Mitigating Racial And Ethnic Bias And Advancing Health Equity In Clinical Algorithms: A Scoping Review.

In August 2022 the Department of Health and Human Services (HHS) issued a notice of proposed rulemaking prohibiting covered entities, which include health care providers and health plans, from discriminating against individuals when using clinical algorithms in decision making. However, HHS did not provide specific guidelines on how covered entities should prevent discrimination. We conducted a scoping review of literature published during the period 2011-22 to identify health care applications, frameworks, reviews and perspectives, and assessment tools that identify and mitigate bias in clinical algorithms, with a specific focus on racial and ethnic bias. Our scoping review encompassed 109 articles comprising 45 empirical health care applications that included tools tested in health care settings, 16 frameworks, and 48 reviews and perspectives. We identified a wide range of technical, operational, and systemwide bias mitigation strategies for clinical algorithms, but there was no consensus in the literature on a single best practice that covered entities could employ to meet the HHS requirements. Future research should identify optimal bias mitigation methods for various scenarios, depending on factors such as patient population, clinical setting, algorithm design, and types of bias to be addressed.

Considerations for addressing bias in artificial intelligence for health equity.

Health equity is a primary goal of healthcare stakeholders: patients and their advocacy groups, clinicians, other providers and their professional societies, bioethicists, payors and value based care organizations, regulatory agencies, legislators, and creators of artificial intelligence/machine learning (AI/ML)-enabled medical devices. Lack of equitable access to diagnosis and treatment may be improved through new digital health technologies, especially AI/ML, but these may also exacerbate disparities, depending on how bias is addressed. We propose an expanded Total Product Lifecycle (TPLC) framework for healthcare AI/ML, describing the sources and impacts of undesirable bias in AI/ML systems in each phase, how these can be analyzed using appropriate metrics, and how they can be potentially mitigated. The goal of these "Considerations" is to educate stakeholders on how potential AI/ML bias may impact healthcare outcomes and how to identify and mitigate inequities; to initiate a discussion between stakeholders on these issues, in order to ensure health equity along the expanded AI/ML TPLC framework, and ultimately, better health outcomes for all.

Implications of predicting race variables from medical images.

AI-predicted race variables pose risks and opportunities for studying health disparities.

Receipt of Medications for Chronic Disease During the First 2 Years of the COVID-19 Pandemic Among Enrollees in Fee-for-Service Medicare.

Importance: During the first 2 years of the COVID-19 pandemic, inpatient and ambulatory care declined dramatically. Little is known about prescription drug receipt during this period, particularly for populations with chronic illness and with high risk of adverse COVID-19 outcomes and decreased access to care. Objective: To investigate whether receipt of medications was maintained during the first 2 years of the COVID-19 pandemic among older people with chronic diseases, particularly Asian, Black, and Hispanic populations and people with dementia, who faced pandemic-related care disruptions. Design, Setting, and Participants: This cohort study used a 100% sample of US Medicare fee-for-service administrative data from 2019 to 2021 for community-dwelling beneficiaries aged 65 years or older. Population-based prescription fill rates were compared for 2020 and 2021 vs 2019. Data were analyzed from July 2022 to March 2023. Exposure: The COVID-19 pandemic. Main Outcomes and Measures: Age- and sex-adjusted monthly prescription fill rates were calculated for 5 groups of medications commonly prescribed for chronic disease : angiotensin-converting enzyme inhibitors and angiotensin receptor blockers, 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase inhibitors (statins), oral diabetes medications, asthma and chronic obstructive pulmonary disease medications, and antidepressants. Measurements were stratified by race and ethnicity group and dementia diagnosis. Secondary analyses measured changes in the proportion of prescriptions dispensed as a 90-day or greater supply. Results: Overall, the mean monthly cohort included 18 113 000 beneficiaries (mean [SD] age, 74.5 [7.4] years; 10 520 000 females [58.1%]; 587 000 Asian [3.2%], 1 069 000 Black [5.9%], 905 000 Hispanic [5.0%], and 14 929 000 White [82.4%]); 1 970 000 individuals (10.9%) were diagnosed with dementia. Across 5 drug classifications, mean fill rates increased by 2.07% (95% CI, 2.01% to 2.12%) in 2020 and decreased by 2.61% (95% CI, -2.67% to -2.56%) in 2021 compared with 2019. Fill rates decreased by less than the mean overall decrease for Black enrollees (-1.42%; 95% CI, -1.64% to -1.20%) and Asian enrollees (-1.05%; 95% CI, -1.36% to -0.77%) and people diagnosed with dementia (-0.38%; 95% CI, -0.54% to -0.23%). The proportion of fills dispensed as 90-day or greater supplies increased during the pandemic for all groups, with an increase per 100 fills of 3.98 fills (95% CI, 3.94 to 4.03 fills) overall. Conclusions and Relevance: This study found that, in contrast to in-person health services, receipt of medications for chronic conditions was relatively stable in the first 2 years of the COVID-19 pandemic overall, across racial and ethnic groups, and for community-dwelling patients with dementia. This finding of stability may hold lessons for other outpatient services during the next pandemic.

