Full-text vs Abstract advantage: Causaly identifies 3x as many relevant articles by machine-reading the full-text

Danai Mavreli
published on October 31, 2020

Introduction

Literature research is an ongoing, iterative process for all scientists. A literature review is a large task that can require 60–80 hours of focused effort for regulatory professionals (1). This includes scanning the medical literature to collect adverse events and analyzing the results.

A PubMed search considers abstracts and indexed MeSH terms but not full-text articles. However, safety outcomes are predominantly reported in full-text articles and can be easily missed when relying only on abstracts. The potential of machine-reading full-text articles addresses the human challenge of limited time and can facilitate the identification of most relevant evidence.

The case study below compares full-text articles and abstracts as sources for finding the side effects of Idelalisib.

More specifically Causaly identifies 3x as many articles, based on signals from full-text papers, that would otherwise have been missed.

Results also include twice as many side effects, when searching through full-text papers compared to abstracts.

Side effects of Idelalisib: Full-text vs Abstracts

A search using Causaly was performed to identify the side effects of idelalisib, a phosphoinositide 3-kinase inhibitor used to treat certain types of blood cancers, such as chronic lymphocytic leukemia and small lymphocytic lymphoma. The aim was to investigate the different safety signals identified through full-text articles, compared to those found in abstracts.

Our analysis found 87 full-text articles from PubMed Central that contain side effects for Idelalisib, compared to 30 abstracts from MEDLINE. This means that approximately 3 times as many full-text articles were found compared to abstracts. This is due to the fact that side effects are more prevalent in the body of a full-text paper than in the abstract.

Furthermore full-text articles yielded significantly more side effects: 65 were identified in full-texts compared to 35 identified in abstracts i.e. approximately twice as many as in abstracts. Overall, from the 84 unique side effects that were found, 49 of them only exist in full-texts and would otherwise have been missed with abstract search alone (Figure 1). Indicative side effects that only exist in full-texts are shown below (Figure 2).

Venn-diagram-v3
Figure 1: Full-text articles contained 49 additional side effects. 16 side effects are common between PubMed Central and MEDLINE. 19 side effects were found only in MEDLINE abstracts.

2_side_effects_v6
Figure 2. Indicative side effects that would have been missed using traditional literature review methods, as evidence exists only in full-text.

Conclusion

Causaly enables regulatory experts to reduce time spent scanning research literature, while at the same time increase the yield from full-text articles, which typically are not selected due to unsuspecting abstracts.

The application of AI on full-texts removes the bottleneck of human reading time and helps detect relevant safety signals from the ever increasing research literature.

The Search Outcomes were improved compared to a typical PubMed search. More specifically focusing on full-text papers we found:

  • 3x as many articles
  • 2x as many side effects
  • 49 safety signals which only exist in the full-texts and missed when reading only abstracts

Knowledge Management departments in Pharmaceutical companies with access to vast document holdings can benefit from Causaly’s AI technology by machine-reading their full-text documents and provide more relevant evidence to researchers and decision makers in the organization.

References

  1. “Seven Steps to Systematic Literature Reviews.” Mddionline.Com, 1 Feb. 2007, www.mddionlinae.com/testing/seven-steps-systematic-literature-reviews. Accessed 23 Oct. 2020.
  2. “Unified Medical Language System (UMLS).” National Library of Medicine, www.nlm.nih.gov/research/umls/index.html. Accessed 27 Oct. 2020.
Causaly vs PubMed®: 2x as many relevant articles identified by Causaly using the same data
use case

Causaly vs PubMed®: 2x as many relevant articles identified by Causaly using the same data

Causaly AI finds more relevant articles than PubMed alone, using its advanced machine-reading technology.

Target identification and validation using AI for literature-based insights: Causaly & Pierre Fabre Joint Webinar

Target identification and validation using AI for literature-based insights: Causaly & Pierre Fabre Joint Webinar

Causaly and Pierre Fabre joint webinar Causaly and Pierre Fabre co-hosted a joint webinar on the 28th of October, addressing how...

Preclinical safety analysis using Artificial Intelligence on the example of Alzheimer’s Disease.
Application

Preclinical safety analysis using Artificial Intelligence on the example of Alzheimer’s Disease.

Causaly AI enables researchers to identify safety-relevant information in medical literature regarding a drug candidate. Preclinical experts can include this data in the preclinical study design to minimize the risks of unforeseen toxicities and increase chances of approval.

Sign up for Causaly newsletter