Introducing
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
Bias refers to an inequity of either perception or treatment towards a certain group. Bias can be social-- referring to the adoption of stereotypes and prejudices that affect how one treats others-- and statistical-- referring to data sets that misrepresent or exclude certain populations and lead to skewed conclusions and application.(1)
An algorithm, when used in medicine, is a calculator that can determine risk, priority, and qualification for various medical procedures and practices. These calculations are based on formulas that algorithm developers input, and those formulas are meant to help physicians make the most practical and economical decisions for their benefit of both their patients and medical facility. (3)
Data bias, otherwise known as algorithmic bias, is within the statistical embodiment of bias. In the realm of medicine, it describes how an algorithm's data set or computing process may calculate recommendations that can influence a physician to make poor decisions-- including the misdiagnosis and mistreatment of a certain group of people. (1-3).
Algorithms and artificial intelligence (AI) are used in healthcare to help evaluate or determine(1):
When these algorithms, though created to be resourceful, are either built on data from non-inclusive populations or use formulas that reiterate current trends of societal inequity, they become "biased".
For example, an cost-saving commercial algorithm was created to calculate health risks in an effort to assist in the determination of patients in need of advanced care treatment programs. The algorithm was formulated from vast data sets that excluded patient race, and instead used medical cost trends to determine health risks. It assists 200 million patients annually. (3)
At first glance, the algorithm would be unbiased, as it calculated very similar healthcare costs across patient races. But, since black patients generate lower health costs on average-- possibly due to socioeconomic hardship, careers without insurance benefits, and general distrust in the healthcare system and its professionals-- yet have over 26% more chronic illnesses such as diabetes, hypertension, renal failure and anemia compared to their white counterparts, the algorithm created a higher threshold for black patients to be advised intensive treatment. (3)
This calculated cardiac surgery complication risk based on the type of operation and patient characteristics.
Post-surgery mortality risk is increased by up to 20% if the patient is identified as black. Other nonwhite minorities are affected as well with higher risk of surgical complication rather than mortality.
These calculations could lessen recommendation for necessary heart operations. (4).
This predicts mortality in patients with heart failure.
It adds 3 points for non-black patients. A higher scores mean a higher prediction of mortality.
This calculation regarded black patients as of lower risk than white counterparts with similar symptoms, making it harder for them to be recommended therapy treatments.(4).
This calculator helps determine if a mother can safely undergo a vaginal birth after having previously given birth through cesarean section. It assesses age, race, and the experience of previous births.
Being African American or Hispanic increases risk for VBAC--and does so at a similar interval to the amount that risk is lowered in patients who have undergone previous vaginal birth or VBACs.
This may cause clinicians to unnecessarily push surgical births for certain minority patients. (4)
This algorithm measures the risk that a patient has for developing breast cancer. It assesses previous breast screening results, familial history, birthing experiences, age, and race.
Nonwhite patients (Asian, Black, and Hispanic) are given lower risk scores that their White counterparts.
If used, this tool may deter physicians from more intensely screening minority patients for breast cancer. (4)
This calculator forecasts the chance of survival a patient has after being diagnosed with rectal cancer. It takes into consideration the characteristics and severity of the cancer, as well as patient characteristics like age and race.
Black patients are given higher regression coefficients that their White counterparts (1.18- 1.72 compared to 1).
This increases the perception of cancer-related mortality in Black patients, which may decrease clinical efforts for treatment.(4)
This test uses spirometers to measure lung capacity and airway flow to diagnose and monitor lung disease.
Black patients have a 15% correction on spirometers and Asian patients have a 4-6% correction on spirometers in the US.
This can lower physician perception of the true severity of a patient's lung disease and may lead to misdiagnosis or insufficient treatment. (4)
This index predicts the success of kidney grafting and donation. It determines this using the graft source's medical history as well as their personal characteristics i.e. race and age.
If the donor was Black, the associated risk for donation failure is increased at a figure similar to that of a patient being diabetic or having hypertension.
This can reduce donor options for Black patients interested in receiving kidneys from Black donors. (4)
This calculator uses creatine levels as well as age, sex and race of the patient to determine glomerular filtration rate.
Both the MDRD and CKD-EPI equations have similar race corrections that report higher rates for Black patients. These race corrections affect eGFR at a figure higher than gender does.
These calculations make kidney performance in Black patients seem higher compared to their non=black counterparts, thus potentially lowering the recommendation of further treatment when necessary. (4)
That is completely up to the developer. These algorithms are commercial, meaning that they are made independently and sold to various medical systems for profit. They are often formulated to target specific decision-making tasks in an effort to conserve time and resources. Whether or not medical systems want to incorporate them is a voluntary decision. (5)
Though there is some regulation of medical algorithms-- particularly by the Food and Drug Administration (FDA) and its Digital Health Innovation Action Plan to ensure proper development and operation-- there is still a lack of official, proactive measure to combat the occasional occurrence of racial, gender and sexual bias. (6)
B
A
A) This graph refers to an algorithm that calculated health risk in an effort to target candidates for intensive treatment programs. As you can see, Black patients were sicker and had more chronic conditions that their White counterparts of the same assigned risk score. Black patients also had a higher threshold for chronic conditions in order to be referred for both screenings and treatment programs. (3).
B) This graph refers to the spirometry-based lung function scores of patients categorized by race. White patients were calculated to have higher lung function than their non-black counterparts at all stages of age. (8).
The patient is a 35 year old Black woman with a 10 year health history of wheezing, coughing, and shortness of breath. She is a nonsmoker whose symptoms seem to be brought on by environmental allergies and upper respiratory infections. She uses her prescribed albuterol inhaler twice weekly.
Using the Global Lung Initiative (GLI) spirometry, her Forced Vital Capacity (FVC) was calculated to be 15% higher when under African American presets than White presets, and her Forced Expiratory Volume (FEV) was calculated to be 12% higher when under African American presets than White presets.
Since the African American preset calculations required more restricted breathing in order to be considered asthmatic compared to those of the White race, she was not determined to have asthma, and further medication was not prescribed. (8)
The following article, from the American Medical Association, details algorithmic bias as well as the role of the American government surrounding this novel form of discrimination. (10)
https://www.ama-assn.org/delivering-care/health-equity/why-clinical-algorithms-fall-short-race
Now that I am familiar with Data Bias, what should I do as a medical professional?
Awareness is the first step to mitigating bias. (7) Continue to research the technology used in your office, its calculations, data pool used, and its effects on patients of all backgrounds. If race-based adjustments are potentially harmful to patients under an algorithm, alert administration. (9)
Race is more societal than biological. Environmental factors and socioeconomic status have the potential to be better health determinants than ethnicity. Always consider the effects of using race as a factor when determining risk, resource allocation, and treatment options. (11)
You do not have to follow the advice of potentially biased algorithms. If you see that a patient needs to be advocated for in spite of an algorithm's recommendations, do so. (3)
(1) Norori N, Hu Q, Aellen FM, Faraci FD, Tzovara A. Addressing bias in big data and AI for health care: A call for open science. Patterns (New York, N.Y.) 2021 Oct 8,;2(10):100347.
(2) Thomasian NM, Eickhoff C, Adashi EY. Advancing health equity with artificial intelligence. J Public Health Pol 2021 Nov 22,;42(4):602-611.
(3) Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science (American Association for the Advancement of Science) 2019 Oct 25,;366(6464):447-453.
(4) Vyas DA, Eisenstein LG, Jones DS. Hidden in Plain Sight — Reconsidering the Use of Race Correction in Clinical Algorithms. The New England journal of medicine 2020 Aug 27,;383(9):874-882.
(5) Omar RA. Unabashed bias: How health-care organizations can significantly reduce bias in the face of unaccountable AI. Denver Law Review 2021;98(4):807-837.
(6) Zou J, Schiebinger L. Ensuring that biomedical AI benefits diverse populations. EBioMedicine 2021 May;67:103358.
(7) Thomas B, Booth-McCoy AN. Blackface, Implicit Bias, and the Informal Curriculum: Shaping the Healthcare Workforce, and Improving Health. Journal of the National Medical Association 2020 Oct;112(5):533-540.
(8) Ramsey NB, Apter AJ, Israel E, Louisias M, Noroski LM, Nyenhuis SM, et al. Deconstructing the Way We Use Pulmonary Function Test Race-Based Adjustments. The journal of allergy and clinical immunology in practice (Cambridge, MA) 2022 Feb 17,.
(9) Wiggins O. University of Maryland Medical System drops race-based algorithm officials say harms Black patients. The Washington post 2021 Nov 17,.
(10) Robeznieks A. Why clinical algorithms fall short on race. 2020; Available at: https://www.ama-assn.org/delivering-care/health-equity/why-clinical-algorithms-fall-short-race.
(11) Thomasian NM, Eickhoff C, Adashi EY. Advancing health equity with artificial intelligence. J Public Health Pol 2021 Nov 22,;42(4):602-611.