The Potential For Bias In Machine Learning And Opportunities For Health Insurers To Address It.

As the use of machine learning algorithms in health care continues to expand, there are growing concerns about equity, fairness, and bias in the ways in which machine learning models are developed and used in clinical and business decisions. We present a guide to the data ecosystem used by health insurers to highlight where bias can arise along machine learning pipelines. We suggest mechanisms for identifying and dealing with bias and discuss challenges and opportunities to increase fairness through analytics in the health insurance industry.

Pooled testing efficiency increases with test frequency.

Pooled testing increases efficiency by grouping individual samples and testing the combined sample, such that many individuals can be cleared with one negative test. This short paper demonstrates that pooled testing is particularly advantageous in the setting of pandemics, given repeated testing, rapid spread, and uncertain risk. Repeated testing mechanically lowers the infection probability at the time of the next test by removing positives from the population. This effect alone means that increasing frequency by x times only increases expected tests by around [Formula: see text] However, this calculation omits a further benefit of frequent testing: Removing infections from the population lowers intragroup transmission, which lowers infection probability and generates further efficiency. For this reason, increasing testing frequency can paradoxically reduce total testing cost. Our calculations are based on the assumption that infection rates are known, but predicting these rates is challenging in a fast-moving pandemic. However, given that frequent testing naturally suppresses the mean and variance of infection rates, we show that our results are very robust to uncertainty and misprediction. Finally, we note that efficiency further increases given natural sampling pools (e.g., workplaces, classrooms) that induce correlated risk via local transmission. We conclude that frequent pooled testing using natural groupings is a cost-effective way to provide consistent testing of a population to suppress infection risk in a pandemic.

An algorithmic approach to reducing unexplained pain disparities in underserved populations.

Underserved populations experience higher levels of pain. These disparities persist even after controlling for the objective severity of diseases like osteoarthritis, as graded by human physicians using medical images, raising the possibility that underserved patients' pain stems from factors external to the knee, such as stress. Here we use a deep learning approach to measure the severity of osteoarthritis, by using knee X-rays to predict patients' experienced pain. We show that this approach dramatically reduces unexplained racial disparities in pain. Relative to standard measures of severity graded by radiologists, which accounted for only 9% (95% confidence interval (CI), 3-16%) of racial disparities in pain, algorithmic predictions accounted for 43% of disparities, or 4.7× more (95% CI, 3.2-11.8×), with similar results for lower-income and less-educated patients. This suggests that much of underserved patients' pain stems from factors within the knee not reflected in standard radiographic measures of severity. We show that the algorithm's ability to reduce unexplained disparities is rooted in the racial and socioeconomic diversity of the training set. Because algorithmic severity measures better capture underserved patients' pain, and severity measures influence treatment decisions, algorithmic predictions could potentially redress disparities in access to treatments like arthroplasty.

Variation in common laboratory test results caused by ambient temperature.

BACKGROUND: Laboratory tests measure important aspects of physiology, but their results also vary for idiosyncratic reasons. We explore an underappreciated source of variation: ambient temperature on the day blood is drawn. METHODS: In a sample of 4,877,039 individuals between 2009-2015, we model 215,234,179 test results as a function of temperature, controlling for individual and city-week fixed effects. This measures how day-to-day temperature fluctuations affect results over and above the individual's mean values, and seasonal variation. FINDINGS: 51 of 75 assays are significantly affected by temperature, including measures of kidney function (increased creatinine, urea nitrogen, and urine specific gravity), cellular blood components (decreased neutrophils, erythrocytes, and platelets), and lipids (increased high-density lipoprotein [HDL] and decreased total cholesterol, triglycerides, and low-density lipoprotein [LDL]). These small, day-to-day fluctuations are unlikely to correlate with long-term physiological trends; for example, lipid panels checked on cooler days look lower risk, but these short-term changes probably do not reflect stable changes in cardiovascular risk. Nonetheless, doctors appear to treat these individuals differently. We observe 9.7% fewer statin prescriptions for individuals checked on the coolest versus the warmest days (-0.42% versus baseline of 4.34%, p < 0.001). CONCLUSIONS: Ambient temperature affects the results of many laboratory tests. These distortions, in turn, affect medical decision-making. Statistical adjustment in reporting is feasible and could limit undesired temperature-driven variability. FUNDING: None.

Association of Clinical Characteristics With Variation in Emergency Physician Preferences for Patients.

