The true cost of developing new drugs has been debated for the last 40 years. A broadly accepted number for developing a new drug is $1-2 Billion and 10+ years. But does it really cost so much? Similar things have been said about the cost of developing Artificial Intelligence (AI)-chatbots like ChatGPT until DeepSeek showed publicly how an identical chatbot could be created for a fraction of the broadly accepted cost of such tools. But is it fair to compare the cost of new drugs with that of an AI chatbot?
OpenAI spent about $540 million to create ChatGPT and it costs about $700,000 a day to run, according to publicly available information. Most of the development and maintenance cost is due to the immense computing power needed to train the large language model (LLM) on a massive dataset requiring very expensive hardware and engineering expertise. DeepSeek was created using a creative workaround that needed far fewer resources (allegedly about $6 million) and yielded a product that gives similar outputs.
New drugs go through a comparable scenario. The bulk of the cost of developing new drugs is for clinical trials that require expensive resources and expertise. The high cost estimates deter many developers from even initiating the translation of research to clinical products and hence many discoveries never even see a clinic. Can new pharma developers take some lessons from the developers of DeepSeek? Can a new drug be developed at a fraction of the $1-2 Billion cost?
Yes, it is possible to reduce or defer the overall cost of developing a new drug using creative out-of-the-box thinking and project execution. Since the bulk of the cost of developing a new drug is for conducting clinical trials, this is where the focus of cost reduction should reside. Analysis by some groups has shown that the total cost of clinical trials could be far less than generally accepted. The cost of a clinical trial includes three major areas; the design of the trial and logistics, patient handling, and data analysis.
A very common error made by sponsors is the poor design of their trials. Several smart trial designs such as adaptive clinical trials, basket trials, enrichment trials, personalized or targeted designs, and others could de-risk the clinical program and save on the overall cost and time. For logistics, an FTE and virtual models for skills and resources could save anywhere from 20-40% of the cost of each trial. Patient handling costs are primarily attributed to the high cost of renting clinics and medical personnel for conducting the trial, mostly due to the scarcity of available sites. Since only about 2% of physicians in the US participate in clinical trials, there exists an opportunity with the remaining 98% of physicians to expand the site availability and access bringing down its cost and competitiveness. Data analysis and regulatory strategies can minimize the compliance and regulatory acceptability risks. Together these strategies could do to the costs of new drug development what DeepSeek did to ChatGPT.
We have been told over the last couple of years how expensive it is to create tools such as ChatGPT, that it takes years and costs hundreds of millions of dollars, till a couple of weeks back when a Chinese company with a ragtag group of young programmers released DeepSeek, created for about 1% of what it cost OpenAI to create ChatGPT. There have been questions raised about the cost to develop DeepSeek, but there is no question that it was far lower than what it cost to develop ChatGPT. To its credit, for a small company in China, the goal of beating ChatGPT may have sounded ridiculous at the onset. Similarly, new small drug developers often get intimidated by the projected cost of their programs. A look under the hood, matter of fact, of the cost attributes could change that. We have been told for decades that new drugs can only be developed by those with very deep pockets. We need to think outside the box to find ways to “DeepSeek” new drugs.
