Diversity is not just important for reasons of equality; it is essential to counteract potential biases in data and in human judgments.

It’s well established that clinicians can be influenced by subconscious bias. Often these biases are so deep-set that we are blind to them [1]. Biases in health data are also common and can be life threatening when not addressed properly.

For example, heart attacks are more common in men, but are also the leading cause of death in women – at least in Western society. Despite this, it is more common for heart disease in women to be overlooked by doctors, unrecognized, and therefore untreated [2]. This isn’t just because it’s considered less likely, but also because symptomatically it often manifests differently in women than in men [3].

Similarly, AI in healthcare is only as good as the people and data it learns from. This means a lack of diversity in the development of AI models can drastically reduce its effectiveness [4]. AI trained on biased data will simply amplify that bias [5]. For example, IBM Watson’s recommendations for cancer treatments have been based on training by just a small number of physicians at one medical institution [6]. This creates biased recommendations based not on official guidelines but on the experience and opinions of a few, probably quite similar, people.

This makes diversity particularly critical in healthcare and healthtech. Having diverse teams – in terms of gender, ethnicity, training and background – will increase the likelihood of unconscious biases being recognized and addressed, rather than encoded within the next generation of AI technologies. This will in-turn improve the impartiality of the data upon which care decisions are based.

See More on the topic of biases, diversity and inclusion:

Part 1: The Gender Imabalance in AIMed

Part 2: The business case for diversity

Part 4: Intentional Inclusivity: a new strategy for solving medical problems

or see the whole feature on Diversity & Inclusion in AIMed in AIMed Magazine issue 04 here.

References

[1] https://www.newscientist.com/article/mg21028122-200-the-grand-delusion-blind-to-bias/

[2] https://blog.thesullivangroup.com/rsqsolutions/heart-disease-in-women-underdiagnosed-undertreated

[3] https://www.nature.com/collections/qdxbjfddqb

[4] https://publications.parliament.uk/pa/ld201719/ldselect/ldai/100/10002.htm

[5] http://www.pnas.org/content/early/2018/03/30/1720347115

[6] https://paulvanderlaken.com/2017/09/12/ibms-watson-for-oncology-a-biased-and-unproven-recommendation-system-in-cancer-treatment/

Bio

biases artificial intelligence medicine data healthcare diversityDr. Claire Novorol, Chief Medical Officer, Ada Health

Claire is the co-founder and Chief Medical Officer for Ada Health, a personal health guide that uses AI and machine learning to help people to understand and manage their health. Before founding Ada, Claire worked as a Paediatrician in London before specialising in Clinical Genetics. She has degrees in Pathology and Medicine as well as a PhD in Neuroscience from the University of Cambridge.