discussion 4

Amie1997

GervasiBiasinMachineLearning.pdf

Home >Applied Sciences homework help >discussion 4

By Stephanie S. Gervasi, Irene Y. Chen, Aaron Smith-McLallen, David Sontag, Ziad Obermeyer, Michael Vennera, and Ravi Chawla

ANALYSIS

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

ABSTRACT 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.

T he amount of data collected about health care in the United States is enormous1 and continues to grow rapidly. Machine learning has be- come embedded in the health in-

surance industry for tasks such as predicting early disease onset,2 determining the likelihood of future hospitalizations,3 and predictingwhich members will be medication noncompliant. Al- gorithms are often developed to optimize inter- ventions to drive improved health outcomes. As machine learning is increasingly used in

health care settings, there is growing concern that it can reflect and perpetuate past and pres- ent systemic inequities and biases. Researchers have begun to evaluate algorithms and their ef- fects on disadvantaged or marginalized popula- tions. In one notable study, algorithms used to identifypatients for a caremanagementprogram perpetuated racial disparities,4 further contrib- uting to racial inequities in health care use and disease outcomes.5–8 This research led to imme- diate calls for greater transparency and account- ability across the health care industry in how the use of algorithms is audited and how to avoid bias in predictive models.9

We examine issues of bias and fairness from thehealth care payer perspective, outlining com- mon sources of and potential solutions to bias in algorithms. These concerns are applicable to any

computational tools used by insurers, from line- ar models to neural networks, but we focus on machine learningmethods because of their com- plexity and opacity. We outline three use cases common among health insurers for identifying and stratifying members who may benefit from care management programs. We then address how entities in the health insurance ecosystem can identify and remediate bias in these cases and beyond. See the online appendix for a sum- mary of the health care data collected by the US health insurance industry, the main stages of machine learning pipelines where bias arises, common sources of bias in predictive health care models, and potential solutions.10

Common Uses Of Predictive Modeling By Insurers Health insurers use predictive modeling to iden- tify members with complex health needs for interventions and outreach, including care coor- dination and conditionmanagement. To identify and prioritize members for outreach, most health plans rely on some combination of risk scores from commercial vendors, outputs from one or more predictive models, and “if-then” type business rules. Because these risk-based prioritization strate-

gies drive the allocation of valuable health care

doi: 10.1377/hlthaff.2021.01287 HEALTH AFFAIRS 41, NO. 2 (2022): 212–218 This open access article is distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license.

Stephanie S. Gervasi, Independence Blue Cross, Philadelphia, Pennsylvania.

Irene Y. Chen (iychen@csail .mit.edu), Massachusetts Institute of Technology, Cambridge, Massachusetts.

Aaron Smith-McLallen, Independence Blue Cross.

David Sontag, Massachusetts Institute of Technology.

Ziad Obermeyer, University of California Berkeley, Berkeley, California.

Michael Vennera, Independence Blue Cross.

Ravi Chawla, Independence Blue Cross.

212 Health Affairs February 2022 41 :2

Health Information Technology

Downloaded from HealthAffairs.org on February 22, 2022. Copyright Project HOPE—The People-to-People Health Foundation, Inc.

resources, the underlying algorithmic processes should undergo regular audits to identify poten- tial biases.We describe how sources of bias relat- ed to problem selection, outcomedefinition, and data availability and reliability manifest across three models commonly used among health in- surers to prioritize care management.

Disease Onset Six in ten US adults have a chronic disease, and four in tenhave twoormore chronic diseases.11 Chronic diseases are signifi- cant causes of death, disability, and reduced quality of life and account for trillions of dollars in annual health care costs. Many chronic dis- easesmay be effectivelymanaged through smok- ing cessation, nutrition counseling, or medica- tion adherence programs. As a result, models predicting the onset of the most prevalent dis- eases, especially those tracked by the Centers for Medicare andMedicaid Services (CMS) for qual- ity performance assessments,12 are common among health insurers. When a predictive model is being developed, a

