A substantial portion of drug discovery projects do not progress beyond the clinical trial stage. In a recent review published in Nature Genetics, researchers from various institutes at the Wellcome Genome Campus in the United Kingdom used a natural language processing model to examine over 28,000 clinical trials that had been terminated before achieving the endpoint to determine the cause of trial discontinuation and investigate the role of genetic information in the success of clinical trials.
Study: Genetic factors associated with reasons for clinical trial stoppage. Image Credit: create jobs 51/Shutterstock.com
Background
The failure rates in the drug discovery process are often high due to safety issues associated with adverse reactions, and a lack of efficacy, which contribute to approximately 79% of the failed clinical trials.
The systematic assessment of available evidence through clinical and research methods is being adopted in many of the new drug development approaches in the industry to increase success rates.
Furthermore, clinical trials for therapeutic hypotheses supported by human genetics findings have been associated with higher success rates.
While it is essential to understand the underlying causes of clinical trial failures, factors such as biased reporting of positive results and the absence of published negative results from clinical trials make it difficult to understand the factors that impact efficacy and safety profiles.
About the Study
In the present review, the researchers trained a natural language processing model to classify the reasons for discontinuing a clinical trial. Then they used the model to classify close to 29,000 discontinued trials.
This information was then integrated with evidence on disease and drug targets obtained from the Open Targets Platform to determine the reasons for trial discontinuation.
The study used clinical and genetic data from public resources such as ClinicalTrials.gov, Open Targets Platform, and ChEMBL. A long short-term memory network was then trained to classify the various reasons for trial discontinuation, and then agglomerative hierarchical clustering was used to merge semantically similar classes. Human annotations were performed to validate the model.
Clinical trials were collated and classified, and the clinical indications and genetic traits from these trials were harmonized using Experimental Factor Ontology or EFO. The ChEMBL database was then used to map the drugs in these trials to their pharmacological targets.
The researchers extracted data on gene-disease associations using the Open Targets Platform and mapped the genetic evidence to the identifiers from EFO and Ensembl.
The Open Targets Genetics platform was also used to include common disease genetics, while other germline genetics sources such as Gene Burden, Uniprot, Gene2Phenotype, etc., were also included. A dataset of over 3.5 million genetically supported pairs of genes and traits was compiled.
Furthermore, to examine the factors that could influence the clinical trials that were discontinued due to side effects, the researchers included target annotations that were independent of the indicators examined in the study. The target tissue specificity data was obtained from the Human Protein Atlas.
The trained model was then applied to the 28,561 clinical trials that were identified as discontinued, and the integrated clinical and genetic evidence was used to analyze the reasons for stopping the trials, with a focus on safety and efficacy.
Major Findings
The study found that the natural language processing model was able to classify the reasons for discontinuation of clinical trials with high accuracy, especially for straightforward reasons such as insufficient enrollment or the coronavirus disease 2019 (COVID-19) pandemic. For more complex reasons of discontinuation, the performance accuracy was a little lower.
The model classified the reasons for discontinuation into 15 major classes, of which insufficient enrollment was the most common reason, reported by 36.67% of the discontinued studies. About 3.38% of the discontinued trials stated side effects or safety concerns as the reason and 7.6% were discontinued due to inadequate indicators of the efficacy of the trial.
While safety-related trial discontinuations occurred largely in phase I studies, negative outcomes were more prevalent in trials that were in phases II and III. The discontinuation of various trials in early phase I due to logistical concerns also highlighted the importance of comprehensive clinical practices.
Almost half (48%) of the discontinued trials were in the field of oncology, indicating that cancer research trials had a high failure rate. While most oncology clinical trials were discontinued due to safety concerns, trials involving respiratory diseases frequently cited COVID-19 as the reason for discontinuation.
The study also found that trials where the pharmacological target was strongly supported by genetic evidence had a high probability of success. The discontinued trials showed a significant lack of genetic support for targets, and the trends were observed in both oncological and non-oncological trials.
The safety concerns in most discontinued trials were linked to genes that were intolerant of loss-of-function changes or had the drug targets interacting with multiple proteins.
Conclusions
To summarize, the study trained a natural language processing model to classify the reasons for clinical trial discontinuation and applied to it classify over 28,000 discontinued clinical trials.
The findings suggested that lack of efficacy and safety concerns or side effects were the most likely reasons for trial discontinuation. However, trials where the drug targets were strongly supported by genetics were more likely to succeed.
New GIFT Method Outshines GWAS in Genetic Breakthrough