Riskology

Risk-management is a foundational competency of not only the emergency physician, but the department (ED) as a whole (future essay). Utilizing a suite of tools and processes, the department aims to identify and stratify – often surreptitious – risk in an environment that is time, attention, and informationally constrained. It is tasked to rule-in/rule-out high-morbidity diseases and safely differentiate between the sick, the potentially sick, and the worried well. Two emergency department specific tools that are explicitly designed for risk-management are triage screens and risk-stratification scores. Risk-assessment begins at the point of contact between the patient and the ED – at triage. In addition to the Emergency Severity Index (ESI), EDs utilize screening algorithms to identify and anticipate high-risk and time-sensitive diseases. These screens are top-of-the-funnel identification algorithms to identify patients with high-risk diseases such as sepsis and stroke. On the other hand, risk-stratification scores are designed to integrate patient generated data such as the history, physical exam, and biomarker results to quantify and categorize risk. Both these tools are key components of ED risk-management toolkit. They influence patient trajectories, enable next-best actions, and anchor medical decisions.

When available, risk-stratification scores are pushed top-down on emergency physicians, and thus strongly influence physician decision making. The perceived value of these tools comes from the fact that they integrate data sources and are designed to explicitly quantify and categorize risk. For example, the HEART score uses the history, historical data with EKG and biomarker testing to risk-stratify patients into low-moderate-high categories for predicted major adverse cardiac events (MACE). The pulmonary embolism rule-out criteria (PERC) uses historical data and vital signs to screen-out patients for a pulmonary embolism without additional testing. The pediatric appendicitis score uses the history, physical-exam findings, and biomarker testing to identify pediatric patients who can be safely ruled-out for appendicitis without the ionizing radiation risks of diagnostic imaging. In general, these tools have utility because they are perceived to offer prescriptive and predictive power. They reduce ambiguity and enable decision-making to be grounded in quantified probabilities.

Despite the obvious value of these tools, they suffer from the tyrannical constraints of an overwhelming false-positive (FP) bias that exists in healthcare. Operating within such a framework, all prescriptive and predictive tools are either explicitly designed or become stretched and degraded to fit the singular goal of false-negative (FN) minimization. For example, utilizing the HEART score, any patient that is sixty-five or older with non-specific repolarization changes, with ubiquitous risk factors such as hypertension, diabetes, and atherosclerotic disease would be considered moderate risk with a 12 – 16.6% risk of MACE. All such patients require either serial biomarker testing or admission into the hospital. On the other side of the edger, there are no risk-stratification tools to adequately rule-out low-risk patients or rule-in either atypical presentations or outlier – conventionally designated as without risk factors – patients. In the absence of these tools, all patients with the chief-complaint of “chest pain” are screened with the HEART score requiring the excessive utilization of biomarker testing. Similarly, every female patient on hormone based contraceptives or over the age of forty-nine requires biomarker testing to rule-out a pulmonary embolism. However, the next-best action – d-dimer – is itself a non-specific test and prone to false-positives, potentially leading to the downstream effects of false-positive for an individual and unnecessary resource utilization on a population and system level. In essence, risk-stratification tools are theoretically useful, however, they degrade to the status of ignored artifacts and post-hoc justifications at best, to wasteful and inappropriately utilized tools at worst.

The FP-FN asymmetry is further amplified in the risk-identification algorithms utilized in triage. Since treatment effects and outcomes of disease such as sepsis and stroke are strongly time-dependent, time-to-identification and time-to-treatment for these diseases have become nationally measured metrics. However, both of these diseases have significant knowledge-gaps and definitional ambiguity. Stroke presentations have significant heterogeneity, sepsis definitions covers a continuum of dysfunction from the mild to the severe. Evidence of effectiveness of interventions and testing vary, often with level of dysfunction. Nonetheless, these imposed metrics have spurred the proliferation of “cast-a-wide-net” sepsis and stroke screens. Any patient within the orbit of these diseases “rules-in,” triggering a cascade of testing and treatment. For example, a positive sepsis screen results in reflexive biomarker testing, intravenous fluid administration, and the use of broad-spectrum antibiotic. The biomarkers are non-specific, intravenous fluids are wasteful, and minimally discriminated antibiotic use insidious to the patient and the system. The downstream iatrogenic effects of these screens is hard to measure and remains mostly unmeasured (future essay).

This lack of consideration for false-positive associated iatrogenesis is further amplified when you consider the group of patients who preferentially screen-in. The riskiest patients – elderly patients, patients with chronic diseases, nursing home residents, homeless patients, and patients with history of polysubstance use – are also the most vulnerable to pernicious effects of false-positive screens. It is the elderly and multi-morbid nursing home patient with a diminished physiological reserve who is subjected to microbiome compromising or antibiotic resistance selecting broad-spectrum antibiotics. It is the cancer patient receiving debilitating chemotherapy who presents with vaguely described symptoms of “weakness” or “dizziness” who is subjected to malignancy causing ionizing radiation from CT imaging. It is the immunocompromised patient who screens-in and triggers a set of actions that ends in admission of the patient into an environment teeming with virulent microorganisms.  There are no screening or identification tools that exist that can safely rule-out the vulnerable. There are no intermediate actions to take. As currently designed, next-best actions are invasive, aggressive, and coarse-grained.

Risk-management – what I term riskology – is the foundational function of the emergency department. Starting at triage, identifying risk and risk-stratification are the goals of the ED. Risk-stratification scores combine subjective input – history and physical exam – with more objective data sources – biomarker testing – to quantify and categorize risk. Screening algorithms at triage, can identify high-morbidity conditions such as sepsis and stroke early in the disease trajectory. These tools identify, integrate, categorize, and quantify risk. The value proposition of these tools is undeniable. In fact, emergency physicians require more such tools for not only specific diseases but also a global risk score that dynamically updates with additional data (future essay). However, there remains a significant gap between the theoretical and realized value proposition of these tools. Constrained by the invisible and tyrannical fear of the false negative, the current iteration of tools lack sufficient granularity, cannot safely rule-out vulnerable subpopulations, and do not explicitly identify potential anomalies (future essay). They are unfit for function, generate affordances that skew towards the aggressive, indiscriminate, and inapplicable.