SCIENTIA.

Artificial Intelligence 

Often, we regard AI as a fair, unbiased and non-discriminatory technology. Major factors why we trust AI include its incapability of consciousness, sensing morals and processing emotions the way we do. Perhaps this allows them to make fair decisions - uninfluenced by human bias. 

 

Wait, is AI completely fair?

Let’s think about a scenario: a company is hiring new employees amongst an endless list of applicants. Before they start interviewing the candidates, they want to have an elimination stage in which the AI selects a narrower range of qualified applicants. Is this a totally efficient and fair system? Have a think.

 

Definitely, one may argue. 

It’s probably better than having humans select candidates who are susceptible to various prejudices and biases. The first round- with only the document of their academics, nationality and gender- should not be influenced by inclination towards a group, intended or not. Interviewing can be carried out by humans later on, allowing employers to interact with applicants in greater depth.

 

Actually, no. 

Here comes the alarm: AI can discriminate. Several years ago, Amazon failed to adapt a similar recruitment tool based on AI technology. It had shown gender bias against women and inclination towards men candidates. For example, applications including gender key words like “women’s golf club” and graduates from women colleges were penalized. The reason was simply because it reflected data collected, prepared and selected from a male-dominant tech industry. 

 

Here, we can apply the concept of “garbage in, garbage out”; incomplete, flawed data results in a flawed output. In other words, the bias already rooted in our society over preferred race, gender and degrees is input into the AI as data - the AI is only trained to apply the concept. 

 

The lack of diversity in data also affects many other platforms, for example facial recognition disproportionately misidentifying women (especially of colour). Yet, it becomes a serious issue when flawed data can further interfere with our living and health.

 

Gender bias in clinical trials

 

Are women receiving optimal treatment?

 

Research shows that women are more likely to suffer from side effects of drug usage because of the clinical trials relying heavily on male participants. Gender is a fundamental, yet underappreciated variable in clinical development of drugs. 

 

Diversity in research is crucial as patients from different subgroups (age,gender,race,disabilities,etc) can experience varying treatment outcomes to the same medication. 

 

“Men and women have different concentrations of digestive enzymes starting in the mouth, stomach and liver, which affect the metabolism of that drug,” says McGregor of Brown University.

 

However, there are other significant differences between male and females that must be considered, for example :

 

• Fluctuating hormone levels from menstruation each month (female)

• Potential to becoming pregnant (female)

• Physical differences (body composition, size, weight)

• Patterns in genetic expression 

• Social, cultural factors

 

So what hinders women participation rates?

 

Until the early 1990s, women of childbearing age were banned from participating in medical trials due to concerns over potential damage to pregnancy and their fetuses, regardless of their capabilities to become pregnant, sexual orientations and desire to carry a child. 

 

In return, these efforts to protect women resulted in women being underrepresented in clinical trials, therefore resulting in unreliable results of side effects, correct dosage and drug efficacy for different subgroups. 

 

In later stage trials (Phase 3), there is an increased enrollment of women; these numbers significantly drop in Phase 1 trials. This is concerning because the stage allows researchers details on how the drug works- how it is broken down, absorbed, processed, distributed, excreted etc.

 

For example, Zolpidem was originally approved with equal doses for male and female.Yet alarmingly, women experienced serious side effects like increased traffic accidents in the morning after taking the medication. After realizing the same doses in women caused double the drug levels (due to differences in metabolism), the FDA later directed that women should receive half the standard dose given to men [2].

 

ACTIONS?

 

Actions to enhance the collection and availability of subgroup data may include:

 

1. Participation [3] - improving community outreach and health literacy 

• Identifying barriers to enrollment for members of subgroup 

• Implementing programs to encourage enrollment

• Outreach to communities lacking access or knowledge to clinical trials 

 

  2. Transparency [3]

• Providing clear information with patients about the trial processes and potential side effects, so patients feel confident being aware of the details. 

  

  3. Technology 

• FDA’s project Critical Path Initiative aims to modernize drug development through use of innovative tools and techniques [4] such as: 

  • Biomarkers to estimate efficacy and safety outcomes 

  • Advanced technology - study designs

  • New trials designs

  • New data analysis techniques 

  • Collectively, these can help identify womens’ different responses accurately

  

  4. Pregnancy 

• Fertile women participating in the trials must be warned to avoid pregnancy during the period of drug exposure

• Women should have access to counseling and medical care for contraception

• Researchers must confirm the participants are not pregnant, and continue to monitor for pregnancy

• Detailed information about potential risks to a fetus must be stated clearly for women participants’ confidence

 

Will gender discrimination in trials be solved by simply achieving a 50:50 split ratio?

 

Rather than a simple 50:50 ratio, a participation to prevalence ratio (PPR) higher than 0.8 is more commonly calculated to confirm an appropriate participation ratio of women.

The participation to prevalence ratio (PPR) is a metric used to describe representation of women in a trial relative to their representation in the disease population [1] [2]  The PPR was calculated as follows: