Publication bias

From Wikipedia, the free encyclopedia
Jump to: navigation, search

Publication bias is a bias with regard to what is likely to be published, among what is available to be published. Not all bias is inherently problematic – for instance, a bias against publishing lies is often a desirable bias – but one problematic and much-discussed bias is the tendency of researchers, editors, and pharmaceutical companies to handle the reporting of experimental results that are positive (i.e. showing a significant finding) differently from results that are negative (i.e. supporting the null hypothesis) or inconclusive, leading to a misleading bias in the overall published literature.[1]

This is usually a bias towards reporting significant results, despite the fact that studies with significant results do not appear to be superior to studies with a null result with respect to quality of design.[2] It has been found that statistically significant results are three times more likely to be published than papers affirming a null result.[3] It also has been found that the most common reason for non-publication is an investigator's declining to submit results for publication (because of the investigator's loss of interest in the topic, the investigator's anticipation that others will not be interested in null results, etc.), a finding that underlines researchers’ role in publication bias phenomena.[2]

In an effort to decrease this problem, some prominent medical journals require registration of a trial before it commences so that unfavorable results are not withheld from publication. Several such registries exist, but researchers are often unaware of them. In addition, attempts to identify unpublished studies have proved very difficult and often unsatisfactory. Another strategy suggested by a meta-analysis is caution in the use of small and non-randomised clinical trials because of their demonstrated high susceptibility to error and bias.[2]

Definition[edit]

According to one publication:

Publication bias occurs when the publication of research results depends on their nature and direction.[4]

Positive results bias, a type of publication bias, occurs when authors are more likely to submit, or editors accept, positive than null (negative or inconclusive) results.[5] A related term, "the file drawer problem", refers to the tendency for negative or inconclusive results to remain unpublished by their authors.[6]

Outcome reporting bias occurs when several outcomes within a trial are measured but are reported selectively depending on the strength and direction of those results. A related term that has been coined is HARKing (Hypothesizing After the Results are Known).[7]

File drawer effect[edit]

The file drawer effect, or file drawer problem, is that many studies in a given area of research may be conducted but never reported, and those that are not reported may on average report different results from those that are reported. An extreme scenario is that a given null hypothesis of interest is in fact true, i.e. the association being studied does not exist, but the 5% of studies that by chance show a statistically significant result are published, while the remaining 95% where the null hypothesis was not rejected languish in researchers' file drawers. Even a small number of studies lost "in the file drawer" can result in a significant bias.[8] The term "file drawer problem" was coined by the psychologist Robert Rosenthal in 1979.[6]

Effect on meta-analysis[edit]

The effect of this is that published studies may not be truly representative of all valid studies undertaken, and this bias may distort meta-analyses and systematic reviews of large numbers of studies—on which evidence-based medicine, for example, increasingly relies. The problem may be particularly significant when the research is sponsored by entities that may have a financial or ideological interest in achieving favorable results.

Indeed, a recent study has shown that clinical trials showing statistically significant results favoring the treatment and observational studies showing plausible statistically significant outcomes often had a higher probability of being included in the recent meta-analyses published in major general medical journals than studies showing other results.[9]

Those undertaking meta-analyses and systematic reviews need to take account of publication bias in the methods they use for identifying the studies to include in the review. Among other techniques to minimize the effects of publication bias, they may need to perform a thorough search for unpublished studies, and to use such analytical tools as a Begg's funnel plot or Egger's plot to quantify the potential presence of publication bias. Tests for publications bias rely on the underlying theory that small studies with small sample size (and large variance) would be more prone to publication bias, while large-scale studies would be less likely to escape public knowledge and more likely to be published regardless of significance of findings. Thus, when overall estimates are plotted against the variance (sample size), a symmetrical funnel is usually formed in the absence of publication bias, while a skewed asymmetrical funnel is observed in presence of potential publication bias.

Extending the funnel plot, the "trim and fill" method has also been suggested as a method to infer the existence of unpublished hidden studies as determined from a funnel plot, and subsequently correct the meta-analysis by imputing the presence of missing studies to yield an unbiased pooled estimate.

Additionally, selection models are available, which allow to estimate the function describing the probability of being included in a meta-analysis for different outcomes. Selection models can also be used to conduct a meta-analysis when a publication bias is present.

However, since all publication bias methods are characterized by a relatively low power and are based on strong and unverifiable assumptions, a negative result of a publication bias test does not guarantee the validity of conclusions from a meta-analysis.[10]

Examples[edit]

The antidepressant Reboxetine provides an example of experimental bias in clinical trials. It was originally passed as effective for treatment of depression in many countries in Europe in the UK in 2001 (though in practise it is rarely used). It was later (in 2010) found to be ineffective due to publication bias in the original trials published by the drug manufacturer Pfizer. A later (2011) meta analysis of the original data found flaws in the 2010 meta analysis and suggests that it can be effective after all, in severe cases of depression. See Reboxetine - Efficacy. Whatever the final outcome for Reboxetine, the original trials show a clear case of publication bias. More examples of publication bias are given by Ben Goldacre [11] and Peter Wilmhurst.[12]

In the social sciences, a study looks at published papers on the relationship between Corporate Social and Financial Performance, and found that "In economics, finance, and accounting journals, the average correlations were only about half the magnitude of the findings published in Social Issues Management, Business Ethics, or Business and Society journals".[13]

Publication bias is often cited in investigations of papers on the Paranormal. A recent example is a paper by Daryl Bem, which showed evidence of short term pre-cognition. Negative results by other researchers that attempted to duplicate his work were not published in the journals that published the original results.[14]

One study[15] compared Chinese and non-Chinese studies of gene-disease associations and found that "Chinese studies in general reported a stronger gene-disease association and more frequently a statistically significant result".[16] One possible interpretation of this result is selective publication (publication bias).

