Adjusting for publication bias: modelling the selection process

Authors

  • Carrol Preston BSc PhD,

    Corresponding author
    1. Lecturer in Medical Statistics, Centre for Medical Statistics and Health Evaluation, University of Liverpool, Liverpool, UK
    Search for more papers by this author
  • Deborah Ashby BSc MSc PhD CStat,

    1. Professor of Medical Statistics, Wolfson Institute of Preventive Medicine, Queen Mary, University of London, London, UK
    Search for more papers by this author
  • Rosalind Smyth MA MB BS MD FRCPCH

    1. Brough Professor of Paediatric Medicine, Institute of Child Health, Alder Hey Childrens Hospital, Liverpool, UK
    Search for more papers by this author

Dr. Carrol Preston
Centre for Medical Statistics and Health Evaluation
School of Health Sciences
Faculty of Medicine
University of Liverpool
Liverpool L69 3BX
UK
E-mail: c.preston@liverpool.ac.uk

Abstract

Rationale, aims and background  Systematic review with meta-analysis, a statistical technique for combining results of several studies, is progressively being used to guide decisions in medicine. Publication bias is acknowledged as a threat to the validity of systematic reviews and its existence may lead to inappropriate decisions about patient management or health policy. It is said to occur when the results of research available in the literature are not representative of the totality of all research. The selection mechanism that causes publication bias is complex, yet despite an extensive literature of empirical research identifying risk factors for publication, little work has been done to improve models of selection.

Methods  Methods that adjust combined meta-analytic estimates for publication bias are compared and applied to a systematic review of oral rehydration solution in the treatment of dehydration. Within a weighted distributions framework models of the selection process are considered and developed further.

Conclusions  Weighted distributions offer a flexible approach that allows the potential to modify the selection function to incorporate other factors. Methods that adjust combined estimates should not be used to provide an alternative answer but to consider the robustness of the combined estimate to publication bias.

Ancillary