Evidence of publication bias in reporting acute stroke clinical trials.

OBJECTIVE

To ascertain the extent of publication bias in the reporting of acute stroke clinical trials.

METHODS

We identified controlled acute ischemic stroke clinical trials reported in English over a 45-year period from 1955 to 1999 through systematic search of MEDLINE, the Cochrane Controlled Stroke Trials Register, and additional databases. We analyzed trial methodology, quality, outcome, study sponsorship, and timing of publication to identify various forms of publication bias, including nonpublication bias, abbreviated publication bias, and time-lag bias.

RESULTS

One hundred seventy-eight acute ischemic stroke trials, enrolling 73,949 subjects, evaluated 75 agents or nonpharmacologic interventions. A greater proportion of harmful outcomes in unpublished studies (n = 4) compared with published trials (0.75 vs 0.06, p < 0.0001) and underreporting of smaller, nonbeneficial studies in acute stroke suggest nonpublication bias. Although a definite time-lag bias was not evident, nonbeneficial studies were slower to proceed from enrollment completion to publication (2.3 vs 2.0 years, p = 0.207), with an even longer delay for nonbeneficial corporate pharmaceutical sponsored trials (2.8 vs 2.1 years, p = 0.086), despite superior trial report quality scores for corporate-sponsored studies when compared with nonprofit/governmental studies (mean 69.2 +/- 95% CI 3.9 vs 53.4 +/- 95% CI 9.2, p < 0.005).

CONCLUSION

Publication bias is evident in the acute stroke research literature, supporting the need for prospective trial registration.