Can IT Help to Remove Racism from Healthcare?

There are 2 types of racism that impact healthcare: institutional racism and personally-mediated racism. The first refers to the “differential
access to goods, services, and opportunities by race” – including access to health insurance and therefore preventive and acute care. The second type refers to individual prejudice or discrimination based on race. Personally-mediated racism can impact the types of treatment options offered to patients and the level of shared decision-making.

Evidence-based practice seeks to guide care decisions through best evidence, physician experience, and patient preferences. To best embody evidence-based practice, physicians should look at racial background in terms of understanding how each patient may be impacted by systemic racism and discrimination. For example, physicians should be aware of historical context, including medical experimentation on slaves and the Tuskegee experiment in the 1900s, in which a group of Black men were unknowingly infected with syphilis by physicians, who observed their symptoms over 40 years without providing any treatment.

These events have led to deep mistrust of physicians within the Black community. In light of this, non-Black physicians should understand these historical impacts and seek to develop good relationships with their patients to build trust. Furthermore, physicians should know about trauma-induced risk factors resulting from stress responses to racial discrimination. These physiological changes can build over time and lead to increased risk of conditions such as cardiovascular disease. Lastly, care providers should work to overcome limitations to access due to lack of insurance coverage. Free options should be presented for testing and other services where patients are not required to have insurance.

When it comes to practice-based evidence, most research focusing on racism in healthcare has used data on patient perceptions of discrimination. These perceptions can have impacts on health, and there has not been much progress in developing accurate methods for assessing physician bias. As long as data on race is captured accurately for patients, quality of care and the services utilized by patients can be analyzed across racial categories. Implicit bias, a major way in which racism impacts quality of care, can be assessed through data collection on the conversations physicians have with patients and what treatments they offer to patients.

So, where does informatics come into the picture?

Informatics plays a huge role in identifying inequities in healthcare practice, as well as collecting data that can be used to understand the causes of these inequities. For example, in order to analyze the quality of care and the services provided to patients across different racial backgrounds, researchers need access to EMR data that has ICD-10 diagnosis codes, CPT codes, and racial data.

Natural language processing (NLP) is a developing area of informatics using machine learning to analyze voice or written conversation data. As voice-driven EMR documentation increases to improve physician user experience, the data collected can be analyzed through NLP to determine attitudes, relationship closeness, and balance of conversation between patients and providers. Coded language terms that indicate bias or discrimination can also be tagged and identified through data mining of the conversation data. This would help to improve the measurement of implicit bias within physicians.

Another approach is to update the standards for social determinants of health to include racism. Based on the potential health impacts of facing discrimination, updating EMRs and other informatics systems to collect patient data on experienced discrimination would help to ensure proper evidence-based practice.

This update to SDOH data could help to improve race-based clinical decision support algorithms that have been criticized for perpetuating system racism through harmful “race-correcting” of outputs. See the chart below from The New England Journal of Medicine for examples of how these algorithms are potentially contributing to disparities in quality of care, rather than mending them.