fundamental source of bias is the initial selection of the prediction problem. Models are less com- mon for diseases that tend to affect smaller or minority segments of the member population (such as sickle cell anemia) or that might not have well-defined or easily scalable interven- tions.Yet targeting such conditions could greatly impact morbidity, mortality, and health care costs for those with the condition. Another bias common in disease onsetmodels

is the availability of data required to identify a target outcome and generate features for predic- tions. Clinical indicators in claims and in elec- tronicmedical record (EMR)data aremore likely to bemissingor populated at lower frequency for members with less health care use. Moreover, the data reported on the claim reflect disparities in provider treatment and diagnosis stemming from implicit and explicit bias, including rac- ism.5 Further, data related to previous diagnoses and procedures, other medical history, or stage of disease may be missing differentially across groups, adversely affectingpredictions. Incorpo- rating data on the social determinants of health, including health care access; poverty; education level; employment; housing; exposure to haz- ards in living and occupational environments; and access to transportation, food, and health clinics, may improve the performance of disease onset models and reduce the reliance on utiliza- tion patterns alone for need-based optimization.

Likelihood Of HospitalizationAccording to CMS, hospitalizations represented the largest component of national health care expenditures in 2017 and 2018.13 While many acute inpatient events such asmaternity and trauma admissions are unavoidable, others are preventable through

effective primary and specialty care, disease management, availability of interventions at outpatient facilities, or all of the above. In 2017 the Agency for Healthcare Research and Quality (AHRQ) estimated that 3.5 million pre- ventable inpatient hospitalizations accounted for $33.7 billion in hospital costs.14

Machine learningmodels that predict the like- lihood of an avoidable inpatient hospitalization (known as likelihood of hospitalizationmodels) can help target interventions, prevent adverse health outcomes, and reduce individual andpop- ulationhealth care costs.15–18 However, observing an acute hospitalization event in the data is con- tingent on access to and use of health care ser- vices, both of which are influenced by racial and socioeconomic disparities.11,19 Disparities in ac- cess and use mean that some subpopulations are underrepresented in the target population and in the data used to predict the outcome of interest. Thus, the resulting model output may reflect those systemic biases, and interventions or policy decisions based on the model outputs risk reinforcing and exacerbating existing in- equities. Similar to disease onset models, one way to

address the data disparities in likelihood of hos- pitalizationmodels is through inclusion of addi- tional data sources that showpatterns inprimary or preventative care that can prevent unplanned hospitalization. EMRdata can add granularity to clinical events, capturing diagnostic and other health information that may not be recorded on claims. However, integrating EMR and claims data can introduce additional bias20 stemming frommissing or incomplete records for patients who experience barriers to consistent care. Im- portantly, missing clinical codes can indicate lack of key diagnostics, procedures, or primary care support along a patient’s health care jour- ney that might have precluded the need for inpa- tient hospitalization. Similar symptoms may be treated differently among providers, leading to downstream effects on hospitalization. Data on social determinants of health can also improve the performance and potentially interpretations of likelihood of hospitalization prediction tasks. Medication Adherence In 2003 the World

Health Organization noted that approximately 50 percent of patients with chronic illnesses do not takemedications as prescribed.21 In theUnit- ed States, lack of medication adherence can lead to morbidity and mortality and is estimated to cost $100 billion per year.22 CMS also considers medication adherence to be a critical component of Medicare health plan performance ratings, making predictive models for medication adher- ence common across the health insurance indus- try. Adherence is also associated with reduced

February 2022 41 :2 Health Affairs 213 Downloaded from HealthAffairs.org on February 22, 2022.

