Artificial Intelligence in Clinical Trials: Balancing Innovation and Accuracy

The use of artificial intelligence (AI) has become pervasive throughout different fields from agriculture to finance, and healthcare is no exception. While AI has several usages in clinical settings and public health, such as diagnostics, robot-assisted surgery, patient medication reminders, and COVID-19 case surge prediction, here we focus on areas in which AI tools can be used to facilitate clinical research.

Risk Prediction

The development of machine learning (ML) models has enabled professionals to perform risk prediction for events such as trial failure. The process starts with data engineers training ML models on data from trials that were performed in the past, the information for which is publicly available on repositories such as ClinicalTrials.gov.

The advantage of ML models over traditional statistical models used for risk prediction is that ML models can be taught on both structured data, which have “simple” answers (e.g., the number of countries in which the trial was conducted or the funding source) and unstructured data, which are complex (e.g., trial protocol eligibility criteria).

Another AI-driven tool—natural language processing (NLP)—is used to extract unstructured data in a usable format for the ML model. The model can then be shown a new study and make a prediction based on thousands of factors on which it was taught.

Finally, if a trial is predicted to fail, interpretability methods can be used to visualize the factors that the model labeled as the most important contributors to failure in this specific trial, opening possibilities for protocol alterations in order to decrease the chances of this unwelcome scenario.

Patient Engagement

Another promising usage of AI in clinical trials is the use of chatbots for trial recruitment and patient engagement.

Patient recruitment has traditionally been the pitfall of clinical trials, with most prematurely terminated trials citing it as the main reason. Chatbots can help improve trial participation and retention in several ways, including:

• Facilitating the informed consent process
• Sending personalized reminders about upcoming visits
• Providing educational materials and counseling prior to visits

These abilities make the clinical trial process more comfortable for patients, as they have a constant opportunity to have their questions answered and their worries relieved. Use of AI reduces the burden on the healthcare workers as well, leaving more time for detailed physical exams and similar medical duties.

Medical Writing Support

The rise of generative AI has opened opportunities for accelerating clinical trial initiation and conduct by speeding up the exhaustive protocol writing process. The protocol is integral in that it consists of all the necessary information for conducting a specific trial (such as the endpoints being assessed, eligibility criteria, dosing, visiting schedules, etc.).

The protocol is usually dozens of pages long, and AI can significantly speed up the writing process, drafting the initial document in a matter of minutes based on the investigators’ input.

However, confidentiality remains a critical barrier, as protocols contain highly sensitive information that can be exposed when input into online AI systems. Organizations must carefully weigh the efficiency gains against the risks of compromising competitive advantages through external AI platforms.

Medical Scribes

Careful oversight and the right expertise are necessary for AI to be used to its full potential. For example, the use of AI medical scribes for documenting patient encounters in clinical practice in real time has been tested with conflicting results.

On one hand, AI scribes have the potential to relieve a lot of the clinicians’ burden, leading to less manual data input time during patient care, which patients find an improvement. On the other hand, these are still error-prone tools—for instance, they can hallucinate, leading to confusion when sentences no one spoke appear in the transcript.

Critically, omissions are the most frequent errors introduced by these tools, leading to questionable data quality and integrity. Since it is the clinician’s responsibility to ensure the accuracy of the medical documentation, AI scribes might finally lead to increased burden due to data cleaning being required on the clinician’s part.

Conclusion

While AI offers tremendous potential in the landscape of clinical trials, the challenge of data integrity remains universal. Whether it’s medical scribes introducing transcription errors, predictive models trained on biased datasets, or AI writing tools generating plausible but inaccurate content, data errors don’t magically disappear when AI steps in. In fact, AI can introduce new types of errors if not properly implemented.

This means that successful AI integration in clinical settings requires more than just adding AI tools as an afterthought. Clinical data management systems need robust validation frameworks, semantic technologies for better AI communication, and intuitive interfaces that support human oversight.

A truly user-friendly system for data management of clinical trials can apply AI-driven tools like NLP and automated data validation to support clinical workflows, ensuring fully structured data. Such systems should be built on flexible models and semantic technologies that enable seamless AI communication.

The next era of clinical research will be driven by AI systems designed not only to automate tasks but also to enhance data integrity and user experience. By embracing these intelligent, interoperable solutions, researchers can accelerate breakthroughs while ensuring trials remain both efficient and patient‑focused.

Resources

Agarwal P, Lall R, Girdhari R. 2024. Artificial intelligence scribes in primary care. CMAJ 196(30):E1042.

Askin S, Burkhalter D, Calado G, El Dakrouni S. 2023. Artificial Intelligence Applied to clinical trials: opportunities and challenges. Health Technol 13(2):203–13.

Biro J, Handley JL, Cobb NK, Kottamasu V, Collins J, Krevat S, Ratwani RM. 2025.
Accuracy and Safety of AI-Enabled Scribe Technology: Instrument Validation Study.
J Med Internet Res 27:e64993.

Cihoric N, Gavric S, Dennstädt F, Jovanovic A. 2025. A Multi-Factor Machine Learning Model for Predicting and Preventing Clinical Trial Failures (preprint available at Research Square). https://doi.org/10.21203/rs.3.rs-5932404/v1

Foote HP, Hong C, Anwar M, Borentain M, Bugin K, Dreyer N, Fessel J, Goyal N, Hanger M, Hernandez AF, Hornik CP, Jackman JG, Lindsay AC, Matheny ME, Ozer K, Seidel J, Stockbridge N, Embi PJ, Lindsell CJ. 2025. Embracing Generative Artificial Intelligence in Clinical Research and Beyond: Opportunities, Challenges, and Solutions. JACC 4(3):101593.

Mohanty N, Yang TY, Morrison J, Hossain T, Wilson A, Ekong A. 2022. Chec-up: a digital intervention to reduce disparities in well-child and immunization completion in community health. Telehealth Med Today 7(5).

This article originally appeared in the June 2025 issue of Clinical Researcher, the official journal of the Association of Clinical Research Professionals (ACRP), and is presented here with permission from ACRP.

Written by Aleksa Jovanovic, MD, PhD, Scientific Engagement and Innovation Specialist