Table 1. Examples Of Race Correction in Clinical Medicine* 
Tool and Clinical Utility 
Cardiology 
The American Heart Association's Get with the 
Guidelines—Heart Failure9 (https://wv.æ.' 
.mdcalc.com/gwtg-heart-failure-risk-score) 
Predicts in-hospital mortality in patients with 
acute heart failure. Clinicians are advised to use 
this risk stratification to guide decisions regarding 
initiating medical therapy. 
Cardiac surgery 
The Society of Thoracic Surgeons Short Term 
Risk Calculator10 (http://riskcalc.sts.org/ 
stswebriskcalc/calculate) 
Calculates a patient's risks of complications and 
death with the most Common cardiac surgeries. 
Considers variables, some of which are listed 
here. 
Nephrology 
Estimated glomerular filtration rate (eGFR) 
MDRD and CKD-EPI equationsn (https:// 
ukidney.com/nephrology-resources/egfr 
-calculator) 
Estimates glomerularfiltration rate on the basis 
of a measurement of serum creatinine. 
Organ Procurement and Transplantation Network: 
Kidney Donor Risk Index (KDRl)L2 (https:// 
optn.transplant.hrsa.gov/resources/allocation 
-calculators/kdpi-calculator/) 
Estimates predicted risk of donor kidney graft fail- 
ure, which is used to predict viability ofpotential 
kidney donor. T 
Obstetrics 
Vaginal Birth after Cesarean (VBAC) Risk 
Calculator13'1' (https://mfmunetwork.bsc.gwu 
.edu/PublicBSC/MFMU/VGBirthCalc/vagbirth 
.html) 
Estimates the probability of successful paginal 
birth after prior cesarean section. Clinicians can 
use this estimate to counsel people who have to 
decide whether to attempt a trial Of labor rather 
than undergo a repeat cesarean section. 
Urology 
STONE ScorelS,16 
Predicts the risk of a ureteral stone in patients 
who present with flank pain 
Urinary tract infection (UTI) calculatorl' (https:// 
uticalc.pitt.edu/) 
Estimates the risk of UTI in children 2—23 mo 
Of age to guide decisions about when to pursue 
urine testing for definitive diagnosis 
Oncology 
Rectal Cancer Survival Calculator18 (http:// 
wævw3.mdanderson.0rg/app/medcalc/index 
Estimates conditional survival 1—5 yr after diag- 
nosis with rectal cancer 
National Cancer Institute Breast Cancer Risk 
Assessment Tool (https://bcrisktool.cancer 
.gov/calculator.html) 
Estimates 5-yr and lifetime risk of developing 
breast cancer, for women without prior history of 
breast cancer, DCIS, or LCIS. 
Breast Cancer Surveillance Consortium Risk 
Calculator19 (https://tools.bcsc-scc.org/ 
BC5yearRisk/calculator.htm) 
Estimates 5- and 10-yr risk of developing breast 
cancer in women with no previous diagnosis Of 
breast cancer, DCIS, prior breast augmentation, 
or prior mastectomy 
Endocrinology 
Osteoporosis Risk SCORE (Simple Calculated 
Osteoporosis Risk Estimation)20 (https://www 
.mdapp.co/osteoporosis-risk-score-calculator 
-3161) 
Determines whether a woman is at low, moder- 
ate, or high risk for low bone density in order to 
guide decisions about screening with DXA scan 
Fracture Risk Assessment Tool (FRAX)21 (https:// 
www.sheffield.ac.uk/FRAX/tool.aspx) 
Estimates 1 Oyr risk Of a hip fracture or Other ma- 
jor osteoporoticfracture on the basis of patient 
demographics and risk-factor profile. Calculators 
are country-specific. 
Pu Imonology 
Pulmonary-function tests22 
Uses spirometry to measure lung volume and the 
rate Of flow through airways in order to diagnose 
and monitor pulmonary disease 
Input Variables 
Systolic blood pressure 
Blood urea nitrogen 
Sodium 
Age 
Heart rate 
History Of COPD 
Race: black or nonblack 
Operation type 
Age and sex 
Race: black/African American, Asian, 
American Indian/Alaskan Native, 
Native Hawaiian/Pacific Islander, or 
"Hispanic, Latino or Spanish ethnic- 
ity"; white race is the default setting. 
BMI 
Serum creatinine 
Age and sex 
Race: black vs. white or other 
Age 
Hypertension, diabetes 
Serum creatinine level 
Cause of death (e.g., cerebrovascular 
accident) 
Donation after cardiac death 
Hepatitis C 
Height and weight 
HLA matching 
Cold ischemia 
En bloc transplantation 
Double kidney transplantation 
Race: African American 
Age 
BMI 
Prior vaginal delivery 
Prior VBAC 
Recurring indication for cesarean sec- 
tion 
African-American race 
Hispanic ethnicity 
Sex 
Acute onset of pain 
Race: black or nonblack 
Nausea or vomiting 
Hematuria 
Age months 
