Article type
Year
Abstract
Background: Selective publication can seriously threaten the validity of the results of a systematic review. For intervention studies various mechanisms of selective publication have been identified. The mechanisms underlying selective publication of studies on diagnostic test accuracy (DTA), however, are not well understood. There might be some analogies with intervention studies: small sample sizes tend to have a larger and more positive effect compared to studies with larger sample sizes. In addition, the first studies reporting results of a novel intervention, tend to have more positive findings, though subsequent publications commonly show less optimistic results.
Objectives: To explore whether sample size or time of publication affects the results of DTA reviews.
Methods: We have searched MEDLINE and EMBASE to identify DTA reviews with a meta-analysis published between May and September 2012. From the largest meta-analysis of every review we have extracted characteristics of the primary studies: total sample size, year of publication, and two-by-two tables. The existence of sample size effect or time effect will be investigated by meta-regression for each meta-analysis by the use of the bivariate model. The resulting regression coefficients and corresponding variances will be combined in a random effects model to investigate the overall association.
Results: We have included 53 DTA reviews. So far, we have done data extraction for 28 reviews. The meta-analyses contained a median of 12 primary studies (IQR = 7–17) and a median sample size of 81 participants (IQR = 42–153). Median difference in publication year since first publication was 7 years (IQR = 4–11). Data collection is currently ongoing and results of the meta-regression will be presented at the Colloquium.
Conclusions: This study aims to investigate sample size and time of publication as possible mechanisms of selective publication in the DTA setting. We will provide authors of DTA reviews with guidance regarding the investigation of publication bias.
Objectives: To explore whether sample size or time of publication affects the results of DTA reviews.
Methods: We have searched MEDLINE and EMBASE to identify DTA reviews with a meta-analysis published between May and September 2012. From the largest meta-analysis of every review we have extracted characteristics of the primary studies: total sample size, year of publication, and two-by-two tables. The existence of sample size effect or time effect will be investigated by meta-regression for each meta-analysis by the use of the bivariate model. The resulting regression coefficients and corresponding variances will be combined in a random effects model to investigate the overall association.
Results: We have included 53 DTA reviews. So far, we have done data extraction for 28 reviews. The meta-analyses contained a median of 12 primary studies (IQR = 7–17) and a median sample size of 81 participants (IQR = 42–153). Median difference in publication year since first publication was 7 years (IQR = 4–11). Data collection is currently ongoing and results of the meta-regression will be presented at the Colloquium.
Conclusions: This study aims to investigate sample size and time of publication as possible mechanisms of selective publication in the DTA setting. We will provide authors of DTA reviews with guidance regarding the investigation of publication bias.