Importance: Much of the wide variation in health care has been associated with practice variation among physicians. Physicians choosing to see patients with more (or fewer) care needs could also produce variations in care observed across physicians. Objective: To quantify emergency physician preferences by measuring nonrandom variations in patients they choose to see. Design, Setting, and Participants: This cross-sectional study used a large, detailed clinical data set from an electronic health record system of a single academic hospital. The data set included all emergency department (ED) encounters of adult patients from January 1, 2010, to May 31, 2015, as well as ED visits information. Data were analyzed from September 1, 2018, to March 31, 2019. Exposure: Patient assignment to a particular emergency physician. Main Outcomes and Measures: Variation in patient characteristics (age, sex, acuity [Emergency Severity Index score], and comorbidities) seen by emergency physicians before patient selection, adjusted for temporal factors (seasonal, weekly, and hourly variation in patient mix). Results: This study analyzed 294 915 visits to the ED seen by 62 attending physicians. Of the 294 915 patients seen, the mean (SD) age was 48.6 (19.8) years and 176 690 patients (59.9%) were women. Many patient characteristics, such as age (F = 2.2; P < .001), comorbidities (F = 1.7; P < .001), and acuity (F = 4.7; P < .001), varied statistically significantly. Compared with the lowest-quintile physicians for each respective characteristic, the highest-quintile physicians saw patients who were older (mean age, 47.9 [95% CI, 47.8-48.1] vs 49.7 [95% CI, 49.5-49.9] years, respectively; difference, +1.8 years; 95% CI, 1.5-2.0 years) and sicker (mean comorbidity score: 0.4 [95% CI, 0.3-0.5] vs 1.8 [95% CI, 1.7-1.8], respectively; difference, +1.3; 95% CI, 1.2-1.4). These differences were absent or highly attenuated during overnight shifts, when only 1 physician was on duty and there was limited room for patient selection. Compared with earlier in the shift, the same physician later in the shift saw patients who were younger (mean age, 49.7 [95% CI, 49.4-49.7] vs 44.6 [95 % CI, 44.3-44.9] years, respectively; difference, -5.1 years; 95% CI, 4.8-5.5) and less sick (mean comorbidity score: 0.7 [95% CI, 0.7-0.8] vs 1.1 [95% CI, 1.1-1.1], respectively; difference, -0.4; 95% CI, 0.4-0.4). Accounting for preference variation resulted in substantial reordering of physician ranking by care intensity, as measured by ED charges, with 48 of 62 physicians (77%) being reclassified into a different quintile and 9 of 12 physicians (75%) in the highest care intensity quintile moving into a lower quintile. A regression model demonstrated that 22% of reported ED charges were associated with physician preference. Conclusions and Relevance: This study found preference variation across physicians and within physicians during the course of a shift. These findings suggest that current efforts to reduce practice variation may not affect the variation associated with physician preferences, which reflect underlying differences in patient needs and not physician practice.

Characteristics and determinants of high-risk unscheduled return visits to the emergency department.

BACKGROUND: High-risk unscheduled return visits (HRURVs), defined as return visits within 72 hours that require admission or die in the emergency department (ED) on representation, are a key quality metric in the ED. The objective of this study was to determine the incidence and describe the characteristics and predictors of HRURVs to the ED. METHODS: Case-control study, conducted between 1 November 2014 and 31 October 2015. Cases included all HRURVs over the age of 18 that presented to the ED. Controls were selected from patients who were discharged from the ED during the study period and did not return in the next 72 hours. Controls were matched to cases based on gender, age (±5 years) and date of presentation. RESULTS: Out of 38 886 ED visits during the study period, 271 are HRURVs, giving an incidence of HRURV of 0.70% (95% CI 0.62% to 0.78%). Our final analysis includes 270 HRURV cases and 270 controls, with an in-ED mortality rate of 0.7%, intensive care unit admission of 11.1% and need for surgical intervention of 22.2%. After adjusting for other factors, HRURV cases are more likely to be discharged with a diagnosis related to digestive system or infectious disease (OR 1.64, 95% CI 1.02 to 2.65 and OR 2.81, 95% CI 1.05 to 7.51, respectively). Furthermore, presentation to the ED during off-hours is a significant predictor of HRURV (OR 1.64, 95% CI 1.11 to 2.43) as is the presence of a handover during the patient visit (OR 1.68, 95% CI 1.02 to 2.75). CONCLUSION: HRURV is an important key quality outcome metric that reflects a subgroup of ED patients with specific characteristics and predictors. Efforts to reduce this HRURV rate should focus on interventions targeting patients discharged with digestive system, kidney and urinary tract and infectious diseases diagnosis as well as exploring the role of handover tools in reducing HRURVs.

Putting decisions under the microscope.

Ziad Obermeyer is an acting associate professor at the University of California, Berkeley, and an emergency physician. His teaching and research focus on how algorithms can aid in human decision-making in health care. Previously, he taught at Harvard Medical School, where he received the Early Independence Award, the most prestigious award for early-career scientists given by the US National Institutes of Health.