The implications of ChatGPT for pharma

The State of the Biopharmaceutical Industry report revealed that 39% of healthcare professionals see AI as the most disruptive emerging tech in 2023. ChatGPT is one such aspect of artificial intelligence, disrupting our discussions if nothing else. The buzz that ChatGPT is generating is on par with the vast reams of content it’s producing as we excitedly test it out to see its capabilities. Undoubtedly, it has implications for pharma, but what are these away from the hype?

Sam Altman, CEO of OpenAI (the company behind ChatGPT), tweeted in December, “ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness.” So should the pharma industry, reliant on non-negotiable ethics, regulations, and reliability touch it with a barge pole?
Here I take a non-ChatGPT-generated look at the buzz to reveal the likely implications of large language models (LLMs) like ChatGPT in the pharma industry.

What is ChatGPT in the context of the pharma industry?

GPT stands for Generative Pre-trained Transformer. This is a type of large language model (LLM) – a machine learning system (which is a type of Artificial Intelligence) that takes the reams of data it’s trained on and produces text outputs. In 2020, OpenAI first released GPT-3; in late 2022, the San Francisco company released ChatGPT, a type of this AI that is free.

Anyone who has a play with ChatGPT is bound to marvel at it like a child in a toy store. It is enticing and certainly flags up the potential to significantly impact the pharma industry in multiple ways.
However, the potential of ChatGBT faces hurdles, hindrances and warnings. Let’s look at the potential implications of ChatGPT for pharma and the issues presented in these areas.

The ways that ChatGPT is relevant to the pharma industry

1. Drug discovery and development

ChatGPT can analyse vast amounts of data from research papers, clinical trials, and other sources to identify patterns and insights that could inform drug discovery. It can help identify new drug targets and predict the efficacy and safety of potential drugs, reducing the time and cost of the drug discovery process. Drug discovery and development is a slow and painstaking process, so an Artificial Intelligence (AI) tool that speeds up elements of the process would be welcome.

The issue: Clean, accurate and complete data is required and chatGPT does not always provide accurate summaries of data. This is likely to improve but currently is not reliable.

2. Clinical trials

Again, through its power of analysis and ability to identify patterns, ChatGPT could help improve clinical trial design and patient recruitment and create more effective and efficient clinical trials. There are several ways in which ChatGPT may increase effectiveness and efficiency in clinical trials:

· Predictive analysis: With their predictive analysis capabilities, GPT models can analyse vast amounts of patient data, identifying potential risks and predicting treatment outcomes.

· Efficient data management: By automating the processing of large amounts of data, clinical trial staff can focus on other essential aspects of their work while generating results more quickly.

· Improved patient engagement: By utilising GPT-powered chatbots, patients can receive personalised information and support during their clinical trial journey, leading to increased patient engagement and, ultimately, improved trial quality.

· Automated reporting: GPT models can potentially generate reports from the data provided from the clinical trial. This frees up researcher time for other, less admin-heavy tasks.

The issue: Accuracy is vital in clinical trials, and even the CEO of ChatGPT points out that it is inaccurate. This is likely to change but for now I would recommend caution. It can speed up processes but would require accuracy checking.

3. Pharmacovigilance

ChatGPT holds promise for drug safety monitoring, a complex and laborious process. It could potentially be used to monitor adverse drug reactions (ADRs) and identify potential safety concerns. It may help identify drug interactions and possible side effects (if it has accurate non-biased data), helping healthcare providers make more informed decisions about prescribing medication.

Because they are text-trained, GPT models hold the potential for the level of text-mining required in pharmacovigilance. Artificial Intelligence (AI) could review case reports to find ADRs in seconds in a way that would take a human far longer. It’s possible that these models could manage the issue of causality more easily than a human.

The issue: Pharmacovigilance’s mainstay is in its name: vigilance. The accuracy and fullness of the data are paramount and inherent biases make this impossibly complex.