health services use and lower medical costs for many chronic conditions.23

Predictive models often help health insurers’ pharmacydepartments designmember outreach strategies to improve adherence. These models can be developed using regression or classifica- tion approaches. Regression-based approaches typically predict the proportion of days covered, defined as the proportion of days during a calen- dar year that a member has access to their med- ications, and classification approaches use a proportion of days covered of greater than 80 percent as a target threshold. Medication adherence can be influenced by

many factors, including dosing frequency, side effects, and routes of administration. However, differences in diagnosis, treatment, and pre- scribing are also well documented. Compared with White patients, members of racial and eth- nic minority groups are less likely to be pre- scribed opioids for chronic pain and less likely to receive evidence-based prescribing practices related to antidepressants, anticoagulants, dia- betes medications, drugs for dementia, and sta- tins.24–32Whenmedication adherencemodels are being designed, a different target definition of whether a member should have a prescription for a condition based on clinical care guidelines may be more appropriate. Using machine learning to identify patients at

risk for being noncompliant with a new medica- tion regimenor for falling below an optimal level of adherence over time canbe valuable for target- ing resources and programs. However, health plans and other entities that develop and use medication adherence models (such as pharma- cy benefit managers and health systems) must recognize how systemic biases in access to phar- macies and prescription drugs, prescribing pat- terns, and utilization in Black and Brown com- munities affect problem formulation, algorithm development and interpretation, and interven- tion strategies.33–36

Understanding why a member was predicted to be noncompliant is particularly relevant when medication adherence interventions are being selected and implemented. Collaborations be- tween interventionists and data scientists can ensure that relevant contextual information is used to refine the predictive model at hand. For example, instead of predicting medication ad- herence directly, data scientists can identify members most receptive to lower-cost medica- tion alternatives or nontraditional delivery methods, as these are likely to be patients strug- gling with financial or transportation barriers.

Auditing Machine Learning Pipelines For Bias Fortunately, there are several ways to check pre- dictive models and business processes for bias, and health insurers should establish standard but flexible protocols for auditing their models and processes. Here we outline several practical approaches, and we note that there is likely no “one-size-fits-all” solution. Representational Fairness One way to

check for bias is to examine rates of outreach and engagement in care management programs relative to the proportions of subgroups in the data. For example, an eligible populationmay be observed that is 40 percent White, 30 percent Black or African American, 20 percent Hispanic or Latino, and 10 percent Asian. If the propor- tions of those targeted for outreach and engaged in care management do not reflect the underly- ing population distribution, onemight conclude that there was an element of representational bias.37 Note, however, that this method does not report whether resources were appropriately allocated. That is, there may be reasons to dis- tribute resources equitably based on true care needs, with higher rates of engagement from some subpopulations than others, rather than equally. Counterfactual Reasoning Counterfactual

reasoning asks the question, If a given person was from a different subpopulation but with the samehealthprofile,would theyhave received the same predicted probability of an outcome? For caremanagement, the analogous question could be comparing care management program mem- bership for Black and White patients. Research- ers found that when patients were prioritized by risk scores—representing patient medical costs—from a predictive algorithm, only 17 per- cent of the patients eligible for a care manage- ment programwere Black.4 To simulate a correc- tion, researchers swapped sicker Black patients for less sick White patients at each level of risk until nomore swapswerepossible, with sickness measured by total number of chronic conditions. In this synthetic correction, 46 percent of the patients qualifying for the care management programwereBlack. By assessing counterfactual fairness,38 it is possible to examine how a model treats both race and other potentially unmea- sured confounding factors that may be correlat- ed to race. Error Rate Balance And Error Analysis

Error rate balance involves comparing false pos- itive and falsenegative rates forpredictionswith- in specified subpopulations.39 Analyses might compare the rates of false positives and false negatives by race, ethnicity, or gender. For ex- ample, a chi-square test can be used to compare

Health Information Technology

214 Health Affairs February 2022 41 :2 Downloaded from HealthAffairs.org on February 22, 2022.

the rates of false positives (and false negatives) by gender. A statistically significant result would indicate that the model does not predict equally well for both groups and therefore has some degree of bias vis-à-vis the error rate balance criterion. Error rate balance reports patterns that the

model is detecting and missing. It increases un- derstanding of why the model is making classifi- cation errors by examiningmembers and groups who are most likely to receive an incorrect pre- diction. For example, amodel predicting chronic disease occurrence may be less accurate for members with specific conditions, for members of certain races or ethnicities or who live in cer- tain geographies or see certain providers. Re- searchers can then investigate where the ma- chine learning pipeline can be improved and, in the context of a chronic disease occurrence prediction task,maydecide tooptimize to reduce false negative rates over false positive rates. Potential strategies are to adjust upsampling or downsampling rates in the training data or generate differentmodels for different subpopu- lations. In addition to data-based solutions, re- viewing errors with a diverse set of stakeholders who can provide context from lived experience about why specific types of errors are observed and what impact they have can reduce uninten- tional harm that could be caused when different types of errors are made. When bias is identified, it is important for

stakeholders to have transparent discussions about whether and how the biases are problem- atic, and the potential gaps in data or other as- pects of model development that could have led to the bias. Stakeholders should strategize about differentmodeling approaches that could reduce bias, including redefining the target outcome; experimenting with sampling methods, data augmentation, or restriction; and model class selection. In some instances, solutions may lead to models that have poorer fit but that may be

fairer, inwhich case stakeholders need to adhere to ethical principles in balancing model perfor- mance, business needs, and health equity.

Addressing Bias In Machine Learning As An Industry Health insurers share several challenges in as- sessing and reducing bias that could be ad- dressed collaboratively as an industry. While these themes are not exhaustive, we believe that they represent primary areas where the field of fair machine learning has the potential to make major advances in the comingmonths and years. Industry Vigilance Algorithmovigilance re-

fers to scientific methods and activities relating to the evaluation, monitoring, understanding, and prevention of adverse effects of algorithms in health care.40 Calls for the health care industry, including health insurers, to monitor and evalu- ate machine learning models for bias have been increasing from several sectors. In January 2021 Pennsylvania’s new Interagency Health Reform Council recommended that payers and providers review and revise their predictive analytics and algorithms to remove bias.41 The National Com- mittee for Quality Assurance (NCQA) and AHRQ also have taken an interest in the impact of health care algorithms on racial disparities in health and health care. For example, the NCQA is incorporating evaluation of racial bias into accreditation standards.42 In addition, legisla- tion introduced in the House and Senate in 2019—the Algorithmic Accountability Act— would have required certain commercial entities to conduct assessments of high-risk systems that involve personal information or make automat- ed decisions, such as machine learning. This attention to bias in health care algorithms has led to the development of and renewed attention to guidelines, best practices, and analytics tools related to the evaluation and use of algorithms in predictive analytics.43 These tools have the potential to inform and unify the entire payer space to combat bias and enable health insurers to more effectively provide high-quality, equita- ble care and services to members. Ultimately, these tools will require testing at scale and con- stant and rigorous evaluation to ensure that they are having the intended positive impacts on member populations and that models tuned for fairness do not undergo “bias drift” over time or during business implementation. Algorithmovigilance requires that machine

learning models be designed in ways that can be empirically examined.Health care companies should incorporate knownmethods for identify- ing and remediating algorithmic bias into their machine learning pipelines and participate in

Health insurers share several challenges in assessing and reducing bias that could be addressed collaboratively.

February 2022 41 :2 Health Affairs 215 Downloaded from HealthAffairs.org on February 22, 2022.

the ongoing development and dissemination of new methods. Regular assessment of whether models are generating insight and result in ac- tions that maximize the intended outcome, such as reducing acute hospitalizations in a popula- tion, should take place. Evaluations should not be limited to the model output but should also assess the impact of actions taken based onmod- el results and should examine whether impacts were differential across relevant subgroups. Models that areboth accurate and fairwill lead

to interventions and business practices that ulti- mately benefit members at the highest levels of risk and need and lead to better outcomes and lower costs. Obtaining And Ethically Using Race And

Ethnicity Data Data onmembers’ race and eth- nicity could enhance medical management pro- grams and facilitate audits for possible racial bias in both algorithmic output and care man- agement outreach.Yet most health plans do not collect race, ethnicity, or primary or preferred language data as part of the enrollment process or in any other systematic way. CMS has recently made race, ethnicity, and

