Visualizing Clinical Trial Design Trends by Mapping the
Alzheimer’s Disease Trial Space
Timothy Schultz M.S., Chaomei Chen Ph.D., Neal Handly M.D.
BACKGROUND
METHODOLOGY (cont.)
CURRENT PROGRESS (cont.)
• Ultimate goal is optimization of clinical trial design and evaluation in
light of rising cost and failure rate [1]
• Lack of historical understanding of therapeutic space, what has been
done, what has worked, and what has failed
• Research aims to provide a novel approach to identify promising
clinical trial design patterns across time by qualifying and quantifying
the evolution of clinical trials within a therapeutic space
• Utilize a graph-based nonnegative Matrix Factorization (NMF)
approach for identifying latent themes based on annotated graph [3]
• Calculate trial similarities based on projected lower-dimensional space
• Sort similarities temporally to establish direction of influence between
similar clinical trials, resulting in a weighted directed graph
• Calculate graph metrics over sub-graphs obtained by expanding time
window (yearly), such as betweenness centrality and PageRank
• Utilize Kleinberg’s Burst Detection algorithm (Fig.2) against graph
metrics within each window to identify emergence of pivotal clinical
trials
• Utilizing numerous visual tools to qualify dynamics of therapeutic
area:
• Sankey diagrams to visualize change in graph dynamics over time
windows [4]
• Swim-lane timeline visualizations to visualize emergence of pivotal
clinical trials
• Interactive graphs for exploration of conceptual graph space
• Ability to export to other applications, such as CiteSpace [5] (Fig. 4)
PURPOSE AND RATIONALE
• Past work has focused on clustering similar clinical trials [2]
• Latent topical structures are contained within plain-text documents
such as clinical trial protocols, which are pervasive throughout time
• Plain-text protocol repositories (e.g. ClinicalTrials.gov) can be
extended from a simple document repository by providing analytical
capabilities and insights
• Enhance ability to recognize and exploit dependency of key clinical
trial design variables based on past experience
• Ultimate goal is to provide insights into the temporal dynamics of
research being conducted within therapeutic spaces
METHODOLOGY
• Semantically annotate protocol corpus with biomedical ontologies
• Express each document as a weighted graph of concepts (Fig. 1)
CURRENT PROGRESS
0
0
0
0
Semantically
annotated
documents
Figure 2. Kleinberg’s Burst Algorithm reveals influential drivers of change
within the network across periods of time.
Express documents
as a weighted graph
of concepts
(affinity matrix)
Vector of cooccurring concepts
(edge list)
• Utilizing Titan graph storage to capture discovered connections
between clinical trials and biomedical concepts
• Developed web application (D3, Sigma JS) to provide front-end tools to
explore therapeutic area (Fig. 3)
NCT000 , NCT001, 0.85
NCT002 , NCT003, 0.75
…
NCT00X , NCT00Y, 0.98
Calculate similarities
based on lower
dimensional space,
temporally sort
Establish graph of
interconnected
clinical trials
2003
2004
2005
…
2014
Calculate sub-graph dynamics over
expanding time windows
Figure 1. A graphical overview of the process for identifying thematic
clusters of thought within a therapeutic space.
drexel.edu/cci
FUTURE DIRECTION
• Anticipate ability to map underlying ontological concepts to
biomedical endpoints found within publicly-available observational
datasets (i.e. Alzheimer’s Disease Neuroimaging Initiative)
• Quantify within and between cluster variance of endpoints to further
visualize influence key clinical trial design considerations have on
prospective patient populations over time
REFERENCES
Dimensionality
reduction via NMF
Utilize a suite of tools to visualize
the dynamics of a therapeutic area
(i.e. Sankey Diagrams)
Figure 4. Identifying key themes at varying levels of granularity. Shift in
treatment from targeting moderate AD (early 2000’s) to more recently
early AD (2012) is apparent.
Figure 3. Querying and exploring the connections which exist between
similar clinical trial protocols.
1. S. L. Mercer, B. J. DeVinney, L. J. Fine, L. W. Green, and D. Dougherty, “Study Designs for
Effectiveness and Translation Research: Identifying Trade-offs,” Am. J. Prev. Med., vol. 33, no.
2, pp. 139–154.e2, Aug. 2007.
2. M. R. Boland, R. Miotto, J. Gao, and C. Weng, “Feasibility of feature-based indexing,clustering,
and search of clinical trials: A case study of breast cancer trials from ClinicalTrials.gov,”
Methods Inf. Med., vol. 52, no. 5, pp. 382–394, Oct. 2013.
3. C.-J. Lin, “Projected Gradient Methods for Nonnegative Matrix Factorization,” Neural Comput,
vol. 19, no. 10, pp. 2756–2779, Oct. 2007.
4. M. Rosvall and C. T. Bergstrom, “Mapping Change in Large Networks,” PLoS ONE, vol. 5, no.
1, p. e8694, Jan. 2010.
5. M. B. Synnestvedt, C. Chen, and J. H. Holmes, “CiteSpace II: Visualization and Knowledge
Discovery in Bibliographic Databases,” AMIA. Annu. Symp. Proc., vol. 2005, pp. 724–728,
2005.
drexel.edu/cci