Machine Learning and the Future of Systematic Literature Review

Laurie Mitchell, President of Criterion Edge, recently sat down for an interview to discuss the uses and benefits of Systematic Literature Review (SLR) in meeting the EU MDR and IVDR compliance requirements, as well as the dynamic role that technologies like Machine Learning (ML) can play as SLR evolves. 

"SLR is a clear, sound methodology used for the process of identification, retrieval selection, appraisal, and weighting of published data," Mitchell said. "MedTech manufacturers should consider conducting an SLR when there is insufficient manufacturer-held data to address organizational needs or regulatory requirements."

Within the MedTech industry, she added, current best practices are evolving faster than ever, including the adoption of automation, ML, and artificial intelligence (AI) into the processes of SLR.

“We are very excited about the promise that AI holds for literature review and data identification. This is a rapidly developing field and we've got our eye on how it can be deployed in our organization in a cost-effective way to improve efficiency and produce accurate high quality search and review results,” said Mitchell. 

The Fatal Flaws of Internet Search

According to Mitchell, SLR goes far beyond what is possible with an internet search, as the functionality of an internet search is very limited. The use of MESH terminology, search syntax including Boolean operators, and stepwise full-search protocols, is only possible in a more robust search platform.

For the purposes of MDR and IVDR compliance requirements, Mitchell noted, internet searches are insufficient, in that they are not considered reproducible. This means that notified bodies and other regulatory agencies who attempt to run the same search query that was used as a basis for compliance would not be able to reproduce the initial results. Thus, Mitchell said, when the articles included to back up a claim are questioned, there is no method to arrive at the internet search used to support them. This is where alternative approaches to internet searches come into play, offering results that are auditable and transparent.

“The reproducibility factor is very big,” said Mitchell, “because the results can vary from search to search, and if you're pulling data, then the regulatory authorities are going to worry that the search is not reproducible; thus the data change from search to search - and that is a non-starter for them.” 

Another critical issue, according to Mitchell is objectivity. She said that documenting the searcher’s objectivity is a daunting challenge, primarily because their search terms and search strings are difficult to create and document. Alternatively, the documentation of an SLR search protocol is very straightforward and clear. 

If conducting a regulatory search or collecting data in any official context, Mitchell added, an internet search cannot be used to document and gather this data.

The Evolution of SLR 

Mitchell maintained that SLR is a current, complex, and evolving best practice that makes for an effective and efficient way to abide by the regulatory compliance requirements put forth in EU MDR and IVDR. 

“SLR is used to, and this is a growing field, to augment and automate the company's post-market surveillance activities,” said Mitchell. Not only does SLR help medical device and in vitro diagnostics manufacturers align with the new regulations, Mitchell said, it also assists in the key corporate initiatives such as developing white papers and other clinical marketing material, as well as assessing the current therapeutic landscape of devices. Mitchell’s team currently uses industry-standard automation processes, allowing them to screen literature more efficiently, which results in shorter review times with fewer errors noted during the quality control processes. 

“We're seeing how current best practice is evolving with the adoption of automation and artificial intelligence into this field. Those new technologies are disrupting the human centric process of literature screening and the identification of relevant data points for extraction,” said Mitchell.

To learn more about how ML can accelerate SLR, please request a discovery call with Huma.AI below.