Reporting of confounding bias in observational intervention studies: are we making progress?
Previously, poor quality of reporting of confounding in articles on observational medical interventions has been observed in a systematic review. Included articles were published before the STROBE statement and it was suggested that this statement could have a considerable impact on the reporting of such studies. We aimed to assess whether the reporting of confounding improved.
METHODS:
Articles on observational studies on medical interventions in five general medical and five epidemiological journals published between January 2011 and December 2012 were systematically reviewed. Relevant items related to confounding bias were scored for each article. Overall quality of reporting was based on a 7-point score. A comparison was made with the previously published systematic review based on articles published before the STROBE statement. Risk ratios (RR) with 95% confidence intervals (CI) were calculated to represent changes in two prespecified items: reporting on likelihood of unobserved confounding and reporting on sensitivity analysis to estimate the potential impact of unobserved confounders.
RESULTS:
Preliminary results are based on 153 included articles. The majority of studies provided details on the distribution of key confounders (81%), methods used to control for observed confounding (99%) and the potential for unobserved confounding (90%). Details on the selection and inclusion of confounders for the final model were provided in 22% and 41%, respectively. The overall quality of reporting was moderate (median 4 points; interquartile range 3 to 5). Articles published after the STROBE statement commented more often on the potential for unobserved confounding (RR 1.52, 95% CI 1.33-1.75). Sensitivity analyses to estimate the potential impact of unobserved confounders were not significantly more reported after the STROBE statement (4.6% vs. 2.3%; RR 1.99, 95% CI 0.59-6.67).
CONCLUSIONS:
The quality of reporting of confounding in articles on observational medical interventions remains suboptimal and further efforts are needed to improve reporting.