TY - JOUR
T1 - Improving data sharing in research with context-free encoded missing data
AU - Hoevenaar-Blom, Marieke P.
AU - Guillemont, Juliette
AU - Ngandu, Tiia
AU - Beishuizen, Cathrien R. L.
AU - Coley, Nicola
AU - Moll van Charante, Eric P.
AU - Andrieu, Sandrine
AU - Kivipelto, Miia
AU - Soininen, Hilkka
AU - Brayne, Carol
AU - Meiller, Yannick
AU - Richard, Edo
PY - 2017
Y1 - 2017
N2 - 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
AB - 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
U2 - https://doi.org/10.1371/journal.pone.0182362
DO - https://doi.org/10.1371/journal.pone.0182362
M3 - Article
C2 - 28898245
SN - 1932-6203
VL - 12
SP - e0182362
JO - PLOS ONE
JF - PLOS ONE
IS - 9
ER -