skip to main content

Title: Predictive modeling of clinical trial terminations using feature engineering and embedding learning
Abstract In this study, we propose to use machine learning to understand terminated clinical trials. Our goal is to answer two fundamental questions: (1) what are common factors/markers associated to terminated clinical trials? and (2) how to accurately predict whether a clinical trial may be terminated or not? The answer to the first question provides effective ways to understand characteristics of terminated trials for stakeholders to better plan their trials; and the answer to the second question can direct estimate the chance of success of a clinical trial in order to minimize costs. By using 311,260 trials to build a testbed with 68,999 samples, we use feature engineering to create 640 features, reflecting clinical trial administration, eligibility, study information, criteria etc. Using feature ranking, a handful of features, such as trial eligibility, trial inclusion/exclusion criteria, sponsor types etc. , are found to be related to the clinical trial termination. By using sampling and ensemble learning, we achieve over 67% Balanced Accuracy and over 0.73 AUC (Area Under the Curve) scores to correctly predict clinical trial termination, indicating that machine learning can help achieve satisfactory prediction results for clinical trial study.
Award ID(s):
2027339 1763452 1828181
Publication Date:
Journal Name:
Scientific Reports
Sponsoring Org:
National Science Foundation
More Like this
  1. Gadekallu, Thippa Reddy (Ed.)
    As of March 30 2021, over 5,193 COVID-19 clinical trials have been registered through Among them, 191 trials were terminated, suspended, or withdrawn (indicating the cessation of the study). On the other hand, 909 trials have been completed (indicating the completion of the study). In this study, we propose to study underlying factors of COVID-19 trial completion vs . cessation, and design predictive models to accurately predict whether a COVID-19 trial may complete or cease in the future. We collect 4,441 COVID-19 trials from to build a testbed, and design four types of features to characterize clinical trial administration, eligibility, study information, criteria, drug types, study keywords, as well as embedding features commonly used in the state-of-the-art machine learning. Our study shows that drug features and study keywords are most informative features, but all four types of features are essential for accurate trial prediction. By using predictive models, our approach achieves more than 0.87 AUC (Area Under the Curve) score and 0.81 balanced accuracy to correctly predict COVID-19 clinical trial completion vs . cessation. Our research shows that computational methods can deliver effective features to understand difference between completed vs . ceased COVID-19 trials. In addition, such modelsmore »can also predict COVID-19 trial status with satisfactory accuracy, and help stakeholders better plan trials and minimize costs.« less
  2. Clinical trials are crucial for the advancement of treatment and knowledge within the medical community. Since 2007, US federal government took the initiative and requires organizations sponsoring clinical trials with at least one site in the United States to submit information on these clinical trials to the database, resulting in a rich source of information for clinical trial research. Nevertheless, only a handful of analytic studies have been carried out to understand this valuable data source. In this study, we propose to use network analysis to understand infectious disease clinical trial research. Our goal is to answer two important questions: (1) what are the concentrations and characteristics of infectious disease clinical trail research? and (2) how to accurately predict what type of clinical trials a sponsor (or an investigator) is interested in? The answers to the first question provide effective ways to summarize clinical trial research related to particular disease(s), and the answers to the second question help match clinical trial sponsors and investigators for information recommendation. By using 4,228 clinical trails as the test bed, our study involves 4,864 sponsors and 1,879 research areas characterized by Medical Subject Heading (MeSH) keywords. We extract a set of network measuresmore »to show patterns of infectious disease clinical trials, and design a new community based link prediction approach to predict sponsors' interests, with significant improvement compared to baselines. This trans-formative study concludes that using network analysis can tremendously help the understanding of clinical trial research for effective summarization, characterization, and prediction.« less
  3. Abstract Background:

    Randomized controlled trials (RCT) play a central role in evidence-based healthcare. However, the clinical and policy implications of implementing RCTs in clinical practice are difficult to predict as the studied population is often different from the target population where results are being applied. This study illustrates the concepts of generalizability and transportability, demonstrating their utility in interpreting results from the National Lung Screening Trial (NLST).


    Using inverse-odds weighting, we demonstrate how generalizability and transportability techniques can be used to extrapolate treatment effect from (i) a subset of NLST to the entire NLST population and from (ii) the entire NLST to different target populations.


    Our generalizability analysis revealed that lung cancer mortality reduction by LDCT screening across the entire NLST [16% (95% confidence interval [CI]: 4–24)] could have been estimated using a smaller subset of NLST participants. Using transportability analysis, we showed that populations with a higher prevalence of females and current smokers had a greater reduction in lung cancer mortality with LDCT screening [e.g., 27% (95% CI, 11–37) for the population with 80% females and 80% current smokers] than those with lower prevalence of females and current smokers.


    This article illustrates how generalizability and transportability methods extend estimation ofmore »RCTs' utility beyond trial participants, to external populations of interest, including those that more closely mirror real-world populations.


    Generalizability and transportability approaches can be used to quantify treatment effects for populations of interest, which may be used to design future trials or adjust lung cancer screening eligibility criteria.

    « less
  4. ABSTRACT IMPACT: Understanding how spinal cord stimulation works and who it works best for will improve clinical trial efficacy and prevent unnecessary surgeries. OBJECTIVES/GOALS: Spinal cord stimulation (SCS) is an intervention for chronic low back pain where standard interventions fail to provide relief. However, estimates suggest only 58% of patients achieve at least 50% reduction in their pain. There is no non-invasive method for predicting relief provided by SCS. We hypothesize neural activity in the brain can fill this gap. METHODS/STUDY POPULATION: We tested SCS patients at 3 times points: baseline (pre-surgery), at day 7 during the trial period (post-trial), and 6 months after a permanent system had been implanted. At each time point participants completed 10 minutes of eyes closed, resting electroencephalography (EEG) and self-reported their pain. EEG was collected with the ActiveTwo system and a 128-electrode cap. Patients were grouped based on the percentage change of their pain from baseline to the final visit using a median split (super responders > average responders). Spectral density powerbands were extracted from resting EEG to use as input features for machine learning analyses. We used support vector machines to predict response to SCS. RESULTS/ANTICIPATED RESULTS: Baseline and post-trial EEG data predictedmore »SCS response at 6-months with 95.56% and 100% accuracy, respectively. The gamma band had the highest performance in differentiating responders. Post-trial EEG data best differentiated the groups with feature weighted dipoles being more highly localized in sensorimotor cortex. DISCUSSION/SIGNIFICANCE OF FINDINGS: Understanding how SCS works and who it works best for is the long-term objective of our collaborative research program. These data provide an important first step towards this goal.« less
  5. Statistical inference has strongly relied on the use of p-values to draw conclusions. For over a decade this reliance on the p-value has been questioned by researches and academics. The question of whether p-values are truly the best standard, and what other possible statistics could replace p-values l has been discussed deeply. We set out to understand the amount of variation within p-values, and to find if they really are as reliable as the frequency of their use would suggest. To answer this question, we studied a set of clinical trials over the past two years. We also aim to describe the variety of information included in drag labels, and determine whether this information conforms to FDA guidelines. We found a large variation in the presentation of clinical trial data, much of which was not in line with the guidelines of the FDA. Our findings also show that among the clinical trials we studied there is more variation among the p-values than among the estimates. From this, we can conclude that the estimates from clinical trials should hold a heavy weight in the decision of whether or not to approve the drug. This finding suggests that there is validity tomore »the skepticism of the reliance on p-values, and that further studies need to be done to find a new, more reliable, standard in statistical inference.« less