AI: a tool to optimize the design and facilitate the conduct of clinical trials

Published on 27 April 2020 Read 25 min

Approximately 14% of drugs entering the clinical phase receive marketing authorization [1]. Although this figure seems high, as it was estimated at only 9.6% in 2016, the success rate of clinical trials remains quite low. Poor study design (selection of endpoints, choice of sub-populations, protocol design, etc.) as well as difficulties in providing meaningful results may be among the reasons for trial failure.

The costs associated with this high failure rate and with the long duration (up to 10 or 15 years) of clinical studies have made pharmaceutical companies turn to new solutions for studies design and roll-out. Thanks to the digitization of clinical studies’ results and to the data sharing initiatives, the development of Artificial Intelligence (AI) tools in the field of clinical studies is now possible. At Alcimed, we asked ourselves how AI can improve clinical development in the pharmaceutical industry, and we focused on the three most advanced use cases.

Faster and more efficient protocol design thanks to AI

In order to efficiently conceive a clinical study, several elements must be defined, such as the problem, the target patients’ profile, the sites where the study will be conducted and the number of patients to be included. Analyzing the results of previous studies can guide these choices. However, since the sources of information on clinical studies are numerous, defining these elements can be laborious.

Natural Language Processing (NLP) methods are AI tools that extract essential information from unstructured documents, such as medical reports. By applying NLP to databases of clinical studies and real-world data, it is possible to better design protocols for new studies.

Another potential use of AI in protocol design was developed by the Massachusetts Institute of Technology (MIT), which used forced feature-learning, an algorithmic method based on agent learning thanks to receiving rewards for success and penalties for failure, to determine the dose of chemo and radiation therapies to be administered to patients in a clinical study.

Patients recruitment facilitated by AI

The lack of patients involved in a clinical study is sometimes the reason for its failure, while often not all patients eligible for a clinical trial have access to it. According to an article from the American Cancer Society, less than 5% of adults eligible for a cancer clinical trial participate [2].

AI could address these problems of identifying, and subsequently recruiting, eligible patients for a study by providing tools to analyze patient information faster. The Cincinnati Children’s Hospital Medical Center has set up an AI system called ACTES (Automated Clinical Trial Eligibility Screener) based on NLP to identify patients meeting the criteria of a study 34% faster than with a manual method. This approach has also made it possible to diversify the profiles of recruited patients [3].

Several companies already offer the reverse service in order to facilitate the access to clinical studies for patients. Such applications as NAVIFY Clinical Trial Match, developed by Roche, or MyStudyWindow, launched by Boehringer Ingelheim in March 2020, search for clinical studies for a given patient by using NLP to analyze several parameters, such as genomics.

Better follow-up of clinical trials thanks to new data sources

In the course of a clinical study, patients are usually required to travel to the study site to be followed-up by qualified personnel. The use of connected mobile devices, such as connected watches and mobile phones, allows patients to reduce the frequency of visits to the study site, the disruption to their daily lives, thus making patients more motivated to stay in the study until the end. For example, Janssen has recently launched a clinical study on the impact of heart failure on the health and quality of life of patients, whether or not combined with type 2 diabetes, which will use patients’ personal mobile phones to retrieve information remotely.

An additional benefit of using connected devices is that they provide real-time data. With AI methods, it is possible to process this data as it is collected, in order to detect particular events, such as medication discontinuation or an adverse event with the drug, that may occur during the study. However, the reliability of the data from the connected devices must be guaranteed in order to ensure the validation of all results by the health authorities.

Other uses of AI are possible in this context, such as chatbots that allow patients to ask questions and get answers instantly, as well as to communicate the adverse effects of a drug and alert the medical team more quickly. There are also other innovations such as the assistant developed by AiCure in order to verify the correct drug intake based on a video of the patient taking his or her medication.

To sum it up, the use of AI makes it possible to speed up the design of clinical study protocols, to carry out better patient recruitment, to ensure more personalized follow-up of each patient, and to retrieve data throughout the study. These AI use cases are important in today’s context where clinical studies can last more than a decade and failures can cost billions of dollars.

Although these applications are promising, their use should be considered with caution. For example, the number of digitized studies may be insufficient for the construction of the protocol of a new study, the analysis of data from previous studies may introduce biases, while the reliability of the data provided by connected device sensors may be questionable.

It is then up to each pharmaceutical company to evaluate the opportunities related to the use of AI in clinical trials, as well as to contribute to the increase of the amount of available data.

About the author

Amélie, Head of the Data activity in Alcimed’s Healthcare team in France
Axelle, Data Analyst in Alcimed’s Healthcare team in France

[1] Hale C. New MIT Study Puts Clinical Research Success Rate at 14 Percent | 2018-02-05 | CenterWatch. Cent Watch. 2018.
[2] The Basics of Clinical Trials. Am Cancer Soc. 2020:1-7.
[3] Ni Y, Bermudez M, Kennebeck S, Liddy-Hicks S, Dexheimer J. Designing and evaluating a real-time automated patient screening system in an emergency department. J Med Internet Res. 2019;21(7). doi:10.2196/14185

You have a project?

    Tell us about your uncharted territory

    You have a project and want to discuss it with our explorers, write us!

    One of our explorers will contact you shortly.

    To go further