Risks[edit]

According to John Ioannidis, negative papers are most likely to be suppressed when:[17]

  1. the studies conducted in a field are smaller
  2. effect sizes are smaller
  3. there is a greater number and lesser preselection of tested relationships
  4. there is greater flexibility in designs, definitions, outcomes, and analytical modes
  5. there is greater financial and other interest and prejudice
  6. more teams are involved in a scientific field in chase of statistical significance.

Ioannidis further asserts that "claimed research findings may often be simply accurate measures of the prevailing bias".

Remedies[edit]

Ioannidis' remedies include:

  1. Better powered studies
    • Low-bias meta-analysis
    • Large studies where they can be expected to give very definitive results or test major, general concepts
  2. Enhanced research standards including
    • Pre-registration of protocols (as for randomized trials)
    • Registration or networking of data collections within fields (as in fields where researchers are expected to generate hypotheses after collecting data)
    • Adopting from randomized controlled trials the principles of developing and adhering to a protocol.
  3. Considering, before running an experiment, what they believe the chances are that they are testing a true or non-true relationship.
    • Properly assessing the false positive report probability based on the statistical power of the test[18]
    • Reconfirming (whenever ethically acceptable) established findings of "classic" studies, using large studies designed with minimal bias

Study registration[edit]

In September 2004, editors of several prominent medical journals (including the New England Journal of Medicine, The Lancet, Annals of Internal Medicine, and JAMA) announced that they would no longer publish results of drug research sponsored by pharmaceutical companies unless that research was registered in a public database from the start.[19] Furthermore, some journals, e.g. Trials, encourage publication of study protocols in their journals.[20]

See also[edit]

References[edit]

  1. ^ Song, F.; Parekh, S.; Hooper, L.; Loke, Y. K.; Ryder, J.; Sutton, A. J.; Hing, C.; Kwok, C. S.; Pang, C.; Harvey, I. (2010). "Dissemination and publication of research findings: An updated review of related biases". Health technology assessment (Winchester, England) 14 (8): iii, iix–xi, iix–193. doi:10.3310/hta14080. PMID 20181324.  edit
  2. ^ a b c Easterbrook, P. J.; Berlin, J. A.; Gopalan, R.; Matthews, D. R. (1991). "Publication bias in clinical research". Lancet 337 (8746): 867–872. doi:10.1016/0140-6736(91)90201-Y. PMID 1672966. 
  3. ^ Dickersin, K.; Chan, S.; Chalmers, T. C.; et al. (1987). "Publication bias and clinical trials". Controlled Clinical Trials 8 (4): 343–353. doi:10.1016/0197-2456(87)90155-3. 
  4. ^ K. Dickersin (March 1990). "The existence of publication bias and risk factors for its occurrence". JAMA 263 (10): 1385–1359. doi:10.1001/jama.263.10.1385. PMID 2406472. 
  5. ^ D.L. Sackett (1979). "Bias in analytic research". J Chronic Dis 32 (1–2): 51–63. doi:10.1016/0021-9681(79)90012-2. PMID 447779. 
  6. ^ a b Robert Rosenthal (May 1979). "The file drawer problem and tolerance for null results". Psychological Bulletin 86 (3): 638–641. doi:10.1037/0033-2909.86.3.638. 
  7. ^ N.L. Kerr (1998). "HARKing: Hypothesizing After the Results are Known". Personality and Social Psychology Review 2 (3): 196–217. doi:10.1207/s15327957pspr0203_4. PMID 15647155. 
  8. ^ Jeffrey D. Scargle (2000). "Publication bias: the "file-drawer problem" in scientific inference". Journal of Scientific Exploration 14 (2): 94–106. 
  9. ^ Kicinski M. Publication bias in recent meta-analyses. PLoS ONE 2013;8:e81823 http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0081823
  10. ^ Sutton AJ, Song F, Gilbody SM, Abrams KR (2000)Modelling publication bias in meta-analysis: a review. Stat Methods Med Res 9:421-445.
  11. ^ Ben Goldacre What doctors don't know about the drugs they prescribe
  12. ^ Wilmshurst, Peter. "Dishonesty in Medical Research". 
  13. ^ Marc Orlitzky Institutional Logics in the Study of Organizations: The Social Construction of the Relationship between Corporate Social and Financial Performance
  14. ^ Ben Goldacre Backwards step on looking into the future The Guardian, Saturday 23 April 2011
  15. ^ Zhenglun Pan, Thomas A. Trikalinos, Fotini K. Kavvoura, Joseph Lau, John P.A. Ioannidis, "Local literature bias in genetic epidemiology: An empirical evaluation of the Chinese literature". PLoS Medicine, 2(12):e334, 2005 December.
  16. ^ Jin Ling Tang, "Selection Bias in Meta-Analyses of Gene-Disease Associations", PLoS Medicine, 2(12):e409, 2005 December.
  17. ^ Ioannidis J (2005). "Why most published research findings are false". PLoS Med 2 (8): e124. doi:10.1371/journal.pmed.0020124. PMC 1182327. PMID 16060722. 
  18. ^ Wacholder, S.; Chanock, S; Garcia-Closas, M; El Ghormli, L; Rothman, N (March 2004). "Assessing the Probability That a Positive Report is False: An Approach for Molecular Epidemiology Studies". JNCI 96 (6): 434–42. doi:10.1093/jnci/djh075. PMID 15026468. 
  19. ^ (The Washington Post) (2004-09-10). "Medical journal editors take hard line on drug research". smh.com.au. Retrieved 2008-02-03. 
  20. ^ "Instructions for Trials authors — Study protocol". 2009-02-15. 

External links[edit]