Maximum temperature >390C 
Race: Describes self as black (fully or 
partially) 
Female or uncircumcised male 
Other fever source 
Age and sex 
Race: white, black, Other 
Grade 
Stage 
Surgical history 
Current age, age at menarche, and age 
at first live birth 
First-degree relatives with breast cancer 
Prior benign biopsies, atypical biopsies 
Race/ethnicity: white, African American, 
Hispanic/Latina, Asian American, 
American Indian/Alaska Native, 
unknown 
Age 
Race/ethnicity: white, black, Asian, 
Native American, other/multiple 
races, unknown 
BIRADS breast density score 
First-degree relative with breast cancer 
Pathology results from prior biopsies 
Rheumatoid arthritis 
History of fracture 
Age 
Estrogen use 
Weight 
Race: black or not black 
Age and sex 
Weight and height 
Previous fracture 
Parent who had a hip fracture 
Current smoking 
Glucocorticoid use 
Rheumatoid arthritis 
Secondary osteoporosis 
Alcohol use, 23 drinks per day 
Femoral neck bone mineral density 
Age and sex 
Height 
Race/ethnicity 
Use Of Race 
Adds 3 points to the risk score ifthe patient 
is identified as nonblack. This addition 
increases the estimated probability Of 
death (higher scores predict higher 
mortality). 
The risk score for operative mortality and 
major complications increases (in some 
cases, by 20%) ifa patient is identified 
as black. Identification as another non- 
white race or ethnicity does not increase 
the risk score for death, but it does 
change the risk score for major compli- 
cations such as renal failure, stroke, and 
prolonged ventilation. 
The MDRD equation reports a higher eGFR 
(by a factor of 1.210) ifthe patient is 
identified as black. This adjustment is 
similar in magnitude to the correction 
for sex (0.742 if female). 
The CKD-EPI equation (which included a 
larger number ofblack patients in the 
study population), proposes a more 
modest race correction (by a factor 
of 1.159) if the patient is identified as 
black. This correction is larger than the 
correction for sex (1.018 iffemale). 
Increases the predicted risk of kidney graft 
failure ifthe potential donor is identified 
as African American (coefficient, 0.179), 
a risk adjustment intermediate between 
those for hypertension (0.126) and 
diabetes (0.130) and that for elevated 
creatinine (0.209-0.220). 
The African-American and Hispanic correc- 
tion factors subtract from the estimated 
success rate for any person identified 
as black or Hispanic. The decrement 
for black (0.671) or Hispanic (0.680) is 
almost as large as the benefit from prior 
vaginal delivery (0.888) or prior VBAC 
(1.003). 
Produces a score on a 13-point scale, with 
a higher score indicating a higher risk of 
a ureteral stone; 3 points are added for 
nonblack race. This adjustment is the 
same magnitude as for hematuria. 
Assigns a lower likelihood of UTI ifthe child 
is black (i.e., reports a roughly 2.5-times 
increased risk in patients who do not 
describe themselves as black). 
White patients are assigned a regression 
coeffcient Of l, with higher coefficients 
(depending on stage) assigned to black 
patients (1.18—1.72). 
The calculator returns lower risk estimates 
for women who are African American, 
Hispanic/Latina, or Asian American 
(e.g., Chinese). 
The coefficients rank the race/ethnicity 
categories in the following descending 
order Of risk: white, American Indian, 
black, Hispanic, Asian. 
Assigns 5 additional points (maximum 
score of50, indicating highest risk) ifthe 
patient is identified as nonblack 
The U.S. calculator returns a lower fracture 
risk ifa female patient is identified as 
black (by a factor Of 0.43) , Asian (0.50), 
or Hispanic (0.53). Estimates are not 
provided for Native American patients 
Or for multiracial patients. 
In the U.S., spirometers use correction 
factors for persons labeled as black 
(10-15%) or Asian (4-6%). 
Equity Concern 
The original study envisioned using this score 
to "increase the use of recommended 
medical therapy in high-risk patients and 
reduce resource utilization in those at low 
risk. "9 The race correction regards black 
patients as lower risk and may raise the 
threshold for using clinical resources for 
black patients. 
When used preoperatively to assess a patient's 
risk, these calculations could steer minority 
