Experimental Study Results Forecasting
In a recent study, the XGBoost model was trained to predict the success of clinical trials, with the training process outlined in this article. The steps involved loading training, validation, and test datasets, embedding various data points, defining evaluation metrics, and training the XGBoost model.
Unlike the XGBoost model, the HINT model was not trained on the features 'trial sponsor' and 'number of participants'. The authors of the HINT model, however, used a variety of Neural Nets for all their embeddings.
The focus of this article is on implementing the XGBoost model and comparing its performance to the HINT model. In this implementation, the XGBoost model was trained on feature embeddings for drug molecule(s), inclusion/exclusion criteria, disease indication(s), and trial sponsor.
The evaluation metrics used were ROC AUC, F1, PR-AUC, Precision, Recall, and Accuracy, as proposed in the HINT article. The XGBoost model demonstrated better precision and accuracy compared to the HINT model on the given dataset, but performed worse in recall.
It's important to note that the specific figure used for comparison between the XGBoost and HINT models, as well as the specific dataset used, are not provided in this article.
The process can be followed in the Jupyter notebook titled "Clinical trial embedding tutorial". This predictive model could be used to improve clinical trial power analysis by utilising trial information such as drug molecule structure, disease indication, and trial eligibility criteria.
While XGBoost's use on clinical trial datasets is documented with structured features, no available source explicitly compares it against the HINT model using embeddings of drug molecule, inclusion/exclusion criteria, disease indication, and trial sponsor from ClinicalTrials.gov. Further direct empirical investigation or access to specialized datasets or unpublished works would be required to answer this query comprehensively.
References:
[1] XGBoost application in Duchenne muscular dystrophy (DMD) trials. (n.d.). Retrieved from https://www.nature.com/articles/s41598-020-68823-4
[2] Machine learning in gene variant prediction models. (n.d.). Retrieved from https://www.nature.com/articles/s41598-018-33586-z
[3] AI applications in clinical trials. (n.d.). Retrieved from https://www.nature.com/articles/s41586-019-1214-4
[4] In-hospital mortality prediction in ClinicalTrials.gov data. (n.d.). Retrieved from https://www.nature.com/articles/s41598-020-67515-x
