Mapping the Alzheimer's disease clinical trial space

Timothy J. Schultz

doi:10.17918/etd-7096

Over the past 20 years, the pharmaceutical industry has realized a disproportional trend in new drug approvals at the cost of increased spending. While insights into underlying mechanisms of disease continues to grow, the accumulation and integration of this knowledge complicates its application to clinical practice in the form of clinical trial protocol design. As knowledge is gained from the successes and failures of basic research and subsequent clinical trials, the landscape of the therapeutic space evolves: novel drug targets are assessed, new treatment approaches are developed, and patient populations are refined. This has forced researchers to rethink strategies which facilitate the efficient development of next-generation clinical trials, by leveraging lessons-learned from outputs of past research. Conventional protocol repositories such as ClinicalTrials.gov do not allow for this level of precise inquiry from its plain-text search and document retrieval functionality. Instead, it is useful to devise analytical tools which quantify and visualize temporal shifts in treatment strategies for the purpose of better understanding the evolution of complex therapeutic spaces at various degrees of resolution, through the development of a novel protocol mining framework. This framework is explored through its application to the complex Alzheimer's Disease (AD) therapeutic space. In light of recent high-profile AD clinical trial failures, the need for leveraging lessons-learned from trial results is apparent, since the century-old therapeutic space is dynamic and multi-faceted. ARRASTRA, a Python framework, has been developed for mining and analyzing large corpora of semantically annotated documents, specifically for this purpose, clinical trial protocols. Plain-text protocols are augmented with natural language processing (NLP) and semantic annotation routines. By mining protocol collections and expressing them as temporally-directed similarity networks, graph theory can be employed to quantify the evolution of a therapeutic space at both the individual protocol level (nodes), and high-level themes (clusters). Firstly, the ability of a protocol to influence the genesis of new trials are quantified by their position in this network, and are analyzed over time by identifying significant "bursts" within those metrics. Secondly, the evolution of a therapeutic space itself is visualized as temporal changes at the graph cluster level using a "MetroMap" visualization, which tracks sub-therapeutic themes as new ideas form and old ideas are abandoned. Finally, these graph metrics when integrated with additional protocol metadata, are used to predict the viability of themes using survival models. Overall, the ARRASTRA framework provides novel fine-grained insights into the changing nature of a therapeutic space by repurposing the content of clinical trial protocol repositories in ways not previously explored.

Mapping the Alzheimer's disease clinical trial space

Files and links (1)

Abstract

Metrics

Details

Mapping the Alzheimer's disease clinical trial space

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media