AI and automation in clinical trials

Artificial Intelligence (AI) is the future of clinical research. This is a common phrase, but what does it really mean? You’ve all heard about AI being used in basic research to identify molecules and find disease patterns in potential patient groups. Machine Learning (ML), a branch in AI, deals with the application of algorithms to data. This allows the system to ‘learn and improve.’ ML allows users the ability to analyse large amounts of data, make intelligent inferences and predict future outcomes. This information can be used to automate certain parts of the system, resulting in a faster and more efficient clinical trial system. Automating allows ML predictions to be fed back to the system, and specific actions taken. This reduces the need to human intervention and improves quality and speed. Automation and ML can be used at every stage of the trial process.

Study design
ML can be used to design protocols and translate languages. The system can generate a protocol for a new research using the existing protocol data and health library information for specific therapeutic areas. The ML algorithms could design the optimal protocol using existing knowledge, resulting in shorter design times, protocol amendments and study disruptions. The ML model would also have a domain-specific language knowledge base that can be used to translate languages. This could make it faster and more accurate than traditional methods.

Study setup
ML can be used for automating the creation and maintenance of the study database and case report forms. The protocol can be used to train the ML model to create the optimal CRF. This output can be converted into study setup and validation using automation, which allows database designers to make adjustments as needed. This allows for an optimised design that also includes edit checks, which might not be possible if the design was created by humans. This ML-designed study can also be automated and validated. Designers can use the validation report to help them finish the work before going live. ML can be used to create SDTM-annotated study or automate SDTM mapping.

Trial management
Trial management can be automated using ML. Site selection, patient enrollment, Risk-Based Monitoring (RBM) and chatbots are just a few of the many uses that this technology can be used for.

Site selection: ML models can be used to optimise site selection. These models can be used to analyse site parameters like enrolment and safety, compliance, data quality and make predictions about which sites are good candidates for new studies in a specific specialty. These parameters are prioritised based on the type of trial and sponsor. This algorithm could be trained using data from previous studies and be able to predict the performance of a site for a new one.

Patient enrollment: Predictive analytics for patient enrollment is a very popular use case. Variables such as study duration, disease prevalence (from health economics), study difficulty, adverse events and randomisation are all used to calculate this. The ML algorithm would evaluate all variables and identify the ones that have the greatest impact. The model could be used in future studies to predict patient enrollment. This is a very popular use case. However, it is difficult to predict patient enrollment due to the large number of factors.

Risk-Based Monitoring: RBM can be used at different stages of a clinical trial to identify and reduce risks. One type of RBM uses some of the factors of site selection such as enrollment safety compliance and data quality. Other variables include therapeutic area, multi-centricity, trials, etc. to predict the performance of a site during a clinical study. These predictors can help to reduce the risk of the trial by helping to identify potential risks and to work with them to mitigate them.

Chatbots: Chatbots are one of the easiest and most straight-forward examples of machine learning. These can be used for different purposes – patients, site users and CRO staff. These can be contacted by voice or text. They understand natural language and context, and are able to provide accurate replies. This increases user experience and decreases support staff burden.

Data management
Data management has tremendous potential for AI-enabled automation. Here are some of the most important:

Smart queries: The ML algorithm in smart querying reads trial data and determines possible queries that can be raised for different field items. This is possible by combining previous study data with the therapeutic area. If the algorithm identifies an error, it will raise a query. The query is then reviewed by a data manager, who either qualifies it as a valid query or discards it. This decision is used to help the ML algorithm improve its classification.

Medical coding: Medical terms can be coded automatically using regular programming up to a certain amount. Above that, a medical coder must review the data and manually code any remaining terms. The coding libraries of various therapeutic areas can be used to teach ML algorithms. They will then be able to match the required text with the correct term in the dictionary for the specialty. ML is capable of this kind of coding with high accuracy.

Queries management: Many queries are raised in clinical trials, and it takes a lot of time to respond to them. Many of these queries are redundant and arise because of misconfigurations in the EDC edit checks. These queries can be identified by machine learning and can be managed in bulk. Or, appropriate edit checks can also be set up mid-study to address the problem. ML uses clustering to identify queries that can be grouped together and the issue. These clusters can be handled in bulk.

Smart DDV: Organisations invest a lot in trial monitoring. To monitor the study, CRAs need to travel to the sites and do source data verification. All of these manual tasks can be greatly reduced by ML. Site staff can upload images of source documents to the server. These images can be extracted by machine learning algorithms and sent to the EDC. If there is a match, the EDC compares the data with the input data and marks them as verified source data. If not, the EDC raises a question that would need to be manually verified.

Data analysis
ML can offer many insights into clinical data, both during and after a trial. Data analysis can provide critical insights into large data sets by using classification, prediction and clustering, patient behaviour, adverse events, etc. ML can predict patient behaviour, adverse events, etc.

Regulatory submission
A lot of documentation is required for regulatory submissions in clinical trials. These documents can be templated and automated with ML.

CSR automation: The clinical study report can automatically be generated using ML. This is possible by reading the Study Protocol (SAR) and the Study Analysis Reports (SAR). Most of the CSR can easily be generated using ICH GCP templates. Natural Language Processing (NLP), algorithms can be used for changing the language of the CSR. They can also be used in the generation of narratives. The medical writer can then review and edit these to produce the final CSR. This can all be done in a matter of days. This reduces the time it takes to submit a regulatory submission and increases submission quality.