Soutenance de thèse de SMELA Beata
Titre de thèse
The methodology of literature reviews: current state of the art, their place in health policy and perspectives in the era of new technologies.
Methodology for literature reviews: current state of the art, their place in healthcare policy and perspectives in the era of new technologies.
Résumé de la thèse
This thesis aimed to examine the place of systematic literature reviews (SLRs) and rapid reviews (RRs) in health policies and to assess perspectives in the era of new technologies.
The first part of the work was to assess the requirements of health technology assessments (HTAs) for SLRs, trends in published SLRs, and the status of knowledge on RRs.
HTA recommendations for SLRs are heterogeneous and lacking in rigour; For economic evidence, a systematic approach is not necessary.
The volume of scientific publications is exploding, requiring efficient methodologies for SLRs.
RRs have been proposed as a solution. SLRs and RRs are often used interchangeably; It is necessary to establish a unified definition and specify the indication for each type of journal.
The objective of the second part of my work was to evaluate three methodologies of RR, the software that supports the LR process, and to create and test filters on the study designs.
The methodologies tested for RRs have efficacy results of 84%, 89% and 100%; The latter result was achieved by limiting human intervention and searching a single database.
A total of 6 of the 56 trials were not recovered in the 3 RRs: 1) five were of poor quality, one with unclear risk of bias, 2) four studies were not available in Medline and Embase and two were poorly indexed, 3) Cochrane included missed studies in 28 meta-analyses; however, their absence in RRs affected only the conclusion of 4 meta-analyses.
Creating and testing filters for observational studies has shown that these types of studies are indexed with irrelevant terms or searched with terms that are too broad.
Using our SLR filter, 84-100% of the studies identified by the Cochrane Group were captured.
Improved indexing and good filtering should improve the performance of RRs. AI-backed LRs are effective for screening. Promising developments in machine learning technology will support desk retrieval and data extraction.
Thesis resume
This thesis aimed to review the place of SLRs and RRs in healthcare policy and assess perspectives in the era of new technologies.
The first part of the work was done to assess HTA requirements for LRs, published SLRs' trends, and the knowledge status on RRs.
HTAs' recommendations regarding LRs are heterogeneous and lack stringency; for the economic evidence, a systematic approach is not required. The scientific publications volume is exploding requiring efficient methodologies for LRs. RRs have been proposed as the solution. SLRs and RRs both are often used interchangeably; there is a need for a unified definition and specifying the indication for each reviews' type.
The objective of the second part of my work was to assess three RRs' methodologies, software supporting the LRs process, and create and test study design filters.
The tested methodologies for RRs resulted in 84, 89, and 100% of efficacy; the last was obtained by limiting human intervention and one database searching. In total, 6 out of 56 trials were not retrieved in all 3 RRs: 1) five were of poor quality, one with a not clear risk of bias, 2) four studies were not available in Medline and Embase, and two were inappropriately indexed, 3) Cochrane included missed studies in 28 meta-analyses; however, their lack in RRs did impact only conclusion of 4 meta-analyses.
Creating and testing filters for observational studies showed that this type of studies is being indexed with irrelevant terms or searched with too broad terms. Using our SLR filter, 84 to 100% of studies identified by the Cochrane group were captured. Improved indexing and good filtering should improve the performance of RRs.
LRs' supported by AI software are effective for screening. Promising developments in machine learning technology will support literature searching and data extraction.