Inverse probability weighting to handle attrition in cohort studies: some guidance and a call for caution.
Attrition
Cohort studies
Complete-case analysis
Inverse probability weighting
Missing outcome
Selection bias
Journal
BMC medical research methodology
ISSN: 1471-2288
Titre abrégé: BMC Med Res Methodol
Pays: England
ID NLM: 100968545
Informations de publication
Date de publication:
16 02 2022
16 02 2022
Historique:
received:
02
06
2021
accepted:
25
01
2022
entrez:
17
2
2022
pubmed:
18
2
2022
medline:
22
3
2022
Statut:
epublish
Résumé
Attrition in cohort studies challenges causal inference. Although inverse probability weighting (IPW) has been proposed to handle attrition in association analyses, its relevance has been little studied in this context. We aimed to investigate its ability to correct for selection bias in exposure-outcome estimation by addressing an important methodological issue: the specification of the response model. A simulation study compared the IPW method with complete-case analysis (CCA) for nine response-mechanism scenarios (3 missing at random - MAR and 6 missing not at random - MNAR). Eighteen response models differing by the type of variables included were assessed. The IPW method was equivalent to CCA in terms of bias and consistently less efficient in all scenarios, regardless of the response model tested. The most effective response model included only the confounding factors of the association model. Our study questions the ability of the IPW method to correct for selection bias in situations of attrition leading to missing outcomes. If the method is to be used, we encourage including only the confounding variables of the association of interest in the response model.
Sections du résumé
BACKGROUND
Attrition in cohort studies challenges causal inference. Although inverse probability weighting (IPW) has been proposed to handle attrition in association analyses, its relevance has been little studied in this context. We aimed to investigate its ability to correct for selection bias in exposure-outcome estimation by addressing an important methodological issue: the specification of the response model.
METHODS
A simulation study compared the IPW method with complete-case analysis (CCA) for nine response-mechanism scenarios (3 missing at random - MAR and 6 missing not at random - MNAR). Eighteen response models differing by the type of variables included were assessed.
RESULTS
The IPW method was equivalent to CCA in terms of bias and consistently less efficient in all scenarios, regardless of the response model tested. The most effective response model included only the confounding factors of the association model.
CONCLUSION
Our study questions the ability of the IPW method to correct for selection bias in situations of attrition leading to missing outcomes. If the method is to be used, we encourage including only the confounding variables of the association of interest in the response model.
Identifiants
pubmed: 35172753
doi: 10.1186/s12874-022-01533-9
pii: 10.1186/s12874-022-01533-9
pmc: PMC8848672
doi:
Types de publication
Journal Article
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
45Informations de copyright
© 2022. The Author(s).
Références
Stat Med. 2011 Feb 20;30(4):377-99
pubmed: 21225900
Epidemiology. 1999 Jan;10(1):37-48
pubmed: 9888278
Clin Trials. 2014 May 28;11(4):494-502
pubmed: 24872362
J Pulm Respir Med. 2017 Aug;7(4):
pubmed: 28966879
Matern Child Nutr. 2016 Oct;12(4):918-28
pubmed: 25753294
Neurotoxicology. 2013 Mar;35:162-8
pubmed: 23376090
Alcohol Alcohol. 2018 May 1;53(3):251-258
pubmed: 29329371
Eur Child Adolesc Psychiatry. 2014 Dec;23(12):1219-30
pubmed: 25130265
Stat Methods Med Res. 2013 Jun;22(3):278-95
pubmed: 21220355
Am J Epidemiol. 2014 Mar 1;179(5):536-44
pubmed: 24401561
Int J Epidemiol. 2019 Aug 1;48(4):1294-1304
pubmed: 30879056
Environ Res. 2012 Oct;118:79-85
pubmed: 22910562
Am J Epidemiol. 2006 Jun 15;163(12):1149-56
pubmed: 16624967
Soc Sci Med. 2018 Jan;197:136-143
pubmed: 29232621
BMC Med Res Methodol. 2018 Jun 20;18(1):60
pubmed: 29925318
Epidemiology. 2004 Sep;15(5):615-25
pubmed: 15308962
Epidemiology. 2018 Jan;29(1):87-95
pubmed: 28926372
Environ Res. 2016 Nov;151:436-444
pubmed: 27560981
Stat Med. 2007 Feb 20;26(4):734-53
pubmed: 16708349
Eur J Epidemiol. 2019 Oct;34(10):927-938
pubmed: 31451995
Epidemiology. 2012 Jan;23(1):119-28
pubmed: 21989136
Eur J Epidemiol. 2005;20(8):657-62
pubmed: 16151878
Eur J Epidemiol. 2019 Jul;34(7):661-673
pubmed: 31062119
Multivariate Behav Res. 2011 May;46(3):399-424
pubmed: 21818162
Stat Med. 2015 Dec 10;34(28):3661-79
pubmed: 26238958
Biometrics. 2012 Mar;68(1):129-37
pubmed: 22050039
Soc Sci Med. 2019 Jan;220:371-378
pubmed: 30513487