4. Patient care, customer service and engagement via chatbots

Modern medical systems are fighting a battle to meet individual patient engagement needs, and existing chatbot technology falls woefully short as a nurse and physician. ChatGPT could revolutionise this. Virtual assistants developed by ChatGPT can provide patients with personalised care, including medication management and symptom tracking. This could improve patient outcomes and reduce the burden on healthcare providers.

There’s also some potential for GPT to be useful for diagnoses and treatment advice. For example, Google and DeepMind are already conceiving Med-PaLM, a medical query tool.

The issues: The data must be accurate for accurate output. Additionally, as Chris Stokel-Walker and Richard Van Noorden in Nature point out, “without output controls LLMs can easily be used to generate hate speech and spam, as well as racist, sexist and other harmful associations that might be implicit in their training data.” This has no place in healthcare. Let’s also remember that this is healthcare, where we must recognise the power of human connection in patient outcomes. At the moment the technology underperforms compared with human clinician in a few areas (including incorrect retrieval of information, incorrect reasoning, and incorrect comprehension) – but that can change in time.

5. Regulatory and medical affairs

ChatGPT models offer some promise for regulatory affairs regarding regulatory document analysis (such as drug approval submissions), predictive modelling for regulatory processes, and automated reporting.

The issues: There are already emerging regulatory concerns about the use of GPT models, akin to the legal disputes already being raised over copyright issues in digital art due to the data used to train these models. Within pharma, regulatory matters are even more expansive and raise pertinent ethical questions.

6. Medical writing

There is scope for ChatGPT models to take over the labour-intensive task of some medical writing. As a text generator, ChatGPT creates unique passable prose on various subject matters. The benefit of using ChatGPT for medical writing is that it frees up expert time for other high-value tasks.

The issues: There are valid questions about authorship and integrity. Additionally, texts produced by ChatGPT sound plausible. It can even come up with references and citations, but these are fake, as pointed out in an article in Nature). The problem is that it could be unsubstantiated bunkum delivered with such confidence that it dupes the audience.

7. Sales and marketing

ChatGPT is likely to change the landscape of sales and marketing significantly, which applies as much to the pharma industry as elsewhere.

With clean and accurate data, GPT models can undertake market research more quickly and accurately than humans. ChatGPT can write a passable blog article on basic topics or other marketing materials. It can be used as a sales channel chatbot to field enquiries with personalised responses or can even write a good enough pitch personalised to a particular company. It does a plausible job. It can help automate and streamline time-consuming low-skill elements of sales and marketing processes.

The issues: ChatGPT can only draw on what already exists in its data. It’s also a tool that is only as good as the human using it. It can pretend, but it can’t be. Furthermore, effective market research requires excellent questions created within context and objectives.

The future of ChatGPT in pharma

If I let ChatGPT come up with a conclusion for this article, it tells you:

“Overall, ChatGPT has the potential to revolutionise the pharma industry, improving drug discovery, patient care, clinical trial design, drug safety monitoring, and customer service. Its ability to analyse vast amounts of data and provide personalised insights may transform how drugs are developed, tested, and prescribed, ultimately improving patient outcomes and reducing healthcare costs.”

And that is all true. But it’s not the whole picture.

ChatGPT’s confidence undermines the truth of the situation and my conclusion. GPT models will have their place. But they will only ever be as effective as the data they are trained on, which, in reality, is full of bias and pitfalls. Regulation has no precedence but needs addressing, not least because when it comes to the pharma industry, so many of the above implications ultimately impact in life or death ways. So it’s no surprise that ChatGPT itself will now caveat responses more and more with warnings and disclaimers.

Artificial intelligence (AI), and iterations of GPT models, will revolutionise multiple aspects of pharma, without a doubt. Indeed, using GPT models like ChatGPT could be an excellent first step for many within the industry to open their eyes to the potential that lies ahead. 

But it’s not there yet, and we need to take a questioning approach that ensures we use it ethically, reliably and safely, including dealing with the conundrum of regulation head-on.