patients, deemed higher risk, away from 
these procedures. 
Both equations report higher eGFR values 
(given the same creatinine measurement) 
for patients identified as black, suggesting 
better kidney function. These higher eGFR 
values may delay referral to specialist care 
or listing for kidney transplantation. 
Use ofthis tool may reduce the pool of African- 
American kidney donors in the United 
States. Since African-American patients are 
more likely to receive kidneys from African- 
American donors, by reducing the pool of 
available kidneys, the KDRI could exacer- 
bate this racial inequity in access to kidneys 
for transplantation. 
The VBAC score predicts a lower chance Of suc- 
cess ifthe person is identified as black or 
Hispanic. These lower estimates may dis- 
suade clinicians from offering trials of labor 
to people of color. 
By systematically reporting lower risk for black 
patients than for all nonblack patients, this 
calculator may steer clinicians away from 
aggressive evaluations Of black patients. 
By systematically reporting lower risk for black 
children than for all nonblack children, this 
calculator may deter clinicians from pursu- 
ing definitive diagnostic testing for black 
children presenting with symptoms of UT l. 
The calculator predicts that black patients will 
have shorter cancer-specific survival from 
rectal cancer than white patients. Clinicians 
might be more or less likely to offer inter- 
ventions to patients with lower predicted 
survival rates. 
Though the model is intended to help concep- 
tualize risk and guide screening decisions, 
it may inappropriately discourage more ag- 
gressive screening among some groups of 
nonwhite women. 
Returns lower risk estimates for all nonwhite 
race/ethnicity categories, potentially reduc- 
ing the likelihood Of close surveillance in 
these patients. 
By systematically lowering the estimated risk 
ofosteoporosis in black patients, SCORE 
may discourage clinicians from pursuing 
further evaluation (e.g., DXA scan) in black 
patients, potentially delaying diagnosis and 
intervention. 
The calculator reports 10-yr risk Of major os- 
teoporotic fracture for black women as less 
than halfthat for white women with identi- 
cal risk factors. For Asian and Hispanic 
women, risk is estimated at about halfthat 
for white women. This lower risk reported 
for nonwhite women may delay intervention 
with osteoporosis therapy. 
Inaccurate estimates of lung function may 
result in the misclassification of disease 
severity and impairment for racial/ethnic 
minorities (e.g., in asthma and COPD).23 
* BIRADS denotes Breast Imaging Reporting and Data System, BMI body-mass index (the weight in kilograms divided by the square of the height in meters), CKD-EPI Chronic Kidney 
Disease Epidemiology Collaboration, COPD chronic obstructive pulmonary disease, DCIS ductal carcinoma in situ, DXA dual-energy x-ray absorptiometry, LCIS lobular carcinoma in 
situ, and MDRD Modification of Diet in Renal Disease study. 
-i- The current calculator uses Ethnicity/Race, with the following options: American Indian or Alaska Native, Asian, Black or African American, Hispanic/Latino, Native Hawaiian or Other 
Pacific Islander, White, and Multiracial. 
Three countries' calculators are further subcategorized by race, ethnicity, or location: China (Mainland China, Hong Kong), Singapore (Chinese, Malay, Indian), and the United States 
(Caucasian, black, Hispanic, Asian).

Inaccurately identifying patients as high or low-risk based on race can lead to changes in the services or treatments offered. These algorithms need to be trained to incorporate the nuances of the combined effects of systemic racism and discrimination. As with all informatics research and innovation, sample data used to train any healthcare AI models needs to include equal representation for individuals of all races and demographics to ensure that the technology benefits all demographic groups equally.

References:

Vyas, D. A., M.D., Eisenstein, L. G., M.D., & Jones, D. S., M.D. Ph.D. (2020). Hidden in Plain Sight — Reconsidering the Use of Race Correction in Clinical Algorithms. The New England Journal ofMedicine, (383). https://doi.org/10.1056/NEJMms2004740

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s