Methodological quality of test accuracy studies included in systematic reviews in obstetrics and gynaecology: sources of bias

A prospective protocol was developed to assess the impact of QUADAS on ten systematic reviews performed over the period 2004-2007.We investigated whether there was an improvement in study quality since the introduction of QUADAS, whether a correlation existed between study sample size, country of origin of study and its quality. We also investigated whether there was a correlation between reporting and methodological quality and by the use of meta-regression analyses explored for items of quality that were associated with bias.A total of 300 studies were included. The overall quality of included studies was poor (> 50% compliance with 57.1% of quality items). However, the mean compliance with QUADAS showed an improvement post-publication of QUADAS (54.9% versus 61.4% p = 0.002). There was no correlation with study sample size. Gynaecology studies published from the United States of America showed higher quality (USA versus Western Europe p = 0.002; USA versus Asia p = 0.004). Meta-regression analysis showed that no individual quality item had a significant impact on accuracy. There was an association between reporting and methodological quality (r = 0.51 p < 0.0001 for obstetrics and r = 0.56 p < 0.0001 for gynaecology).A combination of poor methodological quality and poor reporting affects the inferences that can be drawn from test accuracy studies. Further compliance with quality checklists is required to ensure that bias is minimised.Obstetrics and gynaecology have seen rapid growth in the development of new tests [1-4]. For instance, tests designed to detect small for gestational age fetuses and to improve the staging of cancers have grown in recent years [5-9]. A key aspect of research on these is presented in the form of test accuracy studies [10], which generate a comparison of measurements made by an index test against those of an accepted reference standard test - the "gold standard". These comparisons enable an assessment of the accuracy of an index test,