Did you know that you’re likely using artificial intelligence (AI) in your everyday life?
For example, the digital music service, Spotify, creates “mood-based” playlists that are curated to users’ musical preferences. Spotify generates these customized play lists using a machine learning algorithm that has learned your unique musical preferences based on your previous interactions with songs, musicians, and playlists.
AI technology is driving innovation for multiple industries including pharma. In this blog post, I’ll discuss how using AI for regulatory writing is reshaping drug development.
The ROI on AI technology
The timeline for launching a drug to market typically involves a decade of discovery and pre-clinical research followed by another eight years for clinical trials. AI could streamline that process dramatically by cutting time and costs spent on clinical trials. In this respect, investing in AI technology could yield a significant return on investment (ROI).
AI: from skepticism to enthusiasm
Until recently, pharmaceutical companies employed structured authoring to streamline document writing. Structured authoring defines the structure of a document and what content should go in each section.
But all the money and time invested in structured authoring hasn’t provided a sustainable solution because of the industry trend towards mergers and acquisitions. Each time one company acquires another, you end up with reports in many different formats. The resulting document heterogeneity wipes out any efficiency that structured authoring provides.
Currently, the main uses of AI are in basic drug development research. But there is a renewed desire to use these advances in late stage development.
With the use of AI machine learning in helping us write regulatory documents to smart objects that redact sensitive data for publication, we are moving from relying on structured content and methods to contextual-based understanding. These techniques will only get more sophisticated from smart, rule-based objects to learning objects that can adjust approaches and interpretation. So instead of using AI-configured smart objects to develop specific clinical reports, we will move towards the sole use of learning objects that automatically interpret the type of input data to develop the appropriate output reports. It won’t matter if we insert a SAS dataset for narrative generation, protocol and SAP for clinical report writing, or full clinical study report (CSR) and submission documents. AI will interpret the input and self-generate full narratives, complete study reports, or a redacted sets of reports for publication.
Our capabilities to merge structured and unstructured data and the ability to use datasets, reports and external sources to seamlessly cross check or enhance your internal analysis will revolutionize clinical development. The benefits of this approach include smarter data interpretation, faster data manipulation, and more efficient and cost effective generation of the outputs needed to support getting drugs to market.
AI crunches research time
Unlike humans, AI can process huge amounts of data and find and manipulate valuable information. It can interpret contextual information and use natural language processing to combine phrases and statements to understand user’s commands or self-interpreted decision trees. This ability combined with business and writing rules enables AI tech to generate draft regulatory writing documents.
We started using AI for CSR writing. CSRs are enormous reports that comprise part of the submission package. Writing these documents is labor-intensive and tedious. Much of the effort in writing a CSR involves identifying information in previous study documents and putting it in the right tense.
These mundane activities don’t utilize the scientific knowledge and talent of your medical writing team. AI technology can expedite CSR writing by taking information from previously authored study documents (the trial protocol; the statistical analysis plan; and tables, listings and figures) and putting it into the right places in the CSR. Like a person, AI understands the context of information in study documents and interprets where it belongs in each study report section. Our AI system also evaluates data in tables to create fact-based, non-interpretive results text. Using this technology can automate up to 80 percent of time spent writing CSRs. Now, the medical writers are freed to focus on the parts of the CSR that require higher level scientific interpretation.
The purpose of AI is to aid medical writers, not to replace them. Without AI technology, you could spend weeks just generating the CSR draft. AI tech can generate a draft report in 24 to 48 hours. Then, the writers only have to complete the final 10 to 20 percent of effort. This time savings can help your submission be the first to market. Accelerating your marketing authorization ultimately impacts how much potential revenue an asset can generate.
Using AI to support transparency and disclosure activities
In addition to expediting document writing, AI technology is also being leveraged for redacting sensitive information from clinical trial documents. This application is booming with the emergence of EMA Policy 0070. Every pharmaceutical company submitting to the EU must comply with this requirement to publish CSRs and summary reports while ensuring that they don’t risk re-identifying any patients or study administrators. Sponsors must accurately and consistently redact personal protected data (PPD) from these documents.
And that’s where AI technology provides tremendous value. This technology understands the context of the sensitive information in clinical documents as well as business rules to identify and redact PPD as well as support the process to redact company confidential information (CCI).
AI also provides greater accuracy and consistency than manual approaches. When it comes to protecting patients’ privacy, good is just not good enough. The liability of accidently exposing even one patient is huge.
Unlike conventional manual approaches, using AI for redaction is a scalable solution.
When we first started using AI for redaction, sponsors were spending over six months to redact just four documents. Currently, we’re working with one sponsor and redacting 50 documents per week. In total, we are redacting 100s of documents per month across our sponsor base. Achieving that level of productivity and consistency with a manual effort is impossible.
AI tech helps optimize resource allocation
Entire study teams review the data to be redacted from clinical trial documents. For example, medical writers are typically central to the redaction process. However, having highly trained medical writers spent inordinate amounts of time manually removing information from thousands of pages of documents is a poor use of this resource.
Likewise, the legal team often defines CCI. Again, manually identifying CCI is an inefficient use of this high-value resource. The reason you have these high-value resources working on redacting documents is because the impact to the organization is so significant. And that’s where AI really provides value: it automates much of these manual processes. Thus, the impact on these resources for achieving compliance with regulations like Policy 0070 is minimized.
Bigger players changing the technology landscape, tools and infrastructure
Over the last few years, the big technology giants have invested heavily in AI. This is exciting news as new tools, techniques, and data infrastructure become more readily available for pharma to use.
Microsoft is currently researching “automated reasoning, adaptation, and the theories and applications of decision making and learning.”
According to CB Insights, Google― the most prominent global AI player― has completed five acquisitions in the space since 2013.
The tech giant, which acquired London-based AI start-up DeepMind in 2014 for £400m, is exploring different aspects of machine learning including deep learning and neural networks.
Steve Wozniak, Apple’s co-founder acclaimed AI’s transformative potential during an innovation summit in Brisbane, Australia. Quoted by the Sidney Morning Herald, Wozniak said, “Until recently … artificial intelligence really didn’t make much difference in life, but now we’re getting to the point where we’re getting closer to what the brain is.”
He concluded, “I looked at the brain my whole life thinking we would never understand how it’s wired, never know what consciousness is, we would never know what intuition is. And now we’re seeing so many signs that are getting so close – we speak to our phones, we can get answers.”
Amazon became the latest tech giant to give away some of its most sophisticated technology by unveiling DSSTNE, an open-source AI framework that runs its recommendation system.
The news comes after a Wall Street Journal report claimed Amazon was “boosting its artificial intelligence chops” last year.
According to the article, Amazon had hired AI developers in Europe and data scientists for its New York and Berlin offices.
Like its peers, Amazon has also made AI-tech acquisitions such as Silicon Valley-based Orbeus, a recognized API focused on visual recognition technology using deep learning.
Watson, IBM’s AI computer system famed for beating some of the world’s best chess players, can answer questions posed by humans. Developed by IBM’s DeepQA research team, the tech giant announced it would use Watson to solve cyber-crime “once and for all.”
IBM is now expected to spend next year collaborating with eight universities to teach Watson to detect potential cyber threats.
Learn how AI technology can help you meet T&D mandates
Our regulatory and medical writing AI solution meets the promise of automated authoring documents such as patient narratives. This is also the most effective and efficient approach for meeting data transparency requirements. To learn more about how AI can streamline transparency and disclosure, please read this case study!