language data available to health plans forMedi- care Advantage enrollees. For commercially insured members, individual-level data may be available in EMR data from provider health sys- tems, although not all health systems provide EMR data to payers. Health plans may also ob- tain these data from surveys, although surveys are usually administered to subsets of the mem- ber population. Third-party vendor data also contain information on race, ethnicity and lan- guage, but match rates with health plan mem- bership varies, as does the specificity of the data. Race imputation using statistical estimation techniques such as Bayesian Improved Surname Geocoding or Bayesian Improved First Name Surname Geocoding44,45 may also be embedded with bias. Data on race, ethnicity, and language can also be obtained at the census block or tract level through the American Community Survey, but these data sources don’t provide individual- level specificity and are limited to five single-race groups, which does not sufficiently capture het- erogeneity within a community. Many health plans are hesitant to collect and

use data on race, ethnicity, and language, even whenprovided voluntarily, because of the lack of established regulatory and oversight policies on how to ethically collect, aggregate, use, and re- port data on race and ethnicity. Establishing these policies at the federal or state level would provide guidance and protections, but this will likely take years to develop and implement. The health insurance industry should coalesce around ethical principles and standards for col-

lecting and using data on race, ethnicity, and language, aswell as onother social determinants of health. Entities such as America’s Health In- surance Plans or the NCQA could also establish standard practice protocols, which may include establishing a reviewboard or oversight commit- tee at each health plan that would govern the use of race and ethnicity data in analytics and reporting. Addressing Missing Data And Bad Proxies

Member health data are not collected unless a provider is seen, resulting in more missing data onpopulations thathaveobstacles toaccess care. Even when care is delivered, disparities in treat- ment and diagnosis contribute to incomplete and even incorrect data.5,6 Sometimes, proxies for a particular target variable or for individual features are used, but they also can be flawed and exacerbate bias.4 For example,member raceused as a feature in amodel for conditiononset should not be used to make claims about underlying genetic differences. Race is a proxy for systemic racism and should be considered interactively with other data including social determinants of health. As another example, health care costs are not an optimal or complete representation of condition complexity. To facilitate fair machine learning, better

methodologies for evaluating and addressing data missingness, sparsity, and irregularities are needed. For example, computers can gener- ate realistic health care data to rebalance data sets, but the synthetic datamay in fact perpetuate existing biases.46 Health-related behaviors for high-riskmemberswhounderuse care aredriven by a multitude of social determinants of health and other environmental factors not captured in data commonly available to health plans. The next generation of machine learning and artificial intelligence in the health insurance in- dustry needs to explicitly consider how to incor-

Opportunities exist to ensure that machine learning is fair, not only on ethical grounds but also on strong operational and business grounds.

Health Information Technology

216 Health Affairs February 2022 41 :2 Downloaded from HealthAffairs.org on February 22, 2022.

porate outside sources of data from social media platforms, wearable devices, crowdsourcing, and other types of small- and large-scale commu- nity-level resources. Cross-plan collaborations could also lead to robust insights—for example, across members insured through Medicare, Medicaid, and commercial plans across the US.

Including All Relevant Voices Machine learning in health care is developed in response to a business or clinical question. Fairness in machine learning is facilitated by collaborative conversations between machine learning scien- tists and clinical experts, supplemented bymem- ber voices, and guided by the expertise of equity experts. Diverse data science teams, including practitioners with lived experience—especially thosewho are disproportionately affected by sys- temic inequities in the health care system—must be intentionally created. Collaboration within and across such teams can reveal blind spots and impediments47 in efforts to promote health equity through predictive analytics.

Conclusion The responsibility for building and implement- ing equitable machine learning models lies with the broader health insurance community. Con- tinuedmachine learning development is inevita- ble. Opportunities exist to ensure that machine learning is fair, not only on ethical grounds but also on strong operational and business grounds. With recent calls for active vigilance of machine learning and its implementations, institutional and industry commitments to in- crease equity in health care are needed. This includes developing and disseminating best practices in bias detection and remediation as well as the development of targeted programs to reduce bias and promote equity, and deeper involvement and communication with the mem- bers and communities served by health plans. With these combined efforts, more equitable health care can be achieved. ▪

Stephanie S. Gervasi and Irene Y. Chen are co–first authors of this work. The authors are grateful to Alya Nadji and two anonymous reviewers for feedback that greatly improved their manuscript.

This is an open access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt, and

build upon this work, for commercial use, provided the original work is properly cited. See https://creative commons.org/licenses/by/4.0/.

NOTES

1 Raghupathi W, Raghupathi V. Big data analytics in healthcare: promise and potential. Health Inf Sci Syst. 2014;2:3.

2 Razavian N, Blecker S, Schmidt AM, Smith-McLallen A, Nigam S, Sontag D. Population-level prediction of type 2 diabetes from claims data and analysis of risk factors. Big Data. 2015;3(4):277–87.

3 Kansagara D, Englander H, Salanitro A, Kagen D, Theobald C, Freeman M, et al. Risk prediction models for hospital readmission: a systematic review. JAMA. 2011; 306(15):1688–98.

4 Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366(6464):447–53.

5 Fiscella K, Franks P, Gold MR, Clancy CM. Inequality in quality: addressing socioeconomic, racial, and ethnic disparities in health care. JAMA. 2000;283(19):2579–84.

6 Cleveland Manchanda E, Couillard C, Sivashanker K. Inequity in crisis standards of care. N Engl J Med. 2020;383(4):e16.

7 Ashana DC, Anesi GL, Liu VX, Escobar GJ, Chesley C, Eneanya ND, et al. Equitably allocating resources during crises: racial differences in mortality prediction models. Am J Respir Crit Care Med. 2021;204(2):

178–86. 8 Oxman D. The crisis in crisis

standards of care. Ann Am Thorac Soc. 2021 Feb 5(ja).

9 Ron Wyden [Internet]. Washington (DC): Office of Sen. Ron Wyden. Press release, Wyden, Booker de- mand answers on biased health care algorithms; 2019 Dec 3 [cited 2022 Jan 11]. Available from: https://www .wyden.senate.gov/news/press- releases/wyden-booker-demand- answers-on-biased-health-care- algorithms

10 To access the appendix, click on the Details tab of the article online.

11 Rodriguez-Gutierrez R, Herrin J, Lipska KJ, Montori VM, Shah ND, McCoy RG. Racial and ethnic differ- ences in 30-day hospital readmis- sions among US adults with diabe- tes. JAMA Netw Open. 2019;2(10): e1913249.

12 Centers for Medicare and Medicaid Services. Healthcare Effectiveness Data and Information Set (HEDIS) [Internet]. Baltimore (MD): CMS; [last modified 2021 Dec 1; cited 2022 Jan 12]. Available from: https:// www.cms.gov/Medicare/Health- Plans/SpecialNeedsPlans/SNP- HEDIS

13 Centers for Medicare and Medicaid Services. NHE fact sheet [Internet]. Baltimore (MD): CMS; [last modi- fied 2021 Dec 15; cited 2022 Jan 3].

Available from: https://www.cms .gov/Research-Statistics-Data-and- Systems/Statistics-Trends-and- Reports/NationalHealthExpend Data/NHE-Fact-Sheet

14 McDermott KW, Jiang HJ. Charac- teristics and costs of potentially preventable inpatient stays, 2017 [Internet]. Rockville (MD): Agency for Healthcare Research and Quality; 2020 Jun [cited 2021 Dec 23]. (HCUP Statistical Brief No. 259). Available from: www.hcup-us.ahrq .gov/reports/statbriefs/sb259- Potentially-Preventable- Hospitalizations-2017.pdf

15 Kansagara D, Englander H, Salanitro A, Kagen D, Theobald C, Freeman M, et al. Risk prediction models for hospital readmission: a systematic review. JAMA. 2011; 306(15):1688–98.

16 Holloway J, Neely C, Yuan X, Zhang Y, Ouyang J, Cantrell D, et al. Eval- uating the performance of a predic- tive modeling approach to identify- ing members at high-risk of hospitalization. J Med Econ. 2020; 23(3):228–34.

17 Ukert B, David G, Smith-McLallen A, Chawla R. Do payor‐based outreach programs reduce medical cost and utilization? Health Econ. 2020; 29(6):671–82.

18 David G, Smith-McLallen A, Ukert B. The effect of predictive analytics–

February 2022 41 :2 Health Affairs 217 Downloaded from HealthAffairs.org on February 22, 2022.

driven interventions on healthcare utilization. J Health Econ. 2019; 64:68–79.

19 Laditka JN, Laditka SB. Race, eth- nicity, and hospitalization for six chronic ambulatory care sensitive conditions in the USA. Ethn Health. 2006;11(3):247–63.

20 Gianfrancesco MA, Tamang S, Yazdany J, Schmajuk G. Potential biases in machine learning algo- rithms using electronic health record data. JAMA Intern Med. 2018;178(11):1544–7.

21 Sabaté E. Adherence to long-term therapies: evidence for action [In- ternet]. Geneva: World Health Or- ganization; 2003 [cited 2022 Jan 3]. Available from: https://apps.who .int/iris/bitstream/handle/10665/ 42682/9241545992.pdf

22 Benjamin RM. Medication adher- ence: helping patients take their medicines as directed. Public Health Rep. 2012;127(1):2–3.

23 Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005; 353(5):487–97.

24 Ringwalt C, Roberts AW, Gugelmann H, Skinner AC. Racial disparities across provider specialties in opioid prescriptions dispensed to Medicaid beneficiaries with chronic non- cancer pain. Pain Med. 2015;16(4): 633–40.

25 Singhal A, Tien YY, Hsia RY. Racial- ethnic disparities in opioid pre- scriptions at emergency department visits for conditions commonly as- sociated with prescription drug abuse. PloS One. 2016 ;11(8): e0159224.

26 Jung K, Lim D, Shi Y. Racial-ethnic disparities in use of antidepressants in private coverage: implications for the Affordable Care Act. Psychiatr Serv. 2014;65(9):1140–6.

27 Cruz-Flores S, Rabinstein A, Biller J, Elkind MS, Griffith P, Gorelick PB, et al. Racial-ethnic disparities in stroke care: the American experi- ence: a statement for healthcare professionals from the American Heart Association/American Stroke Association. Stroke. 2011;42(7): 2091–116.

28 Trinacty CM, Adams AS, Soumerai SB, Zhang F, Meigs JB, Piette JD, et al. Racial differences in long-term adherence to oral antidiabetic drug therapy: a longitudinal cohort study. BMC Health Serv Res. 2009;9:24.

29 Mehta KM,Yin M, Resendez C,Yaffe K. Ethnic differences in acetyl- cholinesterase inhibitor use for Alzheimer disease. Neurology. 2005;65(1):159–62.

30 Karpinski SA, Look KA, Vanness DJ. Factors associated with county-level variation in the prescription of statins. J Manag Care Spec Pharm. 2019;25(12):1358–65.

31 Qato DM, Alexander GC, Chakraborty A, Guadamuz JS, Jackson JW. Association between pharmacy closures and adherence to cardiovascular medications among older US adults. JAMA Netw Open. 2019;2(4):e192606.

32 Wisseh C, Hildreth K, Marshall J, Tanner A, Bazargan M, Robinson P. Social determinants of pharmacy deserts in Los Angeles County. J Racial Ethn Health Disparities. 2020;8(6):1424–34.

33 Bryant-Stephens T. Asthma dispar- ities in urban environments. J Al- lergy Clin Immunol. 2009;123(6): 1199–206.

34 Hall WJ, Chapman MV, Lee KM, Merino YM, Thomas TW, Payne BK, et al. Implicit racial/ethnic bias among health care professionals and its influence on health care out- comes: a systematic review. Am J Public Health. 2015;105(12):e60–76.

35 McQuaid EL, Landier W. Cultural issues in medication adherence: disparities and directions. J Gen In- tern Med. 2018;33(2):200–6.

