Format

Send to

Choose Destination
PLoS One. 2017 Sep 12;12(9):e0182362. doi: 10.1371/journal.pone.0182362. eCollection 2017.

Improving data sharing in research with context-free encoded missing data.

Author information

1
Department of Neurology, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands.
2
INSERM, University of Toulouse, Toulouse, France.
3
Chronic Disease Prevention Unit, National Institute for Health and Welfare, Helsinki, Finland.
4
Department of Epidemiology and Public Health, Toulouse University Hospital, Toulouse, France.
5
INSERM, University of Toulouse UMR1027, Toulouse, France.
6
Department of General Practice, Academic Medical Centre, University of Amsterdam, Amsterdam, The Netherlands.
7
Institute of Clinical Medicine/Neurology, University of Eastern Finland, Kuopio, Finland.
8
Aging Research Center, Karolinska Institutet/ Stockholm University, Stockholm, Sweden.
9
Karolinska Institutet Center for Alzheimer Research, Stockholm, Sweden.
10
Neurocenter, neurology, Kuopio University Hospital, Kuopio, Finland.
11
Department of Public Health and Primary Care, Cambridge Institute of Public Health, University of Cambridge, United Kingdom.
12
Department of Information and Operations Management, ESCP Europe, Paris, France.
13
Department of Neurology, Radboud University Medical Center, Nijmegen, The Netherlands.

Abstract

Lack of attention to missing data in research may result in biased results, loss of power and reduced generalizability. Registering reasons for missing values at the time of data collection, or-in the case of sharing existing data-before making data available to other teams, can save time and efforts, improve scientific value and help to prevent erroneous assumptions and biased results. To ensure that encoding of missing data is sufficient to understand the reason why data are missing, it should ideally be context-free. Therefore, 11 context-free codes of missing data were carefully designed based on three completed randomized controlled clinical trials and tested in a new randomized controlled clinical trial by an international team consisting of clinical researchers and epidemiologists with extended experience in designing and conducting trials and an Information System expert. These codes can be divided into missing due to participant and/or participation characteristics (n = 6), missing by design (n = 4), and due to a procedural error (n = 1). Broad implementation of context-free missing data encoding may enhance the possibilities of data sharing and pooling, thus allowing more powerful analyses using existing data.

PMID:
28898245
PMCID:
PMC5595279
DOI:
10.1371/journal.pone.0182362
[Indexed for MEDLINE]
Free PMC Article

Supplemental Content

Full text links

Icon for Public Library of Science Icon for PubMed Central
Loading ...
Support Center