Article review: Designing Efficient Systematic Reviews Using Economical Allocation, Creation and Synthesis of Medical Evidence

Designing Efficient Systematic Reviews Using Economical Allocation, Creation and Synthesis of Medical Evidence. Scarpati M. RAND Corporation. 2014

At 140 pages this is a significant publication, produced as part of the Pardee RAND Graduate School dissertation series – so it’s Scarpati’s PhD thesis! The table of contents highlights the following chapters:

  1. Introduction
  2. Estimating the Value of Systematic Reviews: Osteoporosis Case Study
  3. Screening Times in Systematic Reviews
  4. De Novo Reviews
  5. Updating Existing Reviews
  6. Conclusion

It’s impossible to summarise this publication any better than the RAND landing page which reports:

“Medical literature and the actions of policymakers have emphasized the importance of evidence-based medicine in recent years, but basing clinical practice on an exploding base of evidence is challenging. Systematic reviews, which are very resource-intensive, are a crucial channel in the pathway from medical literature to clinical practice. This thesis begins by estimating the value of one systematic review, finding that synthesized evidence regarding treatments to prevent osteoporotic fractures generated a net benefit of approximately $450M. Next, the time taken to screen articles in systematic reviews is analyzed, showing that user interface changes can result in significant reductions in resource requirements. Presenting multiple articles on one screen while reviewing titles leads to a seven-fold reduction in time taken per article. Experience and mental state are also related to screening times, with abstracts reviewed at ideal session lengths requiring 33% less time than those at the beginning of a session.

To further increase the speed at which articles can be screened and decrease the cost of preparing systematic reviews, machine learning techniques allow avoidance of up to 80% of articles. When updating an existing review, savings are increased by utilizing the information present in original screening decisions to train the machine learning model. Finally, implementation issues are addressed, paying attention to technical, organizational, and institutional challenges and opportunities.”

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s