36 Kamishima T, Akaho S, Sakuma J. Fairness-aware learning through regularization approach. In: Spiliopoulou M, Wang H, Cook D, Pei J, Wang W, Zaïane O, et al. edi- tors. ICDMW 2011: 11th IEEE Inter- national Conference on Data Mining Workshops; 2011 Dec; Vancouver, British Columbia [Internet]. Piscataway (NJ): Institute of Elec- trical and Electronics Engineers, Inc.; 2011 [cited 2021 Dec 23]. p. 643–50. Available from: https:// ieeexplore.ieee.org/xpl/conhome/ 6136510/proceeding

37 Zafar MB, Valera I, Rodriguez MG, Gummadi KP. Fairness constraints: mechanisms for fair classification. In: Proceedings of the 20th Inter- national Conference on Artificial Intelligence and Statistics 2017; 2011 Apr 20–22; Fort Lauderdale, Florida. Proceedings of Machine Learning Research [serial on the Internet]. Vol. 54 (2017); [cited 2021 Dec 23]. Available from: http://proceedings .mlr.press/v54/zafar17a/zafar17a .pdf

38 Kusner M, Loftus J, Russell C, Silva R. Counterfactual fairness. In: Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S, et al. editors. Advances in Neural Information Processing Systems 30 (NIPS 2017); 2017 Dec 4–9; Long Beach, California [Internet]. [place unknown]: Neural Information Processing Systems Foundation; [cited 2021 Dec 23]. Available from: https://proceedings.neurips.cc/ paper/2017/file/a486cd07e4ac3d 270571622f4f316ec5-Paper.pdf

39 Hardt M, Price E, Srebro N. Equality of opportunity in supervised learn-

ing. In: Lee D, Sugiyama M, Luxburg U, Guyon I, Garnett R editors. Advances in Neural Information Processing Systems 29 (NIPS 2016); 2016 Dec 5–10; Barcelona, Spain [Internet]. [place unknown]: Neural Information Processing Systems Foundation; [cited 2021 Dec 23]. Available from: https://proceedings .neurips.cc/paper/2016/file/ 9d2682367c3935defcb1f9e247a97 c0d-Paper.pdf

40 Embi PJ. Algorithmovigilance— advancing methods to analyze and monitor artificial intelligence– driven health care for effectiveness and equity. JAMA Netw Open. 2021;4(4):e214622.

41 Interagency Health Reform Council. Health care reform recommenda- tions [Internet]. Harrisburg (PA): Office of Governor Tom Wolf; 2020 Dec [cited 2021 Dec 15]. Available from: https://www.governor.pa .gov/wp-content/uploads/2021/01/ IHRC-HCR-Recommendations.pdf

42 National Committee for Quality As- surance. Health equity [Internet]. Washington (DC): NCQA; [cited 2022 Jan 12]. Available from: https://www.ncqa.org/about-ncqa/ health-equity/

43 Obermeyer Z, Nissan R, Stern M, Eaneff S, Bembeneck EJ, Mullainathan S. Algorithm bias playbook [Internet]. Chicago (IL): Center for Applied Artificial Intelli- gence; [cited 2021 Dec 23]. Available for download from: https://www .chicagobooth.edu/research/center- for-applied-artificial-intelligence/ research/algorithmic-bias

44 Voicu I. Using first name informa- tion to improve race and ethnicity classification. Statistics and Public Policy. 2018;5(1):1–3.

45 Haas A, Elliott MN, Dembosky JW, Adams JL, Wilson‐Frederick SM, Mallett JS, et al. Imputation of race/ ethnicity to enable measurement of HEDIS performance by race/ethnic- ity. Health Serv Res. 2019;54(1): 13–23.

46 Zhang H, Lu AX, Abdalla M, McDermott M, Ghassemi M. Hurtful words: quantifying biases in clinical contextual word embeddings. In: CHIL ’20: Proceedings of the ACM Conference on Health, Inference, and Learning; 2020 Apr 2–4; Toronto, Ontario [Internet]. New York (NY): Association for Comput- ing Machinery; [cited 2021 Dec 23]. Available from: https://dl.acm.org/ doi/10.1145/3368555.3384448

47 Pollard TJ, Chen I,Wiens J, Horng S, Wong D, Ghassemi M, et al. Turning the crank for machine learning: ease, at what expense? Lancet Digit Health. 2019;1(5):e198–9.

Health Information Technology

218 Health Affairs February 2022 41 :2 Downloaded from HealthAffairs.org on February 22, 2022.