This blog was featured on Clinical Trial AI Weekly.
For centuries, science fiction writers have toyed with the endless possibilities of artificial intelligence (AI). And now we’re living the future those writers celebrated and also feared. How might AI apply in the realities of our day-to-day jobs, especially the use of artificial intelligence for clinical trials?
First, let us establish a shared understanding of AI for the purposes of this post. Generally, AI refers to the process of building or programming machines to replicate human intelligence. Programs of the past merely completed tasks based on human-provided rules. Today, leveraging complex algorithms allows machines to learn, change, and adapt to new situations without direct programming. These applications need only be exposed to large quantities of data.
What does artificial intelligence for clinical trials offer?
Within AI, there are four main categories of functionalities:
- Reactive machines – a machine with abilities to perceive situations within their task set and react to situations as they are presented.
- Limited memory – trains a machine to make decisions based on past data.
- Theory of mind – we haven’t gotten there yet, but this is the stage where machines can replicate and react to the thoughts and emotions of others.
- Self-awareness – is purely hypothetical currently, however, this is where a machine develops an independent intelligence, where self-awareness drives a desire for autonomy.
Today, we are exploring the young lands of limited memory. There is no shortage of people tinkering and testing “new tricks” we can get these machines to perform for us. Over the past decade, we’ve seen plenty of marketing and commercial applications. Software that uses our past behavior to predict and modify it for shareholder and political gain are hot topics.
It’s a classic question of good vs. evil: AI opens the possibility to discover new insights into ourselves and use behavior modification for positive social change. But it can also exploit weaknesses in our institutional systems for personal gain. There is also the possibility that AI will become too big or too smart to control.
Best uses of artificial intelligence for clinical studies
AI has strong potential in patient recruitment and retention. The most tried and true applications are in marketing and behavior prediction/modification.
Many clinical trials are engaging social media and marketing experts to improve their recruitment campaigns. Client relations management applications can be used to assist in patient retention in the same way companies in other industries use AI to retain clients.
Critics suggest it may be unethical to use algorithms to drive behavior. Yes, there are important questions to ask:
- Are algorithms used in marketing exerting undue influence by pushing participants into trials?
- Is there enough educated skepticism protecting people from known marketing influence?
A review and revision of the consent process could mitigate these risks.
Data mining is more efficient
Emerging technologies also use large data resources in electronic health information to target qualified participants based on inclusion/exclusion criteria. However, with past data mining methods, only structured databases with particular fields were useful. It was also labor intensive to create structured databases by taking multiple data resources and consolidating the information. Now, unstructured data with disparate fields and information can be used. This opens up an ocean of resources that weren’t useful 10 years ago.
With emerging technology, AI can be used to sort through data from medical records and social media to determine geographic areas where the trial indication is more widespread. Content from social media forums, where patients discuss their conditions, can be mined to determine where patients are located for specific cohorts.
Ensuring patient safety: are machines superior to humans?
Adverse events are currently monitored through human data entry systems that have notification and escalation processes. Most of the action-based monitoring is still completed by human oversight. With machine learning, there is potential to teach machines how to take in past study data along with adverse event information to create predictive models. In the future, machines could flag trends in adverse events sooner than human oversight or even predict participants at higher risk so they can be monitored more frequently.
There is also the potential that smaller actual patient data samples will be needed to predict safety and efficacy in the future. Using all past research data, it is possible machines could show us what the most efficient population sample comprises. This could reduce the overall time and costs associated with clinical research. It may also better include historically underrepresented participant demographics. Validation of these types of predictive programs appears to remain a long way off.
Challenges in using artificial intelligence for clinical trials
Some areas that present challenges for advancing the use of AI in medicine and clinical trials:
- Electronic health records (EHRs) are not universal in their format or quality
- Sometimes data is siloed in medical imaging or pathology systems in addition to the EHR
- Regulations surrounding the use of data differ globally
- Bias exists, for example, currently available genetic data is mainly from patients of European descent
- Ethics committees and laws need to keep pace with the changes and advancements in AI use
With time, diligence, and cooperation, these challenges can be reduced, and the use and applicability of AI in clinical studies will increase.
Technology, just like money or beauty, can be used to benefit or to exploit. With AI, let’s keep one eye on the dire predictions of those imaginative writers, poets, and moviemakers, and another eye on the potential for good.
For now, it is up to us to drive the technology we develop and use to right past wrongs and improve the overall health and lives of others. Our investment choices today will drive the reality of our future.
I’m sure that when applying the use of artificial intelligence for clinical trials, if we keep patient safety and overall well-being top among our priorities, we’ll fair well. What are your thoughts? Scroll down to the comment box to let me know what you think. Also, check out our blog about using machine translation for clinical trial materials.