Bias in artificial intelligence algorithms and recommendations for mitigation

被引：159

作者：

Nazer, Lama H. ^{[1
]}

Zatarah, Razan ^{[1
]}

Waldrip, Shai ^{[2
]}

Ke, Janny Xue Chen ^{[3
]}

Moukheiber, Mira ^{[4
]}

Khanna, Ashish K. ^{[5
,6
,7
]}

Hicklen, Rachel S. ^{[8
]}

Moukheiber, Lama ^{[4
]}

Moukheiber, Dana ^{[4
]}

Ma, Haobo ^{[9
]}

Mathur, Piyush ^{[10
]}

机构：

[1] King Hussein Canc Ctr, Dept Pharm, Amman, Jordan

[2] Morehouse Sch Med, Dept Med, Atlanta, GA USA

[3] Univ British Columbia, Dalhousie Univ, St Pauls Hosp, Dept Med, Vancouver, BC, Canada

[4] MIT, Cambridge, MA USA

[5] Atrium Hlth Wake Forest Baptist Med Ctr, Dept Anaesthesiol, Winston Salem, NC USA

[6] Perioperat Outcomes & Informat Collaborat, Winston Salem, NC USA

[7] Outcomes Res Consortium, Cleveland, OH USA

[8] Univ Texas MD Anderson Canc Ctr, Res Med Lib, Houston, TX USA

[9] Beth Israel Deaconess Med Ctr, Dept Anaesthesia & Crit Care Med, Boston, MA USA

[10] Cleveland Clin, Dept Anaesthesia & Crit Care Med, Cleveland, OH USA

来源：

PLOS DIGITAL HEALTH | 2023年 / 2卷 / 06期

关键词：

PREDICTION; HEALTH; DISPARITIES; RISK; TOOL; AI;

D O I：

10.1371/journal.pdig.0000278

中图分类号：

R19 [保健组织与事业（卫生事业管理）];

学科分类号：

摘要：

The adoption of artificial intelligence (AI) algorithms is rapidly increasing in healthcare. Such algorithms may be shaped by various factors such as social determinants of health that can influence health outcomes. While AI algorithms have been proposed as a tool to expand the reach of quality healthcare to underserved communities and improve health equity, recent literature has raised concerns about the propagation of biases and healthcare disparities through implementation of these algorithms. Thus, it is critical to understand the sources of bias inherent in AI-based algorithms. This review aims to highlight the potential sources of bias within each step of developing AI algorithms in healthcare, starting from framing the problem, data collection, preprocessing, development, and validation, as well as their full implementation. For each of these steps, we also discuss strategies to mitigate the bias and disparities. A checklist was developed with recommendations for reducing bias during the development and implementation stages. It is important for developers and users of AI- based algorithms to keep these important considerations in mind to advance health equity for all populations.

引用

页数：14

共 59 条

[1]

[Anonymous], 53. A health telematics policy in support of WHOs Health-for-all strategy for global health development: report of the WHO Group Consultation on Health Telematics, 11-16 December, Geneva, 1997 World Health Organization. 1998 URL: https://apps.who.int/iris/handle/10665/63857 [accessed 2021-04-15]

[2]

[Anonymous], 2011, Action plan to reduce racial and ethnic disparities: A Nation free of disparities in health and healthcare

[3]

[Anonymous], 174. Open-Label, Single Ascending Dose Study to Evaluate the Safety, Tolerability, and Efficacy of AGN-151587 (EDIT-101) in Adult and Pediatric Participants With Leber Congenital Amaurosis Type 10 (LCA10), With Centrosomal Protein 290 (CEP290)-Related Retinal Degeneration Caused by a Compound Heterozygous or Homozygous Mutation Involving c.2991+1655Agt

[4]

G in Intron 26 (IVS26) of the CEP290 Gene ("LCA10-IVS26") 2019 [Available from: https://clinicaltrials.gov/ct2/show/NCT03872479?term=CRISPRdraw=6rank=30.

[5]

[Anonymous], 10. Centers for Disease Control and Prevention. Available from: https://www.cdc.gov.

[6] Lessons and tips for designing a machine learning study using EHR data [J].