societal health goals. Because scientific inferences are considered theory-laden, the exact technical derivation of the threshold will depend on a given theoretical framework.2 As a result, what is considered overuse under one decision theory may be regarded as underuse under another. Table 1 illustrates how using four different models results in four different assessments of overuse versus underuse in the management of patients with pulmonary embolism. Because V&P is one of the critical determinants of thresholds, the lack of agreed-upon methods to elicit patients' V&P accurately makes the consistent derivation of decision thresholds extremely challenging. An additional problem is that most bedside decisions cannot be evaluated outside the context in which the decisions are made. This also means that whether a bedside clinical decision reflects the right care, overuse, or underuse is impossible to define prospectively and can only be judged after we have ascertained the patient's V&P. In theory, measuring under and overuse can be conducted by comparing (not) recommended treatment against evidence-based practice guidelines augmented with decision models. However, such an approach still requires specifying a theoretical framework and method for eliciting V&P. Finally, technical aspects of decision-making under uncertainty cannot be divorced from consideration of the ethics of our actions. The health goals and the risk assessment of individual clinicians and their patients and the goals of public-health officials and society are often in tension. This has been vividly displayed during the COVID-19 pandemic when many patients declined to receive the COVID-19 vaccine despite public officials' widespread endorsement. From individual perspectives, the decision threshold at which public health officials' have promoted vaccination against COVID-19 was too low, amounting to overuse. On the other hand, from public health agencies' perspective, failure to vaccinate represents underuse, resulting in preventable deaths. Because achieving ethical triple aim7—alignment between utilitarian (society-oriented), duty-bound (individual-oriented), and right-based (autonomy, ‘no decision about me, without me’) ethics—is impossible, different people will inevitably be affected differently by our decisions. That is, the same course of action may be perceived as undertreatment from one perspective and overtreatment from another resulting in unavoidable (real or perceived) injustice as we navigate optimal decision trade-offs under uncertainty. We conclude that having insights into how relationships between evidence and decision-making shape the correctness of our decisions may help us improve suboptimal quality of care in today's practice.4 We argue that the first step to address underuse and overuse that plague contemporary practice is to become aware of its underlying philosophical underpinning rooted in the Sorites paradox. Such awareness will help improve the formulations of transparent and explicit frameworks necessary to link evidence and decision-making, eventually leading to better patient care. The authors declare no conflict of interest. Data sharing not applicable to this article as no datasets were generated or analysed during the current study. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.