The successful development of a new drug takes 10-15 years and costs US$1.5-2.0 billion. Clinical trials (CT) consume almost half of the time and cost. Today, high failure rates of CTs are a major obstacle in successful and fast drug development.
Only 10% of new compounds entering the CT phase advance to the regulatory marketing approval stage. Over 86% of clinical trials do not meet timelines for subject enrolment.
Major challenges in the successful completion of CT are the recruitment of patients and finding the technology to manage the complexity of running a CT. Complex selection criteria and an increasing burden of protocol procedures have increased screen failure rates from 25% to 41% and study completion failure rates from 31% to 48%.
Artificial Intelligence (AI) has the potential to help in overcoming the limitations of the current clinical trial process. Machine learning (ML) can automatically detect patterns of meaning in large databases of text, voice recordings or images. Natural language processing (NLP) can comprehend and correlate diverse content, e.g., written or spoken language. Such AI capabilities can be useful in analysing and correlating large and diverse medical information from electronic medical records (EMRs), medical literature, and clinical trial registries/databases for improving the process of matching patient data–trial design and planning recruitment strategy prior to starting of CT. AI can also facilitate CT conduct by regular monitoring of CT participants. This can improve adherence to protocol requirements, encourage retention of CT participants, and lead to more reliable and efficient assessment of efficacy and safety endpoints.
The application of AI for CT relates to the exploitation of EMR and clinical trial data and employing a variety of high-tech approaches to oversee the CT process. A diversity of sources, and the veracity and volume of EMR data make a comprehensive analysis by AI quite challenging. EMR interoperability dilemma as well as concerns about the protection of sensitive health data is a major stumbling block in the implementation of AI for CT. In India, detailed documentation of medical history and treatment is frequently missing. Dr Handa’s review of 28 rheumatoid arthritis studies revealed that only 25-35% of the studies described the epidemiology, comorbidities, extra-articular manifestations, functioning abilities and the quality of life among patients, and provided information on treatments. Such gaps in paper or EMR records won’t be helpful in developing AI solutions.
In CTs, patients often drop out due to a variety of reasons, e.g., the burden of protocol procedures to capture efficacy/safety data – symptoms, adverse events, drug intake, patient diaries and follow-up visits. AI, coupled with wearable sensors and video monitoring, can obtain real-time individual patient data and lessen the patient’s burden. However, wearable technology and devices face the challenge of interoperability and standardization of data and methodology. In India, illiteracy would be a major challenge in the deployment of such devices.
The utility of AI technology in CT is yet to be proven and validated in a comprehensible, ethical, replicable, and scalable way for pharma companies and regulators. Unless the industry is willing to invest time — 5 to 8 years — and money in developing this novel endeavour, AI will not be able to contribute towards improving the efficiency of the drug development process.
Writer is a consultant on clinical research & development from Mumbai.