Article type
Year
Abstract
Background: PubMed includes the 'Article Types' filter for focused literature search, but it is not clear whether this yields output according to methodological quality (risk of bias) as listed in the evidence hierarchy pyramid.
Objectives: To assess the utility of PubMed 'Article Types' filter for literature search.
Methods: PubMed was searched on 31 March 2015 to i) categorize the risk of bias of study types listed in the 'Article Types' filter; ii) calculate the relative proportions of citations conforming to low, moderate and high risk of bias; and iii) observe time-trends from 1900 to 2015 (data for 2015 was annualized).
Results: There were 71 'Article Types' listed in PubMed, but the basis of classification of citations was not described. It was also not mentioned whether individual citations were categorized into single or multiple slots. Of the 71 types, two were identified as having low risk of bias (a systematic review and a randomized controlled trial), four as moderate risk (observational study, comparative study, clinical trial, controlled clinical trial), and three as high risk (case report, editorial, review). The risk of bias of the other 62 article types was unclear. It was also unclear whether all 71 types were created at the same time. PubMed did not categorize citations as per the evidence-hierarchy pyramid. There was no category for cohort study or case-control study. Even the category 'Observational study' yielded the first citation in 2001. Analysis of time-trends showed a steady increase in low risk of bias citations (from 2.5% in 1976 to 1980 to 16.6% in 2011 to 2015). The proportion of high risk of bias citations appears unchanged over the past 65 years (range 53% to 59%). However, the proportion of moderate risk of bias citations declined from 42% to 28% during over 65 years. There were several obvious instances of misclassification and duplicate classifications.
Conclusions: Searching PubMed using the 'Article types' filter does not yield citations based on methodological quality described in the evidence-hierarchy pyramid. There are several additional limitations, on account of which search using this tool may be unreliable.
Objectives: To assess the utility of PubMed 'Article Types' filter for literature search.
Methods: PubMed was searched on 31 March 2015 to i) categorize the risk of bias of study types listed in the 'Article Types' filter; ii) calculate the relative proportions of citations conforming to low, moderate and high risk of bias; and iii) observe time-trends from 1900 to 2015 (data for 2015 was annualized).
Results: There were 71 'Article Types' listed in PubMed, but the basis of classification of citations was not described. It was also not mentioned whether individual citations were categorized into single or multiple slots. Of the 71 types, two were identified as having low risk of bias (a systematic review and a randomized controlled trial), four as moderate risk (observational study, comparative study, clinical trial, controlled clinical trial), and three as high risk (case report, editorial, review). The risk of bias of the other 62 article types was unclear. It was also unclear whether all 71 types were created at the same time. PubMed did not categorize citations as per the evidence-hierarchy pyramid. There was no category for cohort study or case-control study. Even the category 'Observational study' yielded the first citation in 2001. Analysis of time-trends showed a steady increase in low risk of bias citations (from 2.5% in 1976 to 1980 to 16.6% in 2011 to 2015). The proportion of high risk of bias citations appears unchanged over the past 65 years (range 53% to 59%). However, the proportion of moderate risk of bias citations declined from 42% to 28% during over 65 years. There were several obvious instances of misclassification and duplicate classifications.
Conclusions: Searching PubMed using the 'Article types' filter does not yield citations based on methodological quality described in the evidence-hierarchy pyramid. There are several additional limitations, on account of which search using this tool may be unreliable.